187 lines
23 KiB
XML
187 lines
23 KiB
XML
<?xml version='1.0' encoding='UTF-8' standalone='no'?>
|
|
<doxygen xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:noNamespaceSchemaLocation="compound.xsd" version="1.9.1" xml:lang="en-US">
|
|
<compounddef id="matrix_multiplication_cudaflow" kind="page">
|
|
<compoundname>matrix_multiplication_cudaflow</compoundname>
|
|
<title>Matrix Multiplication (cudaFlow)</title>
|
|
<tableofcontents>
|
|
<tocsect>
|
|
<name>Define a Matrix Multiplication Kernel</name>
|
|
<reference>matrix_multiplication_cudaflow_1GPUAcceleratedMatrixMultiplication</reference>
|
|
</tocsect>
|
|
<tocsect>
|
|
<name>Define a cudaFlow for Matrix Multiplication</name>
|
|
<reference>matrix_multiplication_cudaflow_1DefineAcudaFlowForMatrixMultiplication</reference>
|
|
</tocsect>
|
|
<tocsect>
|
|
<name>Benchmarking</name>
|
|
<reference>matrix_multiplication_cudaflow_1MatrixMultiplicationcudaFlowBenchmarking</reference>
|
|
</tocsect>
|
|
</tableofcontents>
|
|
<briefdescription>
|
|
</briefdescription>
|
|
<detaileddescription>
|
|
<para>Following up on <ref refid="matrix_multiplication" kindref="compound">Matrix Multiplication</ref>, this page studies how to accelerate a matrix multiplication workload on a GPU using <ref refid="classtf_1_1cudaFlow" kindref="compound">tf::cudaFlow</ref>.</para>
|
|
<sect1 id="matrix_multiplication_cudaflow_1GPUAcceleratedMatrixMultiplication">
|
|
<title>Define a Matrix Multiplication Kernel</title>
|
|
<para>GPU can perform a lot of parallel computations more than CPUs. It is especially useful for data-intensive computing such as matrix multiplication. With GPU, we express the parallel patterns at a fine-grained level. The kernel, written in CUDA, is described as follows:</para>
|
|
<para><programlisting filename=".cpp"><codeline><highlight class="comment">//<sp/>CUDA<sp/>kernel<sp/>to<sp/>perform<sp/>matrix<sp/>multiplication</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal">__global__<sp/></highlight><highlight class="keywordtype">void</highlight><highlight class="normal"><sp/>matmul(</highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>*A,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>*B,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>*C,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>M,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>K,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>N)<sp/>{</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>row<sp/>=<sp/>blockIdx.y<sp/>*<sp/>blockDim.y<sp/>+<sp/>threadIdx.y;</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>col<sp/>=<sp/>blockIdx.x<sp/>*<sp/>blockDim.x<sp/>+<sp/>threadIdx.x;</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>sum<sp/>=<sp/>0;</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="keywordflow">if</highlight><highlight class="normal">(col<sp/><<sp/>N<sp/>&&<sp/>row<sp/><<sp/>M)<sp/>{</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/></highlight><highlight class="keywordflow">for</highlight><highlight class="normal">(</highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>i<sp/>=<sp/>0;<sp/>i<sp/><<sp/>K;<sp/>i++)<sp/>{</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><sp/><sp/>sum<sp/>+=<sp/>a[row<sp/>*<sp/>K<sp/>+<sp/>i]<sp/>*<sp/>b[i<sp/>*<sp/>N<sp/>+<sp/>col];</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>}</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>c[row<sp/>*<sp/>N<sp/>+<sp/>col]<sp/>=<sp/>sum;</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>}</highlight></codeline>
|
|
<codeline><highlight class="normal">}</highlight></codeline>
|
|
</programlisting></para>
|
|
<para>Each CUDA thread corresponds to an element of <computeroutput>C</computeroutput> and compute its result. Instead of storing each matrix in a 2D array, we use 1D layout to ease the data transfer between CPU and GPU. In a row-major layout, an element <computeroutput>(x, y)</computeroutput> in the 2D matrix can be addressed at <computeroutput>x * width + y</computeroutput> in the transformed 1D layout.</para>
|
|
<para><image type="html" name="matrix_multiplication_4.png" width="70%"></image>
|
|
</para>
|
|
</sect1>
|
|
<sect1 id="matrix_multiplication_cudaflow_1DefineAcudaFlowForMatrixMultiplication">
|
|
<title>Define a cudaFlow for Matrix Multiplication</title>
|
|
<para>The next step is to allocate memory for <computeroutput>A</computeroutput>, <computeroutput>B</computeroutput>, and <computeroutput>C</computeroutput> at a GPU. We create three tasks each calling <computeroutput>cudaMalloc</computeroutput> to allocate space for one matrix. Then, we create a cudaFlow to offload matrix multiplication to a GPU. The entire code is described as follows:</para>
|
|
<para><programlisting filename=".cpp"><codeline><highlight class="keywordtype">void</highlight><highlight class="normal"><sp/>matrix_multiplication(</highlight><highlight class="keywordtype">int</highlight><highlight class="normal">*<sp/>A,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal">*<sp/>B,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal">*<sp/>C,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>M,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>K,<sp/></highlight><highlight class="keywordtype">int</highlight><highlight class="normal"><sp/>N)<sp/>{</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><ref refid="classtf_1_1Taskflow" kindref="compound">tf::Taskflow</ref><sp/>taskflow;</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><ref refid="classtf_1_1Executor" kindref="compound">tf::Executor</ref><sp/>executor;</highlight></codeline>
|
|
<codeline><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>allocate<sp/>the<sp/>host<sp/>and<sp/>gpu<sp/>storage<sp/>for<sp/>A</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><ref refid="classtf_1_1Task" kindref="compound">tf::Task</ref><sp/>allocate_a<sp/>=<sp/>taskflow.<ref refid="classtf_1_1FlowBuilder_1a60d7a666cab71ecfa3010b2efb0d6b57" kindref="member">emplace</ref>([&](){</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>cudaMalloc(&da,<sp/>M*K*</highlight><highlight class="keyword">sizeof</highlight><highlight class="normal">(</highlight><highlight class="keywordtype">int</highlight><highlight class="normal">));</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>}).name(</highlight><highlight class="stringliteral">"allocate_a"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>allocate<sp/>the<sp/>host<sp/>and<sp/>gpu<sp/>storage<sp/>for<sp/>B</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><ref refid="classtf_1_1Task" kindref="compound">tf::Task</ref><sp/>allocate_b<sp/>=<sp/>taskflow.<ref refid="classtf_1_1FlowBuilder_1a60d7a666cab71ecfa3010b2efb0d6b57" kindref="member">emplace</ref>([&](){</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>cudaMalloc(&db,<sp/>K*N*</highlight><highlight class="keyword">sizeof</highlight><highlight class="normal">(</highlight><highlight class="keywordtype">int</highlight><highlight class="normal">));</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>}).name(</highlight><highlight class="stringliteral">"allocate_b"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>allocate<sp/>the<sp/>host<sp/>and<sp/>gpu<sp/>storage<sp/>for<sp/>C</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><ref refid="classtf_1_1Task" kindref="compound">tf::Task</ref><sp/>allocate_c<sp/>=<sp/>taskflow.<ref refid="classtf_1_1FlowBuilder_1a60d7a666cab71ecfa3010b2efb0d6b57" kindref="member">emplace</ref>([&](){</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>cudaMalloc(&dc,<sp/>M*N*</highlight><highlight class="keyword">sizeof</highlight><highlight class="normal">(</highlight><highlight class="keywordtype">int</highlight><highlight class="normal">));</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>}).name(</highlight><highlight class="stringliteral">"allocate_c"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>create<sp/>a<sp/>cudaFlow<sp/>task<sp/>to<sp/>run<sp/>the<sp/>matrix<sp/>multiplication</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><ref refid="classtf_1_1Task" kindref="compound">tf::Task</ref><sp/>cudaFlow<sp/>=<sp/>taskflow.<ref refid="classtf_1_1FlowBuilder_1a60d7a666cab71ecfa3010b2efb0d6b57" kindref="member">emplace</ref>([&](){</highlight></codeline>
|
|
<codeline><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><ref refid="classtf_1_1cudaFlow" kindref="compound">tf::cudaFlow</ref><sp/>cf;</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/></highlight><highlight class="comment">//<sp/>copy<sp/>data<sp/>to<sp/>da,<sp/>db,<sp/>and<sp/>dc</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><ref refid="classtf_1_1cudaTask" kindref="compound">tf::cudaTask</ref><sp/>copy_da<sp/>=<sp/>cf.<ref refid="classtf_1_1cudaFlow_1af03e04771b655f9e629eb4c22e19b19f" kindref="member">copy</ref>(da,<sp/>A,<sp/>M*K).<ref refid="classtf_1_1cudaTask_1ab81b4f71a44af8d61758524f0c274962" kindref="member">name</ref>(</highlight><highlight class="stringliteral">"H2D_A"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><ref refid="classtf_1_1cudaTask" kindref="compound">tf::cudaTask</ref><sp/>copy_db<sp/>=<sp/>cf.<ref refid="classtf_1_1cudaFlow_1af03e04771b655f9e629eb4c22e19b19f" kindref="member">copy</ref>(db,<sp/>B,<sp/>K*N).<ref refid="classtf_1_1cudaTask_1ab81b4f71a44af8d61758524f0c274962" kindref="member">name</ref>(</highlight><highlight class="stringliteral">"H2D_B"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><ref refid="classtf_1_1cudaTask" kindref="compound">tf::cudaTask</ref><sp/>copy_hc<sp/>=<sp/>cf.<ref refid="classtf_1_1cudaFlow_1af03e04771b655f9e629eb4c22e19b19f" kindref="member">copy</ref>(C,<sp/>dc,<sp/>M*N).<ref refid="classtf_1_1cudaTask_1ab81b4f71a44af8d61758524f0c274962" kindref="member">name</ref>(</highlight><highlight class="stringliteral">"D2H_C"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>dim3<sp/>grid<sp/><sp/>((K+16-1)/16,<sp/>(M+16-1)/16);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>dim3<sp/>block<sp/>(16,<sp/>16);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><ref refid="classtf_1_1cudaTask" kindref="compound">tf::cudaTask</ref><sp/>kmatmul<sp/>=<sp/>cf.<ref refid="classtf_1_1cudaFlow_1a68f666503d13a7b80fb7399fb2f0c153" kindref="member">kernel</ref>(grid,<sp/>block,<sp/>0,<sp/>matmul,<sp/>da,<sp/>db,<sp/>dc,<sp/>M,<sp/>K,<sp/>N)</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/>.<ref refid="classtf_1_1cudaTask_1ab81b4f71a44af8d61758524f0c274962" kindref="member">name</ref>(</highlight><highlight class="stringliteral">"matmul"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>kmatmul.<ref refid="classtf_1_1cudaTask_1a4a9ca1a34bac47e4c9b04eb4fb2f7775" kindref="member">succeed</ref>(copy_da,<sp/>copy_db)</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/>.<ref refid="classtf_1_1cudaTask_1abdd68287ec4dff4216af34d1db44d1b4" kindref="member">precede</ref>(copy_hc);</highlight></codeline>
|
|
<codeline><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/></highlight><highlight class="comment">//<sp/>launch<sp/>the<sp/>cudaFlow</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><ref refid="classtf_1_1cudaStream" kindref="compound">tf::cudaStream</ref><sp/>stream;</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>cf.<ref refid="classtf_1_1cudaFlow_1ae6810f7de27e5a347331aacfce67bea1" kindref="member">run</ref>(stream);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>stream.<ref refid="classtf_1_1cudaStream_1a1a81d6005e8d60ad082dba2303a8aa30" kindref="member">synchronize</ref>();</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>}).name(</highlight><highlight class="stringliteral">"cudaFlow"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>free<sp/>the<sp/>gpu<sp/>storage</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="keyword">auto</highlight><highlight class="normal"><sp/><ref refid="cpp/memory/c/free" kindref="compound" external="/home/thuang295/Code/taskflow/doxygen/cppreference-doxygen-web.tag.xml">free</ref><sp/>=<sp/>taskflow.<ref refid="classtf_1_1FlowBuilder_1a60d7a666cab71ecfa3010b2efb0d6b57" kindref="member">emplace</ref>([&](){</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>cudaFree(da);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>cudaFree(db);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/>cudaFree(dc);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>}).name(</highlight><highlight class="stringliteral">"free"</highlight><highlight class="normal">);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>create<sp/>dependency</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>cudaFlow.<ref refid="classtf_1_1Task_1a331b1b726555072e7c7d10941257f664" kindref="member">succeed</ref>(allocate_a,<sp/>allocate_b,<sp/>allocate_c)</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/><sp/>.<ref refid="classtf_1_1Task_1a8c78c453295a553c1c016e4062da8588" kindref="member">precede</ref>(free);</highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>dump<sp/>the<sp/>graph<sp/>without<sp/>unfolding<sp/>the<sp/>cudaFlow</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>taskflow.<ref refid="classtf_1_1Taskflow_1ac433018262e44b12c4cc9f0c4748d758" kindref="member">dump</ref>(<ref refid="cpp/io/basic_ostream" kindref="compound" external="/home/thuang295/Code/taskflow/doxygen/cppreference-doxygen-web.tag.xml">std::cout</ref>);</highlight></codeline>
|
|
<codeline><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>run<sp/>the<sp/>taskflow</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>executor.<ref refid="classtf_1_1Executor_1a8d08f0cb79e7b3780087975d13368a96" kindref="member">run</ref>(taskflow).wait();</highlight></codeline>
|
|
<codeline><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/></highlight><highlight class="comment">//<sp/>dump<sp/>the<sp/>entire<sp/>execution<sp/>graph<sp/>including<sp/>unfolded<sp/>cudaFlow</highlight><highlight class="normal"></highlight></codeline>
|
|
<codeline><highlight class="normal"><sp/><sp/>taskflow.<ref refid="classtf_1_1Taskflow_1ac433018262e44b12c4cc9f0c4748d758" kindref="member">dump</ref>(<ref refid="cpp/io/basic_ostream" kindref="compound" external="/home/thuang295/Code/taskflow/doxygen/cppreference-doxygen-web.tag.xml">std::cout</ref>);</highlight></codeline>
|
|
<codeline><highlight class="normal">}</highlight></codeline>
|
|
</programlisting></para>
|
|
<para>Within the cudaFlow, we create two host-to-device (H2D) tasks that copy data from <computeroutput>A</computeroutput> and <computeroutput>B</computeroutput> to <computeroutput>da</computeroutput> and <computeroutput>db</computeroutput>, one device-to-host (D2H) task that copies the result from <computeroutput>dc</computeroutput> to <computeroutput>C</computeroutput>, and one kernel task that launches <computeroutput>matmul</computeroutput> on the GPU (by default, GPU 0). H2D tasks precede the kernel and the kernel precedes the D2H task. These GPU operations form a GPU task graph managed by a cudaFlow. The first dump of the taskflow gives the following graph:</para>
|
|
<para><dotfile name="/home/thuang295/Code/taskflow/doxygen/images/matrix_multiplication_5.dot"></dotfile>
|
|
</para>
|
|
<para>A cudaFlow encapsulates a GPU task dependency graph similar to a <ref refid="classtf_1_1Subflow" kindref="compound">tf::Subflow</ref> (see <ref refid="SubflowTasking" kindref="compound">Subflow Tasking</ref>). In order to visualize it, we need to execute the graph first and then dump the taskflow.</para>
|
|
<para><dotfile name="/home/thuang295/Code/taskflow/doxygen/images/matrix_multiplication_6.dot"></dotfile>
|
|
</para>
|
|
</sect1>
|
|
<sect1 id="matrix_multiplication_cudaflow_1MatrixMultiplicationcudaFlowBenchmarking">
|
|
<title>Benchmarking</title>
|
|
<para>We run three versions of matrix multiplication, sequential CPU, parallel CPUs, and one GPU, on a machine of 12 Intel i7-8700 CPUs at 3.20 GHz and a Nvidia RTX 2080 GPU using various matrix sizes of <computeroutput>A</computeroutput>, <computeroutput>B</computeroutput>, and <computeroutput>C</computeroutput>.</para>
|
|
<para> <table rows="7" cols="6"><row>
|
|
<entry thead="yes" align='center'><para>A </para>
|
|
</entry><entry thead="yes" align='center'><para>B </para>
|
|
</entry><entry thead="yes" align='center'><para>C </para>
|
|
</entry><entry thead="yes" align='center'><para>CPU Sequential </para>
|
|
</entry><entry thead="yes" align='center'><para>CPU Parallel </para>
|
|
</entry><entry thead="yes" align='center'><para>GPU Parallel </para>
|
|
</entry></row>
|
|
<row>
|
|
<entry thead="no" align='center'><para>10x10 </para>
|
|
</entry><entry thead="no" align='center'><para>10x10 </para>
|
|
</entry><entry thead="no" align='center'><para>10x10 </para>
|
|
</entry><entry thead="no" align='center'><para>0.142 ms </para>
|
|
</entry><entry thead="no" align='center'><para>0.414 ms </para>
|
|
</entry><entry thead="no" align='center'><para>82 ms </para>
|
|
</entry></row>
|
|
<row>
|
|
<entry thead="no" align='center'><para>100x100 </para>
|
|
</entry><entry thead="no" align='center'><para>100x100 </para>
|
|
</entry><entry thead="no" align='center'><para>100x100 </para>
|
|
</entry><entry thead="no" align='center'><para>1.641 ms </para>
|
|
</entry><entry thead="no" align='center'><para>0.733 ms </para>
|
|
</entry><entry thead="no" align='center'><para>83 ms </para>
|
|
</entry></row>
|
|
<row>
|
|
<entry thead="no" align='center'><para>1000x1000 </para>
|
|
</entry><entry thead="no" align='center'><para>1000x1000 </para>
|
|
</entry><entry thead="no" align='center'><para>1000x1000 </para>
|
|
</entry><entry thead="no" align='center'><para>1532 ms </para>
|
|
</entry><entry thead="no" align='center'><para>504 ms </para>
|
|
</entry><entry thead="no" align='center'><para>85 ms </para>
|
|
</entry></row>
|
|
<row>
|
|
<entry thead="no" align='center'><para>2000x2000 </para>
|
|
</entry><entry thead="no" align='center'><para>2000x2000 </para>
|
|
</entry><entry thead="no" align='center'><para>2000x2000 </para>
|
|
</entry><entry thead="no" align='center'><para>25688 ms </para>
|
|
</entry><entry thead="no" align='center'><para>4387 ms </para>
|
|
</entry><entry thead="no" align='center'><para>133 ms </para>
|
|
</entry></row>
|
|
<row>
|
|
<entry thead="no" align='center'><para>3000x3000 </para>
|
|
</entry><entry thead="no" align='center'><para>3000x3000 </para>
|
|
</entry><entry thead="no" align='center'><para>3000x3000 </para>
|
|
</entry><entry thead="no" align='center'><para>104838 ms </para>
|
|
</entry><entry thead="no" align='center'><para>16170 ms </para>
|
|
</entry><entry thead="no" align='center'><para>214 ms </para>
|
|
</entry></row>
|
|
<row>
|
|
<entry thead="no" align='center'><para>4000x4000 </para>
|
|
</entry><entry thead="no" align='center'><para>4000x4000 </para>
|
|
</entry><entry thead="no" align='center'><para>4000x4000 </para>
|
|
</entry><entry thead="no" align='center'><para>250133 ms </para>
|
|
</entry><entry thead="no" align='center'><para>39646 ms </para>
|
|
</entry><entry thead="no" align='center'><para>427 ms </para>
|
|
</entry></row>
|
|
</table>
|
|
</para>
|
|
<para>As the matrix size increases, the speed-up of GPU over CPUs becomes prominent. For example, at <computeroutput>4000x4000</computeroutput>, the GPU runtime is 585.8 times faster than the sequential CPU runtime and is 92.8 times faster than the parallel CPU solutions. </para>
|
|
</sect1>
|
|
</detaileddescription>
|
|
<location file="doxygen/examples/matrix_multiplication_cudaflow.dox"/>
|
|
</compounddef>
|
|
</doxygen>
|