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

[kpn]

Summary

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.

[spatel]

"X - (-0.0)" should be equivalent to the existing code for "X + 0.0", so LGTM.

llvm/lib/Analysis/InstructionSimplify.cpp

SimplifyFSubInst(Value *Op0, Value *Op1, FastMathFlags FMF,
 
   // fsub X, +0 ==> X
   if (canIgnoreSNaN(ExBehavior, FMF) &&
       (!canRoundingModeBe(Rounding, RoundingMode::TowardNegative) ||
        FMF.noSignedZeros()))
     if (match(Op1, m_PosZeroFP()))
       return Op0;
 
-  if (!isDefaultFPEnvironment(ExBehavior, Rounding))
-    return nullptr;
-
   // 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 -0.0, (fsub -0.0, X) ==> X
   // fsub -0.0, (fneg X) ==> X
   Value *X;
   if (match(Op0, m_NegZeroFP()) &&
       match(Op1, m_FNeg(m_Value(X))))
     return X;
 
   // fsub 0.0, (fsub 0.0, X) ==> X if signed zeros are ignored.

llvm/test/Transforms/InstSimplify/strictfp-fsub.ll

 ; CHECK-LABEL: @fold_fsub_nsz_x_n0_ebmaytrap(
 ; CHECK-NEXT:    [[SUB:%.*]] = call 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]]
 ;
   %sub = call nsz float @llvm.experimental.constrained.fsub.f32(float %a, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #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
 }
 
 ; Missing nnan: must not fire.
 define float @fold_fsub_nsz_x_n0_ebstrict(float %a) #0 {
 ; CHECK-LABEL: @fold_fsub_nsz_x_n0_ebstrict(
 ; CHECK-NEXT:    [[SUB:%.*]] = call 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]]
 ;
   %sub = call nsz float @llvm.experimental.constrained.fsub.f32(float %a, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.strict") #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
 }
 
 ;
 ; fsub X, -0 ==> X, when we know X is not -0
 ; (No "nsz" flags)

 ; CHECK-NEXT:    [[SUB:%.*]] = call float @llvm.experimental.constrained.fsub.f32(float [[ABSA]], float -0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #[[ATTR0]]
 ; CHECK-NEXT:    ret float [[SUB]]
 ;
   %absa = call float @llvm.fabs.f32(float %a)
   %sub = call float @llvm.experimental.constrained.fsub.f32(float %absa, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #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
   ret float %sub
 }
 
 ; Missing nnan: must not fire.
 define float @fold_fsub_fabs_x_n0_ebstrict(float %a) #0 {
 ; CHECK-LABEL: @fold_fsub_fabs_x_n0_ebstrict(
 ; CHECK-NEXT:    [[ABSA:%.*]] = call float @llvm.fabs.f32(float [[A:%.*]])
 ; CHECK-NEXT:    [[SUB:%.*]] = call float @llvm.experimental.constrained.fsub.f32(float [[ABSA]], float -0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR0]]
 ; CHECK-NEXT:    ret float [[SUB]]
 ;
   %absa = call float @llvm.fabs.f32(float %a)
   %sub = call float @llvm.experimental.constrained.fsub.f32(float %absa, float -0.0, metadata !"round.tonearest", metadata !"fpexcept.strict") #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
   ret float %sub
 }
 
 ; TODO: This can't fire currently due to missing constrained support:
 define float @fold_fsub_sitofp_x_n0_defaultenv(i32 %a) #0 {