[llvm] [LoopVectorize] Fix nondeterminism in loop-vectorize (PR #200833)
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Mon Jun 8 10:46:00 PDT 2026
================
@@ -0,0 +1,157 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt --passes=loop-vectorize %s -S | FileCheck %s
+
+; Check that we see expected deterministic (over multiple test runs) output.
+; NOTE: Beware, if this test fails it may be due to non-determinism.
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux"
+
+define void @fun(i64 %0, float %1, ptr %a, ptr %b, i64 %len) #0 {
+; CHECK-LABEL: define void @fun(
+; CHECK-SAME: i64 [[TMP0:%.*]], float [[TMP1:%.*]], ptr [[A:%.*]], ptr [[B:%.*]], i64 [[LEN:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[VLA:%.*]] = alloca float, i64 [[LEN]], align 16
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP0]], 1
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 8
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
+; CHECK: [[VECTOR_MEMCHECK]]:
+; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP0]], 2
+; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP3]], 32
+; CHECK-NEXT: br i1 [[DIFF_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 8
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr [4 x i8], ptr [[VLA]], i64 [[TMP0]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[TMP1]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 3
+; CHECK-NEXT: [[TMP44:%.*]] = add i64 [[INDEX]], 4
+; CHECK-NEXT: [[TMP45:%.*]] = add i64 [[INDEX]], 5
+; CHECK-NEXT: [[TMP46:%.*]] = add i64 [[INDEX]], 6
+; CHECK-NEXT: [[TMP47:%.*]] = add i64 [[INDEX]], 7
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[TMP6]]
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[TMP7]]
+; CHECK-NEXT: [[TMP12:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[TMP8]]
+; CHECK-NEXT: [[TMP48:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[TMP44]]
+; CHECK-NEXT: [[TMP49:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[TMP45]]
+; CHECK-NEXT: [[TMP50:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[TMP46]]
+; CHECK-NEXT: [[TMP51:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[TMP47]]
+; CHECK-NEXT: [[TMP13:%.*]] = load ptr, ptr [[TMP9]], align 8
+; CHECK-NEXT: [[TMP14:%.*]] = load ptr, ptr [[TMP10]], align 8
+; CHECK-NEXT: [[TMP15:%.*]] = load ptr, ptr [[TMP11]], align 8
+; CHECK-NEXT: [[TMP16:%.*]] = load ptr, ptr [[TMP12]], align 8
+; CHECK-NEXT: [[TMP60:%.*]] = load ptr, ptr [[TMP48]], align 8
+; CHECK-NEXT: [[TMP62:%.*]] = load ptr, ptr [[TMP49]], align 8
+; CHECK-NEXT: [[TMP64:%.*]] = load ptr, ptr [[TMP50]], align 8
+; CHECK-NEXT: [[TMP65:%.*]] = load ptr, ptr [[TMP51]], align 8
+; CHECK-NEXT: [[TMP17:%.*]] = load i64, ptr [[TMP13]], align 8
+; CHECK-NEXT: [[TMP18:%.*]] = load i64, ptr [[TMP14]], align 8
+; CHECK-NEXT: [[TMP19:%.*]] = load i64, ptr [[TMP15]], align 8
+; CHECK-NEXT: [[TMP20:%.*]] = load i64, ptr [[TMP16]], align 8
+; CHECK-NEXT: [[TMP66:%.*]] = load i64, ptr [[TMP60]], align 8
+; CHECK-NEXT: [[TMP33:%.*]] = load i64, ptr [[TMP62]], align 8
+; CHECK-NEXT: [[TMP67:%.*]] = load i64, ptr [[TMP64]], align 8
+; CHECK-NEXT: [[TMP35:%.*]] = load i64, ptr [[TMP65]], align 8
+; CHECK-NEXT: [[TMP21:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP17]]
+; CHECK-NEXT: [[TMP22:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP18]]
+; CHECK-NEXT: [[TMP23:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP19]]
+; CHECK-NEXT: [[TMP24:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP20]]
+; CHECK-NEXT: [[TMP68:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP66]]
+; CHECK-NEXT: [[TMP69:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP33]]
+; CHECK-NEXT: [[TMP70:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP67]]
+; CHECK-NEXT: [[TMP43:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP35]]
+; CHECK-NEXT: [[TMP25:%.*]] = load float, ptr [[TMP21]], align 4
+; CHECK-NEXT: [[TMP26:%.*]] = load float, ptr [[TMP22]], align 4
+; CHECK-NEXT: [[TMP27:%.*]] = load float, ptr [[TMP23]], align 4
+; CHECK-NEXT: [[TMP28:%.*]] = load float, ptr [[TMP24]], align 4
+; CHECK-NEXT: [[TMP29:%.*]] = insertelement <4 x float> poison, float [[TMP25]], i32 0
+; CHECK-NEXT: [[TMP30:%.*]] = insertelement <4 x float> [[TMP29]], float [[TMP26]], i32 1
+; CHECK-NEXT: [[TMP31:%.*]] = insertelement <4 x float> [[TMP30]], float [[TMP27]], i32 2
+; CHECK-NEXT: [[TMP32:%.*]] = insertelement <4 x float> [[TMP31]], float [[TMP28]], i32 3
+; CHECK-NEXT: [[TMP52:%.*]] = load float, ptr [[TMP68]], align 4
+; CHECK-NEXT: [[TMP53:%.*]] = load float, ptr [[TMP69]], align 4
+; CHECK-NEXT: [[TMP54:%.*]] = load float, ptr [[TMP70]], align 4
+; CHECK-NEXT: [[TMP55:%.*]] = load float, ptr [[TMP43]], align 4
+; CHECK-NEXT: [[TMP56:%.*]] = insertelement <4 x float> poison, float [[TMP52]], i32 0
+; CHECK-NEXT: [[TMP57:%.*]] = insertelement <4 x float> [[TMP56]], float [[TMP53]], i32 1
+; CHECK-NEXT: [[TMP58:%.*]] = insertelement <4 x float> [[TMP57]], float [[TMP54]], i32 2
+; CHECK-NEXT: [[TMP59:%.*]] = insertelement <4 x float> [[TMP58]], float [[TMP55]], i32 3
+; CHECK-NEXT: [[TMP34:%.*]] = getelementptr [4 x i8], ptr [[VLA]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP61:%.*]] = getelementptr float, ptr [[TMP34]], i64 4
+; CHECK-NEXT: store <4 x float> [[TMP32]], ptr [[TMP34]], align 4
+; CHECK-NEXT: store <4 x float> [[TMP59]], ptr [[TMP61]], align 4
+; CHECK-NEXT: [[TMP36:%.*]] = getelementptr [4 x i8], ptr [[TMP4]], i64 [[INDEX]]
+; CHECK-NEXT: [[TMP63:%.*]] = getelementptr float, ptr [[TMP36]], i64 4
+; CHECK-NEXT: store <4 x float> [[BROADCAST_SPLAT]], ptr [[TMP36]], align 4
+; CHECK-NEXT: store <4 x float> [[BROADCAST_SPLAT]], ptr [[TMP63]], align 4
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
+; CHECK-NEXT: [[TMP37:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP37]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END_LOOPEXIT:.*]], label %[[SCALAR_PH]]
+; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ], [ 0, %[[VECTOR_MEMCHECK]] ]
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ]
+; CHECK-NEXT: [[ARRAYIDX16:%.*]] = getelementptr [8 x i8], ptr [[A]], i64 [[INDVARS_IV]]
+; CHECK-NEXT: [[TMP38:%.*]] = load ptr, ptr [[ARRAYIDX16]], align 8
+; CHECK-NEXT: [[TMP39:%.*]] = load i64, ptr [[TMP38]], align 8
+; CHECK-NEXT: [[ARRAYIDX18:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP39]]
+; CHECK-NEXT: [[TMP40:%.*]] = load float, ptr [[ARRAYIDX18]], align 4
+; CHECK-NEXT: [[ARRAYIDX21:%.*]] = getelementptr [4 x i8], ptr [[VLA]], i64 [[INDVARS_IV]]
+; CHECK-NEXT: store float [[TMP40]], ptr [[ARRAYIDX21]], align 4
+; CHECK-NEXT: [[TMP41:%.*]] = load i64, ptr [[B]], align 8
+; CHECK-NEXT: [[ARRAYIDX27:%.*]] = getelementptr [4 x i8], ptr [[A]], i64 [[TMP41]]
+; CHECK-NEXT: [[TMP42:%.*]] = load float, ptr [[ARRAYIDX27]], align 4
+; CHECK-NEXT: [[ARRAYIDX28:%.*]] = getelementptr [4 x i8], ptr [[VLA]], i64 [[TMP0]]
+; CHECK-NEXT: [[ARRAYIDX30:%.*]] = getelementptr [4 x i8], ptr [[ARRAYIDX28]], i64 [[INDVARS_IV]]
+; CHECK-NEXT: store float [[TMP1]], ptr [[ARRAYIDX30]], align 4
+; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV]], [[TMP0]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK: [[FOR_END_LOOPEXIT]]:
+; CHECK-NEXT: ret void
+;
+entry:
+ %vla = alloca float, i64 %len, align 16
+ br label %loop
+
+loop: ; preds = %loop, %entry
----------------
fhahn wrote:
```suggestion
loop:
```
https://github.com/llvm/llvm-project/pull/200833
More information about the llvm-commits
mailing list