From 8e268d42e2260dfc6200cda723e269b6473ed8fb Mon Sep 17 00:00:00 2001
From: Jarvis Jia <jia44@wisc.edu>
Date: Mon, 10 Jun 2024 20:56:08 -0500
Subject: [PATCH] gpu-compute: Provided m5ops support for gpu

Adding m5 stat dump and reset into python script through different exit
event

Change-Id: I662233ae71e2987d90af1fd0100e29036b2ef1c6
---
 configs/example/apu_se.py | 45 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 44 insertions(+), 1 deletion(-)

diff --git a/configs/example/apu_se.py b/configs/example/apu_se.py
index 2d3a849df0..6f23565f28 100644
--- a/configs/example/apu_se.py
+++ b/configs/example/apu_se.py
@@ -707,7 +707,7 @@ render_driver = GPURenderDriver(filename=f"dri/renderD{renderDriNum}")
 gpu_hsapp = HSAPacketProcessor(
     pioAddr=hsapp_gpu_map_paddr, numHWQueues=args.num_hw_queues
 )
-dispatcher = GPUDispatcher()
+dispatcher = GPUDispatcher(kernel_exit_events=True)
 gpu_cmd_proc = GPUCommandProcessor(hsapp=gpu_hsapp, dispatcher=dispatcher)
 gpu_driver.device = gpu_cmd_proc
 shader.dispatcher = dispatcher
@@ -834,6 +834,8 @@ if fast_forward:
 # configure the TLB hierarchy
 GPUTLBConfig.config_tlb_hierarchy(args, system, shader_idx)
 
+system.exit_on_work_items = True
+
 # create Ruby system
 system.piobus = IOXBar(
     width=32, response_latency=0, frontend_latency=0, forward_latency=0
@@ -1008,6 +1010,47 @@ if args.fast_forward:
 
 exit_event = m5.simulate(maxtick)
 
+while True:
+    if (
+        exit_event.getCause() == "m5_exit instruction encountered"
+        or exit_event.getCause() == "user interrupt received"
+        or exit_event.getCause() == "simulate() limit reached"
+        or  "exiting with last active thread context" in exit_event.getCause()
+    ):
+        print(f"breaking loop due to: {exit_event.getCause()}.")
+        break
+    elif "checkpoint" in exit_event.getCause():
+        assert args.checkpoint_dir is not None
+        m5.checkpoint(args.checkpoint_dir)
+        print("breaking loop with checkpoint")
+        break
+    elif "GPU Kernel Completed" in exit_event.getCause():
+        print("GPU Kernel Completed dump and reset")
+        m5.stats.dump()
+        m5.stats.reset()
+    elif "GPU Blit Kernel Completed" in exit_event.getCause():
+        print("GPU Blit Kernel Completed dump and reset")
+        m5.stats.dump()
+        m5.stats.reset()
+    elif "Skipping GPU Kernel" in exit_event.getCause():
+        print("Skipping GPU Kernel dump and reset")
+        m5.stats.dump()
+        m5.stats.reset()
+    elif "workbegin" in exit_event.getCause():
+        print("m5 work begin dump and reset")
+        m5.stats.dump()
+        m5.stats.reset()
+    elif "workend" in exit_event.getCause():
+        print("m5 work end dump and reset")
+        m5.stats.dump()
+        m5.stats.reset()
+    else:
+        print(
+            f"Unknown exit event: {exit_event.getCause()}. Continuing..."
+        )
+
+    exit_event = m5.simulate(maxtick - m5.curTick())
+
 if args.fast_forward:
     if exit_event.getCause() == "a thread reached the max instruction count":
         m5.switchCpus(system, switch_cpu_list)