[SimplifyLibCalls] pow(x, -0.5) -> 1.0 / sqrt(x)

[davide]

Summary

Missing optimization pointed out by David earlier https://godbolt.org/g/PsWrfV

[majnemer]

lib/Transforms/Utils/SimplifyLibCalls.cpp

1084–1085

Doesn't hasUsnafeAlgebra make this irrelevant?

[efriedma]

lib/Transforms/Utils/SimplifyLibCalls.cpp

1089

Missing fast-math flags on sqrt() call and fdiv instruction.

[davide]

Oops, thanks for spotting the mistake, Eli.

[davide]

lib/Transforms/Utils/SimplifyLibCalls.cpp

1084–1085

Does it? I'm not entirely sure of the implications, see e.g. http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160808/380416.html @hfinkel
(I'd leave the lowering to an intrinsic as a follow-up).

1089

Fixed, my bad.

[majnemer]

lib/Transforms/Utils/SimplifyLibCalls.cpp

1084–1085

Of course... Sorry about that!

[majnemer]

LGTM

lib/Transforms/Utils/SimplifyLibCalls.cpp

 
   ConstantFP *Op2C = dyn_cast<ConstantFP>(Op2);
   if (!Op2C)
     return Ret;
 
   if (Op2C->getValueAPF().isZero()) // pow(x, 0.0) -> 1.0
     return ConstantFP::get(CI->getType(), 1.0);
 
+  if (Op2C->isExactlyValue(-0.5) &&
+      hasUnaryFloatFn(TLI, Op2->getType(), LibFunc::sqrt, LibFunc::sqrtf,
+                      LibFunc::sqrtl)) {
+    // If -ffast-math:
+    // pow(x, -0.5) -> 1.0 / sqrt(x)
+    if (CI->hasUnsafeAlgebra()) {
+      IRBuilder<>::FastMathFlagGuard Guard(B);
+      B.setFastMathFlags(CI->getFastMathFlags());
+
+      // Here we cannot lower to an intrinsic because C99 sqrt() and llvm.sqrt
+      // are not guaranteed to have the same semantics.
+      Value *Sqrt = emitUnaryFloatFnCall(Op1, TLI->getName(LibFunc::sqrt), B,
+                                         Callee->getAttributes());
+
+      return B.CreateFDiv(ConstantFP::get(CI->getType(), 1.0), Sqrt, "sqrtrecip");
+    }
+  }
+
   if (Op2C->isExactlyValue(0.5) &&
       hasUnaryFloatFn(TLI, Op2->getType(), LibFunc::sqrt, LibFunc::sqrtf,
                       LibFunc::sqrtl) &&
       hasUnaryFloatFn(TLI, Op2->getType(), LibFunc::fabs, LibFunc::fabsf,
                       LibFunc::fabsl)) {
 
     // In -ffast-math, pow(x, 0.5) -> sqrt(x).
     if (CI->hasUnsafeAlgebra()) {

test/Transforms/InstCombine/pow-sqrt.ll

 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
 define double @pow_half(double %x) {
   %pow = call fast double @llvm.pow.f64(double %x, double 5.000000e-01)
   ret double %pow
 }
 
 ; CHECK-LABEL: define double @pow_half(
 ; CHECK-NEXT:  %sqrt = call fast double @sqrt(double %x)
 ; CHECK-NEXT:  ret double %sqrt
 
-declare double @llvm.pow.f64(double, double)
+define double @pow_neghalf(double %x) {
+  %pow = call fast double @llvm.pow.f64(double %x, double -5.000000e-01)
+  ret double %pow
+}
 
+; CHECK-LABEL: define double @pow_neghalf(
+; CHECK-NEXT: %sqrt = call fast double @sqrt(double %x) #0
+; CHECK-NEXT: %sqrtrecip = fdiv fast double 1.000000e+00, %sqrt
+; CHECK-NEXT: ret double %sqrtrecip
+
+declare double @llvm.pow.f64(double, double)