|
| 1 | += sycl_ext_intel_runtime_buffer_location |
| 2 | + |
| 3 | +:source-highlighter: coderay |
| 4 | +:coderay-linenums-mode: table |
| 5 | + |
| 6 | +// This section needs to be after the document title. |
| 7 | +:doctype: book |
| 8 | +:toc2: |
| 9 | +:toc: left |
| 10 | +:encoding: utf-8 |
| 11 | +:lang: en |
| 12 | +:dpcpp: pass:[DPC++] |
| 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 | +== Notice |
| 20 | + |
| 21 | +[%hardbreaks] |
| 22 | +Copyright (C) 2022-2022 Intel Corporation. All rights reserved. |
| 23 | + |
| 24 | +Khronos(R) is a registered trademark and SYCL(TM) and SPIR(TM) are trademarks |
| 25 | +of The Khronos Group Inc. OpenCL(TM) is a trademark of Apple Inc. used by |
| 26 | +permission by Khronos. |
| 27 | + |
| 28 | +== Contact |
| 29 | + |
| 30 | +To report problems with this extension, please open a new issue at: |
| 31 | + |
| 32 | +https://github.com/intel/llvm/issues |
| 33 | + |
| 34 | +== Dependencies |
| 35 | + |
| 36 | +This extension is written against the SYCL 2020 revision 4 specification. All |
| 37 | +references below to the "core SYCL specification" or to section numbers in the |
| 38 | +SYCL specification refer to that revision. |
| 39 | + |
| 40 | +== Status |
| 41 | +This is a proposed extension specification, intended to gather community |
| 42 | +feedback. Interfaces defined in this specification may not be implemented yet |
| 43 | +or may be in a preliminary state. The specification itself may also change in |
| 44 | +incompatible ways before it is finalized. *Shipping software products should |
| 45 | +not rely on APIs defined in this specification.* |
| 46 | + |
| 47 | +== Overview |
| 48 | + |
| 49 | +This document propose a new buffer_location runtime property that can be |
| 50 | +passed to `malloc_device`. |
| 51 | + |
| 52 | +On targets that provide more than one type of global memory, this provide |
| 53 | +users the flexibility of choosing which memory the device usm should be |
| 54 | +allocated to. |
| 55 | + |
| 56 | +== Specification |
| 57 | + |
| 58 | +=== Feature test macro |
| 59 | + |
| 60 | +This extension provides a feature-test macro as described in the core SYCL |
| 61 | +specification. An implementation supporting this extension must predefine the |
| 62 | +macro `SYCL_EXT_INTEL_RUNTIME_BUFFER_LOCATION` to one of the values defined in the table |
| 63 | +below. Applications can test for the existence of this macro to determine if |
| 64 | +the implementation supports this feature, or applications can test the macro's |
| 65 | +value to determine which of the extension's features the implementation |
| 66 | +supports. |
| 67 | + |
| 68 | +[%header,cols="1,5"] |
| 69 | +|=== |
| 70 | +|Value |
| 71 | +|Description |
| 72 | + |
| 73 | +|1 |
| 74 | +|The APIs of this experimental extension are not versioned, so the |
| 75 | + feature-test macro always has this value. |
| 76 | +|=== |
| 77 | + |
| 78 | +== Examples |
| 79 | + |
| 80 | +[source,c++] |
| 81 | +---- |
| 82 | +array = (int *)sycl::malloc_device<int>( |
| 83 | + N, q, |
| 84 | + sycl::property_list{sycl::ext::intel::experimental::property::usm::buffer_location(2)}); |
| 85 | +
|
| 86 | +sycl::queue q; |
| 87 | +q.parallel_for(sycl::range<1>(N), [=] (sycl::id<1> i){ |
| 88 | + data[i] *= 2; |
| 89 | +}).wait(); |
| 90 | +---- |
| 91 | + |
| 92 | + |
| 93 | +=== Changes to runtime properties |
| 94 | + |
| 95 | +This extension adds the new property |
| 96 | +`sycl::ext::intel::experimental::property::usm::buffer_location` which |
| 97 | +applications can pass in the property_list parameter to all overloads of the |
| 98 | +`sycl::malloc_device()`, `sycl::malloc_shared()`, and `sycl::malloc_host()` |
| 99 | +functions. However, this property has no effect when passed to |
| 100 | +`sycl::malloc_shared()` or `sycl::malloc_host()`. Following is a synopsis of |
| 101 | + this property. |
| 102 | + |
| 103 | +[source,c++] |
| 104 | +---- |
| 105 | +namespace sycl::ext::intel::experimental::property::usm { |
| 106 | +
|
| 107 | +class buffer_location { |
| 108 | + public: |
| 109 | + buffer_location(int location); |
| 110 | + int get_buffer_location() const; |
| 111 | +}; |
| 112 | +
|
| 113 | +} // namespace sycl::ext::intel::experimental::property::usm |
| 114 | +---- |
| 115 | + |
| 116 | +On targets that provide more than one type of global memory, `buffer_location` |
| 117 | +allows user to specify which of the global memory to allocate memory to. This |
| 118 | +provide user the flexibility to choose the global memory that satisfy |
| 119 | +requirements for bandwidth and throughput. |
| 120 | + |
| 121 | +This property is ignored for non-FPGA devices. Attempting to use this |
| 122 | +extension on other devices or backends will result in no effect. |
| 123 | + |
| 124 | + |
| 125 | +== Issues |
| 126 | + |
| 127 | +== Revision History |
| 128 | + |
| 129 | +[cols="5,15,15,70"] |
| 130 | +[grid="rows"] |
| 131 | +[options="header"] |
| 132 | +|======================================== |
| 133 | +|Rev|Date|Author|Changes |
| 134 | +|1|2022-03-01|Sherry Yuan|*Initial public draft* |
| 135 | +|======================================== |
0 commit comments