[AMDGPU] Omit image waits in function prologue on gfx1250

[rampitec]

No description provided.

[rampitec]

• [AMDGPU] Omit image waits in function prologue on gfx1250 #145097 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Stanislav Mekhanoshin (rampitec)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/145097.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp (+4)

diff --git a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
index f7b88bf2d5ebc..a60e2102d4e8c 100644
--- a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
@@ -2681,6 +2681,10 @@ bool SIInsertWaitcnts::run(MachineFunction &MF) {
         if (CT == LOAD_CNT || CT == DS_CNT || CT == STORE_CNT)
           continue;
 
+        if (!ST->hasImageInsts() &&
+            (CT == EXP_CNT || CT == SAMPLE_CNT || CT == BVH_CNT))
+          continue;
+
         BuildMI(EntryBB, I, DebugLoc(),
                 TII->get(instrsForExtendedCounterTypes[CT]))
             .addImm(0);

[jayfoad]

Using hasImageInsts to control EXP_CNT looks weird - will this be changed again later?

[rampitec]

We may add a feature, but the problem with it we do not like negative features. So a FeatureEXPInsts will need to be added almost everywhere. Sound very invasive.

[jayfoad]

Well I guess it's OK for now