[PATCH] D123835: AMDGPU/SDAG: Refine the fold to v_mad_[iu]64_[iu]32
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 25 13:30:56 PDT 2022
arsenm added inline comments.
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:10696
- SDValue MulLHS = LHS.getOperand(0);
- SDValue MulRHS = LHS.getOperand(1);
- SDValue AddRHS = RHS;
+ // Avoid the fold if it would unduly increase the number of multiply due to
+ // multiple uses, except on hardware with full-rate multiply-add (which is
----------------
s/number of multiply/number of multiplies/
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:10702
+ for (auto I = LHS->use_begin(), E = LHS->use_end(); I != E; ++I) {
+ if (I.getUse().getResNo() != 0)
+ continue;
----------------
I don't understand why you're checking this if you bail on not ISD:ADD. I guess it would make sense if you were handling the carry out adds in a separate patch?
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:10747
+
+ SDValue Shift = DAG.getShiftAmountConstant(32, MVT::i64, SL);
+ SDValue AccumLo = DAG.getNode(ISD::TRUNCATE, SL, MVT::i32, Accum);
----------------
Could just hardcode this to i32 instead of going through getShiftAmountConstant
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:10752-10757
+ if (!MulLHSUnsigned32) {
+ SDValue MulLHSHi = DAG.getNode(ISD::SRL, SL, MVT::i64, MulLHS, Shift);
+ MulLHSHi = DAG.getNode(ISD::TRUNCATE, SL, MVT::i32, MulLHSHi);
+ SDValue MulHi = DAG.getNode(ISD::MUL, SL, MVT::i32, MulLHSHi, MulRHSLo);
+ AccumHi = DAG.getNode(ISD::ADD, SL, MVT::i32, MulHi, AccumHi);
+ }
----------------
A comment with the DAG formed here would be helpful
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:10766
+
+ AccumHi = DAG.getNode(ISD::ANY_EXTEND, SL, MVT::i64, AccumHi);
+ AccumHi = DAG.getNode(ISD::SHL, SL, MVT::i64, AccumHi, Shift);
----------------
Ditto
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D123835/new/
https://reviews.llvm.org/D123835
More information about the llvm-commits
mailing list