Skip to content

[InstCombine] Fix FMF propagation in foldSelectWithFCmpToFabs #121580

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
Feb 1, 2025
Merged

[InstCombine] Fix FMF propagation in foldSelectWithFCmpToFabs #121580

merged 1 commit into from
Feb 1, 2025

Conversation

dtcxzyw
Copy link
Member

@dtcxzyw dtcxzyw commented Jan 3, 2025 •
edited
Loading

Consider the following pattern:

%cmp = fcmp <pred> double %x, 0.000000e+00
%negX = fneg <fmf> double %x
%sel = select i1 %cmp, double %x, double %negX

We cannot propagate ninf from fneg to select since %negX may not be chosen. Similarly, we cannot propagate nnan unless %negX is guaranteed to be selected when %x is NaN.
This patch also propagates nnan/ninf from fcmp to avoid regression in PhaseOrdering/generate-fabs.ll.

Alive2: https://alive2.llvm.org/ce/z/t6U-tA
Closes #121430 and #113989.

@dtcxzyw dtcxzyw requested review from arsenm and goldsteinn January 3, 2025 16:21
@dtcxzyw dtcxzyw requested a review from nikic as a code owner January 3, 2025 16:21
@llvmbot
Copy link
Member

llvmbot commented Jan 3, 2025

@llvm/pr-subscribers-llvm-transforms

Author: Yingwei Zheng (dtcxzyw)

Changes

Consider the following pattern:

%cmp = fcmp &lt;pred&gt; double %x, 0.000000e+00
%negX = fneg &lt;fmf&gt; double %x
%sel = select i1 %cmp, double %x, double %negX

We cannot propagate ninf from fneg to select since %negX may not be chosen. Similarly, we cannot propagate nnan unless %negX is guaranteed to be selected when %x is NaN.
This patch also propagates nnan/ninf from fcmp to avoid regression in PhaseOrdering/generate-fabs.ll.

Alive2: https://alive2.llvm.org/ce/z/t6U-tA
Closes #121430.

Full diff: https://github.com/llvm/llvm-project/pull/121580.diff

3 Files Affected:

  • (modified) llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp (+9-2)
  • (modified) llvm/test/Transforms/InstCombine/fabs.ll (+20-20)
  • (modified) llvm/test/Transforms/InstCombine/fneg-fabs.ll (+9-8)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index e7a8e947705f8d..3a1536bd1e8de3 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -2926,10 +2926,10 @@ static Instruction *foldSelectWithFCmpToFabs(SelectInst &SI,
     if (!match(TrueVal, m_FNeg(m_Specific(X))))
       return nullptr;
 
-    // Forward-propagate nnan and ninf from the fneg to the select.
+    // Forward-propagate nnan and ninf from the fcmp to the select.
     // If all inputs are not those values, then the select is not either.
     // Note: nsz is defined differently, so it may not be correct to propagate.
-    FastMathFlags FMF = cast<FPMathOperator>(TrueVal)->getFastMathFlags();
+    FastMathFlags FMF = cast<FPMathOperator>(CondVal)->getFastMathFlags();
     if (FMF.noNaNs() && !SI.hasNoNaNs()) {
       SI.setHasNoNaNs(true);
       ChangedFMF = true;
@@ -2938,6 +2938,13 @@ static Instruction *foldSelectWithFCmpToFabs(SelectInst &SI,
       SI.setHasNoInfs(true);
       ChangedFMF = true;
     }
+    // Forward-propagate nnan from the fneg to the select.
+    // The nnan flag can be propagated iff fneg is selected when X is NaN.
+    if (!SI.hasNoNaNs() && cast<FPMathOperator>(TrueVal)->hasNoNaNs() &&
+        (Swap ? FCmpInst::isOrdered(Pred) : FCmpInst::isUnordered(Pred))) {
+      SI.setHasNoNaNs(true);
+      ChangedFMF = true;
+    }
 
     // With nsz, when 'Swap' is false:
     // fold (X < +/-0.0) ? -X : X or (X <= +/-0.0) ? -X : X to fabs(X)
diff --git a/llvm/test/Transforms/InstCombine/fabs.ll b/llvm/test/Transforms/InstCombine/fabs.ll
index cccf0f4457b6ab..7b9a672f188cab 100644
--- a/llvm/test/Transforms/InstCombine/fabs.ll
+++ b/llvm/test/Transforms/InstCombine/fabs.ll
@@ -484,7 +484,7 @@ define double @select_fcmp_nnan_nsz_olt_zero(double %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_olt_zero(
 ; CHECK-NEXT:    [[LTZERO:%.*]] = fcmp olt double [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz double [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LTZERO]], double [[NEGX]], double [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LTZERO]], double [[NEGX]], double [[X]]
 ; CHECK-NEXT:    ret double [[FABS]]
 ;
   %ltzero = fcmp olt double %x, 0.0
@@ -523,7 +523,7 @@ define double @select_fcmp_nnan_nsz_olt_zero_unary_fneg(double %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_olt_zero_unary_fneg(
 ; CHECK-NEXT:    [[LTZERO:%.*]] = fcmp olt double [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz double [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LTZERO]], double [[NEGX]], double [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LTZERO]], double [[NEGX]], double [[X]]
 ; CHECK-NEXT:    ret double [[FABS]]
 ;
   %ltzero = fcmp olt double %x, 0.0
@@ -553,7 +553,7 @@ define float @select_fcmp_nnan_nsz_olt_negzero(float %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_olt_negzero(
 ; CHECK-NEXT:    [[LTZERO:%.*]] = fcmp olt float [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan ninf nsz float [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LTZERO]], float [[NEGX]], float [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LTZERO]], float [[NEGX]], float [[X]]
 ; CHECK-NEXT:    ret float [[FABS]]
 ;
   %ltzero = fcmp olt float %x, -0.0
@@ -579,7 +579,7 @@ define float @select_fcmp_nnan_nsz_ult_negzero(float %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ult_negzero(
 ; CHECK-NEXT:    [[LTZERO:%.*]] = fcmp ult float [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan ninf nsz float [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LTZERO]], float [[NEGX]], float [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LTZERO]], float [[NEGX]], float [[X]]
 ; CHECK-NEXT:    ret float [[FABS]]
 ;
   %ltzero = fcmp ult float %x, -0.0
@@ -592,7 +592,7 @@ define float @select_fcmp_nnan_nsz_olt_negzero_unary_fneg(float %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_olt_negzero_unary_fneg(
 ; CHECK-NEXT:    [[LTZERO:%.*]] = fcmp olt float [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan ninf nsz float [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LTZERO]], float [[NEGX]], float [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LTZERO]], float [[NEGX]], float [[X]]
 ; CHECK-NEXT:    ret float [[FABS]]
 ;
   %ltzero = fcmp olt float %x, -0.0
@@ -607,7 +607,7 @@ define float @select_fcmp_nnan_nsz_ult_negzero_unary_fneg(float %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ult_negzero_unary_fneg(
 ; CHECK-NEXT:    [[LTZERO:%.*]] = fcmp ult float [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan ninf nsz float [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LTZERO]], float [[NEGX]], float [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LTZERO]], float [[NEGX]], float [[X]]
 ; CHECK-NEXT:    ret float [[FABS]]
 ;
   %ltzero = fcmp ult float %x, -0.0
@@ -622,7 +622,7 @@ define double @select_fcmp_nnan_nsz_ole_zero(double %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ole_zero(
 ; CHECK-NEXT:    [[LEZERO:%.*]] = fcmp ole double [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg fast double [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LEZERO]], double [[NEGX]], double [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LEZERO]], double [[NEGX]], double [[X]]
 ; CHECK-NEXT:    ret double [[FABS]]
 ;
   %lezero = fcmp ole double %x, 0.0
@@ -648,7 +648,7 @@ define double @select_fcmp_nnan_nsz_ule_zero(double %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ule_zero(
 ; CHECK-NEXT:    [[LEZERO:%.*]] = fcmp ule double [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg fast double [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LEZERO]], double [[NEGX]], double [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LEZERO]], double [[NEGX]], double [[X]]
 ; CHECK-NEXT:    ret double [[FABS]]
 ;
   %lezero = fcmp ule double %x, 0.0
@@ -661,7 +661,7 @@ define double @select_fcmp_nnan_nsz_ole_zero_unary_fneg(double %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ole_zero_unary_fneg(
 ; CHECK-NEXT:    [[LEZERO:%.*]] = fcmp ole double [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg fast double [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LEZERO]], double [[NEGX]], double [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LEZERO]], double [[NEGX]], double [[X]]
 ; CHECK-NEXT:    ret double [[FABS]]
 ;
   %lezero = fcmp ole double %x, 0.0
@@ -676,7 +676,7 @@ define double @select_fcmp_nnan_nsz_ule_zero_unary_fneg(double %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ule_zero_unary_fneg(
 ; CHECK-NEXT:    [[LEZERO:%.*]] = fcmp ule double [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg fast double [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[LEZERO]], double [[NEGX]], double [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LEZERO]], double [[NEGX]], double [[X]]
 ; CHECK-NEXT:    ret double [[FABS]]
 ;
   %lezero = fcmp ule double %x, 0.0
@@ -691,7 +691,7 @@ define float @select_fcmp_nnan_nsz_ole_negzero(float %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ole_negzero(
 ; CHECK-NEXT:    [[LEZERO:%.*]] = fcmp ole float [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz float [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LEZERO]], float [[NEGX]], float [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LEZERO]], float [[NEGX]], float [[X]]
 ; CHECK-NEXT:    ret float [[FABS]]
 ;
   %lezero = fcmp ole float %x, -0.0
@@ -730,7 +730,7 @@ define float @select_fcmp_nnan_nsz_ole_negzero_unary_fneg(float %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ole_negzero_unary_fneg(
 ; CHECK-NEXT:    [[LEZERO:%.*]] = fcmp ole float [[X:%.*]], 0.000000e+00
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz float [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[LEZERO]], float [[NEGX]], float [[X]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[LEZERO]], float [[NEGX]], float [[X]]
 ; CHECK-NEXT:    ret float [[FABS]]
 ;
   %lezero = fcmp ole float %x, -0.0
@@ -802,7 +802,7 @@ define <2 x float> @select_fcmp_nnan_nsz_ugt_zero(<2 x float> %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ugt_zero(
 ; CHECK-NEXT:    [[GTZERO:%.*]] = fcmp ugt <2 x float> [[X:%.*]], zeroinitializer
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz arcp <2 x float> [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan <2 x i1> [[GTZERO]], <2 x float> [[X]], <2 x float> [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select <2 x i1> [[GTZERO]], <2 x float> [[X]], <2 x float> [[NEGX]]
 ; CHECK-NEXT:    ret <2 x float> [[FABS]]
 ;
   %gtzero = fcmp ugt <2 x float> %x, zeroinitializer
@@ -830,7 +830,7 @@ define <2 x float> @select_fcmp_nnan_nsz_ugt_zero_unary_fneg(<2 x float> %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ugt_zero_unary_fneg(
 ; CHECK-NEXT:    [[GTZERO:%.*]] = fcmp ugt <2 x float> [[X:%.*]], zeroinitializer
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz arcp <2 x float> [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan <2 x i1> [[GTZERO]], <2 x float> [[X]], <2 x float> [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select <2 x i1> [[GTZERO]], <2 x float> [[X]], <2 x float> [[NEGX]]
 ; CHECK-NEXT:    ret <2 x float> [[FABS]]
 ;
   %gtzero = fcmp ugt <2 x float> %x, zeroinitializer
@@ -845,7 +845,7 @@ define half @select_fcmp_nnan_nsz_ogt_negzero(half %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ogt_negzero(
 ; CHECK-NEXT:    [[GTZERO:%.*]] = fcmp ogt half [[X:%.*]], 0xH0000
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg fast half [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[GTZERO]], half [[X]], half [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[GTZERO]], half [[X]], half [[NEGX]]
 ; CHECK-NEXT:    ret half [[FABS]]
 ;
   %gtzero = fcmp ogt half %x, -0.0
@@ -860,7 +860,7 @@ define half @select_fcmp_nnan_nsz_ugt_negzero(half %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_ugt_negzero(
 ; CHECK-NEXT:    [[GTZERO:%.*]] = fcmp ugt half [[X:%.*]], 0xH0000
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg fast half [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan ninf i1 [[GTZERO]], half [[X]], half [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[GTZERO]], half [[X]], half [[NEGX]]
 ; CHECK-NEXT:    ret half [[FABS]]
 ;
   %gtzero = fcmp ugt half %x, -0.0
@@ -890,7 +890,7 @@ define <2 x double> @select_fcmp_nnan_nsz_uge_zero(<2 x double> %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_uge_zero(
 ; CHECK-NEXT:    [[GEZERO:%.*]] = fcmp uge <2 x double> [[X:%.*]], zeroinitializer
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg reassoc nnan nsz <2 x double> [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan <2 x i1> [[GEZERO]], <2 x double> [[X]], <2 x double> [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select <2 x i1> [[GEZERO]], <2 x double> [[X]], <2 x double> [[NEGX]]
 ; CHECK-NEXT:    ret <2 x double> [[FABS]]
 ;
   %gezero = fcmp uge <2 x double> %x, zeroinitializer
@@ -918,7 +918,7 @@ define <2 x double> @select_fcmp_nnan_nsz_uge_zero_unary_fneg(<2 x double> %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_uge_zero_unary_fneg(
 ; CHECK-NEXT:    [[GEZERO:%.*]] = fcmp uge <2 x double> [[X:%.*]], zeroinitializer
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg reassoc nnan nsz <2 x double> [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan <2 x i1> [[GEZERO]], <2 x double> [[X]], <2 x double> [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select <2 x i1> [[GEZERO]], <2 x double> [[X]], <2 x double> [[NEGX]]
 ; CHECK-NEXT:    ret <2 x double> [[FABS]]
 ;
   %gezero = fcmp uge <2 x double> %x, zeroinitializer
@@ -948,7 +948,7 @@ define half @select_fcmp_nnan_nsz_uge_negzero(half %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_uge_negzero(
 ; CHECK-NEXT:    [[GEZERO:%.*]] = fcmp uge half [[X:%.*]], 0xH0000
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz half [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[GEZERO]], half [[X]], half [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[GEZERO]], half [[X]], half [[NEGX]]
 ; CHECK-NEXT:    ret half [[FABS]]
 ;
   %gezero = fcmp uge half %x, -0.0
@@ -976,7 +976,7 @@ define half @select_fcmp_nnan_nsz_uge_negzero_unary_fneg(half %x) {
 ; CHECK-LABEL: @select_fcmp_nnan_nsz_uge_negzero_unary_fneg(
 ; CHECK-NEXT:    [[GEZERO:%.*]] = fcmp uge half [[X:%.*]], 0xH0000
 ; CHECK-NEXT:    [[NEGX:%.*]] = fneg nnan nsz half [[X]]
-; CHECK-NEXT:    [[FABS:%.*]] = select nnan i1 [[GEZERO]], half [[X]], half [[NEGX]]
+; CHECK-NEXT:    [[FABS:%.*]] = select i1 [[GEZERO]], half [[X]], half [[NEGX]]
 ; CHECK-NEXT:    ret half [[FABS]]
 ;
   %gezero = fcmp uge half %x, -0.0
diff --git a/llvm/test/Transforms/InstCombine/fneg-fabs.ll b/llvm/test/Transforms/InstCombine/fneg-fabs.ll
index fdcdfd123eefa4..c013a2c9f5449c 100644
--- a/llvm/test/Transforms/InstCombine/fneg-fabs.ll
+++ b/llvm/test/Transforms/InstCombine/fneg-fabs.ll
@@ -20,8 +20,8 @@ define double @select_noFMF_nfabs_lt(double %x) {
 ; One test where the neg has fmfs.
 define double @select_nsz_nfabs_lt_fmfProp(double %x) {
 ; CHECK-LABEL: @select_nsz_nfabs_lt_fmfProp(
-; CHECK-NEXT:    [[TMP1:%.*]] = call nnan ninf nsz double @llvm.fabs.f64(double [[X:%.*]])
-; CHECK-NEXT:    [[SEL:%.*]] = fneg nnan ninf nsz double [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call nnan nsz double @llvm.fabs.f64(double [[X:%.*]])
+; CHECK-NEXT:    [[SEL:%.*]] = fneg nnan nsz double [[TMP1]]
 ; CHECK-NEXT:    ret double [[SEL]]
 ;
   %cmp = fcmp olt double %x, 0.000000e+00
@@ -32,8 +32,8 @@ define double @select_nsz_nfabs_lt_fmfProp(double %x) {
 
 define double @select_nsz_nnan_nfabs_lt_fmfProp(double %x) {
 ; CHECK-LABEL: @select_nsz_nnan_nfabs_lt_fmfProp(
-; CHECK-NEXT:    [[TMP1:%.*]] = call nnan ninf nsz double @llvm.fabs.f64(double [[X:%.*]])
-; CHECK-NEXT:    [[SEL:%.*]] = fneg nnan ninf nsz double [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call nnan nsz double @llvm.fabs.f64(double [[X:%.*]])
+; CHECK-NEXT:    [[SEL:%.*]] = fneg nnan nsz double [[TMP1]]
 ; CHECK-NEXT:    ret double [[SEL]]
 ;
   %cmp = fcmp olt double %x, 0.000000e+00
@@ -147,8 +147,9 @@ define double @select_noFMF_nfabs_gt(double %x) {
 ; One test where the neg has fmfs.
 define double @select_nsz_nfabs_gt_fmfProp(double %x) {
 ; CHECK-LABEL: @select_nsz_nfabs_gt_fmfProp(
-; CHECK-NEXT:    [[TMP1:%.*]] = call nnan ninf nsz double @llvm.fabs.f64(double [[X:%.*]])
-; CHECK-NEXT:    [[SEL:%.*]] = fneg nnan ninf nsz double [[TMP1]]
+; CHECK-NEXT:    [[CMP:%.*]] = fcmp ogt double [[X:%.*]], 0.000000e+00
+; CHECK-NEXT:    [[NEGX:%.*]] = fneg fast double [[X]]
+; CHECK-NEXT:    [[SEL:%.*]] = select nsz i1 [[CMP]], double [[NEGX]], double [[X]]
 ; CHECK-NEXT:    ret double [[SEL]]
 ;
   %cmp = fcmp ogt double %x, 0.000000e+00
@@ -159,8 +160,8 @@ define double @select_nsz_nfabs_gt_fmfProp(double %x) {
 
 define double @select_nsz_nnan_nfabs_gt_fmfProp(double %x) {
 ; CHECK-LABEL: @select_nsz_nnan_nfabs_gt_fmfProp(
-; CHECK-NEXT:    [[TMP1:%.*]] = call nnan ninf nsz double @llvm.fabs.f64(double [[X:%.*]])
-; CHECK-NEXT:    [[SEL:%.*]] = fneg nnan ninf nsz double [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call nnan nsz double @llvm.fabs.f64(double [[X:%.*]])
+; CHECK-NEXT:    [[SEL:%.*]] = fneg nnan nsz double [[TMP1]]
 ; CHECK-NEXT:    ret double [[SEL]]
 ;
   %cmp = fcmp ogt double %x, 0.000000e+00

This was referenced Jan 3, 2025
Closed
Closed
@dtcxzyw dtcxzyw merged commit 3ec6a6b into llvm:main Feb 1, 2025
11 checks passed
@dtcxzyw dtcxzyw deleted the select-fneg-fabs-fmf branch February 1, 2025 07:14
@dtcxzyw dtcxzyw added this to the LLVM 20.X Release milestone Feb 1, 2025
@dtcxzyw
Copy link
Member Author

dtcxzyw commented Feb 1, 2025

/cherry-pick 3ec6a6b

@llvmbot
Copy link
Member

llvmbot commented Feb 1, 2025

/pull-request #125338

@dtcxzyw dtcxzyw mentioned this pull request Feb 2, 2025
Closed
swift-ci pushed a commit to swiftlang/llvm-project that referenced this pull request Feb 7, 2025
@dtcxzyw @tstellar
[InstCombine] Fix FMF propagation in foldSelectWithFCmpToFabs (llvm…
857d8d7 
…#121580)

Consider the following pattern:
```
%cmp = fcmp <pred> double %x, 0.000000e+00
%negX = fneg <fmf> double %x
%sel = select i1 %cmp, double %x, double %negX
```
We cannot propagate ninf from fneg to select since `%negX` may not be
chosen. Similarly, we cannot propagate nnan unless `%negX` is guaranteed
to be selected when `%x` is NaN.
This patch also propagates nnan/ninf from fcmp to avoid regression in
`PhaseOrdering/generate-fabs.ll`.

Alive2: https://alive2.llvm.org/ce/z/t6U-tA
Closes llvm#121430 and
llvm#113989.

(cherry picked from commit 3ec6a6b)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@arsenm arsenm arsenm approved these changes

@goldsteinn goldsteinn Awaiting requested review from goldsteinn

@nikic nikic Awaiting requested review from nikic nikic is a code owner

Assignees
No one assigned
Labels
llvm:instcombine llvm:transforms
Projects
LLVM Release Status
Status: Done
Milestone
LLVM 20.X Release
Development

Successfully merging this pull request may close these issues.

[InstCombine] ninf is incorrectly inferred
3 participants
@dtcxzyw @llvmbot @arsenm