[PATCH] D67767: [AMDGPU] Add llvm.amdgcn.wqm.demote intrinsic

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/include/llvm/IR/IntrinsicsAMDGPU.td:1267
+// Like ps.live, but cannot be moved by LICM.
+def int_amdgcn_wqm_helper : Intrinsic <[llvm_i1_ty], [], []>;
+
----------------
I assume this needs to be convergent


================
Comment at: llvm/include/llvm/IR/IntrinsicsAMDGPU.td:1485
+// If false, mark all active lanes as helper lanes until the end of program.
+def int_amdgcn_wqm_demote : Intrinsic<[], [llvm_i1_ty], []>;
+
----------------
Ditto, and nomem


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td:242
 def : SourceOfDivergence<int_amdgcn_ps_live>;
+def : SourceOfDivergence<int_amdgcn_wqm_helper>;
 def : SourceOfDivergence<int_amdgcn_ds_swizzle>;
----------------
Missing DivergenceAnalysis test


================
Comment at: llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp:778-779
+          break;
+        unsigned Src = MBBI->getOperand(1).getReg();
+        if (!Register::isVirtualRegister(Src) && MBB.isLiveIn(Src)) {
+          MBBI++;
----------------
use Register


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D67767/new/

https://reviews.llvm.org/D67767