llvm/test/CodeGen/AMDGPU/promote-alloca-vector-gep-of-gep.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=amdgpu-promote-alloca < %s | FileCheck %s
define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_of_gep(i32 %idx, ptr addrspace(1) %output) #0 {
; CHECK-LABEL: define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_of_gep(
; CHECK-SAME: i32 [[IDX:%.*]], ptr addrspace(1) [[OUTPUT:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[BUF:%.*]] = freeze <20 x i32> poison
; CHECK-NEXT:    [[TMP0:%.*]] = mul i32 [[IDX]], 2
; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <20 x i32> [[BUF]], i32 1, i32 [[TMP0]]
; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[TMP0]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <20 x i32> [[TMP1]], i32 2, i32 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = mul i32 [[IDX]], 2
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 1, [[TMP4]]
; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <20 x i32> [[TMP3]], i32 [[TMP5]]
; CHECK-NEXT:    store i32 [[TMP6]], ptr addrspace(1) [[OUTPUT]], align 4
; CHECK-NEXT:    ret void
;
entry:
  %alloca = alloca [10 x <2 x i32>], align 8, addrspace(5)
  %row = getelementptr [10 x <2 x i32>], ptr addrspace(5) %alloca, i32 0, i32 %idx
  store <2 x i32> <i32 1, i32 2>, ptr addrspace(5) %row, align 8
  %elt1 = getelementptr i8, ptr addrspace(5) %row, i32 4
  %val = load i32, ptr addrspace(5) %elt1, align 4
  store i32 %val, ptr addrspace(1) %output
  ret void
}

define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_of_gep3(i32 %idx, ptr addrspace(1) %output) #0 {
; CHECK-LABEL: define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_of_gep3(
; CHECK-SAME: i32 [[IDX:%.*]], ptr addrspace(1) [[OUTPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[ALLOCA:%.*]] = freeze <16 x i32> poison
; CHECK-NEXT:    [[TMP0:%.*]] = mul i32 [[IDX]], 2
; CHECK-NEXT:    [[TMP1:%.*]] = add i32 8, [[TMP0]]
; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <16 x i32> [[ALLOCA]], i32 10, i32 [[TMP1]]
; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[TMP1]], 1
; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <16 x i32> [[TMP2]], i32 20, i32 [[TMP3]]
; CHECK-NEXT:    [[TMP5:%.*]] = mul i32 [[IDX]], 2
; CHECK-NEXT:    [[TMP6:%.*]] = add i32 9, [[TMP5]]
; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <16 x i32> [[TMP4]], i32 [[TMP6]]
; CHECK-NEXT:    store i32 [[TMP7]], ptr addrspace(1) [[OUTPUT]], align 4
; CHECK-NEXT:    ret void
;
entry:
  %alloca = alloca [2 x [4 x <2 x i32>]], align 8, addrspace(5)
  %lvl1 = getelementptr inbounds [2 x [4 x <2 x i32>]], ptr addrspace(5) %alloca, i32 0, i32 1
  %lvl2 = getelementptr inbounds [4 x <2 x i32>], ptr addrspace(5) %lvl1, i32 0, i32 %idx
  store <2 x i32> <i32 10, i32 20>, ptr addrspace(5) %lvl2, align 8
  %byte = getelementptr inbounds i8, ptr addrspace(5) %lvl2, i32 4
  %val = load i32, ptr addrspace(5) %byte, align 4
  store i32 %val, ptr addrspace(1) %output
  ret void
}

define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_twice_idx(i32 %idx, ptr addrspace(1) %out) #0 {
; CHECK-LABEL: define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_twice_idx(
; CHECK-SAME: i32 [[IDX:%.*]], ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[BUF:%.*]] = freeze <20 x i32> poison
; CHECK-NEXT:    [[TMP0:%.*]] = mul i32 [[IDX]], 2
; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <20 x i32> [[BUF]], i32 1, i32 [[TMP0]]
; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[TMP0]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <20 x i32> [[TMP4]], i32 2, i32 [[TMP5]]
; CHECK-NEXT:    [[TMP1:%.*]] = mul i32 [[IDX]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <20 x i32> [[TMP3]], i32 [[TMP1]]
; CHECK-NEXT:    store i32 [[TMP2]], ptr addrspace(1) [[OUT]], align 4
; CHECK-NEXT:    ret void
;
entry:
  %alloca = alloca [10 x [2 x i32]], align 8, addrspace(5)
  %row = getelementptr inbounds [10 x [2 x i32]], ptr addrspace(5) %alloca, i32 0, i32 %idx
  store <2 x i32> <i32 1, i32 2>, ptr addrspace(5) %row, align 8
  %elt = getelementptr inbounds [2 x i32], ptr addrspace(5) %row, i32 0, i32 %idx
  %val = load i32, ptr addrspace(5) %elt, align 4
  store i32 %val, ptr addrspace(1) %out
  ret void
}

attributes #0 = { "amdgpu-promote-alloca-to-vector-max-regs"="32" }