[llvm] 560292f - [AMDGPU] Update isFMAFasterThanFMulAndFAdd assumptions

[Carl Ritson via llvm-commits]

Author: Carl Ritson
Date: 2020-07-07T15:40:44+09:00
New Revision: 560292fa990a2bfcf8415f07a166393beff667f6

URL: https://github.com/llvm/llvm-project/commit/560292fa990a2bfcf8415f07a166393beff667f6
DIFF: https://github.com/llvm/llvm-project/commit/560292fa990a2bfcf8415f07a166393beff667f6.diff

LOG: [AMDGPU] Update isFMAFasterThanFMulAndFAdd assumptions

MAD/MAC is no longer always available.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D83207

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/SIISelLowering.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index d90272848500..79204180540f 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -4277,10 +4277,13 @@ bool SITargetLowering::isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
 
   switch (VT.getSimpleVT().SimpleTy) {
   case MVT::f32: {
-    // This is as fast on some subtargets. However, we always have full rate f32
-    // mad available which returns the same result as the separate operations
-    // which we should prefer over fma. We can't use this if we want to support
-    // denormals, so only report this in these cases.
+    // If mad is not available this depends only on if f32 fma is full rate.
+    if (!Subtarget->hasMadMacF32Insts())
+      return Subtarget->hasFastFMAF32();
+
+    // Otherwise f32 mad is always full rate and returns the same result as
+    // the separate operations so should be preferred over fma.
+    // However does not support denomals.
     if (hasFP32Denormals(MF))
       return Subtarget->hasFastFMAF32() || Subtarget->hasDLInsts();