|
| 1 | += SYCL_INTEL_reqd_work_group_size |
| 2 | +:source-highlighter: coderay |
| 3 | +:coderay-linenums-mode: table |
| 4 | + |
| 5 | +// This section needs to be after the document title. |
| 6 | +:doctype: book |
| 7 | +:toc2: |
| 8 | +:toc: left |
| 9 | +:encoding: utf-8 |
| 10 | +:lang: en |
| 11 | + |
| 12 | +:blank: pass:[ +] |
| 13 | + |
| 14 | +// Set the default source code type in this document to C++, |
| 15 | +// for syntax highlighting purposes. This is needed because |
| 16 | +// docbook uses c++ and html5 uses cpp. |
| 17 | +:language: {basebackend@docbook:c++:cpp} |
| 18 | + |
| 19 | +// This is necessary for asciidoc, but not for asciidoctor |
| 20 | +:cpp: C++ |
| 21 | + |
| 22 | +== Introduction |
| 23 | +IMPORTANT: This specification is a draft. |
| 24 | + |
| 25 | +NOTE: Khronos(R) is a registered trademark and SYCL(TM) and SPIR(TM) are trademarks of The Khronos Group Inc. OpenCL(TM) is a trademark of Apple Inc. used by permission by Khronos. |
| 26 | + |
| 27 | +NOTE: This document is better viewed when rendered as html with asciidoctor. GitHub does not render image icons. |
| 28 | + |
| 29 | +This document describes an extension that introduces a +reqd_work_group_size+ attribute for SYCL kernels. This attribute enables developers to declare that a kernel may be invoked only with a specific work-group size. |
| 30 | + |
| 31 | +== Name Strings |
| 32 | + |
| 33 | ++SYCL_INTEL_reqd_work_group_size+ |
| 34 | + |
| 35 | +== Notice |
| 36 | + |
| 37 | +Copyright (c) 2019 Intel Corporation. All rights reserved. |
| 38 | + |
| 39 | +== Status |
| 40 | + |
| 41 | +Working Draft |
| 42 | + |
| 43 | +This is a preview extension specification, intended to provide early access to a feature for review and community feedback. When the feature matures, this specification may be released as a formal extension. |
| 44 | + |
| 45 | +Because the interfaces defined by this specification are not final and are subject to change they are not intended to be used by shipping software products. |
| 46 | + |
| 47 | +== Version |
| 48 | + |
| 49 | +Built On: {docdate} + |
| 50 | +Revision: 1 |
| 51 | + |
| 52 | +== Contact |
| 53 | +John Pennycook, Intel (john 'dot' pennycook 'at' intel 'dot' com) |
| 54 | + |
| 55 | +== Dependencies |
| 56 | + |
| 57 | +This extension is written against the SYCL 1.2.1 specification, Revision v1.2.1-6. |
| 58 | + |
| 59 | +== Overview |
| 60 | + |
| 61 | +The +reqd_work_group_size+ attribute is declared in the +intel+ namespace, and can be attached to the declaration of kernel functors and kernel lambdas using {cpp}11 attribute specifier syntax: |
| 62 | + |
| 63 | +[source,c++] |
| 64 | +---- |
| 65 | +struct Functor |
| 66 | +{ |
| 67 | + void operator()(nd_item<1> it) [[intel::reqd_work_group_size(32)]] |
| 68 | + { ... } |
| 69 | +}; |
| 70 | +---- |
| 71 | + |
| 72 | +[source,c++] |
| 73 | +---- |
| 74 | +auto kernel = [](nd_item<1> it) [[intel::reqd_work_group_size(32)]] { ... }; |
| 75 | +---- |
| 76 | + |
| 77 | +The attribute-token +intel::reqd_work_group_size+ shall appear at most once in each _attribute-list_. The attribute may be applied to the function-type in a function declaration. The first declaration of a function shall specify the +intel::reqd_work_group_size+ attribute if any declaration of that function specifies the +intel::reqd_work_group_size+ attribute. If a function is declared with the +intel::reqd_work_group_size+ attribute in one translation unit and the same function is declared without the +intel::reqd_work_group_size+ attribute in another translation unit, the program is ill-formed and no diagnostic is required. |
| 78 | + |
| 79 | +SYCL device compilers should give a compilation error if the required work-group size is unsupported. If the kernel is submitted for execution using an incompatible work-group size, the SYCL runtime must throw a SYCL +nd_range_error+. |
| 80 | + |
| 81 | +== Attribute Syntax |
| 82 | + |
| 83 | +The attribute has three variants, corresponding to one-, two- and three-dimensional work-groups: |
| 84 | + |
| 85 | +[source,c++] |
| 86 | +---- |
| 87 | +[[intel::reqd_work_group_size(dim0)]] |
| 88 | +[[intel::reqd_work_group_size(dim0, dim1)]] |
| 89 | +[[intel::reqd_work_group_size(dim0, dim1, dim2)]] |
| 90 | +---- |
| 91 | + |
| 92 | +The sizes are written in row-major format. Each argument to the attribute must be a +constexpr+ integral value. The dimensionality of the attribute variant used must match the dimensionality of the work-group used to invoke the kernel. |
| 93 | + |
| 94 | +== Issues |
| 95 | + |
| 96 | +None. |
| 97 | + |
| 98 | +//. asd |
| 99 | +//+ |
| 100 | +//-- |
| 101 | +//*RESOLUTION*: Not resolved. |
| 102 | +//-- |
| 103 | +
|
| 104 | +== Revision History |
| 105 | +
|
| 106 | +[cols="5,15,15,70"] |
| 107 | +[grid="rows"] |
| 108 | +[options="header"] |
| 109 | +|======================================== |
| 110 | +|Rev|Date|Author|Changes |
| 111 | +|1|2020-01-24|John Pennycook|*Initial public working draft* |
| 112 | +|======================================== |
| 113 | +
|
| 114 | +//************************************************************************ |
| 115 | +//Other formatting suggestions: |
| 116 | +// |
| 117 | +//* Use *bold* text for host APIs, or [source] syntax highlighting. |
| 118 | +//* Use +mono+ text for device APIs, or [source] syntax highlighting. |
| 119 | +//* Use +mono+ text for extension names, types, or enum values. |
| 120 | +//* Use _italics_ for parameters. |
| 121 | +//************************************************************************ |
0 commit comments