Skip to content

[SLP]Add debug counter support #110734

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
Merged
merged 1 commit into from
Oct 1, 2024
Merged

[SLP]Add debug counter support #110734

merged 1 commit into from
Oct 1, 2024

Conversation

alexey-bataev
Copy link
Member

@alexey-bataev alexey-bataev commented Oct 1, 2024 •
edited
Loading

Fixes #110725

@alexey-bataev
[𝘀𝗽𝗿] initial version
6f23e91 
Created using spr 1.3.5
@alexey-bataev alexey-bataev requested a review from aeubanks October 1, 2024 20:10
@llvmbot
Copy link
Member

llvmbot commented Oct 1, 2024

@llvm/pr-subscribers-llvm-transforms

Author: Alexey Bataev (alexey-bataev)

Changes

Patch is 20.09 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/110734.diff

2 Files Affected:

  • (modified) llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp (+7)
  • (added) llvm/test/Transforms/SLPVectorizer/debug-counter.ll (+235)
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 15e798bd6c98f9..b68dfbaa6c3075 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -79,6 +79,7 @@
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/DOTGraphTraits.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/DebugCounter.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/GraphWriter.h"
 #include "llvm/Support/InstructionCost.h"
@@ -109,6 +110,9 @@ using namespace slpvectorizer;
 
 STATISTIC(NumVectorInstructions, "Number of vector instructions generated");
 
+DEBUG_COUNTER(VectorizedGraphs, "slp-vectorized",
+              "Controls which SLP graphs should be vectorized.");
+
 static cl::opt<bool>
     RunSLPVectorization("vectorize-slp", cl::init(true), cl::Hidden,
                         cl::desc("Run the SLP vectorization passes"));
@@ -11175,6 +11179,9 @@ bool BoUpSLP::isLoadCombineCandidate(ArrayRef<Value *> Stores) const {
 }
 
 bool BoUpSLP::isTreeTinyAndNotFullyVectorizable(bool ForReduction) const {
+  if (!DebugCounter::shouldExecute(VectorizedGraphs))
+    return true;
+
   // No need to vectorize inserts of gathered values.
   if (VectorizableTree.size() == 2 &&
       isa<InsertElementInst>(VectorizableTree[0]->Scalars[0]) &&
diff --git a/llvm/test/Transforms/SLPVectorizer/debug-counter.ll b/llvm/test/Transforms/SLPVectorizer/debug-counter.ll
new file mode 100644
index 00000000000000..006b6ef33964e5
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/debug-counter.ll
@@ -0,0 +1,235 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt -S -passes=slp-vectorizer -debug-counter=slp-vectorized=0 -slp-threshold=-99999 < %s | FileCheck %s --check-prefix=COUNT0
+; RUN: opt -S -passes=slp-vectorizer -debug-counter=slp-vectorized=1 -slp-threshold=-99999 < %s | FileCheck %s --check-prefix=COUNT1
+; RUN: opt -S -passes=slp-vectorizer -debug-counter=slp-vectorized=2 -slp-threshold=-99999 < %s | FileCheck %s --check-prefix=COUNT2
+; RUN: opt -S -passes=slp-vectorizer -debug-counter=slp-vectorized=0-1 -slp-threshold=-99999 < %s | FileCheck %s --check-prefix=COUNT-1
+
+define void @blam(ptr %arg, double %load2, i1 %fcmp3) {
+; CHECK-LABEL: define void @blam
+; CHECK-SAME: (ptr [[ARG:%.*]], double [[LOAD2:%.*]], i1 [[FCMP3:%.*]]) {
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    [[GETELEMENTPTR13:%.*]] = getelementptr double, ptr [[ARG]], i64 3
+; CHECK-NEXT:    [[TMP0:%.*]] = load <2 x double>, ptr [[ARG]], align 8
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i1> poison, i1 [[FCMP3]], i32 0
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i1> [[TMP1]], <2 x i1> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x double> zeroinitializer, <2 x double> [[TMP0]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x double> [[TMP0]], double [[LOAD2]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = fcmp olt <2 x double> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = select <2 x i1> [[TMP5]], <2 x double> zeroinitializer, <2 x double> [[TMP0]]
+; CHECK-NEXT:    [[TMP7:%.*]] = fcmp olt <2 x double> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = select <2 x i1> [[TMP7]], <2 x double> <double 0.000000e+00, double 1.000000e+00>, <2 x double> <double 1.000000e+00, double 0.000000e+00>
+; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <2 x double> [[TMP8]], <2 x double> poison, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT:    [[TMP10:%.*]] = fcmp olt <2 x double> [[TMP9]], [[TMP6]]
+; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <2 x double> [[TMP4]], <2 x double> <double poison, double 0.000000e+00>, <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <2 x double> [[TMP4]], <2 x double> <double 0.000000e+00, double poison>, <2 x i32> <i32 2, i32 0>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <2 x i1> [[TMP10]], <2 x double> [[TMP11]], <2 x double> [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = fcmp olt <2 x double> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = select <2 x i1> [[TMP14]], <2 x double> zeroinitializer, <2 x double> <double 1.000000e+00, double 1.000000e+00>
+; CHECK-NEXT:    [[TMP16:%.*]] = fcmp ogt <2 x double> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = shufflevector <2 x double> [[TMP4]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = select <2 x i1> [[TMP16]], <2 x double> zeroinitializer, <2 x double> [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = fcmp olt <2 x double> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = select <2 x i1> [[TMP19]], <2 x double> <double 1.000000e+00, double 1.000000e+00>, <2 x double> zeroinitializer
+; CHECK-NEXT:    store <2 x double> [[TMP20]], ptr [[GETELEMENTPTR13]], align 8
+; CHECK-NEXT:    ret void
+;
+; COUNT0-LABEL: define void @blam
+; COUNT0-SAME: (ptr [[ARG:%.*]], double [[LOAD2:%.*]], i1 [[FCMP3:%.*]]) {
+; COUNT0-NEXT:  bb:
+; COUNT0-NEXT:    [[GETELEMENTPTR:%.*]] = getelementptr double, ptr [[ARG]], i64 1
+; COUNT0-NEXT:    [[LOAD:%.*]] = load double, ptr [[GETELEMENTPTR]], align 8
+; COUNT0-NEXT:    [[FCMP:%.*]] = fcmp olt double [[LOAD]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT3:%.*]] = select i1 [[FCMP]], double 0.000000e+00, double [[LOAD]]
+; COUNT0-NEXT:    [[SELECT4:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD]]
+; COUNT0-NEXT:    [[LOAD7:%.*]] = load double, ptr [[ARG]], align 8
+; COUNT0-NEXT:    [[SELECT10:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD7]]
+; COUNT0-NEXT:    [[GETELEMENTPTR13:%.*]] = getelementptr double, ptr [[ARG]], i64 3
+; COUNT0-NEXT:    [[GETELEMENTPTR21:%.*]] = getelementptr double, ptr [[ARG]], i64 4
+; COUNT0-NEXT:    [[FCMP23:%.*]] = fcmp olt double [[SELECT10]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT24:%.*]] = select i1 [[FCMP23]], double 0.000000e+00, double 1.000000e+00
+; COUNT0-NEXT:    [[TMP0:%.*]] = insertelement <2 x double> poison, double [[LOAD2]], i32 1
+; COUNT0-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> [[TMP0]], double [[SELECT4]], i32 0
+; COUNT0-NEXT:    [[TMP2:%.*]] = fcmp olt <2 x double> [[TMP1]], zeroinitializer
+; COUNT0-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> <double 0.000000e+00, double poison>, double [[LOAD7]], i32 1
+; COUNT0-NEXT:    [[TMP4:%.*]] = select <2 x i1> [[TMP2]], <2 x double> <double 1.000000e+00, double 0.000000e+00>, <2 x double> [[TMP3]]
+; COUNT0-NEXT:    [[FCMP33:%.*]] = fcmp olt double [[SELECT24]], [[SELECT3]]
+; COUNT0-NEXT:    [[SELECT34:%.*]] = select i1 [[FCMP33]], double 0.000000e+00, double [[LOAD2]]
+; COUNT0-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 0
+; COUNT0-NEXT:    [[TMP6:%.*]] = extractelement <2 x double> [[TMP4]], i32 1
+; COUNT0-NEXT:    [[FCMP39:%.*]] = fcmp olt double [[TMP5]], [[TMP6]]
+; COUNT0-NEXT:    [[SELECT40:%.*]] = select i1 [[FCMP39]], double [[LOAD2]], double 0.000000e+00
+; COUNT0-NEXT:    [[FCMP62:%.*]] = fcmp olt double [[SELECT34]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT639:%.*]] = select i1 [[FCMP62]], double 0.000000e+00, double 1.000000e+00
+; COUNT0-NEXT:    [[FCMP76:%.*]] = fcmp olt double [[SELECT40]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT77:%.*]] = select i1 [[FCMP76]], double 0.000000e+00, double 1.000000e+00
+; COUNT0-NEXT:    [[FCMP90:%.*]] = fcmp ogt double [[SELECT639]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT91:%.*]] = select i1 [[FCMP90]], double 0.000000e+00, double [[LOAD2]]
+; COUNT0-NEXT:    [[FCMP92:%.*]] = fcmp ogt double [[SELECT77]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT93:%.*]] = select i1 [[FCMP92]], double 0.000000e+00, double [[LOAD2]]
+; COUNT0-NEXT:    [[FCMP108:%.*]] = fcmp olt double [[SELECT93]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT109:%.*]] = select i1 [[FCMP108]], double 1.000000e+00, double 0.000000e+00
+; COUNT0-NEXT:    [[FCMP110:%.*]] = fcmp olt double [[SELECT91]], 0.000000e+00
+; COUNT0-NEXT:    [[SELECT111:%.*]] = select i1 [[FCMP110]], double 1.000000e+00, double 0.000000e+00
+; COUNT0-NEXT:    store double [[SELECT111]], ptr [[GETELEMENTPTR21]], align 8
+; COUNT0-NEXT:    store double [[SELECT109]], ptr [[GETELEMENTPTR13]], align 8
+; COUNT0-NEXT:    ret void
+;
+; COUNT1-LABEL: define void @blam
+; COUNT1-SAME: (ptr [[ARG:%.*]], double [[LOAD2:%.*]], i1 [[FCMP3:%.*]]) {
+; COUNT1-NEXT:  bb:
+; COUNT1-NEXT:    [[GETELEMENTPTR:%.*]] = getelementptr double, ptr [[ARG]], i64 1
+; COUNT1-NEXT:    [[LOAD:%.*]] = load double, ptr [[GETELEMENTPTR]], align 8
+; COUNT1-NEXT:    [[SELECT4:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD]]
+; COUNT1-NEXT:    [[LOAD7:%.*]] = load double, ptr [[ARG]], align 8
+; COUNT1-NEXT:    [[SELECT10:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD7]]
+; COUNT1-NEXT:    [[FCMP11:%.*]] = fcmp olt double [[LOAD2]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT128:%.*]] = select i1 [[FCMP11]], double 0.000000e+00, double [[LOAD7]]
+; COUNT1-NEXT:    [[GETELEMENTPTR13:%.*]] = getelementptr double, ptr [[ARG]], i64 3
+; COUNT1-NEXT:    [[GETELEMENTPTR21:%.*]] = getelementptr double, ptr [[ARG]], i64 4
+; COUNT1-NEXT:    [[TMP0:%.*]] = insertelement <2 x double> poison, double [[SELECT10]], i32 0
+; COUNT1-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> [[TMP0]], double [[LOAD]], i32 1
+; COUNT1-NEXT:    [[TMP2:%.*]] = fcmp olt <2 x double> [[TMP1]], zeroinitializer
+; COUNT1-NEXT:    [[TMP3:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> <double 1.000000e+00, double poison>, <2 x i32> <i32 2, i32 1>
+; COUNT1-NEXT:    [[TMP4:%.*]] = select <2 x i1> [[TMP2]], <2 x double> zeroinitializer, <2 x double> [[TMP3]]
+; COUNT1-NEXT:    [[FCMP29:%.*]] = fcmp olt double [[SELECT4]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT30:%.*]] = select i1 [[FCMP29]], double 1.000000e+00, double 0.000000e+00
+; COUNT1-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 0
+; COUNT1-NEXT:    [[TMP6:%.*]] = extractelement <2 x double> [[TMP4]], i32 1
+; COUNT1-NEXT:    [[FCMP33:%.*]] = fcmp olt double [[TMP5]], [[TMP6]]
+; COUNT1-NEXT:    [[SELECT34:%.*]] = select i1 [[FCMP33]], double 0.000000e+00, double [[LOAD2]]
+; COUNT1-NEXT:    [[FCMP39:%.*]] = fcmp olt double [[SELECT30]], [[SELECT128]]
+; COUNT1-NEXT:    [[SELECT40:%.*]] = select i1 [[FCMP39]], double [[LOAD2]], double 0.000000e+00
+; COUNT1-NEXT:    [[FCMP62:%.*]] = fcmp olt double [[SELECT34]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT639:%.*]] = select i1 [[FCMP62]], double 0.000000e+00, double 1.000000e+00
+; COUNT1-NEXT:    [[FCMP76:%.*]] = fcmp olt double [[SELECT40]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT77:%.*]] = select i1 [[FCMP76]], double 0.000000e+00, double 1.000000e+00
+; COUNT1-NEXT:    [[FCMP90:%.*]] = fcmp ogt double [[SELECT639]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT91:%.*]] = select i1 [[FCMP90]], double 0.000000e+00, double [[LOAD2]]
+; COUNT1-NEXT:    [[FCMP92:%.*]] = fcmp ogt double [[SELECT77]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT93:%.*]] = select i1 [[FCMP92]], double 0.000000e+00, double [[LOAD2]]
+; COUNT1-NEXT:    [[FCMP108:%.*]] = fcmp olt double [[SELECT93]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT109:%.*]] = select i1 [[FCMP108]], double 1.000000e+00, double 0.000000e+00
+; COUNT1-NEXT:    [[FCMP110:%.*]] = fcmp olt double [[SELECT91]], 0.000000e+00
+; COUNT1-NEXT:    [[SELECT111:%.*]] = select i1 [[FCMP110]], double 1.000000e+00, double 0.000000e+00
+; COUNT1-NEXT:    store double [[SELECT111]], ptr [[GETELEMENTPTR21]], align 8
+; COUNT1-NEXT:    store double [[SELECT109]], ptr [[GETELEMENTPTR13]], align 8
+; COUNT1-NEXT:    ret void
+;
+; COUNT2-LABEL: define void @blam
+; COUNT2-SAME: (ptr [[ARG:%.*]], double [[LOAD2:%.*]], i1 [[FCMP3:%.*]]) {
+; COUNT2-NEXT:  bb:
+; COUNT2-NEXT:    [[GETELEMENTPTR:%.*]] = getelementptr double, ptr [[ARG]], i64 1
+; COUNT2-NEXT:    [[LOAD:%.*]] = load double, ptr [[GETELEMENTPTR]], align 8
+; COUNT2-NEXT:    [[FCMP:%.*]] = fcmp olt double [[LOAD]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT3:%.*]] = select i1 [[FCMP]], double 0.000000e+00, double [[LOAD]]
+; COUNT2-NEXT:    [[SELECT4:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD]]
+; COUNT2-NEXT:    [[LOAD7:%.*]] = load double, ptr [[ARG]], align 8
+; COUNT2-NEXT:    [[SELECT10:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD7]]
+; COUNT2-NEXT:    [[FCMP11:%.*]] = fcmp olt double [[LOAD2]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT128:%.*]] = select i1 [[FCMP11]], double 0.000000e+00, double [[LOAD7]]
+; COUNT2-NEXT:    [[GETELEMENTPTR13:%.*]] = getelementptr double, ptr [[ARG]], i64 3
+; COUNT2-NEXT:    [[GETELEMENTPTR21:%.*]] = getelementptr double, ptr [[ARG]], i64 4
+; COUNT2-NEXT:    [[FCMP23:%.*]] = fcmp olt double [[SELECT10]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT24:%.*]] = select i1 [[FCMP23]], double 0.000000e+00, double 1.000000e+00
+; COUNT2-NEXT:    [[FCMP29:%.*]] = fcmp olt double [[SELECT4]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT30:%.*]] = select i1 [[FCMP29]], double 1.000000e+00, double 0.000000e+00
+; COUNT2-NEXT:    [[FCMP33:%.*]] = fcmp olt double [[SELECT24]], [[SELECT3]]
+; COUNT2-NEXT:    [[SELECT34:%.*]] = select i1 [[FCMP33]], double 0.000000e+00, double [[LOAD2]]
+; COUNT2-NEXT:    [[FCMP39:%.*]] = fcmp olt double [[SELECT30]], [[SELECT128]]
+; COUNT2-NEXT:    [[SELECT40:%.*]] = select i1 [[FCMP39]], double [[LOAD2]], double 0.000000e+00
+; COUNT2-NEXT:    [[FCMP62:%.*]] = fcmp olt double [[SELECT34]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT639:%.*]] = select i1 [[FCMP62]], double 0.000000e+00, double 1.000000e+00
+; COUNT2-NEXT:    [[FCMP76:%.*]] = fcmp olt double [[SELECT40]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT77:%.*]] = select i1 [[FCMP76]], double 0.000000e+00, double 1.000000e+00
+; COUNT2-NEXT:    [[FCMP90:%.*]] = fcmp ogt double [[SELECT639]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT91:%.*]] = select i1 [[FCMP90]], double 0.000000e+00, double [[LOAD2]]
+; COUNT2-NEXT:    [[FCMP92:%.*]] = fcmp ogt double [[SELECT77]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT93:%.*]] = select i1 [[FCMP92]], double 0.000000e+00, double [[LOAD2]]
+; COUNT2-NEXT:    [[FCMP108:%.*]] = fcmp olt double [[SELECT93]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT109:%.*]] = select i1 [[FCMP108]], double 1.000000e+00, double 0.000000e+00
+; COUNT2-NEXT:    [[FCMP110:%.*]] = fcmp olt double [[SELECT91]], 0.000000e+00
+; COUNT2-NEXT:    [[SELECT111:%.*]] = select i1 [[FCMP110]], double 1.000000e+00, double 0.000000e+00
+; COUNT2-NEXT:    store double [[SELECT111]], ptr [[GETELEMENTPTR21]], align 8
+; COUNT2-NEXT:    store double [[SELECT109]], ptr [[GETELEMENTPTR13]], align 8
+; COUNT2-NEXT:    ret void
+;
+; COUNT-1-LABEL: define void @blam
+; COUNT-1-SAME: (ptr [[ARG:%.*]], double [[LOAD2:%.*]], i1 [[FCMP3:%.*]]) {
+; COUNT-1-NEXT:  bb:
+; COUNT-1-NEXT:    [[GETELEMENTPTR:%.*]] = getelementptr double, ptr [[ARG]], i64 1
+; COUNT-1-NEXT:    [[LOAD:%.*]] = load double, ptr [[GETELEMENTPTR]], align 8
+; COUNT-1-NEXT:    [[SELECT4:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD]]
+; COUNT-1-NEXT:    [[LOAD7:%.*]] = load double, ptr [[ARG]], align 8
+; COUNT-1-NEXT:    [[SELECT10:%.*]] = select i1 [[FCMP3]], double 0.000000e+00, double [[LOAD7]]
+; COUNT-1-NEXT:    [[GETELEMENTPTR13:%.*]] = getelementptr double, ptr [[ARG]], i64 3
+; COUNT-1-NEXT:    [[GETELEMENTPTR21:%.*]] = getelementptr double, ptr [[ARG]], i64 4
+; COUNT-1-NEXT:    [[TMP0:%.*]] = insertelement <2 x double> poison, double [[SELECT10]], i32 0
+; COUNT-1-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> [[TMP0]], double [[LOAD]], i32 1
+; COUNT-1-NEXT:    [[TMP2:%.*]] = fcmp olt <2 x double> [[TMP1]], zeroinitializer
+; COUNT-1-NEXT:    [[TMP3:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> <double 1.000000e+00, double poison>, <2 x i32> <i32 2, i32 1>
+; COUNT-1-NEXT:    [[TMP4:%.*]] = select <2 x i1> [[TMP2]], <2 x double> zeroinitializer, <2 x double> [[TMP3]]
+; COUNT-1-NEXT:    [[TMP5:%.*]] = insertelement <2 x double> poison, double [[LOAD2]], i32 1
+; COUNT-1-NEXT:    [[TMP6:%.*]] = insertelement <2 x double> [[TMP5]], double [[SELECT4]], i32 0
+; COUNT-1-NEXT:    [[TMP7:%.*]] = fcmp olt <2 x double> [[TMP6]], zeroinitializer
+; COUNT-1-NEXT:    [[TMP8:%.*]] = insertelement <2 x double> <double 0.000000e+00, double poison>, double [[LOAD7]], i32 1
+; COUNT-1-NEXT:    [[TMP9:%.*]] = select <2 x i1> [[TMP7]], <2 x double> <double 1.000000e+00, double 0.000000e+00>, <2 x double> [[TMP8]]
+; COUNT-1-NEXT:    [[TMP10:%.*]] = extractelement <2 x double> [[TMP4]], i32 0
+; COUNT-1-NEXT:    [[TMP11:%.*]] = extractelement <2 x double> [[TMP4]], i32 1
+; COUNT-1-NEXT:    [[FCMP33:%.*]] = fcmp olt double [[TMP10]], [[TMP11]]
+; COUNT-1-NEXT:    [[SELECT34:%.*]] = select i1 [[FCMP33]], double 0.000000e+00, double [[LOAD2]]
+; COUNT-1-NEXT:    [[TMP12:%.*]] = extractelement <2 x double> [[TMP9]], i32 0
+; COUNT-1-NEXT:    [[TMP13:%.*]] = extractelement <2 x double> [[TMP9]], i32 1
+; COUNT-1-NEXT:    [[FCMP39:%.*]] = fcmp olt double [[TMP12]], [[TMP13]]
+; COUNT-1-NEXT:    [[SELECT40:%.*]] = select i1 [[FCMP39]], double [[LOAD2]], double 0.000000e+00
+; COUNT-1-NEXT:    [[FCMP62:%.*]] = fcmp olt double [[SELECT34]], 0.000000e+00
+; COUNT-1-NEXT:    [[SELECT639:%.*]] = select i1 [[FCMP62]], double 0.000000e+00, double 1.000000e+00
+; COUNT-1-NEXT:    [[FCMP76:%.*]] = fcmp olt double [[SELECT40]], 0.000000e+00
+; COUNT-1-NEXT:    [[SELECT77:%.*]] = select i1 [[FCMP76]], double 0.000000e+00, double 1.000000e+00
+; COUNT-1-NEXT:    [[FCMP90:%.*]] = fcmp ogt double [[SELECT639]], 0.000000e+00
+; COUNT-1-NEXT:    [[SELECT91:%.*]] = select i1 [[FCMP90]], double 0.000000e+00, double [[LOAD2]]
+; COUNT-1-NEXT:    [[FCMP92:%.*]] = fcmp ogt double [[SELECT77]], 0.000000e+00
+; COUNT-1-NEXT:    [[SELECT93:%.*]] = select i1 [[FCMP92]], double 0.000000e+00, double [[LOAD2]]
+; COUNT-1-NEXT:    [[FCMP108:%.*]] = fcmp olt double [[SELECT93]], 0.000000e+00
+; COUNT-1-NEXT:    [[SELECT109:%.*]] = select i1 [[FCMP108]], double 1.000000e+00, double 0.000000e+00
+; COUNT-1-NEXT:    [[FCMP110:%.*]] = fcmp olt double [[SELECT91]], 0.000000e+00
+; COUNT-1-NEXT:    [[SELECT111:%.*]] = select i1 [[FCMP110]], double 1.000000e+00, double 0.000000e+00
+; COUNT-1-NEXT:    store double [[SELECT111]], ptr [[GETELEMENTPTR21]], align 8
+; COUNT-1-NEXT:    store double [[SELECT109]], ptr [[GETELEMENTPTR13]], align 8
+; COUNT-1-NEXT:    ret void
+;
+bb:
+  %getelementptr = getelementptr double, ptr %arg, i64 1
+  %load = load double, ptr %getelementptr, align 8
+  %fcmp = fcmp olt double %load, 0.000000e+00
+  %select3 = select i1 %fcmp, double 0.000000e+00, double %load
+  %select4 = select i1 %fcmp3, double 0.000000e+00, double %load
+  %load7 = load double, ptr %arg, align 8
+  %select10 = select i1 %fcmp3, double 0.000000e+00, double %load7
+  %fcmp11 = fcmp olt double %load2, 0.000000e+00
+  %select128 = select i1 %fcmp11, double 0.000000e+00, double %load7
+  %getelementptr13 = getelementptr double, ptr %arg, i64 3
+  %getelementptr21 = getelementptr double, ptr %arg, i64 4
+  %fcmp23 = fcmp olt double %select10, 0.000000e+00
+  %select24 = select i1 %fcmp23, double 0.000000e+00, double 1.000000e+00
+  %fcmp29 = fcmp olt double %select4, 0.000000e+00
+  %select30 = select i1 %fcmp29, double 1.000000e+00, double 0.000000e+00
+  %fcmp33 = fcmp olt double %select24, %select3
+  %select34 = select i1 %fcmp33, double 0.000000e+00, double %load2
+  %fcmp39 = fcmp olt double %select30, %select128
+  %select40 = select i1 %fcmp39, double %load2, double 0.000000e+00
+  %fcmp62 = fcmp olt double %select34, 0.000000e+00
+  %select639 = select i1 %fcmp62, double 0.000000e+00, double 1.000000e+00
+  %fcmp76 = fcmp olt double %select40, 0.000000e+00
+  %select77 = select i1 %fcmp76, double 0.000000e+00, double 1.000000e+00
+  %fcmp90 = fcmp ogt double %select639, 0.000000e+00
+  %select91 = select i1 %fcmp90, double 0.000000e+00, double %load2
+  %fcmp92 = fcmp ogt double %select77, 0.000000e+00
+  %select93 = select i1 %fcmp92, double 0.000000e+00, double %load2
+  %fcmp108 = fcmp olt double %select93, 0.000000e+00
+  %select109 = select i1 %fcmp108, double 1.000000e+00, double 0.000000e+00
+  %fcmp110 = fcmp olt double %select91, 0.000000e+00
+  %select111 = select i1 %fcmp110, double 1.000000e+00, double 0.000000e+00
+  store double %select111, ptr %g...
[truncated]

aeubanks
aeubanks reviewed Oct 1, 2024
View reviewed changes
Copy link
Contributor

@aeubanks aeubanks left a comment

Choose a reason for hiding this comment

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

thanks for the very quick PR!

can you add Fixes #110725 in the description?

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -11175,6 +11179,9 @@ bool BoUpSLP::isLoadCombineCandidate(ArrayRef<Value *> Stores) const {
}

bool BoUpSLP::isTreeTinyAndNotFullyVectorizable(bool ForReduction) const {
Copy link
Contributor

Choose a reason for hiding this comment

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

I think ideally we'd only increment the counter and force return true in the cases where this function would return false, but ultimately it doesn't matter too much, it just makes the counter ranges larger

@alexey-bataev alexey-bataev merged commit f3c408d into main Oct 1, 2024
8 of 10 checks passed
@alexey-bataev alexey-bataev deleted the users/alexey-bataev/spr/slpadd-debug-counter-support branch October 1, 2024 20:21
puja2196 pushed a commit to puja2196/LLVM-tutorial that referenced this pull request Oct 2, 2024
@alexey-bataev @puja2196
[SLP]Add debug counter support
a5cf414 
Fixes #110725

Reviewers: aeubanks

Reviewed By: aeubanks

Pull Request: llvm/llvm-project#110734
alexey-bataev added a commit that referenced this pull request Oct 2, 2024
@alexey-bataev
[SLP]Add debug counter support
67dd9d2 
Fixes #110725

Reviewers: aeubanks

Reviewed By: aeubanks

Pull Request: #110734
alexey-bataev added a commit that referenced this pull request Oct 2, 2024
@alexey-bataev
[SLP]Add debug counter support
4197e73 
Fixes #110725

Reviewers: aeubanks

Reviewed By: aeubanks

Pull Request: #110734
Sterling-Augustine pushed a commit to Sterling-Augustine/llvm-project that referenced this pull request Oct 3, 2024
@alexey-bataev @Sterling-Augustine
[SLP]Add debug counter support
2487b2d 
Fixes llvm#110725

Reviewers: aeubanks

Reviewed By: aeubanks

Pull Request: llvm#110734
Sterling-Augustine pushed a commit to Sterling-Augustine/llvm-project that referenced this pull request Oct 3, 2024
@alexey-bataev @Sterling-Augustine
[SLP]Add debug counter support
dd7ed23 
Fixes llvm#110725

Reviewers: aeubanks

Reviewed By: aeubanks

Pull Request: llvm#110734
Sterling-Augustine pushed a commit to Sterling-Augustine/llvm-project that referenced this pull request Oct 3, 2024
@alexey-bataev @Sterling-Augustine
[SLP]Add debug counter support
96db61b 
Fixes llvm#110725

Reviewers: aeubanks

Reviewed By: aeubanks

Pull Request: llvm#110734
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@aeubanks aeubanks aeubanks approved these changes

Assignees
No one assigned
Labels
llvm:transforms vectorizers
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Consider adding DEBUG_COUNTER in SLPVectorizer
3 participants
@alexey-bataev @llvmbot @aeubanks