[llvm] 22bd65f - [FPEnv][InstSimplify] Fold fsub X, +0 ==> X

[Kevin P. Neal via llvm-commits]

Author: Kevin P. Neal
Date: 2022-02-14T11:56:45-05:00
New Revision: 22bd65fbe7ecf06b9c0a6df212985943de04cd32

URL: https://github.com/llvm/llvm-project/commit/22bd65fbe7ecf06b9c0a6df212985943de04cd32
DIFF: https://github.com/llvm/llvm-project/commit/22bd65fbe7ecf06b9c0a6df212985943de04cd32.diff

LOG: [FPEnv][InstSimplify] Fold fsub X, +0 ==> X

Currently the fsub optimizations in InstSimplify don't know how to fold X
- +0.0 to X when using the constrained intrinsics. This adds the support.

This review is split out from D107285.

Differential Revision: https://reviews.llvm.org/D118928

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 8679cd6881bf7..5d7efac9a2c7b 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -5136,13 +5136,16 @@ SimplifyFSubInst(Value *Op0, Value *Op1, FastMathFlags FMF,
   if (Constant *C = simplifyFPOp({Op0, Op1}, FMF, Q, ExBehavior, Rounding))
     return C;
 
+  // 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
-  if (match(Op1, m_PosZeroFP()))
-    return Op0;
-
   // fsub X, -0 ==> X, when we know X is not -0
   if (match(Op1, m_NegZeroFP()) &&
       (FMF.noSignedZeros() || CannotBeNegativeZero(Op0, Q.TLI)))

diff  --git a/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll b/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll
index 35ee46e7b8ff4..f72304166d56b 100644
--- a/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll
+++ b/llvm/test/Transforms/InstSimplify/strictfp-fsub.ll
@@ -27,11 +27,9 @@ define float @fsub_x_p0_ebmaytrap(float %a) #0 {
   ret float %ret
 }
 
-; TODO: This will fold if we allow non-default floating point environments.
 define float @fsub_nnan_x_p0_ebmaytrap(float %a) #0 {
 ; CHECK-LABEL: @fsub_nnan_x_p0_ebmaytrap(
-; CHECK-NEXT:    [[RET:%.*]] = call nnan float @llvm.experimental.constrained.fsub.f32(float [[A:%.*]], float 0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #[[ATTR0]]
-; CHECK-NEXT:    ret float [[RET]]
+; CHECK-NEXT:    ret float [[A:%.*]]
 ;
   %ret = call nnan float @llvm.experimental.constrained.fsub.f32(float %a, float 0.0, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #0
   ret float %ret
@@ -47,12 +45,11 @@ define float @fsub_x_p0_ebstrict(float %a) #0 {
   ret float %ret
 }
 
-; 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 @fsub_nnan_x_p0_ebstrict(float %a) #0 {
 ; CHECK-LABEL: @fsub_nnan_x_p0_ebstrict(
 ; CHECK-NEXT:    [[RET:%.*]] = call nnan float @llvm.experimental.constrained.fsub.f32(float [[A:%.*]], float 0.000000e+00, metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR0]]
-; CHECK-NEXT:    ret float [[RET]]
+; CHECK-NEXT:    ret float [[A]]
 ;
   %ret = call nnan float @llvm.experimental.constrained.fsub.f32(float %a, float 0.0, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
   ret float %ret
@@ -68,6 +65,45 @@ define float @fsub_ninf_x_p0_ebstrict(float %a) #0 {
   ret float %ret
 }
 
+; Round to -inf and if x is zero then the result is -0.0: must not fire
+define float @fsub_x_p0_neginf(float %a) #0 {
+; CHECK-LABEL: @fsub_x_p0_neginf(
+; CHECK-NEXT:    [[RET:%.*]] = call float @llvm.experimental.constrained.fsub.f32(float [[A:%.*]], float 0.000000e+00, metadata !"round.downward", metadata !"fpexcept.ignore") #[[ATTR0]]
+; CHECK-NEXT:    ret float [[RET]]
+;
+  %ret = call float @llvm.experimental.constrained.fsub.f32(float %a, float 0.0, metadata !"round.downward", metadata !"fpexcept.ignore") #0
+  ret float %ret
+}
+
+; Dynamic rounding means the rounding mode might be to -inf:
+; Round to -inf and if x is zero then the result is -0.0: must not fire
+define float @fsub_x_p0_dynamic(float %a) #0 {
+; CHECK-LABEL: @fsub_x_p0_dynamic(
+; CHECK-NEXT:    [[RET:%.*]] = call float @llvm.experimental.constrained.fsub.f32(float [[A:%.*]], float 0.000000e+00, metadata !"round.dynamic", metadata !"fpexcept.ignore") #[[ATTR0]]
+; CHECK-NEXT:    ret float [[RET]]
+;
+  %ret = call float @llvm.experimental.constrained.fsub.f32(float %a, float 0.0, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
+  ret float %ret
+}
+
+; With nsz we don't have to worry about -0.0 so the transform is valid.
+define float @fsub_nsz_x_p0_neginf(float %a) #0 {
+; CHECK-LABEL: @fsub_nsz_x_p0_neginf(
+; CHECK-NEXT:    ret float [[A:%.*]]
+;
+  %ret = call nsz float @llvm.experimental.constrained.fsub.f32(float %a, float 0.0, metadata !"round.downward", metadata !"fpexcept.ignore") #0
+  ret float %ret
+}
+
+; With nsz we don't have to worry about -0.0 so the transform is valid.
+define float @fsub_nsz_x_p0_dynamic(float %a) #0 {
+; CHECK-LABEL: @fsub_nsz_x_p0_dynamic(
+; CHECK-NEXT:    ret float [[A:%.*]]
+;
+  %ret = call nsz float @llvm.experimental.constrained.fsub.f32(float %a, float 0.0, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
+  ret float %ret
+}
+
 ;
 ; fsub X, -0 ==> X, when we know X is not -0
 ; (fast math flag: nsz)