diff options
Diffstat (limited to 'archive')
| -rw-r--r-- | archive/2025/README.md | 1 | ||||
| -rw-r--r-- | archive/2025/summer/bsc_bambou/README.md | 34 | ||||
| -rw-r--r-- | archive/2025/summer/bsc_bambou/evaluation_utils.ipynb | 784 |
3 files changed, 819 insertions, 0 deletions
diff --git a/archive/2025/README.md b/archive/2025/README.md index 5477db5b5..32c1d8195 100644 --- a/archive/2025/README.md +++ b/archive/2025/README.md @@ -11,6 +11,7 @@ | Maximilian Jäcklein | Protecting H/W and S/W Interactions for Network-Attached Accelerators | BA | Harshavardhan Unnibhavi | [Source](/archive/2025/summer/bsc_jaecklein) | | Hristina Yordanova Grigorova | Online OS reconfiguration for Cloud DBMS | BA | Ilya Meignan--Masson | [Source](/archive/2025/summer/bsc_grigorova) | | Jonas Ludwig Gerg | Automatic Term Extraction based on GenAI | BA | Bernd Brügge | [Source](summer/bsc_gerg) +| Steve Bambou Biangang | Evaluation of Multiple Branch Prediction's Potential in the Context of Wide-Pipeline Architectures | BA | Dr. David Schall | [Source](/archive/2025/summer/bsc_bambou) | ## Winter semester diff --git a/archive/2025/summer/bsc_bambou/README.md b/archive/2025/summer/bsc_bambou/README.md new file mode 100644 index 000000000..a06ee73af --- /dev/null +++ b/archive/2025/summer/bsc_bambou/README.md @@ -0,0 +1,34 @@ +# Artifact Archive: BSc Thesis Steve Bambou + +This folder contains the research artifacts for my Bachelor thesis at [TUM DSE](https://dse.in.tum.de/). + +## Code Base + +The main source code for my thesis is available in the following repository: + +- [Code Repository](https://github.com/dhschall/gem5-fdp/tree/pf-rework-multi-pred) + +This repository includes: +- Source code +- Build system files + +## Evaluation Code + +The evaluation scripts and analysis tools are maintained separately in: + +- [Evaluation Repository](https://github.com/dhschall/gem5-svr-bench/tree/multi-pred) + +This repository contains: +- Gem5 configuration scripts +- Benchmark configuration scripts +- Data processing scripts + + +## How to Reproduce Results + +1. Clone the code base repository. +2. Clone the evaluation repository inside of the code base repository. +3. Follow the instructions in each repository’s `README.md` to build and run the code. +4. Use the evaluation scripts to run the benchmarks. +5. Use the data collection script to aggregate the benchmark results in single .csv files for SPEC and non SPEC workloads. +6. Use this [script](evaluation_utils.ipynb) to generate plots. \ No newline at end of file diff --git a/archive/2025/summer/bsc_bambou/evaluation_utils.ipynb b/archive/2025/summer/bsc_bambou/evaluation_utils.ipynb new file mode 100644 index 000000000..b2def4301 --- /dev/null +++ b/archive/2025/summer/bsc_bambou/evaluation_utils.ipynb @@ -0,0 +1,784 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": null, + "id": "c0f20eed-308c-4f28-824b-657d808788a5", + "metadata": {}, + "outputs": [], + "source": [ + "\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "import numpy as np\n", + "import re\n", + "import seaborn as sns\n", + "from matplotlib.transforms import offset_copy\n", + "import matplotlib.patches as mpatches\n", + "\n", + "def column_mapping(cores):\n", + " return {'finalTick' : 'finalTick',\n", + " 'benchmark' : 'benchmark',\n", + " 'experiment' : 'experiment',\n", + " f'board.processor.cores{cores}.core.ipc': 'ipc',\n", + " f'board.processor.cores{cores}.core.branchPred.mispredictDueToBTBMiss_0::total': 'btb_mispredicts',\n", + " f'board.processor.cores{cores}.core.branchPred.mispredictDueToPredictor_0::DirectCond' : 'cond_mispredicts',\n", + " f'board.processor.cores{cores}.core.branchPred.mispredictDueToPredictor_0::total': 'bp_mispredicts',\n", + " f'board.processor.cores{cores}.core.branchPred.mispredicted_0::total' : 'total_mispredicts',\n", + " f'board.processor.cores{cores}.core.TopDownL1.frontendBound' : 'TopDownL1_frontendBound',\n", + " f'board.processor.cores{cores}.core.TopDownL1.badSpeculation' : 'TopDownL1_badSpeculation',\n", + " f'board.processor.cores{cores}.core.TopDownL1.backendBound' : 'TopDownl1_backendBound',\n", + " f'board.processor.cores{cores}.core.TopDownL1.retiring' : 'TopDownl1_retiring',\n", + " f'board.processor.cores{cores}.core.rename.serializeStallCycles' : 'serializingStallCycles',\n", + " f'board.processor.cores{cores}.core.rename.ROBFullEvents' : 'rob_full',\n", + " f'board.processor.cores{cores}.core.rename.IQFullEvents':'iq_full',\n", + " f'board.processor.cores{cores}.core.rename.LQFullEvents' : 'lq_full',\n", + " f'board.processor.cores{cores}.core.rename.SQFullEvents' : 'sq_full',\n", + " f'board.processor.cores{cores}.core.rename.fullRegistersEvents' : 'register_full',\n", + " f'board.processor.cores{cores}.core.fetchStats0.icacheStallCycles' : 'icache_stall_cycles',\n", + " f'board.processor.cores{cores}.core.iew.memOrderViolationEvents' : 'memOrderViolations',\n", + " f'board.processor.cores{cores}.core.TopDownL2_BackendBound.serializeStalls' : 'TopDownL2_serializingStalls',\n", + " f'board.processor.cores{cores}.core.TopDownL2_BackendBound.memoryBound' : 'TopDownL2_memoryBound',\n", + " f'board.processor.cores{cores}.core.TopDownL2_BackendBound.coreBound' : 'TopDownL2_coreBound',\n", + " f'board.processor.cores{cores}.core.TopDownL2_BadSpeculation.machineClears' : 'TopDownL2_machineClears',\n", + " f'board.processor.cores{cores}.core.TopDownL2_BadSpeculation.branchMissPredicts' : 'TopDownL2_Mispredicts',\n", + " f'board.processor.cores{cores}.core.issueRate' : 'issueRate',\n", + " f'board.processor.cores{cores}.core.fetchStats0.fetchRate' : 'fetchRate',\n", + " f'board.processor.cores{cores}.core.commitStats0.numInsts' : 'insts',\n", + " f'board.processor.cores{cores}.core.cpi' : 'cpi',\n", + " f'board.processor.cores{cores}.core.idleCycles' : 'idleCycles',\n", + " f'board.processor.cores{cores}.core.numCycles' : 'numCycles', \n", + " f'board.processor.cores{cores}.core.robOccupancy::mean' : 'ROB mean occupancy',\n", + " f'board.cache_hierarchy.l1icaches{cores}.demandMshrMisses::total' : 'l1icacheMisses',\n", + " f'board.cache_hierarchy.l1dcaches{cores}.demandMshrMisses::total' : 'l1dcacheMisses',\n", + " f'board.processor.cores{cores}.core.bac.ftSizeDist::mean' : 'mean_ftSize',\n", + " f'board.processor.cores{cores}.core.bac.ftNumber::mean' : 'mean_ftNumber',\n", + " f'board.processor.cores{cores}.core.numIssuedDist::mean' : 'mean_issuedInsts',\n", + " f'board.processor.cores{cores}.core.iew.dispInstDist::mean' : 'mean_dispatchedInsts',\n", + " f'board.processor.cores{cores}.core.iew.wbRate' : 'wbRate',\n", + " f'board.processor.cores{cores}.core.ftq.occupancy::mean' : 'Mean FTQ Occupancy',\n", + " f'board.processor.cores{cores}.core.instSquashedAexperimenttIssueDist::mean' : 'mean_squashedInstsAtIssue',\n", + " f'board.processor.cores{cores}.core.lsq0.blockedByCache' : 'lsqBlockebByCache',\n", + " f'board.processor.cores{cores}.core.rob.robSquashCycles' : 'robSquashCycles',\n", + " f'board.processor.cores{cores}.core.rob.independentInstDelta::mean' : 'independentInstDelta',\n", + " f'board.processor.cores{cores}.core.rob.independentInst' : 'independentInst'\n", + " }\n", + "\n", + "\n", + "\n", + "def loadAndPrepare(fileNames, cores=[], experiment_marker = []):\n", + " dfs = []\n", + " for idx, fileName in enumerate(fileNames):\n", + " tmp_df = pd.read_csv(fileName).fillna(0)\n", + " if len(experiment_marker) > idx:\n", + " tmp_df['set'] = experiment_marker[idx] \n", + " tmp_df.rename(columns=column_mapping(cores[idx] if len (cores) > 0 else \"\"), inplace=True) \n", + " dfs.append(tmp_df)\n", + " df = pd.concat(dfs , ignore_index=True)\n", + " \n", + " \n", + " #df.drop(columns=[col for col in df.columns if not (col.startswith('board.processor.cores1.core') or col in column_mapping.values())], inplace=True)\n", + "\n", + " new_cols = {\n", + " 'mpki': df['total_mispredicts'] * 1000 / df['insts'],\n", + " 'btb_mpki': df['btb_mispredicts'] * 1000 / df['insts'],\n", + " 'cond_mpki': df['cond_mispredicts'] * 1000 / df['insts'],\n", + " 'bp_mpki': df['bp_mispredicts'] * 1000 / df['insts'],\n", + " 'icache_mpki': df['l1icacheMisses'] * 1000 / df['insts'],\n", + " 'IPC': df['insts'] / (df['numCycles'] - df['idleCycles']),\n", + " 'm_cpi': 1 / (df['insts'] / (df['numCycles'] - df['idleCycles'])),\n", + " 'L1IStallCycleRate': df['icache_stall_cycles'] / (df['numCycles'] - df['idleCycles']),\n", + " 'fetchRate' : df['fetchRate'] * df['numCycles'] / (df['numCycles'] - df['idleCycles']),\n", + " 'serializingStallCycleRate': df['serializingStallCycles'] / (df['numCycles'] - df['idleCycles']),\n", + " 'ppc' : df[\"experiment\"].str.extract(r\"(ppc\\d*)\") \n", + "}\n", + " \n", + " df['benchmark'] = df['benchmark'].replace('502.gcc_r.gcc-pp.opts-O3_-finline-limit_36000', '502.gcc_r')\n", + " df = df.assign(**new_cols)\n", + " \n", + " \n", + " return df\n", + "\n", + "def removeWarmup(df):\n", + " n = 2\n", + " mask = df.groupby(['benchmark', 'experiment']).cumcount() >= n\n", + " df = df[mask].reset_index(drop=True)\n", + " return df\n", + "\n", + "\n", + "def filterBenchExperiment(df, experiment_like, exclude_benchs = [] , only_benchs = []):\n", + " if len(only_benchs) > 0:\n", + " df = df[df['benchmark'].isin(only_benchs)]\n", + " else: \n", + " df = df[~df['benchmark'].isin(exclude_benchs)]\n", + " \n", + " df = df[df['experiment'].str.contains(experiment_like, regex = True, na = False)]\n", + " return df\n", + "\n", + "def plotYoverTime(df, y, experiment_like = '', exclude_benchs = [], only_benchs = []):\n", + " x = 'finalTick'\n", + " \n", + " df = filterBenchExperiment(df, experiment_like, exclude_benchs , only_benchs)\n", + "\n", + " for benchmark, group in df.groupby('benchmark'):\n", + " plt.figure(figsize=(8, 5))\n", + " plt.title(f\"Benchmark: {benchmark}\")\n", + " \n", + " # Group by experiment within the benchmark group\n", + " for experiment, subgrp in group.groupby('experiment'):\n", + " plt.plot(subgrp[x], subgrp[y], label=experiment) # Line\n", + " plt.scatter(subgrp[x], subgrp[y]) # Scatter points\n", + " \n", + " plt.xlabel(x)\n", + " plt.ylabel(y)\n", + " plt.legend(title=\"Experiment\")\n", + " plt.grid(True)\n", + " plt.tight_layout()\n", + " plt.show()\n", + "\n", + "\n", + "\n", + "def barPlotY_B(\n", + " dfs,\n", + " y,\n", + " label=None,\n", + " experiment_like='',\n", + " exclude_benchs=[],\n", + " only_benchs=[],\n", + " save_as=\"\",\n", + " ylim=None,\n", + " legend_out=False,\n", + " show_means=False, # mean bars across benchmarks\n", + " ref_value=None # percentage ref value\n", + "):\n", + " # --- If a single df is passed instead of list, wrap it ---\n", + " if isinstance(dfs, pd.DataFrame):\n", + " dfs = [dfs]\n", + "\n", + " # --- Concatenate with explicit set ordering preserved ---\n", + " df = pd.concat(dfs, ignore_index=True)\n", + "\n", + " # --- Maintain set order as provided ---\n", + " ordered_sets = []\n", + " for d in dfs:\n", + " for s in d['set'].unique():\n", + " if s not in ordered_sets:\n", + " ordered_sets.append(s)\n", + "\n", + " df = filterBenchExperiment(df, experiment_like, exclude_benchs, only_benchs)\n", + "\n", + " # --- Group by benchmark, set, ppc ---\n", + " grouped = df.groupby(['benchmark', 'set', 'ppc'])[y].mean().reset_index()\n", + "\n", + " # --- Optionally add \"Mean\" row across benchmarks ---\n", + " if show_means:\n", + " mean_rows = (\n", + " grouped.groupby(['set', 'ppc'])[y]\n", + " .mean()\n", + " .reset_index()\n", + " .assign(benchmark=\"Mean\")\n", + " )\n", + " grouped = pd.concat([grouped, mean_rows], ignore_index=True)\n", + "\n", + " # --- Plot ---\n", + " benchmarks = grouped['benchmark'].unique()\n", + " sets = ordered_sets # use preserved order\n", + " ppcs = grouped['ppc'].unique()\n", + "\n", + " x = np.arange(len(benchmarks))\n", + " width = 0.8 / len(sets) # width per set group\n", + "\n", + " hatch_patterns = [\"//\" , \"...\", \"\\\\\\\\\", \"xx\", \"++\"] # extend if needed\n", + " colors = sns.color_palette(\"tab10\", len(ppcs))\n", + "\n", + " fig, ax = plt.subplots(figsize=(12, 6))\n", + "\n", + " for i, s in enumerate(sets):\n", + " for j, p in enumerate(ppcs):\n", + " vals = []\n", + " for b in benchmarks:\n", + " row = grouped[(grouped['benchmark'] == b) & (grouped['set'] == s) & (grouped['ppc'] == p)]\n", + " vals.append(row[y].values[0] if not row.empty else 0)\n", + "\n", + " bars = ax.bar(\n", + " x + i * width + j * (width / len(ppcs)),\n", + " vals,\n", + " width / len(ppcs),\n", + " color=colors[j],\n", + " hatch=hatch_patterns[i % len(hatch_patterns)],\n", + " edgecolor=\"black\"\n", + " )\n", + "\n", + " # --- Add percentage labels if ref_value is given ---\n", + " if ref_value is not None:\n", + " for bar in bars:\n", + " height = bar.get_height()\n", + " if height > 0:\n", + " perc = (height / ref_value) * 100\n", + " ax.text(\n", + " bar.get_x() + bar.get_width() / 2,\n", + " height + (ylim * 0.01 if ylim else height * 0.01),\n", + " f\"{perc:.1f}%\",\n", + " ha=\"center\", va=\"bottom\", fontsize=8\n", + " )\n", + "\n", + " # --- Beautify ---\n", + " ax.set_xticks(x + (len(sets) - 1) * width / 2)\n", + " ax.set_xticklabels(benchmarks, rotation=45, ha=\"right\")\n", + "\n", + " if ylim:\n", + " plt.ylim(0, ylim)\n", + "\n", + " ax.set_title(f'{label if label else y} per Benchmark')\n", + " ax.set_xlabel('Benchmark')\n", + " ax.set_ylabel(f'{label if label else y}')\n", + "\n", + " # --- Single combined legend ---\n", + " color_patches = [mpatches.Patch(color=colors[j], label=f\"{p}\") for j, p in enumerate(ppcs)]\n", + " hatch_patches = [mpatches.Patch(facecolor=\"white\", hatch=hatch_patterns[i % len(hatch_patterns)], edgecolor=\"black\", label=f\"{s}\") for i, s in enumerate(sets)]\n", + "\n", + " all_handles = color_patches + hatch_patches\n", + "\n", + " ax.legend(\n", + " handles=all_handles,\n", + " title=\"Legend\",\n", + " loc=\"best\",\n", + " bbox_to_anchor=(1.0, 0.5) if legend_out else None\n", + " )\n", + "\n", + " plt.grid(axis='y', linestyle='--', alpha=0.7)\n", + " plt.tight_layout()\n", + "\n", + " if len(save_as) > 0:\n", + " plt.savefig(f\"bar_plots/{save_as}.png\", bbox_inches='tight')\n", + "\n", + " plt.show()\n", + "\n", + "\n", + "\n", + "\n", + "def barPlotY(\n", + " df, \n", + " y, \n", + " label=None, \n", + " experiment_like='', \n", + " exclude_benchs=[], \n", + " only_benchs=[], \n", + " save_as=\"\", \n", + " ylim=None, \n", + " legend_out=False,\n", + " show_means=False, # NEW flag for mean bars\n", + " ref_value=None # NEW: reference value for percentage\n", + "):\n", + "\n", + " df = filterBenchExperiment(df, experiment_like, exclude_benchs, only_benchs)\n", + " \n", + " # Mean per benchmark per experiment\n", + " grouped = df.groupby(['benchmark', 'experiment'])[f'{y}'].mean().unstack()\n", + "\n", + " # Optionally add a \"Mean\" row across benchmarks\n", + " if show_means:\n", + " means = grouped.mean(axis=0)\n", + " grouped.loc[\"Mean\"] = means\n", + "\n", + " # Step 2: Plot as bar chart\n", + " ax = grouped.plot(kind='bar', figsize=(10, 6))\n", + " \n", + " # Step 3: Beautify\n", + " if ylim:\n", + " plt.ylim(0, ylim)\n", + " plt.title(f'{label if label else y} per Benchmark per Experiment')\n", + " plt.xlabel('Benchmark')\n", + " plt.ylabel(f'{label if label else y}')\n", + " plt.legend(\n", + " title='Experiment', \n", + " loc=\"best\",\n", + " bbox_to_anchor=(1.0, 0.5) if legend_out else None\n", + " )\n", + " plt.ylim(bottom=0)\n", + " plt.grid(axis='y', linestyle='--', alpha=0.7)\n", + " plt.tight_layout()\n", + "\n", + " # Step 4: Add percentages if ref_value is given\n", + " if ref_value is not None:\n", + " for container in ax.containers: # each experiment\n", + " for bar in container:\n", + " height = bar.get_height()\n", + " if height > 0:\n", + " perc = (height / ref_value) * 100\n", + " ax.text(\n", + " bar.get_x() + bar.get_width() / 2,\n", + " height + (ylim * 0.01 if ylim else height * 0.01), # offset a bit above\n", + " f\"{perc:.1f}%\",\n", + " ha=\"center\", va=\"bottom\", fontsize=8\n", + " )\n", + "\n", + " if len(save_as) > 0:\n", + " plt.savefig(f\"bar_plots/{save_as}.png\", bbox_inches='tight')\n", + " plt.show()\n", + "\n", + "\n", + "\n", + "def plotMeanY(df, y, experiment_like = '', exclude_benchs = [], only_benchs = [], save_as = \"\"):\n", + "\n", + " df = filterBenchExperiment(df, experiment_like, exclude_benchs , only_benchs)\n", + " \n", + " grouped = df.groupby(['benchmark', 'experiment'])[f'{y}'].mean().unstack()\n", + "\n", + " # Step 2: Plot as bar chart\n", + " grouped.plot(kind='bar', figsize=(10, 6))\n", + " \n", + " # Step 3: Beautify\n", + " plt.title(f'Mean of {y} per Benchmark per Experiment')\n", + " plt.xlabel('Benchmark')\n", + " plt.ylabel(f'Mean {y}')\n", + " plt.legend(title='Experiment', \n", + " bbox_to_anchor=(1.0, 0.5) \n", + " )\n", + " plt.grid(axis= 'y', linestyle='--', alpha=0.7)\n", + " plt.tight_layout()\n", + " if len(save_as) > 0:\n", + " plt.savefig(f\"bar_plots/{save_as}.png\", bbox_inches='tight')\n", + " plt.show()\n", + "\n", + "def plotYDensity(df, y , cores=\"\", experiment_like = '', exclude_benchs = [], only_benchs = [], label = None , save_as = \"\"):\n", + "\n", + " variable_supported = {\n", + " 'robOccupancy': fr'board\\.processor\\.cores{cores}\\.core\\.robOccupancy::\\d+(-\\d+)?$', \n", + " 'ftNumber' : fr'board\\.processor\\.cores{cores}\\.core\\.bac\\.ftNumber::\\d+(-\\d+)?$',\n", + " 'BB_Size' : fr'board\\.processor\\.cores{cores}\\.core\\.bac\\.ftSizeDist::\\d+(-\\d+)?$',\n", + " 'issuedInsts': fr'board\\.processor\\.cores{cores}\\.core\\.numIssuedDist::\\d+(-\\d+)?$', \n", + " 'ftqOccupancy' : fr'board\\.processor\\.cores{cores}\\.core\\.ftq\\.occupancy::\\d+(-\\d+)?$'\n", + " }\n", + "\n", + " if y not in variable_supported.keys():\n", + " print(\"The variable entered is not yet supported!\")\n", + " return\n", + " \n", + " df = filterBenchExperiment(df, experiment_like, exclude_benchs , only_benchs)\n", + "\n", + " # Find all columns that represent ROB occupancy buckets (excluding 'total', 'mean', 'stddev', 'samples')\n", + " bucket_cols = [\n", + " col for col in df.columns\n", + " if re.match(variable_supported[y] , col)\n", + " ]\n", + " \n", + " if not bucket_cols:\n", + " print(f\"No {y} bucket columns found.\")\n", + " return\n", + "\n", + " # Prepare data for plotting\n", + " df_plot = df[['benchmark', 'experiment', 'finalTick'] + bucket_cols].copy()\n", + " df_plot = df_plot.loc[df_plot.groupby(['experiment', 'benchmark'])['finalTick'].idxmax()]\n", + "\n", + " # Melt for density plotting\n", + " df_melted = df_plot.melt(\n", + " id_vars=['benchmark', 'experiment'],\n", + " value_vars=bucket_cols,\n", + " var_name=f'{y} Bucket',\n", + " value_name='Frequency'\n", + " )\n", + "\n", + " # Extract bucket start as numeric value for plotting\n", + " df_melted['Bucket Start'] = df_melted[f'{y} Bucket'].apply(lambda x: int(x.split('::')[-1].split('-')[0]))\n", + "\n", + " for benchmark, group_df in df_melted.groupby('benchmark'):\n", + " # Plot\n", + " plt.figure(figsize=(10, 6))\n", + " sns.kdeplot(\n", + " data=group_df,\n", + " x='Bucket Start',\n", + " weights='Frequency',\n", + " hue='experiment',\n", + " common_norm=False,\n", + " fill=True,\n", + " alpha=0.4,\n", + " linewidth=1.5, \n", + " cut=0,\n", + " clip=(0, df_melted['Bucket Start'].max())\n", + " )\n", + "\n", + " \n", + " plt.title(f\" {label if label else y} Density by Experiment benchmark: {benchmark}\")\n", + " plt.xlabel(f'{label if label else y} Bucket Start')\n", + " plt.ylabel('Density (weighted)')\n", + " plt.tight_layout()\n", + " if len(save_as) > 0:\n", + " plt.savefig(f\"density_plots/{save_as}.png\", bbox_inches='tight')\n", + " plt.show()\n", + "\n", + "\n", + "def plot_incremental_from_dfs(\n", + " dfs, labels,\n", + " benchmarks_col=\"benchmark\", experiment_col=\"experiment\", value_col=\"IPC\",\n", + " ylabel=\"Performance (IPC)\", title=None , save_as=\"\", ymax=None,\n", + " show_means=False # NEW argument\n", + "):\n", + " \"\"\"\n", + " Plot grouped stacked bars from multiple DataFrames, each representing\n", + " a performance scenario. The DataFrames must have columns:\n", + " [benchmark, experiment, IPC].\n", + " \n", + " Parameters:\n", + " dfs (list[pd.DataFrame]): list of DataFrames in scenario order\n", + " labels (list[str]): names of each scenario\n", + " \"\"\"\n", + " experiments = dfs[0][experiment_col].unique()\n", + " \n", + " # --- aggregate each df ---\n", + " grouped_dfs = []\n", + " for df in dfs:\n", + " grouped = df.groupby([benchmarks_col, experiment_col])[value_col].mean().reset_index()\n", + " grouped_dfs.append(grouped)\n", + " \n", + " # --- merge into one wide table ---\n", + " merged = grouped_dfs[0].rename(columns={value_col: labels[0]})\n", + " for i, df in enumerate(grouped_dfs[1:], 1):\n", + " merged = merged.merge(df.rename(columns={value_col: labels[i]}),\n", + " on=[benchmarks_col, experiment_col])\n", + " \n", + " # --- optionally add a \"Mean\" benchmark ---\n", + " if show_means:\n", + " mean_rows = []\n", + " for exp in experiments:\n", + " sub = merged[merged[experiment_col] == exp]\n", + " means = sub[labels].mean()\n", + " row = {benchmarks_col: \"Mean\", experiment_col: exp}\n", + " row.update(means.to_dict())\n", + " mean_rows.append(row)\n", + " merged = pd.concat([merged, pd.DataFrame(mean_rows)], ignore_index=True)\n", + " \n", + " benchmarks = merged[benchmarks_col].unique()\n", + " x = np.arange(len(benchmarks))\n", + " width = 0.75 / len(experiments)\n", + " fig, ax = plt.subplots(figsize=(16, 7))\n", + " \n", + " # softer, more readable colors\n", + " colors = [\"#183ab8\", \"#bf8970\", \"#FF2400\", \"#ffd700\", \"#50C878\", \"#C0C0C0\", \"#7851a9\", \"#f77b07\"]\n", + " hatches = [\"//\", \"\\\\\\\\\", \"xx\", \"oo\", \"..\", \"++\"] \n", + " \n", + " # --- plot each experiment as grouped stacked bar ---\n", + " for i, exp in enumerate(experiments):\n", + " sub = merged[merged[experiment_col]==exp]\n", + " xpos = x - 0.375 + i*width + width/2\n", + " \n", + " bottom = np.zeros(len(sub))\n", + " prev_vals = bottom.copy()\n", + " for j, lab in enumerate(labels):\n", + " vals = sub[lab].to_numpy()\n", + " inc = vals - prev_vals # true difference\n", + "\n", + " # only plot positive increments\n", + " pos_mask = inc > 0\n", + " if np.any(pos_mask):\n", + " ax.bar(\n", + " xpos[pos_mask], inc[pos_mask], width,\n", + " bottom=prev_vals[pos_mask],\n", + " color=colors[j % len(colors)],\n", + " hatch=hatches[i % len(hatches)],\n", + " edgecolor=\"black\"\n", + " )\n", + "\n", + " prev_vals = vals # always update (so baseline is correct)\n", + "\n", + " # --- formatting ---\n", + " ax.set_ylabel(ylabel, fontsize=12)\n", + " if title: ax.set_title(title, fontsize=14)\n", + " ax.set_xticks(x)\n", + " ax.set_xticklabels(benchmarks, rotation=45, ha=\"right\")\n", + " ax.yaxis.grid(True, which=\"both\", linestyle=\"--\", alpha=0.6)\n", + " ax.margins(x=0.12)\n", + " if ymax:\n", + " ax.set_ylim(0, ymax)\n", + " \n", + " # --- build combined legend ---\n", + " scenario_patches = [mpatches.Patch(color=colors[j], label=lab) for j, lab in enumerate(labels)]\n", + " hatch_patches = [mpatches.Patch(facecolor=\"white\", edgecolor=\"black\", hatch=hatches[i], label=exp) \n", + " for i, exp in enumerate(experiments)]\n", + " ax.legend(handles=scenario_patches + hatch_patches, title=\"Legend\", ncol=2)\n", + " \n", + " if len(save_as) > 0:\n", + " plt.savefig(f\"IPC_plots/{save_as}.png\", bbox_inches='tight')\n", + " plt.tight_layout()\n", + " plt.show()\n", + "\n", + "def plotMPKIStack(\n", + " df, \n", + " mean=False, \n", + " experiment_like='', \n", + " exclude_benchs=[], \n", + " only_benchs=[], \n", + " rm_components=[], \n", + " save_as=\"\", \n", + " legend_out=False,\n", + " show_means=False, # NEW ARGUMENT (already added)\n", + " reverse_experiments=False # NEW ARGUMENT\n", + "):\n", + " latest = df\n", + " latest = filterBenchExperiment(latest, experiment_like, exclude_benchs, only_benchs)\n", + "\n", + " if mean:\n", + " latest = latest.groupby(['experiment', 'benchmark'])[\n", + " ['btb_mpki', 'cond_mpki', 'mpki', 'bp_mpki', 'icache_mpki']\n", + " ].mean().reset_index(drop=False)\n", + " else:\n", + " latest = latest.loc[latest.groupby(['experiment', 'benchmark'])['finalTick'].idxmax()]\n", + " \n", + " component_cols = {\n", + " \"icache_mpki\": \"L1ICache\",\n", + " \"btb_mpki\": \"BTB\",\n", + " \"bp_mpki\": \"Branch Predictor\"\n", + " }\n", + "\n", + " for comp in rm_components:\n", + " component_cols.pop(comp, None) \n", + " \n", + " # --- Optionally add \"Mean\" row for each experiment ---\n", + " if show_means:\n", + " mean_rows = []\n", + " for exp in latest['experiment'].unique():\n", + " sub = latest[latest['experiment'] == exp]\n", + " means = sub[list(component_cols.keys())].mean()\n", + " row = {\"benchmark\": \"Mean\", \"experiment\": exp}\n", + " row.update(means.to_dict())\n", + " mean_rows.append(row)\n", + " latest = pd.concat([latest, pd.DataFrame(mean_rows)], ignore_index=True)\n", + "\n", + " # Prepare axis categories\n", + " latest['benchmark'] = latest['benchmark'].astype(str)\n", + " benchmarks = list(latest['benchmark'].unique())\n", + " if \"Mean\" in benchmarks:\n", + " benchmarks = [b for b in benchmarks if b != \"Mean\"] + [\"Mean\"]\n", + " \n", + " experiments = sorted(latest['experiment'].unique())\n", + " if reverse_experiments: # <-- invert order if requested\n", + " experiments = experiments[::-1]\n", + " \n", + " # Define colors and hatches\n", + " colors = {\n", + " \"L1ICache\": \"#14a31b\",\n", + " \"BTB\": \"#3734eb\",\n", + " \"Branch Predictor\": \"#eb5334\"\n", + " }\n", + " hatches = ['', '*', 'o', '-', 'o+*', '-+.', '.', '+'] \n", + " \n", + " # Plot\n", + " fig, ax = plt.subplots(figsize=(16, 6))\n", + " bar_width = 0.6 / len(experiments)\n", + " x = np.arange(len(benchmarks))\n", + " \n", + " # Offset bars by experiment\n", + " for i, exp in enumerate(experiments):\n", + " subset = latest[latest['experiment'] == exp].set_index('benchmark')\n", + " bottom = np.zeros(len(benchmarks))\n", + " for col, name in component_cols.items():\n", + " values = [subset.loc[b, col] if b in subset.index else 0 for b in benchmarks]\n", + " bars = ax.bar(\n", + " x + i * bar_width, values, bar_width,\n", + " bottom=bottom,\n", + " color=colors[name],\n", + " hatch=hatches[i % len(hatches)],\n", + " edgecolor='black',\n", + " label=name if i == 0 else None\n", + " )\n", + " bottom += values\n", + " \n", + " # Labels and ticks\n", + " ax.set_xticks(x + bar_width * (len(experiments) - 1) / 2)\n", + " ax.set_xticklabels(benchmarks, rotation=45, ha='right')\n", + " ax.set_ylabel(\"MPKI\")\n", + " ax.set_title(\"MPKI Breakdown per Benchmark (Per Experiment)\")\n", + " \n", + " # Legends\n", + " component_handles = [plt.Rectangle((0, 0), 1, 1, color=colors[name]) for name in component_cols.values()]\n", + " experiment_handles = [\n", + " plt.Rectangle((0, 0), 1, 1, facecolor='white', edgecolor='black', hatch=hatches[i % len(hatches)])\n", + " for i, exp in enumerate(experiments)\n", + " ]\n", + " all_handles = component_handles + experiment_handles\n", + " all_labels = list(component_cols.values()) + experiments\n", + " \n", + " ax.legend(\n", + " all_handles, all_labels,\n", + " title=\"Component / Experiment\",\n", + " loc='best',\n", + " bbox_to_anchor=(1.0, 0.5) if legend_out else None\n", + " )\n", + " \n", + " plt.tight_layout()\n", + " if len(save_as) > 0:\n", + " plt.savefig(f\"mpki_plots/{save_as}.png\", bbox_inches='tight')\n", + " plt.show()\n", + "\n", + "\n", + " \n", + "\n", + "def plotCPIStack(df, mean = False, experiment_like = '', exclude_benchs = [], only_benchs = [], m_cpi = False, detailed = True, save_as = \"\", ymax=None):\n", + " latest = df\n", + "\n", + " latest = filterBenchExperiment(latest, experiment_like, exclude_benchs, only_benchs)\n", + "\n", + " cpi_var = 'm_cpi' if m_cpi else 'cpi'\n", + "\n", + " \n", + " \n", + " if mean:\n", + " latest = latest.groupby(['experiment', 'benchmark'])[['TopDownl1_retiring', 'TopDownL1_frontendBound', 'TopDownL1_badSpeculation', 'TopDownl1_backendBound', \"TopDownL2_Mispredicts\", \"TopDownL2_machineClears\", \"TopDownL2_serializingStalls\", \"TopDownL2_coreBound\", \"TopDownL2_memoryBound\" , cpi_var]].mean().reset_index(drop=False)\n", + " else:\n", + " latest = latest.loc[latest.groupby(['experiment', 'benchmark'])['finalTick'].idxmax()]\n", + " \n", + " \n", + " #latest = latest.map(lambda x: 0 if isinstance(x, (int, float)) and x < 0 else x)\n", + " \n", + " #Print the CPI Stack for specific benchmarks and experiments\n", + "\n", + " # Define component columns and friendly display names\n", + " component_cols_detailed = {\n", + " \"TopDownl1_retiring\": \"Retiring\",\n", + " \"TopDownL1_frontendBound\": \"Frontend Bound\",\n", + " \"TopDownL2_Mispredicts\": \"BadSpec Mispredicts\",\n", + " \"TopDownL2_machineClears\" : \"BadSpec memOrderViolations\",\n", + " \"TopDownL2_serializingStalls\" : \"Backend SerializingStalls\",\n", + " \"TopDownL2_coreBound\" : \"Backend coreBound\",\n", + " \"TopDownL2_memoryBound\": \"Backend memoryBound\"\n", + " }\n", + "\n", + " component_cols_L1 = {\n", + " \"TopDownl1_retiring\": \"Retiring\",\n", + " \"TopDownL1_frontendBound\": \"Frontend Bound\",\n", + " \"TopDownL1_badSpeculation\": \"Bad Speculation\",\n", + " \"TopDownl1_backendBound\": \"Backend Bound\"\n", + " }\n", + "\n", + " component_cols = component_cols_detailed if detailed else component_cols_L1\n", + " \n", + " # Multiply each component % by total CPI\n", + " for col in component_cols:\n", + " latest[col] = latest[col] * latest[cpi_var]\n", + " \n", + " # Prepare axis categories\n", + " latest['benchmark'] = latest['benchmark'].astype(str)\n", + " benchmarks = sorted(latest['benchmark'].unique())\n", + " experiments = sorted(latest['experiment'].unique())\n", + " \n", + " # Define colors and hatches\n", + " colors = {\n", + " \"Retiring\": \"#7B3294\",\n", + " \"Frontend Bound\": \"#F0E442\",\n", + " \"Bad Speculation\": \"#999999\",\n", + " \"BadSpec Mispredicts\": \"#999999\",\n", + " \"BadSpec memOrderViolations\": \"#0dd617\",\n", + " \"Backend Bound\": \"#D7191C\",\n", + " \"Backend SerializingStalls\": \"#f7760c\",\n", + " \"Backend coreBound\": \"#1939d7\", \n", + " \"Backend memoryBound\" : \"#D7191C\"\n", + " }\n", + " hatches = ['', '*', 'oo', 'X', 'x', '.', '.', '+'] # One per experiment\n", + " \n", + " # Plot\n", + " fig, ax = plt.subplots(figsize=(16, 6))\n", + " bar_width = 0.6 / len(experiments)\n", + " x = np.arange(len(benchmarks))\n", + " \n", + " # Offset bars by experiment\n", + " for i, exp in enumerate(experiments):\n", + " subset = latest[latest['experiment'] == exp].set_index('benchmark')\n", + "\n", + " for j, b in enumerate(benchmarks):\n", + " if b not in subset.index:\n", + " continue\n", + "\n", + " total_cpi = subset.loc[b, cpi_var]\n", + " bottom = 0.0\n", + " drawn = 0.0\n", + "\n", + " for col, name in component_cols.items():\n", + " value = subset.loc[b, col] if col in subset.columns else 0.0\n", + "\n", + " # Nothing left to allocate? skip\n", + " if drawn >= total_cpi:\n", + " continue\n", + "\n", + " # If the component would overshoot, clip it\n", + " if drawn + value > total_cpi:\n", + " value = max(0, total_cpi - drawn)\n", + "\n", + " if value > 0:\n", + " ax.bar(\n", + " j + i * bar_width,\n", + " value,\n", + " bar_width,\n", + " bottom=bottom,\n", + " color=colors[name],\n", + " hatch=hatches[i % len(hatches)],\n", + " edgecolor='black',\n", + " label=name if (i == 0 and bottom == 0) else None\n", + " )\n", + "\n", + " # Update positions\n", + " drawn += value\n", + " bottom += value\n", + " \n", + " # Labels and ticks\n", + " ax.set_xticks(x + bar_width * (len(experiments) - 1) / 2)\n", + " ax.set_xticklabels(benchmarks, rotation=45, ha='right')\n", + " ax.set_ylabel(\"CPI\")\n", + " ax.set_title(\"CPI Breakdown per Benchmark (Per Experiment)\")\n", + " if ymax:\n", + " ax.set_ylim(0, ymax)\n", + " \n", + " # Legend for components (only once)\n", + " # Component legend (already exists)\n", + " component_handles = [\n", + " plt.Rectangle((0, 0), 1, 1, color=colors[name])\n", + " for name in component_cols.values()\n", + " ]\n", + " \n", + " # Experiment legend (hatch only, gray fill to avoid visual conflict)\n", + " experiment_handles = [\n", + " plt.Rectangle((0, 0), 1, 1, facecolor='white', edgecolor='black', hatch=hatches[i % len(hatches)])\n", + " for i, exp in enumerate(experiments)\n", + " ]\n", + " \n", + " # Combine both\n", + " all_handles = component_handles + experiment_handles\n", + " all_labels = list(component_cols.values()) + experiments\n", + " \n", + " # Show both in a unified legend\n", + " ax.legend(\n", + " all_handles,\n", + " all_labels,\n", + " title=\"Component / Experiment\",\n", + " loc='center left',\n", + " bbox_to_anchor=(1.0, 0.5) # x=1.0 means just outside the plot on the right\n", + " )\n", + " \n", + " plt.tight_layout()\n", + " if len(save_as) > 0:\n", + " plt.savefig(f\"CPI_Stacks/{save_as}.png\", bbox_inches='tight')\n", + " plt.show()" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3 (ipykernel)", + "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": "3.12.3" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} |