Codebase Browser
llvm
Go to App

llvm/llvm/test/Transforms/SLPVectorizer/AMDGPU/address-space-ptr-sze-gep-index-assert.ll 

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=slp-vectorizer -slp-threshold=-18 < %s | FileCheck %s

; Make sure there's no SCEV assert when the indexes are for different
; sized address spaces

target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5"

define void @slp_scev_assert(i32 %idx, i64 %tmp3) #0 {
; CHECK-LABEL: @slp_scev_assert(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[TMP:%.*]] = addrspacecast ptr addrspace(5) undef to ptr
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i8, ptr addrspace(5) undef, i32 [[IDX:%.*]]
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[TMP]], i64 [[TMP3:%.*]]
; CHECK-NEXT:    store i8 0, ptr addrspace(5) [[TMP2]], align 1
; CHECK-NEXT:    store i8 0, ptr [[TMP4]], align 1
; CHECK-NEXT:    ret void
;
bb:
  %tmp = addrspacecast ptr addrspace(5) undef to ptr
  %tmp2 = getelementptr inbounds i8, ptr addrspace(5) undef, i32 %idx
  %tmp4 = getelementptr inbounds i8, ptr %tmp, i64 %tmp3
  store i8 0, ptr addrspace(5) %tmp2
  store i8 0, ptr %tmp4
  ret void
}

define void @multi_as_reduction_different_sized(ptr addrspace(3) %lds, i32 %idx0, i64 %idx1) #0 {
; CHECK-LABEL: @multi_as_reduction_different_sized(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[FLAT:%.*]] = addrspacecast ptr addrspace(3) [[LDS:%.*]] to ptr
; CHECK-NEXT:    [[ADD0:%.*]] = add i32 [[IDX0:%.*]], 2
; CHECK-NEXT:    [[ADD1:%.*]] = add i64 [[IDX1:%.*]], 1
; CHECK-NEXT:    [[LDS_1:%.*]] = getelementptr inbounds i32, ptr addrspace(3) [[LDS]], i32 [[ADD0]]
; CHECK-NEXT:    [[FLAT_1:%.*]] = getelementptr inbounds i32, ptr [[FLAT]], i64 [[ADD1]]
; CHECK-NEXT:    [[LOAD_LDS_0:%.*]] = load i32, ptr addrspace(3) [[LDS]], align 4
; CHECK-NEXT:    [[LOAD_LDS_1:%.*]] = load i32, ptr addrspace(3) [[LDS_1]], align 4
; CHECK-NEXT:    [[LOAD_FLAT_0:%.*]] = load i32, ptr [[FLAT]], align 4
; CHECK-NEXT:    [[LOAD_FLAT_1:%.*]] = load i32, ptr [[FLAT_1]], align 4
; CHECK-NEXT:    [[SUB0:%.*]] = sub i32 [[LOAD_FLAT_0]], [[LOAD_LDS_0]]
; CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[LOAD_FLAT_1]], [[LOAD_LDS_1]]
; CHECK-NEXT:    store i32 [[SUB0]], ptr undef, align 4
; CHECK-NEXT:    store i32 [[SUB1]], ptr undef, align 4
; CHECK-NEXT:    ret void
;
bb:
  %flat = addrspacecast ptr addrspace(3) %lds to ptr
  %add0 = add i32 %idx0, 2
  %add1 = add i64 %idx1, 1

  %lds.1 = getelementptr inbounds i32, ptr addrspace(3) %lds, i32 %add0
  %flat.1 = getelementptr inbounds i32, ptr %flat, i64 %add1

  %load.lds.0 = load i32, ptr addrspace(3) %lds, align 4
  %load.lds.1 = load i32, ptr addrspace(3) %lds.1, align 4

  %load.flat.0 = load i32, ptr %flat, align 4
  %load.flat.1 = load i32, ptr %flat.1, align 4

  %sub0 = sub i32 %load.flat.0, %load.lds.0
  %sub1 = sub i32 %load.flat.1, %load.lds.1

  store i32 %sub0, ptr undef
  store i32 %sub1, ptr undef
  ret void
}

; This should vectorize if using getUnderlyingObject
define void @multi_as_reduction_same_size(ptr addrspace(1) %global, i64 %idx0, i64 %idx1) #0 {
; CHECK-LABEL: @multi_as_reduction_same_size(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[FLAT:%.*]] = addrspacecast ptr addrspace(1) [[GLOBAL:%.*]] to ptr
; CHECK-NEXT:    [[ADD0:%.*]] = add i64 [[IDX0:%.*]], 2
; CHECK-NEXT:    [[ADD1:%.*]] = add i64 [[IDX1:%.*]], 1
; CHECK-NEXT:    [[GLOBAL_1:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[GLOBAL]], i64 [[ADD0]]
; CHECK-NEXT:    [[FLAT_1:%.*]] = getelementptr inbounds i32, ptr [[FLAT]], i64 [[ADD1]]
; CHECK-NEXT:    [[LOAD_GLOBAL_0:%.*]] = load i32, ptr addrspace(1) [[GLOBAL]], align 4
; CHECK-NEXT:    [[LOAD_GLOBAL_1:%.*]] = load i32, ptr addrspace(1) [[GLOBAL_1]], align 4
; CHECK-NEXT:    [[LOAD_FLAT_0:%.*]] = load i32, ptr [[FLAT]], align 4
; CHECK-NEXT:    [[LOAD_FLAT_1:%.*]] = load i32, ptr [[FLAT_1]], align 4
; CHECK-NEXT:    [[SUB0:%.*]] = sub i32 [[LOAD_FLAT_0]], [[LOAD_GLOBAL_0]]
; CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[LOAD_FLAT_1]], [[LOAD_GLOBAL_1]]
; CHECK-NEXT:    store i32 [[SUB0]], ptr undef, align 4
; CHECK-NEXT:    store i32 [[SUB1]], ptr undef, align 4
; CHECK-NEXT:    ret void
;
bb:
  %flat = addrspacecast ptr addrspace(1) %global to ptr
  %add0 = add i64 %idx0, 2
  %add1 = add i64 %idx1, 1

  %global.1 = getelementptr inbounds i32, ptr addrspace(1) %global, i64 %add0
  %flat.1 = getelementptr inbounds i32, ptr %flat, i64 %add1

  %load.global.0 = load i32, ptr addrspace(1) %global, align 4
  %load.global.1 = load i32, ptr addrspace(1) %global.1, align 4

  %load.flat.0 = load i32, ptr %flat, align 4
  %load.flat.1 = load i32, ptr %flat.1, align 4

  %sub0 = sub i32 %load.flat.0, %load.global.0
  %sub1 = sub i32 %load.flat.1, %load.global.1

  store i32 %sub0, ptr undef
  store i32 %sub1, ptr undef
  ret void
}

; This should vectorize if using getUnderlyingObject
; The add is done in the same width, even though the address space size is smaller
define void @multi_as_reduction_different_sized_noncanon(ptr addrspace(3) %lds, i64 %idx0, i64 %idx1) #0 {
; CHECK-LABEL: @multi_as_reduction_different_sized_noncanon(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    [[FLAT:%.*]] = addrspacecast ptr addrspace(3) [[LDS:%.*]] to ptr
; CHECK-NEXT:    [[ADD0:%.*]] = add i64 [[IDX0:%.*]], 2
; CHECK-NEXT:    [[ADD1:%.*]] = add i64 [[IDX1:%.*]], 1
; CHECK-NEXT:    [[LDS_1:%.*]] = getelementptr inbounds i32, ptr addrspace(3) [[LDS]], i64 [[ADD0]]
; CHECK-NEXT:    [[FLAT_1:%.*]] = getelementptr inbounds i32, ptr [[FLAT]], i64 [[ADD1]]
; CHECK-NEXT:    [[LOAD_LDS_0:%.*]] = load i32, ptr addrspace(3) [[LDS]], align 4
; CHECK-NEXT:    [[LOAD_LDS_1:%.*]] = load i32, ptr addrspace(3) [[LDS_1]], align 4
; CHECK-NEXT:    [[LOAD_FLAT_0:%.*]] = load i32, ptr [[FLAT]], align 4
; CHECK-NEXT:    [[LOAD_FLAT_1:%.*]] = load i32, ptr [[FLAT_1]], align 4
; CHECK-NEXT:    [[SUB0:%.*]] = sub i32 [[LOAD_FLAT_0]], [[LOAD_LDS_0]]
; CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[LOAD_FLAT_1]], [[LOAD_LDS_1]]
; CHECK-NEXT:    store i32 [[SUB0]], ptr undef, align 4
; CHECK-NEXT:    store i32 [[SUB1]], ptr undef, align 4
; CHECK-NEXT:    ret void
;
bb:
  %flat = addrspacecast ptr addrspace(3) %lds to ptr
  %add0 = add i64 %idx0, 2
  %add1 = add i64 %idx1, 1

  %lds.1 = getelementptr inbounds i32, ptr addrspace(3) %lds, i64 %add0
  %flat.1 = getelementptr inbounds i32, ptr %flat, i64 %add1

  %load.lds.0 = load i32, ptr addrspace(3) %lds, align 4
  %load.lds.1 = load i32, ptr addrspace(3) %lds.1, align 4

  %load.flat.0 = load i32, ptr %flat, align 4
  %load.flat.1 = load i32, ptr %flat.1, align 4

  %sub0 = sub i32 %load.flat.0, %load.lds.0
  %sub1 = sub i32 %load.flat.1, %load.lds.1

  store i32 %sub0, ptr undef
  store i32 %sub1, ptr undef
  ret void
}

define void @slp_crash_on_addrspacecast() {
; CHECK-LABEL: @slp_crash_on_addrspacecast(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i64, ptr addrspace(3) undef, i32 undef
; CHECK-NEXT:    [[P0:%.*]] = addrspacecast ptr addrspace(3) [[TMP0]] to ptr
; CHECK-NEXT:    store i64 undef, ptr [[P0]], align 8
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i64, ptr addrspace(3) undef, i32 undef
; CHECK-NEXT:    [[P1:%.*]] = addrspacecast ptr addrspace(3) [[TMP1]] to ptr
; CHECK-NEXT:    store i64 undef, ptr [[P1]], align 8
; CHECK-NEXT:    ret void
;
entry:
  %0 = getelementptr inbounds i64, ptr addrspace(3) undef, i32 undef
  %p0 = addrspacecast ptr addrspace(3) %0 to ptr
  store i64 undef, ptr %p0, align 8
  %1 = getelementptr inbounds i64, ptr addrspace(3) undef, i32 undef
  %p1 = addrspacecast ptr addrspace(3) %1 to ptr
  store i64 undef, ptr %p1, align 8
  ret void
}