In finite-only mode, transform pow(x, 0.5) directly to fabs(sqrt(x)).

[davide]

Summary

This patch implements a TODO.

Currently, the expansion of pow(x, 0.5) always checks for + and - infinity. Instead if x is finite, we could simply transform it to fabs(sqrt(x)).

Patch by Mandeep Singh Grang.

[weimingz]

lib/Transforms/Utils/SimplifyLibCalls.cpp

1245

you are checking the op1, which is the 2nd parameter (e.g. 0.5), but the comments says "If x is finite".

[majnemer]

This change has no tests.

[mgrang]

lib/Transforms/Utils/SimplifyLibCalls.cpp

1245

Weimingz, Op1 is the first parameter and Op2 is the second. Look at lines 1180 and 1229:

if (Op1C->isExactlyValue(1.0)) // pow(1.0, x) -> 1.0

if (Op2C->getValueAPF().isZero()) // pow(x, 0.0) -> 1.0

[davide]

Are you still planning to work on this? Thanks, --Davide

[mgrang]

Hi Davide,

I got busy with other work so did not get time to look into it.
Please feel free to work on this.

Thanks,
Mandeep

lib/Transforms/Utils/SimplifyLibCalls.cpp

1245

Also :

Value *Op1 = CI->getArgOperand(0), *Op2 = CI->getArgOperand(1);

[davide]

Continuing somewhere else

[davide]

Continuing somewhere else

lib/Transforms/Utils/SimplifyLibCalls.cpp

if (Op2C->isExactlyValue(0.5) &&
     // In -ffast-math, pow(x, 0.5) -> sqrt(x).
     if (CI->hasUnsafeAlgebra()) {
       IRBuilder<>::FastMathFlagGuard Guard(B);
       B.setFastMathFlags(CI->getFastMathFlags());
       return emitUnaryFloatFnCall(Op1, TLI->getName(LibFunc::sqrt), B,
                                   Callee->getAttributes());
     }
 
-    // Expand pow(x, 0.5) to (x == -infinity ? +infinity : fabs(sqrt(x))).
+    // If x is finite, then pow(x, 0.5) ==> fabs(sqrt(x)).
+    // Else expand pow(x, 0.5) to (x == -infinity ? +infinity : fabs(sqrt(x))).
     // This is faster than calling pow, and still handles negative zero
     // and negative infinity correctly.
-    // TODO: In finite-only mode, this could be just fabs(sqrt(x)).
-    Value *Inf = ConstantFP::getInfinity(CI->getType());
-    Value *NegInf = ConstantFP::getInfinity(CI->getType(), true);
     Value *Sqrt = emitUnaryFloatFnCall(Op1, "sqrt", B, Callee->getAttributes());
     Value *FAbs =
-        emitUnaryFloatFnCall(Sqrt, "fabs", B, Callee->getAttributes());
+    if (ConstantFP *Op1C = dyn_cast<ConstantFP>(Op1))
+      if (Op1C->getValueAPF().isFinite())
+        return FAbs;
+    Value *Inf = ConstantFP::getInfinity(CI->getType());
+    Value *NegInf = ConstantFP::getInfinity(CI->getType(), true);
     Value *FCmp = B.CreateFCmpOEQ(Op1, NegInf);
     Value *Sel = B.CreateSelect(FCmp, Inf, FAbs);
     return Sel;
   }
 
   if (Op2C->isExactlyValue(1.0)) // pow(x, 1.0) -> x
     return Op1;
   if (Op2C->isExactlyValue(2.0)) // pow(x, 2.0) -> x*x