commit cb73fe195955afc4d4b15584159ec8def781448a
author Kyle Roarty <kyleroarty1716@gmail.com> Thu Jul 01 21:53:46 2021 -0500
committer Matt Sinclair <mattdsinclair@gmail.com> Fri Jul 09 03:49:29 2021 +0000
tree 00e401317523b898b54721cf0dbacc6af847e4c1
parent 8134902e8b8e1d5953845b0b73782f4cec091a8a

arch-vega: Fix s_endpgm instruction

Copy over changes that had been made to s_engpgm in GCN3
but weren't added to the Vega implementation

Change-Id: I1063f83b1ce8f7c5e451c8c227265715c8f725b9
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/47519
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Reviewed-by: Alex Dutu <alexandru.dutu@amd.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

src/arch/amdgpu/vega/insts/instructions.cc

diff --git a/src/arch/amdgpu/vega/insts/instructions.cc b/src/arch/amdgpu/vega/insts/instructions.cc
index 47ea892..6e8c854 100644
--- a/src/arch/amdgpu/vega/insts/instructions.cc
+++ b/src/arch/amdgpu/vega/insts/instructions.cc
@@ -4137,7 +4137,12 @@
         ComputeUnit *cu = gpuDynInst->computeUnit();
 
         // delete extra instructions fetched for completed work-items
-        wf->instructionBuffer.clear();
+        wf->instructionBuffer.erase(wf->instructionBuffer.begin() + 1,
+            wf->instructionBuffer.end());
+
+        if (wf->pendingFetch) {
+            wf->dropFetch = true;
+        }
 
         wf->computeUnit->fetchStage.fetchUnit(wf->simdId)
             .flushBuf(wf->wfSlotId);
@@ -4215,8 +4220,11 @@
             bool kernelEnd =
                 wf->computeUnit->shader->dispatcher().isReachingKernelEnd(wf);
 
+            bool relNeeded =
+                wf->computeUnit->shader->impl_kern_end_rel;
+
             //if it is not a kernel end, then retire the workgroup directly
-            if (!kernelEnd) {
+            if (!kernelEnd || !relNeeded) {
                 wf->computeUnit->shader->dispatcher().notifyWgCompl(wf);
                 wf->setStatus(Wavefront::S_STOPPED);
                 wf->computeUnit->completedWGs++;
@@ -4232,6 +4240,7 @@
              * the complex
              */
             setFlag(MemSync);
+            setFlag(GlobalSegment);
             // Notify Memory System of Kernel Completion
             // Kernel End = isKernel + isMemSync
             wf->setStatus(Wavefront::S_RETURNING);