[PATCH] D35218: [AMDGPU] fcanonicalize elimination optimization

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:4638
+
+  case ISD::FCANONICALIZE:
+    return true;
----------------
fcanonicalize should have the idempotent optimization applied so there shouldn't be a need to handle it


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:4662
+  case ISD::FCOS:
+  case ISD::FSINCOS:
+    return Op.getValueType().getScalarType() != MVT::f16;
----------------
We don't handle this


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:4665-4668
+  // In pre-GFX9 targets V_MIN_F32 and others do not flush denorms.
+  // For such targets need to check their input recursively.
+  case ISD::FMINNUM:
+  case ISD::FMAXNUM:
----------------
The output is never flushed anywhere?


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:4669-4670
+  case ISD::FMAXNUM:
+  case ISD::FMINNAN:
+  case ISD::FMAXNAN:
+    if (ST->getGeneration() >= SISubtarget::GFX9)
----------------
We don't handle these


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:4671-4672
+  case ISD::FMAXNAN:
+    if (ST->getGeneration() >= SISubtarget::GFX9)
+      return true;
+
----------------
I don't think this is true, but should have a named check in the subtarget


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:4678-4680
+  case ISD::ConstantFP: {
+    auto F = cast<ConstantFPSDNode>(Op)->getValueAPF();
+    return !F.isDenormal() && !(F.isNaN() && F.isSignaling());
----------------
We constant fold this already so this shouldn't be necessary


https://reviews.llvm.org/D35218