Matthew Poremba has submitted this change. (
https://gem5-review.googlesource.com/c/public/gem5/+/71418?usp=email )
Change subject: configs,gpu-compute: Kernel dispatch-based exit events
......................................................................
configs,gpu-compute: Kernel dispatch-based exit events
Add two kernel dispatch-based exit events that are useful for limiting
the simulation and enabling debug flags at specific GPU kernels. Since
the KVM CPU typically used with GPUFS is not deterministic, this help
with enabling debug flags when the Tick number may vary. The exit at GPU
kernel option can also limit simulation by only simulating a few hundred
kernels, for example, and exit at a determined point.
Change-Id: I81bae92a80c25fc38c41e999aa662e1417b7a20d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/71418
Maintainer: Matt Sinclair <mattdsincl...@gmail.com>
Tested-by: kokoro <noreply+kok...@google.com>
Reviewed-by: Matt Sinclair <mattdsincl...@gmail.com>
---
M configs/example/gpufs/runfs.py
M src/gpu-compute/dispatcher.cc
2 files changed, 30 insertions(+), 0 deletions(-)
Approvals:
Matt Sinclair: Looks good to me, approved; Looks good to me, approved
kokoro: Regressions pass
diff --git a/configs/example/gpufs/runfs.py b/configs/example/gpufs/runfs.py
index f8ef70d..01203bb 100644
--- a/configs/example/gpufs/runfs.py
+++ b/configs/example/gpufs/runfs.py
@@ -137,6 +137,20 @@
"MI200 (gfx90a)",
)
+ parser.add_argument(
+ "--debug-at-gpu-kernel",
+ type=int,
+ default=-1,
+ help="Turn on debug flags starting with this kernel",
+ )
+
+ parser.add_argument(
+ "--exit-at-gpu-kernel",
+ type=int,
+ default=-1,
+ help="Exit simulation after running this many kernels",
+ )
+
def runGpuFSSystem(args):
"""
@@ -184,6 +198,9 @@
print("Running the simulation")
sim_ticks = args.abs_max_tick
+ kernels_launched = 0
+ if args.debug_at_gpu_kernel != -1:
+ m5.trace.disable()
exit_event = m5.simulate(sim_ticks)
@@ -199,11 +216,21 @@
assert args.checkpoint_dir is not None
m5.checkpoint(args.checkpoint_dir)
break
+ elif "GPU Kernel Completed" in exit_event.getCause():
+ kernels_launched += 1
else:
print(
f"Unknown exit event: {exit_event.getCause()}.
Continuing..."
)
+ if kernels_launched == args.debug_at_gpu_kernel:
+ m5.trace.enable()
+ if kernels_launched == args.exit_at_gpu_kernel:
+ print(f"Exiting @ GPU kernel {kernels_launched}")
+ break
+
+ exit_event = m5.simulate(sim_ticks - m5.curTick())
+
print(
"Exiting @ tick %i because %s" % (m5.curTick(),
exit_event.getCause())
)
diff --git a/src/gpu-compute/dispatcher.cc b/src/gpu-compute/dispatcher.cc
index a76ba7c..b19bccc 100644
--- a/src/gpu-compute/dispatcher.cc
+++ b/src/gpu-compute/dispatcher.cc
@@ -40,6 +40,7 @@
#include "gpu-compute/hsa_queue_entry.hh"
#include "gpu-compute/shader.hh"
#include "gpu-compute/wavefront.hh"
+#include "sim/sim_exit.hh"
#include "sim/syscall_emul_buf.hh"
#include "sim/system.hh"
@@ -330,6 +331,8 @@
DPRINTF(GPUWgLatency, "Kernel Complete ticks:%d kernel:%d\n",
curTick(), kern_id);
DPRINTF(GPUKernelInfo, "Completed kernel %d\n", kern_id);
+
+ exitSimLoop("GPU Kernel Completed");
}
if (!tickEvent.scheduled()) {
--
To view, visit
https://gem5-review.googlesource.com/c/public/gem5/+/71418?usp=email
To unsubscribe, or for help writing mail filters, visit
https://gem5-review.googlesource.com/settings?usp=email
Gerrit-MessageType: merged
Gerrit-Project: public/gem5
Gerrit-Branch: develop
Gerrit-Change-Id: I81bae92a80c25fc38c41e999aa662e1417b7a20d
Gerrit-Change-Number: 71418
Gerrit-PatchSet: 2
Gerrit-Owner: Matthew Poremba <matthew.pore...@amd.com>
Gerrit-Reviewer: Jason Lowe-Power <ja...@lowepower.com>
Gerrit-Reviewer: Matt Sinclair <mattdsincl...@gmail.com>
Gerrit-Reviewer: Matthew Poremba <matthew.pore...@amd.com>
Gerrit-Reviewer: kokoro <noreply+kok...@google.com>
Gerrit-CC: VISHNU RAMADAS <vrama...@wisc.edu>
Gerrit-CC: kokoro <noreply+kok...@google.com>
_______________________________________________
gem5-dev mailing list -- gem5-dev@gem5.org
To unsubscribe send an email to gem5-dev-le...@gem5.org
