shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll
Florian Hahn 2e4c06780c
[LV] Add extra X86 cost tests for any_of reduction and multi-exit loops. 
Add extra test coverage to ensure decisions do not change when
transitioning to a VPlan-based cost model.
2024-06-10 13:13:04 +01:00

483 lines
30 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=loop-vectorize -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx -S | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.8.0"
@c = common global [2048 x i32] zeroinitializer, align 16
@b = common global [2048 x i32] zeroinitializer, align 16
@d = common global [2048 x i32] zeroinitializer, align 16
@a = common global [2048 x i32] zeroinitializer, align 16
; The program below gathers and scatters data. We better not vectorize it.
define void @cost_model_1() nounwind uwtable noinline ssp {
; CHECK-LABEL: @cost_model_1(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-NEXT: [[TMP0:%.*]] = shl nsw i64 [[INDVARS_IV]], 1
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [2048 x i32], ptr @c, i64 0, i64 [[TMP0]]
; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX]], align 8
; CHECK-NEXT: [[IDXPROM1:%.*]] = sext i32 [[TMP1]] to i64
; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2048 x i32], ptr @b, i64 0, i64 [[IDXPROM1]]
; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4
; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [2048 x i32], ptr @d, i64 0, i64 [[INDVARS_IV]]
; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
; CHECK-NEXT: [[IDXPROM5:%.*]] = sext i32 [[TMP3]] to i64
; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds [2048 x i32], ptr @a, i64 0, i64 [[IDXPROM5]]
; CHECK-NEXT: store i32 [[TMP2]], ptr [[ARRAYIDX6]], align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], 256
; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
; CHECK: for.end:
; CHECK-NEXT: ret void
;
entry:
br label %for.body
for.body: ; preds = %for.body, %entry
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%0 = shl nsw i64 %indvars.iv, 1
%arrayidx = getelementptr inbounds [2048 x i32], ptr @c, i64 0, i64 %0
%1 = load i32, ptr %arrayidx, align 8
%idxprom1 = sext i32 %1 to i64
%arrayidx2 = getelementptr inbounds [2048 x i32], ptr @b, i64 0, i64 %idxprom1
%2 = load i32, ptr %arrayidx2, align 4
%arrayidx4 = getelementptr inbounds [2048 x i32], ptr @d, i64 0, i64 %indvars.iv
%3 = load i32, ptr %arrayidx4, align 4
%idxprom5 = sext i32 %3 to i64
%arrayidx6 = getelementptr inbounds [2048 x i32], ptr @a, i64 0, i64 %idxprom5
store i32 %2, ptr %arrayidx6, align 4
%indvars.iv.next = add i64 %indvars.iv, 1
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
%exitcond = icmp eq i32 %lftr.wideiv, 256
br i1 %exitcond, label %for.end, label %for.body
for.end: ; preds = %for.body
ret void
}
; This function uses a stride that is generally too big to benefit from vectorization without
; really good support for a gather load. But if we don't vectorize the pointer induction,
; then we don't need to extract the pointers out of vector of pointers,
; and the vectorization becomes profitable.
define float @PR27826(ptr nocapture readonly %a, ptr nocapture readonly %b, i32 %n) {
; CHECK-LABEL: @PR27826(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N:%.*]], 0
; CHECK-NEXT: br i1 [[CMP]], label [[PREHEADER:%.*]], label [[FOR_END:%.*]]
; CHECK: preheader:
; CHECK-NEXT: [[T0:%.*]] = sext i32 [[N]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[T0]], -1
; CHECK-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 5
; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 16
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 16
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
; CHECK-NEXT: [[IND_END:%.*]] = mul i64 [[N_VEC]], 32
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP119:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP120:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP121:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP122:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 32
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 0
; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 32
; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 64
; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 96
; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 128
; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[OFFSET_IDX]], 160
; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], 192
; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[OFFSET_IDX]], 224
; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], 256
; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[OFFSET_IDX]], 288
; CHECK-NEXT: [[TMP13:%.*]] = add i64 [[OFFSET_IDX]], 320
; CHECK-NEXT: [[TMP14:%.*]] = add i64 [[OFFSET_IDX]], 352
; CHECK-NEXT: [[TMP15:%.*]] = add i64 [[OFFSET_IDX]], 384
; CHECK-NEXT: [[TMP16:%.*]] = add i64 [[OFFSET_IDX]], 416
; CHECK-NEXT: [[TMP17:%.*]] = add i64 [[OFFSET_IDX]], 448
; CHECK-NEXT: [[TMP18:%.*]] = add i64 [[OFFSET_IDX]], 480
; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP4]]
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP5]]
; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP6]]
; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP7]]
; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP8]]
; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP9]]
; CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP10]]
; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP11]]
; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP12]]
; CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP13]]
; CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP14]]
; CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP15]]
; CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP16]]
; CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP17]]
; CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP18]]
; CHECK-NEXT: [[TMP35:%.*]] = load float, ptr [[TMP19]], align 4
; CHECK-NEXT: [[TMP36:%.*]] = load float, ptr [[TMP20]], align 4
; CHECK-NEXT: [[TMP37:%.*]] = load float, ptr [[TMP21]], align 4
; CHECK-NEXT: [[TMP38:%.*]] = load float, ptr [[TMP22]], align 4
; CHECK-NEXT: [[TMP39:%.*]] = insertelement <4 x float> poison, float [[TMP35]], i32 0
; CHECK-NEXT: [[TMP40:%.*]] = insertelement <4 x float> [[TMP39]], float [[TMP36]], i32 1
; CHECK-NEXT: [[TMP41:%.*]] = insertelement <4 x float> [[TMP40]], float [[TMP37]], i32 2
; CHECK-NEXT: [[TMP42:%.*]] = insertelement <4 x float> [[TMP41]], float [[TMP38]], i32 3
; CHECK-NEXT: [[TMP43:%.*]] = load float, ptr [[TMP23]], align 4
; CHECK-NEXT: [[TMP44:%.*]] = load float, ptr [[TMP24]], align 4
; CHECK-NEXT: [[TMP45:%.*]] = load float, ptr [[TMP25]], align 4
; CHECK-NEXT: [[TMP46:%.*]] = load float, ptr [[TMP26]], align 4
; CHECK-NEXT: [[TMP47:%.*]] = insertelement <4 x float> poison, float [[TMP43]], i32 0
; CHECK-NEXT: [[TMP48:%.*]] = insertelement <4 x float> [[TMP47]], float [[TMP44]], i32 1
; CHECK-NEXT: [[TMP49:%.*]] = insertelement <4 x float> [[TMP48]], float [[TMP45]], i32 2
; CHECK-NEXT: [[TMP50:%.*]] = insertelement <4 x float> [[TMP49]], float [[TMP46]], i32 3
; CHECK-NEXT: [[TMP51:%.*]] = load float, ptr [[TMP27]], align 4
; CHECK-NEXT: [[TMP52:%.*]] = load float, ptr [[TMP28]], align 4
; CHECK-NEXT: [[TMP53:%.*]] = load float, ptr [[TMP29]], align 4
; CHECK-NEXT: [[TMP54:%.*]] = load float, ptr [[TMP30]], align 4
; CHECK-NEXT: [[TMP55:%.*]] = insertelement <4 x float> poison, float [[TMP51]], i32 0
; CHECK-NEXT: [[TMP56:%.*]] = insertelement <4 x float> [[TMP55]], float [[TMP52]], i32 1
; CHECK-NEXT: [[TMP57:%.*]] = insertelement <4 x float> [[TMP56]], float [[TMP53]], i32 2
; CHECK-NEXT: [[TMP58:%.*]] = insertelement <4 x float> [[TMP57]], float [[TMP54]], i32 3
; CHECK-NEXT: [[TMP59:%.*]] = load float, ptr [[TMP31]], align 4
; CHECK-NEXT: [[TMP60:%.*]] = load float, ptr [[TMP32]], align 4
; CHECK-NEXT: [[TMP61:%.*]] = load float, ptr [[TMP33]], align 4
; CHECK-NEXT: [[TMP62:%.*]] = load float, ptr [[TMP34]], align 4
; CHECK-NEXT: [[TMP63:%.*]] = insertelement <4 x float> poison, float [[TMP59]], i32 0
; CHECK-NEXT: [[TMP64:%.*]] = insertelement <4 x float> [[TMP63]], float [[TMP60]], i32 1
; CHECK-NEXT: [[TMP65:%.*]] = insertelement <4 x float> [[TMP64]], float [[TMP61]], i32 2
; CHECK-NEXT: [[TMP66:%.*]] = insertelement <4 x float> [[TMP65]], float [[TMP62]], i32 3
; CHECK-NEXT: [[TMP67:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[TMP3]]
; CHECK-NEXT: [[TMP68:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP4]]
; CHECK-NEXT: [[TMP69:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP5]]
; CHECK-NEXT: [[TMP70:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP6]]
; CHECK-NEXT: [[TMP71:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP7]]
; CHECK-NEXT: [[TMP72:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP8]]
; CHECK-NEXT: [[TMP73:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP9]]
; CHECK-NEXT: [[TMP74:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP10]]
; CHECK-NEXT: [[TMP75:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP11]]
; CHECK-NEXT: [[TMP76:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP12]]
; CHECK-NEXT: [[TMP77:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP13]]
; CHECK-NEXT: [[TMP78:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP14]]
; CHECK-NEXT: [[TMP79:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP15]]
; CHECK-NEXT: [[TMP80:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP16]]
; CHECK-NEXT: [[TMP81:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP17]]
; CHECK-NEXT: [[TMP82:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP18]]
; CHECK-NEXT: [[TMP83:%.*]] = load float, ptr [[TMP67]], align 4
; CHECK-NEXT: [[TMP84:%.*]] = load float, ptr [[TMP68]], align 4
; CHECK-NEXT: [[TMP85:%.*]] = load float, ptr [[TMP69]], align 4
; CHECK-NEXT: [[TMP86:%.*]] = load float, ptr [[TMP70]], align 4
; CHECK-NEXT: [[TMP87:%.*]] = insertelement <4 x float> poison, float [[TMP83]], i32 0
; CHECK-NEXT: [[TMP88:%.*]] = insertelement <4 x float> [[TMP87]], float [[TMP84]], i32 1
; CHECK-NEXT: [[TMP89:%.*]] = insertelement <4 x float> [[TMP88]], float [[TMP85]], i32 2
; CHECK-NEXT: [[TMP90:%.*]] = insertelement <4 x float> [[TMP89]], float [[TMP86]], i32 3
; CHECK-NEXT: [[TMP91:%.*]] = load float, ptr [[TMP71]], align 4
; CHECK-NEXT: [[TMP92:%.*]] = load float, ptr [[TMP72]], align 4
; CHECK-NEXT: [[TMP93:%.*]] = load float, ptr [[TMP73]], align 4
; CHECK-NEXT: [[TMP94:%.*]] = load float, ptr [[TMP74]], align 4
; CHECK-NEXT: [[TMP95:%.*]] = insertelement <4 x float> poison, float [[TMP91]], i32 0
; CHECK-NEXT: [[TMP96:%.*]] = insertelement <4 x float> [[TMP95]], float [[TMP92]], i32 1
; CHECK-NEXT: [[TMP97:%.*]] = insertelement <4 x float> [[TMP96]], float [[TMP93]], i32 2
; CHECK-NEXT: [[TMP98:%.*]] = insertelement <4 x float> [[TMP97]], float [[TMP94]], i32 3
; CHECK-NEXT: [[TMP99:%.*]] = load float, ptr [[TMP75]], align 4
; CHECK-NEXT: [[TMP100:%.*]] = load float, ptr [[TMP76]], align 4
; CHECK-NEXT: [[TMP101:%.*]] = load float, ptr [[TMP77]], align 4
; CHECK-NEXT: [[TMP102:%.*]] = load float, ptr [[TMP78]], align 4
; CHECK-NEXT: [[TMP103:%.*]] = insertelement <4 x float> poison, float [[TMP99]], i32 0
; CHECK-NEXT: [[TMP104:%.*]] = insertelement <4 x float> [[TMP103]], float [[TMP100]], i32 1
; CHECK-NEXT: [[TMP105:%.*]] = insertelement <4 x float> [[TMP104]], float [[TMP101]], i32 2
; CHECK-NEXT: [[TMP106:%.*]] = insertelement <4 x float> [[TMP105]], float [[TMP102]], i32 3
; CHECK-NEXT: [[TMP107:%.*]] = load float, ptr [[TMP79]], align 4
; CHECK-NEXT: [[TMP108:%.*]] = load float, ptr [[TMP80]], align 4
; CHECK-NEXT: [[TMP109:%.*]] = load float, ptr [[TMP81]], align 4
; CHECK-NEXT: [[TMP110:%.*]] = load float, ptr [[TMP82]], align 4
; CHECK-NEXT: [[TMP111:%.*]] = insertelement <4 x float> poison, float [[TMP107]], i32 0
; CHECK-NEXT: [[TMP112:%.*]] = insertelement <4 x float> [[TMP111]], float [[TMP108]], i32 1
; CHECK-NEXT: [[TMP113:%.*]] = insertelement <4 x float> [[TMP112]], float [[TMP109]], i32 2
; CHECK-NEXT: [[TMP114:%.*]] = insertelement <4 x float> [[TMP113]], float [[TMP110]], i32 3
; CHECK-NEXT: [[TMP115:%.*]] = fadd fast <4 x float> [[TMP42]], [[VEC_PHI]]
; CHECK-NEXT: [[TMP116:%.*]] = fadd fast <4 x float> [[TMP50]], [[VEC_PHI1]]
; CHECK-NEXT: [[TMP117:%.*]] = fadd fast <4 x float> [[TMP58]], [[VEC_PHI2]]
; CHECK-NEXT: [[TMP118:%.*]] = fadd fast <4 x float> [[TMP66]], [[VEC_PHI3]]
; CHECK-NEXT: [[TMP119]] = fadd fast <4 x float> [[TMP115]], [[TMP90]]
; CHECK-NEXT: [[TMP120]] = fadd fast <4 x float> [[TMP116]], [[TMP98]]
; CHECK-NEXT: [[TMP121]] = fadd fast <4 x float> [[TMP117]], [[TMP106]]
; CHECK-NEXT: [[TMP122]] = fadd fast <4 x float> [[TMP118]], [[TMP114]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT: [[TMP123:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP123]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[BIN_RDX:%.*]] = fadd fast <4 x float> [[TMP120]], [[TMP119]]
; CHECK-NEXT: [[BIN_RDX4:%.*]] = fadd fast <4 x float> [[TMP121]], [[BIN_RDX]]
; CHECK-NEXT: [[BIN_RDX5:%.*]] = fadd fast <4 x float> [[TMP122]], [[BIN_RDX4]]
; CHECK-NEXT: [[TMP124:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[BIN_RDX5]])
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[PREHEADER]] ]
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ 0.000000e+00, [[PREHEADER]] ], [ [[TMP124]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: br label [[FOR:%.*]]
; CHECK: for:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR]] ]
; CHECK-NEXT: [[S_02:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ADD4:%.*]], [[FOR]] ]
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDVARS_IV]]
; CHECK-NEXT: [[T1:%.*]] = load float, ptr [[ARRAYIDX]], align 4
; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDVARS_IV]]
; CHECK-NEXT: [[T2:%.*]] = load float, ptr [[ARRAYIDX3]], align 4
; CHECK-NEXT: [[ADD:%.*]] = fadd fast float [[T1]], [[S_02]]
; CHECK-NEXT: [[ADD4]] = fadd fast float [[ADD]], [[T2]]
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 32
; CHECK-NEXT: [[CMP1:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], [[T0]]
; CHECK-NEXT: br i1 [[CMP1]], label [[FOR]], label [[LOOPEXIT]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: loopexit:
; CHECK-NEXT: [[ADD4_LCSSA:%.*]] = phi float [ [[ADD4]], [[FOR]] ], [ [[TMP124]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: br label [[FOR_END]]
; CHECK: for.end:
; CHECK-NEXT: [[S_0_LCSSA:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD4_LCSSA]], [[LOOPEXIT]] ]
; CHECK-NEXT: ret float [[S_0_LCSSA]]
;
entry:
%cmp = icmp sgt i32 %n, 0
br i1 %cmp, label %preheader, label %for.end
preheader:
%t0 = sext i32 %n to i64
br label %for
for:
%indvars.iv = phi i64 [ 0, %preheader ], [ %indvars.iv.next, %for ]
%s.02 = phi float [ 0.0, %preheader ], [ %add4, %for ]
%arrayidx = getelementptr inbounds float, ptr %a, i64 %indvars.iv
%t1 = load float, ptr %arrayidx, align 4
%arrayidx3 = getelementptr inbounds float, ptr %b, i64 %indvars.iv
%t2 = load float, ptr %arrayidx3, align 4
%add = fadd fast float %t1, %s.02
%add4 = fadd fast float %add, %t2
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 32
%cmp1 = icmp slt i64 %indvars.iv.next, %t0
br i1 %cmp1, label %for, label %loopexit
loopexit:
%add4.lcssa = phi float [ %add4, %for ]
br label %for.end
for.end:
%s.0.lcssa = phi float [ 0.0, %entry ], [ %add4.lcssa, %loopexit ]
ret float %s.0.lcssa
}
define void @multi_exit(ptr %dst, ptr %src.1, ptr %src.2, i64 %A, i64 %B) #0 {
; CHECK-LABEL: @multi_exit(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[UMAX6:%.*]] = call i64 @llvm.umax.i64(i64 [[B:%.*]], i64 1)
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[UMAX6]], -1
; CHECK-NEXT: [[TMP1:%.*]] = freeze i64 [[TMP0]]
; CHECK-NEXT: [[UMIN7:%.*]] = call i64 @llvm.umin.i64(i64 [[TMP1]], i64 [[A:%.*]])
; CHECK-NEXT: [[TMP2:%.*]] = add nuw i64 [[UMIN7]], 1
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP2]], 30
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
; CHECK: vector.scevcheck:
; CHECK-NEXT: [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[B]], i64 1)
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[UMAX]], -1
; CHECK-NEXT: [[TMP4:%.*]] = freeze i64 [[TMP3]]
; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[TMP4]], i64 [[A]])
; CHECK-NEXT: [[TMP5:%.*]] = icmp ugt i64 [[UMIN]], 4294967295
; CHECK-NEXT: [[TMP6:%.*]] = trunc i64 [[UMIN]] to i32
; CHECK-NEXT: [[TMP7:%.*]] = add i32 1, [[TMP6]]
; CHECK-NEXT: [[TMP8:%.*]] = icmp ult i32 [[TMP7]], 1
; CHECK-NEXT: [[TMP9:%.*]] = icmp ugt i64 [[UMIN]], 4294967295
; CHECK-NEXT: [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
; CHECK-NEXT: br i1 [[TMP10]], label [[SCALAR_PH]], label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST:%.*]], i64 1
; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[SRC_1:%.*]], i64 8
; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[SRC_2:%.*]], i64 8
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP1]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SRC_1]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: [[BOUND03:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP2]]
; CHECK-NEXT: [[BOUND14:%.*]] = icmp ult ptr [[SRC_2]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT5:%.*]] = and i1 [[BOUND03]], [[BOUND14]]
; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT5]]
; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
; CHECK-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i64 4, i64 [[N_MOD_VF]]
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[TMP12]]
; CHECK-NEXT: [[IND_END:%.*]] = trunc i64 [[N_VEC]] to i32
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP13:%.*]] = load i64, ptr [[SRC_1]], align 8, !alias.scope [[META4:![0-9]+]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT9:%.*]] = insertelement <2 x i64> poison, i64 [[TMP13]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT10:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT9]], <2 x i64> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP14:%.*]] = load i64, ptr [[SRC_2]], align 8, !alias.scope [[META7:![0-9]+]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT13:%.*]] = insertelement <2 x i64> poison, i64 [[TMP14]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT14:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT13]], <2 x i64> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP15:%.*]] = icmp eq <2 x i64> [[BROADCAST_SPLAT10]], zeroinitializer
; CHECK-NEXT: [[TMP16:%.*]] = icmp eq <2 x i64> [[BROADCAST_SPLAT10]], zeroinitializer
; CHECK-NEXT: [[TMP17:%.*]] = icmp ne <2 x i64> [[BROADCAST_SPLAT14]], zeroinitializer
; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <2 x i64> [[BROADCAST_SPLAT14]], zeroinitializer
; CHECK-NEXT: [[TMP19:%.*]] = and <2 x i1> [[TMP17]], [[TMP15]]
; CHECK-NEXT: [[TMP20:%.*]] = and <2 x i1> [[TMP18]], [[TMP16]]
; CHECK-NEXT: [[TMP21:%.*]] = zext <2 x i1> [[TMP19]] to <2 x i8>
; CHECK-NEXT: [[TMP22:%.*]] = zext <2 x i1> [[TMP20]] to <2 x i8>
; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x i8> [[TMP22]], i32 1
; CHECK-NEXT: store i8 [[TMP23]], ptr [[DST]], align 1, !alias.scope [[META9:![0-9]+]], !noalias [[META11:![0-9]+]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: br label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ]
; CHECK-NEXT: [[BC_RESUME_VAL8:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ]
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV_1_WIDE:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT_WIDE:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT: [[IV_1:%.*]] = phi i32 [ [[BC_RESUME_VAL8]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP_LATCH]] ]
; CHECK-NEXT: [[EC_1:%.*]] = icmp ult i64 [[IV_1_WIDE]], [[A]]
; CHECK-NEXT: br i1 [[EC_1]], label [[LOOP_LATCH]], label [[EXIT:%.*]]
; CHECK: loop.latch:
; CHECK-NEXT: [[L_1:%.*]] = load i64, ptr [[SRC_1]], align 8
; CHECK-NEXT: [[L_2:%.*]] = load i64, ptr [[SRC_2]], align 8
; CHECK-NEXT: [[CMP55_US:%.*]] = icmp eq i64 [[L_1]], 0
; CHECK-NEXT: [[CMP_I_US:%.*]] = icmp ne i64 [[L_2]], 0
; CHECK-NEXT: [[AND:%.*]] = and i1 [[CMP_I_US]], [[CMP55_US]]
; CHECK-NEXT: [[EXT:%.*]] = zext i1 [[AND]] to i8
; CHECK-NEXT: store i8 [[EXT]], ptr [[DST]], align 1
; CHECK-NEXT: [[IV_1_NEXT]] = add i32 [[IV_1]], 1
; CHECK-NEXT: [[IV_1_NEXT_WIDE]] = zext i32 [[IV_1_NEXT]] to i64
; CHECK-NEXT: [[EC_2:%.*]] = icmp ult i64 [[IV_1_NEXT_WIDE]], [[B]]
; CHECK-NEXT: br i1 [[EC_2]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP13:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
entry:
br label %loop
loop:
%iv.1.wide = phi i64 [ 0, %entry ], [ %iv.1.next.wide, %loop.latch ]
%iv.1 = phi i32 [ 0, %entry ], [ %iv.1.next, %loop.latch ]
%ec.1 = icmp ult i64 %iv.1.wide, %A
br i1 %ec.1, label %loop.latch, label %exit
loop.latch:
%l.1 = load i64, ptr %src.1, align 8
%l.2 = load i64, ptr %src.2, align 8
%cmp55.us = icmp eq i64 %l.1, 0
%cmp.i.us = icmp ne i64 %l.2, 0
%and = and i1 %cmp.i.us, %cmp55.us
%ext = zext i1 %and to i8
store i8 %ext, ptr %dst, align 1
%iv.1.next = add i32 %iv.1, 1
%iv.1.next.wide = zext i32 %iv.1.next to i64
%ec.2 = icmp ult i64 %iv.1.next.wide, %B
br i1 %ec.2, label %loop, label %exit
exit:
ret void
}
define i1 @any_of_cost(ptr %start, ptr %end) #0 {
; CHECK-LABEL: @any_of_cost(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[START2:%.*]] = ptrtoint ptr [[START:%.*]] to i64
; CHECK-NEXT: [[END1:%.*]] = ptrtoint ptr [[END:%.*]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[END1]], [[START2]]
; CHECK-NEXT: [[TMP1:%.*]] = udiv i64 [[TMP0]], 40
; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP2]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
; CHECK-NEXT: [[TMP4:%.*]] = select i1 [[TMP3]], i64 4, i64 [[N_MOD_VF]]
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[TMP4]]
; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[N_VEC]], 40
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP5]]
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i1> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP26:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i1> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP27:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 40
; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 0
; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 40
; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[OFFSET_IDX]], 80
; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], 120
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]]
; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP7]]
; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP8]]
; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP9]]
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 8
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP4]], i64 8
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[NEXT_GEP5]], i64 8
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[NEXT_GEP6]], i64 8
; CHECK-NEXT: [[TMP14:%.*]] = load ptr, ptr [[TMP10]], align 8
; CHECK-NEXT: [[TMP15:%.*]] = load ptr, ptr [[TMP11]], align 8
; CHECK-NEXT: [[TMP16:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP14]], i32 0
; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x ptr> [[TMP16]], ptr [[TMP15]], i32 1
; CHECK-NEXT: [[TMP18:%.*]] = load ptr, ptr [[TMP12]], align 8
; CHECK-NEXT: [[TMP19:%.*]] = load ptr, ptr [[TMP13]], align 8
; CHECK-NEXT: [[TMP20:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP18]], i32 0
; CHECK-NEXT: [[TMP21:%.*]] = insertelement <2 x ptr> [[TMP20]], ptr [[TMP19]], i32 1
; CHECK-NEXT: [[TMP22:%.*]] = icmp eq <2 x ptr> [[TMP17]], zeroinitializer
; CHECK-NEXT: [[TMP23:%.*]] = icmp eq <2 x ptr> [[TMP21]], zeroinitializer
; CHECK-NEXT: [[TMP24:%.*]] = xor <2 x i1> [[TMP22]], <i1 true, i1 true>
; CHECK-NEXT: [[TMP25:%.*]] = xor <2 x i1> [[TMP23]], <i1 true, i1 true>
; CHECK-NEXT: [[TMP26]] = or <2 x i1> [[VEC_PHI]], [[TMP24]]
; CHECK-NEXT: [[TMP27]] = or <2 x i1> [[VEC_PHI3]], [[TMP25]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP28:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[BIN_RDX:%.*]] = or <2 x i1> [[TMP27]], [[TMP26]]
; CHECK-NEXT: [[TMP29:%.*]] = call i1 @llvm.vector.reduce.or.v2i1(<2 x i1> [[BIN_RDX]])
; CHECK-NEXT: [[TMP30:%.*]] = freeze i1 [[TMP29]]
; CHECK-NEXT: [[RDX_SELECT:%.*]] = select i1 [[TMP30]], i1 false, i1 false
; CHECK-NEXT: br label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START]], [[ENTRY:%.*]] ]
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i1 [ false, [[ENTRY]] ], [ [[RDX_SELECT]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[ANY_OF:%.*]] = phi i1 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ANY_OF_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, ptr [[PTR_IV]], i64 8
; CHECK-NEXT: [[L:%.*]] = load ptr, ptr [[GEP]], align 8
; CHECK-NEXT: [[CMP13_NOT_NOT:%.*]] = icmp eq ptr [[L]], null
; CHECK-NEXT: [[ANY_OF_NEXT]] = select i1 [[CMP13_NOT_NOT]], i1 [[ANY_OF]], i1 false
; CHECK-NEXT: [[PTR_IV_NEXT]] = getelementptr inbounds i8, ptr [[PTR_IV]], i64 40
; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq ptr [[PTR_IV]], [[END]]
; CHECK-NEXT: br i1 [[CMP_NOT]], label [[EXIT:%.*]], label [[LOOP]], !llvm.loop [[LOOP15:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: [[ANY_OF_NEXT_LCSSA:%.*]] = phi i1 [ [[ANY_OF_NEXT]], [[LOOP]] ]
; CHECK-NEXT: ret i1 [[ANY_OF_NEXT_LCSSA]]
;
entry:
br label %loop
loop:
%any.of = phi i1 [ false, %entry ], [ %any.of.next, %loop ]
%ptr.iv = phi ptr [ %start, %entry ], [ %ptr.iv.next, %loop ]
%gep = getelementptr i8, ptr %ptr.iv, i64 8
%l = load ptr, ptr %gep, align 8
%cmp13.not.not = icmp eq ptr %l, null
%any.of.next = select i1 %cmp13.not.not, i1 %any.of, i1 false
%ptr.iv.next = getelementptr inbounds i8, ptr %ptr.iv, i64 40
%cmp.not = icmp eq ptr %ptr.iv, %end
br i1 %cmp.not, label %exit, label %loop
exit:
ret i1 %any.of.next
}
attributes #0 = { "target-cpu"="penryn" }
Reference in New Issue View Git Blame Copy Permalink