Perf

Simulator/simulator.py

@@ -398,6 +398,7 @@ def get_result_from_file(result_path):
         dram_channel_bw = {}
         avg_dram_bw = None
         simulation_time = None
+        total_cycle = None
 
         # Read and find total stat position
         with open(result_path, "r") as f:

tutorial/session1/CompilerOptimization.ipynb

@@ -0,0 +1,179 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Compiler Optimization"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "import os\n",
+    "import sys\n",
+    "base_dir = os.environ.get('TORCHSIM_DIR', default='/workspace/PyTorchSim')\n",
+    "sys.path.append(base_dir)\n",
+    "os.environ['TORCHSIM_DUMP_PATH']=base_dir\n",
+    "os.environ['TORCHSIM_FUNCTIONAL_MODE']=\"0\"\n",
+    "os.environ['TORCHSIM_TIMING_MODE']=\"1\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### GeMM + ReLU fusion"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Using /root/.cache/torch_extensions/py310_cu121 as PyTorch extensions root...\n",
+      "Emitting ninja build file /root/.cache/torch_extensions/py310_cu121/npu/build.ninja...\n",
+      "Building extension module npu...\n",
+      "Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)\n",
+      "Loading extension module npu...\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "ninja: no work to do.\n",
+      "Wrapper Codegen Path = /tmp/torchinductor_root/3z/c3zx4dfsx2o24goyevxgy4upevdsyxegbadiquz7z33ttsbs22a6.py\n",
+      "[Gem5] Gem5 is running.   \n",
+      "[TOGSim] TOGSim is running.   \n",
+      "[TOGSim] Simulation of \"/root/workspace/PyTorchSim/tmp/5o2xythi5z3/tile_graph.onnx\" is stored to \"/root/workspace/PyTorchSim/tmp/5o2xythi5z3/togsim_result/0\"\n"
+     ]
+    }
+   ],
+   "source": [
+    "from Scheduler.scheduler import PyTorchSimRunner\n",
+    "device = PyTorchSimRunner.setup_device().custom_device()\n",
+    "\n",
+    "input = torch.randn(1024, 1024).to(device=device)\n",
+    "weight = torch.randn(1024, 1024).to(device=device)\n",
+    "\n",
+    "def gemm_relu(a, b):\n",
+    "    return torch.relu(torch.matmul(a, b))\n",
+    "\n",
+    "opt_fn = torch.compile(dynamic=False)(gemm_relu)\n",
+    "out = opt_fn(input, weight)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[2025-11-30 20:03:06.792] [info] Total execution cycle: 50316\n"
+     ]
+    }
+   ],
+   "source": [
+    "!cat /root/workspace/PyTorchSim/tmp/5o2xythi5z3/togsim_result/0 | grep \"Total execution cycle\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Disable fusion"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Wrapper Codegen Path = /tmp/torchinductor_root/6s/c6sm56korxlpospvj434xqahpah3qitivib3y66vxn6teip2uh32.py\n",
+      "[Gem5] Gem5 is running... \n",
+      "[Gem5] Gem5 is running..  \n",
+      "[TOGSim] TOGSim is running.   \n",
+      "[TOGSim] Simulation of \"/root/workspace/PyTorchSim/tmp/4q4qv6gbpia/tile_graph.onnx\" is stored to \"/root/workspace/PyTorchSim/tmp/4q4qv6gbpia/togsim_result/4\"\n",
+      "[TOGSim] TOGSim is running..  \n",
+      "[TOGSim] Simulation of \"/root/workspace/PyTorchSim/tmp/37dfo4nczcq/tile_graph.onnx\" is stored to \"/root/workspace/PyTorchSim/tmp/37dfo4nczcq/togsim_result/0\"\n"
+     ]
+    }
+   ],
+   "source": [
+    "os.environ['TORCHSIM_COMPILER_OPTIMIZATION']=\"none\"\n",
+    "\n",
+    "input = torch.randn(1024, 1024).to(device=device)\n",
+    "weight = torch.randn(1024, 1024).to(device=device)\n",
+    "\n",
+    "def gemm_relu(a, b):\n",
+    "    return torch.relu(torch.matmul(a, b))\n",
+    "\n",
+    "opt_fn = torch.compile(dynamic=False)(gemm_relu)\n",
+    "out = opt_fn(input, weight)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[2025-11-30 20:03:30.690] [info] Total execution cycle: 46996\n",
+      "[2025-11-30 20:03:32.178] [info] Total execution cycle: 58455\n"
+     ]
+    }
+   ],
+   "source": [
+    "!cat /root/workspace/PyTorchSim/tmp/4q4qv6gbpia/togsim_result/4 | grep \"Total execution cycle\"\n",
+    "!cat /root/workspace/PyTorchSim/tmp/37dfo4nczcq/togsim_result/0 | grep \"Total execution cycle\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "base",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

tutorial/session1/DNNServing.ipynb

@@ -0,0 +1,130 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## DNN Serving System"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "import os\n",
+    "import sys\n",
+    "base_dir = os.environ.get('TORCHSIM_DIR', default='/workspace/PyTorchSim')\n",
+    "sys.path.append(base_dir)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Scheduler"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "from torchvision.models import resnet18\n",
+    "from Scheduler.scheduler import Scheduler, SchedulerDNNModel, Request\n",
+    "from PyTorchSimFrontend.extension_config import CONFIG_TORCHSIM_BACKEND_CONFIG\n",
+    "\n",
+    "scheduler = Scheduler(num_request_queue=1, engine_select=Scheduler.FIFO_ENGINE, backend_config=CONFIG_TORCHSIM_BACKEND_CONFIG)\n",
+    "device = scheduler.execution_engine.module.custom_device()\n",
+    "\n",
+    "model = resnet18().eval()\n",
+    "input = torch.randn(1, 3, 224, 224).to(device=device)\n",
+    "opt_fn = torch.compile(dynamic=False)(model.to(device, memory_format=torch.channels_last))\n",
+    "\n",
+    "SchedulerDNNModel.register_model(\"resnet18\", opt_fn)\n",
+    "request = Request(\"resnet18\", [input], [], request_queue_idx=0)\n",
+    "scheduler.add_request(request, request_time=0)\n",
+    "\n",
+    "# Run scheduler\n",
+    "while not scheduler.is_finished():\n",
+    "    with torch.no_grad():\n",
+    "        scheduler.schedule()\n",
+    "\n",
+    "print(\"ResNet18 Simulation Done\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Load Generator"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import torch\n",
+    "from torchvision.models import resnet18\n",
+    "\n",
+    "from Scheduler.scheduler import Scheduler, SchedulerDNNModel, Request, poisson_request_generator\n",
+    "CONFIG_TORCHSIM_DIR = os.environ.get('TORCHSIM_DIR', default='/workspace/PyTorchSim')\n",
+    "\n",
+    "lambda_requests = 10\n",
+    "max_time = 30\n",
+    "\n",
+    "target_model1 = resnet18().eval()\n",
+    "\n",
+    "# Init scheduler\n",
+    "scheduler = Scheduler(num_request_queue=1, max_batch=32, engine_select=Scheduler.FIFO_ENGINE, backend_config=f\"{CONFIG_TORCHSIM_DIR}/PyTorchSimBackend/configs/systolic_ws_128x128_c2_simple_noc_tpuv2.json\")\n",
+    "# Register compiled model\n",
+    "opt_model1 = torch.compile(target_model1.to(device=scheduler.execution_engine.module.custom_device(), memory_format=torch.channels_last), dynamic=False)\n",
+    "SchedulerDNNModel.register_model(\"resnet18\", opt_model1)\n",
+    "\n",
+    "# Generate time stamp\n",
+    "for request_time in poisson_request_generator(lambda_requests, max_time):\n",
+    "    # Init input data\n",
+    "    model_input1 = torch.randn(1, 3, 224, 224)\n",
+    "\n",
+    "    # Init request\n",
+    "    new_request1 = Request(\"resnet18\", [model_input1], [], request_queue_idx=0)\n",
+    "\n",
+    "    # Add request to scheduler\n",
+    "    print(\"[Reqest] Resnet18 request time: \", request_time, flush=True)\n",
+    "    scheduler.add_request(new_request1, request_time=request_time)\n",
+    "\n",
+    "# Run scheduler\n",
+    "while not scheduler.is_finished():\n",
+    "    scheduler.schedule()"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "base",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}