{"id":259589,"date":"2024-05-15T13:10:38","date_gmt":"2024-05-15T05:10:38","guid":{"rendered":"https:\/\/docs.pingcode.com\/ask\/ask-ask\/259589.html"},"modified":"2024-05-15T13:10:47","modified_gmt":"2024-05-15T05:10:47","slug":"%e5%a6%82%e4%bd%95%e7%94%a8clang%e5%b0%86opencl-kernel%e7%bc%96%e8%af%91%e6%88%90ptx%e4%bb%a3%e7%a0%81","status":"publish","type":"post","link":"https:\/\/docs.pingcode.com\/ask\/259589.html","title":{"rendered":"\u5982\u4f55\u7528clang\u5c06opencl kernel\u7f16\u8bd1\u6210ptx\u4ee3\u7801"},"content":{"rendered":"<p style=\"text-align:center\"><img decoding=\"async\" src=\"https:\/\/cdn-kb.worktile.com\/kb\/wp-content\/uploads\/2024\/04\/27111019\/eb9930aa-ed86-439f-9519-182995692da9.webp\" alt=\"\u5982\u4f55\u7528clang\u5c06opencl kernel\u7f16\u8bd1\u6210ptx\u4ee3\u7801\" \/><\/p>\n<p><p>OpenCL\uff08Open Computing Language\uff09\u662f\u4e00\u4e2a\u7528\u4e8e\u7f16\u7a0b\u5f02\u6784\u8ba1\u7b97\u7cfb\u7edf\u7684\u6846\u67b6\uff0c\u800cPTX\uff08Parallel Thread Execution\uff09\u662fNVIDIA GPU\u7684\u4e00\u79cd\u4e2d\u95f4\u4ee3\u7801\uff08Intermediate Representation\uff0cIR\uff09\u3002\u8981\u5c06OpenCL kernel\u7f16\u8bd1\u6210PTX\u4ee3\u7801\uff0c\u60a8\u9700\u8981\u4f7f\u7528\u652f\u6301\u8be5\u529f\u80fd\u7684\u5de5\u5177\u94fe\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528NVIDIA\u7684CUDA\u5de5\u5177\u5305\u4e2d\u7684nvcc\u7f16\u8bd1\u5668\u6216Clang\u7f16\u8bd1\u5668\uff0c\u8fd9\u4e24\u79cd\u7f16\u8bd1\u5668\u90fd\u63d0\u4f9b\u4e86\u6784\u5efaOpenCL\u7a0b\u5e8f\u5e76\u5c06\u5176\u7f16\u8bd1\u6210PTX\u4ee3\u7801\u7684\u80fd\u529b\u3002<strong>\u9996\u5148\u786e\u5b9a\u5b89\u88c5\u4e86NVIDIA CUDA Toolkit\uff0c\u7136\u540e\u4f7f\u7528Clang\u7f16\u8bd1\u5668\uff0c\u914d\u5408\u7279\u5b9a\u7684\u53c2\u6570\u6765\u7f16\u8bd1OpenCL\u6e90\u4ee3\u7801\uff0c\u751f\u6210\u5bf9\u5e94\u7684PTX\u4ee3\u7801\u3002<\/strong><\/p>\n<\/p>\n<p><p><strong>\u4e00\u3001CUDA TOOLKIT\u4e0eCLANG\u7684\u5b89\u88c5\u4e0e\u914d\u7f6e<\/strong><\/p>\n<\/p>\n<p><p>\u9996\u5148\uff0c\u8bf7\u786e\u4fdd\u60a8\u7684\u7cfb\u7edf\u4e2d\u5b89\u88c5\u4e86NVIDIA CUDA Toolkit\uff0c\u5b83\u63d0\u4f9b\u4e86\u7f16\u8bd1OpenCL\u7a0b\u5e8f\u6240\u5fc5\u9700\u7684\u5934\u6587\u4ef6\u548c\u5e93\u6587\u4ef6\u3002\u540c\u65f6\uff0c\u8fd8\u9700\u8981\u5b89\u88c5\u652f\u6301OpenCL\u7684Clang\u7248\u672c\u3002\u5728\u5b89\u88c5\u65f6\uff0c\u5e94\u5f53\u6ce8\u610f\u68c0\u67e5Clang\u7684\u7248\u672c\u662f\u5426\u652f\u6301OpenCL\u4ee5\u53caCUDA\u3002<\/p>\n<\/p>\n<p><p>\u5b89\u88c5\u5b8c\u6210\u540e\uff0c\u60a8\u9700\u8981\u914d\u7f6e\u73af\u5883\u53d8\u91cf\uff0c\u786e\u4fdd\u5728\u547d\u4ee4\u884c\u4e2d\u80fd\u591f\u8bbf\u95ee\u5230Clang\u7f16\u8bd1\u5668\u548cCUDA Toolkit\u5305\u542b\u7684\u5404\u79cd\u5de5\u5177\u3002\u73af\u5883\u53d8\u91cf\u901a\u5e38\u5305\u62ec<code>PATH<\/code>\u3001<code>LD_LIBRARY_PATH<\/code>\u7b49\uff0c\u7528\u4e8e\u6307\u5b9a\u53ef\u6267\u884c\u6587\u4ef6\u548c\u5e93\u6587\u4ef6\u7684\u641c\u7d22\u8def\u5f84\u3002<\/p>\n<\/p>\n<p><p><strong>\u4e8c\u3001\u7406\u89e3OPENCL\u4e0ePTX\u7684\u5173\u7cfb<\/strong><\/p>\n<\/p>\n<p><p>OpenCL\u7a0b\u5e8f\u901a\u5e38\u5305\u542b\u4e00\u4e2a\u6216\u591a\u4e2a\u5728GPU\u4e0a\u8fd0\u884c\u7684\u8ba1\u7b97\u6838\u5fc3\uff08kernel\uff09\uff0c\u4ee5\u53ca\u7528\u4e8e\u7ba1\u7406\u8fd9\u4e9b\u6838\u5fc3\u6267\u884c\u73af\u5883\u7684\u5bbf\u4e3b\u7a0b\u5e8f\u4ee3\u7801\u3002\u800cPTX\u5219\u662f\u8fd0\u884c\u5728NVIDIA GPU\u4e0a\u7684\u4f4e\u7ea7\u4e2d\u95f4\u4ee3\u7801\uff0c\u53ef\u4ee5\u88abCUDA\u9a71\u52a8\u7f16\u8bd1\u6210\u786c\u4ef6\u53ef\u4ee5\u76f4\u63a5\u6267\u884c\u7684\u4e8c\u8fdb\u5236\u4ee3\u7801\u3002<\/p>\n<\/p>\n<p><p>OpenCL\u548cPTX\u7684\u5173\u7cfb\u5728\u4e8e\uff0c\u901a\u8fc7\u7f16\u8bd1\uff0c\u4e00\u4e2aOpenCL kernel\u53ef\u4ee5\u8f6c\u6362\u6210PTX\u4ee3\u7801\uff0c\u7136\u540e\u88abNVIDIA GPU\u6267\u884c\u3002PTX\u4f5c\u4e3a\u4e00\u4e2a\u4e2d\u95f4\u5c42\uff0c\u5b83\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u4e86\u76f4\u63a5\u5f71\u54cdGPU\u5982\u4f55\u6267\u884c\u4ee3\u7801\u7684\u80fd\u529b\uff0c\u4e5f\u4f7f\u5f97\u8de8\u67b6\u6784\u7684\u7f16\u7a0b\u6a21\u578b\u80fd\u591f\u8fd0\u884c\u5728NVIDIA\u7684GPU\u67b6\u6784\u4e0a\u3002<\/p>\n<\/p>\n<p><p><strong>\u4e09\u3001\u51c6\u5907OPENCL KERNEL\u6e90\u4ee3\u7801<\/strong><\/p>\n<\/p>\n<p><p>\u5728\u5c06OpenCL kernel\u7f16\u8bd1\u6210PTX\u4ee3\u7801\u4e4b\u524d\uff0c\u60a8\u9700\u8981\u51c6\u5907\u597dOpenCL kernel\u7684\u6e90\u4ee3\u7801\u3002OpenCL\u6e90\u4ee3\u7801\u6587\u4ef6\u901a\u5e38\u5e26\u6709<code>.cl<\/code>\u6269\u5c55\u540d\uff0c\u5e76\u5305\u542b\u4e86\u7528\u4e8e\u5728GPU\u4e0a\u6267\u884c\u7684\u51fd\u6570\uff08\u5373\u8ba1\u7b97\u6838\u5fc3\uff09\u3002<\/p>\n<\/p>\n<p><p>\u786e\u4fddOpenCL kernel\u4ee3\u7801\u6b63\u786e\u65e0\u8bef\uff0c\u5e76\u4e14\u7b26\u5408NVIDIA GPU\u652f\u6301\u7684OpenCL\u7248\u672c\u89c4\u8303\u3002\u56e0\u4e3a\u4e0d\u540c\u7248\u672c\u7684OpenCL\u53ef\u80fd\u5b58\u5728\u8bed\u6cd5\u548c\u7279\u6027\u4e0a\u7684\u5dee\u5f02\uff0c\u4e0d\u540cGPU\u67b6\u6784\u4e5f\u6709\u53ef\u80fd\u9650\u5b9a\u4e86\u6240\u652f\u6301\u7684OpenCL\u529f\u80fd\u3002<\/p>\n<\/p>\n<p><p><strong>\u56db\u3001\u4f7f\u7528CLANG\u7f16\u8bd1OPENCL KERNEL\u6e90\u4ee3\u7801<\/strong><\/p>\n<\/p>\n<p><p>\u5728\u7f16\u8bd1OpenCL kernel\u65f6\uff0c\u4f7f\u7528Clang\u5e76\u642d\u914d\u9002\u5f53\u7684\u547d\u4ee4\u884c\u53c2\u6570\u3002\u8fd9\u4e9b\u53c2\u6570\u5c06\u544a\u8bc9Clang\u5c06OpenCL\u6e90\u4ee3\u7801\u7f16\u8bd1\u6210PTX\u4e2d\u95f4\u4ee3\u7801\u3002\u4e0b\u9762\u7684\u547d\u4ee4\u63d0\u4f9b\u4e86\u4e00\u4e2a\u57fa\u672c\u793a\u4f8b\uff1a<\/p>\n<\/p>\n<p><pre><code class=\"language-sh\">clang -x cl -cl-std=CL1.2 -target nvptx64-nvidia-cuda --cuda-gpu-arch=sm_35 -S -o kernel.ptx kernel.cl<\/p>\n<p><\/code><\/pre>\n<\/p>\n<p><p>\u8fd9\u6761\u547d\u4ee4\u6307\u5b9a\u4e86Clang\u5e94\u5f53\u4ee5OpenCL 1.2\u6807\u51c6\u6765\u5904\u7406\u7ed9\u5b9a\u7684<code>kernel.cl<\/code>\u6587\u4ef6\uff0c\u5e76\u5c06\u76ee\u6807\u8bbe\u4e3aNVIDIA GPU\u7684PTX\uff0c\u540c\u65f6\u6307\u5b9a\u4e86GPU\u67b6\u6784\u3002<\/p>\n<\/p>\n<p><p>\u6267\u884c\u8be5\u547d\u4ee4\u540e\uff0c\u82e5\u6ca1\u6709\u8bef\uff0c\u60a8\u5c06\u5f97\u5230\u4e00\u4e2a<code>kernel.ptx<\/code>\u6587\u4ef6\uff0c\u8fd9\u5c31\u662f\u8f6c\u6362\u540e\u7684PTX\u4ee3\u7801\u3002<\/p>\n<\/p>\n<p><p><strong>\u4e94\u3001\u7406\u89e3\u7f16\u8bd1\u547d\u4ee4\u53ca\u53c2\u6570<\/strong><\/p>\n<\/p>\n<p><p>\u8ba9\u6211\u4eec\u6df1\u5165\u7406\u89e3\u4e0a\u8ff0\u547d\u4ee4\u4e2d\u7684\u6bcf\u4e2a\u53c2\u6570\u7684\u610f\u4e49\uff1a<\/p>\n<\/p>\n<ul>\n<li><strong>-x cl<\/strong>\uff1a\u6307\u793aClang\u7f16\u8bd1\u5668\u6587\u4ef6\u7c7b\u578b\u4e3aOpenCL\u3002<\/li>\n<li><strong>-cl-std=CL1.2<\/strong>\uff1a\u6307\u5b9aOpenCL\u7684\u7248\u672c\u6807\u51c6\uff0c\u4f8b\u5982CL1.2\u8868\u793a\u9075\u5faaOpenCL 1.2\u6807\u51c6\u3002<\/li>\n<li><strong>-target<\/strong>\uff1a\u6307\u5b9a\u76ee\u6807\u5e73\u53f0\uff0c\u8fd9\u91cc\u662fNVIDIA GPU\u7684PTX\u3002<\/li>\n<li><strong>nvptx64-nvidia-cuda<\/strong>\uff1a\u662f\u76ee\u6807\u4e09\u91cd\u7ec4\u5408\uff08triple\uff09\uff0c\u6307\u5b9a\u4e86\u67b6\u6784\u548cCUDA\u540e\u7aef\u3002<\/li>\n<li><strong>&#8211;cuda-gpu-arch=sm_35<\/strong>\uff1a\u6307\u5b9a\u4e86NVIDIA GPU\u7684\u67b6\u6784\u7248\u672c\uff0c<code>sm_35<\/code>\u662f\u4ee3\u8868\u7279\u5b9a\u7684\u8ba1\u7b97\u80fd\u529b\u3002<\/li>\n<li><strong>-S<\/strong>\uff1a\u544a\u8bc9Clang\u8f93\u51fa\u6c47\u7f16\u4ee3\u7801\uff0c\u5373PTX\u4ee3\u7801\u3002<\/li>\n<li><strong>-o kernel.ptx<\/strong>\uff1a\u6307\u5b9a\u4e86\u8f93\u51fa\u6587\u4ef6\u540d\u3002<\/li>\n<\/ul>\n<p><p><strong>\u516d\u3001PTX\u4ee3\u7801\u7684\u540e\u7eed\u4f7f\u7528<\/strong><\/p>\n<\/p>\n<p><p>\u7f16\u8bd1\u5f97\u5230\u7684PTX\u4ee3\u7801\u662f\u4e0d\u80fd\u76f4\u63a5\u5728GPU\u4e0a\u6267\u884c\u7684\uff0c\u9700\u8981\u8fdb\u4e00\u6b65\u901a\u8fc7NVIDIA\u7684CUDA\u9a71\u52a8\u7f16\u8bd1\u6210\u53ef\u6267\u884c\u7684\u4e8c\u8fdb\u5236\u4ee3\u7801\u3002\u901a\u5e38\uff0c\u5f00\u53d1\u8005\u4f1a\u5c06PTX\u4ee3\u7801\u4e0e\u5bbf\u4e3b\u7a0b\u5e8f\u7ed3\u5408\uff0c\u901a\u8fc7CUDA\u8fd0\u884c\u65f6API\u6216\u8005OpenCL API\u6765\u52a8\u6001\u7f16\u8bd1\u548c\u6267\u884c\u8fd9\u4e9b\u4ee3\u7801\u3002<\/p>\n<\/p>\n<p><p>\u5982\u679c\u60a8\u6b63\u5728\u5f00\u53d1CUDA\u7a0b\u5e8f\uff0c\u53ef\u4ee5\u4f7f\u7528CUDA\u9a71\u52a8API\u6765\u76f4\u63a5\u52a0\u8f7dPTX\u4ee3\u7801\u3002\u5982\u679c\u60a8\u5de5\u4f5c\u5728OpenCL\u6846\u67b6\u4e2d\uff0c\u5f80\u5f80\u4e0d\u9700\u8981\u76f4\u63a5\u5904\u7406PTX\u4ee3\u7801\uff0c\u56e0\u4e3aOpenCL\u7684\u5b9e\u73b0\u901a\u5e38\u4f1a\u628a\u8fd9\u4e00\u6b65\u9690\u85cf\u8d77\u6765\uff0c\u81ea\u52a8\u4e3a\u60a8\u5b8c\u6210\u3002\u4e0d\u8fc7\uff0c\u6709\u65f6\u5019\u786e\u4fdd\u60a8\u7684OpenCL kernel\u80fd\u591f\u7f16\u8bd1\u6210PTX\u4ee3\u7801\u662f\u6709\u76ca\u7684\uff0c\u8fd9\u610f\u5473\u7740\u60a8\u7684kernel\u66f4\u6709\u53ef\u80fd\u6210\u529f\u6267\u884c\u5728NVIDIA\u7684GPU\u4e0a\u3002<\/p>\n<\/p>\n<p><p><strong>\u4e03\u3001\u53ef\u80fd\u7684\u95ee\u9898\u4e0e\u89e3\u51b3<\/strong><\/p>\n<\/p>\n<p><p>\u5f53\u60a8\u5c1d\u8bd5\u5c06OpenCL kernel\u7f16\u8bd1\u6210PTX\u4ee3\u7801\u65f6\uff0c\u53ef\u80fd\u4f1a\u9047\u5230\u5404\u79cd\u95ee\u9898\uff0c\u6bd4\u5982\u7f16\u8bd1\u9519\u8bef\u3001\u517c\u5bb9\u6027\u95ee\u9898\u6216\u6027\u80fd\u4e0d\u4f73\u3002\u9762\u5bf9\u9519\u8bef\u4fe1\u606f\uff0c\u68c0\u67e5\u6e90\u4ee3\u7801\u4ee5\u907f\u514d\u8bed\u6cd5\u6216API\u4f7f\u7528\u4e0a\u7684\u95ee\u9898\uff1b\u517c\u5bb9\u6027\u95ee\u9898\u53ef\u80fd\u6765\u81ea\u4e8eGPU\u67b6\u6784\u4e0eOpenCL\u7248\u672c\u4e4b\u95f4\u7684\u4e0d\u5339\u914d\uff0c\u9700\u8981\u4ed4\u7ec6\u9009\u62e9\u6b63\u786e\u7684\u76ee\u6807\u67b6\u6784\u548cOpenCL\u6807\u51c6\uff1b\u6027\u80fd\u95ee\u9898\u53ef\u80fd\u6765\u81ea\u4e8e\u4ee3\u7801\u672c\u8eab\u4e0d\u9002\u5408GPU\u67b6\u6784\u6216\u8005\u9700\u8981\u8fdb\u4e00\u6b65\u7684\u4f18\u5316\u3002<\/p>\n<\/p>\n<p><p><strong>\u516b\u3001\u603b\u7ed3<\/strong><\/p>\n<\/p>\n<p><p>\u5c06OpenCL kernel\u7f16\u8bd1\u6210PTX\u4ee3\u7801\u662f\u4e00\u79cd\u63d0\u9ad8\u4ee3\u7801\u517c\u5bb9\u6027\u548c\u6027\u80fd\u7684\u65b9\u6cd5\uff0c\u5c24\u5176\u662f\u5728NVIDIA GPU\u4e0a\u6267\u884c\u65f6\u3002\u901a\u8fc7\u672c\u6587\u7684\u8be6\u5c3d\u8bf4\u660e\uff0c\u60a8\u5e94\u8be5\u80fd\u7406\u89e3\u4e0e\u6267\u884c\u7f16\u8bd1\u8fc7\u7a0b\uff0c\u5e76\u4e3a\u53ef\u80fd\u9047\u5230\u7684\u6311\u6218\u505a\u597d\u51c6\u5907\u3002<\/p>\n<\/p>\n<h2><strong>\u76f8\u5173\u95ee\u7b54FAQs\uff1a<\/strong><\/h2>\n<p><strong>1. \u5982\u4f55\u4f7f\u7528clang\u5c06OpenCL Kernel\u7f16\u8bd1\u6210PTX\u4ee3\u7801\uff1f<\/strong><\/p>\n<p>\u8981\u4f7f\u7528Clang\u5c06OpenCL Kernel\u7f16\u8bd1\u6210PTX\u4ee3\u7801\uff0c\u60a8\u53ef\u4ee5\u6309\u7167\u4ee5\u4e0b\u6b65\u9aa4\u8fdb\u884c\u64cd\u4f5c\uff1a<\/p>\n<p>\u6b65\u9aa41: \u5b89\u88c5CUDA Toolkit<br \/>\n\u9996\u5148\uff0c\u786e\u4fdd\u60a8\u5df2\u5b89\u88c5NVIDIA CUDA Toolkit\u3002\u8fd9\u4e2a\u5de5\u5177\u5305\u5305\u542b\u4e86\u4f7f\u7528clang\u5c06OpenCL\u5185\u6838\u7f16\u8bd1\u6210PTX\u4ee3\u7801\u6240\u9700\u7684\u7ec4\u4ef6\u3002<\/p>\n<p>\u6b65\u9aa42\uff1a\u6dfb\u52a0OpenCL\u7f16\u8bd1\u5668\u6807\u5fd7<br \/>\n\u60a8\u9700\u8981\u4f7f\u7528\u4e00\u7ec4\u7279\u5b9a\u7684\u6807\u5fd7\u6765\u544a\u8bc9clang\u5c06OpenCL\u5185\u6838\u7f16\u8bd1\u6210PTX\u4ee3\u7801\u3002\u5728\u7f16\u8bd1\u547d\u4ee4\u4e2d\u6dfb\u52a0\u4ee5\u4e0b\u6807\u5fd7\uff1a<\/p>\n<p><code>clang -Xclang -target -Xclang nvptx64-nvidia-cuda -include opencl-c.h -cl-std=CL2.0 -S -emit-llvm -o output.ptx input.cl<\/code><\/p>\n<p>\u5176\u4e2d\uff0c<code>-Xclang -target -Xclang nvptx64-nvidia-cuda<\/code>\u6807\u5fd7\u544a\u8bc9clang\u4f7f\u7528NVIDIA PTX\u4f5c\u4e3a\u76ee\u6807\u8bbe\u5907\u3002<br \/>\n<code>-include opencl-c.h<\/code>\u6807\u5fd7\u5305\u542bOpenCL\u5934\u6587\u4ef6\uff0c\u4ee5\u786e\u4fdd\u5728\u7f16\u8bd1\u65f6\u53ef\u4ee5\u4f7f\u7528OpenCL\u51fd\u6570\u548c\u6570\u636e\u7c7b\u578b\u3002<br \/>\n<code>-cl-std=CL2.0<\/code>\u6807\u5fd7\u6307\u5b9aOpenCL\u7248\u672c\u4e3a2.0\u3002<br \/>\n<code>-S<\/code>\u6807\u5fd7\u544a\u8bc9clang\u5c06\u8f93\u5165\u6587\u4ef6\u7f16\u8bd1\u4e3a\u6c47\u7f16\u8bed\u8a00\u3002<br \/>\n<code>-emit-llvm<\/code>\u6807\u5fd7\u5c06\u8f93\u51faPTX\u4ee3\u7801\u8f6c\u6362\u6210LLVM IR\u3002<br \/>\n<code>-o output.ptx<\/code>\u6807\u5fd7\u6307\u5b9a\u8f93\u51fa\u6587\u4ef6\u7684\u540d\u79f0\u3002<br \/>\n<code>input.cl<\/code>\u662f\u5305\u542bOpenCL\u5185\u6838\u4ee3\u7801\u7684\u8f93\u5165\u6587\u4ef6\u3002<\/p>\n<p>\u6b65\u9aa43: \u7f16\u8bd1OpenCL\u5185\u6838<br \/>\n\u8fd0\u884c\u4e0a\u8ff0\u547d\u4ee4\u540e\uff0cclang\u5c06\u4f1a\u4f7f\u7528\u6307\u5b9a\u7684\u53c2\u6570\u5c06OpenCL\u5185\u6838\u7f16\u8bd1\u6210PTX\u4ee3\u7801\uff0c\u5e76\u5c06\u7ed3\u679c\u4fdd\u5b58\u5728output.ptx\u6587\u4ef6\u4e2d\u3002<\/p>\n<p><strong>2. \u5982\u4f55\u4f7f\u7528clang\u7f16\u8bd1\u5668\u5c06OpenCL\u5185\u6838\u8f6c\u6362\u6210PTX\u4ee3\u7801\uff1f<\/strong><\/p>\n<p>\u8981\u5c06OpenCL\u5185\u6838\u4ee3\u7801\u8f6c\u6362\u6210PTX\u4ee3\u7801\uff0c\u53ef\u4ee5\u6309\u7167\u4ee5\u4e0b\u6b65\u9aa4\u64cd\u4f5c\uff1a<\/p>\n<p>\u6b65\u9aa41\uff1a\u5b89\u88c5Clang\u7f16\u8bd1\u5668\u548cOpenCL\u8fd0\u884c\u65f6<br \/>\n\u9996\u5148\uff0c\u786e\u4fdd\u60a8\u5df2\u7ecf\u5b89\u88c5\u4e86Clang\u7f16\u8bd1\u5668\u548c\u76f8\u5e94\u7684OpenCL\u8fd0\u884c\u65f6\u3002\u8fd9\u4e9b\u8f6f\u4ef6\u5305\u63d0\u4f9b\u4e86\u5c06OpenCL\u4ee3\u7801\u8f6c\u6362\u4e3a\u4e0d\u540c\u5e73\u53f0\u7684\u8bbe\u5907\u4ee3\u7801\u7684\u5de5\u5177\u548c\u5e93\u3002<\/p>\n<p>\u6b65\u9aa42\uff1a\u7f16\u5199OpenCL\u5185\u6838\u4ee3\u7801<br \/>\n\u7f16\u5199\u60a8\u7684OpenCL\u5185\u6838\u4ee3\u7801\u6587\u4ef6\uff0c\u4f7f\u7528\u6269\u5c55\u540d\u4e3a<code>.cl<\/code>\u3002\u60a8\u53ef\u4ee5\u5728\u6b64\u6587\u4ef6\u4e2d\u5b9a\u4e49\u8981\u5728GPU\u4e0a\u5e76\u884c\u6267\u884c\u7684\u4efb\u52a1\u3002<\/p>\n<p>\u6b65\u9aa43\uff1a\u4f7f\u7528Clang\u7f16\u8bd1OpenCL\u5185\u6838\u4ee3\u7801<br \/>\n\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u5c06OpenCL\u5185\u6838\u4ee3\u7801\u8f6c\u6362\u4e3aPTX\u4ee3\u7801\uff1a<\/p>\n<p><code>clang -x cl -target x86_64-pc-linux-gnu -o output.ptx input.cl<\/code><\/p>\n<p>\u5176\u4e2d\uff1a<br \/>\n<code>-x cl<\/code>\u6307\u5b9a\u8f93\u5165\u6587\u4ef6\u7684\u7c7b\u578b\u4e3aOpenCL\u4ee3\u7801\u3002<br \/>\n<code>-target x86_64-pc-linux-gnu<\/code>\u6307\u5b9a\u76ee\u6807\u5e73\u53f0\u4e3ax86_64\u67b6\u6784\u7684Linux\u3002<\/p>\n<p>\u6b65\u9aa44\uff1a\u68c0\u67e5\u8f93\u51fa\u6587\u4ef6<br \/>\n\u8fd0\u884c\u4e0a\u8ff0\u547d\u4ee4\u540e\uff0cClang\u5c06\u751f\u6210\u4e00\u4e2a\u540d\u4e3a<code>output.ptx<\/code>\u7684\u6587\u4ef6\uff0c\u5176\u4e2d\u5305\u542b\u4e86\u5c06OpenCL\u5185\u6838\u4ee3\u7801\u8f6c\u6362\u4e3aPTX\u4ee3\u7801\u7684\u7ed3\u679c\u3002<\/p>\n<p><strong>3. \u4f7f\u7528clang\u5c06\u81ea\u5b9a\u4e49OpenCL\u5185\u6838\u7f16\u8bd1\u4e3aPTX\u4ee3\u7801\u7684\u6b65\u9aa4\u662f\u4ec0\u4e48\uff1f<\/strong><\/p>\n<p>\u8981\u4f7f\u7528Clang\u5c06\u81ea\u5b9a\u4e49OpenCL\u5185\u6838\u7f16\u8bd1\u6210PTX\u4ee3\u7801\uff0c\u53ef\u4ee5\u9075\u5faa\u4ee5\u4e0b\u6b65\u9aa4\uff1a<\/p>\n<p>\u6b65\u9aa41\uff1a\u521b\u5efa\u81ea\u5b9a\u4e49OpenCL\u5185\u6838\u4ee3\u7801<br \/>\n\u9996\u5148\uff0c\u521b\u5efa\u4e00\u4e2aOpenCL\u5185\u6838\u4ee3\u7801\u6587\u4ef6\uff0c\u60a8\u53ef\u4ee5\u5728\u5176\u4e2d\u5b9a\u4e49\u81ea\u5df1\u7684\u5185\u6838\u51fd\u6570\u3001\u5de5\u4f5c\u7ec4\u548c\u53d8\u91cf\u3002<\/p>\n<p>\u6b65\u9aa42\uff1a\u5b89\u88c5Clang\u7f16\u8bd1\u5668\u548cOpenCL\u8fd0\u884c\u65f6<br \/>\n\u786e\u4fdd\u60a8\u5df2\u7ecf\u5b89\u88c5\u4e86Clang\u7f16\u8bd1\u5668\u4ee5\u53ca\u4e0e\u60a8\u7684\u76ee\u6807\u5e73\u53f0\u517c\u5bb9\u7684OpenCL\u8fd0\u884c\u65f6\u3002<\/p>\n<p>\u6b65\u9aa43\uff1a\u4f7f\u7528Clang\u5c06OpenCL\u5185\u6838\u4ee3\u7801\u7f16\u8bd1\u4e3aPTX\u4ee3\u7801<br \/>\n\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u5c06\u81ea\u5b9a\u4e49OpenCL\u5185\u6838\u4ee3\u7801\u8f6c\u6362\u4e3aPTX\u4ee3\u7801\uff1a<\/p>\n<p><code>clang -x cl -target ptx -o output.ptx input.cl<\/code><\/p>\n<p>\u5176\u4e2d\uff1a<br \/>\n<code>-x cl<\/code>\u6307\u5b9a\u8f93\u5165\u6587\u4ef6\u7684\u7c7b\u578b\u4e3aOpenCL\u4ee3\u7801\u3002<br \/>\n<code>-target ptx<\/code>\u6307\u5b9a\u8981\u5c06\u4ee3\u7801\u7f16\u8bd1\u4e3aPTX\u4ee3\u7801\u3002<\/p>\n<p>\u6b65\u9aa44\uff1a\u68c0\u67e5\u751f\u6210\u7684PTX\u4ee3\u7801<br \/>\n\u8fd0\u884c\u4e0a\u8ff0\u547d\u4ee4\u540e\uff0cClang\u5c06\u751f\u6210\u4e00\u4e2a\u540d\u4e3a<code>output.ptx<\/code>\u7684\u6587\u4ef6\uff0c\u5176\u4e2d\u5305\u542b\u5c06\u81ea\u5b9a\u4e49OpenCL\u5185\u6838\u4ee3\u7801\u8f6c\u6362\u4e3aPTX\u4ee3\u7801\u7684\u7ed3\u679c\u3002\u60a8\u53ef\u4ee5\u68c0\u67e5\u8be5\u6587\u4ef6\u4ee5\u67e5\u770b\u751f\u6210\u7684PTX\u4ee3\u7801\u3002<\/p>\n<p>\u8bf7\u6ce8\u610f\uff0c\u8fd9\u53ea\u662f\u4e00\u4e2a\u7b80\u5355\u793a\u4f8b\uff0c\u5e76\u4e14\u5177\u4f53\u7684\u547d\u4ee4\u53ef\u80fd\u56e0\u60a8\u7684\u7f16\u8bd1\u5668\u548c\u8fd0\u884c\u65f6\u7248\u672c\u800c\u5f02\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"OpenCL\uff08Open Computing Language\uff09\u662f\u4e00\u4e2a\u7528\u4e8e\u7f16\u7a0b\u5f02\u6784\u8ba1\u7b97\u7cfb\u7edf\u7684\u6846\u67b6\uff0c\u800cPTX\uff08Pa [&hellip;]","protected":false},"author":3,"featured_media":259605,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[37],"tags":[],"acf":[],"_links":{"self":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/259589"}],"collection":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/comments?post=259589"}],"version-history":[{"count":0,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/259589\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media\/259605"}],"wp:attachment":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media?parent=259589"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/categories?post=259589"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/tags?post=259589"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}