InstCombine: Add constant folding/simplify for amdgcn.ldexp intrinsic

This really belongs in InstructionSimplify since it doesn't introduce
new instructions. Put it in instcombine to avoid increasing the number
of passes considering target intrinsics.

I also noticed that we seem to now be interpreting strictfp attributes
on call sites, so try to handle that.

1 files changed, 11 insertions, 0 deletions

diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index b2bde95..4fdc73c 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1457,6 +1457,7 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
   case Intrinsic::amdgcn_cubetc:
   case Intrinsic::amdgcn_fmul_legacy:
   case Intrinsic::amdgcn_fract:
+  case Intrinsic::amdgcn_ldexp:
   case Intrinsic::x86_sse_cvtss2si:
   case Intrinsic::x86_sse_cvtss2si64:
   case Intrinsic::x86_sse_cvttss2si:
@@ -2224,6 +2225,16 @@ static Constant *ConstantFoldScalarCall2(StringRef Name,
         return ConstantFP::get(Ty->getContext(),
                                APFloat((double)std::pow((double)Op1V,
                                                  (int)Op2C->getZExtValue())));
+
+      if (IntrinsicID == Intrinsic::amdgcn_ldexp) {
+        // FIXME: Should flush denorms depending on FP mode, but that's ignored
+        // everywhere else.
+
+        // scalbn is equivalent to ldexp with float radix 2
+        APFloat Result = scalbn(Op1->getValueAPF(), Op2C->getSExtValue(),
+                                APFloat::rmNearestTiesToEven);
+        return ConstantFP::get(Ty->getContext(), Result);
+      }
     }
     return nullptr;
   }