[PATCH] D69231: AMDGPU/GlobalISel: Legalize fast unsafe FDIV

Mon Oct 21 08:51:49 PDT 2019

arsenm added inline comments.

================
Comment at: llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp:1832
+}
+
+bool AMDGPULegalizerInfo::legalizeFastUnsafeFDIV(MachineInstr &MI,
----------------
I’m debating if this should really go in the combiner, but this is fine for now

================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:7550
   if (const ConstantFPSDNode *CLHS = dyn_cast<ConstantFPSDNode>(LHS)) {
-    if (Unsafe || VT == MVT::f32 || VT == MVT::f16) {
-      if (CLHS->isExactlyValue(1.0)) {
----------------
Unrelated change 

================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fast-unsafe-fdiv.mir:2
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -enable-unsafe-fp-math -global-isel-abort=0 -o - %s  | FileCheck -check-prefix=GFX9-UNSAFE %s
+
----------------
I don’t think this needs -global-isel-abort

================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fast-unsafe-fdiv.mir:46
+
+---
+name: test_fast_unsafe_fdiv_s32_arcp
----------------
This test isn’t doing much with the global flag enabled

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69231/new/

https://reviews.llvm.org/D69231