[PATCH] D63941: [Float2Int] Add support for unary FNeg to Float2Int

[Cameron McInally via Phabricator via llvm-commits]

cameron.mcinally created this revision.
cameron.mcinally added reviewers: spatel, arsenm, andrew.w.kaylor, craig.topper, kpn, jmolloy.
Herald added subscribers: llvm-commits, hiraditya, wdng.
Herald added a project: LLVM.

I'm not that familiar with ConstantRange or F2I, so a thorough review will be appreciated.

Note that I chose to fake a binary FNeg for ConstantRange purposes. I think this is safe and should limit regressions. Opposing views welcome...


Repository:
  rL LLVM

https://reviews.llvm.org/D63941

Files:
  llvm/lib/Transforms/Scalar/Float2Int.cpp
  llvm/test/Transforms/Float2Int/basic.ll


Index: llvm/test/Transforms/Float2Int/basic.ll
===================================================================
--- llvm/test/Transforms/Float2Int/basic.ll
+++ llvm/test/Transforms/Float2Int/basic.ll
@@ -80,12 +80,11 @@
 }
 
 ; CHECK-LABEL: @simple6
-; CHECK:  %1 = uitofp i8 %a to float
-; CHECK:  %2 = uitofp i8 %b to float
-; CHECK:  %3 = fneg float %1
-; CHECK:  %4 = fmul float %3, %2
-; CHECK:  %5 = fptoui float %4 to i32
-; CHECK:  ret i32 %5
+; CHECK:  %1 = zext i8 %a to i32
+; CHECK:  %2 = zext i8 %b to i32
+; CHECK:  %3 = sub i32 0, %1
+; CHECK:  %4 = mul i32 %3, %2
+; CHECK:  ret i32 %4
 define i32 @simple6(i8 %a, i8 %b) {
   %1 = uitofp i8 %a to float
   %2 = uitofp i8 %b to float
Index: llvm/lib/Transforms/Scalar/Float2Int.cpp
===================================================================
--- llvm/lib/Transforms/Scalar/Float2Int.cpp
+++ llvm/lib/Transforms/Scalar/Float2Int.cpp
@@ -200,6 +200,7 @@
       continue;
     }
 
+    case Instruction::FNeg:
     case Instruction::FAdd:
     case Instruction::FSub:
     case Instruction::FMul:
@@ -240,6 +241,16 @@
     case Instruction::SIToFP:
       llvm_unreachable("Should have been handled in walkForwards!");
 
+    case Instruction::FNeg:
+      Op = [I](ArrayRef<ConstantRange> Ops) {
+        assert(Ops.size() == 1 && "its a unary operator!");
+        auto BinOp = (Instruction::BinaryOps) Instruction::FSub;
+        unsigned Size = Ops[0].getBitWidth();
+        auto Zero = ConstantRange(APInt::getNullValue(Size));
+        return Zero.binaryOp(BinOp, Ops[0]);
+      };
+      break;
+
     case Instruction::FAdd:
     case Instruction::FSub:
     case Instruction::FMul:
@@ -466,6 +477,10 @@
     NewV = IRB.CreateSExtOrTrunc(NewOperands[0], ToTy);
     break;
 
+  case Instruction::FNeg:
+    NewV = IRB.CreateNeg(NewOperands[0], I->getName());
+    break;
+
   case Instruction::FAdd:
   case Instruction::FSub:
   case Instruction::FMul:


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D63941.207090.patch
Type: text/x-patch
Size: 1937 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190628/9f39e3fd/attachment.bin>