; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=SIVI,SI %s
; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=SIVI,VI %s
; RUN: llc -march=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX11 %s

declare double @llvm.copysign.f64(double, double) #0
declare <2 x double> @llvm.copysign.v2f64(<2 x double>, <2 x double>) #0
declare <3 x double> @llvm.copysign.v3f64(<3 x double>, <3 x double>) #0
declare <4 x double> @llvm.copysign.v4f64(<4 x double>, <4 x double>) #0

define amdgpu_kernel void @s_test_copysign_f64(ptr addrspace(1) %out, [8 x i32], double %mag, [8 x i32], double %sign) {
; SI-LABEL: s_test_copysign_f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x13
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x1d
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    v_mov_b32_e32 v0, s3
; SI-NEXT:    v_mov_b32_e32 v1, s1
; SI-NEXT:    v_bfi_b32 v1, s0, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s2
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x74
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_brev_b32 s4, -2
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_bfi_b32 v1, s4, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x2
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x74
; GFX11-NEXT:    s_load_b64 s[4:5], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, s3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, s5, v0
; GFX11-NEXT:    v_mov_b32_e32 v0, s4
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double %mag, double %sign)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_0(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SI-LABEL: s_test_copysign_f64_0:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x13
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bitset0_b32 s5, 31
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v1, s5
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bitset0_b32 s3, 31
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_0:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_bitset0_b32 s3, 31
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double %mag, double 0.0)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_1(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SI-LABEL: s_test_copysign_f64_1:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x13
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bitset0_b32 s5, 31
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v1, s5
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bitset0_b32 s3, 31
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_1:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_bitset0_b32 s3, 31
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double %mag, double 1.0)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_10(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SI-LABEL: s_test_copysign_f64_10:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x13
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bitset0_b32 s5, 31
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v1, s5
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_10:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bitset0_b32 s3, 31
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_10:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_bitset0_b32 s3, 31
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double %mag, double 10.0)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_neg1(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SI-LABEL: s_test_copysign_f64_neg1:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x13
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bitset1_b32 s5, 31
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v1, s5
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_neg1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bitset1_b32 s3, 31
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_neg1:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_bitset1_b32 s3, 31
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double %mag, double -1.0)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_neg10(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SI-LABEL: s_test_copysign_f64_neg10:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x13
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bitset1_b32 s5, 31
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v1, s5
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_neg10:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bitset1_b32 s3, 31
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_neg10:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_bitset1_b32 s3, 31
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double %mag, double -10.0)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_f32(ptr addrspace(1) %out, [8 x i32], double %mag, [8 x i32], float %sign) {
; SI-LABEL: s_test_copysign_f64_f32:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x13
; SI-NEXT:    s_load_dword s0, s[0:1], 0x1d
; SI-NEXT:    s_brev_b32 s1, -2
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_mov_b32_e32 v0, s3
; SI-NEXT:    v_mov_b32_e32 v1, s0
; SI-NEXT:    v_bfi_b32 v1, s1, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s2
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dword s4, s[0:1], 0x74
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_brev_b32 s5, -2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    v_mov_b32_e32 v1, s4
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_bfi_b32 v1, s5, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x2
; GFX11-NEXT:    s_load_b32 s4, s[0:1], 0x74
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, s3, v0
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %sign.ext = fpext float %sign to double
  %result = call double @llvm.copysign.f64(double %mag, double %sign.ext)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_f16(ptr addrspace(1) %out, [8 x i32], double %mag, [8 x i32], half %sign) {
; SI-LABEL: s_test_copysign_f64_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[0:1], 0x1d
; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x13
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_mov_b32_e32 v1, s1
; SI-NEXT:    v_bfi_b32 v1, s2, v1, v0
; SI-NEXT:    v_mov_b32_e32 v0, s0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s4, s[0:1], 0x74
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_brev_b32 s5, -2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_lshlrev_b32_e64 v0, 16, s4
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_bfi_b32 v1, s5, v1, v0
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x2
; GFX11-NEXT:    s_load_b32 s4, s[0:1], 0x74
; GFX11-NEXT:    s_load_b64 s[2:3], s[0:1], 0x4c
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_lshlrev_b32_e64 v0, 16, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, s3, v0
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %sign.ext = fpext half %sign to double
  %result = call double @llvm.copysign.f64(double %mag, double %sign.ext)
  store double %result, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_0_mag(ptr addrspace(1) %out, double %sign) {
; SI-LABEL: s_test_copysign_f64_0_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    v_mov_b32_e32 v0, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_and_b32 s0, s3, 0x80000000
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v1, s0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_0_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v2, 0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    s_and_b32 s0, s3, 0x80000000
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v3, s0
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_0_mag:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_and_b32 s2, s3, 0x80000000
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX11-NEXT:    global_store_b64 v0, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double 0.0, double %sign)
  store double %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_1_mag(ptr addrspace(1) %out, double %sign) {
; SI-LABEL: s_test_copysign_f64_1_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    v_mov_b32_e32 v0, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_and_b32 s0, s3, 0x80000000
; SI-NEXT:    s_or_b32 s0, s0, 0x3ff00000
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v1, s0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_1_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v2, 0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    s_and_b32 s0, s3, 0x80000000
; VI-NEXT:    s_or_b32 s0, s0, 0x3ff00000
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v3, s0
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_1_mag:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_and_b32 s2, s3, 0x80000000
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    s_or_b32 s2, s2, 0x3ff00000
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX11-NEXT:    global_store_b64 v0, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double 1.0, double %sign)
  store double %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_10_mag(ptr addrspace(1) %out, double %sign) {
; SI-LABEL: s_test_copysign_f64_10_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    v_mov_b32_e32 v0, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_and_b32 s0, s3, 0x80000000
; SI-NEXT:    s_or_b32 s0, s0, 0x40240000
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v1, s0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_10_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v2, 0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    s_and_b32 s0, s3, 0x80000000
; VI-NEXT:    s_or_b32 s0, s0, 0x40240000
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v3, s0
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_10_mag:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_and_b32 s2, s3, 0x80000000
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    s_or_b32 s2, s2, 0x40240000
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX11-NEXT:    global_store_b64 v0, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double 10.0, double %sign)
  store double %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_neg1_mag(ptr addrspace(1) %out, double %sign) {
; SI-LABEL: s_test_copysign_f64_neg1_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    v_mov_b32_e32 v0, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_and_b32 s0, s3, 0x80000000
; SI-NEXT:    s_or_b32 s0, s0, 0x3ff00000
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v1, s0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_neg1_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v2, 0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    s_and_b32 s0, s3, 0x80000000
; VI-NEXT:    s_or_b32 s0, s0, 0x3ff00000
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v3, s0
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_neg1_mag:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_and_b32 s2, s3, 0x80000000
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    s_or_b32 s2, s2, 0x3ff00000
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX11-NEXT:    global_store_b64 v0, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double -1.0, double %sign)
  store double %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f64_neg10_mag(ptr addrspace(1) %out, double %sign) {
; SI-LABEL: s_test_copysign_f64_neg10_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    v_mov_b32_e32 v0, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_and_b32 s0, s3, 0x80000000
; SI-NEXT:    s_or_b32 s0, s0, 0x40240000
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v1, s0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f64_neg10_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v2, 0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    s_and_b32 s0, s3, 0x80000000
; VI-NEXT:    s_or_b32 s0, s0, 0x40240000
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v3, s0
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_f64_neg10_mag:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_and_b32 s2, s3, 0x80000000
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    s_or_b32 s2, s2, 0x40240000
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX11-NEXT:    global_store_b64 v0, v[0:1], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call double @llvm.copysign.f64(double -10.0, double %sign)
  store double %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_v2f64(ptr addrspace(1) %out, <2 x double> %mag, <2 x double> %sign) {
; SI-LABEL: s_test_copysign_v2f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0xd
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_brev_b32 s8, -2
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    v_mov_b32_e32 v0, s7
; SI-NEXT:    v_mov_b32_e32 v1, s11
; SI-NEXT:    v_bfi_b32 v3, s8, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s5
; SI-NEXT:    v_mov_b32_e32 v1, s9
; SI-NEXT:    v_bfi_b32 v1, s8, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v2, s6
; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_v2f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_brev_b32 s2, -2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s7
; VI-NEXT:    v_mov_b32_e32 v1, s11
; VI-NEXT:    v_mov_b32_e32 v2, s5
; VI-NEXT:    v_bfi_b32 v3, s2, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s9
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_bfi_b32 v1, s2, v2, v0
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_v2f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b256 s[4:11], s[0:1], 0x34
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v1, s11
; GFX11-NEXT:    v_mov_b32_e32 v2, s9
; GFX11-NEXT:    v_mov_b32_e32 v0, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT:    v_bfi_b32 v3, 0x7fffffff, s7, v1
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, s5, v2
; GFX11-NEXT:    v_mov_b32_e32 v2, s6
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call <2 x double> @llvm.copysign.v2f64(<2 x double> %mag, <2 x double> %sign)
  store <2 x double> %result, ptr addrspace(1) %out, align 16
  ret void
}

define amdgpu_kernel void @s_test_copysign_v3f64(ptr addrspace(1) %out, <3 x double> %mag, <3 x double> %sign) {
; SI-LABEL: s_test_copysign_v3f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x11
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_brev_b32 s10, -2
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    v_mov_b32_e32 v0, s7
; SI-NEXT:    v_mov_b32_e32 v1, s15
; SI-NEXT:    v_bfi_b32 v3, s10, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s5
; SI-NEXT:    v_mov_b32_e32 v1, s13
; SI-NEXT:    v_bfi_b32 v1, s10, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s9
; SI-NEXT:    v_mov_b32_e32 v2, s17
; SI-NEXT:    v_bfi_b32 v5, s10, v0, v2
; SI-NEXT:    v_mov_b32_e32 v4, s8
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v2, s6
; SI-NEXT:    buffer_store_dwordx2 v[4:5], off, s[0:3], 0 offset:16
; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_v3f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x44
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_brev_b32 s2, -2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s7
; VI-NEXT:    v_mov_b32_e32 v1, s15
; VI-NEXT:    v_mov_b32_e32 v2, s5
; VI-NEXT:    v_bfi_b32 v3, s2, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s13
; VI-NEXT:    v_bfi_b32 v1, s2, v2, v0
; VI-NEXT:    v_mov_b32_e32 v0, s9
; VI-NEXT:    v_mov_b32_e32 v2, s17
; VI-NEXT:    v_bfi_b32 v5, s2, v0, v2
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v7, s3
; VI-NEXT:    v_mov_b32_e32 v4, s8
; VI-NEXT:    v_mov_b32_e32 v6, s2
; VI-NEXT:    flat_store_dwordx2 v[6:7], v[4:5]
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_v3f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b512 s[4:19], s[0:1], 0x44
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v6, 0 :: v_dual_mov_b32 v1, s15
; GFX11-NEXT:    v_dual_mov_b32 v5, s17 :: v_dual_mov_b32 v0, s4
; GFX11-NEXT:    v_dual_mov_b32 v7, s13 :: v_dual_mov_b32 v4, s8
; GFX11-NEXT:    v_mov_b32_e32 v2, s6
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_bfi_b32 v5, 0x7fffffff, s9, v5
; GFX11-NEXT:    v_bfi_b32 v3, 0x7fffffff, s7, v1
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, s5, v7
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b64 v6, v[4:5], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v6, v[0:3], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call <3 x double> @llvm.copysign.v3f64(<3 x double> %mag, <3 x double> %sign)
  store <3 x double> %result, ptr addrspace(1) %out, align 32
  ret void
}

define amdgpu_kernel void @s_test_copysign_v4f64(ptr addrspace(1) %out, <4 x double> %mag, <4 x double> %sign) {
; SI-LABEL: s_test_copysign_v4f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_brev_b32 s12, -2
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    v_mov_b32_e32 v0, s7
; SI-NEXT:    v_mov_b32_e32 v1, s15
; SI-NEXT:    v_bfi_b32 v3, s12, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s5
; SI-NEXT:    v_mov_b32_e32 v1, s13
; SI-NEXT:    v_bfi_b32 v1, s12, v0, v1
; SI-NEXT:    v_mov_b32_e32 v0, s11
; SI-NEXT:    v_mov_b32_e32 v2, s19
; SI-NEXT:    v_bfi_b32 v7, s12, v0, v2
; SI-NEXT:    v_mov_b32_e32 v0, s9
; SI-NEXT:    v_mov_b32_e32 v2, s17
; SI-NEXT:    v_bfi_b32 v5, s12, v0, v2
; SI-NEXT:    v_mov_b32_e32 v4, s8
; SI-NEXT:    v_mov_b32_e32 v6, s10
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v2, s6
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_v4f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x44
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_brev_b32 s2, -2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s7
; VI-NEXT:    v_mov_b32_e32 v1, s15
; VI-NEXT:    v_mov_b32_e32 v2, s5
; VI-NEXT:    v_bfi_b32 v3, s2, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s13
; VI-NEXT:    v_bfi_b32 v1, s2, v2, v0
; VI-NEXT:    v_mov_b32_e32 v0, s11
; VI-NEXT:    v_mov_b32_e32 v2, s19
; VI-NEXT:    v_bfi_b32 v7, s2, v0, v2
; VI-NEXT:    v_mov_b32_e32 v0, s9
; VI-NEXT:    v_mov_b32_e32 v2, s17
; VI-NEXT:    v_bfi_b32 v5, s2, v0, v2
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v9, s3
; VI-NEXT:    v_mov_b32_e32 v4, s8
; VI-NEXT:    v_mov_b32_e32 v6, s10
; VI-NEXT:    v_mov_b32_e32 v8, s2
; VI-NEXT:    flat_store_dwordx4 v[8:9], v[4:7]
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: s_test_copysign_v4f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b512 s[4:19], s[0:1], 0x44
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v8, 0 :: v_dual_mov_b32 v1, s15
; GFX11-NEXT:    v_dual_mov_b32 v3, s19 :: v_dual_mov_b32 v2, s10
; GFX11-NEXT:    v_dual_mov_b32 v9, s17 :: v_dual_mov_b32 v4, s4
; GFX11-NEXT:    v_dual_mov_b32 v5, s13 :: v_dual_mov_b32 v0, s8
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_bfi_b32 v7, 0x7fffffff, s7, v1
; GFX11-NEXT:    v_bfi_b32 v3, 0x7fffffff, s11, v3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, s9, v9
; GFX11-NEXT:    v_mov_b32_e32 v6, s6
; GFX11-NEXT:    v_bfi_b32 v5, 0x7fffffff, s5, v5
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[0:1]
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
  %result = call <4 x double> @llvm.copysign.v4f64(<4 x double> %mag, <4 x double> %sign)
  store <4 x double> %result, ptr addrspace(1) %out, align 32
  ret void
}

define double @v_test_copysign_f64(ptr addrspace(1) %out, [8 x i32], double %mag, [8 x i32], double %sign) {
; SIVI-LABEL: v_test_copysign_f64:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    s_brev_b32 s4, -2
; SIVI-NEXT:    v_mov_b32_e32 v0, v10
; SIVI-NEXT:    v_bfi_b32 v1, s4, v11, v21
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, v10
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, v11, v21
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call double @llvm.copysign.f64(double %mag, double %sign)
  ret double %result
}

define double @v_test_copysign_f64_0(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SIVI-LABEL: v_test_copysign_f64_0:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    v_mov_b32_e32 v0, v10
; SIVI-NEXT:    v_and_b32_e32 v1, 0x7fffffff, v11
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64_0:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, v10 :: v_dual_and_b32 v1, 0x7fffffff, v11
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call double @llvm.copysign.f64(double %mag, double 0.0)
  ret double %result
}

define double @v_test_copysign_f64_1(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SIVI-LABEL: v_test_copysign_f64_1:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    v_mov_b32_e32 v0, v10
; SIVI-NEXT:    v_and_b32_e32 v1, 0x7fffffff, v11
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64_1:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, v10 :: v_dual_and_b32 v1, 0x7fffffff, v11
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call double @llvm.copysign.f64(double %mag, double 1.0)
  ret double %result
}

define double @v_test_copysign_f64_10(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SIVI-LABEL: v_test_copysign_f64_10:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    v_mov_b32_e32 v0, v10
; SIVI-NEXT:    v_and_b32_e32 v1, 0x7fffffff, v11
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64_10:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, v10 :: v_dual_and_b32 v1, 0x7fffffff, v11
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call double @llvm.copysign.f64(double %mag, double 10.0)
  ret double %result
}

define double @v_test_copysign_f64_neg1(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SIVI-LABEL: v_test_copysign_f64_neg1:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    v_mov_b32_e32 v0, v10
; SIVI-NEXT:    v_or_b32_e32 v1, 0x80000000, v11
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64_neg1:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, v10
; GFX11-NEXT:    v_or_b32_e32 v1, 0x80000000, v11
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call double @llvm.copysign.f64(double %mag, double -1.0)
  ret double %result
}

define double @v_test_copysign_f64_neg10(ptr addrspace(1) %out, [8 x i32], double %mag) {
; SIVI-LABEL: v_test_copysign_f64_neg10:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    v_mov_b32_e32 v0, v10
; SIVI-NEXT:    v_or_b32_e32 v1, 0x80000000, v11
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64_neg10:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, v10
; GFX11-NEXT:    v_or_b32_e32 v1, 0x80000000, v11
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call double @llvm.copysign.f64(double %mag, double -10.0)
  ret double %result
}

define double @v_test_copysign_f64_f32(ptr addrspace(1) %out, [8 x i32], double %mag, [8 x i32], float %sign) {
; SIVI-LABEL: v_test_copysign_f64_f32:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    s_brev_b32 s4, -2
; SIVI-NEXT:    v_mov_b32_e32 v0, v10
; SIVI-NEXT:    v_bfi_b32 v1, s4, v11, v20
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, v10
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, v11, v20
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %sign.ext = fpext float %sign to double
  %result = call double @llvm.copysign.f64(double %mag, double %sign.ext)
  ret double %result
}

define double @v_test_copysign_f64_f16(ptr addrspace(1) %out, [8 x i32], double %mag, [8 x i32], half %sign) {
; SI-LABEL: v_test_copysign_f64_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    s_brev_b32 s4, -2
; SI-NEXT:    v_mov_b32_e32 v0, v10
; SI-NEXT:    v_bfi_b32 v1, s4, v11, v20
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: v_test_copysign_f64_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_lshlrev_b32_e32 v1, 16, v20
; VI-NEXT:    s_brev_b32 s4, -2
; VI-NEXT:    v_mov_b32_e32 v0, v10
; VI-NEXT:    v_bfi_b32 v1, s4, v11, v1
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_f64_f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, v10 :: v_dual_lshlrev_b32 v1, 16, v20
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, v11, v1
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %sign.ext = fpext half %sign to double
  %result = call double @llvm.copysign.f64(double %mag, double %sign.ext)
  ret double %result
}

define <2 x double> @v_test_copysign_v2f64(ptr addrspace(1) %out, <2 x double> %mag, <2 x double> %sign) {
; SIVI-LABEL: v_test_copysign_v2f64:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    s_brev_b32 s4, -2
; SIVI-NEXT:    v_mov_b32_e32 v0, v2
; SIVI-NEXT:    v_bfi_b32 v1, s4, v3, v7
; SIVI-NEXT:    v_bfi_b32 v3, s4, v5, v9
; SIVI-NEXT:    v_mov_b32_e32 v2, v4
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_v2f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, v2
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, v3, v7
; GFX11-NEXT:    v_bfi_b32 v3, 0x7fffffff, v5, v9
; GFX11-NEXT:    v_mov_b32_e32 v2, v4
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call <2 x double> @llvm.copysign.v2f64(<2 x double> %mag, <2 x double> %sign)
  ret <2 x double> %result
}

define <3 x double> @v_test_copysign_v3f64(ptr addrspace(1) %out, <3 x double> %mag, <3 x double> %sign) {
; SIVI-LABEL: v_test_copysign_v3f64:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    s_brev_b32 s4, -2
; SIVI-NEXT:    v_mov_b32_e32 v0, v2
; SIVI-NEXT:    v_bfi_b32 v1, s4, v3, v9
; SIVI-NEXT:    v_bfi_b32 v3, s4, v5, v11
; SIVI-NEXT:    v_bfi_b32 v5, s4, v7, v13
; SIVI-NEXT:    v_mov_b32_e32 v2, v4
; SIVI-NEXT:    v_mov_b32_e32 v4, v6
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_v3f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, v2
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, v3, v9
; GFX11-NEXT:    v_bfi_b32 v3, 0x7fffffff, v5, v11
; GFX11-NEXT:    v_bfi_b32 v5, 0x7fffffff, v7, v13
; GFX11-NEXT:    v_mov_b32_e32 v2, v4
; GFX11-NEXT:    v_mov_b32_e32 v4, v6
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call <3 x double> @llvm.copysign.v3f64(<3 x double> %mag, <3 x double> %sign)
  ret <3 x double> %result
}

define <4 x double> @v_test_copysign_v4f64(ptr addrspace(1) %out, <4 x double> %mag, <4 x double> %sign) {
; SIVI-LABEL: v_test_copysign_v4f64:
; SIVI:       ; %bb.0:
; SIVI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SIVI-NEXT:    s_brev_b32 s4, -2
; SIVI-NEXT:    v_mov_b32_e32 v0, v2
; SIVI-NEXT:    v_bfi_b32 v1, s4, v3, v11
; SIVI-NEXT:    v_bfi_b32 v3, s4, v5, v13
; SIVI-NEXT:    v_bfi_b32 v5, s4, v7, v15
; SIVI-NEXT:    v_bfi_b32 v7, s4, v9, v17
; SIVI-NEXT:    v_mov_b32_e32 v2, v4
; SIVI-NEXT:    v_mov_b32_e32 v4, v6
; SIVI-NEXT:    v_mov_b32_e32 v6, v8
; SIVI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_test_copysign_v4f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, v2
; GFX11-NEXT:    v_bfi_b32 v1, 0x7fffffff, v3, v11
; GFX11-NEXT:    v_bfi_b32 v3, 0x7fffffff, v5, v13
; GFX11-NEXT:    v_bfi_b32 v5, 0x7fffffff, v7, v15
; GFX11-NEXT:    v_bfi_b32 v7, 0x7fffffff, v9, v17
; GFX11-NEXT:    v_mov_b32_e32 v2, v4
; GFX11-NEXT:    v_mov_b32_e32 v4, v6
; GFX11-NEXT:    v_mov_b32_e32 v6, v8
; GFX11-NEXT:    s_setpc_b64 s[30:31]
  %result = call <4 x double> @llvm.copysign.v4f64(<4 x double> %mag, <4 x double> %sign)
  ret <4 x double> %result
}

attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }