imported>ExultantEditor 来自 2025年12月26日 (五) 16:36

2025-12-26T16:36:55Z

新页面

{{noteTA
|G1 = IT
}}

{{distinguish|OpenGL}}

{{Infobox software
| name = OpenCL API
| title = OpenCL API
| logo = OpenCL logo.svg
| logo caption =
| logo size = 200px
| logo alt = OpenCL logo
| screenshot = 
| caption =
| screenshot size =
| screenshot alt =
| collapsible =
| author = [[苹果公司]]
| developer = [[科纳斯组织]]
| released = {{Start date and age|2009|08|28}}
| discontinued =
| latest release version = {{wikidata|property|preferred|references|edit|P348|P548=Q2804309}}
| latest release date = {{wikidata|qualifier|preferred|single|P348|P548=Q2804309|P577}}
| latest preview version =
| latest preview date =
| programming language = [[C语言|C]]，具有[[C++]]绑定
| operating system = [[Android]]（厂商依赖）<ref>{{cite web |title=Android Devices With OpenCL support |url=https://docs.google.com/a/arrayfire.com/spreadsheets/d/1Mpzfl2NmLUVSAjIph77-FOsJeuyD9Xjha89r5iHw1hI/edit?pli=1#gid=0 |website=Google Docs |publisher=ArrayFire |access-date=April 28, 2015 |archive-date=2021-02-25 |archive-url=https://web.archive.org/web/20210225193922/https://docs.google.com/a/arrayfire.com/spreadsheets/d/1Mpzfl2NmLUVSAjIph77-FOsJeuyD9Xjha89r5iHw1hI/edit?pli=1#gid=0 |dead-url=no }}</ref>、[[FreeBSD]]<ref>{{cite web |title=FreeBSD Graphics/OpenCL |url=https://wiki.freebsd.org/Graphics/OpenCL |publisher=FreeBSD |access-date=December 23, 2015 |archive-date=2021-02-08 |archive-url=https://web.archive.org/web/20210208150957/https://wiki.freebsd.org/Graphics/OpenCL |dead-url=no }}</ref>、[[Linux]]、[[macOS]]（通过PoCL）、[[Windows]]
| platform = [[ARMv7]]、[[ARMv8]]<ref name="conformant-products">{{cite web |title=Conformant Products |url=https://www.khronos.org/conformance/adopters/conformant-products/opencl |publisher=Khronos Group |access-date=May 9, 2015 |archive-date=2018-06-29 |archive-url=https://web.archive.org/web/20180629130825/https://www.khronos.org/conformance/adopters/conformant-products/opencl |dead-url=no }}</ref>、[[Cell (微處理器)|Cell]]、[[IA-32]]、 [[IBM POWER微处理器|Power]]、[[x86-64]]
| size =
| language =
| language count = 
| language footnote =
| genre = [[异构计算]][[API]]
| license = OpenCL规范许可证
| alexa =
| website = {{URL|https://www.khronos.org/opencl/}}
| standard =
}}
{{Infobox programming language
| name = OpenCL C和C++ for OpenCL
| logo =
| logo caption =
| paradigm = [[指令式编程|指令式]]（[[过程式编程|过程式]]）、[[结构化编程|结构化]]、（仅C++）[[面向对象编程|面向对象]]、[[泛型编程|泛型]]
| family = {{le|C家族编程语言列表|List of C-family programming languages|C}}
| designer =
| developer =
| released =
| latest release version =
| latest release date =
| latest preview version =
| latest preview date =
| typing = [[类型系统|静态]]、[[弱类型]]、{{le|明示类型|Manifest typing|明示}}、[[名義型別系統|名义]]
| scope =
| programming language = 特定于实现
| platform =
| operating_system =
| license =
| file_ext = .cl .clcpp
| file format =
| website =
| implementations = [[AMD]]、[[Mesa 3D|Gallium]] Compute、[[IBM]]、[[Intel]] NEO、Intel SDK、[[Texas Instruments]]、[[Nvidia]]、PoCL、[[ARM架構|ARM]]
| dialects =
| influenced_by = [[C99]]、[[CUDA]]、[[C++14]]、[[C++17]]
| influenced =
}}

'''OpenCL'''（{{langx|en|Open Computing Language}}，中譯：'''开放计算语言'''），是一个为[[异构计算|异构平台]]编写程序的[[軟體框架|框架]]，此异构平台可由[[CPU]]、[[GPU]]、[[數位訊號處理器|DSP]]、[[FPGA]]或其他类型的处理器與硬體加速器所组成。OpenCL由一门用于编写kernels（在OpenCL设备上运行的函数）的语言（基于[[C99]]）和一组用于定义并控制平台的API组成。OpenCL提供了基于任务分割和数据分割的[[并行计算]]机制。

OpenCL类似于另外两个开放的工业标准[[OpenGL]]和[[OpenAL]]，这两个标准分别用于三维图形和计算机音频方面。OpenCL擴充了GPU圖形生成之外的能力。OpenCL由非盈利性技术组织[[Khronos Group]]掌管。

== 历史 ==
OpenCL最初由[[苹果公司]]开发，拥有其商标权，并在与[[AMD]]，[[IBM]]，[[Intel]]和[[NVIDIA]]技术团队的合作之下初步完善。随后，苹果将这一草案提交至[[Khronos Group]]。2008年6月16日，Khronos的通用计算工作小组成立<ref>{{cite press release |url=http://www.khronos.org/news/press/releases/khronos_launches_heterogeneous_computing_initiative/ |title=Khronos Launches Heterogeneous Computing Initiative |accessdate=2008-06-18 |publisher=Khronos Group |date=2008-06-16 |deadurl=yes |archiveurl=https://web.archive.org/web/20080620123431/http://www.khronos.org/news/press/releases/khronos_launches_heterogeneous_computing_initiative/ |archivedate=2008-06-20 }}</ref>。5个月后的2008年11月18日，该工作组完成了OpenCL 1.0规范的技术细节<ref name=macWorld>{{cite web | url=http://www.macworld.com/article/136921/2008/11/opencl.html?lsrc=top_2 | title=OpenCL gets touted in Texas | publisher=MacWorld | date=2008-11-20 | accessdate=2009-06-12 | archive-date=2009-02-18 | archive-url=https://web.archive.org/web/20090218165557/http://www.macworld.com/article/136921/2008/11/opencl.html?lsrc=top_2 | dead-url=no }}</ref>。该技术规范在由Khronos成员进行审查之后，于2008年12月8日公开发表<ref name=khronosGroup>{{cite press release | url=http://www.khronos.org/news/press/releases/the_khronos_group_releases_opencl_1.0_specification/ | title=The Khronos Group Releases OpenCL 1.0 Specification | publisher=Khronos Group | date=2008-12-08 | accessdate=2009-06-12 | deadurl=yes | archiveurl=https://web.archive.org/web/20100713014204/http://www.khronos.org/news/press/releases/the_khronos_group_releases_opencl_1.0_specification/ | archivedate=2010-07-13 }}</ref>。

2010年6月14日，OpenCL 1.1发布<ref>{{cite press release | url=http://www.khronos.org/news/press/releases/khronos-group-releases-opencl-1-1-parallel-computing-standard/ | title=Khronos Drives Momentum of Parallel Computing Standard with Release of OpenCL 1.1 Specification | publisher=Khronos Group | date=2010-06-14 | accessdate=2010-10-13 | deadurl=yes | archiveurl=https://web.archive.org/web/20100923101844/http://www.khronos.org/news/press/releases/khronos-group-releases-opencl-1-1-parallel-computing-standard | archivedate=2010-09-23 }}</ref>。2011年11月15日，OpenCL 1.2发布<ref>{{cite web |url=https://www.khronos.org/news/press/khronos-releases-opencl-1.2-specification |title=Khronos Releases OpenCL 1.2 Specification |publisher=Khronos Group |date=November 15, 2011 |access-date=June 23, 2015 |archive-date=2020-12-02 |archive-url=https://web.archive.org/web/20201202103240/https://www.khronos.org/news/press/khronos-releases-opencl-1.2-specification |dead-url=no }}</ref>。2013年11月18日，OpenCL 2.0发布<ref>{{cite web |url=https://www.khronos.org/news/press/khronos-finalizes-opencl-2.0-specification-for-heterogeneous-computing |title=Khronos Finalizes OpenCL 2.0 Specification for Heterogeneous Computing |date=November 18, 2013 |access-date=February 10, 2014 |publisher=Khronos Group |archive-date=2020-11-11 |archive-url=https://web.archive.org/web/20201111201527/https://www.khronos.org/news/press/khronos-finalizes-opencl-2.0-specification-for-heterogeneous-computing |dead-url=no }}</ref>。2015年11月16日，OpenCL 2.1发布<ref>{{cite web |title=Khronos Releases OpenCL 2.1 and SPIR-V 1.0 Specifications for Heterogeneous Parallel Programming |url=https://www.khronos.org/news/press/khronos-releases-opencl-2.1-and-spir-v-1.0-specifications-for-heterogeneous |publisher=Khronos Group |access-date=November 16, 2015 |date=November 16, 2015 |archive-date=2020-11-08 |archive-url=https://web.archive.org/web/20201108122543/https://www.khronos.org/news/press/khronos-releases-opencl-2.1-and-spir-v-1.0-specifications-for-heterogeneous |dead-url=no }}</ref>。2017年5月16日，OpenCL 2.2发布<ref>{{cite web |url=https://www.khronos.org/news/press/khronos-releases-opencl-2.2-with-spir-v-1.2 |title=Khronos Releases OpenCL 2.2 With SPIR-V 1.2 |date=May 16, 2017 |publisher=[[Khronos Group]] |access-date=2024-10-13 |archive-date=2021-01-18 |archive-url=https://web.archive.org/web/20210118110308/https://www.khronos.org/news/press/khronos-releases-opencl-2.2-with-spir-v-1.2 |dead-url=no }}</ref>。

=== 路线图 ===
在2017年5月发行OpenCL 2.2之时，Khronos Group宣布OpenCL将尽可能的汇合于[[Vulkan]]，以确使OpenCL软件在这两种API上灵活部署<ref>{{Cite web|url=https://www.pcper.com/reviews/General-Tech/Breaking-OpenCL-Merging-Roadmap-Vulkan|title=Breaking: OpenCL Merging Roadmap into Vulkan | PC Perspective|website=www.pcper.com|access-date=May 17, 2017|archive-url=https://web.archive.org/web/20171101062642/https://www.pcper.com/reviews/General-Tech/Breaking-OpenCL-Merging-Roadmap-Vulkan|archive-date=November 1, 2017|url-status=dead}}</ref>。这已经由Adobe的Premiere Rush展示出来，它使用clspv开源编译器<ref name=":4">{{Citation|title=Clspv is a compiler for OpenCL C to Vulkan compute shaders|date=2019-08-17|url=https://github.com/google/clspv|access-date=2024-10-14|archive-date=2020-11-28|archive-url=https://web.archive.org/web/20201128054700/https://github.com/google/clspv|dead-url=no}}</ref>，编译了大量OpenCL C内核代码，使其在部署于Android的Vulkan运行时系统上运行<ref>{{Cite web|url=https://www.khronos.org/assets/uploads/developers/library/2019-siggraph/Vulkan-01-Update-SIGGRAPH-Jul19.pdf|title=Vulkan Update SIGGRAPH 2019|access-date=2024-10-13|archive-date=2019-08-20|archive-url=https://web.archive.org/web/20190820222456/https://www.khronos.org/assets/uploads/developers/library/2019-siggraph/Vulkan-01-Update-SIGGRAPH-Jul19.pdf|dead-url=no}}</ref>。

OpenCL拥有独立于Vulkan的前瞻性路线图<ref>{{Cite web|url=https://www.phoronix.com/scan.php?page=article&item=siggraph-2018-khr&num=2|title=SIGGRAPH 2018: OpenCL-Next Taking Shape, Vulkan Continues Evolving – Phoronix|website=www.phoronix.com|access-date=2024-10-13|archive-date=2020-10-28|archive-url=https://web.archive.org/web/20201028152042/https://www.phoronix.com/scan.php?page=article&item=siggraph-2018-khr&num=2|dead-url=no}}</ref>，即曾意图在2020年发行的“OpenCL Next”<ref>{{Cite web|url=https://www.khronos.org/assets/uploads/developers/library/2019-embedded-vision-summit/1b%20Khronos-and-OpenCL-Overview-EVS-Workshop_May19.pdf|title=Khronos and OpenCL Overview EVS Workshop May19|last=Trevett|first=Neil|date=May 23, 2019|website=Khronos Group|access-date=2024-10-13|archive-date=2020-07-17|archive-url=https://web.archive.org/web/20200717035519/https://www.khronos.org/assets/uploads/developers/library/2019-embedded-vision-summit/1b%20Khronos-and-OpenCL-Overview-EVS-Workshop_May19.pdf|dead-url=no}}</ref>，它可以集成于扩展诸如Vulkan/OpenCL互操作、Scratch-Pad内存管理、扩展子组、SPIR-V 1.4摄入和SPIR-V扩展调试信息；OpenCL还在考虑类似Vulkan的装载器和分层以及“灵活配置”，以便在多种加速类型上灵活部署。

=== OpenCL 3.0 ===
在2020年8月30日，发行了最终的OpenCL 3.0规范<ref>{{Cite web|url = https://www.khronos.org/blog/opencl-3.0-specification-finalized-and-initial-khronos-open-source-opencl-sdk-released|title = OpenCL 3.0 Specification Finalized and Initial Khronos Open Source OpenCL SDK Released|date = September 30, 2020|access-date = 2024-10-14|archive-date = 2020-09-30|archive-url = https://web.archive.org/web/20200930143119/https://www.khronos.org/blog/opencl-3.0-specification-finalized-and-initial-khronos-open-source-opencl-sdk-released|dead-url = no}}</ref>。OpenCL 1.2功能已经成为强制性基准，而所有OpenCL 2.x和OpenCL 3.0特征变为可选项<ref>{{Cite web|url=https://www.phoronix.com/review/opencl-30-spec|title=OpenCL 3.0 Bringing Greater Flexibility, Async DMA Extensions|website=www.phoronix.com|access-date=2024-10-13|archive-date=2024-05-06|archive-url=https://web.archive.org/web/20240506194730/https://www.phoronix.com/review/opencl-30-spec|dead-url=no}}</ref>。这个规范保留了“OpenCL C”语言<ref>{{cite web |last1=Munshi |first1=Aaftab |last2=Howes |first2=Lee |last3=Sochaki |first3=Barosz |title=The OpenCL C Specification Version: 3.0 Document Revision: V3.0.7 |url=https://www.khronos.org/registry/OpenCL/specs/3.0-unified/pdf/OpenCL_C.pdf |publisher=Khronos OpenCL Working Group |access-date=Apr 28, 2021 |date=Apr 27, 2020 |archive-date=September 20, 2020 |archive-url=https://web.archive.org/web/20200920173143/https://www.khronos.org/registry/OpenCL/specs/3.0-unified/pdf/OpenCL_C.pdf |url-status=dead }}</ref>，并废弃了版本2.1介入的“OpenCL C++”内核语言<ref>{{cite web|last1=Sochacki|first1=Bartosz|title=The OpenCL C++ 1.0 Specification|url=https://www.khronos.org/registry/OpenCL/specs/opencl-2.2-cplusplus.pdf|publisher=Khronos OpenCL Working Group|access-date=Jul 19, 2019|date=Jul 19, 2019|archive-date=2021-03-06|archive-url=https://web.archive.org/web/20210306212145/https://www.khronos.org/registry/OpenCL/specs/2.2/pdf/OpenCL_Cxx.pdf|dead-url=no}}</ref>，将其替代为“C++ for OpenCL”语言<ref name=":0">{{Cite web|title=C++ for OpenCL, OpenCL-Guide|url=https://github.com/KhronosGroup/OpenCL-Guide/blob/main/chapters/cpp_for_opencl.md|access-date=2021-04-18|website=GitHub|language=en|archive-date=2025-01-08|archive-url=https://web.archive.org/web/20250108213646/https://github.com/KhronosGroup/OpenCL-Guide/blob/main/chapters/cpp_for_opencl.md|dead-url=no}}</ref>，它基于了[[Clang]]/[[LLVM]]编译器，实现了[[C++17]]的子集和{{le|标准可移植中间表示|Standard Portable Intermediate Representation|SPIR-V}}中间代码。C++ for OpenCL版本1.0的官方文档在2020年12月发表<ref>{{Cite web|date=December 2020|title=Release of Documentation of C++ for OpenCL kernel language, version 1.0, revision 1 · KhronosGroup/OpenCL-Docs|url=https://github.com/KhronosGroup/OpenCL-Docs/releases/tag/cxxforopencl-v1.0-r1|access-date=2021-04-18|website=GitHub|language=en|archive-date=2024-12-16|archive-url=https://web.archive.org/web/20241216064559/https://github.com/KhronosGroup/OpenCL-Docs/releases/tag/cxxforopencl-v1.0-r1|dead-url=no}}</ref>，它后向兼容于OpenCL C 2.0。

在{{le|IWOCL}} 21上发布的OpenCL 3.0.7，提出了C++ for OpenCL的新版本和一些Khronos openCL扩展<ref>{{Cite web |last=Trevett |first=Neil |date=2021 |title=State of the Union: OpenCL Working Group |url=https://www.iwocl.org/wp-content/uploads/k03-iwocl-syclcon-2021-trevett-updated.mp4.pdf |page=9 |access-date=2024-10-13 |archive-date=2024-07-16 |archive-url=https://web.archive.org/web/20240716180733/https://www.iwocl.org/wp-content/uploads/k03-iwocl-syclcon-2021-trevett-updated.mp4.pdf |dead-url=no }}</ref>。在2021年12月，发行了C++ for OpenCL版本2021<ref>{{cite web |title=The C++ for OpenCL 1.0 and 2021 Programming Language Documentation |url=https://github.com/KhronosGroup/OpenCL-Docs/releases/tag/cxxforopencl-docrev2021.12 |publisher=Khronos OpenCL Working Group |access-date=Dec 2, 2022 |date=Dec 20, 2021 |archive-date=2025-01-26 |archive-url=https://web.archive.org/web/20250126162657/https://github.com/KhronosGroup/OpenCL-Docs/releases/tag/cxxforopencl-docrev2021.12 |dead-url=no }}</ref>，它完全兼容于OpenCL 3.0标准。NVIDIA密切协作于Khronos OpenCL工作组，通过信号量和内存共享改进了[[Vulkan]]互操作<ref>{{Cite web|url=https://developer.nvidia.com/blog/using-semaphore-and-memory-sharing-extensions-for-vulkan-interop-with-opencl/|title=Using Semaphore and Memory Sharing Extensions for Vulkan Interop with NVIDIA OpenCL|date=February 24, 2022|access-date=2024-10-13|archive-date=2022-07-06|archive-url=https://web.archive.org/web/20220706061023/https://developer.nvidia.com/blog/using-semaphore-and-memory-sharing-extensions-for-vulkan-interop-with-opencl/|dead-url=no}}</ref>。小更新3.0.14版本，具有缺陷修正和针对多设备的一个新扩展<ref>{{cite web | url=https://www.phoronix.com/news/OpenCL-3.0.14 | title=OpenCL 3.0.14 Released with New Extension for Command Buffer Multi-Device | access-date=2024-10-13 | archive-date=2024-10-08 | archive-url=https://web.archive.org/web/20241008120142/https://www.phoronix.com/news/OpenCL-3.0.14 | dead-url=no }}</ref>。

== 範例 ==
=== 快速傅立葉變換 ===
一個[[快速傅立葉變換]]的式子：
<ref name=siggraph>{{cite web | url=http://s08.idav.ucdavis.edu/munshi-opencl.pdf | title=OpenCL | accessdate=2008-08-14 | publisher=SIGGRAPH2008 | date=2008-08-14 | archive-url=https://www.webcitation.org/66GmScoh5?url=http://s08.idav.ucdavis.edu/munshi-opencl.pdf | archive-date=2012-03-19 | dead-url=yes }}</ref>
<syntaxhighlight lang="c">
// create a compute context with GPU device
context = clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU, NULL, NULL, NULL);

// create a command queue
queue = clCreateCommandQueue(context, NULL, 0, NULL);

// allocate the buffer memory objects
memobjs[0] = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float)*2*num_entries, srcA, NULL);
memobjs[1] = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float)*2*num_entries, NULL, NULL);

// create the compute program
program = clCreateProgramWithSource(context, 1, &fft1D_1024_kernel_src, NULL, NULL);

// build the compute program executable
clBuildProgram(program, 0, NULL, NULL, NULL, NULL);

// create the compute kernel
kernel = clCreateKernel(program, "fft1D_1024", NULL);

// set the args values
clSetKernelArg(kernel, 0, sizeof(cl_mem),(void *)&memobjs[0]);
clSetKernelArg(kernel, 1, sizeof(cl_mem),(void *)&memobjs[1]);
clSetKernelArg(kernel, 2, sizeof(float)*(local_work_size[0]+1)*16, NULL);
clSetKernelArg(kernel, 3, sizeof(float)*(local_work_size[0]+1)*16, NULL);

// create N-D range object with work-item dimensions and execute kernel
global_work_size[0] = num_entries;
local_work_size[0] = 64;
clEnqueueNDRangeKernel(queue, kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL);
</syntaxhighlight>

真正的運算：（基於Fitting FFT onto the G80 Architecture）<ref name=VolkovKazianFFTG80>{{cite web | url=http://www.cs.berkeley.edu/~kubitron/courses/cs258-S08/projects/reports/project6_report.pdf | title=Fitting FFT onto G80 Architecture | accessdate=2008-11-14 | publisher=Vasily Volkov and Brian Kazian, UC Berkeley CS258 project report | format=PDF | date=May 2008 | archive-date=2012-03-19 | archive-url=https://www.webcitation.org/66GmTA1HM?url=http://www.cs.berkeley.edu/~kubitron/courses/cs258-S08/projects/reports/project6_report.pdf | dead-url=no }}</ref>
<syntaxhighlight lang="c">
// This kernel computes FFT of length 1024. The 1024 length FFT is decomposed into
// calls to a radix 16 function, another radix 16 function and then a radix 4 function

__kernel void fft1D_1024(__global float2 *in, __global float2 *out,
__local float *sMemx, __local float *sMemy){
int tid = get_local_id(0);
int blockIdx = get_group_id(0) * 1024 + tid;
float2 data[16];

// starting index of data to/from global memory
in = in + blockIdx; out = out + blockIdx;

globalLoads(data, in, 64); // coalesced global reads
fftRadix16Pass(data); // in-place radix-16 pass
twiddleFactorMul(data, tid, 1024, 0);

// local shuffle using local memory
localShuffle(data, sMemx, sMemy, tid, (((tid & 15)* 65) +(tid >> 4)));
fftRadix16Pass(data); // in-place radix-16 pass
twiddleFactorMul(data, tid, 64, 4); // twiddle factor multiplication

localShuffle(data, sMemx, sMemy, tid, (((tid >> 4)* 64) +(tid & 15)));

// four radix-4 function calls
fftRadix4Pass(data); // radix-4 function number 1
fftRadix4Pass(data + 4); // radix-4 function number 2
fftRadix4Pass(data + 8); // radix-4 function number 3
fftRadix4Pass(data + 12); // radix-4 function number 4

// coalesced global writes
globalStores(data, out, 64);
}
</syntaxhighlight>
Apple的網站上可以發現傅立葉變換的例子<ref name=AppleOpenCLFFT>.
{{cite web | url=https://developer.apple.com/mac/library/samplecode/OpenCL_FFT/index.html | title=OpenCL on FFT | accessdate=2009-12-07 | publisher=Apple | date=16 Nov 2009 | archive-date=2009-11-30 | archive-url=https://web.archive.org/web/20091130085543/http://developer.apple.com/mac/library/samplecode/OpenCL_FFT/index.html | dead-url=no }}</ref>

=== 平行合併排序法 ===
使用 Python 3.x 搭配 PyOpenCL 與 NumPy
<div style="height:400px; overflow-y: scroll;">
<syntaxhighlight lang="python3">
import io
import random
import numpy as np
import pyopencl as cl

def dump_step(data, chunk_size):
"""顯示排序過程"""
msg = io.StringIO('')
div = io.StringIO('')
for idx, item in enumerate(data):
if idx % chunk_size == 0:
if idx > 0:
msg.write(' ||')
div.write(' ')
div.write(' --')
else:
msg.write(' ')
div.write('------')
msg.write(' {:2d}'.format(item))

out = msg.getvalue()
if chunk_size == 1: print(' ' + '-' * (len(out) - 1))
print(out)
print(div.getvalue())
msg.close()
div.close()

def cl_merge_sort_sbs(data_in):
"""平行合併排序"""
# OpenCL kernel 函數程式碼
CL_CODE = '''
kernel void merge(int chunk_size, int size, global long* data, global long* buff) {
// 取得分組編號
const int gid = get_global_id(0);

// 根據分組編號計算責任範圍
const int offset = gid * chunk_size;
const int real_size = min(offset + chunk_size, size) - offset;
global long* data_part = data + offset;
global long* buff_part = buff + offset;

// 設定合併前的初始狀態
int r_beg = chunk_size >> 1;
int b_ptr = 0;
int l_ptr = 0;
int r_ptr = r_beg;

// 進行合併
while (b_ptr < real_size) {
if (r_ptr >= real_size) {
// 若右側沒有資料，取左側資料堆入緩衝區
buff_part[b_ptr] = data_part[l_ptr++];
} else if (l_ptr == r_beg) {
// 若左側沒有資料，取右側資料堆入緩衝區
buff_part[b_ptr] = data_part[r_ptr++];
} else {
// 若兩側都有資料，取較小資料堆入緩衝區
if (data_part[l_ptr] < data_part[r_ptr]) {
buff_part[b_ptr] = data_part[l_ptr++];
} else {
buff_part[b_ptr] = data_part[r_ptr++];
}
}
b_ptr++;
}
}
'''

# 配置計算資源，編譯 OpenCL 程式
ctx = cl.Context(dev_type=cl.device_type.GPU)
prg = cl.Program(ctx, CL_CODE).build()
queue = cl.CommandQueue(ctx)
mf = cl.mem_flags

# 資料轉換成 numpy 形式以利轉換為 OpenCL Buffer
data_np = np.int64(data_in)
buff_np = np.empty_like(data_np)

# 建立緩衝區，並且複製數值到緩衝區
data = cl.Buffer(ctx, mf.READ_WRITE | mf.COPY_HOST_PTR, hostbuf=data_np)
buff = cl.Buffer(ctx, mf.READ_WRITE | mf.COPY_HOST_PTR, hostbuf=buff_np)

# 設定合併前初始狀態
data_len = np.int32(len(data_np))
chunk_size = np.int32(1)

dump_step(data_np, chunk_size)
while chunk_size < data_len:
# 更新分組大小，每一回合變兩倍
chunk_size <<= 1
# 換算平行作業組數
group_size = ((data_len - 1) // chunk_size) + 1
# 進行分組合併作業
prg.merge(queue, (group_size,), (1,), chunk_size, data_len, data, buff)
# 將合併結果作為下一回合的原始資料
temp = data
data = buff
buff = temp
# 顯示此回合狀態
cl.enqueue_copy(queue, data_np, data)
dump_step(data_np, chunk_size)

queue.finish()
data.release()
buff.release()

def main():
n = random.randint(5, 16)
data = []
for i in range(n):
data.append(random.randint(1, 99))
cl_merge_sort_sbs(data)

if __name__ == '__main__':
main()
</syntaxhighlight>
</div>

執行結果：
<syntaxhighlight lang="text">
--------------------------------------------------------------------------------------
85 || 41 || 64 || 40 || 90 || 29 || 38 || 41 || 64 || 17 || 20 || 41 || 16 || 65 || 83
-- -- -- -- -- -- -- -- -- -- -- -- -- -- --
41 85 || 40 64 || 29 90 || 38 41 || 17 64 || 20 41 || 16 65 || 83
-------- -------- -------- -------- -------- -------- -------- --
40 41 64 85 || 29 38 41 90 || 17 20 41 64 || 16 65 83
-------------------- -------------------- -------------------- --------------
29 38 40 41 41 64 85 90 || 16 17 20 41 64 65 83
-------------------------------------------- --------------------------------------
16 17 20 29 38 40 41 41 41 64 64 65 83 85 90
--------------------------------------------------------------------------------------
</syntaxhighlight>

== 参见 ==
{{Portal|信息技术}}
* [[GPGPU]]
* [[CUDA]]
* [[DirectCompute]]
* [[C++ AMP]]

== 參考文獻 ==
{{Reflist|30em}}

== 外部連結 ==
* [https://www.khronos.org/conformance/adopters/conformant-products/opencl 支援OpenCL的產品]{{Wayback|url=https://www.khronos.org/conformance/adopters/conformant-products/opencl |date=20180629130825 }}
* [https://web.archive.org/web/20190122085224/http://www.opengpu.org/ 开源GPU社区]{{zh-cn}}

{{-}}
{{Khronos Group standards}}
{{并行计算}}

[[Category:C語言家族]]
[[Category:应用程序接口]]
[[Category:GPGPU]]
[[Category:GPGPU函式庫|OpenCL]]
[[Category:并行计算]]
[[Category:跨平台軟體]]

OpenCL - 版本历史

imported>ExultantEditor 来自 2025年12月26日 (五) 16:36