[PATCH] D154438: [AMDGPU] Fix incorrect hazard mitigation

[Stephen Thomas via Phabricator via llvm-commits]

stepthomas created this revision.
Herald added subscribers: foad, kerbowa, hiraditya, tpr, dstuttard, yaxunl, jvesely, kzhuravl, arsenm.
Herald added a project: All.
stepthomas requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

GCNHazardRecognizer::fixVcmpxExecWARHazard() mitigates a specific hazard
by inserting a wait on sa_sdst==0 if such a wait isn't already present.
Unfortunately, the check for an existing wait incorrectly checks for one
that doesn't actually care about sa_sdst itself, but requires that no
other counters are waited for.

Once the check is performed correctly, a lit test needs to be updated,
since it is currently testing for the incorrect behaviour.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D154438

Files:
  llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
  llvm/test/CodeGen/AMDGPU/vcmpx-exec-war-hazard.mir


Index: llvm/test/CodeGen/AMDGPU/vcmpx-exec-war-hazard.mir
===================================================================
--- llvm/test/CodeGen/AMDGPU/vcmpx-exec-war-hazard.mir
+++ llvm/test/CodeGen/AMDGPU/vcmpx-exec-war-hazard.mir
@@ -126,12 +126,13 @@
     S_ENDPGM 0
 ...
 
-# GCN-LABEL: name: no_hazard_vcmpx_smov_exec_lo_depctr_ffff
-# GCN:      $sgpr0 = S_MOV_B32 $exec_lo
-# GCN-NEXT: S_WAITCNT_DEPCTR 65535
-# GCN-NEXT: V_CMPX_LE_F32_nosdst_e32
+# GCN-LABEL: name: hazard_vcmpx_smov_exec_lo_depctr_ffff
+# GCN:        $sgpr0 = S_MOV_B32 $exec_lo
+# GCN-NEXT:   S_WAITCNT_DEPCTR 65535
+# GFX10-NEXT: S_WAITCNT_DEPCTR 65534
+# GCN-NEXT:   V_CMPX_LE_F32_nosdst_e32
 ---
-name:            no_hazard_vcmpx_smov_exec_lo_depctr_ffff
+name:            hazard_vcmpx_smov_exec_lo_depctr_ffff
 body:            |
   bb.0:
     successors: %bb.1
@@ -148,7 +149,6 @@
 # GCN-LABEL: name: hazard_vcmpx_smov_exec_lo_depctr_effe
 # GCN:        $sgpr0 = S_MOV_B32 $exec_lo
 # GCN-NEXT:   S_WAITCNT_DEPCTR 61438
-# GFX10-NEXT: S_WAITCNT_DEPCTR 65534
 # GCN-NEXT:   V_CMPX_LE_F32_nosdst_e32
 ---
 name:            hazard_vcmpx_smov_exec_lo_depctr_effe
Index: llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -1293,8 +1293,7 @@
           return true;
     }
     if (MI.getOpcode() == AMDGPU::S_WAITCNT_DEPCTR &&
-        AMDGPU::DepCtr::encodeFieldSaSdst(MI.getOperand(0).getImm(), 0) ==
-            0xfffe)
+        AMDGPU::DepCtr::decodeFieldSaSdst(MI.getOperand(0).getImm()) == 0)
       return true;
     return false;
   };


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D154438.537069.patch
Type: text/x-patch
Size: 1694 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230704/9fca2b65/attachment.bin>