[llvm] c740089 - [FPEnv][InstSimplify] Fold fsub X, -0 ==> X, when we know X is not -0

[Kevin P. Neal via llvm-commits]

Author: Kevin P. Neal
Date: 2022-02-16T10:10:13-05:00
New Revision: c7400892caa0a3a2ca7d0d4b918da03b8bac947c

URL: https://github.com/llvm/llvm-project/commit/c7400892caa0a3a2ca7d0d4b918da03b8bac947c
DIFF: https://github.com/llvm/llvm-project/commit/c7400892caa0a3a2ca7d0d4b918da03b8bac947c.diff

LOG: [FPEnv][InstSimplify] Fold fsub X, -0 ==> X, when we know X is not -0

Currently the fsub optimizations in InstSimplify don't know how to fold
X - -0.0 to X when we know X is not zero and the constrained intrinsics
are used. This adds the support.

This review is split out from D107285.

Differential Revision: https://reviews.llvm.org/D119746

Added: 
    

Modified: 
    llvm/lib/Analysis/InstructionSimplify.cpp
    llvm/test/Transforms/InstSimplify/strictfp-fsub.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index 305c3a8dbe125..a2d2b2ae27025 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -5138,14 +5138,15 @@ SimplifyFSubInst(Value *Op0, Value *Op1, FastMathFlags FMF,
     if (match(Op1, m_PosZeroFP()))
       return Op0;
 
+  // fsub X, -0 ==> X, when we know X is not -0
+  if (canIgnoreSNaN(ExBehavior, FMF))
+    if (match(Op1, m_NegZeroFP()) &&
+        (FMF.noSignedZeros() || CannotBeNegativeZero(Op0, Q.TLI)))
+      return Op0;
+
   if (!isDefaultFPEnvironment(ExBehavior, Rounding))
     return nullptr;
 
-  // fsub X, -0 ==> X, when we know X is not -0
-  if (match(Op1, m_NegZeroFP()) &&
-      (FMF.noSignedZeros() || CannotBeNegativeZero(Op0, Q.TLI)))
-    return Op0;
-
   // fsub -0.0, (fsub -0.0, X) ==> X
   // fsub -0.0, (fneg X) ==> X
   Value *X;

diff  --git a/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll b/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll
index f72304166d56b..f391560b3422d 100644
--- a/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll
+++ b/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll
@@ -127,11 +127,9 @@ define float @fold_fsub_nsz_x_n0_ebmaytrap(float %a) #0 {
   ret float %sub
 }
 
-; TODO: This will fold if we allow non-default floating point environments.
 define float @fold_fsub_nnan_nsz_x_n0_ebmaytrap(float %a) #0 {
 ; CHECK-LABEL: @fold_fsub_nnan_nsz_x_n0_ebmaytrap(
-; CHECK-NEXT:    [[SUB:%.*]] = call nnan nsz float @llvm.experimental.constrained.fsub.f32(float [[A:%.*]], float -0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #[[ATTR0]]
-; CHECK-NEXT:    ret float [[SUB]]
+; CHECK-NEXT:    ret float [[A:%.*]]
 ;
   %sub = call nnan nsz float @llvm.experimental.constrained.fsub.f32(float %a, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #0
   ret float %sub
@@ -147,12 +145,11 @@ define float @fold_fsub_nsz_x_n0_ebstrict(float %a) #0 {
   ret float %sub
 }
 
-; TODO: This will fold if we allow non-default floating point environments.
 ; TODO: The instruction is expected to remain, but the result isn't used.
 define float @fold_fsub_nsz_nnan_x_n0_ebstrict(float %a) #0 {
 ; CHECK-LABEL: @fold_fsub_nsz_nnan_x_n0_ebstrict(
 ; CHECK-NEXT:    [[SUB:%.*]] = call nnan nsz float @llvm.experimental.constrained.fsub.f32(float [[A:%.*]], float -0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR0]]
-; CHECK-NEXT:    ret float [[SUB]]
+; CHECK-NEXT:    ret float [[A]]
 ;
   %sub = call nsz nnan float @llvm.experimental.constrained.fsub.f32(float %a, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
   ret float %sub
@@ -185,12 +182,10 @@ define float @fold_fsub_fabs_x_n0_ebmaytrap(float %a) #0 {
   ret float %sub
 }
 
-; TODO: This will fold if we allow non-default floating point environments.
 define float @fold_fsub_fabs_nnan_x_n0_ebmaytrap(float %a) #0 {
 ; CHECK-LABEL: @fold_fsub_fabs_nnan_x_n0_ebmaytrap(
 ; CHECK-NEXT:    [[ABSA:%.*]] = call float @llvm.fabs.f32(float [[A:%.*]])
-; CHECK-NEXT:    [[SUB:%.*]] = call nnan float @llvm.experimental.constrained.fsub.f32(float [[ABSA]], float -0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #[[ATTR0]]
-; CHECK-NEXT:    ret float [[SUB]]
+; CHECK-NEXT:    ret float [[ABSA]]
 ;
   %absa = call float @llvm.fabs.f32(float %a)
   %sub = call nnan float @llvm.experimental.constrained.fsub.f32(float %absa, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #0
@@ -209,13 +204,12 @@ define float @fold_fsub_fabs_x_n0_ebstrict(float %a) #0 {
   ret float %sub
 }
 
-; TODO: This will fold if we allow non-default floating point environments.
 ; TODO: The instruction is expected to remain, but the result isn't used.
 define float @fold_fsub_fabs_nnan_x_n0_ebstrict(float %a) #0 {
 ; CHECK-LABEL: @fold_fsub_fabs_nnan_x_n0_ebstrict(
 ; CHECK-NEXT:    [[ABSA:%.*]] = call float @llvm.fabs.f32(float [[A:%.*]])
 ; CHECK-NEXT:    [[SUB:%.*]] = call nnan float @llvm.experimental.constrained.fsub.f32(float [[ABSA]], float -0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR0]]
-; CHECK-NEXT:    ret float [[SUB]]
+; CHECK-NEXT:    ret float [[ABSA]]
 ;
   %absa = call float @llvm.fabs.f32(float %a)
   %sub = call nnan float @llvm.experimental.constrained.fsub.f32(float %absa, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.strict") #0