[PATCH] D86253: [AMDGPU] Disable stall weight calculation in regbank-reassign
Stanislav Mekhanoshin via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Thu Aug 20 00:20:19 PDT 2020
rampitec added inline comments.
================
Comment at: llvm/lib/Target/AMDGPU/GCNRegBankReassign.cpp:605
unsigned Weight = getOperandGatherWeight(MI, Reg1, Reg2, StallCycles);
Weight += MLI->getLoopDepth(MI.getParent()) * 10;
----------------
rampitec wrote:
> critson wrote:
> > Presumably we don't need to do this if we are not using weights at all (i.e. no sorting)?
> > Perhaps "if (UseWeight)" all of this?
> Good point. It is cheap, but still.
Actually since it is cheap it makes sense to keep loop depth weight. The operand forwarding part is expensive, but the sort is not. I will experiment tomorrow and return sort and MLI part, just disable operand scan part.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D86253/new/
https://reviews.llvm.org/D86253
More information about the llvm-commits
mailing list