From: Kyle Roarty <kyleroarty1716@gmail.com>
Date: Thu, 24 Dec 2020 01:30:37 +0000 (-0600)
Subject: gpu-compute: Fix FLAT insts decrementing lgkm count early
X-Git-Tag: develop-gem5-snapshot~315
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=f6ec145fc0;p=gem5.git

gpu-compute: Fix FLAT insts decrementing lgkm count early

FLAT instructions used to decrement lgkm count on execute, while the
GCN3 ISA specifies that lgkm count should be decremented on data being
returned or data being written.

This patch changes it so that lgkm is decremented after initiateAcc (for
stores) and after completeAcc (for loads) to better reflect the ISA
definition.

This fixes a bug where waitcnts would be satisfied even though the
memory access wasn't completed, which lead to instructions using the
wrong data.

Change-Id: I596cb031af9cda8d47a1b5e146e4a4ffd793d36c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38696
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
---

diff --git a/src/gpu-compute/global_memory_pipeline.cc b/src/gpu-compute/global_memory_pipeline.cc
index bcd93f886..a2b24e4b4 100644
--- a/src/gpu-compute/global_memory_pipeline.cc
+++ b/src/gpu-compute/global_memory_pipeline.cc
@@ -130,6 +130,9 @@ GlobalMemPipeline::exec()
         DPRINTF(GPUMem, "CU%d: WF[%d][%d]: Completing global mem instr %s\n",
                 m->cu_id, m->simdId, m->wfSlotId, m->disassemble());
         m->completeAcc(m);
+        if (m->isFlat() && m->isLoad()) {
+            w->decLGKMInstsIssued();
+        }
         w->decVMemInstsIssued();
 
         if (m->isLoad() || m->isAtomicRet()) {
@@ -193,6 +196,10 @@ GlobalMemPipeline::exec()
                 mp->disassemble(), mp->seqNum());
         mp->initiateAcc(mp);
 
+        if (mp->isFlat() && mp->isStore()) {
+            mp->wavefront()->decLGKMInstsIssued();
+        }
+
         if (mp->isStore() && mp->isGlobalSeg()) {
             mp->wavefront()->decExpInstsIssued();
         }
diff --git a/src/gpu-compute/gpu_dyn_inst.cc b/src/gpu-compute/gpu_dyn_inst.cc
index 03ed68951..38e4ecf49 100644
--- a/src/gpu-compute/gpu_dyn_inst.cc
+++ b/src/gpu-compute/gpu_dyn_inst.cc
@@ -819,7 +819,6 @@ GPUDynInst::resolveFlatSegment(const VectorMask &mask)
     if (executedAs() == Enums::SC_GLOBAL) {
         // no transormation for global segment
         wavefront()->execUnitId =  wavefront()->flatGmUnitId;
-        wavefront()->decLGKMInstsIssued();
         if (isLoad()) {
             wavefront()->rdLmReqsInPipe--;
         } else if (isStore()) {