Octavian!

Octavian M=K=N=1024.ipynb

{
  "cells": [
    {
      "metadata": {},
      "cell_type": "markdown",
      "source": "https://discourse.julialang.org/t/intel-c-c-compiler-performance-versus-julia/61929/18"
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "versioninfo()",
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Julia Version 1.7.0-DEV.1129\nCommit 9117b4d6d6 (2021-05-20 16:42 UTC)\nPlatform Info:\n  OS: Windows (x86_64-w64-mingw32)\n  CPU: Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz\n  WORD_SIZE: 64\n  LIBM: libopenlibm\n  LLVM: libLLVM-11.0.1 (ORCJIT, skylake)\nEnvironment:\n  JULIA_NUM_THREADS = 12\n  JULIA_PYTHONCALL_EXE = C:\\Users\\genkuroki\\.julia\\conda\\3\\python.exe\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "using LinearAlgebra\nusing BLASBenchmarksCPU\nusing Octavian\nusing BenchmarkHistograms",
      "execution_count": 2,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "M = K = N = 1024\nA = rand(M, K)\nB = rand(K, N)\nC1 = @time(A * B)\nC0 = similar(C1);",
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": "  0.630451 seconds (2.53 M allocations: 142.127 MiB, 8.36% gc time, 97.14% compilation time)\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "@benchmark mul!($C0, $A, $B)",
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 4,
          "data": {
            "text/plain": "samples: 311; evals/sample: 1; memory estimate: 0 bytes; allocs estimate: 0\nns\n\n (1.2e7 - 1.4e7]  \u001b[32m██▏\u001b[39m13\n (1.4e7 - 1.6e7]  \u001b[32m████████████████▍\u001b[39m103\n (1.6e7 - 1.8e7]  \u001b[32m██████████████████████████████ \u001b[39m189\n (1.8e7 - 2.0e7]  \u001b[32m▉\u001b[39m5\n (2.0e7 - 2.2e7]  \u001b[32m \u001b[39m0\n (2.2e7 - 2.4e7]  \u001b[32m \u001b[39m0\n (2.4e7 - 2.6e7]  \u001b[32m \u001b[39m0\n (2.6e7 - 2.8e7]  \u001b[32m \u001b[39m0\n (2.8e7 - 3.0e7]  \u001b[32m \u001b[39m0\n (3.0e7 - 3.4e7]  \u001b[32m▎\u001b[39m1\n\n                  Counts\n\nmin: 11.855 ms (0.00% GC); mean: 16.063 ms (0.00% GC); median: 16.017 ms (0.00% GC); max: 34.308 ms (0.00% GC)."
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "# MKL dgemm\n@benchmark gemmmkl!($C0, $A, $B)",
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 5,
          "data": {
            "text/plain": "samples: 344; evals/sample: 1; memory estimate: 0 bytes; allocs estimate: 0\nns\n\n (1.0e7 - 1.1e7]  \u001b[32m▋\u001b[39m4\n (1.1e7 - 1.3e7]  \u001b[32m█▌\u001b[39m11\n (1.3e7 - 1.4e7]  \u001b[32m███████▍\u001b[39m58\n (1.4e7 - 1.5e7]  \u001b[32m██████████████████████████████ \u001b[39m239\n (1.5e7 - 1.7e7]  \u001b[32m███▉\u001b[39m30\n (1.7e7 - 1.8e7]  \u001b[32m \u001b[39m0\n (1.8e7 - 1.9e7]  \u001b[32m \u001b[39m0\n (1.9e7 - 2.1e7]  \u001b[32m \u001b[39m0\n (2.1e7 - 2.2e7]  \u001b[32m \u001b[39m0\n (2.2e7 - 2.4e7]  \u001b[32m▍\u001b[39m2\n\n                  Counts\n\nmin: 9.866 ms (0.00% GC); mean: 14.514 ms (0.00% GC); median: 14.654 ms (0.00% GC); max: 23.579 ms (0.00% GC)."
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "# MKL dgemm_direct\n@benchmark gemmmkl_direct!($C0, $A, $B)",
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 6,
          "data": {
            "text/plain": "samples: 323; evals/sample: 1; memory estimate: 0 bytes; allocs estimate: 0\nns\n\n (1.0e7  - 1.09e7]  \u001b[32m█▊\u001b[39m5\n (1.09e7 - 1.18e7]  \u001b[32m▊\u001b[39m2\n (1.18e7 - 1.26e7]  \u001b[32m██▍\u001b[39m7\n (1.26e7 - 1.35e7]  \u001b[32m████\u001b[39m12\n (1.35e7 - 1.43e7]  \u001b[32m████████▉\u001b[39m27\n (1.43e7 - 1.52e7]  \u001b[32m██████████████████████████████ \u001b[39m92\n (1.52e7 - 1.6e7 ]  \u001b[32m████████████████████████▎\u001b[39m74\n (1.6e7  - 1.69e7]  \u001b[32m██████████▉\u001b[39m33\n (1.69e7 - 1.78e7]  \u001b[32m████████████▏\u001b[39m37\n (1.78e7 - 1.86e7]  \u001b[32m███████████▏\u001b[39m34\n\n                  Counts\n\nmin: 10.048 ms (0.00% GC); mean: 15.499 ms (0.00% GC); median: 15.280 ms (0.00% GC); max: 18.620 ms (0.00% GC)."
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "# Octavian.jl\n@benchmark matmul!($C0, $A, $B)",
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 7,
          "data": {
            "text/plain": "samples: 265; evals/sample: 1; memory estimate: 0 bytes; allocs estimate: 0\nns\n\n (1.4e7 - 1.7e7]  \u001b[32m███▊\u001b[39m24\n (1.7e7 - 2.0e7]  \u001b[32m██████████████████████████████ \u001b[39m198\n (2.0e7 - 2.3e7]  \u001b[32m██████▍\u001b[39m42\n (2.3e7 - 2.5e7]  \u001b[32m \u001b[39m0\n (2.5e7 - 2.8e7]  \u001b[32m \u001b[39m0\n (2.8e7 - 3.1e7]  \u001b[32m \u001b[39m0\n (3.1e7 - 3.4e7]  \u001b[32m \u001b[39m0\n (3.4e7 - 3.7e7]  \u001b[32m \u001b[39m0\n (3.7e7 - 4.0e7]  \u001b[32m \u001b[39m0\n (4.0e7 - 4.6e7]  \u001b[32m▎\u001b[39m1\n\n                  Counts\n\nmin: 14.231 ms (0.00% GC); mean: 18.854 ms (0.00% GC); median: 18.904 ms (0.00% GC); max: 45.967 ms (0.00% GC)."
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "",
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "@webio": {
      "lastKernelId": null,
      "lastCommId": null
    },
    "kernelspec": {
      "name": "julia-1.7-depwarn-o3",
      "display_name": "Julia 1.7.0-DEV depwarn -O3",
      "language": "julia"
    },
    "language_info": {
      "file_extension": ".jl",
      "name": "julia",
      "mimetype": "application/julia",
      "version": "1.7.0"
    },
    "toc": {
      "nav_menu": {},
      "number_sections": true,
      "sideBar": true,
      "skip_h1_title": false,
      "base_numbering": 1,
      "title_cell": "Table of Contents",
      "title_sidebar": "Contents",
      "toc_cell": false,
      "toc_position": {},
      "toc_section_display": true,
      "toc_window_display": false
    },
    "gist": {
      "id": "6123aef79488bc20b52047656fc6f015",
      "data": {
        "description": "Octavian!",
        "public": true
      }
    },
    "_draft": {
      "nbviewer_url": "https://gist.github.com/6123aef79488bc20b52047656fc6f015"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}

Octavian M=K=N=128.ipynb

Octavian M=K=N=32.ipynb

Octavian M=K=N=512.ipynb

Octavian M=K=N=8.ipynb