process more perf data into df, export to matlab

target/sim/bench.ipynb

@@ -15,7 +15,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!python -m pip install pandas plotly pyyaml"
+    "!python -m pip install pandas plotly pyyaml scipy"
    ]
   },
   {
@@ -330,22 +330,6 @@
    "outputs": [],
    "source": []
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "813a82e1-5683-497a-8603-545cb55baada",
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "46335c5f-0802-4541-8ebe-5f3ef2de12e4",
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
   {
    "cell_type": "markdown",
    "id": "db94fbb4-44d3-4ca1-a271-5be3d0f904f5",
@@ -616,32 +600,33 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "6186458e-484c-495a-8b5a-1bd5883667ac",
+   "id": "75cb302d-07aa-49e9-819a-0487738bcea2",
    "metadata": {},
    "outputs": [],
    "source": [
     "import hjson\n",
     "from pathlib import Path\n",
-    "dfs = []\n",
-    "for i in range(4):\n",
-    "    with Path(f'logs_{i}/params.hjson').open() as f:\n",
-    "        param = hjson.loads(f.read())\n",
-    "        df1 = pd.json_normalize(param)\n",
-    "        df1.index = df1.index + i\n",
-    "        dfs.append(df1)\n",
-    "runs = pd.concat(dfs)\n",
-    "runs"
+    "import pandas as pd\n",
+    "import plotly.express as px"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "481d6636-026f-4111-88e6-b86773d1aa03",
+   "id": "74f75c0a-47e4-432a-bf91-3750988e64ed",
    "metadata": {},
    "outputs": [],
    "source": [
-    "import pandas as pd\n",
-    "import plotly.express as px"
+    "dfs = []\n",
+    "n_runs = 11\n",
+    "for i in range(n_runs):\n",
+    "    with Path(f'logs_{i}/params.hjson').open() as f:\n",
+    "        param = hjson.loads(f.read())\n",
+    "        df1 = pd.json_normalize(param)\n",
+    "        df1.index = df1.index + i\n",
+    "        dfs.append(df1)\n",
+    "runs = pd.concat(dfs)\n",
+    "runs"
    ]
   },
   {
@@ -657,7 +642,7 @@
     "end_region = -2\n",
     "\n",
     "dfs = []\n",
-    "for run_id in range(4):\n",
+    "for run_id in range(n_runs):\n",
     "    for i in range(9):\n",
     "        hart_type = 'dma' if i % 9 == 8 else 'fpu'\n",
     "        p = [i % 9, int(i / 9), 0]\n",
@@ -703,15 +688,35 @@
     "perf"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "015c33a5-e12c-459f-8a7c-fa0980fe66f0",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "perf[(perf['hart_type'] == 'fpu') & (perf['region_id'] == 2)][['fpss_fpu_occupancy','gemmInfo.K','gemmInfo.N']]"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
    "id": "05415f8f-b48f-4e4c-a783-35a0e4575ca0",
    "metadata": {},
    "outputs": [],
    "source": [
-    "# px.violin(perf[(perf['hart_type'] == 'fpu') & (perf['region_id'] == 2)], 'gemmInfo.K', 'fpss_fpu_occupancy', violinmode='overlay')\n",
-    "px.bar(perf[(perf['hart_type'] == 'fpu')], 'gemmInfo.K', 'fpss_fpu_occupancy', color='region_id')"
+    "# px.violin(perf[(perf['hart_type'] == 'fpu') & (perf['region_id'] == 2)], color='gemmInfo.K', y='fpss_fpu_occupancy', violinmode='overlay')\n",
+    "px.scatter(perf[(perf['hart_type'] == 'fpu')], 'gemmImpl.L1_K', 'fpss_fpu_occupancy', color='gemmImpl.L1_N', log_x=True, range_y=[0,1])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "710efc36-a758-4571-83db-bd46b3ce6105",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "px.scatter(perf[(perf['hart_type'] == 'fpu')], 'gemmImpl.L1_K', 'cycles', color='gemmImpl.L1_N', log_x=True)"
    ]
   },
   {
@@ -754,6 +759,36 @@
     "px.violin(perf[(perf['cycles'] < 1000) & (perf['fpss_occupancy'] == 0)], 'cycles')"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6b80cbd5-49a7-4067-bc25-4b846f7ae433",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "cf2bab5d-ee65-4dca-84fa-e067d44f8cd8",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "4d4799a3-c653-438b-b4b1-4d209a7deb43",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Export df to matlab\n",
+    "import scipy.io as sio\n",
+    "import re\n",
+    "out = perf[(perf['hart_type'] == 'fpu') & (perf['region_id'] == 2)]\n",
+    "sio.savemat('./logs/perf.mat', out.rename(columns=lambda x: re.sub('[\\. -]','_',x)).to_dict('list'))"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "925ac4a8-de21-4203-b52a-185aaa3c212c",