[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:

> On the other hand VA_VDST_RD is a "made up" counter that has no hardware equivalent,

No, I completely disagree with this description. There is one hardware counter. The _only_ reason for having two separate `InstCounterType`s is that we want to remember two different values of this counter for each VGPR. But neither of these two uses is more "real" or "made up" than the other.

> I don't want to block this, feel free to merge. We can revisit the names if needed in a future patch.

Thanks :)

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