{"id":619,"date":"2025-09-29T14:23:16","date_gmt":"2025-09-29T06:23:16","guid":{"rendered":"https:\/\/ferina.top\/?p=619"},"modified":"2025-09-29T14:23:18","modified_gmt":"2025-09-29T06:23:18","slug":"cuda%e5%ad%a6%e4%b9%a0%e7%ac%94%e8%ae%b0","status":"publish","type":"post","link":"https:\/\/ferina.top\/index.php\/2025\/09\/29\/cuda%e5%ad%a6%e4%b9%a0%e7%ac%94%e8%ae%b0\/","title":{"rendered":"CUDA\u5b66\u4e60\u7b14\u8bb0"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\u4e00. \u5b89\u88c5<\/h2>\n\n\n\n<pre class=\"wp-block-code\"><code><code>\nwget https:\/\/developer.download.nvidia.com\/compute\/cuda\/repos\/wsl-ubuntu\/x86_64\/cuda-wsl-ubuntu.pin\nsudo mv cuda-wsl-ubuntu.pin \/etc\/apt\/preferences.d\/cuda-repository-pin-600\nwget https:\/\/developer.download.nvidia.com\/compute\/cuda\/12.2.0\/local_installers\/cuda-repo-wsl-ubuntu-12-2-local_12.2.0-1_amd64.deb\nsudo dpkg -i cuda-repo-wsl-ubuntu-12-2-local_12.2.0-1_amd64.deb\nsudo cp \/var\/cuda-repo-wsl-ubuntu-12-2-local\/cuda-*-keyring.gpg \/usr\/share\/keyrings\/\nsudo apt-get update\nsudo apt-get -y install cuda\n<\/code><\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\"><br>Q\uff1aThe libtinfo5 package isn&#8217;t available<br><br>A: Open a terminal window and run:<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>\nsudo nano \/etc\/apt\/sources.list<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\"><br>Add this line:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>\n\ndeb http:\/\/old-releases.ubuntu.com\/ubuntu\/ lunar universe\n<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Save and exit, then run:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>\n\nsudo apt update<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\"><br>&#8230;and now the install command for CUDA should work, automatically downloading and installing libtinfo5 while installing CUDA.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><br><br>Q: how to change env variable?<br>A: <\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>\n\ncd ~\nnano .bashrc\nexport LD_LIBRARY_PATH=$LD_LIBRARY_PATH:\/usr\/local\/cuda-12.2\/lib64\nexport PATH=$PATH:\/usr\/local\/cuda-12.2\/bin\nexport CUDA_HOME=$CUDA_HOME:\/usr\/local\/cuda-12.2\nexport PATH=\/usr\/local\/cuda\/bin:$PATH\n\nsource ~\/.bashrc\n\nQ\uff1a\u7f16\u8bd1\u7684\u65f6\u5019\u51fa\u73b0\u9519\u8bef\uff0c\u63d0\u793agcc\u95ee\u9898\nA\uff1a\u6211\u4f7f\u7528\u7684cuda12.2\u4e0d\u652f\u6301\u8f83\u65b0\u7684gcc\u7248\u672c\uff0c\u53ef\u4ee5\u652f\u6301gcc-11\u3002\u5b89\u88c5\u540e\uff0c\u5728\u7f16\u8bd1\u65f6\u901a\u8fc7\u6307\u793a\u7f16\u8bd1\u5668\u7248\u672c\u6765\u7f16\u8bd1\u3002\n\ne.g.\n\n<strong>nvcc -ccbin gcc-11 test.cu -o test<\/strong><\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e8c. \u7406\u8bba\u77e5\u8bc6 <\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u67e5\u8be2 <\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">nvidia-smi  \u67e5\u8be2\u57fa\u672c\u4fe1\u606f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">nvidia-smi -q \u67e5\u8be2\u8be6\u7ec6\u4fe1\u606f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">nvidia-smi -q -i 0 \u67e5\u8be2\u7b2c\u4e00\u5757GPU\u7684\u4fe1\u606f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">nvidia-smi -q -i 0 -d MEMORY \u67e5\u8be2\u7b2c\u4e00\u5757GPU\u7684\u7279\u5b9a\u4fe1\u606f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">nvidia-smi -h \u67e5\u8be2\u6307\u4ee4\u4f7f\u7528\u65b9\u5f0f<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2.\u8ba4\u8bc6\u6838\u51fd\u6570<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\">2.1 \u4e00\u7ef4\u7ebf\u7a0b\u6a21\u578b\uff08grid-block-thread\uff09<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e3a\u4ec0\u4e48\u8981\u6709block-&gt;\u52a0\u4e00\u4e2a\u4e2d\u95f4\u5c42\uff0c\u65b9\u4fbf\u7edf\u4e00\u7ec4\u7ec7\u4e00\u4e9b\u4efb\u52a1<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e0b\u9762\u4ee5\u4e00\u7ef4\u6a21\u578b\u4e3a\u4f8b\u5b50<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">gridDim.x      grid\u7684\u7ef4\u5ea6\uff08\u6709\u51e0\u4e2a\u7ebf\u7a0b\u5757\uff09<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">blockDim.x    block\u7ef4\u5ea6\uff08\u6709\u51e0\u4e2a\u7ebf\u7a0b\uff09<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">blockIdx.x      block\u5e8f\u53f7<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">threadIdx.x    \u6bcf\u4e2ablock\u4e2dthread\u7684\u5e8f\u53f7<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Idx = threadIdx.x + blockIdx.x * blockDim.x \uff08\u5f88\u597d\u7406\u89e3\uff0c\u4ece\u57570\u7684\u7b2c0\u7ebf\u7a0b\u5230\u505ai\u540e\u4e00\u4e2a\u5757\u7684\u6700\u540e\u4e00\u4e2a\u7ebf\u7a0b\u987a\u5e8f\u7f16\u53f7\uff09<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h4 class=\"wp-block-heading\">2.2 \u9ad8\u7ef4\u7ebf\u7a0b\u6a21\u578b<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">cuda\u6700\u591a\u652f\u6301\u7ec4\u7ec7\u4e09\u4e2a\u7ef4\u5ea6\u7684grid\u548cblock\uff0c\u663e\u7136\uff0c\u53ef\u4ee5\u901a\u8fc7\u8c03\u7528xyz\u6765\u83b7\u5f97\u4e09\u4e2a\u7ef4\u5ea6\u7684\u4fe1\u606f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f53\u62d3\u5c55\u5230\u591a\u7ef4\u7684\u65f6\u5019\uff0c\u518d\u89c2\u5bdf\u4e4b\u524d\u7684\u7c7b\u4f3c helloFromGPU&lt;&lt;&lt;4, 4&gt;&gt;&gt;(); \u7684\u5f62\u5f0f\uff0c\u53ef\u4ee5\u53d1\u73b0\u5b9e\u9645\u4e0a\u6bcf\u4e2a\u4f4d\u7f6e\u4e0a\u5e94\u8be5\u662f\u4e00\u4e2auint3\uff0c\u8fd9\u91cc\u518d\u6267\u884c\u8fc7\u7a0b\u4e2d\u662f\u5c06\u6ca1\u6709\u6307\u5b9a\u7684\u5176\u4ed6\u6240\u6709\u4fe1\u606f\u9ed8\u8ba4\u4e3a1\uff0c\u5373\u780d\u6389\u4e86\u8be5\u7ef4\u5ea6<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">bid\uff1a grid\u4e2d\u7b2c\u51e0\u4e2ablock<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">tid\uff1a  block\u4e2d\u7b2c\u51e0\u4e2athread<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u6ce8\u610f\uff01<\/strong>\u548c\u591a\u7ef4\u6570\u7ec4\u7f16\u7801\u4e0d\u4e00\u6837\u7684\u4e00\u70b9\uff1a\u5728\u7d22\u5f15\u8ba1\u7b97\u7684\u65f6\u5019\uff0c\u662f\u6309\u7167x\uff0cy\uff0cz\u7684\u987a\u5e8f\u53d8\u5316\u7684\uff0c\u800c\u4e0d\u662fz\uff0cy\uff0cx<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e5f\u5c31\u662f\u8bf4\uff0c(0\uff0c0\uff0c0)\u7684\u4e0b\u4e00\u4e2a\u662f\uff081\uff0c0\uff0c0\uff09\u800c\u4e0d\u662f\uff080\uff0c0\uff0c1\uff09\u3002\u8fd9\u4e00\u70b9\u8981\u4e60\u60ef\u4e00\u4e0b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6309\u7167\u5c42\uff0c\u884c\uff0c\u5217\u7684\u8bf4\u6cd5\u6765\u8bf4\uff0c\u5e94\u8be5\u662fz\u5c42\uff0cy\u884c\uff0cx\u5217\uff0cx\u6700\u591a\u53d8\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">gridDim\u7684\u9650\u5236\uff1ax\uff0cy\uff0cz\u5206\u522b\u662f2^32-1, 2^16-1,2^16-1<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">blockDim\u6709\u4e24\u4e2a\u9650\u5236\uff1a1. x\uff0cy\uff0cz\u5206\u522b\u662f1024, 1024, 64  2. \u603b\u7684blockDim\uff08\u4e5f\u5c31\u662f\u4e58\u79ef\uff09\u4e0d\u5927\u4e8e1024<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">3. nvcc\u7f16\u8bd1<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\">3.1 \u7f16\u8bd1\u6d41\u7a0b\uff1a.cu -&gt; .ptx -&gt; .cubin -&gt; .exe<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Q\uff1a\u4e3a\u4ec0\u4e48\u8981\u6709ptx\u6587\u4ef6<\/strong>\uff1f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>PTX\uff08Parallel Thread eXecution\uff09\u662f CUDA \u7684&#8221;\u4e2d\u95f4\u6c47\u7f16\u8bed\u8a00&#8221;\uff0c\u76f8\u5f53\u4e8e CPU \u4e16\u754c\u7684&#8221;\u5b57\u8282\u7801(bytecode)&#8221;\u6216&#8221;\u4e2d\u95f4\u8868\u793a(IR)&#8221;\u3002<\/strong><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u6838\u5fc3\u4f5c\u7528\uff1a<strong>\u786c\u4ef6\u517c\u5bb9\u6027<\/strong><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e3a\u4ec0\u4e48\u9700\u8981 PTX\uff1a<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>&#8220;\u4e00\u6b21\u7f16\u8bd1\uff0c\u591a\u5904\u8fd0\u884c&#8221;<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>   \/\/ \u4f60\u7684 .cu \u6e90\u7801 \u2192 \u7f16\u8bd1\u4e3a PTX \u2192 \u5728\u4e0d\u540c\u67b6\u6784\u7684 GPU \u4e0a\u8fd0\u884c\n   \/\/ PTX \u53ef\u4ee5\u5728 RTX 3060\u30014090\u3001A100 \u7b49\u4e0d\u540c GPU \u4e0a\u6267\u884c<\/code><\/pre>\n\n\n\n<ol start=\"2\" class=\"wp-block-list\">\n<li><strong>\u89e3\u8026\u7f16\u8bd1\u65f6\u548c\u8fd0\u884c\u65f6<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u7f16\u8bd1\u65f6<\/strong>\uff1a\u751f\u6210\u4e0e\u5177\u4f53\u786c\u4ef6\u65e0\u5173\u7684 PTX<\/li>\n\n\n\n<li><strong>\u8fd0\u884c\u65f6<\/strong>\uff1aGPU \u9a71\u52a8\u5c06 PTX \u5b9e\u65f6\u7f16\u8bd1\u4e3a\u5bf9\u5e94\u786c\u4ef6\u7684\u673a\u5668\u7801<\/li>\n<\/ul>\n\n\n\n<ol start=\"2\" class=\"wp-block-list\">\n<li><strong>\u652f\u6301\u672a\u6765\u786c\u4ef6<\/strong><\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4eca\u5929\u7f16\u8bd1\u7684 PTX \u4ee3\u7801\uff0c\u660e\u5929\u7684\u65b0 GPU \u4e5f\u80fd\u8fd0\u884c<\/li>\n\n\n\n<li>NVIDIA \u65e0\u9700\u4e3a\u6bcf\u4e2a\u65b0\u67b6\u6784\u91cd\u65b0\u53d1\u5e03\u7f16\u8bd1\u5668<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">\u7f16\u8bd1\u6d41\u7a0b\u5bf9\u6bd4\uff1a<\/h2>\n\n\n\n<pre class=\"wp-block-code\"><code>\u4f20\u7edf CPU \u7f16\u8bd1\uff1a\n\u6e90\u7801(.c) \u2192 \u6c47\u7f16(.s) \u2192 \u673a\u5668\u7801(.exe) \u2192 \u76f4\u63a5\u5728 CPU \u8fd0\u884c\n\nCUDA \u7f16\u8bd1\uff1a\n\u6e90\u7801(.cu) \u2192 PTX(.ptx) \u2192 \u8fd0\u884c\u65f6\u7f16\u8bd1 \u2192 CUBIN \u2192 \u5728 GPU \u8fd0\u884c\n                      \u2191\n                (\u7531 GPU \u9a71\u52a8\u5b8c\u6210)<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\">\u5b9e\u9645\u4f8b\u5b50\uff1a<\/h2>\n\n\n\n<pre class=\"wp-block-code\"><code># \u751f\u6210 PTX \u6587\u4ef6\uff08\u53ef\u8de8\u4ee3 GPU \u4f7f\u7528\uff09\nnvcc -ptx mycode.cu -o mycode.ptx\n\n# \u751f\u6210\u7279\u5b9a GPU \u7684 CUBIN\uff08\u6027\u80fd\u6700\u4f18\uff09\nnvcc -cubin mycode.cu -o mycode.cubin<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u603b\u7ed3\uff1aPTX \u662f NVIDIA \u89e3\u51b3&#8221;\u4e0d\u540c\u4ee3 GPU \u67b6\u6784\u5dee\u5f02&#8221;\u7684\u667a\u80fd\u4e2d\u95f4\u5c42\uff0c\u5e73\u8861\u4e86\u517c\u5bb9\u6027\u548c\u6027\u80fd\u3002<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h4 class=\"wp-block-heading\">3.2 \u6307\u5b9a\u865a\u62df\u67b6\u6784\u8ba1\u7b97\u80fd\u529b<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>C\/C++\u6e90\u7801\u7f16\u8bd1\u4e3aPTX\u7684\u65f6\u5019\uff0c\u53ef\u4ee5\u6307\u5b9a\u865a\u62df\u67b6\u6784\u7684\u8ba1\u7b97\u80fd\u529b\uff0c\u7528\u6765\u786e\u5b9a\u4ee3\u7801\u4e2d\u53ef\u4ee5\u4f7f\u7528\u7684CUDA\u529f\u80fd<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>C\/C++\u6e90\u7801\u7f16\u8bd1\u4e3aPTX\uff0c\u8fd9\u4e00\u6b65\u9aa4\u548cGPU\u786c\u4ef6\u65e0\u5173\uff0c\u53ef\u4ee5\u770b\u4f5c\u58f0\u660e\uff08\u544a\u8bc9\u8bbe\u5907\uff0c\u8fd9\u4e2a\u6587\u4ef6\u8d85\u51fa\u4f60\u7684\u80fd\u529b\u8303\u56f4\u4e86\uff0c\u4e0d\u80fd\u6267\u884c\uff09<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7f16\u8bd1\u6307\u4ee4\uff1a \u7f16\u8bd1\u51fa\u7684\u53ef\u6267\u884c\u6587\u4ef6helloworld\u53ea\u80fd\u518d\u8ba1\u7b97\u80fd\u529b\u5927\u4e8e\u7b49\u4e8e6.1\u7684GPU\u4e0a\u8dd1<\/li>\n<\/ul>\n\n\n\n<pre class=\"wp-block-code\"><code>nvcc helloworld.cu -o helloworld -arch=compute_61<\/code><\/pre>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8865\u5145\u6709\u5173\u8ba1\u7b97\u80fd\u529b\uff1a\u4ee3\u8868\u67b6\u6784\u7684\u5b9a\u4e49\uff0c\u4e0d\u4ee3\u8868\u5b9e\u9645\u6027\u80fd\u3002\u5b9e\u9645\u6027\u80fd\u591a\u7528TFLOPS\uff08\u6d6e\u70b9\u6570\u8fd0\u7b97\u5cf0\u503c\u6765\u5b9a\u4e49\uff09\u3002\u524d\u9762\u4e3a\u4e3b\u7248\u672c\uff0c\u540e\u9762\u4e3a\u6b21\u7248\u672c\u3002\u4e3b\u7248\u672c\u53f7\u4e0d\u540c\u7684GPU\u4e00\u822c\u5dee\u5f02\u5f88\u5927\uff0c\u673a\u5668\u7801\u4e5f\u5927\u6982\u7387\u4e0d\u80fd\u5171\u7528\uff0c\u6b21\u7248\u672c\u5219\u4e00\u822c\u53ea\u6709\u7ec6\u5fae\u5dee\u5f02<\/li>\n\n\n\n<li>\u5b9e\u9645\u4e0a\u8fd9\u91cc\u4e5f\u4f53\u73b0\u51faPTX\u7684\u4ef7\u503c\uff0c\u9488\u5bf9\u4e0d\u540c\u67b6\u6784\u7684GPU\uff0c\u4e3a\u4e86\u786e\u4fdd\u4ee3\u7801\u7684\u517c\u5bb9\u6027\uff0c\u9700\u8981\u8fd9\u6837\u7684\u4e2d\u95f4\u5c42<\/li>\n\n\n\n<li>\u4e00\u822c\u60c5\u51b5\u4e0b\uff0c\u4e3a\u4e86\u66f4\u517c\u5bb9\uff0c\u865a\u62df\u67b6\u6784\u7684\u6307\u5b9a\u4f1a\u5c3d\u53ef\u80fd\u4f4e<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">3.3 \u6307\u5b9a\u771f\u5b9e\u67b6\u6784\u8ba1\u7b97\u80fd\u529b<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>PTX\u7f16\u8bd1\u4e3acubin\u7684\u65f6\u5019\uff0c\u53ef\u4ee5\u6307\u5b9a\u771f\u5b9e\u67b6\u6784\u7684\u8ba1\u7b97\u80fd\u529b\u3002<strong>\u8fd9\u4e0e\u5177\u4f53\u7684GPU\u67b6\u6784\u6709\u5173\u3002<\/strong><\/li>\n\n\n\n<li>\u5927\u7248\u672c\u4e4b\u95f4<strong>\u4e0d\u517c\u5bb9\uff01\uff01\uff01<\/strong>\u56e0\u4e3a\u6d89\u53ca\u5230\u4e8c\u8fdb\u5236\u7684cubin\u4ee3\u7801<\/li>\n\n\n\n<li>\u6307\u5b9a\u771f\u5b9e\u67b6\u6784\u8ba1\u7b97\u80fd\u529b\u65f6\uff0c\u5fc5\u987b\u6307\u5b9a\u865a\u62df\u67b6\u6784\u8ba1\u7b97\u80fd\u529b<\/li>\n\n\n\n<li>\u6307\u5b9a\u7684\u771f\u5b9e\u67b6\u6784\u8ba1\u7b97\u80fd\u529b\u5fc5\u987b\u5927\u4e8e\u865a\u62df\u67b6\u6784\u8ba1\u7b97\u80fd\u529b\uff08\u5f88\u597d\u7406\u89e3\uff0c\u865a\u62df\u67b6\u6784\u6307\u5b9a\u58f0\u660e\u4e86\u6700\u4f4e\u7248\u672c\uff0c\u771f\u5b9e\u7684\u80af\u5b9a\u8981\u5927\u4e8e\u5b83\u624d\u80fd\u8dd1\uff09<\/li>\n\n\n\n<li>\u5c0f\u7248\u672c\u7684\u517c\u5bb9\u6027\uff1a\u6bd4\u5982\u6307\u5b9a\u4e3a6.0\uff0c\u90a3\u4e48\u5b9e\u9645\u8dd1\u57286.1\u8fd8\u53ef\u4ee5\u5de5\u4f5c\u3002\u4f46\u662f\u6307\u5b9a\u4e3a6.1\uff0c\u5b9e\u9645\u8dd1\u57286.0\u4e0a\u5c31\u4e0d\u80fd\u5de5\u4f5c\uff08\u901a\u4fd7\u7684\u7406\u89e3\uff1a\u8d8a\u5927\u5c31\u8d8a\u5f3a\uff0c\u6307\u5b9a\u8f83\u5f31\u7684\u4ee3\u7801\u53ef\u4ee5\u8dd1\u5728\u5f3a\u673a\u5668\u4e0a\uff0c\u4f46<strong>\u4ec5\u9650\u5c0f\u7248\u672c<\/strong>\uff09<\/li>\n\n\n\n<li>\u8865\u5145\u4e00\u70b9\uff1aNVIDIA GeForce RTX 3060 = sm_86<\/li>\n<\/ul>\n\n\n\n<pre class=\"wp-block-code\"><code>nvcc helloworld.cu -o helloworld -arch=compute_61 -code=sm_62<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">3.4 \u6307\u5b9a\u591a\u4e2aGPU\u7248\u672c\u7f16\u8bd1<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4f7f\u7f16\u8bd1\u51fa\u6765\u7684\u53ef\u6267\u884c\u6587\u4ef6\u53ef\u4ee5\u5728\u591aGPU\u4e2d\u6267\u884c<\/li>\n\n\n\n<li>\u540c\u65f6\u6307\u5b9a\u591a\u7ec4\u8ba1\u7b97\u80fd\u529b(\u4e0d\u540c\u7684\u6307\u5b9a\u4e2d\u95f4\u4ee5\u7a7a\u683c\u9694\u5f00)<\/li>\n<\/ul>\n\n\n\n<pre class=\"wp-block-code\"><code>-gencode arch=compute_50,code=sm_50\n-gencode arch=compute_60,code=sm_60<\/code><\/pre>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728\u6307\u5b9a\u65f6\uff0c\u9700\u8981\u8003\u8651cuda\u7248\u672c\u652f\u6301\uff0c\u6bd4\u5982cuda12.2\u5c31\u4e0d\u518d\u652f\u6301\u8ba1\u7b97\u80fd\u529b3.5\u7684GPU\u4e86<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">3.5 cuda\u5728\u6ca1\u6709\u6307\u5b9a\u7684\u65f6\u5019\uff0c\u4f1a\u6709\u4e00\u4e2a\u9ed8\u8ba4\u8ba1\u7b97\u80fd\u529b\uff0c\u76ee\u524d10.0\u523012.4\u4e4b\u95f4\u7684\u9ed8\u8ba4\u7f16\u8bd1\u76ee\u6807\u90fd\u662fsm_52<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>sm\u662f\u4ec0\u4e48\uff1aStreaming Multiprocessor  <\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e09. \u4ee3\u7801\u6846\u67b6demo<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u7a0b\u5e8f\u6846\u67b6<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>include &lt;\u5934\u6587\u4ef6&gt;<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u6838\u51fd\u6570\u90e8\u5206<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>int main(void) <\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bbe\u7f6eGPU\u8bbe\u5907<\/li>\n\n\n\n<li>\u5206\u914d\u4e3b\u673a\u4e0e\u8bbe\u5907\u5185\u5b58<\/li>\n\n\n\n<li>\u521d\u59cb\u5316\u4e3b\u673a\u4e2d\u7684\u6570\u636e<\/li>\n\n\n\n<li>\u6570\u636e\u4ece\u4e3b\u673a\u590d\u5236\u5230\u8bbe\u5907<\/li>\n\n\n\n<li>\u8c03\u7528\u6838\u51fd\u6570\u5728\u8bbe\u5907\u4e2d\u8fdb\u884c\u8ba1\u7b97<\/li>\n\n\n\n<li>\u5c06\u8ba1\u7b97\u5f97\u5230\u7684\u6570\u636e\u4ece\u8bbe\u5907\u4f20\u7ed9\u4e3b\u673a<\/li>\n\n\n\n<li>\u91ca\u653e\u4e3b\u673a\u4e0e\u8bbe\u5907\u7684\u5185\u5b58<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">e.g.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>#include &lt;stdio.h&gt;\n#include &lt;cuda_runtime.h&gt;\n\n\/\/ ===================== \u6838\u51fd\u6570\u90e8\u5206 =====================\n__global__ void myKernel(float *d_out, const float *d_in, int N) {\n    int idx = blockIdx.x * blockDim.x + threadIdx.x;  \/\/ \u5168\u5c40\u7ebf\u7a0b\u7d22\u5f15\n    if (idx &lt; N) {\n        d_out&#91;idx] = d_in&#91;idx] * 2.0f;  \/\/ \u793a\u4f8b\uff1a\u6bcf\u4e2a\u5143\u7d20\u4e58 2\n    }\n}\n\n\/\/ ===================== \u4e3b\u51fd\u6570 =====================\nint main(void) {\n    \/\/ 1. \u8bbe\u7f6e GPU \u8bbe\u5907\uff08\u53ef\u9009\uff0c\u591a GPU \u65f6\u6307\u5b9a\uff09\n    int deviceId = 0;\n    cudaSetDevice(deviceId);\n\n    \/\/ 2. \u5b9a\u4e49\u95ee\u9898\u89c4\u6a21\n    int N = 1 &lt;&lt; 16;  \/\/ 65536 \u4e2a\u5143\u7d20\n    size_t size = N * sizeof(float);\n\n    \/\/ 3. \u5206\u914d\u4e3b\u673a\u5185\u5b58\n    float *h_in = (float*)malloc(size);\n    float *h_out = (float*)malloc(size);\n\n    \/\/ 4. \u521d\u59cb\u5316\u4e3b\u673a\u6570\u636e\n    for (int i = 0; i &lt; N; i++) {\n        h_in&#91;i] = static_cast&lt;float&gt;(i);\n    }\n\n    \/\/ 5. \u5206\u914d\u8bbe\u5907\u5185\u5b58\n    float *d_in, *d_out;\n    cudaMalloc((void**)&amp;d_in, size);\n    cudaMalloc((void**)&amp;d_out, size);\n\n    \/\/ 6. \u5c06\u6570\u636e\u4ece\u4e3b\u673a\u590d\u5236\u5230\u8bbe\u5907\n    cudaMemcpy(d_in, h_in, size, cudaMemcpyHostToDevice);\n\n    \/\/ 7. \u914d\u7f6e\u7ebf\u7a0b\u5757\u4e0e\u7f51\u683c\n    int threadsPerBlock = 256;\n    int blocksPerGrid = (N + threadsPerBlock - 1) \/ threadsPerBlock;\n\n    \/\/ 8. \u8c03\u7528\u6838\u51fd\u6570\uff08\u5728 GPU \u4e0a\u5e76\u884c\u6267\u884c\uff09\n    myKernel&lt;&lt;&lt;blocksPerGrid, threadsPerBlock&gt;&gt;&gt;(d_out, d_in, N);\n\n    \/\/ 9. \u540c\u6b65\u7b49\u5f85 GPU \u5b8c\u6210\u8ba1\u7b97\n    cudaDeviceSynchronize();\n\n    \/\/ 10. \u5c06\u8ba1\u7b97\u7ed3\u679c\u4ece\u8bbe\u5907\u590d\u5236\u56de\u4e3b\u673a\n    cudaMemcpy(h_out, d_out, size, cudaMemcpyDeviceToHost);\n\n    \/\/ 11. \u68c0\u67e5\u90e8\u5206\u7ed3\u679c\n    for (int i = 0; i &lt; 10; i++) {\n        printf(\"h_out&#91;%d] = %f\\n\", i, h_out&#91;i]);\n    }\n\n    \/\/ 12. \u91ca\u653e\u8bbe\u5907\u4e0e\u4e3b\u673a\u5185\u5b58\n    cudaFree(d_in);\n    cudaFree(d_out);\n    free(h_in);\n    free(h_out);\n\n    \/\/ 13. \u91cd\u7f6e\u8bbe\u5907\uff08\u91ca\u653e\u4e0a\u4e0b\u6587\uff09\n    cudaDeviceReset();\n\n    return 0;\n}\n\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u8bbe\u7f6eGPU\u8bbe\u5907\u7684\u5e38\u7528\u51fd\u6570<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4e00\u4e2a\u5c0f\u5e38\u8bc6\uff1a\u5f88\u591acuda\u7684\u51fd\u6570\u4f1a\u8fd4\u56decudaError_t\u7c7b\u578b\uff0c\u7528\u6765\u505a\u9519\u8bef\u68c0\u67e5\u3002\u9700\u8981\u53d6\u6570\u636e\u5f80\u5f80\u7528\u5f15\u7528<\/li>\n<\/ul>\n\n\n\n<pre class=\"wp-block-code\"><code>\/\/ \u8fd4\u56deGPU\u8bbe\u5907\u6570\u91cf\nint iDeviceCount = 0;\ncudaGetDeviceCount(&amp;iDeviceCount);\n\n\/\/ \u8bbe\u7f6eGPU\u6267\u884c\u65f6\u4f7f\u7528\u7684\u8bbe\u5907\uff08\u53ea\u6709\u4e00\u4e2a\uff0c\u56e0\u6b64\u75280\uff09\nint iDev = 0;\ncudaSetDevice(iDev);\n<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u5185\u5b58\u7ba1\u7406<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u5185\u5b58\u5206\u914d\uff08cudaMalloc\uff09\uff0c\u6570\u636e\u4f20\u9012(cudaMemcpy)\uff0c\u5185\u5b58\u521d\u59cb\u5316(cudaMemset)\uff0c\u5185\u5b58\u91ca\u653e(cudaFree)<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">3.1 \u5185\u5b58\u5206\u914d<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>float *fpDevice_A;\ncudaMalloc((float**)&amp;fpDevice_A,nBytes);<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">3.2 \u6570\u636e\u62f7\u8d1d<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>\n\/\/ \u56db\u79cd\u62f7\u8d1d\uff1a\n\/\/ cudaMemcpyHostToHost\uff0ccudaMemcpyHostToDevice\uff0ccudaMemcpyDeviceToHost\uff0ccudaMemcpyDeviceToDevice\n\/\/ \u6ce8\u610f\u8fd9\u91cc\u7684\u7b2c\u4e00\u4e2a\u53c2\u6570\u662fdst\uff0c\u7b2c\u4e8c\u4e2a\u53c2\u6570\u662fsrc\uff0c\u6240\u4ee5\u4ec0\u4e48to\u4ec0\u4e48\u7684\u987a\u5e8f\u5b9e\u9645\u4e0a\u662f\u76f8\u53cd\u7684\n\/\/ \u5982HostToDevice\uff0c \u90a3\u4e48\u7b2c\u4e00\u4e2a\u53c2\u6570\u662fDevice\u5730\u5740\uff0c\u7b2c\u4e8c\u4e2a\u53c2\u6570\u662fHost\u5730\u5740\ncudaMemcpy(Device_A, Host_A, nBytes, cudaMemcpyHostToHost)\n\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">3.3 \u5185\u5b58\u521d\u59cb\u5316<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>\/\/ \u6309\u7167\u5b57\u8282\u6570\u8fdb\u884c\u521d\u59cb\u5316\uff0c\u521d\u59cb\u5316\u4e3a0\u9632\u6b62\u6570\u636e\u8bbf\u95ee\u65f6\u51fa\u73b0\u9519\u8bef\ncudaMemset(fpDevice_A,0,nBytes);\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">3.4 \u5185\u5b58\u91ca\u653e<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>cudaFree(pDevice_A);\n<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">4. \u81ea\u5b9a\u4e49\u8bbe\u5907\u51fd\u6570\/\u6838\u51fd\u6570\/\u4e3b\u673a\u51fd\u6570<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\">4.1 \u81ea\u5b9a\u4e49\u8bbe\u5907\u51fd\u6570<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5b9a\u4e49\uff1a\u53ea\u80fd\u6267\u884c\u5728GPU\u8bbe\u5907\u4e0a\u7684\u51fd\u6570\u4e3a\u8bbe\u5907\u51fd\u6570<\/li>\n\n\n\n<li>\u8bbe\u5907\u51fd\u6570\u53ea\u80fd\u88ab\u6838\u51fd\u6570\u6216\u5176\u4ed6\u8bbe\u5907\u51fd\u6570\u8c03\u7528<\/li>\n\n\n\n<li>\u8bbe\u5907\u51fd\u6570\u7528__device__\u4fee\u9970<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">4.2 \u6838\u51fd\u6570<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>__global__\u4fee\u9970\uff0c\u4e3b\u673a\u4e2d\u8c03\u7528\uff0c\u8bbe\u5907\u4e2d\u6267\u884c<\/li>\n\n\n\n<li>__global__\u4fee\u9970\u7b26\u4e0d\u80fd\u4e0e\u5176\u4ed6\u4e24\u4e2a\u6df7\u7528<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">4.3 \u4e3b\u673a\u51fd\u6570<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4e3b\u673a\u7aef\u666e\u901ac++\u51fd\u6570\u53ef\u4ee5\u7528__host__\u4fee\u9970\uff0c\u4e5f\u53ef\u4ee5\u7701\u7565<\/li>\n\n\n\n<li>\u53ef\u4ee5\u540c\u65f6\u7528host\u548cdevice\u4fee\u9970\u4e00\u4e2a\u51fd\u6570\u6765\u51cf\u5c11\u5197\u4f59\u4ee3\u7801\uff08\u56e0\u4e3a\u8fd9\u4e24\u4e2a\u51fd\u6570\u4e00\u822c\u90fd\u4e0d\u6d89\u53ca\u8bbe\u5907\u4e0e\u4e3b\u673a\u7684\u4ea4\u4e92\uff0c\u901a\u5e38\u90fd\u662f\u5728\u5404\u81ea\u7684\u672c\u5730\u4e0a\u8fdb\u884c\uff09\uff0c\u7f16\u8bd1\u5668\u4f1a\u9488\u5bf9\u4e3b\u673a\u548c\u8bbe\u5907\u5206\u522b\u7f16\u8bd1\u8be5\u51fd\u6570<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">5. \u9519\u8bef\u68c0\u67e5<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\">5.1 cudaError_t<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>cudaError_t\u662f\u5927\u90e8\u5206cuda runtime API \u7684\u8fd4\u56de\u503c\u7c7b\u578b,\u8fd9\u4e2a\u8fd4\u56de\u7684\u6267\u884c\u72b6\u6001\u503c\u662f\u4e00\u4e2a\u679a\u4e3e\u53d8\u91cf<\/li>\n\n\n\n<li>cudaSuccess = 0<\/li>\n\n\n\n<li>cudaErrorInvalidValue = 1 <\/li>\n\n\n\n<li>cudaErrorMemoryAllocation = 2<\/li>\n\n\n\n<li>cudaErrorInitializationError = 3<\/li>\n\n\n\n<li>cudaErrorCudartUnloading = 4<\/li>\n\n\n\n<li>&#8230;<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">5.2 \u9519\u8bef\u68c0\u67e5\u51fd\u6570<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>\/\/ \u83b7\u53d6\u9519\u8bef\u4ee3\u7801\u5bf9\u5e94\u540d\u79f0\ncudaGetErrorName(cudaError_t error_code);\n\/\/ \u83b7\u53d6\u9519\u8bef\u4ee3\u7801\u63cf\u8ff0\u4fe1\u606f\ncudaGetErrorString(cudaError_t error_code);\n\n\/\/ CUDA error checking\n\/\/ filename: use __FILE__\n\/\/ lineNumber: use __LINE__\ncudaError_t ErrorCheck(cudaError_t error_code, const char* filename, const int lineNumber){\n    if (error_code != cudaSuccess)\n    {\n        printf(\"CUDA error:\\r\\ncode=%d, name=%s, description=%s\\r\\nfile=%s, line=%d\\r\\n\",\n               error_code, cudaGetErrorName(error_code), cudaGetErrorString(error_code), filename, lineNumber);\n    }\n    return error_code;\n}<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">5.3 \u68c0\u67e5\u6838\u51fd\u6570\u7684\u9519\u8bef<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Q: \u4e3a\u4ec0\u4e48\u6838\u51fd\u6570\u662f\u7279\u6b8a\u7684\uff1f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A: \u6240\u6709\u7684\u6838\u51fd\u6570\u7684\u8fd4\u56de\u503c\u9ed8\u8ba4\u4e3avoid\uff0c\u4e0d\u53cd\u6094cudaError_t\uff0c\u56e0\u6b64\u4e0d\u80fd\u7528\u4ee5\u4e0a\u7684\u65b9\u6cd5\u8fdb\u884c\u68c0\u67e5<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>ErrorCheck(cudaGetLastError(),__FILE__,__LINE__);\nErrorCheck(cudaDeviceSynchronize(),__FILE__,__LINE__);<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">6. cuda\u8ba1\u65f6\uff1a\u6027\u80fd\u68c0\u6d4b<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\">6.1 \u4e8b\u52a1\u7cfb\u7edf\u8ba1\u65f6<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>cudaEvent_t start,stop;\nErrorCheck(cudaEventCreate(&amp;start), __FILE__, __LINE__);\nErrorCheck(cudaEventCreate(&amp;stop), __FILE__, __LINE__);\nErrorCheck(cudaEventRecord(start), __FILE__, __LINE__);\ncudaEventQuery(start); \/\/ \u6b64\u5904\u4e0d\u53ef\u7528\u9519\u8bef\u68c0\u6d4b\u51fd\u6570\n\n\/**************************************************************\n\u9700\u8981\u8ba1\u65f6\u7684\u4ee3\u7801\n**************************************************************\/\n\nErrorCheck(cudaEventRecord(stop), __FILE__, __LINE__);\nErrorCheck(cudaEventSynchronize(stop), __FILE__, __LINE__);\nfloat elapsed_time;\nErrorCheck(cudaEventElapsedTime(&amp;elapsed_time, start, stop), __FILE__, __LINE__);\nprintf(\"Time = %g ms.\\n\", elapsed_time);\n\nErrorCheck(cudaEventDestroy(start), __FILE__, __LINE__);\nErrorCheck(cudaEventDestroy(stop), __FILE__, __LINE__);<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">6.2 nvsight<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u5feb\u901f\u770b\u6027\u80fd\u74f6\u9888<\/strong>  <code>nsys profile .\/a <\/code>+ <code>nsys stats .\/report1.nsys-rep<\/code><\/li>\n\n\n\n<li><strong>\u8be6\u7ec6\u6838\u51fd\u6570\u6027\u80fd<\/strong> <code>ncu .\/a.out<\/code><\/li>\n\n\n\n<li><strong>\u5b8c\u6574\u6307\u6807\u91c7\u96c6<\/strong> <code>ncu --set full .\/a.out<\/code><\/li>\n\n\n\n<li><strong>\u53ea\u5173\u6ce8\u4e00\u4e2a kernel<\/strong> <code>ncu --kernel-name regex:myKernel .\/a.out<\/code><\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u56db. \u7ec4\u7ec7\u7ebf\u7a0b\u6a21\u578b\uff08\u5149\u7ebf\u8ffd\u8e2a\u5668\uff1f\uff09<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u4e8c\u7ef4\u7f51\u683c\u4e8c\u7ef4\u7ebf\u7a0b\u5757<\/h3>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-1024x623.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"623\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-1024x623.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-702\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-3-1024x374.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"374\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-3-1024x374.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-705\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u4e8c\u7ef4\u7f51\u683c\u4e00\u7ef4\u7ebf\u7a0b\u5757<\/h3>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-1-1024x617.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"617\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-1-1024x617.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-703\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-2-1024x330.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"330\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-2-1024x330.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-704\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u4e00\u7ef4\u7f51\u683c\u4e00\u7ef4\u7ebf\u7a0b\u5757<\/h3>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-4-1024x471.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"471\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-4-1024x471.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-707\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-5-1024x417.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"417\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-5-1024x417.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-708\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e94. GPU\u786c\u4ef6\u8d44\u6e90<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. SM-Streaming Multiprocessor<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>SIMT \u5355\u6307\u4ee4\u591a\u7ebf\u7a0b\u67b6\u6784<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4e00\u4e2a\u7ebf\u7a0b\u675f\u542b\u670932\u4e2a\u7ebf\u7a0b\u3002\u5728\u4e00\u4e2a\u7ebf\u7a0b\u5757\u4e2d\uff0c0-31\u5c5e\u4e8e\u7b2c0\u4e2a\u7ebf\u7a0b\u675f\uff0c\u4f9d\u6b21\u7c7b\u63a8<\/li>\n\n\n\n<li>\u6bcf\u4e2a\u7ebf\u7a0b\u675f\u4e2d\u53ea\u80fd\u5305\u542b\u540c\u4e00\u7ebf\u7a0b\u5757\u4e2d\u7684\u7ebf\u7a0b<\/li>\n\n\n\n<li>\u91cd\u70b9\uff1a\u7ebf\u7a0b\u675f\u662fGPU\u786c\u4ef6\u4e0a\u771f\u6b63\u505a\u5230\u4e86\u5e76\u884c\uff0c\u5728\u8fd9\u4e4b\u4e0a\u7684\u7ed3\u6784\u672c\u8d28\u4e0a\u90fd\u662f\u5728\u505a\u5e76\u53d1<\/li>\n\n\n\n<li>\u4e3a\u4e86\u6700\u5927\u5316\u6548\u7387\uff0c\u9700\u8981\u4f7f\u7ebf\u7a0b\u5757\u4e2d\u7684\u7ebf\u7a0b\u6570\u91cf\u662f32\u7684\u6574\u6570\u500d<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7ebf\u7a0b\u5757\u4e2d\u7684\u6240\u6709\u7ebf\u7a0b\u5fc5\u987b\u5168\u90e8\u88ab\u5206\u914d\u5230\u4e00\u4e2aSM\u4e0a\uff0c\u4f46\u4e00\u4e2aSM\u53ef\u4ee5\u88ab\u5206\u914d\u591a\u4e2a\u7ebf\u7a0b\u5757<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4e00\u4e2a\u7ebf\u7a0b\u5757\u4e0d\u80fd\u88ab\u5206\u914d\u5230\u5f88\u591aSM\u4e0a<\/li>\n\n\n\n<li>\u5728\u4e00\u4e2aSM\u4e0a\uff0c\u652f\u6301\u5f88\u591a\u7ebf\u7a0b<strong>\u5e76\u53d1<\/strong>\u6267\u884c\u3002<\/li>\n\n\n\n<li>\u5e76\u53d1\u4e0e\u5e76\u884c\u7684\u533a\u522b\uff1a\u5e76\u884c\u6307\u4e92\u4e0d\u5e72\u6270\u7684\u5e73\u884c\u6267\u884c\u3002\u5e76\u53d1\u6307\u901a\u8fc7\u8c03\u5ea6\u7684\u65b9\u5f0f\uff0c\u597d\u50cf\u5728\u5e76\u884c\u6267\u884c\uff0c\u4f46\u5b9e\u9645\u5728\u6bcf\u4e2a\u65f6\u523b\u53ea\u6267\u884c\u4e00\u4e2a\u4efb\u52a1\u3002<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u5185\u5b58\u6a21\u578b<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5bc4\u5b58\u5668-\u7f13\u5b58-\u4e3b\u5b58\uff08\u8fd0\u884c\u5185\u5b58\uff09-\u78c1\u76d8\u5b58\u50a8\u5668<\/li>\n\n\n\n<li>\u4e3b\u5b58-DRAM\uff08\u52a8\u6001\u968f\u673a\u5b58\u50a8\u5668\uff09 \u4f4e\u5ef6\u8fdf\u5185\u5b58-SRAM\uff08\u9759\u6001\u968f\u673a\u5b58\u50a8\u5668\uff09<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">GPU\u5185\u5b58\u67b6\u6784<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5bc4\u5b58\u5668(register) &#8211; \u5c5e\u4e8e\u7ebf\u7a0b\u7684\u6982\u5ff5\uff0c\u4e0d\u80fd\u5171\u4eab<\/li>\n\n\n\n<li>\u5171\u4eab\u5185\u5b58(shared memory) &#8211; \u5c5e\u4e8e\u7ebf\u7a0b\u5757\u7684\u6982\u5ff5\uff0c\u4e00\u4e2a\u7ebf\u7a0b\u5757\u4e2d\u7684\u7ebf\u7a0b\u53ef\u4ee5\u5171\u4eab\u8fd9\u90e8\u5206\u5185\u5b58\u3002\u7531\u4e8e\u5171\u4eab\u5185\u5b58\u662f\u7247\u4e0a\u5185\u5b58\uff0c\u8ddd\u79bb\u5904\u7406\u5668\u6bd4\u8f83\u8fd1\uff0c\u5ef6\u8fdf\u4f4e\uff0c\u5e26\u5bbd\u5927\uff0c\u5177\u6709\u9ad8\u901f\u8bbf\u5b58\u7684\u7279\u6027\uff0c\u56e0\u6b64\u9700\u8981\u9ad8\u901f\u8bbf\u5b58\u7684\u6570\u636e\u53ef\u4ee5\u4fdd\u5b58\u5728\u5171\u4eab\u5185\u5b58\u4e2d\u3002<\/li>\n\n\n\n<li>\u672c\u5730\/\u5c40\u90e8\u5185\u5b58(local memory) &#8211; \u5c5e\u4e8e\u7ebf\u7a0b\u7684\u6982\u5ff5\uff0c\u4e0d\u80fd\u5171\u4eab\uff0c\u76f8\u5bf9\u4e8e\u5bc4\u5b58\u5668\u8f83\u6162\uff0c\u76f8\u5bf9\u4e8e\u5171\u4eab\u5185\u5b58\u4e5f\u8f83\u6162\uff0c\u56e0\u4e3a\u5b83\u4e0d\u662f\uff08\u82af\uff09\u7247\u4e0a\u7684\u5185\u5b58<\/li>\n\n\n\n<li>\u5e38\u91cf\u5185\u5b58(constant memory) &#8211; \u5c5e\u4e8e\u5168\u5c40\u7684\u6982\u5ff5\uff0c\u53ea\u80fd\u88ab\u8bfb\u4e0d\u80fd\u88ab\u7ebf\u7a0b\u5199\uff08\u53ef\u4ee5\u88abCPU\u5199\uff09<\/li>\n\n\n\n<li>\u7eb9\u7406\u5185\u5b58(texture memory) &#8211; \u5c5e\u4e8e\u5168\u5c40\u7684\u6982\u5ff5\uff0c\u53ea\u80fd\u88ab\u8bfb\u4e0d\u80fd\u88ab\u7ebf\u7a0b\u5199\uff08\u53ef\u4ee5\u88abCPU\u5199\uff09<\/li>\n\n\n\n<li>\u5168\u5c40\u5185\u5b58(global memory) &#8211; \u5c5e\u4e8e\u5168\u5c40\u7684\u6982\u5ff5\uff0c\u662fGPU\u4e2d\u6700\u5927\u7684\u5185\u5b58\uff0c\u53ef\u8bfb\u5199<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-6-1024x804.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"804\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-6-1024x804.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-715\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-7-1024x328.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"328\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-7-1024x328.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-718\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h4 class=\"wp-block-heading\">2.1 register, local memory<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6838\u51fd\u6570\u4e2d\u5b9a\u4e49\u7684\uff0c\u4e0d\u52a0\u4efb\u4f55\u9650\u5b9a\u7b26\u7684\u53d8\u91cf\u4e00\u822c\u5b58\u653e\u5728\u5bc4\u5b58\u5668\u4e2d\uff08\u5982__shared__\u4fdd\u5b58\u5728shared memory\u4e2d\uff09<\/li>\n\n\n\n<li>\u5185\u5efa\u53d8\u91cf\u5b58\u653e\u5728\u5bc4\u5b58\u5668\u4e2d\uff0cgridDim\uff0cblockDim\uff0cblockIdx<\/li>\n\n\n\n<li>\u6838\u51fd\u6570\u4e2d\u5b9a\u4e49\u7684\uff0c\u4e0d\u52a0\u4efb\u4f55\u9650\u5b9a\u7b26\u7684\u6570\u7ec4\u53ef\u80fd\u5728register\uff0c\u4e5f\u53ef\u80fd\u662flocal<\/li>\n\n\n\n<li>\u5bc4\u5b58\u5668\u662f32\u4f4d\u7684\uff0c\u76f8\u5f53\u4e8e\u4e00\u4e2adouble\u9700\u8981\u4e24\u4e2a\u5bc4\u5b58\u5668<\/li>\n\n\n\n<li>\u5bf9\u4e8e\u6211\u7684sm_86,\u6bcf\u4e2aSM\u4e2d\u542b\u670964K\u4e2a\u5bc4\u5b58\u5668\uff0c\u4e00\u4e2a\u5757\u6700\u591a\u540c\u65f6\u4f7f\u752864K\uff08\u4e5f\u5c31\u662f\u5168\u7528\uff09\uff0c\u5bf9\u4e8e\u6bcf\u4e2a\u7ebf\u7a0b\u6765\u8bf4\uff0c\u6700\u5927\u7684\u5bc4\u5b58\u5668\u6570\u91cf\u662f255\u4e2a<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-8-1024x422.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"422\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-8-1024x422.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-721\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5bc4\u5b58\u5668\u653e\u4e0d\u4e0b\uff0c\u5c31\u4f1a\u653e\u5728local memory<\/li>\n\n\n\n<li>\u5bf9\u4e8e\u4e00\u4e2a\u7ebf\u7a0b\uff0c\u53ef\u4ee5\u4f7f\u7528<strong>512KB<\/strong>\u7684\u672c\u5730\u5185\u5b58\uff0c\u800c\u5bf9\u5e94\u7684\u5bc4\u5b58\u5668\u5185\u5b58\u5c31\u8981\u5c0f\u7684\u591a\uff0c\u6700\u591a\u53ea\u6709\uff08255\/2\uff09B<\/li>\n\n\n\n<li>\u4ece\u786c\u4ef6\u89d2\u5ea6\u770b\uff0clocal\u662fglobal\u7684\u4e00\u90e8\u5206\uff0c\u5c5e\u4e8e\u7247\u5916\u5185\u5b58\uff0c\u5ef6\u8fdf\u5f88\u9ad8\uff0c\u56e0\u6b64\u8981\u5c3d\u91cf\u907f\u514d\u4f7f\u7528\u3002\u56e0\u6b64\uff0c\u5728\u7a0b\u5e8f\u8bbe\u8ba1\u65f6\u8981\u5c3d\u91cf\u907f\u514d\u6ea2\u51fa\uff0c\u591a\u7528\u5bc4\u5b58\u5668<\/li>\n\n\n\n<li>\u5bf9\u4e8e\u8ba1\u7b97\u80fd\u529b2.0\u4ee5\u4e0a\u7684\u8bbe\u5907\uff0clocal\u5b58\u50a8\u5728\u6bcf\u4e2aSM\u7684\u4e00\u7ea7\u7f13\u5b58\u548c\u8bbe\u5907\u7684\u4e8c\u7ea7\u7f13\u5b58\u4e2d<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5bc4\u5b58\u5668\u6ea2\u51fa\uff1a\u4e00\u4e2aSM\u5e76\u53d1\u8fd0\u884c\u5927\u91cf\u7684\u7ebf\u7a0b\u5757\/\u7ebf\u7a0b\u675f\uff0c\u603b\u7684\u9700\u6c42\u5bb9\u91cf\u5927\u4e8e64KB\uff0c\u6216\u8005\u5355\u4e2a\u7ebf\u7a0b\u8fd0\u884c\u6240\u9700\u7684\u5bc4\u5b58\u5668\u6570\u91cf\u5927\u4e8e255\u4e2a\uff0c\u5c31\u4f1a\u53d1\u751f\u5bc4\u5b58\u5668\u6ea2\u51fa\u3002\u8fd9\u4f1a\u964d\u4f4e\u7a0b\u5e8f\u7684\u8fd0\u884c\u6027\u80fd<\/li>\n\n\n\n<li>\u8fd9\u4e2a\u6ea2\u51fa\u7684\u90e8\u5206\u4e5f\u53ef\u4ee5\u4fdd\u5b58\u5728\u7f13\u5b58\u4e2d\uff0c\u4ece\u800c\u63d0\u9ad8\u6548\u7387<\/li>\n\n\n\n<li>\u68c0\u67e5\u5bc4\u5b58\u5668\u547d\u4ee4 nvcc &#8211;resource-usage hello.cu -o hello -arch=sm_86<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h4 class=\"wp-block-heading\">2.2 global \u5168\u5c40\u5185\u5b58\uff0c\u9759\u6001\u521d\u59cb\u5316\u4e0e\u52a8\u6001\u521d\u59cb\u5316<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7279\u70b9\uff1a\u5bb9\u91cf\u6700\u5927\uff0c\u5ef6\u8fdf\u6700\u591a\uff0c\u4f7f\u7528\u6700\u591a<\/li>\n\n\n\n<li>\u6570\u636e\u6240\u6709\u7ebf\u7a0b\u53ef\u89c1\uff0chost\u7aef\u53ef\u89c1\uff0c\u4e14\u5177\u6709\u4e0e\u7a0b\u5e8f\u76f8\u540c\u7684\u751f\u547d\u5468\u671f\uff08\u5373\u7531\u4e3b\u673a\u7aef\u8fdb\u884c\u51b3\u5b9a\uff09<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5168\u5c40\u5185\u5b58\u521d\u59cb\u5316\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u52a8\u6001\u5168\u5c40\u5185\u5b58\uff1ahost\u7aef\u4f7f\u7528runtime API cudaMalloc\u52a8\u6001\u58f0\u660e\u5185\u5b58\u7a7a\u95f4\uff0c\u7531cudaFree\u91ca\u653e<\/li>\n\n\n\n<li>\u9759\u6001\u5168\u5c40\u5185\u5b58\uff1a\u4f7f\u7528__device__\u5173\u952e\u5b57\u9759\u6001\u58f0\u660e\u5168\u5c40\u5185\u5b58\uff0c\u5728\u7f16\u8bd1\u5668\u7f16\u8bd1\u7684\u65f6\u5019\u5c31\u786e\u5b9a\u4e0b\u6765\u3002<\/li>\n\n\n\n<li>\u4e00\u4e2a\u9700\u8981\u6ce8\u610f\u7684\u70b9\uff1a\u9759\u6001\u5168\u5c40\u53d8\u91cf\u5fc5\u987b\u8981\u5728\u6240\u6709\u7684\u4e3b\u673a\u51fd\u6570\u548c\u6838\u51fd\u6570\u4e4b\u5916\u58f0\u660e\uff0c\u8fd9\u6837\u624d\u80fd\u4f53\u73b0\u5b83\u7684\u5168\u5c40\u6027\u3002\u6838\u51fd\u6570\u53ef\u4ee5\u76f4\u63a5\u8bbf\u95ee\u9759\u6001\u53d8\u91cf\uff0c\u4e0d\u9700\u8981\u53d8\u91cf\u7684\u4f20\u9012\u8fc7\u7a0b\u3002\u4f46\u662f\uff01\u4e3b\u673a\u4e2d\u7684\u51fd\u6570\u662f\u4e0d\u80fd\u76f4\u63a5\u8bbf\u95ee\u9759\u6001\u5168\u5c40\u53d8\u91cf\u7684\u3002\u9700\u8981\u901a\u8fc7\u4ee5\u4e0b\u4e24\u4e2aAPI\u8fdb\u884c\u8bfb\u5199\uff1a<\/li>\n<\/ul>\n\n\n\n<pre class=\"wp-block-code\"><code>cudaMemcpyToSymbol(const void *symbol. const *src, size_t count, size_t offset, cudaMemcpyKind kind)\ncudaMemcpyFromSymbol(void *dst, const void *symbol, size_t count, size_t offset, cudaMemcpyKind kind)\n\n\/\/ \u4e00\u4e2a\u793a\u4f8b\n__device__ int d_x = 1;\n__device__ int d_y&#91;2];\n\n__global__ void kernel(void){\n    \/\/ \u5c55\u793akernel\u53ef\u4ee5\u76f4\u63a5\u8bbf\u95ee\n    d_y&#91;0] += d_x;\n    d_y&#91;1] += d_x;\n}\n\nint main(int argc, char**argv){\n    int h_y&#91;2] = {10,20};\n    cudaMemcpyToSymbol(d_y, h_y, sizeof(int)*2);\n\n    \/*\u4e2d\u95f4\u63d2\u5165kernel\u51fd\u6570\u7684\u76f8\u5173\u64cd\u4f5c*\/\n\n    cudaMemcpyFromSymbol(h_y, d_y, sizeof(int)*2);\n\n}\n<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h4 class=\"wp-block-heading\">2.3 \u5171\u4eab\u5185\u5b58\uff08shared memory\uff09<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5bf9\u4e8e\u6211\u7684\u663e\u5361sm_86\u6765\u8bf4\uff0c\u6bcf\u4e00\u4e2aSM\u5bf9\u5e94\u7684\u5171\u4eab\u5185\u5b58\u5927\u5c0f\u4e3a100KB\uff0c\u6bcf\u4e00\u4e2a\u7ebf\u7a0b\u5757\u80fd\u591f\u62e5\u6709\u7684\u6700\u5927\u7684\u5171\u4eab\u5185\u5b58\u4e3a99KB<\/li>\n\n\n\n<li>shared\u6709\u4ec5\u6b21\u4e8eregister\u7684\u901f\u5ea6(on-chip\u7684\u542b\u91d1\u91cf)<\/li>\n\n\n\n<li>\u4f7f\u7528__shared__\u4fee\u9970\u7684\u53d8\u91cf\u5b58\u653e\u5728shared\u4e2d\uff0c\u53ef\u5b9a\u4e49\u52a8\u6001\u4e0e\u9759\u6001\u4e24\u79cd<\/li>\n\n\n\n<li>\u5982\u679c\u5728\u5355\u4e2a\u7ebf\u7a0b\u5757\u4e2d\u5206\u914d\u8fc7\u91cf\u7684\u5171\u4eab\u5185\u5b58\uff08\u60f3\u8c61\u76f4\u63a5\u7ed9\u4e00\u4e2ablock\u5206\u914d99KB\uff09\uff0c\u5c06\u4f1a\u9650\u5236\u6d3b\u8dc3\u7ebf\u7a0b\u675f\u7684\u6570\u91cf\uff08\u53ea\u6709\u8fd9\u4e00\u4e2ablock\u80fd\u6709\u8db3\u591f\u7684shared memory\u4e86\uff09<\/li>\n\n\n\n<li>\u8bbf\u95eeshared memory\u5fc5\u987b\u52a0\u5165\u540c\u6b65\u673a\u5236\uff1avoid __syncthreads();<\/li>\n\n\n\n<li>\u8fd9\u4e2a\u51fd\u6570\u7684\u4f5c\u7528\u4ec5\u9650\u4e8e\u4fdd\u8bc1\u4e00\u4e2ablock\u4e2d\u7684\u7ebf\u7a0b\u540c\u6b65\uff0c\u4e0d\u540cblock\u662f\u4e0d\u80fd\u4fdd\u8bc1\u7684<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">shared memory\u7684\u4e3b\u8981\u4f5c\u7528\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u51cf\u5c11kernel\u5bf9global\u7684\u8bbf\u95ee\uff0c\u66f4\u52a0\u9ad8\u6548<\/li>\n\n\n\n<li>\u6539\u53d8\u5168\u5c40\u5185\u5b58\u8bbf\u95ee\u5185\u5b58\u7684\u5185\u5b58\u4e8b\u52a1\u65b9\u5f0f\uff0c\u63d0\u9ad8\u6570\u636e\u8bbf\u95ee\u7684\u5e26\u5bbd<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u9759\u6001\u5171\u4eab\u5185\u5b58\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u58f0\u660e\u65b9\u5f0f\uff1a __shared__ float tile[size,size];<\/li>\n\n\n\n<li>\u9759\u6001\u5171\u4eab\u5185\u5b58\u7684\u4f5c\u7528\u57df\u968f\u7740\u58f0\u660e\u4f4d\u7f6e\u53d8\u5316\uff1a<\/li>\n\n\n\n<li>1. \u5728kernel\u4e2d\u58f0\u660e\u65f6\uff0c\u9759\u6001\u5171\u4eab\u5185\u5b58\u4f5c\u7528\u57df\u5c40\u9650\u5728\u8fd9\u4e2a\u6838\u51fd\u6570\u4e2d<\/li>\n\n\n\n<li>2. \u5728\u6587\u4ef6\u4e2d\uff0c\u4efb\u4f55kernel\u5916\u58f0\u660e\u65f6\uff0c\u9759\u6001\u5171\u4eab\u5185\u5b58\u4f5c\u7528\u57df\u5bf9\u6240\u6709\u6838\u51fd\u6570\u6709\u6548<\/li>\n\n\n\n<li>\u540c\u9759\u6001global\u4e00\u6837\uff0c\u5728\u7f16\u8bd1\u7684\u65f6\u5019\u5c31\u8981\u786e\u5b9a\u5185\u5b58\u5927\u5c0f<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u52a8\u6001\u5171\u4eab\u5185\u5b58\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7528\u7c7b\u4f3cextern __shared__ float s_array[]; \u5b9a\u4e49<\/li>\n\n\n\n<li>\u4e0d\u80fd\u4f7f\u7528 *s_array\uff0c\u4e0d\u80fd\u5ffd\u7565extern\uff0c\u4e0d\u80fd\u5f80\u62ec\u53f7\u91cc\u52a0\u6570 <\/li>\n\n\n\n<li>\u5206\u914d\u5185\u5b58\u5927\u5c0f\u5728\u4ec0\u4e48\u5730\u65b9\u5b9a\u4e49\uff1akernel&lt;&lt;&lt;grid,block,nBytes&gt;&gt;&gt;,\u8fd9\u91cc\u7684\u7b2c\u4e09\u4e2a\u53c2\u6570\u7528\u6765\u6307\u5b9a\u52a8\u6001\u5171\u4eab\u5185\u5b58<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h4 class=\"wp-block-heading\">2.4 \u5e38\u91cf\u5185\u5b58<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5e38\u91cf\u5185\u5b58\u662f\u7531\u5168\u5c40\u7f13\u5b58\u7684\u5168\u5c40\u5185\u5b58\uff0c\u6570\u91cf\u6709\u9650\uff0c\u5927\u5c0f\u4ec5\u4e3a64KB\u3002\u7531\u4e8e\u6709\u7f13\u5b58\uff0c\u7ebf\u7a0b\u675f\u5728\u8bfb\u53d6\u76f8\u540c\u7684\u5e38\u91cf\u5185\u5b58\u6570\u636e\u65f6\uff0c\u8bbf\u95ee\u901f\u5ea6\u6bd4\u5168\u5c40\u5185\u5b58\u5feb\u3002<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5e38\u91cf\u5185\u5b58\u4e2d\u7684\u6570\u636e\u5bf9\u540c\u4e00\u7f16\u8bd1\u5355\u5143\u5185\u6240\u6709\u7ebf\u7a0b\u53ef\u89c1\u3002<\/li>\n\n\n\n<li>\u4f7f\u7528__constant__\u4fee\u9970\u3002\u4e0d\u80fd\u5b9a\u4e49\u5728kernel\u51fd\u6570\u4e2d\u3002\u5e38\u91cf\u5185\u5b58\u662f\u9759\u6001\u5b9a\u4e49\u7684\u3002<\/li>\n\n\n\n<li>\u53ea\u53ef\u8bfb\uff0c\u4e0d\u53ef\u5199<\/li>\n\n\n\n<li>\u6211\u4eec\u5728\u7ed9\u6838\u51fd\u6570\u4f20\u9012<strong>\u6570\u503c<\/strong>\u53c2\u6570\u65f6\uff0c\u8fd9\u4e2a\u53d8\u91cf\u5c31\u5b58\u653e\u4e8e\u5e38\u91cf\u5185\u5b58\u3002<\/li>\n\n\n\n<li>\u56e0\u4e3a\u5e38\u91cf\u5185\u5b58\u672c\u8d28\u4e0a\u5c31\u662f\u9759\u6001\u5168\u5c40\u5185\u5b58\uff0c\u56e0\u6b64\u521d\u59cb\u5316\u5fc5\u987b\u8981\u4f7f\u7528cudaMemcpyToSymbol<\/li>\n\n\n\n<li>\u5e38\u91cf\u5185\u5b58\u7684\u4f7f\u7528\u573a\u666f\uff1a\u5f53\u7ebf\u7a0b\u675f\u4e2d\u7684\u7ebf\u7a0b\u4ece\u76f8\u540c\u7684\u5185\u5b58\u5730\u5740\u4e2d\u8bfb\u53d6\u6570\u636e\u65f6\uff0c\u5e38\u91cf\u5185\u5b58\u8868\u73b0\u8f83\u597d\uff0c\u56e0\u4e3a\u8fd9\u4e2a\u8fc7\u7a0b\u4e0d\u9700\u8981\u6240\u6709\u7ebf\u7a0b\u7684\u8bbf\u95ee\uff0c\u53ea\u9700\u8981\u8bfb\u53d6\u4e00\u6b21\u5e76\u5e7f\u64ad\u5c31\u53ef\u4ee5\u3002\u6bd4\u5982\u6570\u5b66\u516c\u5f0f\u4e2d\u7684\u7cfb\u6570\u3002<\/li>\n<\/ul>\n\n\n\n<pre class=\"wp-block-code\"><code>__constant__ float c_data;        \/\/ \u8fd9\u4e2a\u672a\u521d\u59cb\u5316\u7684\u5e38\u91cf\u5185\u5b58\u5fc5\u987b\u7528cudaMemcpyToSymbol\n__constant__ float c_data2 = 6.6f; \/\/ \u5df2\u7ecf\u521d\u59cb\u5316\uff0c\u4e0d\u5fc5\u8981\u4f7f\u7528,\u53ef\u4ee5\u7528cudaMemcpyFromSymbol\u5728 \u4e3b\u673a\u7aef\u8c03\u7528\n\nfloat h_data = 8.8f;\n\/\/ \u7528\u4e00\u4e2ahost\u91cc\u7684float\u521d\u59cb\u5316\u5e38\u91cf\u5185\u5b58\u4e2d\u7684c_data\nCUDA_CHECK(cudaMemcpyToSymbol(c_data, &amp;h_data, sizeof(float)));\n\n__global__ void kernel_2(int N) \/\/ \u8fd9\u4e2a\u91cc\u9762\u7684N\u5c31\u662f\u5b58\u653e\u5728\u5e38\u91cf\u5185\u5b58\u4e2d\uff0c\u56e0\u4e3a\u6240\u6709\u7684kernel\u90fd\u9700\u8981\u8bbf\u95ee\uff0c\u8fd9\u6837\u505a\u5f88\u9ad8\u6548  <\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h4 class=\"wp-block-heading\">2.5 GPU\u7f13\u5b58<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li>GPU\u7f13\u5b58\uff1a\u4e0d\u53ef\u7f16\u7a0b\u7684\u5185\u5b58\uff08\u81ea\u52a8\u5316\u8fd9\u4e00\u5757\uff09<\/li>\n\n\n\n<li>\u6bcf\u4e2aSM\u90fd\u6709\u4e00\u4e2aL1 Cache\uff0c \u6240\u6709\u7684SM\u5171\u4eab\u4e00\u4e2aL2 Cache<\/li>\n\n\n\n<li>L1,L2 Cache \u7528\u6765\u5b58\u653e local memory\uff08\u56de\u5fc6\uff0c\u5c5e\u4e8e\u6bcf\u4e2a\u7ebf\u7a0b\u7684\u72ec\u7acb\u7684\u90e8\u5206\uff0c\u672c\u8d28\u4e0a\u65f6global\uff0c\u5f88\u6162\uff09\u548cglobal memory\u7684\u6570\u636e\uff0c\u4e5f\u5305\u62ec\u5bc4\u5b58\u5668\u6ea2\u51fa\u7684\u90e8\u5206\uff08\u56de\u5fc6\uff0c\u6ea2\u51fa\u7684\u90e8\u5206\u4f1a\u5b58\u5230local\u91cc\uff09<\/li>\n\n\n\n<li>\u5728GPU\u4e0a\u53ea\u6709\u5185\u5b58\u52a0\u8f7d\u53ef\u4ee5\u88ab\u7f13\u5b58\uff0c\u5185\u5b58\u5b58\u50a8\u64cd\u4f5c\u4e0d\u80fd\u88ab\u7f13\u5b58\uff1f\uff1f<\/li>\n\n\n\n<li>\u6bcf\u4e2aSM\u6709\u4e00\u4e2a\u53ea\u8bfb\u5e38\u91cf\u7f13\u5b58\u548c\u53ea\u8bfb\u7eb9\u7406\u7f13\u5b58\uff0c\u7528\u4e8e\u63d0\u5347\u6027\u80fd<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-9-1024x496.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"496\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-9-1024x496.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-752\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5982\u56fe\uff0cglobal memory\u5728\u7269\u7406\u4e0a\u7684\u4f4d\u7f6e\u5728DRAM\u4e0a\uff0c\u56e0\u6b64\u5728\u52a0\u8f7d\u5168\u5c40\u5185\u5b58\u6570\u636e\u7684\u65f6\u5019\u5fc5\u987b\u7ecf\u8fc7L2 Cache\uff0c\u4f46\u662f\u5426\u7ecf\u8fc7L1 Cache\u662f\u53ef\u4ee5\u6539\u53d8\u7684\uff08\u9ed8\u8ba4\u4e0d\u4f1a\u7ecf\u8fc7L1 Cache\uff09<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">L1 Cache \u7684\u67e5\u8be2\u548c\u8bbe\u7f6e<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>GPU\u662f\u5426\u652f\u6301L1\u7f13\u5b58\u67e5\u8be2\u6307\u4ee4\uff1acudaDeviceProp:globalL1CacheSupported<\/li>\n\n\n\n<li>\u9ed8\u8ba4\u60c5\u51b5\uff0c\u6570\u636e\u4e0d\u4f1a\u7f13\u5b58\u5728\u7edf\u4e00\u7684L1\/Texture Cache\u4e2d\uff0c\u4f46\u53ef\u4ee5\u901a\u8fc7\u7f16\u8bd1\u6307\u4ee4\u542f\u7528\u7f13\u5b58\uff1a<\/li>\n\n\n\n<li>\u5f00\u542f\uff1a -Xptxas -dlcm=ca\uff0c\u9664\u4e86\u5e26\u6709\u7981\u7528\u7f13\u5b58\u4fee\u9970\u7b26\u7684\u5185\u8054\u6c47\u7f16\u4fee\u9970\u7684\u6570\u636e\u5916\uff0c\u6240\u6709\u8bfb\u53d6\u90fd\u88ab\u7f13\u5b58<\/li>\n\n\n\n<li>\u5f00\u542f\uff1a -Xptxas -fscm=ca\uff0c\u6240\u6709\u6570\u636e\u8bfb\u53d6\u90fd\u4f1a\u88ab\u7f13\u5b58<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">L1 Cache and shared memory\uff08\u4ee58.9\u4e3a\u4f8b\uff09<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7edf\u4e00\u6570\u636e\u7f13\u5b58\u5927\u5c0f\u4e3a128KB\uff0c\u5305\u62ecshared memory, texture memory \u548c L1 Cache<\/li>\n\n\n\n<li>shared memory \u4ece\u7edf\u4e00\u7684\u6570\u636e\u7f13\u5b58\u4e2d\u5206\u533a\u51fa\u6765\uff0c\u5e76\u4e14\u53ef\u4ee5\u914d\u7f6e\u4e3a\u5404\u79cd\u5927\u5c0f\u3002\u53ef\u4ee5\u8bbe\u5b9a\u4e3a0\uff0c8\uff0c16\uff0c32\uff0c64\u548c100KB\u3002\uff08\u56de\u5fc6\u4e00\u4e0b\uff0c\u4e00\u4e2aSM\u5bf9\u5e94\u7684\u6700\u5927shared memory\u521a\u597d\u662f100KB\uff0c\u4e00\u4e2a\u7ebf\u7a0b\u5757\u6700\u5927\u80fd\u5206\u523099KB\uff09\u3002\u8fd9\u4e4b\u540e\u5269\u4e0b\u7684\u90e8\u5206\u7528\u4f5cL1 Cache\uff0c\u4e5f\u53ef\u4ee5\u7531\u7eb9\u7406\u5355\u5143\u4f7f\u7528\u3002<\/li>\n\n\n\n<li>\u5171\u4eab\u5185\u5b58\u7684\u5927\u5c0f\u53ef\u4ee5\u8bbe\u7f6e\uff0c\u4f46\u4e0d\u4e00\u5b9a\u751f\u6548\uff0cGPU\u6709\u81ea\u52a8\u9009\u62e9\u6700\u4f18\u914d\u7f6e\u7684\u7b56\u7565\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">L2 Cache\u7684\u4e00\u4e9b\u6709\u8da3\u7684\u6570\u636e\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>RTX 3060 \u53ea\u67093MB\u7684 L2 Cache(sm_86)<\/li>\n\n\n\n<li>RTX 4070 \u5374\u670936MB\u7684 L2 Cache(sm_89), \u6269\u5c55\u4e86\u4e00\u4e2a\u6570\u91cf\u7ea7\uff0c\u51cf\u5c11\u663e\u5b58\u5e26\u5bbd\u538b\u529b\uff0c\u63d0\u9ad8\u80fd\u6548<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u8ba1\u7b97\u8d44\u6e90\u5206\u914d<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u7ebf\u7a0b\u675f\u672c\u5730\u6267\u884c\u4e0a\u4e0b\u6587\u4e3b\u8981\u8d44\u6e90\u6784\u6210\uff1a\uff08\u56de\u5fc6\uff0c\u7ebf\u7a0b\u675f\u662f\u771f\u6b63\u5e76\u884c\u7684\uff0c32\u4e2a\u4e3a\u4e00\u675f\uff0c\u7ebf\u7a0b\u675f\u7ec4\u6210\u7ebf\u7a0b\u5757\uff0c\u56e0\u6b64\u7ebf\u7a0b\u5757\u6700\u597d\u5305\u542b32\u7684\u6574\u6570\u500d\u7ebf\u7a0b\uff09<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>1. \u7a0b\u5e8f\u8ba1\u6570\u5668<\/li>\n\n\n\n<li>2. \u5bc4\u5b58\u5668<\/li>\n\n\n\n<li>3. \u5171\u4eab\u5185\u5b58\uff08\u8fd9\u662f\u7ebf\u7a0b\u5757\u7ea7\u522b\u7684\u6982\u5ff5\uff0c\u4e5f\u662fSM\u7ea7\u522b\u7684\u6982\u5ff5\uff0c\u548cL1 Cache\u7269\u7406\u4e0a\u4e00\u8d77\uff0c128KB\u4e2d\u6700\u591a100KB\u53ef\u5206\u914d\uff09<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u7531\u4e8eSM\u5904\u7406\u7684\u6bcf\u4e2a\u7ebf\u7a0b\u675fwarp\u8ba1\u7b97\u6240\u9700\u7684\u8d44\u6e90\u90fd\u662fon-chip\u7684\uff08\u5f88\u663e\u7136\uff0cregister\uff0cshared\u90fd\u5f88\u5feb\uff09\uff0c\u56e0\u6b64\u4ece\u4e00\u4e2a\u6267\u884c\u4e0a\u4e0b\u6587\u5207\u6362\u5230\u53e6\u4e00\u4e2a\u6267\u884c\u4e0a\u4e0b\u6587\u662f\u6ca1\u6709\u65f6\u95f4\u635f\u8017\u7684\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5bf9\u4e8e\u4e00\u4e2a\u7ed9\u5b9a\u7684kernel\uff0c\u540c\u65f6\u5b58\u5728\u4e8e\u540c\u4e00\u4e2aSM\u4e2d\u7684block\u548cwarp\u53d6\u51b3\u4e8e\u5728SM\u4e2dregister\u548cshared memory\u7684\u5206\u914d\uff0c\u5355\u4e2athread\u6d88\u8017\u7684\u8d8a\u591a\uff0c\u53ef\u4ee5\u5b58\u7684\u5c31\u8d8a\u5c11<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5bf9\u4e8esm_86\uff0c\u6bcf\u4e2aSM\u670964KB<strong>\u4e2a<\/strong>register\uff0c\u56e0\u6b64\u5982\u679c\u4e00\u4e2athread\u6d88\u80171kb\uff0c\u90a3\u6700\u591a\u53ea\u80fd\u8dd12\u4e2awarp<\/li>\n\n\n\n<li>\u5f53\u7136\uff0c\u4e00\u4e2athread\u6700\u591a\u53ea\u80fd\u7531255\u4e2aregister\uff08\u6bcf\u4e2a32\u4f4d\uff09\uff0c\u6240\u4ee5\u8fd9\u91cc\u5c31\u662f\u4e3e\u4e2a\u4f8b\u5b50<\/li>\n\n\n\n<li>\u53ef\u4ee5\u60f3\u8c61\uff0c\u6700\u6781\u7aef\u7684\u60c5\u51b5\uff0c\u6700\u4f4e\u8dd18\u4e2awarp<\/li>\n\n\n\n<li>\u540c\u6837\u7684\uff0c\u5bf9\u4e8esm_86\uff0c\u6bcf\u4e2aSM\u81f3\u591a\u6709100KB<strong>\u5927\u5c0f<\/strong>\u7684shared memory\u3002\u533a\u522b\u4e8eregister\u662fthread\u7ea7\u522b\u7684\u6982\u5ff5\uff0cshared\u662fblock\u7ea7\u522b\u7684\u6982\u5ff5\uff0c\u4f46\u9053\u7406\u5dee\u4e0d\u591a<\/li>\n<\/ul>\n\n\n\n<pre class=\"wp-block-code\"><code>========== GPU Device Info ==========\nDevice name: NVIDIA GeForce RTX 3060 Laptop GPU\nCompute capability: 8.6\nSM (multiprocessor) count: 30\n\/\/ \u8fd9\u4e24\u4e2a\u6570\u636e\u5c3d\u53ef\u80fd\u63a5\u8fd1\u6700\u5927\u503c\uff0c\u624d\u80fd\u66f4\u597d\u7684\u5229\u7528\u6027\u80fd\nMax threads per SM: 1536\nMax resident blocks per SM: 16\n\nL2 Cache Size: 3 MB\nGlobal L1 Cache: Supported\nLocal L1 Cache: Supported\nMax shared memory per SM: 100 KB\nMax registers per SM: 64 K (32-bit)\nStream priorities: Supported\nMemory clock rate: 7001.00 MHz\nMemory bus width: 192 bits\nTheoretical memory bandwidth: 336.05 GB\/s<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f53\u8ba1\u7b97\u8d44\u6e90\uff08register\uff0cshared\u8fd9\u4e9b\uff09\u88ab\u5206\u914d\u7ed9block\uff0c\u8fd9\u4e2ablock\u88ab\u79f0\u4e3aactive\u7684\uff0cwarp\u4e5f\u662f\u8fd9\u6837<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">active warp\u6709\u4e09\u79cd\u7c7b\u578b<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u9009\u5b9a\u7684warp\uff08active + \u6b63\u5728\u6267\u884c\uff09<\/li>\n\n\n\n<li>\u963b\u585e\u7684warp  \uff08active + \u6ca1\u6709\u505a\u597d\u6267\u884c\u7684\u51c6\u5907\uff09<\/li>\n\n\n\n<li>\u7b26\u5408\u6761\u4ef6\u7684warp \uff08active + \u51c6\u5907\u597d\u4e86\uff0c\u4f46\u8fd8\u6ca1\u6709\u6267\u884c\uff09<\/li>\n\n\n\n<li>\u51c6\u5907\uff1a32\u4e2acuda\u6838\u5fc3\u53ef\u7528\u4e8e\u6267\u884c + \u6267\u884c\u6240\u9700\u8981\u7684\u53c2\u6570\u5168\u90e8\u51c6\u5907\u597d<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5360\u7528\u7387 = active warp\/max warp\uff081536\/32 = 48\uff09-> active warp\/48<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ee5sm_86\u7684Nt = 1536, Nb = 16\uff0c\u8fdb\u884c\u4e00\u4e9b\u7406\u60f3\u7684\u8ba1\u7b97<\/li>\n\n\n\n<li>\u5047\u8bbe\u5e76\u884c\u89c4\u6a21\u8db3\u591f\u5927\uff08\u603b\u7ebf\u7a0b\u8db3\u591f\u591a\uff09<\/li>\n\n\n\n<li>1. \u5728shared\u548cregister\u7528\u4e0d\u6ee1\u7684\u60c5\u51b5\u4e0b\uff0c\u662f\u9700\u8981\u5355\u4e2ablock\u542b\u6709\u8d85\u8fc796\u4e2a\u7ebf\u7a0b\uff0c\u5c31\u53ef\u4ee5\u6ee1\u5360\u7528<\/li>\n\n\n\n<li>2. register: \u5f53\u771f\u7684\u67091536\u4e2athread\uff0c\u6bcf\u4e2athread\u6700\u591a\u4f7f\u752842\u4e2a\u5bc4\u5b58\u5668<\/li>\n\n\n\n<li>3. shared memory\uff1a \u6ee1\u5360\u7528\u670916\u4e2ablock\uff0c\u6bcf\u4e2ablock\u53ef\u5206\u914d100\/16 = 6.25 KB\u7684shared<\/li>\n\n\n\n<li>\u5047\u8bbe\u6709\u4e00\u4e2ablock\u9700\u8981\u4f7f\u7528\u8d85\u8fc799KB\u7684shared\uff0ckernel\u65e0\u6cd5\u542f\u52a8<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5bf9Grid\u548cBlock\u8bbe\u8ba1\u7684\u51c6\u5219<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>1. block\u8bbe\u8ba1\u4e3a32\u7684\u500d\u6570<\/li>\n\n\n\n<li>2. block\u4e0d\u80fd\u592a\u5c0f\uff0c\u4e0d\u7136\u7531\u4e8eblock\u6570\u91cf\u7684\u9650\u5236\uff0c\u5f88\u96be\u8fbe\u5230\u9ad8\u5360\u7528<\/li>\n\n\n\n<li>3. \u6839\u636ekernel\u8c03\u6574block\u5927\u5c0f\uff08\u5c31\u662f\u8003\u8651register\u548cshared\uff09<\/li>\n\n\n\n<li>4. block\u8981\u8fdc\u591a\u4e8eSM\u7684\u6570\u91cf\uff08\u8fd9\u91cc\u4e3a30\uff09\uff0c\u4e5f\u5373grid\u8981\u8fdc\u5927\u4e8e\u8fd9\u4e2a\u6570\uff0c\u4fdd\u8bc1\u5e76\u884c\u8db3\u591f<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">4. \u5ef6\u8fdf\u9690\u85cf<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6307\u4ee4\u5ef6\u8fdf\uff1a\u6307\u4ee4\u4ece\u53d1\u51fa\u5230\u5b8c\u6210\u4e4b\u95f4\u7684\u65f6\u949f\u5468\u671f<\/li>\n\n\n\n<li>\u5728\u6bcf\u4e2a\u65f6\u949f\u5468\u671f\u4e2d\uff0c\u6240\u6709\u7684warp\u8c03\u5ea6\u5668\u90fd\u6709\u4e00\u4e2a\u7b26\u5408\u6761\u4ef6\u7684warp\u7684\u65f6\u5019\uff0c\u53ef\u4ee5\u8fbe\u5230\u8ba1\u7b97\u8d44\u6e90\u7684\u5b8c\u5168\u5229\u7528<\/li>\n\n\n\n<li>GPU\u7684\u6307\u4ee4\u5ef6\u8fdf\u88ab\u5176\u4ed6warp\u7684\u8ba1\u7b97\u9690\u85cf\uff0c\u88ab\u79f0\u4e3a\u8ba1\u7b97\u9690\u85cf\uff08pipeline\u7684\u601d\u60f3\uff0c\u4e0d\u7a7a\u7b49\uff09<\/li>\n\n\n\n<li>\u6307\u4ee4\u53ef\u4ee5\u5206\u4e3a\u7b97\u6570\u6307\u4ee4\u548c\u5185\u5b58\u6307\u4ee4<\/li>\n\n\n\n<li>\u7b97\u6570\u8fd0\u7b97\u7684\u6307\u4ee4\u5ef6\u8fdf\u662f\u4ece\u5f00\u59cb\u8fd0\u7b97\u5230\u5f97\u5230\u8ba1\u7b97\u7ed3\u679c\u7684\u59cb\u7ec8\u5468\u671f\uff0c\u901a\u5e38\u4e3a4<\/li>\n\n\n\n<li>\u6ee1\u8db3\u5ef6\u8fdf\u9690\u85cf\u7684\u6240\u9700\u7ebf\u7a0b\u675f\u6570\u91cf\uff1a\u6240\u9700warp\u7684\u6570\u91cf = \u541e\u5410\u91cf * \u5ef6\u8fdf<\/li>\n\n\n\n<li>\u5e26\u5bbd\u548c\u541e\u5410\u91cf\u7684\u533a\u522b\uff1a\u5e26\u5bbd\u4e00\u822c\u6307\u7406\u8bba\u503c\uff0c\u541e\u5410\u91cf\u6307\u5b9e\u9645\u8fbe\u5230\u7684\u503c<\/li>\n\n\n\n<li>\u4e00\u4e9b\u4f8b\u5b50\uff1a<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-11-1024x538.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"538\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-11-1024x538.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-770\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-12-1024x508.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"508\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-12-1024x508.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-771\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-13-1024x541.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"541\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-13-1024x541.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-772\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">5. \u7ebf\u7a0b\u675f\u5206\u5316<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u5bf9\u4e8e\u4e00\u4e2awarp\uff0c\u6700\u597d\u8ba9\u91cc\u9762\u7684\u6240\u6709\u7ebf\u7a0b\u5728\u540c\u4e00\u4e2a\u5468\u671f\u4e2d\u6267\u884c\u76f8\u540c\u7684\u6307\u4ee4\uff0c\u5426\u5219\u5c31\u4f1a\u56e0\u4e3a\u5206\u652f\u9884\u6d4b\u673a\u5236\uff0c\u9020\u6210\u7ebf\u7a0b\u675f\u5206\u5316\uff0c\u5bfc\u81f4\u7a0b\u5e8f\u7684\u5e76\u884c\u6027\u80fd\u4e0b\u964d<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u4e2a\u5e38\u89c1\u7684\u89e3\u51b3\u601d\u8def\u662f\uff0c\u4ee532\u4e3a\u57fa\u672c\u5355\u4f4d\u7ec4\u7ec7\u3002\u56e0\u4e3awarp\u5206\u5316\u4ec5\u4ec5\u53d1\u751f\u5728\u4e00\u4e2awarp\u5185\u90e8\uff0c\u56e0\u6b64\u53ea\u8981\u4fdd\u8bc1\u4e00\u4e2awarp\u5185\u90e8\u6267\u884c\u76f8\u540c\u7684\u6307\u4ee4\uff0c\u5c31\u53ef\u4ee5\u907f\u514d\u5206\u5316\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">5.1 <strong>\u5e76\u884c\u89c4\u7ea6\u8ba1\u7b97<\/strong>\uff08\u5206\u6cbb\u6cd5\u7684\u601d\u8def\uff09<\/h4>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-14-1024x454.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"454\" data-original=\"https:\/\/ferina.top\/wp-content\/uploads\/2025\/09\/image-14-1024x454.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-774\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4e00. \u5b89\u88c5 Q\uff1aThe libtinfo5 package isn&#8217;t available A:  [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[28,29],"class_list":["post-619","post","type-post","status-publish","format-standard","hentry","category-8","tag-cuda","tag-gpu"],"_links":{"self":[{"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/posts\/619","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/comments?post=619"}],"version-history":[{"count":142,"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/posts\/619\/revisions"}],"predecessor-version":[{"id":776,"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/posts\/619\/revisions\/776"}],"wp:attachment":[{"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/media?parent=619"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/categories?post=619"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ferina.top\/index.php\/wp-json\/wp\/v2\/tags?post=619"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}