[llvm] 7deea9d - [AMDGPU] Move WaitcntBrackets::simplifyXcnt near other simplify functions. NFC. (#178673)

[via llvm-commits]

Author: Jay Foad
Date: 2026-01-29T15:49:11Z
New Revision: 7deea9db700a26c2b15afebe573b914febe030ac

URL: https://github.com/llvm/llvm-project/commit/7deea9db700a26c2b15afebe573b914febe030ac
DIFF: https://github.com/llvm/llvm-project/commit/7deea9db700a26c2b15afebe573b914febe030ac.diff

LOG: [AMDGPU] Move WaitcntBrackets::simplifyXcnt near other simplify functions. NFC. (#178673)

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 ee59a2e59d4a3..27f4ac389d437 100644
--- a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
@@ -1325,6 +1325,27 @@ void WaitcntBrackets::simplifyWaitcnt(InstCounterType T,
     Count = ~0u;
 }
 
+void WaitcntBrackets::simplifyXcnt(const AMDGPU::Waitcnt &CheckWait,
+                                   AMDGPU::Waitcnt &UpdateWait) const {
+  // Try to simplify xcnt further by checking for joint kmcnt and loadcnt
+  // optimizations. On entry to a block with multiple predescessors, there may
+  // be pending SMEM and VMEM events active at the same time.
+  // In such cases, only clear one active event at a time.
+  // TODO: Revisit xcnt optimizations for gfx1250.
+  // Wait on XCNT is redundant if we are already waiting for a load to complete.
+  // SMEM can return out of order, so only omit XCNT wait if we are waiting till
+  // zero.
+  if (CheckWait.KmCnt == 0 && hasPendingEvent(SMEM_GROUP))
+    UpdateWait.XCnt = ~0u;
+  // If we have pending store we cannot optimize XCnt because we do not wait for
+  // stores. VMEM loads retun in order, so if we only have loads XCnt is
+  // decremented to the same number as LOADCnt.
+  if (CheckWait.LoadCnt != ~0u && hasPendingEvent(VMEM_GROUP) &&
+      !hasPendingEvent(STORE_CNT) && CheckWait.XCnt >= CheckWait.LoadCnt)
+    UpdateWait.XCnt = ~0u;
+  simplifyWaitcnt(X_CNT, UpdateWait.XCnt);
+}
+
 void WaitcntBrackets::simplifyVmVsrc(const AMDGPU::Waitcnt &CheckWait,
                                      AMDGPU::Waitcnt &UpdateWait) const {
   // Waiting for some counters implies waiting for VM_VSRC, since an
@@ -1455,27 +1476,6 @@ void WaitcntBrackets::applyWaitcnt(InstCounterType T, unsigned Count) {
   }
 }
 
-void WaitcntBrackets::simplifyXcnt(const AMDGPU::Waitcnt &CheckWait,
-                                   AMDGPU::Waitcnt &UpdateWait) const {
-  // Try to simplify xcnt further by checking for joint kmcnt and loadcnt
-  // optimizations. On entry to a block with multiple predescessors, there may
-  // be pending SMEM and VMEM events active at the same time.
-  // In such cases, only clear one active event at a time.
-  // TODO: Revisit xcnt optimizations for gfx1250.
-  // Wait on XCNT is redundant if we are already waiting for a load to complete.
-  // SMEM can return out of order, so only omit XCNT wait if we are waiting till
-  // zero.
-  if (CheckWait.KmCnt == 0 && hasPendingEvent(SMEM_GROUP))
-    UpdateWait.XCnt = ~0u;
-  // If we have pending store we cannot optimize XCnt because we do not wait for
-  // stores. VMEM loads retun in order, so if we only have loads XCnt is
-  // decremented to the same number as LOADCnt.
-  if (CheckWait.LoadCnt != ~0u && hasPendingEvent(VMEM_GROUP) &&
-      !hasPendingEvent(STORE_CNT) && CheckWait.XCnt >= CheckWait.LoadCnt)
-    UpdateWait.XCnt = ~0u;
-  simplifyWaitcnt(X_CNT, UpdateWait.XCnt);
-}
-
 // Where there are multiple types of event in the bracket of a counter,
 // the decrement may go out of order.
 bool WaitcntBrackets::counterOutOfOrder(InstCounterType T) const {