[PATCH] D120544: [AMDGPU] Omit unnecessary waitcnt before barriers

[Jay Foad via Phabricator via llvm-commits]

foad added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/GCNSubtarget.h:496
+  /// \returns true if the target supports backing off of s_barrier instructions
+  //  when an exception is raised"
+  bool supportsBackOffBarrier() const {
----------------
Stray `"` character


================
Comment at: llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp:1139
+  // The subtarget may have an implicit S_WAITCNT 0 before barriers. If it does
+  // not, we need to ensure the subatarget is capable of backing off barrier
+  // instructions in case there are any outstanding memory operations that may
----------------
Typo "subatarget"


================
Comment at: llvm/test/CodeGen/AMDGPU/waitcnt-preexisting-vscnt.mir:38
     ; GFX10: GLOBAL_STORE_DWORD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec
-    ; GFX10: S_WAITCNT_VSCNT undef $sgpr_null, 0
+    ; GFX10: S_WAITCNT_VSCNT undef $sgpr_null, 1
     ; GFX10: S_BARRIER
----------------
Isn't this "waitcnt 1" redundant, since we have just done a "waitcnt 0" and then issued one store? Or am I completely misunderstanding?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D120544/new/

https://reviews.llvm.org/D120544