| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2 |
| ; RUN: llc -mtriple=amdgcn -global-isel=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=SI,SI-SDAG %s |
| ; RUN: llc -mtriple=amdgcn -global-isel=1 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=SI,SI-GISEL %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=tonga -global-isel=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=VI-SDAG %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=tonga -global-isel=1 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=VI-GISEL %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -global-isel=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-SDAG %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -global-isel=1 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-GISEL %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -global-isel=0 -mattr=-real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -global-isel=1 -mattr=-real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -global-isel=0 -mattr=+real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s |
| ; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -global-isel=1 -mattr=+real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s |
| |
| define amdgpu_kernel void @v_test_nnan_input_fmed3_r_i_i_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_nnan_input_fmed3_r_i_i_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_nnan_input_fmed3_r_i_i_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_nnan_input_fmed3_r_i_i_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_nnan_input_fmed3_r_i_i_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_nnan_input_fmed3_r_i_i_f32: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_nnan_input_fmed3_r_i_i_f32: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan float %a, 1.0 |
| %max = call float @llvm.maxnum.f32(float %a.add, float 2.0) |
| %med = call float @llvm.minnum.f32(float %max, float 4.0) |
| |
| store float %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_fmed3_nnan_r_i_i_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_fmed3_nnan_r_i_i_f32: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_fmed3_nnan_r_i_i_f32: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan float %a, 1.0 |
| |
| %max = call float @llvm.maxnum.f32(float %a.add, float 2.0) |
| %med = call float @llvm.minnum.f32(float %max, float 4.0) |
| |
| store float %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_fmed3_nnan_r_i_i_commute0_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_commute0_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_commute0_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_commute0_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_commute0_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_fmed3_nnan_r_i_i_commute0_f32: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_fmed3_nnan_r_i_i_commute0_f32: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan float %a, 1.0 |
| |
| %max = call float @llvm.maxnum.f32(float 2.0, float %a.add) |
| %med = call float @llvm.minnum.f32(float 4.0, float %max) |
| |
| store float %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_fmed3_nnan_r_i_i_commute1_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_commute1_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_commute1_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_commute1_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_commute1_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_fmed3_nnan_r_i_i_commute1_f32: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_fmed3_nnan_r_i_i_commute1_f32: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan float %a, 1.0 |
| |
| %max = call float @llvm.maxnum.f32(float %a.add, float 2.0) |
| %med = call float @llvm.minnum.f32(float 4.0, float %max) |
| |
| store float %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_fmed3_nnan_r_i_i_constant_order_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_constant_order_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, 4.0, v2 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, 2.0, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_constant_order_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, 4.0, v2 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, 2.0, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_constant_order_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, 4.0, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_constant_order_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, 4.0, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, 2.0, v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_fmed3_nnan_r_i_i_constant_order_f32: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_max_f32_e32 v1, 4.0, v1 |
| ; GFX9-NEXT: v_min_f32_e32 v1, 2.0, v1 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_fmed3_nnan_r_i_i_constant_order_f32: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-NEXT: v_maxmin_f32 v1, v1, 4.0, 2.0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan float %a, 1.0 |
| |
| %max = call float @llvm.maxnum.f32(float %a.add, float 4.0) |
| %med = call float @llvm.minnum.f32(float %max, float 2.0) |
| |
| store float %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_fmed3_nnan_r_i_i_multi_use_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, 2.0, v2 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v3, 4.0, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v3, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v3, 2.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: buffer_store_dword v3, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v3, 4.0, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v3 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v3, 2.0, v2 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v3 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-SDAG-NEXT: v_max_f32_e32 v1, 2.0, v1 |
| ; GFX9-SDAG-NEXT: v_min_f32_e32 v2, 4.0, v1 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v2, s[0:1] |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-GISEL-NEXT: v_max_f32_e32 v2, 2.0, v1 |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v2, s[0:1] |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-SDAG-NEXT: v_max_f32_e32 v1, 2.0, v1 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_min_f32_e32 v2, 4.0, v1 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v2, s[0:1] dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_fmed3_nnan_r_i_i_multi_use_f32: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-GISEL-NEXT: v_med3_f32 v2, v1, 2.0, 4.0 |
| ; GFX11-GISEL-NEXT: v_max_f32_e32 v1, 2.0, v1 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v2, s[0:1] dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan float %a, 1.0 |
| |
| %max = call float @llvm.maxnum.f32(float %a.add, float 2.0) |
| %med = call float @llvm.minnum.f32(float %max, float 4.0) |
| |
| store volatile float %med, ptr addrspace(1) %outgep |
| store volatile float %max, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_fmed3_r_i_i_f64(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_fmed3_r_i_i_f64: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 3, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f64 v[2:3], v[2:3], 1.0 |
| ; SI-SDAG-NEXT: v_max_f64 v[2:3], v[2:3], 2.0 |
| ; SI-SDAG-NEXT: v_min_f64 v[2:3], v[2:3], 4.0 |
| ; SI-SDAG-NEXT: buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_fmed3_r_i_i_f64: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 3, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f64 v[2:3], v[2:3], 1.0 |
| ; SI-GISEL-NEXT: v_max_f64 v[2:3], v[2:3], 2.0 |
| ; SI-GISEL-NEXT: v_min_f64 v[2:3], v[2:3], 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_fmed3_r_i_i_f64: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 3, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dwordx2 v[0:1], v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f64 v[0:1], v[0:1], 1.0 |
| ; VI-SDAG-NEXT: v_max_f64 v[0:1], v[0:1], 2.0 |
| ; VI-SDAG-NEXT: v_min_f64 v[0:1], v[0:1], 4.0 |
| ; VI-SDAG-NEXT: flat_store_dwordx2 v[2:3], v[0:1] |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_fmed3_r_i_i_f64: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v4, 3, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v4 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dwordx2 v[0:1], v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s1 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s0 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v4 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f64 v[0:1], v[0:1], 1.0 |
| ; VI-GISEL-NEXT: v_max_f64 v[0:1], v[0:1], 2.0 |
| ; VI-GISEL-NEXT: v_min_f64 v[0:1], v[0:1], 4.0 |
| ; VI-GISEL-NEXT: flat_store_dwordx2 v[2:3], v[0:1] |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_fmed3_r_i_i_f64: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v2, 3, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dwordx2 v[0:1], v2, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f64 v[0:1], v[0:1], 1.0 |
| ; GFX9-NEXT: v_max_f64 v[0:1], v[0:1], 2.0 |
| ; GFX9-NEXT: v_min_f64 v[0:1], v[0:1], 4.0 |
| ; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_fmed3_r_i_i_f64: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v2, 3, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b64 v[0:1], v2, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], 1.0 |
| ; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], 2.0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_min_f64 v[0:1], v[0:1], 4.0 |
| ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr double, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr double, ptr addrspace(1) %out, i32 %tid |
| %a = load double, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan double %a, 1.0 |
| |
| %max = call double @llvm.maxnum.f64(double %a.add, double 2.0) |
| %med = call double @llvm.minnum.f64(double %max, double 4.0) |
| |
| store double %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_fmed3_r_i_i_no_nans_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #2 { |
| ; SI-SDAG-LABEL: v_test_fmed3_r_i_i_no_nans_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_fmed3_r_i_i_no_nans_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_fmed3_r_i_i_no_nans_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v3, 2.0, 4.0 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_fmed3_r_i_i_no_nans_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v3, 2.0, 4.0 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_fmed3_r_i_i_no_nans_f32: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_fmed3_r_i_i_no_nans_f32: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| |
| %max = call float @llvm.maxnum.f32(float %a, float 2.0) |
| %med = call float @llvm.minnum.f32(float %max, float 4.0) |
| |
| store float %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_legacy_fmed3_r_i_i_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_max_legacy_f32_e64 v2, v2, 2.0 |
| ; SI-GISEL-NEXT: v_min_legacy_f32_e64 v2, v2, 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_cmp_nlt_f32_e32 vcc, 2.0, v2 |
| ; VI-GISEL-NEXT: v_cndmask_b32_e64 v2, v2, 2.0, vcc |
| ; VI-GISEL-NEXT: v_cmp_ngt_f32_e32 vcc, 4.0, v2 |
| ; VI-GISEL-NEXT: v_cndmask_b32_e64 v2, v2, 4.0, vcc |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-GISEL-NEXT: v_cmp_nlt_f32_e32 vcc, 2.0, v1 |
| ; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v1, 2.0, vcc |
| ; GFX9-GISEL-NEXT: v_cmp_ngt_f32_e32 vcc, 4.0, v1 |
| ; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v1, 4.0, vcc |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, v1, 2.0, 4.0 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_legacy_fmed3_r_i_i_f32: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX11-GISEL-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 2.0, v1 |
| ; GFX11-GISEL-NEXT: v_cndmask_b32_e64 v1, v1, 2.0, vcc_lo |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 4.0, v1 |
| ; GFX11-GISEL-NEXT: v_cndmask_b32_e64 v1, v1, 4.0, vcc_lo |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %a.nnan = fadd nnan float %a, 1.0 |
| |
| ; fmax_legacy |
| %cmp0 = fcmp ule float %a.nnan, 2.0 |
| %max = select i1 %cmp0, float 2.0, float %a.nnan |
| |
| ; fmin_legacy |
| %cmp1 = fcmp uge float %max, 4.0 |
| %med = select i1 %cmp1, float 4.0, float %max |
| |
| store float %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat0_srcmod0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, -v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, -1.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, -v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, -1.0, v7 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, -v1, v2, v3 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, -v1, v2, v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %a.fneg = fsub float -0.0, %a |
| %tmp0 = call float @llvm.minnum.f32(float %a.fneg, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a.fneg, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat0_srcmod1(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, -v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, -1.0, v3 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, -v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v2, -1.0, v2 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, v1, -v2, v3 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v2, -v2, -v2 |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, v1, -v2, v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod1: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v2, -v2, -v2 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %b.fneg = fsub float -0.0, %b |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b.fneg) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b.fneg) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat0_srcmod2(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, -v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v4, -1.0, v4 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, -v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v3, -1.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, v1, v2, -v3 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v3, -v3, -v3 |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, v1, v2, -v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod2: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v3, -v3, -v3 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %c.fneg = fsub float -0.0, %c |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.fneg) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat0_srcmod012(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, -v2, |v3|, -|v4| |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, -1.0, v2 |
| ; SI-GISEL-NEXT: v_mul_f32_e64 v4, -1.0, |v4| |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, |v3|, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, -v7, |v2|, -|v3| |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, -1.0, v7 |
| ; VI-GISEL-NEXT: v_mul_f32_e64 v3, -1.0, |v3| |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, |v2|, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, -v1, |v2|, -|v3| |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v3, -|v3|, -|v3| |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, |v2|, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, -v1, |v2|, -|v3| |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod012: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v3, -|v3|, -|v3| |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_med3_f32 v1, v1, |v2|, v3 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| |
| %a.fneg = fsub float -0.0, %a |
| %b.fabs = call float @llvm.fabs.f32(float %b) |
| %c.fabs = call float @llvm.fabs.f32(float %c) |
| %c.fabs.fneg = fsub float -0.0, %c.fabs |
| |
| %tmp0 = call float @llvm.minnum.f32(float %a.fneg, float %b.fabs) |
| %tmp1 = call float @llvm.maxnum.f32(float %a.fneg, float %b.fabs) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.fabs.fneg) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat0_negabs012(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, -|v2|, -|v3|, -|v4| |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e64 v2, -1.0, |v2| |
| ; SI-GISEL-NEXT: v_mul_f32_e64 v3, -1.0, |v3| |
| ; SI-GISEL-NEXT: v_mul_f32_e64 v4, -1.0, |v4| |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, -|v7|, -|v2|, -|v3| |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e64 v4, -1.0, |v7| |
| ; VI-GISEL-NEXT: v_mul_f32_e64 v2, -1.0, |v2| |
| ; VI-GISEL-NEXT: v_mul_f32_e64 v3, -1.0, |v3| |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, -|v1|, -|v2|, -|v3| |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v1, -|v1|, -|v1| |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v2, -|v2|, -|v2| |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v3, -|v3|, -|v3| |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, -|v1|, -|v2|, -|v3| |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_negabs012: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v1, -|v1|, -|v1| |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v2, -|v2|, -|v2| |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v3, -|v3|, -|v3| |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| |
| %a.fabs = call float @llvm.fabs.f32(float %a) |
| %a.fabs.fneg = fsub float -0.0, %a.fabs |
| %b.fabs = call float @llvm.fabs.f32(float %b) |
| %b.fabs.fneg = fsub float -0.0, %b.fabs |
| %c.fabs = call float @llvm.fabs.f32(float %c) |
| %c.fabs.fneg = fsub float -0.0, %c.fabs |
| |
| %tmp0 = call float @llvm.minnum.f32(float %a.fabs.fneg, float %b.fabs.fneg) |
| %tmp1 = call float @llvm.maxnum.f32(float %a.fabs.fneg, float %b.fabs.fneg) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.fabs.fneg) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_nnan_inputs_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_nnan_inputs_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_nnan_inputs_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_nnan_inputs_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_nnan_inputs_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_nnan_inputs_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; GFX9-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_nnan_inputs_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v2, 2.0, v2 |
| ; GFX11-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| |
| %a.nnan = fadd nnan float %a, 1.0 |
| %b.nnan = fadd nnan float %b, 2.0 |
| %c.nnan = fadd nnan float %c, 4.0 |
| |
| %tmp0 = call float @llvm.minnum.f32(float %a.nnan, float %b.nnan) |
| %tmp1 = call float @llvm.maxnum.f32(float %a.nnan, float %b.nnan) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.nnan) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_nnan_input_calls_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_nnan_input_calls_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_nnan_input_calls_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_nnan_input_calls_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_nnan_input_calls_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_nnan_input_calls_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_nnan_input_calls_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call nnan float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call nnan float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call nnan float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_nnan_call_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_nnan_call_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_nnan_call_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_nnan_call_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_nnan_call_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_nnan_call_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_nnan_call_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call nnan float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call nnan float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call nnan float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call nnan float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_fast_call_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_fast_call_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_fast_call_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_fast_call_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_fast_call_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_fast_call_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_fast_call_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call fast float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call fast float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call fast float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call fast float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| ; 16 combinations |
| |
| ; 0: max(min(x, y), min(max(x, y), z)) |
| ; 1: max(min(x, y), min(max(y, x), z)) |
| ; 2: max(min(x, y), min(z, max(x, y))) |
| ; 3: max(min(x, y), min(z, max(y, x))) |
| ; 4: max(min(y, x), min(max(x, y), z)) |
| ; 5: max(min(y, x), min(max(y, x), z)) |
| ; 6: max(min(y, x), min(z, max(x, y))) |
| ; 7: max(min(y, x), min(z, max(y, x))) |
| ; + commute outermost max |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat1(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat1: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat1: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat1: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat1: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat1: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat1: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat1_srcmod0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, -v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, -1.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, -v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, -1.0, v7 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, -v1, v2, v3 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, -v1, v2, v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_global_nnans_med3_f32_pat1_srcmod0: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %a.fneg = fsub float -0.0, %a |
| %tmp0 = call float @llvm.maxnum.f32(float %a.fneg, float %b) |
| %tmp1 = call float @llvm.minnum.f32(float %a.fneg, float %b) |
| %tmp2 = call float @llvm.maxnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.minnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat2(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat2: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat2: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat2: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat2: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat2: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat2: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat3(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat3: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat3: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat3: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat3: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat3: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat3: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat4(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat4: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat4: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat4: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat4: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat4: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat4: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat5(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat5: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat5: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat5: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat5: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat5: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat5: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat6(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat6: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat6: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat6: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat6: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat6: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat6: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat7(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat7: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat7: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat7: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat7: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat7: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat7: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat8(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat8: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat8: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat8: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat8: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat8: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat8: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat9(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat9: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat9: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat9: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat9: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat9: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat9: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat10(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat10: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat10: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat10: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat10: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat10: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat10: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat11(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat11: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat11: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat11: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat11: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat11: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat11: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat12(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat12: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat12: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat12: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat12: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat12: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat12: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat13(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat13: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat13: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat13: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat13: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat13: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat13: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat14(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat14: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat14: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat14: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat14: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat14: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat14: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat15(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat15: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat15: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v3, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat15: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat15: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v7, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat15: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat15: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v2, v1, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %b, float %a) |
| %tmp1 = call float @llvm.maxnum.f32(float %b, float %a) |
| %tmp2 = call float @llvm.minnum.f32(float %c, float %tmp1) |
| %med3 = call float @llvm.maxnum.f32(float %tmp2, float %tmp0) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| ; Also handle `min` at the root: |
| ; min(max(x, y), max(min(x, y), z)) |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat16(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat16: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat16: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat16: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat16: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v7, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_med3_f32_pat16: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_med3_f32_pat16: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.maxnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.minnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| ; --------------------------------------------------------------------- |
| ; Negative patterns |
| ; --------------------------------------------------------------------- |
| |
| define amdgpu_kernel void @v_test_safe_med3_f32_pat0_multi_use0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_test_safe_med3_f32_pat0_multi_use0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b32 s10, -1 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v4, 1.0, v4 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-SDAG-NEXT: buffer_store_dword v5, off, s[8:11], 0 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, v2, v4 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_safe_med3_f32_pat0_multi_use0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b32 s2, -1 |
| ; SI-GISEL-NEXT: s_mov_b32 s3, s11 |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v4 |
| ; SI-GISEL-NEXT: buffer_store_dword v5, off, s[0:3], 0 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_safe_med3_f32_pat0_multi_use0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v5 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_safe_med3_f32_pat0_multi_use0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v5 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_safe_med3_f32_pat0_multi_use0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v1 |
| ; GFX9-NEXT: v_max_f32_e32 v2, v2, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v3, v3, v3 |
| ; GFX9-NEXT: v_min_f32_e32 v4, v1, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX9-NEXT: global_store_dword v[0:1], v4, off |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_min_f32_e32 v1, v1, v3 |
| ; GFX9-NEXT: v_max_f32_e32 v1, v4, v1 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_safe_med3_f32_pat0_multi_use0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v2, v2, v2 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_min_f32_e32 v4, v1, v2 |
| ; GFX11-NEXT: v_dual_max_f32 v1, v1, v2 :: v_dual_max_f32 v2, v3, v3 |
| ; GFX11-NEXT: v_minmax_f32 v1, v1, v2, v4 |
| ; GFX11-NEXT: global_store_b32 v[0:1], v4, off dlc |
| ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| store volatile float %tmp0, ptr addrspace(1) undef |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_safe_med3_f32_pat0_multi_use1(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_test_safe_med3_f32_pat0_multi_use1: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b32 s10, -1 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v4, 1.0, v4 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, off, s[8:11], 0 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, v2, v4 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_safe_med3_f32_pat0_multi_use1: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b32 s2, -1 |
| ; SI-GISEL-NEXT: s_mov_b32 s3, s11 |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v4 |
| ; SI-GISEL-NEXT: buffer_store_dword v2, off, s[0:3], 0 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_safe_med3_f32_pat0_multi_use1: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_safe_med3_f32_pat0_multi_use1: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_safe_med3_f32_pat0_multi_use1: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v1 |
| ; GFX9-NEXT: v_max_f32_e32 v2, v2, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v3, v3, v3 |
| ; GFX9-NEXT: v_min_f32_e32 v4, v1, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX9-NEXT: global_store_dword v[0:1], v1, off |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_min_f32_e32 v1, v1, v3 |
| ; GFX9-NEXT: v_max_f32_e32 v1, v4, v1 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_safe_med3_f32_pat0_multi_use1: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_max_f32_e32 v1, v1, v1 |
| ; GFX11-SDAG-NEXT: v_dual_max_f32 v3, v3, v3 :: v_dual_max_f32 v2, v2, v2 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_max_f32_e32 v4, v1, v2 |
| ; GFX11-SDAG-NEXT: v_min_f32_e32 v3, v4, v3 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_minmax_f32 v1, v1, v2, v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v[0:1], v4, off dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_safe_med3_f32_pat0_multi_use1: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v2, v2, v2 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_min_f32_e32 v4, v1, v2 |
| ; GFX11-GISEL-NEXT: v_dual_max_f32 v1, v1, v2 :: v_dual_max_f32 v2, v3, v3 |
| ; GFX11-GISEL-NEXT: v_minmax_f32 v2, v1, v2, v4 |
| ; GFX11-GISEL-NEXT: global_store_b32 v[0:1], v1, off dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v2, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| store volatile float %tmp1, ptr addrspace(1) undef |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_safe_med3_f32_pat0_multi_use2(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_test_safe_med3_f32_pat0_multi_use2: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b32 s10, -1 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v4, 1.0, v4 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, off, s[8:11], 0 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_safe_med3_f32_pat0_multi_use2: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b32 s2, -1 |
| ; SI-GISEL-NEXT: s_mov_b32 s3, s11 |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v4 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: buffer_store_dword v2, off, s[0:3], 0 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_safe_med3_f32_pat0_multi_use2: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_safe_med3_f32_pat0_multi_use2: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_safe_med3_f32_pat0_multi_use2: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v1 |
| ; GFX9-NEXT: v_max_f32_e32 v2, v2, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v3, v3, v3 |
| ; GFX9-NEXT: v_min_f32_e32 v4, v1, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX9-NEXT: v_min_f32_e32 v1, v1, v3 |
| ; GFX9-NEXT: global_store_dword v[0:1], v1, off |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_max_f32_e32 v1, v4, v1 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_safe_med3_f32_pat0_multi_use2: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v2, v2, v2 |
| ; GFX11-NEXT: v_max_f32_e32 v3, v3, v3 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) |
| ; GFX11-NEXT: v_maxmin_f32 v3, v1, v2, v3 |
| ; GFX11-NEXT: v_minmax_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v[0:1], v3, off dlc |
| ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| store volatile float %tmp2, ptr addrspace(1) undef |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_safe_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_test_safe_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[6:7] |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v4 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_safe_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v4 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_safe_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v4, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v4 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v4 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: flat_load_dword v6, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s6, v4 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v4 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v4, 1.0, v6 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-SDAG-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_safe_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v2, 1.0, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v5, v4, v2 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v3, 1.0, v3 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_safe_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v1 |
| ; GFX9-NEXT: v_max_f32_e32 v2, v2, v2 |
| ; GFX9-NEXT: v_min_f32_e32 v4, v1, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v2, v3, v3 |
| ; GFX9-NEXT: v_min_f32_e32 v1, v1, v2 |
| ; GFX9-NEXT: v_max_f32_e32 v1, v4, v1 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_safe_med3_f32_pat0: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v2, v2, v2 |
| ; GFX11-SDAG-NEXT: v_max_f32_e32 v3, v3, v3 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_maxmin_f32 v3, v1, v2, v3 |
| ; GFX11-SDAG-NEXT: v_minmax_f32 v1, v1, v2, v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_safe_med3_f32_pat0: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v2, v2, v2 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_min_f32_e32 v4, v1, v2 |
| ; GFX11-GISEL-NEXT: v_dual_max_f32 v1, v1, v2 :: v_dual_max_f32 v2, v3, v3 |
| ; GFX11-GISEL-NEXT: v_minmax_f32 v1, v1, v2, v4 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %tmp0 = call float @llvm.minnum.f32(float %a, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_nnan_inputs_missing0_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_nnan_inputs_missing0_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_nnan_inputs_missing0_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_nnan_inputs_missing0_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_nnan_inputs_missing0_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_nnan_inputs_missing0_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; GFX9-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_nnan_inputs_missing0_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v2, 2.0, v2 |
| ; GFX11-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| |
| %a.nnan = fadd float %a, 1.0 |
| %b.nnan = fadd nnan float %b, 2.0 |
| %c.nnan = fadd nnan float %c, 4.0 |
| |
| %tmp0 = call float @llvm.minnum.f32(float %a.nnan, float %b.nnan) |
| %tmp1 = call float @llvm.maxnum.f32(float %a.nnan, float %b.nnan) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.nnan) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_nnan_inputs_missing1_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_nnan_inputs_missing1_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_nnan_inputs_missing1_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_nnan_inputs_missing1_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_nnan_inputs_missing1_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_nnan_inputs_missing1_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; GFX9-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_nnan_inputs_missing1_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v2, 2.0, v2 |
| ; GFX11-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| |
| %a.nnan = fadd nnan float %a, 1.0 |
| %b.nnan = fadd float %b, 2.0 |
| %c.nnan = fadd nnan float %c, 4.0 |
| |
| %tmp0 = call float @llvm.minnum.f32(float %a.nnan, float %b.nnan) |
| %tmp1 = call float @llvm.maxnum.f32(float %a.nnan, float %b.nnan) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.nnan) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_nnan_inputs_missing2_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_nnan_inputs_missing2_med3_f32_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_nnan_inputs_missing2_med3_f32_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_nnan_inputs_missing2_med3_f32_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_nnan_inputs_missing2_med3_f32_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_add_f32_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_nnan_inputs_missing2_med3_f32_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_add_f32_e32 v2, 2.0, v2 |
| ; GFX9-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_nnan_inputs_missing2_med3_f32_pat0: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v2, 2.0, v2 |
| ; GFX11-NEXT: v_add_f32_e32 v3, 4.0, v3 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| |
| %a.nnan = fadd nnan float %a, 1.0 |
| %b.nnan = fadd nnan float %b, 2.0 |
| %c.nnan = fadd float %c, 4.0 |
| |
| %tmp0 = call float @llvm.minnum.f32(float %a.nnan, float %b.nnan) |
| %tmp1 = call float @llvm.maxnum.f32(float %a.nnan, float %b.nnan) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.nnan) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_nnan_on_call_med3_f32_pat0_srcmod0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_med3_f32 v2, -v2, v3, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v2, -1.0, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_med3_f32 v2, -v7, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, -1.0, v7 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_med3_f32 v1, -v1, v2, v3 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX9-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v1, -v1, v2, v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_nnan_on_call_med3_f32_pat0_srcmod0: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v1, -v1, -v1 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_med3_f32 v1, v1, v2, v3 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %a.fneg = fsub float -0.0, %a |
| %tmp0 = call nnan float @llvm.minnum.f32(float %a.fneg, float %b) |
| %tmp1 = call nnan float @llvm.maxnum.f32(float %a.fneg, float %b) |
| %tmp2 = call nnan float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call nnan float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_min_f32_e64 v5, -v2, v3 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, v2, v4 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_mul_f32_e32 v5, -1.0, v2 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v5, v5, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v4 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v5, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v4, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v4 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v4 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: flat_load_dword v6, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s6, v4 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v4 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_min_f32_e64 v4, -v6, v2 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v6, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v4, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mul_f32_e32 v4, -1.0, v7 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v5, v7, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, v4, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v3, v5, v3 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_min_f32_e64 v4, -v1, v2 |
| ; GFX9-SDAG-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX9-SDAG-NEXT: v_min_f32_e32 v1, v1, v3 |
| ; GFX9-SDAG-NEXT: v_max_f32_e32 v1, v4, v1 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_max_f32_e64 v4, -v1, -v1 |
| ; GFX9-GISEL-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX9-GISEL-NEXT: v_min_f32_e32 v2, v4, v2 |
| ; GFX9-GISEL-NEXT: v_min_f32_e32 v1, v1, v3 |
| ; GFX9-GISEL-NEXT: v_max_f32_e32 v1, v2, v1 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_maxmin_f32 v3, v1, v2, v3 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_minmax_f32 v1, -v1, v2, v3 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: v_test_global_nnans_med3_f32_pat0_srcmod0_mismatch: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_max_f32_e64 v4, -v1, -v1 |
| ; GFX11-GISEL-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_min_f32_e32 v4, v4, v2 |
| ; GFX11-GISEL-NEXT: v_minmax_f32 v1, v1, v3, v4 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %a.fneg = fsub float -0.0, %a |
| %tmp0 = call float @llvm.minnum.f32(float %a.fneg, float %b) |
| %tmp1 = call float @llvm.maxnum.f32(float %a, float %b) |
| %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c) |
| %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2) |
| store float %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| ; A simple min and max is not sufficient |
| define amdgpu_kernel void @v_test_global_nnans_min_max_f32(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #2 { |
| ; SI-SDAG-LABEL: v_test_global_nnans_min_max_f32: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_dword v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, v2, v4 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_global_nnans_min_max_f32: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_dword v3, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_dword v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v4 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_global_nnans_min_max_f32: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, v7, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_global_nnans_min_max_f32: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, v7, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_global_nnans_min_max_f32: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_dword v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_max_f32_e32 v1, v1, v2 |
| ; GFX9-NEXT: v_min_f32_e32 v1, v1, v3 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: v_test_global_nnans_min_max_f32: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v2, v0, s[4:5] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v3, v0, s[6:7] glc dlc |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_maxmin_f32 v1, v1, v2, v3 |
| ; GFX11-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile float, ptr addrspace(1) %gep0 |
| %b = load volatile float, ptr addrspace(1) %gep1 |
| %c = load volatile float, ptr addrspace(1) %gep2 |
| %max = call float @llvm.maxnum.f32(float %a, float %b) |
| %minmax = call float @llvm.minnum.f32(float %max, float %c) |
| store float %minmax, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_test_nnan_input_fmed3_r_i_i_f16(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, 2.0, 4.0 |
| ; SI-SDAG-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-SDAG-NEXT: buffer_store_short v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 1.0 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, 2.0 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v4 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 4.0 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_short v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 1, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_ushort v3, v[0:1] |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f16_e32 v2, 1.0, v3 |
| ; VI-SDAG-NEXT: v_max_f16_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: v_min_f16_e32 v2, 4.0, v2 |
| ; VI-SDAG-NEXT: flat_store_short v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 1, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_ushort v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f16_e32 v2, 1.0, v3 |
| ; VI-GISEL-NEXT: v_max_f16_e32 v2, 2.0, v2 |
| ; VI-GISEL-NEXT: v_min_f16_e32 v2, 4.0, v2 |
| ; VI-GISEL-NEXT: flat_store_short v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_ushort v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f16_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_med3_f16 v1, v1, 2.0, 4.0 |
| ; GFX9-NEXT: global_store_short v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-FAKE16-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; GFX11-SDAG-FAKE16: ; %bb.0: |
| ; GFX11-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-SDAG-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; GFX11-SDAG-FAKE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3] |
| ; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v1, 1.0, v1 |
| ; GFX11-SDAG-FAKE16-NEXT: v_med3_f16 v1, v1, 2.0, 4.0 |
| ; GFX11-SDAG-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] |
| ; GFX11-SDAG-FAKE16-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-FAKE16-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; GFX11-GISEL-FAKE16: ; %bb.0: |
| ; GFX11-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-GISEL-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3] |
| ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v1, 1.0, v1 |
| ; GFX11-GISEL-FAKE16-NEXT: v_med3_f16 v1, v1, 2.0, 4.0 |
| ; GFX11-GISEL-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] |
| ; GFX11-GISEL-FAKE16-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-TRUE16-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; GFX11-SDAG-TRUE16: ; %bb.0: |
| ; GFX11-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 1, v0 |
| ; GFX11-SDAG-TRUE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-TRUE16-NEXT: global_load_u16 v0, v1, s[2:3] |
| ; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, 1.0, v0.l |
| ; GFX11-SDAG-TRUE16-NEXT: v_med3_f16 v0.l, v0.l, 2.0, 4.0 |
| ; GFX11-SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] |
| ; GFX11-SDAG-TRUE16-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-TRUE16-LABEL: v_test_nnan_input_fmed3_r_i_i_f16: |
| ; GFX11-GISEL-TRUE16: ; %bb.0: |
| ; GFX11-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-GISEL-TRUE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 1, v0 |
| ; GFX11-GISEL-TRUE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v0, v1, s[2:3] |
| ; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, 1.0, v0.l |
| ; GFX11-GISEL-TRUE16-NEXT: v_med3_f16 v0.l, v0.l, 2.0, 4.0 |
| ; GFX11-GISEL-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] |
| ; GFX11-GISEL-TRUE16-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr half, ptr addrspace(1) %aptr, i32 %tid |
| %outgep = getelementptr half, ptr addrspace(1) %out, i32 %tid |
| %a = load half, ptr addrspace(1) %gep0 |
| %a.add = fadd nnan half %a, 1.0 |
| %max = call half @llvm.maxnum.f16(half %a.add, half 2.0) |
| %med = call half @llvm.minnum.f16(half %max, half 4.0) |
| |
| store half %med, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @v_nnan_inputs_med3_f16_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) #1 { |
| ; SI-SDAG-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s10, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_mov_b64 s[14:15], s[10:11] |
| ; SI-SDAG-NEXT: s_mov_b64 s[18:19], s[10:11] |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-SDAG-NEXT: s_mov_b64 s[12:13], s[4:5] |
| ; SI-SDAG-NEXT: s_mov_b64 s[16:17], s[6:7] |
| ; SI-SDAG-NEXT: buffer_load_ushort v2, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_ushort v3, v[0:1], s[12:15], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_load_ushort v4, v[0:1], s[16:19], 0 addr64 glc |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-SDAG-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-SDAG-NEXT: v_cvt_f32_f16_e32 v3, v3 |
| ; SI-SDAG-NEXT: v_cvt_f32_f16_e32 v4, v4 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 1.0, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v3, 2.0, v3 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v4, 4.0, v4 |
| ; SI-SDAG-NEXT: v_med3_f32 v2, v2, v3, v4 |
| ; SI-SDAG-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-SDAG-NEXT: buffer_store_short v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s10, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, 1.0 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 2.0 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_ushort v4, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v5, 4.0 |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] |
| ; SI-GISEL-NEXT: buffer_load_ushort v6, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] |
| ; SI-GISEL-NEXT: buffer_load_ushort v7, v[0:1], s[8:11], 0 addr64 glc |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, v4 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, v4, v2 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, v6 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v3, v4, v3 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, v7 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v3, v3 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v4, v4, v5 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v4, v4 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, v3 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v5, v2, v3 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, v4 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v4, v5 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, v4 |
| ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, v3, v2 |
| ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11] |
| ; SI-GISEL-NEXT: buffer_store_short v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v6, 1, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v2, vcc, s4, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v4, vcc, s6, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-SDAG-NEXT: flat_load_ushort v7, v[0:1] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_ushort v2, v[2:3] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_load_ushort v3, v[4:5] glc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v6 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: v_add_f16_e32 v4, 1.0, v7 |
| ; VI-SDAG-NEXT: v_add_f16_e32 v2, 2.0, v2 |
| ; VI-SDAG-NEXT: v_add_f16_e32 v3, 4.0, v3 |
| ; VI-SDAG-NEXT: v_min_f16_e32 v5, v4, v2 |
| ; VI-SDAG-NEXT: v_max_f16_e32 v2, v4, v2 |
| ; VI-SDAG-NEXT: v_min_f16_e32 v2, v2, v3 |
| ; VI-SDAG-NEXT: v_max_f16_e32 v2, v5, v2 |
| ; VI-SDAG-NEXT: flat_store_short v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v6, 1, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v2, s4 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v3, s5 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v2, vcc, v2, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, s6 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, s7 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v4, vcc, v4, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc |
| ; VI-GISEL-NEXT: flat_load_ushort v7, v[0:1] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_ushort v2, v[2:3] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_load_ushort v3, v[4:5] glc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v6 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: v_add_f16_e32 v4, 1.0, v7 |
| ; VI-GISEL-NEXT: v_add_f16_e32 v2, 2.0, v2 |
| ; VI-GISEL-NEXT: v_add_f16_e32 v3, 4.0, v3 |
| ; VI-GISEL-NEXT: v_min_f16_e32 v5, v4, v2 |
| ; VI-GISEL-NEXT: v_max_f16_e32 v2, v4, v2 |
| ; VI-GISEL-NEXT: v_min_f16_e32 v2, v2, v3 |
| ; VI-GISEL-NEXT: v_max_f16_e32 v2, v5, v2 |
| ; VI-GISEL-NEXT: flat_store_short v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_ushort v1, v0, s[10:11] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_ushort v2, v0, s[12:13] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: global_load_ushort v3, v0, s[14:15] glc |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f16_e32 v1, 1.0, v1 |
| ; GFX9-NEXT: v_add_f16_e32 v2, 2.0, v2 |
| ; GFX9-NEXT: v_add_f16_e32 v3, 4.0, v3 |
| ; GFX9-NEXT: v_med3_f16 v1, v1, v2, v3 |
| ; GFX9-NEXT: global_store_short v0, v1, s[8:9] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-FAKE16-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; GFX11-SDAG-FAKE16: ; %bb.0: |
| ; GFX11-SDAG-FAKE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; GFX11-SDAG-FAKE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3] glc dlc |
| ; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] glc dlc |
| ; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-FAKE16-NEXT: global_load_u16 v3, v0, s[6:7] glc dlc |
| ; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v1, 1.0, v1 |
| ; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v2, 2.0, v2 |
| ; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v3, 4.0, v3 |
| ; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-FAKE16-NEXT: v_med3_f16 v1, v1, v2, v3 |
| ; GFX11-SDAG-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] |
| ; GFX11-SDAG-FAKE16-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-FAKE16-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; GFX11-GISEL-FAKE16: ; %bb.0: |
| ; GFX11-GISEL-FAKE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 1, v0 |
| ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3] glc dlc |
| ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] glc dlc |
| ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-FAKE16-NEXT: global_load_u16 v3, v0, s[6:7] glc dlc |
| ; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v1, 1.0, v1 |
| ; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v2, 2.0, v2 |
| ; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v3, 4.0, v3 |
| ; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-FAKE16-NEXT: v_med3_f16 v1, v1, v2, v3 |
| ; GFX11-GISEL-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] |
| ; GFX11-GISEL-FAKE16-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-TRUE16-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; GFX11-SDAG-TRUE16: ; %bb.0: |
| ; GFX11-SDAG-TRUE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 1, v0 |
| ; GFX11-SDAG-TRUE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-TRUE16-NEXT: global_load_u16 v0, v2, s[2:3] glc dlc |
| ; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-TRUE16-NEXT: global_load_u16 v1, v2, s[4:5] glc dlc |
| ; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-TRUE16-NEXT: global_load_u16 v3, v2, s[6:7] glc dlc |
| ; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, 1.0, v0.l |
| ; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l |
| ; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, v3.l |
| ; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) |
| ; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.h, 2.0, v0.h |
| ; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v1.l, 4.0, v1.l |
| ; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-TRUE16-NEXT: v_med3_f16 v0.l, v0.l, v0.h, v1.l |
| ; GFX11-SDAG-TRUE16-NEXT: global_store_b16 v2, v0, s[0:1] |
| ; GFX11-SDAG-TRUE16-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-TRUE16-LABEL: v_nnan_inputs_med3_f16_pat0: |
| ; GFX11-GISEL-TRUE16: ; %bb.0: |
| ; GFX11-GISEL-TRUE16-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 |
| ; GFX11-GISEL-TRUE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 1, v0 |
| ; GFX11-GISEL-TRUE16-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v0, v2, s[2:3] glc dlc |
| ; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v1, v2, s[4:5] glc dlc |
| ; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v3, v2, s[6:7] glc dlc |
| ; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, 1.0, v0.l |
| ; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.h, 2.0, v1.l |
| ; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v1.l, 4.0, v3.l |
| ; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-GISEL-TRUE16-NEXT: v_med3_f16 v0.l, v0.l, v0.h, v1.l |
| ; GFX11-GISEL-TRUE16-NEXT: global_store_b16 v2, v0, s[0:1] |
| ; GFX11-GISEL-TRUE16-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr half, ptr addrspace(1) %aptr, i32 %tid |
| %gep1 = getelementptr half, ptr addrspace(1) %bptr, i32 %tid |
| %gep2 = getelementptr half, ptr addrspace(1) %cptr, i32 %tid |
| %outgep = getelementptr half, ptr addrspace(1) %out, i32 %tid |
| %a = load volatile half, ptr addrspace(1) %gep0 |
| %b = load volatile half, ptr addrspace(1) %gep1 |
| %c = load volatile half, ptr addrspace(1) %gep2 |
| |
| %a.nnan = fadd nnan half %a, 1.0 |
| %b.nnan = fadd nnan half %b, 2.0 |
| %c.nnan = fadd nnan half %c, 4.0 |
| |
| %tmp0 = call half @llvm.minnum.f16(half %a.nnan, half %b.nnan) |
| %tmp1 = call half @llvm.maxnum.f16(half %a.nnan, half %b.nnan) |
| %tmp2 = call half @llvm.minnum.f16(half %tmp1, half %c.nnan) |
| %med3 = call half @llvm.maxnum.f16(half %tmp0, half %tmp2) |
| store half %med3, ptr addrspace(1) %outgep |
| ret void |
| } |
| |
| define amdgpu_kernel void @two_non_inline_constant(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: two_non_inline_constant: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 0.5, v2 |
| ; SI-SDAG-NEXT: v_max_f32_e32 v2, 0x41000000, v2 |
| ; SI-SDAG-NEXT: v_min_f32_e32 v2, 0x41800000, v2 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: two_non_inline_constant: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 0.5, v2 |
| ; SI-GISEL-NEXT: v_max_f32_e32 v2, 0x41000000, v2 |
| ; SI-GISEL-NEXT: v_min_f32_e32 v2, 0x41800000, v2 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: two_non_inline_constant: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 0.5, v3 |
| ; VI-SDAG-NEXT: v_max_f32_e32 v2, 0x41000000, v2 |
| ; VI-SDAG-NEXT: v_min_f32_e32 v2, 0x41800000, v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: two_non_inline_constant: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 0.5, v3 |
| ; VI-GISEL-NEXT: v_max_f32_e32 v2, 0x41000000, v2 |
| ; VI-GISEL-NEXT: v_min_f32_e32 v2, 0x41800000, v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: two_non_inline_constant: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v1, 0.5, v1 |
| ; GFX9-NEXT: v_max_f32_e32 v1, 0x41000000, v1 |
| ; GFX9-NEXT: v_min_f32_e32 v1, 0x41800000, v1 |
| ; GFX9-NEXT: global_store_dword v0, v1, s[0:1] |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: two_non_inline_constant: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-SDAG-NEXT: s_mov_b32 s2, 0x41000000 |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_add_f32_e32 v1, 0.5, v1 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_maxmin_f32 v1, v1, s2, 0x41800000 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: two_non_inline_constant: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: v_mov_b32_e32 v2, 0x41800000 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_add_f32_e32 v1, 0.5, v1 |
| ; GFX11-GISEL-NEXT: v_maxmin_f32 v1, v1, 0x41000000, v2 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v1, s[0:1] |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %out.gep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %add = fadd nnan float %a, 0.5 |
| %max = call float @llvm.maxnum.f32(float %add, float 8.0) |
| %med = call float @llvm.minnum.f32(float %max, float 16.0) |
| |
| store float %med, ptr addrspace(1) %out.gep |
| ret void |
| } |
| |
| ; FIXME: Simple stores do not work as a multiple use because they are bitcasted to integer constants. |
| define amdgpu_kernel void @one_non_inline_constant(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: one_non_inline_constant: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v3, 0x41800000 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_mov_b32 s6, -1 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v4, 0.5, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 0x41800000, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v3, v4, 1.0, v3 |
| ; SI-SDAG-NEXT: buffer_store_dword v3, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: buffer_store_dword v2, off, s[4:7], 0 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: one_non_inline_constant: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v3, 0x41800000 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: s_mov_b32 s6, -1 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v4, 0.5, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 0x41800000, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v3, v4, 1.0, v3 |
| ; SI-GISEL-NEXT: buffer_store_dword v3, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: buffer_store_dword v2, off, s[4:7], 0 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: one_non_inline_constant: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v4, 0x41800000 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 0.5, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, 1.0, v4 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v3, 0x41800000, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v3 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: one_non_inline_constant: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, 0x41800000 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 0.5, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, 1.0, v4 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v3, 0x41800000, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v3 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-LABEL: one_non_inline_constant: |
| ; GFX9: ; %bb.0: |
| ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-NEXT: v_mov_b32_e32 v2, 0x41800000 |
| ; GFX9-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: v_add_f32_e32 v3, 0.5, v1 |
| ; GFX9-NEXT: v_add_f32_e32 v1, 0x41800000, v1 |
| ; GFX9-NEXT: v_med3_f32 v2, v3, 1.0, v2 |
| ; GFX9-NEXT: global_store_dword v0, v2, s[0:1] |
| ; GFX9-NEXT: global_store_dword v[0:1], v1, off |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-NEXT: s_endpgm |
| ; |
| ; GFX11-LABEL: one_non_inline_constant: |
| ; GFX11: ; %bb.0: |
| ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-NEXT: v_add_f32_e32 v2, 0.5, v1 |
| ; GFX11-NEXT: v_add_f32_e32 v1, 0x41800000, v1 |
| ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) |
| ; GFX11-NEXT: v_med3_f32 v2, v2, 1.0, 0x41800000 |
| ; GFX11-NEXT: global_store_b32 v0, v2, s[0:1] |
| ; GFX11-NEXT: global_store_b32 v[0:1], v1, off dlc |
| ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %out.gep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %add = fadd nnan float %a, 0.5 |
| %max = call float @llvm.maxnum.f32(float %add, float 1.0) |
| %med = call float @llvm.minnum.f32(float %max, float 16.0) |
| |
| store float %med, ptr addrspace(1) %out.gep |
| |
| %extra.use = fadd float %a, 16.0 |
| store volatile float %extra.use, ptr addrspace(1) undef |
| ret void |
| } |
| |
| define amdgpu_kernel void @two_non_inline_constant_multi_use(ptr addrspace(1) %out, ptr addrspace(1) %aptr) #1 { |
| ; SI-SDAG-LABEL: two_non_inline_constant_multi_use: |
| ; SI-SDAG: ; %bb.0: |
| ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-SDAG-NEXT: s_mov_b32 s6, 0 |
| ; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-SDAG-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-SDAG-NEXT: s_mov_b32 s4, 0x41000000 |
| ; SI-SDAG-NEXT: v_mov_b32_e32 v3, 0x41800000 |
| ; SI-SDAG-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-SDAG-NEXT: s_mov_b32 s6, -1 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: v_add_f32_e32 v4, 0.5, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v5, 0x41800000, v2 |
| ; SI-SDAG-NEXT: v_add_f32_e32 v2, 0x41000000, v2 |
| ; SI-SDAG-NEXT: v_med3_f32 v3, v4, s4, v3 |
| ; SI-SDAG-NEXT: buffer_store_dword v3, v[0:1], s[0:3], 0 addr64 |
| ; SI-SDAG-NEXT: buffer_store_dword v5, off, s[4:7], 0 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: buffer_store_dword v2, off, s[4:7], 0 |
| ; SI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; SI-SDAG-NEXT: s_endpgm |
| ; |
| ; SI-GISEL-LABEL: two_non_inline_constant_multi_use: |
| ; SI-GISEL: ; %bb.0: |
| ; SI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 |
| ; SI-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s6, 0 |
| ; SI-GISEL-NEXT: s_mov_b32 s7, 0xf000 |
| ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] |
| ; SI-GISEL-NEXT: buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v3, 0x41000000 |
| ; SI-GISEL-NEXT: v_mov_b32_e32 v4, 0x41800000 |
| ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] |
| ; SI-GISEL-NEXT: s_mov_b32 s6, -1 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: v_add_f32_e32 v5, 0.5, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v6, 0x41800000, v2 |
| ; SI-GISEL-NEXT: v_add_f32_e32 v2, 0x41000000, v2 |
| ; SI-GISEL-NEXT: v_med3_f32 v3, v5, v3, v4 |
| ; SI-GISEL-NEXT: buffer_store_dword v3, v[0:1], s[0:3], 0 addr64 |
| ; SI-GISEL-NEXT: buffer_store_dword v6, off, s[4:7], 0 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: buffer_store_dword v2, off, s[4:7], 0 |
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; SI-GISEL-NEXT: s_endpgm |
| ; |
| ; VI-SDAG-LABEL: two_non_inline_constant_multi_use: |
| ; VI-SDAG: ; %bb.0: |
| ; VI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v4, 0x41800000 |
| ; VI-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s2, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-SDAG-NEXT: s_mov_b32 s2, 0x41000000 |
| ; VI-SDAG-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-SDAG-NEXT: v_add_u32_e32 v0, vcc, s0, v2 |
| ; VI-SDAG-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: v_add_f32_e32 v2, 0.5, v3 |
| ; VI-SDAG-NEXT: v_med3_f32 v2, v2, s2, v4 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v5, 0x41800000, v3 |
| ; VI-SDAG-NEXT: v_add_f32_e32 v3, 0x41000000, v3 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v5 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: flat_store_dword v[0:1], v3 |
| ; VI-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; VI-SDAG-NEXT: s_endpgm |
| ; |
| ; VI-GISEL-LABEL: two_non_inline_constant_multi_use: |
| ; VI-GISEL: ; %bb.0: |
| ; VI-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; VI-GISEL-NEXT: v_lshlrev_b32_e32 v2, 2, v0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v4, 0x41000000 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v5, 0x41800000 |
| ; VI-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s2 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s3 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: flat_load_dword v3, v[0:1] |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v0, s0 |
| ; VI-GISEL-NEXT: v_mov_b32_e32 v1, s1 |
| ; VI-GISEL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 |
| ; VI-GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: v_add_f32_e32 v2, 0.5, v3 |
| ; VI-GISEL-NEXT: v_med3_f32 v2, v2, v4, v5 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v6, 0x41800000, v3 |
| ; VI-GISEL-NEXT: v_add_f32_e32 v3, 0x41000000, v3 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v2 |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v6 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: flat_store_dword v[0:1], v3 |
| ; VI-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; VI-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX9-SDAG-LABEL: two_non_inline_constant_multi_use: |
| ; GFX9-SDAG: ; %bb.0: |
| ; GFX9-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x41800000 |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-SDAG-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-SDAG-NEXT: s_mov_b32 s2, 0x41000000 |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: v_add_f32_e32 v3, 0.5, v1 |
| ; GFX9-SDAG-NEXT: v_add_f32_e32 v4, 0x41800000, v1 |
| ; GFX9-SDAG-NEXT: v_add_f32_e32 v1, 0x41000000, v1 |
| ; GFX9-SDAG-NEXT: v_med3_f32 v2, v3, s2, v2 |
| ; GFX9-SDAG-NEXT: global_store_dword v0, v2, s[0:1] |
| ; GFX9-SDAG-NEXT: global_store_dword v[0:1], v4, off |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: global_store_dword v[0:1], v1, off |
| ; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX9-GISEL-LABEL: two_non_inline_constant_multi_use: |
| ; GFX9-GISEL: ; %bb.0: |
| ; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 |
| ; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, 0x41000000 |
| ; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, 0x41800000 |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX9-GISEL-NEXT: global_load_dword v1, v0, s[2:3] |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: v_add_f32_e32 v4, 0.5, v1 |
| ; GFX9-GISEL-NEXT: v_add_f32_e32 v5, 0x41800000, v1 |
| ; GFX9-GISEL-NEXT: v_add_f32_e32 v1, 0x41000000, v1 |
| ; GFX9-GISEL-NEXT: v_med3_f32 v2, v4, v2, v3 |
| ; GFX9-GISEL-NEXT: global_store_dword v0, v2, s[0:1] |
| ; GFX9-GISEL-NEXT: global_store_dword v[0:1], v5, off |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: global_store_dword v[0:1], v1, off |
| ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX9-GISEL-NEXT: s_endpgm |
| ; |
| ; GFX11-SDAG-LABEL: two_non_inline_constant_multi_use: |
| ; GFX11-SDAG: ; %bb.0: |
| ; GFX11-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) |
| ; GFX11-SDAG-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-SDAG-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-SDAG-NEXT: s_mov_b32 s2, 0x41000000 |
| ; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-SDAG-NEXT: v_add_f32_e32 v3, 0x41800000, v1 |
| ; GFX11-SDAG-NEXT: v_add_f32_e32 v2, 0.5, v1 |
| ; GFX11-SDAG-NEXT: v_add_f32_e32 v1, 0x41000000, v1 |
| ; GFX11-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) |
| ; GFX11-SDAG-NEXT: v_med3_f32 v2, v2, s2, 0x41800000 |
| ; GFX11-SDAG-NEXT: global_store_b32 v0, v2, s[0:1] |
| ; GFX11-SDAG-NEXT: global_store_b32 v[0:1], v3, off dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-SDAG-NEXT: global_store_b32 v[0:1], v1, off dlc |
| ; GFX11-SDAG-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-SDAG-NEXT: s_endpgm |
| ; |
| ; GFX11-GISEL-LABEL: two_non_inline_constant_multi_use: |
| ; GFX11-GISEL: ; %bb.0: |
| ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 |
| ; GFX11-GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 |
| ; GFX11-GISEL-NEXT: v_mov_b32_e32 v2, 0x41800000 |
| ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_1) |
| ; GFX11-GISEL-NEXT: v_lshlrev_b32_e32 v0, 2, v0 |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) |
| ; GFX11-GISEL-NEXT: global_load_b32 v1, v0, s[2:3] |
| ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) |
| ; GFX11-GISEL-NEXT: v_add_f32_e32 v3, 0.5, v1 |
| ; GFX11-GISEL-NEXT: v_med3_f32 v2, v3, 0x41000000, v2 |
| ; GFX11-GISEL-NEXT: v_add_f32_e32 v3, 0x41800000, v1 |
| ; GFX11-GISEL-NEXT: v_add_f32_e32 v1, 0x41000000, v1 |
| ; GFX11-GISEL-NEXT: global_store_b32 v0, v2, s[0:1] |
| ; GFX11-GISEL-NEXT: global_store_b32 v[0:1], v3, off dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-GISEL-NEXT: global_store_b32 v[0:1], v1, off dlc |
| ; GFX11-GISEL-NEXT: s_waitcnt_vscnt null, 0x0 |
| ; GFX11-GISEL-NEXT: s_endpgm |
| %tid = call i32 @llvm.amdgcn.workitem.id.x() |
| %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid |
| %out.gep = getelementptr float, ptr addrspace(1) %out, i32 %tid |
| %a = load float, ptr addrspace(1) %gep0 |
| %add = fadd nnan float %a, 0.5 |
| %max = call float @llvm.maxnum.f32(float %add, float 8.0) |
| %med = call float @llvm.minnum.f32(float %max, float 16.0) |
| |
| store float %med, ptr addrspace(1) %out.gep |
| |
| %extra.use0 = fadd float %a, 16.0 |
| store volatile float %extra.use0, ptr addrspace(1) undef |
| %extra.use1 = fadd float %a, 8.0 |
| store volatile float %extra.use1, ptr addrspace(1) undef |
| ret void |
| } |
| |
| declare i32 @llvm.amdgcn.workitem.id.x() #0 |
| declare float @llvm.fabs.f32(float) #0 |
| declare float @llvm.minnum.f32(float, float) #0 |
| declare float @llvm.maxnum.f32(float, float) #0 |
| declare double @llvm.minnum.f64(double, double) #0 |
| declare double @llvm.maxnum.f64(double, double) #0 |
| declare half @llvm.fabs.f16(half) #0 |
| declare half @llvm.minnum.f16(half, half) #0 |
| declare half @llvm.maxnum.f16(half, half) #0 |
| |
| attributes #0 = { nounwind readnone } |
| attributes #1 = { nounwind "unsafe-fp-math"="false" "no-nans-fp-math"="false" } |
| attributes #2 = { nounwind "unsafe-fp-math"="false" "no-nans-fp-math"="true" } |
| ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: |
| ; SI: {{.*}} |