[PATCH] D122501: [AMDGPU] Enable PreRARematerialize scheduling pass with multiple high RP regions
Stanislav Mekhanoshin via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 1 11:44:03 PDT 2022
rampitec added inline comments.
================
Comment at: llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp:788
+ for (auto Idx : ImpactedRegions) {
+ NewPressure[Idx] = Pressure[Idx];
+ NewLiveIns[Idx] = LiveIns[Idx];
----------------
There can be a lot of regions, I'd better switch to a map indexed by Idx rather than live huge unused chunks of vectors.
================
Comment at: llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp:1000
+}
\ No newline at end of file
----------------
Return the newline.
================
Comment at: llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp:739
MachineInstr *UseI = &*MRI.use_instr_begin(Reg);
if (Def->getParent() == UseI->getParent())
----------------
vangthao wrote:
> rampitec wrote:
> > This also needs a nodbg use (use_nodbg_instructions).
> If there is a debug use before the first nondbg use, this will sink the MI after the debug use.
The point is: codegen with and without debug info shall not differ. With this code it will.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D122501/new/
https://reviews.llvm.org/D122501
More information about the llvm-commits
mailing list