[PATCH] D14045: [SimplifyLibCalls] Add a new transform: pow(exp(x), y) -> exp(x*y)

[David Majnemer via llvm-commits]

majnemer added a comment.

I'm not sure what the implications of this transform are with respect to correctness.  My naïve understanding is that this could produce different results because it removes a step where rounding was introduced.  I'd feel more comfortable with this if @scanon, @resistor and/or @escha could take a look.


================
Comment at: lib/Transforms/Utils/SimplifyLibCalls.cpp:1096
@@ +1095,3 @@
+  // pow(exp(x), y) -> exp(x*y)
+  if (CallInst *OpC = dyn_cast<CallInst>(Op1)) {
+    LibFunc::Func Func;
----------------
Please use `auto *OpC` here.

================
Comment at: lib/Transforms/Utils/SimplifyLibCalls.cpp:1102
@@ +1101,3 @@
+    if (TLI->getLibFunc(FuncName, Func) && TLI->has(Func) &&
+        Func == LibFunc::exp && Op2->getType()->isDoubleTy())
+      return EmitUnaryFloatFnCall(
----------------
Any reason in particular why we must limit this to `isDoubleTy`?

================
Comment at: lib/Transforms/Utils/SimplifyLibCalls.cpp:1103-1105
@@ +1102,5 @@
+        Func == LibFunc::exp && Op2->getType()->isDoubleTy())
+      return EmitUnaryFloatFnCall(
+        B.CreateFMul(OpC->getArgOperand(0), Op2, "mul"),
+        TLI->getName(LibFunc::exp), B, Callee->getAttributes());
+  }
----------------
Is this clang-format'd?

================
Comment at: lib/Transforms/Utils/SimplifyLibCalls.cpp:1105
@@ +1104,3 @@
+        B.CreateFMul(OpC->getArgOperand(0), Op2, "mul"),
+        TLI->getName(LibFunc::exp), B, Callee->getAttributes());
+  }
----------------
Wouldn't `TLI->getName(LibFunc::exp)` be equivalent to `FuncName` ?


http://reviews.llvm.org/D14045