[llvm] [AMDGPU] Handle expert scheduling VA_VDST WAR hazards (PR #201619)

[Jay Foad via llvm-commits]

================
@@ -32,8 +32,9 @@ enum InstCounterType {
   ASYNC_CNT,                         // gfx1250.
   TENSOR_CNT,                        // gfx1250.
   NUM_EXTENDED_INST_CNTS,
-  VA_VDST = NUM_EXTENDED_INST_CNTS, // gfx12+ expert mode only.
-  VM_VSRC,                          // gfx12+ expert mode only.
+  VA_VDST_RD = NUM_EXTENDED_INST_CNTS, // gfx12+ expert mode only.
+  VA_VDST_WR,                          // gfx12+ expert mode only.
----------------
jayfoad wrote:

To @Pierre-vh @vporpo and anyone else interested in refactoring of SIInsertWaitcnts, I just wanted to explain what is going on here: there is only a single VA_VDST counter in the hardware, but in WaitcntBrackets I need to be able to track two different scores for each register, representing the VA_VDST count since it was last read and the VA_VDST count since it was last written to.

https://github.com/llvm/llvm-project/pull/201619