Skip to content

[LV] Pre-commit test for #128062 #164801

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[LV] Pre-commit test for #128062 #164801

wants to merge 2 commits into from
+101 −0

Conversation

@artagnon
Copy link
Contributor

@artagnon artagnon commented Oct 23, 2025

In preparation to extend the work done by dfa665f ([VPlan] Add transformation to narrow interleave groups) to make the narrowing more powerful, pre-commit a test case from #128062.

@artagnon artagnon requested review from david-arm, fhahn and nikic October 23, 2025 11:11
@llvmbot
Copy link
Member

llvmbot commented Oct 23, 2025

@llvm/pr-subscribers-llvm-transforms

Author: Ramkumar Ramachandra (artagnon)

Changes

In preparation to extend the work done by dfa665f ([VPlan] Add transformation to narrow interleave groups) to make the narrowing more powerful, pre-commit a test case from #128062.

Patch is 24.22 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/164801.diff

1 Files Affected:

  • (modified) llvm/test/Transforms/LoopVectorize/pointer-induction.ll (+356)
diff --git a/llvm/test/Transforms/LoopVectorize/pointer-induction.ll b/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
index 5c62ca3ff3d01..9a655bba280e4 100644
--- a/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
@@ -700,3 +700,359 @@ exit:
   %result = add i64 %cast.ptr, %0
   ret i64 %result
 }
+
+; FIXME: Unprofitable vectorization.
+; Should be smarter about handling interleave groups.
+define void @pr128062(ptr %dst.start, i64 %n, i8 %a) {
+; DEFAULT-LABEL: @pr128062(
+; DEFAULT-NEXT:  entry:
+; DEFAULT-NEXT:    [[IV_START:%.*]] = and i64 [[N:%.*]], -4
+; DEFAULT-NEXT:    [[A_EXT:%.*]] = zext i8 [[A:%.*]] to i16
+; DEFAULT-NEXT:    [[TMP0:%.*]] = add i64 [[IV_START]], -4
+; DEFAULT-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 2
+; DEFAULT-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
+; DEFAULT-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
+; DEFAULT-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; DEFAULT:       vector.ph:
+; DEFAULT-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
+; DEFAULT-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
+; DEFAULT-NEXT:    [[TMP3:%.*]] = mul i64 [[N_VEC]], -4
+; DEFAULT-NEXT:    [[TMP4:%.*]] = add i64 [[IV_START]], [[TMP3]]
+; DEFAULT-NEXT:    [[TMP5:%.*]] = mul i64 [[N_VEC]], 4
+; DEFAULT-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[DST_START:%.*]], i64 [[TMP5]]
+; DEFAULT-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i16> poison, i16 [[A_EXT]], i64 0
+; DEFAULT-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer
+; DEFAULT-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i8> poison, i8 [[A]], i64 0
+; DEFAULT-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i8> [[BROADCAST_SPLATINSERT1]], <4 x i8> poison, <4 x i32> zeroinitializer
+; DEFAULT-NEXT:    br label [[VECTOR_BODY:%.*]]
+; DEFAULT:       vector.body:
+; DEFAULT-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; DEFAULT-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
+; DEFAULT-NEXT:    [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 0
+; DEFAULT-NEXT:    [[TMP8:%.*]] = add i64 [[OFFSET_IDX]], 4
+; DEFAULT-NEXT:    [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], 8
+; DEFAULT-NEXT:    [[TMP10:%.*]] = add i64 [[OFFSET_IDX]], 12
+; DEFAULT-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP7]]
+; DEFAULT-NEXT:    [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP8]]
+; DEFAULT-NEXT:    [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP9]]
+; DEFAULT-NEXT:    [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP10]]
+; DEFAULT-NEXT:    [[TMP11:%.*]] = load i8, ptr [[NEXT_GEP]], align 1
+; DEFAULT-NEXT:    [[TMP12:%.*]] = load i8, ptr [[NEXT_GEP3]], align 1
+; DEFAULT-NEXT:    [[TMP13:%.*]] = load i8, ptr [[NEXT_GEP4]], align 1
+; DEFAULT-NEXT:    [[TMP14:%.*]] = load i8, ptr [[NEXT_GEP5]], align 1
+; DEFAULT-NEXT:    [[TMP15:%.*]] = insertelement <4 x i8> poison, i8 [[TMP11]], i32 0
+; DEFAULT-NEXT:    [[TMP16:%.*]] = insertelement <4 x i8> [[TMP15]], i8 [[TMP12]], i32 1
+; DEFAULT-NEXT:    [[TMP17:%.*]] = insertelement <4 x i8> [[TMP16]], i8 [[TMP13]], i32 2
+; DEFAULT-NEXT:    [[TMP18:%.*]] = insertelement <4 x i8> [[TMP17]], i8 [[TMP14]], i32 3
+; DEFAULT-NEXT:    [[TMP19:%.*]] = zext <4 x i8> [[TMP18]] to <4 x i16>
+; DEFAULT-NEXT:    [[TMP20:%.*]] = mul nuw <4 x i16> [[TMP19]], [[BROADCAST_SPLAT]]
+; DEFAULT-NEXT:    [[TMP21:%.*]] = udiv <4 x i16> [[TMP20]], splat (i16 255)
+; DEFAULT-NEXT:    [[TMP22:%.*]] = trunc nuw <4 x i16> [[TMP21]] to <4 x i8>
+; DEFAULT-NEXT:    [[TMP23:%.*]] = add <4 x i8> [[BROADCAST_SPLAT2]], [[TMP22]]
+; DEFAULT-NEXT:    [[TMP24:%.*]] = extractelement <4 x i8> [[TMP23]], i32 0
+; DEFAULT-NEXT:    [[TMP25:%.*]] = extractelement <4 x i8> [[TMP23]], i32 1
+; DEFAULT-NEXT:    [[TMP26:%.*]] = extractelement <4 x i8> [[TMP23]], i32 2
+; DEFAULT-NEXT:    [[TMP27:%.*]] = extractelement <4 x i8> [[TMP23]], i32 3
+; DEFAULT-NEXT:    store i8 [[TMP24]], ptr [[NEXT_GEP]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP25]], ptr [[NEXT_GEP3]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP26]], ptr [[NEXT_GEP4]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP27]], ptr [[NEXT_GEP5]], align 1
+; DEFAULT-NEXT:    [[TMP28:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP]], i64 2
+; DEFAULT-NEXT:    [[TMP29:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP3]], i64 2
+; DEFAULT-NEXT:    [[TMP30:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP4]], i64 2
+; DEFAULT-NEXT:    [[TMP31:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP5]], i64 2
+; DEFAULT-NEXT:    [[TMP32:%.*]] = load i8, ptr [[TMP28]], align 1
+; DEFAULT-NEXT:    [[TMP33:%.*]] = load i8, ptr [[TMP29]], align 1
+; DEFAULT-NEXT:    [[TMP34:%.*]] = load i8, ptr [[TMP30]], align 1
+; DEFAULT-NEXT:    [[TMP35:%.*]] = load i8, ptr [[TMP31]], align 1
+; DEFAULT-NEXT:    [[TMP36:%.*]] = insertelement <4 x i8> poison, i8 [[TMP32]], i32 0
+; DEFAULT-NEXT:    [[TMP37:%.*]] = insertelement <4 x i8> [[TMP36]], i8 [[TMP33]], i32 1
+; DEFAULT-NEXT:    [[TMP38:%.*]] = insertelement <4 x i8> [[TMP37]], i8 [[TMP34]], i32 2
+; DEFAULT-NEXT:    [[TMP39:%.*]] = insertelement <4 x i8> [[TMP38]], i8 [[TMP35]], i32 3
+; DEFAULT-NEXT:    [[TMP40:%.*]] = zext <4 x i8> [[TMP39]] to <4 x i16>
+; DEFAULT-NEXT:    [[TMP41:%.*]] = mul nuw <4 x i16> [[TMP40]], [[BROADCAST_SPLAT]]
+; DEFAULT-NEXT:    [[TMP42:%.*]] = udiv <4 x i16> [[TMP41]], splat (i16 255)
+; DEFAULT-NEXT:    [[TMP43:%.*]] = trunc nuw <4 x i16> [[TMP42]] to <4 x i8>
+; DEFAULT-NEXT:    [[TMP44:%.*]] = add <4 x i8> [[BROADCAST_SPLAT2]], [[TMP43]]
+; DEFAULT-NEXT:    [[TMP45:%.*]] = extractelement <4 x i8> [[TMP44]], i32 0
+; DEFAULT-NEXT:    [[TMP46:%.*]] = extractelement <4 x i8> [[TMP44]], i32 1
+; DEFAULT-NEXT:    [[TMP47:%.*]] = extractelement <4 x i8> [[TMP44]], i32 2
+; DEFAULT-NEXT:    [[TMP48:%.*]] = extractelement <4 x i8> [[TMP44]], i32 3
+; DEFAULT-NEXT:    store i8 [[TMP45]], ptr [[TMP28]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP46]], ptr [[TMP29]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP47]], ptr [[TMP30]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP48]], ptr [[TMP31]], align 1
+; DEFAULT-NEXT:    [[TMP49:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP]], i64 3
+; DEFAULT-NEXT:    [[TMP50:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP3]], i64 3
+; DEFAULT-NEXT:    [[TMP51:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP4]], i64 3
+; DEFAULT-NEXT:    [[TMP52:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP5]], i64 3
+; DEFAULT-NEXT:    [[TMP53:%.*]] = load i8, ptr [[TMP49]], align 1
+; DEFAULT-NEXT:    [[TMP54:%.*]] = load i8, ptr [[TMP50]], align 1
+; DEFAULT-NEXT:    [[TMP55:%.*]] = load i8, ptr [[TMP51]], align 1
+; DEFAULT-NEXT:    [[TMP56:%.*]] = load i8, ptr [[TMP52]], align 1
+; DEFAULT-NEXT:    [[TMP57:%.*]] = insertelement <4 x i8> poison, i8 [[TMP53]], i32 0
+; DEFAULT-NEXT:    [[TMP58:%.*]] = insertelement <4 x i8> [[TMP57]], i8 [[TMP54]], i32 1
+; DEFAULT-NEXT:    [[TMP59:%.*]] = insertelement <4 x i8> [[TMP58]], i8 [[TMP55]], i32 2
+; DEFAULT-NEXT:    [[TMP60:%.*]] = insertelement <4 x i8> [[TMP59]], i8 [[TMP56]], i32 3
+; DEFAULT-NEXT:    [[TMP61:%.*]] = zext <4 x i8> [[TMP60]] to <4 x i16>
+; DEFAULT-NEXT:    [[TMP62:%.*]] = mul nuw <4 x i16> [[TMP61]], [[BROADCAST_SPLAT]]
+; DEFAULT-NEXT:    [[TMP63:%.*]] = udiv <4 x i16> [[TMP62]], splat (i16 255)
+; DEFAULT-NEXT:    [[TMP64:%.*]] = trunc nuw <4 x i16> [[TMP63]] to <4 x i8>
+; DEFAULT-NEXT:    [[TMP65:%.*]] = add <4 x i8> [[BROADCAST_SPLAT2]], [[TMP64]]
+; DEFAULT-NEXT:    [[TMP66:%.*]] = extractelement <4 x i8> [[TMP65]], i32 0
+; DEFAULT-NEXT:    [[TMP67:%.*]] = extractelement <4 x i8> [[TMP65]], i32 1
+; DEFAULT-NEXT:    [[TMP68:%.*]] = extractelement <4 x i8> [[TMP65]], i32 2
+; DEFAULT-NEXT:    [[TMP69:%.*]] = extractelement <4 x i8> [[TMP65]], i32 3
+; DEFAULT-NEXT:    store i8 [[TMP66]], ptr [[TMP49]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP67]], ptr [[TMP50]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP68]], ptr [[TMP51]], align 1
+; DEFAULT-NEXT:    store i8 [[TMP69]], ptr [[TMP52]], align 1
+; DEFAULT-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; DEFAULT-NEXT:    [[TMP70:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; DEFAULT-NEXT:    br i1 [[TMP70]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; DEFAULT:       middle.block:
+; DEFAULT-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
+; DEFAULT-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; DEFAULT:       scalar.ph:
+; DEFAULT-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ [[IV_START]], [[ENTRY:%.*]] ]
+; DEFAULT-NEXT:    [[BC_RESUME_VAL6:%.*]] = phi ptr [ [[TMP6]], [[MIDDLE_BLOCK]] ], [ [[DST_START]], [[ENTRY]] ]
+; DEFAULT-NEXT:    br label [[LOOP:%.*]]
+; DEFAULT:       loop:
+; DEFAULT-NEXT:    [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
+; DEFAULT-NEXT:    [[DST:%.*]] = phi ptr [ [[DST_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL6]], [[SCALAR_PH]] ]
+; DEFAULT-NEXT:    [[DST_NEXT]] = getelementptr inbounds nuw i8, ptr [[DST]], i64 4
+; DEFAULT-NEXT:    [[LOAD_DST:%.*]] = load i8, ptr [[DST]], align 1
+; DEFAULT-NEXT:    [[DST_EXT:%.*]] = zext i8 [[LOAD_DST]] to i16
+; DEFAULT-NEXT:    [[MUL_DST_0:%.*]] = mul nuw i16 [[DST_EXT]], [[A_EXT]]
+; DEFAULT-NEXT:    [[UDIV_0:%.*]] = udiv i16 [[MUL_DST_0]], 255
+; DEFAULT-NEXT:    [[TRUNC_0:%.*]] = trunc nuw i16 [[UDIV_0]] to i8
+; DEFAULT-NEXT:    [[VAL_0:%.*]] = add i8 [[A]], [[TRUNC_0]]
+; DEFAULT-NEXT:    store i8 [[VAL_0]], ptr [[DST]], align 1
+; DEFAULT-NEXT:    [[GEP_DST_1:%.*]] = getelementptr inbounds nuw i8, ptr [[DST]], i64 1
+; DEFAULT-NEXT:    [[LOAD_DST_1:%.*]] = load i8, ptr [[GEP_DST_1]], align 1
+; DEFAULT-NEXT:    [[DST_1_EXT:%.*]] = zext i8 [[LOAD_DST_1]] to i16
+; DEFAULT-NEXT:    [[MUL_DST_1:%.*]] = mul nuw i16 [[DST_1_EXT]], [[A_EXT]]
+; DEFAULT-NEXT:    [[UDIV_1:%.*]] = udiv i16 [[MUL_DST_1]], 255
+; DEFAULT-NEXT:    [[TRUNC_1:%.*]] = trunc nuw i16 [[UDIV_1]] to i8
+; DEFAULT-NEXT:    [[VAL_1:%.*]] = add i8 [[A]], [[TRUNC_1]]
+; DEFAULT-NEXT:    [[GEP_DST_2:%.*]] = getelementptr inbounds nuw i8, ptr [[DST]], i64 2
+; DEFAULT-NEXT:    [[LOAD_DST_2:%.*]] = load i8, ptr [[GEP_DST_2]], align 1
+; DEFAULT-NEXT:    [[DST_2_EXT:%.*]] = zext i8 [[LOAD_DST_2]] to i16
+; DEFAULT-NEXT:    [[MUL_DST_2:%.*]] = mul nuw i16 [[DST_2_EXT]], [[A_EXT]]
+; DEFAULT-NEXT:    [[UDIV_2:%.*]] = udiv i16 [[MUL_DST_2]], 255
+; DEFAULT-NEXT:    [[TRUNC_2:%.*]] = trunc nuw i16 [[UDIV_2]] to i8
+; DEFAULT-NEXT:    [[VAL_2:%.*]] = add i8 [[A]], [[TRUNC_2]]
+; DEFAULT-NEXT:    store i8 [[VAL_2]], ptr [[GEP_DST_2]], align 1
+; DEFAULT-NEXT:    [[GEP_DST_3:%.*]] = getelementptr inbounds nuw i8, ptr [[DST]], i64 3
+; DEFAULT-NEXT:    [[LOAD_DST_3:%.*]] = load i8, ptr [[GEP_DST_3]], align 1
+; DEFAULT-NEXT:    [[DST_3_EXT:%.*]] = zext i8 [[LOAD_DST_3]] to i16
+; DEFAULT-NEXT:    [[MUL_DST_3:%.*]] = mul nuw i16 [[DST_3_EXT]], [[A_EXT]]
+; DEFAULT-NEXT:    [[UDIV_3:%.*]] = udiv i16 [[MUL_DST_3]], 255
+; DEFAULT-NEXT:    [[TRUNC_3:%.*]] = trunc nuw i16 [[UDIV_3]] to i8
+; DEFAULT-NEXT:    [[VAL_3:%.*]] = add i8 [[A]], [[TRUNC_3]]
+; DEFAULT-NEXT:    store i8 [[VAL_3]], ptr [[GEP_DST_3]], align 1
+; DEFAULT-NEXT:    [[IV_NEXT]] = add i64 [[IV]], -4
+; DEFAULT-NEXT:    [[EXIT_COND:%.*]] = icmp eq i64 [[IV_NEXT]], 0
+; DEFAULT-NEXT:    br i1 [[EXIT_COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP9:![0-9]+]]
+; DEFAULT:       exit:
+; DEFAULT-NEXT:    ret void
+;
+; STRIDED-LABEL: @pr128062(
+; STRIDED-NEXT:  entry:
+; STRIDED-NEXT:    [[IV_START:%.*]] = and i64 [[N:%.*]], -4
+; STRIDED-NEXT:    [[A_EXT:%.*]] = zext i8 [[A:%.*]] to i16
+; STRIDED-NEXT:    [[TMP0:%.*]] = add i64 [[IV_START]], -4
+; STRIDED-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 2
+; STRIDED-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
+; STRIDED-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
+; STRIDED-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; STRIDED:       vector.ph:
+; STRIDED-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
+; STRIDED-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
+; STRIDED-NEXT:    [[TMP3:%.*]] = mul i64 [[N_VEC]], -4
+; STRIDED-NEXT:    [[TMP4:%.*]] = add i64 [[IV_START]], [[TMP3]]
+; STRIDED-NEXT:    [[TMP5:%.*]] = mul i64 [[N_VEC]], 4
+; STRIDED-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[DST_START:%.*]], i64 [[TMP5]]
+; STRIDED-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i16> poison, i16 [[A_EXT]], i64 0
+; STRIDED-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer
+; STRIDED-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i8> poison, i8 [[A]], i64 0
+; STRIDED-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i8> [[BROADCAST_SPLATINSERT1]], <4 x i8> poison, <4 x i32> zeroinitializer
+; STRIDED-NEXT:    br label [[VECTOR_BODY:%.*]]
+; STRIDED:       vector.body:
+; STRIDED-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; STRIDED-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
+; STRIDED-NEXT:    [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 0
+; STRIDED-NEXT:    [[TMP8:%.*]] = add i64 [[OFFSET_IDX]], 4
+; STRIDED-NEXT:    [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], 8
+; STRIDED-NEXT:    [[TMP10:%.*]] = add i64 [[OFFSET_IDX]], 12
+; STRIDED-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP7]]
+; STRIDED-NEXT:    [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP8]]
+; STRIDED-NEXT:    [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP9]]
+; STRIDED-NEXT:    [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[DST_START]], i64 [[TMP10]]
+; STRIDED-NEXT:    [[TMP11:%.*]] = load i8, ptr [[NEXT_GEP]], align 1
+; STRIDED-NEXT:    [[TMP12:%.*]] = load i8, ptr [[NEXT_GEP3]], align 1
+; STRIDED-NEXT:    [[TMP13:%.*]] = load i8, ptr [[NEXT_GEP4]], align 1
+; STRIDED-NEXT:    [[TMP14:%.*]] = load i8, ptr [[NEXT_GEP5]], align 1
+; STRIDED-NEXT:    [[TMP15:%.*]] = insertelement <4 x i8> poison, i8 [[TMP11]], i32 0
+; STRIDED-NEXT:    [[TMP16:%.*]] = insertelement <4 x i8> [[TMP15]], i8 [[TMP12]], i32 1
+; STRIDED-NEXT:    [[TMP17:%.*]] = insertelement <4 x i8> [[TMP16]], i8 [[TMP13]], i32 2
+; STRIDED-NEXT:    [[TMP18:%.*]] = insertelement <4 x i8> [[TMP17]], i8 [[TMP14]], i32 3
+; STRIDED-NEXT:    [[TMP19:%.*]] = zext <4 x i8> [[TMP18]] to <4 x i16>
+; STRIDED-NEXT:    [[TMP20:%.*]] = mul nuw <4 x i16> [[TMP19]], [[BROADCAST_SPLAT]]
+; STRIDED-NEXT:    [[TMP21:%.*]] = udiv <4 x i16> [[TMP20]], splat (i16 255)
+; STRIDED-NEXT:    [[TMP22:%.*]] = trunc nuw <4 x i16> [[TMP21]] to <4 x i8>
+; STRIDED-NEXT:    [[TMP23:%.*]] = add <4 x i8> [[BROADCAST_SPLAT2]], [[TMP22]]
+; STRIDED-NEXT:    [[TMP24:%.*]] = extractelement <4 x i8> [[TMP23]], i32 0
+; STRIDED-NEXT:    [[TMP25:%.*]] = extractelement <4 x i8> [[TMP23]], i32 1
+; STRIDED-NEXT:    [[TMP26:%.*]] = extractelement <4 x i8> [[TMP23]], i32 2
+; STRIDED-NEXT:    [[TMP27:%.*]] = extractelement <4 x i8> [[TMP23]], i32 3
+; STRIDED-NEXT:    store i8 [[TMP24]], ptr [[NEXT_GEP]], align 1
+; STRIDED-NEXT:    store i8 [[TMP25]], ptr [[NEXT_GEP3]], align 1
+; STRIDED-NEXT:    store i8 [[TMP26]], ptr [[NEXT_GEP4]], align 1
+; STRIDED-NEXT:    store i8 [[TMP27]], ptr [[NEXT_GEP5]], align 1
+; STRIDED-NEXT:    [[TMP28:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP]], i64 2
+; STRIDED-NEXT:    [[TMP29:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP3]], i64 2
+; STRIDED-NEXT:    [[TMP30:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP4]], i64 2
+; STRIDED-NEXT:    [[TMP31:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP5]], i64 2
+; STRIDED-NEXT:    [[TMP32:%.*]] = load i8, ptr [[TMP28]], align 1
+; STRIDED-NEXT:    [[TMP33:%.*]] = load i8, ptr [[TMP29]], align 1
+; STRIDED-NEXT:    [[TMP34:%.*]] = load i8, ptr [[TMP30]], align 1
+; STRIDED-NEXT:    [[TMP35:%.*]] = load i8, ptr [[TMP31]], align 1
+; STRIDED-NEXT:    [[TMP36:%.*]] = insertelement <4 x i8> poison, i8 [[TMP32]], i32 0
+; STRIDED-NEXT:    [[TMP37:%.*]] = insertelement <4 x i8> [[TMP36]], i8 [[TMP33]], i32 1
+; STRIDED-NEXT:    [[TMP38:%.*]] = insertelement <4 x i8> [[TMP37]], i8 [[TMP34]], i32 2
+; STRIDED-NEXT:    [[TMP39:%.*]] = insertelement <4 x i8> [[TMP38]], i8 [[TMP35]], i32 3
+; STRIDED-NEXT:    [[TMP40:%.*]] = zext <4 x i8> [[TMP39]] to <4 x i16>
+; STRIDED-NEXT:    [[TMP41:%.*]] = mul nuw <4 x i16> [[TMP40]], [[BROADCAST_SPLAT]]
+; STRIDED-NEXT:    [[TMP42:%.*]] = udiv <4 x i16> [[TMP41]], splat (i16 255)
+; STRIDED-NEXT:    [[TMP43:%.*]] = trunc nuw <4 x i16> [[TMP42]] to <4 x i8>
+; STRIDED-NEXT:    [[TMP44:%.*]] = add <4 x i8> [[BROADCAST_SPLAT2]], [[TMP43]]
+; STRIDED-NEXT:    [[TMP45:%.*]] = extractelement <4 x i8> [[TMP44]], i32 0
+; STRIDED-NEXT:    [[TMP46:%.*]] = extractelement <4 x i8> [[TMP44]], i32 1
+; STRIDED-NEXT:    [[TMP47:%.*]] = extractelement <4 x i8> [[TMP44]], i32 2
+; STRIDED-NEXT:    [[TMP48:%.*]] = extractelement <4 x i8> [[TMP44]], i32 3
+; STRIDED-NEXT:    store i8 [[TMP45]], ptr [[TMP28]], align 1
+; STRIDED-NEXT:    store i8 [[TMP46]], ptr [[TMP29]], align 1
+; STRIDED-NEXT:    store i8 [[TMP47]], ptr [[TMP30]], align 1
+; STRIDED-NEXT:    store i8 [[TMP48]], ptr [[TMP31]], align 1
+; STRIDED-NEXT:    [[TMP49:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP]], i64 3
+; STRIDED-NEXT:    [[TMP50:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP3]], i64 3
+; STRIDED-NEXT:    [[TMP51:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP4]], i64 3
+; STRIDED-NEXT:    [[TMP52:%.*]] = getelementptr inbounds nuw i8, ptr [[NEXT_GEP5]], i64 3
+; STRIDED-NEXT:    [[TMP53:%.*]] = load i8, ptr [[TMP49]], align 1
+; STRIDED-NEXT:    [[TMP54:%.*]] = load i8, ptr [[TMP50]], align 1
+; STRIDED-NEXT:    [[TMP55:%.*]] = load i8, ptr [[TMP51]], align 1
+; STRIDED-NEXT:    [[TMP56:%.*]] = load i8, ptr [[TMP52]], align 1
+; STRIDED-NEXT:    [[TMP57:%.*]] = insertelement <4 x i8> poison, i8 [[TMP53]], i32 0
+; STRIDED-NEXT:    [[TMP58:%.*]] = insertelement <4 x i8> [[TMP57]], i8 [[TMP54]], i32 1
+; STRIDED-NEXT:    [[TMP59:%.*]] = insertelement <4 x i8> [[TMP58]], i8 [[TMP55]], i32 2
+; STRIDED-NEXT:    [[TMP60:%.*]] = insertelement <4 x i8> [[TMP59]], i8 [[TMP56]], i32 3
+; STRIDED-NEXT:    [[TMP61:%.*]] = zext <4 x i8> [[TMP60]] to <4 x i16>
+; STRIDED-NEXT:    [[TMP62:%.*]] = mul nuw <4 x i16> [[TMP61]], [[BROADCAST_SPLAT]]
+; STRIDED-NEXT:    [[TMP63:%.*]] = udiv <4 x i16> [[TMP62]], splat (i16 255)
+; STRIDED-NEXT:    [[TMP64:%.*]] = trunc nuw <4 x i16> [[TMP63]] to <4 x i8>
+; STRIDED-NEXT:    [[TMP65:%.*]] = add <4 x i8> [[BROADCAST_SPLAT2]], [[TMP64]]
+; STRIDED-NEXT:    [[TMP66:%.*]] = extractelement <4 x i8> [[TMP65]], i32 0
+; STRIDED-NEXT:    [[TMP67:%.*]] = extractelement <4 x i8> [[TMP65]], i32 1
+; STRIDED-NEXT:    [[TMP68:%.*]] = extractelement <4 x i8> [[TMP65]], i32 2
+; STRIDED-NEXT:    [[TMP69:%.*]] = extractelement <4 x i8> [[TMP65]], i32 3
+; STRIDED-NEXT:    store i8 [[TMP66]], ptr [[TMP49]], align 1
+; STRIDED-NEXT:    store i8 [[TMP67]], ptr [[TMP50]], align 1
+; STRIDED-NEXT:    store i8 [[TMP68]], ptr [[TMP51]], align 1
+; STRIDED-NEXT:    store i8 [[TMP69]], ptr [[TMP52]], align 1
+; STRIDED-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; STRIDED-NEXT:    [[TMP70:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; STRIDED-NEXT:    br i1 [[TMP70]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; STRIDED:       middle.block:
+; STRIDED-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
+; STRIDED-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; STRIDED:       scalar.ph:
+; STRIDED-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ [[IV_START]], [[ENTRY:%.*]] ]
+; STRIDED-NEXT:    [[BC_RESUME_VAL6:%.*]] = phi ptr [ [[TMP6]], [[MIDDLE_BLOCK]] ], [ [[DST_START]], [[ENTRY]] ]
+; STRIDED-NEXT:    br label [[LOOP:%.*]]
+; STRIDED:       loop:
+; STRIDED-NEXT:    [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
+; STRIDED...
[truncated]

fhahn
fhahn reviewed Oct 23, 2025
View reviewed changes
@artagnon
[LV] Pre-commit test for llvm#128062
360d886 
In preparation to extend the work done by dfa665f ([VPlan] Add
transformation to narrow interleave groups) to make the narrowing more
powerful, pre-commit a test case from llvm#128062.
@artagnon
[LV] Improve test
456948d
@artagnon
Copy link
Contributor Author

artagnon commented Oct 28, 2025

Gentle ping.

@artagnon artagnon removed the request for review from nikic November 3, 2025 07:26
@artagnon
Copy link
Contributor Author

artagnon commented Nov 3, 2025

Gentle ping.

david-arm
david-arm approved these changes Nov 4, 2025
View reviewed changes
Copy link
Contributor

@david-arm david-arm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM!

fhahn
fhahn reviewed Nov 4, 2025
View reviewed changes
Copy link
Contributor

@fhahn fhahn left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Did you check this works as expected? I think to trigger narrowInterleaveGroups, a target-specific test is needed, as it checks the size of the vector against the target vector size.

@artagnon
Copy link
Contributor Author

artagnon commented Nov 4, 2025

Did you check this works as expected? I think to trigger narrowInterleaveGroups, a target-specific test is needed, as it checks the size of the vector against the target vector size.

I got a diff with #164839, although that PR needs more thought?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@fhahn fhahn fhahn left review comments

@david-arm david-arm david-arm approved these changes

Assignees

No one assigned

Labels

llvm:transforms

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@artagnon @llvmbot @fhahn @david-arm