[llvm] [SLP] NFC: Precommit test for pull/137419 (PR #137730)
Jeffrey Byrnes via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 11 13:24:09 PDT 2025
https://github.com/jrbyrnes updated https://github.com/llvm/llvm-project/pull/137730
>From 28ec8e2772fb27cfb081afdf062260afb8c70e17 Mon Sep 17 00:00:00 2001
From: Jeffrey Byrnes <Jeffrey.Byrnes at amd.com>
Date: Mon, 28 Apr 2025 15:52:39 -0700
Subject: [PATCH 1/2] [SLP] NFC: Precommit test for pull/137419
Change-Id: I1501ad0d1904cb3f3532c2424ac3c523a18f6764
---
.../SLPVectorizer/AMDGPU/external-shuffle.ll | 262 ++++++++++++++++++
1 file changed, 262 insertions(+)
create mode 100644 llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll
diff --git a/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll b/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll
new file mode 100644
index 0000000000000..b5749f7d098f2
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll
@@ -0,0 +1,262 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=slp-vectorizer,dce < %s | FileCheck -check-prefixes=GCN %s
+
+define protected void @phi_4(ptr addrspace(3) %inptr0, ptr addrspace(3) %inptr1, ptr %out, ptr %out1, ptr %out2, i32 %flag) {
+; GCN-LABEL: define protected void @phi_4(
+; GCN-SAME: ptr addrspace(3) [[INPTR0:%.*]], ptr addrspace(3) [[INPTR1:%.*]], ptr [[OUT:%.*]], ptr [[OUT1:%.*]], ptr [[OUT2:%.*]], i32 [[FLAG:%.*]]) #[[ATTR0:[0-9]+]] {
+; GCN-NEXT: [[ENTRY:.*]]:
+; GCN-NEXT: [[GEP0:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 0
+; GCN-NEXT: [[TMP0:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP0]], align 8
+; GCN-NEXT: [[GEP2:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 2
+; GCN-NEXT: [[GEP3:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 3
+; GCN-NEXT: [[TMP1:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP2]], align 2
+; GCN-NEXT: [[GEP4:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 4
+; GCN-NEXT: [[GEP5:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 5
+; GCN-NEXT: [[TMP2:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP4]], align 8
+; GCN-NEXT: [[GEP6:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 6
+; GCN-NEXT: [[GEP7:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 7
+; GCN-NEXT: [[TMP3:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP6]], align 2
+; GCN-NEXT: [[GEP8:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 8
+; GCN-NEXT: [[GEP9:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 9
+; GCN-NEXT: [[TMP4:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP8]], align 8
+; GCN-NEXT: [[GEP10:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 10
+; GCN-NEXT: [[GEP11:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 11
+; GCN-NEXT: [[TMP5:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP10]], align 2
+; GCN-NEXT: [[GEP12:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 12
+; GCN-NEXT: [[GEP13:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 13
+; GCN-NEXT: [[TMP6:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP12]], align 8
+; GCN-NEXT: [[GEP14:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 14
+; GCN-NEXT: [[TMP7:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP14]], align 2
+; GCN-NEXT: [[TMP8:%.*]] = extractelement <2 x i16> [[TMP1]], i32 0
+; GCN-NEXT: [[TMP9:%.*]] = extractelement <2 x i16> [[TMP1]], i32 1
+; GCN-NEXT: [[TMP10:%.*]] = extractelement <2 x i16> [[TMP2]], i32 0
+; GCN-NEXT: [[TMP11:%.*]] = extractelement <2 x i16> [[TMP2]], i32 1
+; GCN-NEXT: [[TMP12:%.*]] = extractelement <2 x i16> [[TMP3]], i32 0
+; GCN-NEXT: [[TMP13:%.*]] = extractelement <2 x i16> [[TMP3]], i32 1
+; GCN-NEXT: [[TMP14:%.*]] = extractelement <2 x i16> [[TMP4]], i32 0
+; GCN-NEXT: [[TMP15:%.*]] = extractelement <2 x i16> [[TMP4]], i32 1
+; GCN-NEXT: [[TMP24:%.*]] = extractelement <2 x i16> [[TMP5]], i32 0
+; GCN-NEXT: [[TMP26:%.*]] = extractelement <2 x i16> [[TMP5]], i32 1
+; GCN-NEXT: [[TMP28:%.*]] = extractelement <2 x i16> [[TMP6]], i32 0
+; GCN-NEXT: [[TMP38:%.*]] = extractelement <2 x i16> [[TMP6]], i32 1
+; GCN-NEXT: br label %[[DO_BODY:.*]]
+; GCN: [[DO_BODY]]:
+; GCN-NEXT: [[PHI2:%.*]] = phi i16 [ [[TMP8]], %[[ENTRY]] ], [ [[TMP30:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI3:%.*]] = phi i16 [ [[TMP9]], %[[ENTRY]] ], [ [[OTHERELE3:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI4:%.*]] = phi i16 [ [[TMP10]], %[[ENTRY]] ], [ [[TMP39:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI5:%.*]] = phi i16 [ [[TMP11]], %[[ENTRY]] ], [ [[OTHERELE5:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI6:%.*]] = phi i16 [ [[TMP12]], %[[ENTRY]] ], [ [[TMP32:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI7:%.*]] = phi i16 [ [[TMP13]], %[[ENTRY]] ], [ [[OTHERELE7:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI8:%.*]] = phi i16 [ [[TMP14]], %[[ENTRY]] ], [ [[TMP40:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI9:%.*]] = phi i16 [ [[TMP15]], %[[ENTRY]] ], [ [[OTHERELE9:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI10:%.*]] = phi i16 [ [[TMP24]], %[[ENTRY]] ], [ [[TMP34:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI11:%.*]] = phi i16 [ [[TMP26]], %[[ENTRY]] ], [ [[OTHERELE11:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI12:%.*]] = phi i16 [ [[TMP28]], %[[ENTRY]] ], [ [[TMP35:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[PHI13:%.*]] = phi i16 [ [[TMP38]], %[[ENTRY]] ], [ [[OTHERELE13:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[TMP41:%.*]] = phi <2 x i16> [ [[TMP0]], %[[ENTRY]] ], [ [[TMP16:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[TMP42:%.*]] = phi <2 x i16> [ [[TMP7]], %[[ENTRY]] ], [ [[TMP23:%.*]], %[[DO_BODY]] ]
+; GCN-NEXT: [[TMP16]] = load <2 x i16>, ptr addrspace(3) [[GEP0]], align 8
+; GCN-NEXT: [[OTHERELE3]] = load i16, ptr addrspace(3) [[GEP3]], align 1
+; GCN-NEXT: [[TMP17:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP2]], align 2
+; GCN-NEXT: [[OTHERELE5]] = load i16, ptr addrspace(3) [[GEP5]], align 1
+; GCN-NEXT: [[TMP18:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP4]], align 8
+; GCN-NEXT: [[OTHERELE7]] = load i16, ptr addrspace(3) [[GEP7]], align 1
+; GCN-NEXT: [[TMP19:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP6]], align 2
+; GCN-NEXT: [[OTHERELE9]] = load i16, ptr addrspace(3) [[GEP9]], align 1
+; GCN-NEXT: [[TMP20:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP8]], align 8
+; GCN-NEXT: [[OTHERELE11]] = load i16, ptr addrspace(3) [[GEP11]], align 1
+; GCN-NEXT: [[TMP21:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP10]], align 2
+; GCN-NEXT: [[OTHERELE13]] = load i16, ptr addrspace(3) [[GEP13]], align 1
+; GCN-NEXT: [[TMP22:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP12]], align 8
+; GCN-NEXT: [[TMP23]] = load <2 x i16>, ptr addrspace(3) [[GEP14]], align 2
+; GCN-NEXT: [[CMP:%.*]] = icmp eq i32 [[FLAG]], 0
+; GCN-NEXT: [[TMP30]] = extractelement <2 x i16> [[TMP17]], i32 0
+; GCN-NEXT: [[TMP39]] = extractelement <2 x i16> [[TMP18]], i32 0
+; GCN-NEXT: [[TMP32]] = extractelement <2 x i16> [[TMP19]], i32 0
+; GCN-NEXT: [[TMP40]] = extractelement <2 x i16> [[TMP20]], i32 0
+; GCN-NEXT: [[TMP34]] = extractelement <2 x i16> [[TMP21]], i32 0
+; GCN-NEXT: [[TMP35]] = extractelement <2 x i16> [[TMP22]], i32 0
+; GCN-NEXT: br i1 [[CMP]], label %[[EXIT:.*]], label %[[DO_BODY]]
+; GCN: [[EXIT]]:
+; GCN-NEXT: [[TMP36:%.*]] = shufflevector <2 x i16> [[TMP16]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[TMP37:%.*]] = shufflevector <2 x i16> [[TMP17]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC038:%.*]] = shufflevector <16 x i16> [[TMP36]], <16 x i16> [[TMP37]], <16 x i32> <i32 0, i32 1, i32 16, i32 17, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP25:%.*]] = shufflevector <2 x i16> [[TMP18]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC059:%.*]] = shufflevector <16 x i16> [[VEC038]], <16 x i16> [[TMP25]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP27:%.*]] = shufflevector <2 x i16> [[TMP19]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC0710:%.*]] = shufflevector <16 x i16> [[VEC059]], <16 x i16> [[TMP27]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 16, i32 17, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP29:%.*]] = shufflevector <2 x i16> [[TMP20]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC0911:%.*]] = shufflevector <16 x i16> [[VEC0710]], <16 x i16> [[TMP29]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP31:%.*]] = shufflevector <2 x i16> [[TMP21]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC01112:%.*]] = shufflevector <16 x i16> [[VEC0911]], <16 x i16> [[TMP31]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 16, i32 17, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP33:%.*]] = shufflevector <2 x i16> [[TMP22]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[TMP58:%.*]] = shufflevector <16 x i16> [[VEC01112]], <16 x i16> [[TMP33]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 14, i32 15>
+; GCN-NEXT: [[TMP60:%.*]] = shufflevector <2 x i16> [[TMP23]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC2157:%.*]] = shufflevector <16 x i16> [[TMP58]], <16 x i16> [[TMP60]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 16, i32 17>
+; GCN-NEXT: [[TMP50:%.*]] = shufflevector <2 x i16> [[TMP0]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[TMP51:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC231:%.*]] = shufflevector <16 x i16> [[TMP50]], <16 x i16> [[TMP51]], <16 x i32> <i32 0, i32 1, i32 16, i32 17, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP52:%.*]] = shufflevector <2 x i16> [[TMP2]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC252:%.*]] = shufflevector <16 x i16> [[VEC231]], <16 x i16> [[TMP52]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP53:%.*]] = shufflevector <2 x i16> [[TMP3]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC273:%.*]] = shufflevector <16 x i16> [[VEC252]], <16 x i16> [[TMP53]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 16, i32 17, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP54:%.*]] = shufflevector <2 x i16> [[TMP4]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC294:%.*]] = shufflevector <16 x i16> [[VEC273]], <16 x i16> [[TMP54]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP55:%.*]] = shufflevector <2 x i16> [[TMP5]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC2115:%.*]] = shufflevector <16 x i16> [[VEC294]], <16 x i16> [[TMP55]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 16, i32 17, i32 12, i32 13, i32 14, i32 15>
+; GCN-NEXT: [[TMP56:%.*]] = shufflevector <2 x i16> [[TMP6]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC2136:%.*]] = shufflevector <16 x i16> [[VEC2115]], <16 x i16> [[TMP56]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 14, i32 15>
+; GCN-NEXT: [[TMP59:%.*]] = shufflevector <2 x i16> [[TMP7]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC2151:%.*]] = shufflevector <16 x i16> [[VEC2136]], <16 x i16> [[TMP59]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 16, i32 17>
+; GCN-NEXT: [[TMP57:%.*]] = shufflevector <2 x i16> [[TMP41]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC22:%.*]] = insertelement <16 x i16> [[TMP57]], i16 [[PHI2]], i64 2
+; GCN-NEXT: [[VEC23:%.*]] = insertelement <16 x i16> [[VEC22]], i16 [[PHI3]], i64 3
+; GCN-NEXT: [[VEC24:%.*]] = insertelement <16 x i16> [[VEC23]], i16 [[PHI4]], i64 4
+; GCN-NEXT: [[VEC25:%.*]] = insertelement <16 x i16> [[VEC24]], i16 [[PHI5]], i64 5
+; GCN-NEXT: [[VEC26:%.*]] = insertelement <16 x i16> [[VEC25]], i16 [[PHI6]], i64 6
+; GCN-NEXT: [[VEC27:%.*]] = insertelement <16 x i16> [[VEC26]], i16 [[PHI7]], i64 7
+; GCN-NEXT: [[VEC28:%.*]] = insertelement <16 x i16> [[VEC27]], i16 [[PHI8]], i64 8
+; GCN-NEXT: [[VEC29:%.*]] = insertelement <16 x i16> [[VEC28]], i16 [[PHI9]], i64 9
+; GCN-NEXT: [[VEC210:%.*]] = insertelement <16 x i16> [[VEC29]], i16 [[PHI10]], i64 10
+; GCN-NEXT: [[VEC211:%.*]] = insertelement <16 x i16> [[VEC210]], i16 [[PHI11]], i64 11
+; GCN-NEXT: [[VEC212:%.*]] = insertelement <16 x i16> [[VEC211]], i16 [[PHI12]], i64 12
+; GCN-NEXT: [[VEC213:%.*]] = insertelement <16 x i16> [[VEC212]], i16 [[PHI13]], i64 13
+; GCN-NEXT: [[TMP61:%.*]] = shufflevector <2 x i16> [[TMP42]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; GCN-NEXT: [[VEC2152:%.*]] = shufflevector <16 x i16> [[VEC213]], <16 x i16> [[TMP61]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 16, i32 17>
+; GCN-NEXT: store <16 x i16> [[VEC2151]], ptr [[OUT]], align 32
+; GCN-NEXT: store <16 x i16> [[VEC2157]], ptr [[OUT1]], align 32
+; GCN-NEXT: store <16 x i16> [[VEC2152]], ptr [[OUT2]], align 32
+; GCN-NEXT: ret void
+;
+entry:
+ %gep0 = getelementptr i16, ptr addrspace(3) %inptr0, i32 0
+ %ele0 = load i16, ptr addrspace(3) %gep0, align 8
+ %gep1 = getelementptr i16, ptr addrspace(3) %inptr0, i32 1
+ %ele1 = load i16, ptr addrspace(3) %gep1, align 1
+ %gep2 = getelementptr i16, ptr addrspace(3) %inptr0, i32 2
+ %ele2 = load i16, ptr addrspace(3) %gep2, align 2
+ %gep3 = getelementptr i16, ptr addrspace(3) %inptr0, i32 3
+ %ele3 = load i16, ptr addrspace(3) %gep3, align 1
+ %gep4 = getelementptr i16, ptr addrspace(3) %inptr0, i32 4
+ %ele4 = load i16, ptr addrspace(3) %gep4, align 8
+ %gep5 = getelementptr i16, ptr addrspace(3) %inptr0, i32 5
+ %ele5 = load i16, ptr addrspace(3) %gep5, align 1
+ %gep6 = getelementptr i16, ptr addrspace(3) %inptr0, i32 6
+ %ele6 = load i16, ptr addrspace(3) %gep6, align 2
+ %gep7 = getelementptr i16, ptr addrspace(3) %inptr0, i32 7
+ %ele7 = load i16, ptr addrspace(3) %gep7, align 1
+ %gep8 = getelementptr i16, ptr addrspace(3) %inptr0, i32 8
+ %ele8 = load i16, ptr addrspace(3) %gep8, align 8
+ %gep9 = getelementptr i16, ptr addrspace(3) %inptr0, i32 9
+ %ele9 = load i16, ptr addrspace(3) %gep9, align 1
+ %gep10 = getelementptr i16, ptr addrspace(3) %inptr0, i32 10
+ %ele10 = load i16, ptr addrspace(3) %gep10, align 2
+ %gep11 = getelementptr i16, ptr addrspace(3) %inptr0, i32 11
+ %ele11 = load i16, ptr addrspace(3) %gep11, align 1
+ %gep12 = getelementptr i16, ptr addrspace(3) %inptr0, i32 12
+ %ele12 = load i16, ptr addrspace(3) %gep12, align 8
+ %gep13 = getelementptr i16, ptr addrspace(3) %inptr0, i32 13
+ %ele13 = load i16, ptr addrspace(3) %gep13, align 1
+ %gep14 = getelementptr i16, ptr addrspace(3) %inptr0, i32 14
+ %ele14 = load i16, ptr addrspace(3) %gep14, align 2
+ %gep15 = getelementptr i16, ptr addrspace(3) %inptr0, i32 15
+ %ele15 = load i16, ptr addrspace(3) %gep15, align 1
+ br label %do.body
+
+do.body:
+ %phi0 = phi i16 [ %ele0, %entry ], [ %otherele0, %do.body ]
+ %phi1 = phi i16 [ %ele1, %entry ], [ %otherele1, %do.body ]
+ %phi2 = phi i16 [ %ele2, %entry ], [ %otherele2, %do.body ]
+ %phi3 = phi i16 [ %ele3, %entry ], [ %otherele3, %do.body ]
+ %phi4 = phi i16 [ %ele4, %entry ], [ %otherele4, %do.body ]
+ %phi5 = phi i16 [ %ele5, %entry ], [ %otherele5, %do.body ]
+ %phi6 = phi i16 [ %ele6, %entry ], [ %otherele6, %do.body ]
+ %phi7 = phi i16 [ %ele7, %entry ], [ %otherele7, %do.body ]
+ %phi8 = phi i16 [ %ele8, %entry ], [ %otherele8, %do.body ]
+ %phi9 = phi i16 [ %ele9, %entry ], [ %otherele9, %do.body ]
+ %phi10 = phi i16 [ %ele10, %entry ], [ %otherele10, %do.body ]
+ %phi11 = phi i16 [ %ele11, %entry ], [ %otherele11, %do.body ]
+ %phi12 = phi i16 [ %ele12, %entry ], [ %otherele12, %do.body ]
+ %phi13 = phi i16 [ %ele13, %entry ], [ %otherele13, %do.body ]
+ %phi14 = phi i16 [ %ele14, %entry ], [ %otherele14, %do.body ]
+ %phi15 = phi i16 [ %ele15, %entry ], [ %otherele15, %do.body ]
+
+ %otherele0 = load i16, ptr addrspace(3) %gep0, align 8
+ %otherele1 = load i16, ptr addrspace(3) %gep1, align 1
+ %otherele2 = load i16, ptr addrspace(3) %gep2, align 2
+ %otherele3 = load i16, ptr addrspace(3) %gep3, align 1
+ %otherele4 = load i16, ptr addrspace(3) %gep4, align 8
+ %otherele5 = load i16, ptr addrspace(3) %gep5, align 1
+ %otherele6 = load i16, ptr addrspace(3) %gep6, align 2
+ %otherele7 = load i16, ptr addrspace(3) %gep7, align 1
+ %otherele8 = load i16, ptr addrspace(3) %gep8, align 8
+ %otherele9 = load i16, ptr addrspace(3) %gep9, align 1
+ %otherele10 = load i16, ptr addrspace(3) %gep10, align 2
+ %otherele11 = load i16, ptr addrspace(3) %gep11, align 1
+ %otherele12 = load i16, ptr addrspace(3) %gep12, align 8
+ %otherele13 = load i16, ptr addrspace(3) %gep13, align 1
+ %otherele14 = load i16, ptr addrspace(3) %gep14, align 2
+ %otherele15 = load i16, ptr addrspace(3) %gep15, align 1
+ %cmp = icmp eq i32 %flag, 0
+ br i1 %cmp, label %exit, label %do.body
+
+exit:
+ %vec00 = insertelement <16 x i16> poison, i16 %otherele0, i64 0
+ %vec01 = insertelement <16 x i16> %vec00, i16 %otherele1, i64 1
+ %vec02 = insertelement <16 x i16> %vec01, i16 %otherele2, i64 2
+ %vec03 = insertelement <16 x i16> %vec02, i16 %otherele3, i64 3
+ %vec04 = insertelement <16 x i16> %vec03, i16 %otherele4, i64 4
+ %vec05 = insertelement <16 x i16> %vec04, i16 %otherele5, i64 5
+ %vec06 = insertelement <16 x i16> %vec05, i16 %otherele6, i64 6
+ %vec07 = insertelement <16 x i16> %vec06, i16 %otherele7, i64 7
+ %vec08 = insertelement <16 x i16> %vec07, i16 %otherele8, i64 8
+ %vec09 = insertelement <16 x i16> %vec08, i16 %otherele9, i64 9
+ %vec010 = insertelement <16 x i16> %vec09, i16 %otherele10, i64 10
+ %vec011 = insertelement <16 x i16> %vec010, i16 %otherele11, i64 11
+ %vec012 = insertelement <16 x i16> %vec011, i16 %otherele12, i64 12
+ %vec013 = insertelement <16 x i16> %vec012, i16 %otherele13, i64 13
+ %vec014 = insertelement <16 x i16> %vec013, i16 %otherele14, i64 14
+ %vec015 = insertelement <16 x i16> %vec014, i16 %otherele15, i64 15
+
+ %vec10 = insertelement <16 x i16> poison, i16 %ele0, i64 0
+ %vec11 = insertelement <16 x i16> %vec10, i16 %ele1, i64 1
+ %vec12 = insertelement <16 x i16> %vec11, i16 %ele2, i64 2
+ %vec13 = insertelement <16 x i16> %vec12, i16 %ele3, i64 3
+ %vec14 = insertelement <16 x i16> %vec13, i16 %ele4, i64 4
+ %vec15 = insertelement <16 x i16> %vec14, i16 %ele5, i64 5
+ %vec16 = insertelement <16 x i16> %vec15, i16 %ele6, i64 6
+ %vec17 = insertelement <16 x i16> %vec16, i16 %ele7, i64 7
+ %vec18 = insertelement <16 x i16> %vec17, i16 %ele8, i64 8
+ %vec19 = insertelement <16 x i16> %vec18, i16 %ele9, i64 9
+ %vec110 = insertelement <16 x i16> %vec19, i16 %ele10, i64 10
+ %vec111 = insertelement <16 x i16> %vec110, i16 %ele11, i64 11
+ %vec112 = insertelement <16 x i16> %vec111, i16 %ele12, i64 12
+ %vec113 = insertelement <16 x i16> %vec112, i16 %ele13, i64 13
+ %vec114 = insertelement <16 x i16> %vec113, i16 %ele14, i64 14
+ %vec115 = insertelement <16 x i16> %vec114, i16 %ele15, i64 15
+
+ %vec20 = insertelement <16 x i16> poison, i16 %phi0, i64 0
+ %vec21 = insertelement <16 x i16> %vec20, i16 %phi1, i64 1
+ %vec22 = insertelement <16 x i16> %vec21, i16 %phi2, i64 2
+ %vec23 = insertelement <16 x i16> %vec22, i16 %phi3, i64 3
+ %vec24 = insertelement <16 x i16> %vec23, i16 %phi4, i64 4
+ %vec25 = insertelement <16 x i16> %vec24, i16 %phi5, i64 5
+ %vec26 = insertelement <16 x i16> %vec25, i16 %phi6, i64 6
+ %vec27 = insertelement <16 x i16> %vec26, i16 %phi7, i64 7
+ %vec28 = insertelement <16 x i16> %vec27, i16 %phi8, i64 8
+ %vec29 = insertelement <16 x i16> %vec28, i16 %phi9, i64 9
+ %vec210 = insertelement <16 x i16> %vec29, i16 %phi10, i64 10
+ %vec211 = insertelement <16 x i16> %vec210, i16 %phi11, i64 11
+ %vec212 = insertelement <16 x i16> %vec211, i16 %phi12, i64 12
+ %vec213 = insertelement <16 x i16> %vec212, i16 %phi13, i64 13
+ %vec214 = insertelement <16 x i16> %vec213, i16 %phi14, i64 14
+ %vec215 = insertelement <16 x i16> %vec214, i16 %phi15, i64 15
+
+
+
+ store <16 x i16> %vec115, ptr %out
+ store <16 x i16> %vec015, ptr %out1
+ store <16 x i16> %vec215, ptr %out2
+
+ ret void
+}
>From 87a2eb04a06784659993e2bf89ff40eab0207cc7 Mon Sep 17 00:00:00 2001
From: Jeffrey Byrnes <Jeffrey.Byrnes at amd.com>
Date: Wed, 11 Jun 2025 13:21:22 -0700
Subject: [PATCH 2/2] Review comments
Change-Id: I461e92334b4ce0a23d093413204d4694075f3fb3
---
.../SLPVectorizer/AMDGPU/external-shuffle.ll | 19 +++++++++----------
1 file changed, 9 insertions(+), 10 deletions(-)
diff --git a/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll b/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll
index b5749f7d098f2..ce9e47a03dee3 100644
--- a/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll
+++ b/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll
@@ -1,12 +1,14 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=slp-vectorizer,dce < %s | FileCheck -check-prefixes=GCN %s
-define protected void @phi_4(ptr addrspace(3) %inptr0, ptr addrspace(3) %inptr1, ptr %out, ptr %out1, ptr %out2, i32 %flag) {
-; GCN-LABEL: define protected void @phi_4(
+; The insertelements in the exit block use the various parts of the vectorized tree. These external uses are just creating an identity vector using a sequence
+; of insert elements. Since these insertelements are just recreating the same vectors that were produced during vectorization, they should not increase the cost of vectorization.
+
+define void @phi_4(ptr addrspace(3) %inptr0, ptr addrspace(3) %inptr1, ptr %out, ptr %out1, ptr %out2, i32 %flag) {
+; GCN-LABEL: define void @phi_4(
; GCN-SAME: ptr addrspace(3) [[INPTR0:%.*]], ptr addrspace(3) [[INPTR1:%.*]], ptr [[OUT:%.*]], ptr [[OUT1:%.*]], ptr [[OUT2:%.*]], i32 [[FLAG:%.*]]) #[[ATTR0:[0-9]+]] {
; GCN-NEXT: [[ENTRY:.*]]:
-; GCN-NEXT: [[GEP0:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 0
-; GCN-NEXT: [[TMP0:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP0]], align 8
+; GCN-NEXT: [[TMP0:%.*]] = load <2 x i16>, ptr addrspace(3) [[INPTR0]], align 8
; GCN-NEXT: [[GEP2:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 2
; GCN-NEXT: [[GEP3:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 3
; GCN-NEXT: [[TMP1:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP2]], align 2
@@ -55,7 +57,7 @@ define protected void @phi_4(ptr addrspace(3) %inptr0, ptr addrspace(3) %inptr1
; GCN-NEXT: [[PHI13:%.*]] = phi i16 [ [[TMP38]], %[[ENTRY]] ], [ [[OTHERELE13:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP41:%.*]] = phi <2 x i16> [ [[TMP0]], %[[ENTRY]] ], [ [[TMP16:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP42:%.*]] = phi <2 x i16> [ [[TMP7]], %[[ENTRY]] ], [ [[TMP23:%.*]], %[[DO_BODY]] ]
-; GCN-NEXT: [[TMP16]] = load <2 x i16>, ptr addrspace(3) [[GEP0]], align 8
+; GCN-NEXT: [[TMP16]] = load <2 x i16>, ptr addrspace(3) [[INPTR0]], align 8
; GCN-NEXT: [[OTHERELE3]] = load i16, ptr addrspace(3) [[GEP3]], align 1
; GCN-NEXT: [[TMP17:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP2]], align 2
; GCN-NEXT: [[OTHERELE5]] = load i16, ptr addrspace(3) [[GEP5]], align 1
@@ -129,8 +131,7 @@ define protected void @phi_4(ptr addrspace(3) %inptr0, ptr addrspace(3) %inptr1
; GCN-NEXT: ret void
;
entry:
- %gep0 = getelementptr i16, ptr addrspace(3) %inptr0, i32 0
- %ele0 = load i16, ptr addrspace(3) %gep0, align 8
+ %ele0 = load i16, ptr addrspace(3) %inptr0, align 8
%gep1 = getelementptr i16, ptr addrspace(3) %inptr0, i32 1
%ele1 = load i16, ptr addrspace(3) %gep1, align 1
%gep2 = getelementptr i16, ptr addrspace(3) %inptr0, i32 2
@@ -181,7 +182,7 @@ do.body:
%phi14 = phi i16 [ %ele14, %entry ], [ %otherele14, %do.body ]
%phi15 = phi i16 [ %ele15, %entry ], [ %otherele15, %do.body ]
- %otherele0 = load i16, ptr addrspace(3) %gep0, align 8
+ %otherele0 = load i16, ptr addrspace(3) %inptr0, align 8
%otherele1 = load i16, ptr addrspace(3) %gep1, align 1
%otherele2 = load i16, ptr addrspace(3) %gep2, align 2
%otherele3 = load i16, ptr addrspace(3) %gep3, align 1
@@ -252,8 +253,6 @@ exit:
%vec214 = insertelement <16 x i16> %vec213, i16 %phi14, i64 14
%vec215 = insertelement <16 x i16> %vec214, i16 %phi15, i64 15
-
-
store <16 x i16> %vec115, ptr %out
store <16 x i16> %vec015, ptr %out1
store <16 x i16> %vec215, ptr %out2
More information about the llvm-commits
mailing list