[SLP]Improve minbitwidth analysis for trun'ed gather nodes. #99072
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Created using spr 1.3.5
|
@llvm/pr-subscribers-llvm-transforms Author: Alexey Bataev (alexey-bataev) ChangesIf the gather node is trunc'ed, better to trunc scalars and then gather Full diff: https://github.com/llvm/llvm-project/pull/99072.diff 6 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 7b981bead6bb8..722590a840a54 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -15502,8 +15502,21 @@ void BoUpSLP::computeMinimumValueSizes() {
auto ComputeMaxBitWidth = [&](const TreeEntry &E, bool IsTopRoot,
bool IsProfitableToDemoteRoot, unsigned Opcode,
unsigned Limit, bool IsTruncRoot,
- bool IsSignedCmp) {
+ bool IsSignedCmp) -> unsigned {
ToDemote.clear();
+ // Check if the root is trunc and the next node is gather/buildvector, then
+ // keep trunc in scalars, which is free in most cases.
+ if (E.isGather() && IsTruncRoot && E.UserTreeIndices.size() == 1 &&
+ E.Idx > (IsStoreOrInsertElt ? 2 : 1)) {
+ ToDemote.push_back(E.Idx);
+ const TreeEntry *UserTE = E.UserTreeIndices.back().UserTE;
+ auto It = MinBWs.find(UserTE);
+ if (It != MinBWs.end())
+ return It->second.first;
+ return DL->getTypeSizeInBits(
+ E.UserTreeIndices.back().UserTE->Scalars.front()->getType());
+ }
+
unsigned VF = E.getVectorFactor();
auto *TreeRootIT = dyn_cast<IntegerType>(E.Scalars.front()->getType());
if (!TreeRootIT || !Opcode)
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/int-bitcast-minbitwidth.ll b/llvm/test/Transforms/SLPVectorizer/X86/int-bitcast-minbitwidth.ll
index 789d73947d1c7..97e505f4319c6 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/int-bitcast-minbitwidth.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/int-bitcast-minbitwidth.ll
@@ -5,9 +5,9 @@ define void @t(i64 %v) {
; CHECK-LABEL: define void @t(
; CHECK-SAME: i64 [[V:%.*]]) {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i64> poison, i64 [[V]], i32 0
-; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i64> [[TMP0]], <4 x i64> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP2:%.*]] = trunc <4 x i64> [[TMP1]] to <4 x i16>
+; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[V]] to i16
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i16> poison, i16 [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: [[TMP3:%.*]] = mul <4 x i16> [[TMP2]], <i16 2, i16 3, i16 6, i16 5>
; CHECK-NEXT: [[TMP4:%.*]] = call i16 @llvm.vector.reduce.or.v4i16(<4 x i16> [[TMP3]])
; CHECK-NEXT: [[TMP5:%.*]] = sext i16 [[TMP4]] to i32
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-transformed-operand.ll b/llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-transformed-operand.ll
index 032625a1199f9..57b5d2af48ee6 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-transformed-operand.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-transformed-operand.ll
@@ -5,20 +5,16 @@ define void @test(i64 %d.promoted.i) {
; CHECK-LABEL: define void @test(
; CHECK-SAME: i64 [[D_PROMOTED_I:%.*]]) {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[AND_1_I:%.*]] = and i64 0, [[D_PROMOTED_I]]
-; CHECK-NEXT: [[TMP0:%.*]] = insertelement <8 x i64> <i64 poison, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0>, i64 [[AND_1_I]], i32 0
-; CHECK-NEXT: [[TMP1:%.*]] = trunc <8 x i64> [[TMP0]] to <8 x i1>
-; CHECK-NEXT: [[TMP2:%.*]] = mul <8 x i1> [[TMP1]], zeroinitializer
-; CHECK-NEXT: [[AND_1_I_1:%.*]] = and i64 0, 0
-; CHECK-NEXT: [[TMP3:%.*]] = insertelement <8 x i64> <i64 poison, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0>, i64 [[AND_1_I_1]], i32 0
-; CHECK-NEXT: [[TMP4:%.*]] = trunc <8 x i64> [[TMP3]] to <8 x i1>
-; CHECK-NEXT: [[TMP5:%.*]] = mul <8 x i1> [[TMP4]], zeroinitializer
-; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v8i1(<8 x i1> [[TMP2]])
-; CHECK-NEXT: [[TMP7:%.*]] = zext i1 [[TMP6]] to i32
-; CHECK-NEXT: [[TMP8:%.*]] = call i1 @llvm.vector.reduce.or.v8i1(<8 x i1> [[TMP5]])
+; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i64> <i64 poison, i64 0>, i64 0, i32 0
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i64> <i64 poison, i64 0>, i64 [[D_PROMOTED_I]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = and <2 x i64> [[TMP0]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = trunc <2 x i64> [[TMP2]] to <2 x i1>
+; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <2 x i1> [[TMP3]], <2 x i1> poison, <16 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <16 x i1> <i1 poison, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 poison, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>, <16 x i1> [[TMP4]], <16 x i32> <i32 16, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 24, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT: [[TMP6:%.*]] = mul <16 x i1> [[TMP5]], zeroinitializer
+; CHECK-NEXT: [[TMP8:%.*]] = call i1 @llvm.vector.reduce.or.v16i1(<16 x i1> [[TMP6]])
; CHECK-NEXT: [[TMP9:%.*]] = zext i1 [[TMP8]] to i32
-; CHECK-NEXT: [[OP_RDX:%.*]] = or i32 [[TMP7]], [[TMP9]]
-; CHECK-NEXT: [[TMP10:%.*]] = and i32 [[OP_RDX]], 0
+; CHECK-NEXT: [[TMP10:%.*]] = and i32 [[TMP9]], 0
; CHECK-NEXT: store i32 [[TMP10]], ptr null, align 4
; CHECK-NEXT: ret void
;
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/resched.ll b/llvm/test/Transforms/SLPVectorizer/X86/resched.ll
index b7237cbb02bb3..4ed52247c2ef3 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/resched.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/resched.ll
@@ -11,26 +11,31 @@ define fastcc void @_ZN12_GLOBAL__N_127PolynomialMultiplyRecognize9recognizeEv()
; CHECK: if.then22.i:
; CHECK-NEXT: [[SUB_I:%.*]] = add nsw i32 undef, -1
; CHECK-NEXT: [[CONV31_I:%.*]] = and i32 undef, [[SUB_I]]
-; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> poison, i32 [[CONV31_I]], i32 0
-; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP2:%.*]] = lshr <4 x i32> [[TMP1]], <i32 1, i32 2, i32 3, i32 4>
-; CHECK-NEXT: [[SHR_4_I_I:%.*]] = lshr i32 [[CONV31_I]], 5
-; CHECK-NEXT: [[SHR_5_I_I:%.*]] = lshr i32 [[CONV31_I]], 6
-; CHECK-NEXT: [[SHR_6_I_I:%.*]] = lshr i32 [[CONV31_I]], 7
-; CHECK-NEXT: [[TMP3:%.*]] = insertelement <8 x i32> poison, i32 [[CONV31_I]], i32 0
-; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <8 x i32> [[TMP3]], <8 x i32> poison, <8 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP5:%.*]] = lshr <8 x i32> [[TMP4]], <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP6:%.*]] = insertelement <16 x i32> poison, i32 [[SUB_I]], i32 0
-; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <16 x i32> [[TMP6]], <16 x i32> [[TMP7]], <16 x i32> <i32 0, i32 16, i32 17, i32 18, i32 19, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT: [[TMP9:%.*]] = insertelement <16 x i32> [[TMP8]], i32 [[SHR_4_I_I]], i32 5
-; CHECK-NEXT: [[TMP10:%.*]] = insertelement <16 x i32> [[TMP9]], i32 [[SHR_5_I_I]], i32 6
-; CHECK-NEXT: [[TMP11:%.*]] = insertelement <16 x i32> [[TMP10]], i32 [[SHR_6_I_I]], i32 7
-; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <8 x i32> [[TMP5]], <8 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <16 x i32> [[TMP11]], <16 x i32> [[TMP12]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
-; CHECK-NEXT: [[TMP14:%.*]] = trunc <16 x i32> [[TMP13]] to <16 x i8>
-; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i8> [[TMP14]], <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
-; CHECK-NEXT: store <16 x i8> [[TMP15]], ptr undef, align 1
+; CHECK-NEXT: [[SHR_I_I:%.*]] = lshr i32 [[CONV31_I]], 1
+; CHECK-NEXT: [[SHR_1_I_I:%.*]] = lshr i32 [[CONV31_I]], 2
+; CHECK-NEXT: [[SHR_2_I_I:%.*]] = lshr i32 [[CONV31_I]], 3
+; CHECK-NEXT: [[TMP0:%.*]] = trunc i32 [[SUB_I]] to i8
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <16 x i8> poison, i8 [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = trunc i32 [[SHR_I_I]] to i8
+; CHECK-NEXT: [[TMP3:%.*]] = insertelement <16 x i8> [[TMP1]], i8 [[TMP2]], i32 1
+; CHECK-NEXT: [[TMP4:%.*]] = trunc i32 [[SHR_1_I_I]] to i8
+; CHECK-NEXT: [[TMP5:%.*]] = insertelement <16 x i8> [[TMP3]], i8 [[TMP4]], i32 2
+; CHECK-NEXT: [[TMP6:%.*]] = trunc i32 [[SHR_2_I_I]] to i8
+; CHECK-NEXT: [[TMP7:%.*]] = insertelement <16 x i8> [[TMP5]], i8 [[TMP6]], i32 3
+; CHECK-NEXT: [[TMP8:%.*]] = insertelement <4 x i32> poison, i32 [[CONV31_I]], i32 0
+; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <4 x i32> [[TMP8]], <4 x i32> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP10:%.*]] = lshr <4 x i32> [[TMP9]], <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT: [[TMP11:%.*]] = trunc <4 x i32> [[TMP10]] to <4 x i8>
+; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <4 x i8> [[TMP11]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <16 x i8> [[TMP7]], <16 x i8> [[TMP12]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP14:%.*]] = insertelement <8 x i32> poison, i32 [[CONV31_I]], i32 0
+; CHECK-NEXT: [[TMP15:%.*]] = shufflevector <8 x i32> [[TMP14]], <8 x i32> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP16:%.*]] = lshr <8 x i32> [[TMP15]], <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT: [[TMP17:%.*]] = trunc <8 x i32> [[TMP16]] to <8 x i8>
+; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <8 x i8> [[TMP17]], <8 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP19:%.*]] = shufflevector <16 x i8> [[TMP13]], <16 x i8> [[TMP18]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT: [[TMP20:%.*]] = and <16 x i8> [[TMP19]], <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
+; CHECK-NEXT: store <16 x i8> [[TMP20]], ptr undef, align 1
; CHECK-NEXT: unreachable
; CHECK: if.end50.i:
; CHECK-NEXT: ret void
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/shuffle-multivector.ll b/llvm/test/Transforms/SLPVectorizer/X86/shuffle-multivector.ll
index 143052a3d9cd0..c2555889f5981 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/shuffle-multivector.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/shuffle-multivector.ll
@@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -passes=slp-vectorizer -S < %s -mtriple=x86_64-unknown-linux -slp-threshold=-160 | FileCheck %s
+; RUN: opt -passes=slp-vectorizer -S < %s -mtriple=x86_64-unknown-linux -slp-threshold=-163 | FileCheck %s
define void @test1(i128 %p0, i128 %p1, i128 %p2, i128 %p3, <4 x i128> %vec) {
; CHECK-LABEL: @test1(
@@ -14,13 +14,14 @@ define void @test1(i128 %p0, i128 %p1, i128 %p2, i128 %p3, <4 x i128> %vec) {
; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
; CHECK-NEXT: [[T5:%.*]] = trunc i128 [[P1]] to i32
; CHECK-NEXT: [[TMP8:%.*]] = sdiv <4 x i32> [[TMP3]], [[TMP7]]
-; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <2 x i128> [[TMP1]], <2 x i128> [[TMP5]], <4 x i32> <i32 poison, i32 0, i32 3, i32 2>
-; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <4 x i128> [[VEC:%.*]], <4 x i128> [[TMP9]], <4 x i32> <i32 0, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP11:%.*]] = trunc <4 x i128> [[TMP10]] to <4 x i32>
-; CHECK-NEXT: [[TMP12:%.*]] = sdiv <4 x i32> [[TMP8]], [[TMP11]]
+; CHECK-NEXT: [[TMP9:%.*]] = trunc <4 x i128> [[VEC:%.*]] to <4 x i32>
+; CHECK-NEXT: [[TMP10:%.*]] = trunc <4 x i128> [[VEC]] to <4 x i32>
+; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <2 x i32> [[TMP2]], <2 x i32> [[TMP6]], <4 x i32> <i32 poison, i32 0, i32 3, i32 2>
+; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <4 x i32> [[TMP10]], <4 x i32> [[TMP11]], <4 x i32> <i32 0, i32 5, i32 6, i32 7>
+; CHECK-NEXT: [[TMP13:%.*]] = sdiv <4 x i32> [[TMP8]], [[TMP12]]
; CHECK-NEXT: br label [[BB:%.*]]
; CHECK: bb:
-; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x i32> [ [[TMP12]], [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[TMP14:%.*]] = phi <4 x i32> [ [[TMP13]], [[ENTRY:%.*]] ]
; CHECK-NEXT: ret void
;
entry:
diff --git a/llvm/test/Transforms/SLPVectorizer/orig-btiwidth-les-projected.ll b/llvm/test/Transforms/SLPVectorizer/orig-btiwidth-les-projected.ll
index 531e964053482..88503aeb6071f 100644
--- a/llvm/test/Transforms/SLPVectorizer/orig-btiwidth-les-projected.ll
+++ b/llvm/test/Transforms/SLPVectorizer/orig-btiwidth-les-projected.ll
@@ -5,10 +5,10 @@ define i32 @test(i4 %0) {
; CHECK-LABEL: define i32 @test(
; CHECK-SAME: i4 [[TMP0:%.*]]) {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[TMP1:%.*]] = trunc i8 0 to i4
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i8 0 to i4
-; CHECK-NEXT: [[ADD_R:%.*]] = or i4 [[TMP1]], [[TMP0]]
-; CHECK-NEXT: [[ADD_R14:%.*]] = or i4 0, [[TMP2]]
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i4> <i4 poison, i4 0>, i4 [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = or <2 x i4> zeroinitializer, [[TMP1]]
+; CHECK-NEXT: [[ADD_R:%.*]] = extractelement <2 x i4> [[TMP2]], i32 0
+; CHECK-NEXT: [[ADD_R14:%.*]] = extractelement <2 x i4> [[TMP2]], i32 1
; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i4 [[ADD_R]], [[ADD_R14]]
; CHECK-NEXT: ret i32 0
;
|
alexey-bataev
deleted the
users/alexey-bataev/spr/slpimprove-minbitwidth-analysis-for-truned-gather-nodes
branch
alexey-bataev
added a commit
that referenced
this pull request
Jul 17, 2024
If the gather node is trunc'ed, better to trunc scalars and then gather them rather than gather and then trunc. Trunc for scalars is free in most cases. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: #99072
alexey-bataev
added a commit
that referenced
this pull request
Jul 17, 2024
If the gather node is trunc'ed, better to trunc scalars and then gather them rather than gather and then trunc. Trunc for scalars is free in most cases. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: #99072
|
I was able to show that this PRs latest commit caused a llvm-test-suite regression on the clang-ppc64le-rhel: build no. 623 and clang-ppc64le-linux-test-suite: build no. 1412 by reproducing build no. 1412's build steps on a branch and then reverting c5c1bd1 and re-testing. I'm sorry for the inconvenience but can you please resolve this error as soon as possible or revert if it will take too long to fix so we can bring the llvm-test-suite back to green status. If I can assist with reproducing the error please let me know. |
|
Do we have to account for the costs of the two different gather types? |
Can you provide a reproducer? I cannot reproduce the issue |
build no. 1412 does not do anything special outside of build the llvm-project and test the c/c++ compilers on lllvm-test-suite, if you were able to run the affected test case on your platform using the bot's cmake configuration and it passed then it is likely a platform specific issue, in this case one of us will have to revert and investigate in the morning (I am EST zone). I can look into providing PowerPC machine access at that point. the llvm-project build command was: cmake-3.28.2/bin/cmake -DLLVM_ENABLE_ASSERTIONS=ON -DCMAKE_BUILD_TYPE=Release -DLLVM_LIT_ARGS=-v -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache '-DLLVM_ENABLE_PROJECTS=clang-tools-extra;llvm;clang' -DLLVM_ENABLE_RUNTIMES=compiler-rt -GNinja ../llvm-project/llvm && ninja -j 128and the test-suite build command was: cmake-3.28.2/bin/cmake -G Ninja -DCMAKE_C_COMPILER=/home/kamaub/Github/llvm/mono2/test-suite-reproducer/bin/clang -DCMAKE_CXX_COMPILER=/home/kamaub/Github/llvm/mono2/test-suite-reproducer/bin/clang++ -DTEST_SUITE_LIT:FILEPATH=/home/kamaub/Github/llvm/mono2/test-suite-reproducer/bin/llvm-lit /home/kamaub/Github/llvm/test-suite && ninja -j 128 && /home/kamaub/Github/llvm/mono2/test-suite-reproducer/bin/llvm-lit -j 128 -v .I do not believe you're required to run the entire suite as |
It is ppc specific issue, no other targets have reported the issue with the patch. That's why I ask for the PPC reproducer |
Were you able to prepare a reproducer? Preprocessed source code and compile line should be enough for initial analaysis |
|
I bisected a miscompilation (giving the wrong result at runtime) back to this commit, reproducing on x86_64 Linux. To repro: Note that the issue only appears in an older version of ffmpeg, it doesn’t seem to happen with latest git. The miscompilation seems to happen in the |
|
Must be fixed in f6e01b9 |
Thanks, that does indeed seem to fix my issue! |
sgundapa
pushed a commit
to sgundapa/upstream_effort
that referenced
this pull request
Jul 23, 2024
If the gather node is trunc'ed, better to trunc scalars and then gather them rather than gather and then trunc. Trunc for scalars is free in most cases. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: llvm#99072
sgundapa
pushed a commit
to sgundapa/upstream_effort
that referenced
this pull request
Jul 23, 2024
If the gather node is trunc'ed, better to trunc scalars and then gather them rather than gather and then trunc. Trunc for scalars is free in most cases. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: llvm#99072
sgundapa
pushed a commit
to sgundapa/upstream_effort
that referenced
this pull request
Jul 23, 2024
If the gather node is trunc'ed, better to trunc scalars and then gather them rather than gather and then trunc. Trunc for scalars is free in most cases. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: llvm#99072
yuxuanchen1997
pushed a commit
that referenced
this pull request
Jul 25, 2024
If the gather node is trunc'ed, better to trunc scalars and then gather them rather than gather and then trunc. Trunc for scalars is free in most cases. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: #99072
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
If the gather node is trunc'ed, better to trunc scalars and then gather
them rather than gather and then trunc. Trunc for scalars is free in
most cases.