[llvm] e151434 - [AMDGPU][InsertWaitCnts][NFC] Merge VMEM_ACCESS and VMEM_READ_ACCESS into a single event type (#171973)

[via llvm-commits]

Author: Pankaj Dwivedi
Date: 2025-12-12T18:26:40+05:30
New Revision: e151434b0f059f844a4e4e73812083f5a34e6524

URL: https://github.com/llvm/llvm-project/commit/e151434b0f059f844a4e4e73812083f5a34e6524
DIFF: https://github.com/llvm/llvm-project/commit/e151434b0f059f844a4e4e73812083f5a34e6524.diff

LOG: [AMDGPU][InsertWaitCnts][NFC] Merge VMEM_ACCESS and VMEM_READ_ACCESS into a single event type (#171973)

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
index 7c0525b9c9957..0a743dbc2e258 100644
--- a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
@@ -144,8 +144,7 @@ struct HardwareLimits {
 };
 
 #define AMDGPU_DECLARE_WAIT_EVENTS(DECL)                                       \
-  DECL(VMEM_ACCESS)              /* vmem read & write */                       \
-  DECL(VMEM_READ_ACCESS)         /* vmem read */                               \
+  DECL(VMEM_ACCESS) /* vmem read & write (pre-gfx10), vmem read (gfx10+) */    \
   DECL(VMEM_SAMPLER_READ_ACCESS) /* vmem SAMPLER read (gfx12+ only) */         \
   DECL(VMEM_BVH_READ_ACCESS)     /* vmem BVH read (gfx12+ only) */             \
   DECL(VMEM_WRITE_ACCESS)        /* vmem write that is not scratch */          \
@@ -369,8 +368,8 @@ class WaitcntGeneratorPreGFX12 : public WaitcntGenerator {
     assert(ST);
 
     static const unsigned WaitEventMaskForInstPreGFX12[NUM_INST_CNTS] = {
-        eventMask({VMEM_ACCESS, VMEM_READ_ACCESS, VMEM_SAMPLER_READ_ACCESS,
-                   VMEM_BVH_READ_ACCESS}),
+        eventMask(
+            {VMEM_ACCESS, VMEM_SAMPLER_READ_ACCESS, VMEM_BVH_READ_ACCESS}),
         eventMask({SMEM_ACCESS, LDS_ACCESS, GDS_ACCESS, SQ_MESSAGE}),
         eventMask({EXP_GPR_LOCK, GDS_GPR_LOCK, VMW_GPR_LOCK, EXP_PARAM_ACCESS,
                    EXP_POS_ACCESS, EXP_LDS_ACCESS}),
@@ -403,7 +402,7 @@ class WaitcntGeneratorGFX12Plus : public WaitcntGenerator {
     assert(ST);
 
     static const unsigned WaitEventMaskForInstGFX12Plus[NUM_INST_CNTS] = {
-        eventMask({VMEM_ACCESS, VMEM_READ_ACCESS}),
+        eventMask({VMEM_ACCESS}),
         eventMask({LDS_ACCESS, GDS_ACCESS}),
         eventMask({EXP_GPR_LOCK, GDS_GPR_LOCK, VMW_GPR_LOCK, EXP_PARAM_ACCESS,
                    EXP_POS_ACCESS, EXP_LDS_ACCESS}),
@@ -537,7 +536,7 @@ class SIInsertWaitcnts {
     switch (Inst.getOpcode()) {
     // FIXME: GLOBAL_INV needs to be tracked with xcnt too.
     case AMDGPU::GLOBAL_INV:
-      return VMEM_READ_ACCESS; // tracked using loadcnt
+      return VMEM_ACCESS; // tracked using loadcnt
     case AMDGPU::GLOBAL_WB:
     case AMDGPU::GLOBAL_WBINV:
       return VMEM_WRITE_ACCESS; // tracked using storecnt
@@ -547,7 +546,7 @@ class SIInsertWaitcnts {
 
     // Maps VMEM access types to their corresponding WaitEventType.
     static const WaitEventType VmemReadMapping[NUM_VMEM_TYPES] = {
-        VMEM_READ_ACCESS, VMEM_SAMPLER_READ_ACCESS, VMEM_BVH_READ_ACCESS};
+        VMEM_ACCESS, VMEM_SAMPLER_READ_ACCESS, VMEM_BVH_READ_ACCESS};
 
     assert(SIInstrInfo::isVMEM(Inst));
     // LDS DMA loads are also stores, but on the LDS side. On the VMEM side
@@ -561,7 +560,7 @@ class SIInsertWaitcnts {
       return VMEM_WRITE_ACCESS;
     }
     if (!ST->hasExtendedWaitCounts() || SIInstrInfo::isFLAT(Inst))
-      return VMEM_READ_ACCESS;
+      return VMEM_ACCESS;
     return VmemReadMapping[getVmemType(Inst)];
   }