llvm/test/Transforms/DeadStoreElimination/dead-vp.store.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=dse -S < %s | FileCheck %s

; Test predicated vector length masked stores for elimination

define void @test1(ptr %a, i32 %vl, <vscale x 8 x i32> %v1, <vscale x 8 x i32> %v2) {
;
; CHECK-LABEL: @test1(
; CHECK-NEXT:    [[VP_OP:%.*]] = call <vscale x 8 x i32> @llvm.vp.add.nxv8i32(<vscale x 8 x i32> [[V1:%.*]], <vscale x 8 x i32> [[V2:%.*]], <vscale x 8 x i1> splat (i1 true), i32 [[VL:%.*]])
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[VP_OP]], ptr nonnull [[A:%.*]], <vscale x 8 x i1> splat (i1 true), i32 [[VL]])
; CHECK-NEXT:    ret void
;
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v1, ptr nonnull %a, <vscale x 8 x i1> splat (i1 true), i32 %vl)
  %vp.op = call <vscale x 8 x i32> @llvm.vp.add.nxv8i32(<vscale x 8 x i32> %v1, <vscale x 8 x i32> %v2, <vscale x 8 x i1> splat (i1 true), i32 %vl)
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %vp.op, ptr nonnull %a, <vscale x 8 x i1> splat (i1 true), i32 %vl)
  ret void
}

; False test for different vector lengths

define void @test2(ptr %a, i32 %vl1, i32 %vl2, <vscale x 8 x i32> %v1, <vscale x 8 x i32> %v2) {
;
; CHECK-LABEL: @test2(
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V1:%.*]], ptr nonnull [[A:%.*]], <vscale x 8 x i1> splat (i1 true), i32 [[VL1:%.*]])
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V2:%.*]], ptr nonnull [[A]], <vscale x 8 x i1> splat (i1 true), i32 [[VL2:%.*]])
; CHECK-NEXT:    ret void
;
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v1, ptr nonnull %a, <vscale x 8 x i1> splat (i1 true), i32 %vl1)
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v2, ptr nonnull %a, <vscale x 8 x i1> splat (i1 true), i32 %vl2)
  ret void
}

; False test for different types

define void @test3(ptr %a, i32 %vl1, i32 %vl2, <vscale x 4 x i32> %v1, <vscale x 8 x i32> %v2) {
;
; CHECK-LABEL: @test3(
; CHECK-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[V1:%.*]], ptr nonnull [[A:%.*]], <vscale x 4 x i1> splat (i1 true), i32 [[VL1:%.*]])
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V2:%.*]], ptr nonnull [[A]], <vscale x 8 x i1> splat (i1 true), i32 [[VL2:%.*]])
; CHECK-NEXT:    ret void
;
  call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> %v1, ptr nonnull %a, <vscale x 4 x i1> splat (i1 true), i32 %vl1)
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v2, ptr nonnull %a, <vscale x 8 x i1> splat (i1 true), i32 %vl2)
  ret void
}

; False test for different element count

define void @test4(ptr %a, i32 %vl, <vscale x 4 x i64> %v1, <vscale x 8 x i32> %v2) {
;
; CHECK-LABEL: @test4(
; CHECK-NEXT:    call void @llvm.vp.store.nxv4i64.p0(<vscale x 4 x i64> [[V1:%.*]], ptr nonnull [[A:%.*]], <vscale x 4 x i1> splat (i1 true), i32 [[VL:%.*]])
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V2:%.*]], ptr nonnull [[A]], <vscale x 8 x i1> splat (i1 true), i32 [[VL]])
; CHECK-NEXT:    ret void
;
  call void @llvm.vp.store.nxv4i64.p0(<vscale x 4 x i64> %v1, ptr nonnull %a, <vscale x 4 x i1> splat (i1 true), i32 %vl)
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v2, ptr nonnull %a, <vscale x 8 x i1> splat (i1 true), i32 %vl)
  ret void
}

; False test for different masks

define void @test5(ptr %a, i32 %vl, <vscale x 8 x i32> %v1, <vscale x 8 x i32> %v2, <vscale x 8 x i1> %m1, <vscale x 8 x i1> %m2) {
;
; CHECK-LABEL: @test5(
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V1:%.*]], ptr nonnull [[A:%.*]], <vscale x 8 x i1> [[M1:%.*]], i32 [[VL:%.*]])
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V2:%.*]], ptr nonnull [[A]], <vscale x 8 x i1> [[M2:%.*]], i32 [[VL]])
; CHECK-NEXT:    ret void
;
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v1, ptr nonnull %a, <vscale x 8 x i1> %m1, i32 %vl)
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v2, ptr nonnull %a, <vscale x 8 x i1> %m2, i32 %vl)
  ret void
}

; False test for different pointers

define void @test6(ptr %a, ptr %b, i32 %vl, <vscale x 8 x i32> %v1, <vscale x 8 x i32> %v2, <vscale x 8 x i1> %m1) {
;
; CHECK-LABEL: @test6(
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V1:%.*]], ptr nonnull [[A:%.*]], <vscale x 8 x i1> [[M1:%.*]], i32 [[VL:%.*]])
; CHECK-NEXT:    call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> [[V2:%.*]], ptr nonnull [[B:%.*]], <vscale x 8 x i1> [[M1]], i32 [[VL]])
; CHECK-NEXT:    ret void
;
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v1, ptr nonnull %a, <vscale x 8 x i1> %m1, i32 %vl)
  call void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32> %v2, ptr nonnull %b, <vscale x 8 x i1> %m1, i32 %vl)
  ret void
}

declare <vscale x 8 x i32> @llvm.vp.add.nxv8i32(<vscale x 8 x i32>, <vscale x 8 x i32>, <vscale x 8 x i1>, i32)
declare void @llvm.vp.store.nxv8i32.p0(<vscale x 8 x i32>, ptr nocapture, <vscale x 8 x i1>, i32)
declare void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32>, ptr nocapture, <vscale x 4 x i1>, i32)
declare void @llvm.vp.store.nxv4i64.p0(<vscale x 4 x i64>, ptr nocapture, <vscale x 4 x i1>, i32)