test/Transforms/DeadStoreElimination/combined-partial-overwrites.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -S -passes=dse -enable-dse-partial-store-merging=false < %s | FileCheck --check-prefixes=CHECK %s
target datalayout = "E-m:e-i64:64-n32:64"
target triple = "powerpc64le-unknown-linux"

%"struct.std::complex" = type { { float, float } }

define void @_Z4testSt7complexIfE(ptr noalias nocapture sret(%"struct.std::complex") %agg.result, i64 %c.coerce) {
; CHECK-LABEL: @_Z4testSt7complexIfE(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[REF_TMP:%.*]] = alloca i64, align 8
; CHECK-NEXT:    [[C_SROA_0_0_EXTRACT_SHIFT:%.*]] = lshr i64 [[C_COERCE:%.*]], 32
; CHECK-NEXT:    [[C_SROA_0_0_EXTRACT_TRUNC:%.*]] = trunc i64 [[C_SROA_0_0_EXTRACT_SHIFT]] to i32
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i32 [[C_SROA_0_0_EXTRACT_TRUNC]] to float
; CHECK-NEXT:    [[C_SROA_2_0_EXTRACT_TRUNC:%.*]] = trunc i64 [[C_COERCE]] to i32
; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32 [[C_SROA_2_0_EXTRACT_TRUNC]] to float
; CHECK-NEXT:    call void @_Z3barSt7complexIfE(ptr nonnull sret(%"struct.std::complex") [[REF_TMP]], i64 [[C_COERCE]])
; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr [[REF_TMP]], align 8
; CHECK-NEXT:    [[TMP3:%.*]] = lshr i64 [[TMP2]], 32
; CHECK-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP4]] to float
; CHECK-NEXT:    [[_M_VALUE_IMAGP_I_I:%.*]] = getelementptr inbounds %"struct.std::complex", ptr [[AGG_RESULT:%.*]], i64 0, i32 0, i32 1
; CHECK-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[TMP6]] to float
; CHECK-NEXT:    [[MUL_AD_I_I:%.*]] = fmul fast float [[TMP5]], [[TMP1]]
; CHECK-NEXT:    [[MUL_BC_I_I:%.*]] = fmul fast float [[TMP7]], [[TMP0]]
; CHECK-NEXT:    [[MUL_I_I_I:%.*]] = fadd fast float [[MUL_AD_I_I]], [[MUL_BC_I_I]]
; CHECK-NEXT:    [[MUL_AC_I_I:%.*]] = fmul fast float [[TMP5]], [[TMP0]]
; CHECK-NEXT:    [[MUL_BD_I_I:%.*]] = fmul fast float [[TMP7]], [[TMP1]]
; CHECK-NEXT:    [[MUL_R_I_I:%.*]] = fsub fast float [[MUL_AC_I_I]], [[MUL_BD_I_I]]
; CHECK-NEXT:    store float [[MUL_R_I_I]], ptr [[AGG_RESULT]], align 4
; CHECK-NEXT:    store float [[MUL_I_I_I]], ptr [[_M_VALUE_IMAGP_I_I]], align 4
; CHECK-NEXT:    ret void
;
entry:

  %ref.tmp = alloca i64, align 8
  %c.sroa.0.0.extract.shift = lshr i64 %c.coerce, 32
  %c.sroa.0.0.extract.trunc = trunc i64 %c.sroa.0.0.extract.shift to i32
  %0 = bitcast i32 %c.sroa.0.0.extract.trunc to float
  %c.sroa.2.0.extract.trunc = trunc i64 %c.coerce to i32
  %1 = bitcast i32 %c.sroa.2.0.extract.trunc to float
  call void @_Z3barSt7complexIfE(ptr nonnull sret(%"struct.std::complex") %ref.tmp, i64 %c.coerce)
  %2 = load i64, ptr %ref.tmp, align 8
  store i64 %2, ptr %agg.result, align 4

  %3 = lshr i64 %2, 32
  %4 = trunc i64 %3 to i32
  %5 = bitcast i32 %4 to float
  %_M_value.imagp.i.i = getelementptr inbounds %"struct.std::complex", ptr %agg.result, i64 0, i32 0, i32 1
  %6 = trunc i64 %2 to i32
  %7 = bitcast i32 %6 to float
  %mul_ad.i.i = fmul fast float %5, %1
  %mul_bc.i.i = fmul fast float %7, %0
  %mul_i.i.i = fadd fast float %mul_ad.i.i, %mul_bc.i.i
  %mul_ac.i.i = fmul fast float %5, %0
  %mul_bd.i.i = fmul fast float %7, %1
  %mul_r.i.i = fsub fast float %mul_ac.i.i, %mul_bd.i.i
  store float %mul_r.i.i, ptr %agg.result, align 4
  store float %mul_i.i.i, ptr %_M_value.imagp.i.i, align 4
  ret void
}

declare void @_Z3barSt7complexIfE(ptr sret(%"struct.std::complex"), i64)

define void @test1(ptr %ptr) {
; CHECK-LABEL: @test1(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    store i16 -30062, ptr [[PTR:%.*]], align 2
; CHECK-NEXT:    [[BPTR3:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 3
; CHECK-NEXT:    store i8 47, ptr [[BPTR3]], align 1
; CHECK-NEXT:    [[BPTR1:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 1
; CHECK-NEXT:    store i16 2020, ptr [[BPTR1]], align 1
; CHECK-NEXT:    ret void
;
entry:

  store i32 5, ptr %ptr
  store i8 7, ptr %ptr
  store i16 -30062, ptr %ptr
  %bptr2 = getelementptr inbounds i8, ptr %ptr, i64 2
  store i8 25, ptr %bptr2
  %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 3
  store i8 47, ptr %bptr3
  %bptr1 = getelementptr inbounds i8, ptr %ptr, i64 1
  store i16 2020, ptr %bptr1, align 1
  ret void


}

define void @test2(ptr %ptr) {
; CHECK-LABEL: @test2(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[BPTRM1:%.*]] = getelementptr inbounds i8, ptr [[PTR:%.*]], i64 -1
; CHECK-NEXT:    [[BPTR1:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 1
; CHECK-NEXT:    [[BPTR2:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 2
; CHECK-NEXT:    [[BPTR3:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 3
; CHECK-NEXT:    store i16 1456, ptr [[BPTRM1]], align 1
; CHECK-NEXT:    store i16 1346, ptr [[PTR]], align 1
; CHECK-NEXT:    store i16 1756, ptr [[BPTR1]], align 1
; CHECK-NEXT:    store i16 1126, ptr [[BPTR2]], align 1
; CHECK-NEXT:    store i16 5656, ptr [[BPTR3]], align 1
; CHECK-NEXT:    ret void
;
entry:

  store i32 5, ptr %ptr

  %bptrm1 = getelementptr inbounds i8, ptr %ptr, i64 -1
  %bptr1 = getelementptr inbounds i8, ptr %ptr, i64 1
  %bptr2 = getelementptr inbounds i8, ptr %ptr, i64 2
  %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 3


  store i16 1456, ptr %bptrm1, align 1
  store i16 1346, ptr %ptr, align 1
  store i16 1756, ptr %bptr1, align 1
  store i16 1126, ptr %bptr2, align 1
  store i16 5656, ptr %bptr3, align 1



  ret void

}

define signext i8 @test3(ptr %ptr) {
; CHECK-LABEL: @test3(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    store i32 5, ptr [[PTR:%.*]], align 4
; CHECK-NEXT:    [[BPTRM1:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 -1
; CHECK-NEXT:    [[BPTR1:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 1
; CHECK-NEXT:    [[BPTR2:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 2
; CHECK-NEXT:    [[BPTR3:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 3
; CHECK-NEXT:    [[V:%.*]] = load i8, ptr [[PTR]], align 1
; CHECK-NEXT:    store i16 1456, ptr [[BPTRM1]], align 1
; CHECK-NEXT:    store i16 1346, ptr [[PTR]], align 1
; CHECK-NEXT:    store i16 1756, ptr [[BPTR1]], align 1
; CHECK-NEXT:    store i16 1126, ptr [[BPTR2]], align 1
; CHECK-NEXT:    store i16 5656, ptr [[BPTR3]], align 1
; CHECK-NEXT:    ret i8 [[V]]
;
entry:

  store i32 5, ptr %ptr

  %bptrm1 = getelementptr inbounds i8, ptr %ptr, i64 -1
  %bptr1 = getelementptr inbounds i8, ptr %ptr, i64 1
  %bptr2 = getelementptr inbounds i8, ptr %ptr, i64 2
  %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 3


  %v = load i8, ptr %ptr, align 1
  store i16 1456, ptr %bptrm1, align 1
  store i16 1346, ptr %ptr, align 1
  store i16 1756, ptr %bptr1, align 1
  store i16 1126, ptr %bptr2, align 1
  store i16 5656, ptr %bptr3, align 1


  ret i8 %v

}

%struct.foostruct = type {
ptr,
ptr,
ptr,
ptr,
ptr
}
declare void @llvm.memset.p0.i64(ptr nocapture, i8, i64, i1)
declare void @goFunc(ptr)
declare i32 @fa(ptr, ptr, i32, i8, ptr)

; We miss this case, because of an aggressive limit of partial overlap analysis.
; With a larger partial store limit, we remove the memset.
define void @test4()  {
; CHECK-LABEL: @test4(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[BANG:%.*]] = alloca [[STRUCT_FOOSTRUCT:%.*]], align 8
; CHECK-NEXT:    store ptr @fa, ptr [[BANG]], align 8
; CHECK-NEXT:    [[V3:%.*]] = getelementptr inbounds [[STRUCT_FOOSTRUCT]], ptr [[BANG]], i64 0, i32 1
; CHECK-NEXT:    store ptr @fa, ptr [[V3]], align 8
; CHECK-NEXT:    [[V4:%.*]] = getelementptr inbounds [[STRUCT_FOOSTRUCT]], ptr [[BANG]], i64 0, i32 2
; CHECK-NEXT:    store ptr @fa, ptr [[V4]], align 8
; CHECK-NEXT:    [[V5:%.*]] = getelementptr inbounds [[STRUCT_FOOSTRUCT]], ptr [[BANG]], i64 0, i32 3
; CHECK-NEXT:    store ptr @fa, ptr [[V5]], align 8
; CHECK-NEXT:    [[V6:%.*]] = getelementptr inbounds [[STRUCT_FOOSTRUCT]], ptr [[BANG]], i64 0, i32 4
; CHECK-NEXT:    store ptr null, ptr [[V6]], align 8
; CHECK-NEXT:    call void @goFunc(ptr [[BANG]])
; CHECK-NEXT:    ret void
entry:

  %bang = alloca %struct.foostruct, align 8
  call void @llvm.memset.p0.i64(ptr align 8 %bang, i8 0, i64 40, i1 false)
  store ptr @fa, ptr %bang, align 8
  %v3 = getelementptr inbounds %struct.foostruct, ptr %bang, i64 0, i32 1
  store ptr @fa, ptr %v3, align 8
  %v4 = getelementptr inbounds %struct.foostruct, ptr %bang, i64 0, i32 2
  store ptr @fa, ptr %v4, align 8
  %v5 = getelementptr inbounds %struct.foostruct, ptr %bang, i64 0, i32 3
  store ptr @fa, ptr %v5, align 8
  %v6 = getelementptr inbounds %struct.foostruct, ptr %bang, i64 0, i32 4
  store ptr null, ptr %v6, align 8
  call void @goFunc(ptr %bang)
  ret void

}

define signext i8 @test5(ptr %ptr) {
; CHECK-LABEL: @test5(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[BPTR1:%.*]] = getelementptr inbounds i8, ptr [[PTR:%.*]], i64 1
; CHECK-NEXT:    [[BPTR2:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 2
; CHECK-NEXT:    [[BPTR3:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 3
; CHECK-NEXT:    store i16 -1, ptr [[BPTR2]], align 1
; CHECK-NEXT:    store i16 1456, ptr [[BPTR1]], align 1
; CHECK-NEXT:    store i16 1346, ptr [[PTR]], align 1
; CHECK-NEXT:    ret i8 0
;
entry:

  store i32 0, ptr %ptr

  %bptr1 = getelementptr inbounds i8, ptr %ptr, i64 1
  %bptr2 = getelementptr inbounds i8, ptr %ptr, i64 2
  %bptr3 = getelementptr inbounds i8, ptr %ptr, i64 3


  store i16 65535, ptr %bptr2, align 1
  store i16 1456, ptr %bptr1, align 1
  store i16 1346, ptr %ptr, align 1


  ret i8 0
}

define signext i8 @test6(ptr %ptr) {
; CHECK-LABEL: @test6(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[BPTR2:%.*]] = getelementptr inbounds i16, ptr [[PTR:%.*]], i64 1
; CHECK-NEXT:    store i16 1456, ptr [[BPTR2]], align 1
; CHECK-NEXT:    store i16 -1, ptr [[PTR]], align 1
; CHECK-NEXT:    ret i8 0
;
entry:

  store i32 0, ptr %ptr

  %bptr2 = getelementptr inbounds i16, ptr %ptr, i64 1

  store i16 1456, ptr %bptr2, align 1
  store i16 65535, ptr %ptr, align 1


  ret i8 0
}

define signext i8 @test7(ptr %ptr) {
; CHECK-LABEL: @test7(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[BPTR2:%.*]] = getelementptr inbounds i16, ptr [[PTR:%.*]], i64 1
; CHECK-NEXT:    [[BPTR3:%.*]] = getelementptr inbounds i16, ptr [[PTR]], i64 2
; CHECK-NEXT:    [[BPTR4:%.*]] = getelementptr inbounds i16, ptr [[PTR]], i64 3
; CHECK-NEXT:    store i16 1346, ptr [[PTR]], align 1
; CHECK-NEXT:    store i16 1756, ptr [[BPTR3]], align 1
; CHECK-NEXT:    store i16 1456, ptr [[BPTR2]], align 1
; CHECK-NEXT:    store i16 5656, ptr [[BPTR4]], align 1
; CHECK-NEXT:    ret i8 0
;
entry:

  store i64 0, ptr %ptr

  %bptr2 = getelementptr inbounds i16, ptr %ptr, i64 1
  %bptr3 = getelementptr inbounds i16, ptr %ptr, i64 2
  %bptr4 = getelementptr inbounds i16, ptr %ptr, i64 3

  store i16 1346, ptr %ptr, align 1
  store i16 1756, ptr %bptr3, align 1
  store i16 1456, ptr %bptr2, align 1
  store i16 5656, ptr %bptr4, align 1


  ret i8 0
}