; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=infer-address-spaces %s | FileCheck %s
; Ports of most of test/CodeGen/NVPTX/access-non-generic.ll

@scalar = internal addrspace(3) global float 0.0, align 4
@array = internal addrspace(3) global [10 x float] zeroinitializer, align 4

define amdgpu_kernel void @load_store_lds_f32(i32 %i, float %v) #0 {
; CHECK-LABEL: define amdgpu_kernel void @load_store_lds_f32(
; CHECK-SAME: i32 [[I:%.*]], float [[V:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP:%.*]] = load float, ptr addrspace(3) @scalar, align 4
; CHECK-NEXT:    call void @use(float [[TMP]])
; CHECK-NEXT:    store float [[V]], ptr addrspace(3) @scalar, align 4
; CHECK-NEXT:    call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT:    [[TMP2:%.*]] = load float, ptr addrspace(3) @scalar, align 4
; CHECK-NEXT:    call void @use(float [[TMP2]])
; CHECK-NEXT:    store float [[V]], ptr addrspace(3) @scalar, align 4
; CHECK-NEXT:    call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT:    [[TMP3:%.*]] = load float, ptr addrspace(3) getelementptr inbounds ([10 x float], ptr addrspace(3) @array, i32 0, i32 5), align 4
; CHECK-NEXT:    call void @use(float [[TMP3]])
; CHECK-NEXT:    store float [[V]], ptr addrspace(3) getelementptr inbounds ([10 x float], ptr addrspace(3) @array, i32 0, i32 5), align 4
; CHECK-NEXT:    call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [10 x float], ptr addrspace(3) @array, i32 0, i32 5
; CHECK-NEXT:    [[TMP5:%.*]] = load float, ptr addrspace(3) [[TMP4]], align 4
; CHECK-NEXT:    call void @use(float [[TMP5]])
; CHECK-NEXT:    store float [[V]], ptr addrspace(3) [[TMP4]], align 4
; CHECK-NEXT:    call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [10 x float], ptr addrspace(3) @array, i32 0, i32 [[I]]
; CHECK-NEXT:    [[TMP8:%.*]] = load float, ptr addrspace(3) [[TMP7]], align 4
; CHECK-NEXT:    call void @use(float [[TMP8]])
; CHECK-NEXT:    store float [[V]], ptr addrspace(3) [[TMP7]], align 4
; CHECK-NEXT:    call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT:    ret void
;
bb:
  %tmp = load float, ptr addrspacecast (ptr addrspace(3) @scalar to ptr), align 4
  call void @use(float %tmp)
  store float %v, ptr addrspacecast (ptr addrspace(3) @scalar to ptr), align 4
  call void @llvm.amdgcn.s.barrier()
  %tmp1 = addrspacecast ptr addrspace(3) @scalar to ptr
  %tmp2 = load float, ptr %tmp1, align 4
  call void @use(float %tmp2)
  store float %v, ptr %tmp1, align 4
  call void @llvm.amdgcn.s.barrier()
  %tmp3 = load float, ptr getelementptr inbounds ([10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i32 0, i32 5), align 4
  call void @use(float %tmp3)
  store float %v, ptr getelementptr inbounds ([10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i32 0, i32 5), align 4
  call void @llvm.amdgcn.s.barrier()
  %tmp4 = getelementptr inbounds [10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i32 0, i32 5
  %tmp5 = load float, ptr %tmp4, align 4
  call void @use(float %tmp5)
  store float %v, ptr %tmp4, align 4
  call void @llvm.amdgcn.s.barrier()
  %tmp6 = addrspacecast ptr addrspace(3) @array to ptr
  %tmp7 = getelementptr inbounds [10 x float], ptr %tmp6, i32 0, i32 %i
  %tmp8 = load float, ptr %tmp7, align 4
  call void @use(float %tmp8)
  store float %v, ptr %tmp7, align 4
  call void @llvm.amdgcn.s.barrier()
  ret void
}

define i32 @constexpr_load_int_from_float_lds() #0 {
; CHECK-LABEL: define i32 @constexpr_load_int_from_float_lds(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP:%.*]] = load i32, ptr addrspace(3) @scalar, align 4
; CHECK-NEXT:    ret i32 [[TMP]]
;
bb:
  %tmp = load i32, ptr addrspacecast (ptr addrspace(3) @scalar to ptr), align 4
  ret i32 %tmp
}

define i32 @load_int_from_global_float(ptr addrspace(1) %input, i32 %i, i32 %j) #0 {
; CHECK-LABEL: define i32 @load_int_from_global_float(
; CHECK-SAME: ptr addrspace(1) [[INPUT:%.*]], i32 [[I:%.*]], i32 [[J:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr float, ptr addrspace(1) [[INPUT]], i32 [[I]]
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr float, ptr addrspace(1) [[TMP1]], i32 [[J]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[TMP2]], align 4
; CHECK-NEXT:    ret i32 [[TMP4]]
;
bb:
  %tmp = addrspacecast ptr addrspace(1) %input to ptr
  %tmp1 = getelementptr float, ptr %tmp, i32 %i
  %tmp2 = getelementptr float, ptr %tmp1, i32 %j
  %tmp4 = load i32, ptr %tmp2
  ret i32 %tmp4
}

define amdgpu_kernel void @nested_const_expr() #0 {
; CHECK-LABEL: define amdgpu_kernel void @nested_const_expr(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT:    store i32 1, ptr addrspace(3) getelementptr ([10 x float], ptr addrspace(3) @array, i64 0, i64 1), align 4
; CHECK-NEXT:    ret void
;
  store i32 1, ptr bitcast (ptr getelementptr ([10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i64 0, i64 1) to ptr), align 4

  ret void
}

define amdgpu_kernel void @rauw(ptr addrspace(1) %input) #0 {
; CHECK-LABEL: define amdgpu_kernel void @rauw(
; CHECK-SAME: ptr addrspace(1) [[INPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[ADDR:%.*]] = getelementptr float, ptr addrspace(1) [[INPUT]], i64 10
; CHECK-NEXT:    [[V:%.*]] = load float, ptr addrspace(1) [[ADDR]], align 4
; CHECK-NEXT:    store float [[V]], ptr addrspace(1) [[ADDR]], align 4
; CHECK-NEXT:    ret void
;
bb:
  %generic_input = addrspacecast ptr addrspace(1) %input to ptr
  %addr = getelementptr float, ptr %generic_input, i64 10
  %v = load float, ptr %addr
  store float %v, ptr %addr
  ret void
}

; FIXME: Should be able to eliminate the cast inside the loop
define amdgpu_kernel void @loop() #0 {
; CHECK-LABEL: define amdgpu_kernel void @loop(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    [[END:%.*]] = getelementptr float, ptr addrspace(3) @array, i64 10
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[I:%.*]] = phi ptr addrspace(3) [ @array, %[[ENTRY]] ], [ [[I2:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[V:%.*]] = load float, ptr addrspace(3) [[I]], align 4
; CHECK-NEXT:    call void @use(float [[V]])
; CHECK-NEXT:    [[I2]] = getelementptr float, ptr addrspace(3) [[I]], i64 1
; CHECK-NEXT:    [[EXIT_COND:%.*]] = icmp eq ptr addrspace(3) [[I2]], [[END]]
; CHECK-NEXT:    br i1 [[EXIT_COND]], label %[[EXIT:.*]], label %[[LOOP]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
entry:
  %p = addrspacecast ptr addrspace(3) @array to ptr
  %end = getelementptr float, ptr %p, i64 10
  br label %loop

loop:                                             ; preds = %loop, %entry
  %i = phi ptr [ %p, %entry ], [ %i2, %loop ]
  %v = load float, ptr %i
  call void @use(float %v)
  %i2 = getelementptr float, ptr %i, i64 1
  %exit_cond = icmp eq ptr %i2, %end
  br i1 %exit_cond, label %exit, label %loop

exit:                                             ; preds = %loop
  ret void
}

@generic_end = external addrspace(1) global ptr

define amdgpu_kernel void @loop_with_generic_bound() #0 {
; CHECK-LABEL: define amdgpu_kernel void @loop_with_generic_bound(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*]]:
; CHECK-NEXT:    [[END:%.*]] = load ptr, ptr addrspace(1) @generic_end, align 8
; CHECK-NEXT:    br label %[[LOOP:.*]]
; CHECK:       [[LOOP]]:
; CHECK-NEXT:    [[I:%.*]] = phi ptr addrspace(3) [ @array, %[[ENTRY]] ], [ [[I2:%.*]], %[[LOOP]] ]
; CHECK-NEXT:    [[V:%.*]] = load float, ptr addrspace(3) [[I]], align 4
; CHECK-NEXT:    call void @use(float [[V]])
; CHECK-NEXT:    [[I2]] = getelementptr float, ptr addrspace(3) [[I]], i64 1
; CHECK-NEXT:    [[TMP0:%.*]] = addrspacecast ptr addrspace(3) [[I2]] to ptr
; CHECK-NEXT:    [[EXIT_COND:%.*]] = icmp eq ptr [[TMP0]], [[END]]
; CHECK-NEXT:    br i1 [[EXIT_COND]], label %[[EXIT:.*]], label %[[LOOP]]
; CHECK:       [[EXIT]]:
; CHECK-NEXT:    ret void
;
entry:
  %p = addrspacecast ptr addrspace(3) @array to ptr
  %end = load ptr, ptr addrspace(1) @generic_end
  br label %loop

loop:                                             ; preds = %loop, %entry
  %i = phi ptr [ %p, %entry ], [ %i2, %loop ]
  %v = load float, ptr %i
  call void @use(float %v)
  %i2 = getelementptr float, ptr %i, i64 1
  %exit_cond = icmp eq ptr %i2, %end
  br i1 %exit_cond, label %exit, label %loop

exit:                                             ; preds = %loop
  ret void
}

define void @select_bug() #0 {
; CHECK-LABEL: define void @select_bug(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT:    [[CMP:%.*]] = icmp ne ptr inttoptr (i64 4873 to ptr), null
; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[CMP]], i64 73, i64 93
; CHECK-NEXT:    [[ADD_PTR157:%.*]] = getelementptr inbounds i64, ptr undef, i64 [[SEL]]
; CHECK-NEXT:    [[CMP169:%.*]] = icmp uge ptr undef, [[ADD_PTR157]]
; CHECK-NEXT:    unreachable
;
  %cmp = icmp ne ptr inttoptr (i64 4873 to ptr), null
  %sel = select i1 %cmp, i64 73, i64 93
  %add.ptr157 = getelementptr inbounds i64, ptr undef, i64 %sel
  %cmp169 = icmp uge ptr undef, %add.ptr157
  unreachable
}

declare void @llvm.amdgcn.s.barrier() #1
declare void @use(float) #0

attributes #0 = { nounwind }
attributes #1 = { convergent nounwind }