Optimize square root squared (PR21126).

When unsafe-fp-math is enabled, we can turn sqrt(X) * sqrt(X) into X.

This can happen in the real world when calculating x ** 3/2. This occurs
in test-suite/SingleSource/Benchmarks/BenchmarkGame/n-body.c.

Differential Revision: http://reviews.llvm.org/D5584

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218906 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
index da5d00dd9a182b5b89f9fb021a4c3d08e3f42a87..0a19a4880f29f4791e228969adedf3c1ac7d450c 100644 (file)
--- a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
+++ b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
@@ -531,6 +531,11 @@ Instruction *InstCombiner::visitFMul(BinaryOperator &I) {
     }
   }
 
+  // sqrt(X) * sqrt(X) -> X
+  if (AllowReassociate && (Op0 == Op1))
+    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Op0))
+      if (II->getIntrinsicID() == Intrinsic::sqrt)
+        return ReplaceInstUsesWith(I, II->getOperand(0));
 
   // Under unsafe algebra do:
   // X * log2(0.5*Y) = X*log2(Y) - X

diff --git a/test/Transforms/InstCombine/fmul.ll b/test/Transforms/InstCombine/fmul.ll
index 18cbf9da53615f12ecc0cc4b30e52e1064e583da..3c8bab9ffb53e467df60781e00b5c841d2ac586c 100644 (file)
--- a/test/Transforms/InstCombine/fmul.ll
+++ b/test/Transforms/InstCombine/fmul.ll
@@ -123,3 +123,32 @@ define float @test11(float %x, float %y) {
 ; CHECK-NOT: fadd float
 ; CHECK: fadd fast float
 }
+
+; PR21126: http://llvm.org/bugs/show_bug.cgi?id=21126
+; With unsafe/fast math, sqrt(X) * sqrt(X) is just X.
+declare double @llvm.sqrt.f64(double)
+
+define double @sqrt_squared1(double %f) #0 {
+  %sqrt = call double @llvm.sqrt.f64(double %f)
+  %mul = fmul fast double %sqrt, %sqrt
+  ret double %mul
+; CHECK-LABEL: @sqrt_squared1(
+; CHECK-NEXT: ret double %f
+}
+
+; With unsafe/fast math, sqrt(X) * sqrt(X) is just X, 
+; but make sure another use of the sqrt is intact.
+; Note that the remaining fmul is altered but is not 'fast'
+; itself because it was not marked 'fast' originally. 
+; Thus, we have an overall fast result, but no more indication of
+; 'fast'ness in the code.
+define double @sqrt_squared2(double %f) #0 {
+  %sqrt = call double @llvm.sqrt.f64(double %f)
+  %mul1 = fmul fast double %sqrt, %sqrt
+  %mul2 = fmul double %mul1, %sqrt
+  ret double %mul2
+; CHECK-LABEL: @sqrt_squared2(
+; CHECK-NEXT: %sqrt = call double @llvm.sqrt.f64(double %f)
+; CHECK-NEXT: %mul2 = fmul double %sqrt, %f
+; CHECK-NEXT: ret double %mul2
+}