Skip to content

Latest commit

 

History

History
5046 lines (4756 loc) · 307 KB

ReleaseNotes.md

File metadata and controls

5046 lines (4756 loc) · 307 KB
Raw

Nov'23 release notes

Release notes for commit range f4e0d3177338..f4ed132f243a

New Features

SYCL Compiler

  • Added -fpreview-breaking-changes option (see "Upcoming API/ABI breakages" section for more details) [08febcfcb64a] [99b97eef4278]
  • Enabled early AOT abilities when creating objects [6f248088541c]
  • Added support for -faltmathlib [32c3e642d644]
  • Added support for -ftarget-register-alloc-mode [8e5275af77af]
  • Added -fsycl-add-default-spec-consts-image command line option to Clang Driver [3847c7c6c5c7]

SYCL Library

  • Added experimental implementation of SYCL bindless images extension [58a8f2001d8b] [380453deeb51] [b1aab046f408]
  • Added support for local accessors to copy_from/copy_to for ESIMD [474461cb2e1d]
  • Added support for local accessors to atomic update for ESIMD [8974cb944b3c]
  • Added support for local accessors to lsc for ESIMD [9dce2d227b31]
  • Added new raw_send APIs moving compile time params to template params for ESIMD [1a5e53216b00]
  • Added SYCLcompat library [bedd818eb281] [465aa56ee5b3] [75b31a4c24b4] [4dfaf20884c7] [47d84897a87b] [8c07a840311a] [ab578107b6f8] [ac5a90e0c1b5] [a733fa74b583] [c39c2138f5a0]
  • Implemented sycl_ext_oneapi_auto_local_range [a6d4c2120b5c]
  • Implemented sycl_ext_oneapi_sub_group_mask version 2 [3bd09b9089c5]
  • Implemented sycl_ext_oneapi_queue_priority [0c33fea5fac0]
  • Implemented optional kernel feature for joint_matrix and joint_matrix_mad [946b1f236167] [c38dfcf64681]
  • Implemented sycl_ext_oneapi_prefetch [e7139b0f15d0] [0229456e19d5] [b5d69df53620]
  • Implemented sycl_ext_oneapi_device_architecture for HIP and CUDA devices [1ad69e59eb13]
  • Improved error message when running ESIMD code on unsupported devices [a5f471ea1535]
  • Added ext_oneapi_non_uniform_groups aspect [de92299c2c09]

Documentation

Improvements

SYCL Compiler

  • Added diagnostics if SYCL-specific command line options are used without -fsycl [bf128c873e74]
  • Enabled INTEL_SYCL_OPAQUEPOINTER_READY [a47705f2d865] [9dded21c5984]
  • Supported multiple SYCL targets with native_cpu in the same compiler invocation [aab5d747c46e]
  • Made -fintelfpga flag imply -fsycl. [ad2f959d8b88] [6d48bc13c456]
  • Ensured debug format for SYCL device on Windows is the expected format [73bba79a3ffb]
  • Disabled parallel for range rounding at -O0 [2c117d7a6104]
  • Fixed missing optnone attribute in -O0 mode on non-FPGA target to improve debuggability [993fe5b31b43]

SYCL Library

  • Supported missing types for the ldg in the sycl_ext_oneapi_cuda_tex_cache_read [aec8a35788c1]
  • Stopped throwing exception when passing empty accessor to handler::require() [7bf1f57bc26e]
  • Moved addc() and subb() functions out of experimental namespace [9e3852bc1c2e]
  • Moved BFN function from experimental namespace [c9da5c7ff7e8]
  • Improved error message when queue::memcpy receives a null pointer by reporting source code info(file/line number) [626c4b5078c1]
  • Improved accuracy of double tanpi host implementation [041a74c0cbda]
  • Added support bitwise and/or/xor and logical and/or subgroup collective functions for AMD[288aeaef03a9]
  • Updated annotated_ptr API to match spec changes [954730e7a823] [fbad42a398ef] [43336a15fcaf]
  • Implemented info::event_profiling::command_submit support for platforms that use OpenCL version less than 2.1 [7cfa951303eb]
  • Fixed required alignment in buffers [093dae1f7214]
  • Improved error message when using sycl_ext_oneapi_graph features that are not yet implemented [8d8d3f4f52b0]
  • Supported get_multi_ptr in ESIMD context [2f0f3814c541]
  • Accept integral type in parallel_for with offsets [ecfc49c825c6]
  • Added support for assert() in ESIMD kernels [545a043ebaf5]
  • Added missing Intel archs to sycl_ext_oneapi_device_architecture [c4db251179dc]
  • Enabled range rounding for unnamed lambdas [4f30e6648d9e]
  • Fixed loading of plugins when path to them contains non-ANSI symbols [5c30815d4821]
  • Improved the error message when the command group function object passed to queue::submit has the wrong signature [5801970585b2]
  • Improved handler::fill performance on 2D and 3D buffers [0249a55f810e]
  • Added defaults for streaming and register_map fpga_kernel_properties [23a6f389c1e4]
  • Moved GRF property conflict check to compile time [4fb92d35c7be]
  • Added error when using invoke_simd with callables with struct arguments [03c13fd53a24]
  • Added missing rintf for CUDA [3c327c736ced]
  • Optimized kernel launch on CUDA by caching the max local mem size so that CUDA driver is called less [1c892f952cae]
  • Made sycl_ext_oneapi_graph functions thread-safe [c8c64a673eeb]
  • Improved diagnostic for nd_item passed to parallel_for with range [e600d7922489]
  • Implemented missing exceptions defined by sycl_ext_oneapi_graph specification [77b794bcaad6]
  • Implemented supported make_device, make_queue, make_event interop functions for HIP backend [5e9d07b14a47]
  • Added missing nearbyint for CUDA [0ef26d3ee808]
  • Implemented runtime query for matrix combinations [a09491f4e389]
  • Supported basic kernel launches over 64-bit global ranges [192c3016e76b]
  • Clarified runtime error for nested kernel submissions [7afc2d03fd00]
  • Fix compiler crash when spec constants are structs with padding in the middle [292309c826a3]
  • Added support for CUDA backend with sycl_ext_oneapi_graph [367b662a49ba]
  • Added support for specialization constants in sycl_ext_oneapi_graph nodes [fca7f5e2e3fb]
  • Improved vec::convert implementation to use a single instruction instead of per-element [f634cc9bf799]
  • Made annotated_arg<T> implicitly device copyable if T is device copyable [9f65e61a2dbf]

Documentation

  • Updated sycl_ext_oneapi_local_memory [457251feceb8]
  • Added get-coord API and general query example to sycl_ext_intel_matrix [579f1dec0b5b] [38ac2124c1e8]
  • Updated design doc for new offloading model [115808f5128f]
  • Updated description of the default value of the environment variable SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS to reflect actual behavior [ad973f3383e2]
  • Added overloads with sycl::queue for some functions in sycl_ext_oneapi_bindless_images [4d0c552985bd]
  • Specify an error in sycl_ext_oneapi_graph if an executable graph is submitted to a queue with a different device or context [2c53815c6c1e]

Bug Fixes

SYCL Compiler

  • Fixed backend option passing issues with multiple targets [12ca2db0e30b]
  • Fixed -fgpu-rdc option for CUDA [f7595ac7527e]
  • Fixed triple in offload mismatch warning [0fd9a4eb17a2]
  • Fixed a compilation issue when -enable-global-offset=false for AMDGU and NVPTX targets [00cf4c29740b]

SYCL Library

  • Fixed stack corruption in sycl::modf for sycl::half type [7c6541dc18e3]
  • Fixed nextafter(half) precision on host [1f10f3588e8c]
  • Fixed identity values for (sub)group collectives min/max for AMD and CUDA [248389371932]
  • Corrected exception thrown when kernel is submitted with a large (unsupported) reqd_work_group_size attribute [25c3666dffa9]
  • Implemented hip prefetch/mem_advise (noop) for rocm < 5.0 [7e4cc1113987]
  • Adjusted range definition so it has a default constructor in accordance with the SYCL spec [0289d2ac94f9]
  • Added missing return statement in the multi_ptr<void..> conversion operator [836469ba88e2]
  • Optimized kernel fusion implementation so that some resources can be released earlier [a33b0a950ae6]
  • Fixed an integer overflow that happened when sub-buffer size exceeds INT_MAX [2fb0b140ad3b]
  • Fixed std::enable_if typo for marray and multi_ptr [0ef25ec3bdf8]
  • Fixed an issue with reqd_work_group_size attribute values bigger than INT_MAX being incorrectly handled [cee07d3d2652]
  • Fixed UB in group sort scratch memory access [ed4c01301a7c]
  • Use primary context in HIP adapter [d1c92cb95b29]
  • Fixed a crash when default-constructed accessor is passed to a kernel [fb8ddbf7b0ea]
  • Stopped sycl_ext_oneapi_graph subgraph nodes from creating incorrect dependencies on subsequent usages of the subgraph [92ddf8d562d7]
  • Fixed ambiguous log2 call in host code [12fa0af28e88]

Documentation

  • Updated device_image_scope property in sycl_ext_oneapi_device_global [be8e031cd078]

API/ABI breakages

  • Deprecated experimental functions for ESIMD: quot, div, mod, dp*, line, lrp [76976a22ba2e]
  • Removed ESIMD Emulator plugin [1ece6da6b113]
  • Deprecated sycl::abs(genfloat) [d15d4409b2cf]

Upcoming API/ABI breakages

The following changes ared only in effect if the -fpreview-breaking-changes flag is set.

  • Fixed sycl::vec ABI inconsistencies and made it trivially copyable [531aabfdee91] [f4ed132f243a]
  • Updated sycl.hpp to not include <complex> and <cmath> [6ed0ab887d18] [d22c9c58d2e7]

Known Issues

  • Having MESA OpenCL implementation which provides no devices on a system may cause incorrect device discovery. As a workaround such an OpenCL implementation can be disabled by removing /etc/OpenCL/vendor/mesa.icd.
  • -fsycl-dead-args-optimization can't help eliminate offset of accessor even though it's created with no offset specified
  • SYCL 2020 barriers show worse performance than SYCL 1.2.1 do. [18c80faa]
  • When using fallback assert in separate compilation flow it requires explicit linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero. 7dfaf3bd
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior.
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows. [50628db1]
  • The support of accessor and local_accessor for ESIMD is still limited comparing to SYCL.
  • sycl_ext_oneapi_root_group implementation has the following limitations: max_num_work_group_sync query always returns 1 and the implemented barrier has work group scope since all work items in a root group are currently in the same work group.

Oct'23 release notes

Release notes for commit range cb91c232c661..f4e0d3177338

New features

SYCL Compiler

  • Enabled Pre-Compiled header (PCH) inclusion while performing host compilation with -fsycl. [7f12b287236f]
  • Added support for -ftarget-export-symbols option which allows to export symbols for AOT. [c33ba5dafe91]
  • Added support for reqd_sub_group_size to sycl::is_compatible and implemented device code split based on reqd-sub-group-size optional kernel feature. [841768794f7a]
  • Added support for -ftarget-compile-fast for spir64_gen and JIT mode. [b6b01b48cfd9][92855bea87f4]
  • Implemented adding C++ libraries to the linker with -fsycl. [da5e1b98f1c3]
  • Added support for multiple call operators in a kernel functor. [84cc1e1eff04]
  • Added support to propagate compile flags to device backend compiler.[f45fb51ba6cf]
  • Added new FPGA loop attribute enable_loop_pipelining. [4147b8ded75f]

SYCL Library

  • Enabled sycl_ext_oneapi_annotated_arg and sycl_ext_oneapi_annotated_ptr experimental extensions. [05d3be63f996][80b7de1d63f4][557204abf242]
  • Implemented sycl_ext_intel_queue_immediate_command_list extension. [2f2560fe74e6]
  • Implemented sycl_ext_oneapi_copy_optimize experimental extension. [65cc0cfe809f]
  • Added initial implementation of the experimental SYCL Graph extension sycl_ext_oneapi_graph [35465da4824a][f9d4830babca][72341ee375d6][96a605095665]
  • Added dimensions member to item/range-like types. [9c7bd9ea046f]
  • Added support of queue::priority_* properties for OpenCL backend. [25fd689999f7]
  • Implemented initial version of the SYCL Native CPU Plug-in designed in SYCLNativeCPU. [b2d0837e9644] [5b501eeca963]
  • Added __imf_max/min/hadd/fast_* functions to imf device libraries. [fe01366e8b08][d96e5075303a]
  • Introduced and implemented new sycl::ext::oneapi::experimental::info::device::architecture device descriptor and device::ext_oneapi_architecture_is(ext::oneapi::experimental::architecture) host API as part of sycl_ext_oneapi_device_architecture extension (for Level Zero and OpenCL). [d0b01b265d92]
  • Added experimental implementation of sycl_ext_intel_grf_size [370aa2a01711]
  • Experimental sycl_ext_oneapi_device_global extension is supported now. [d812d1e13ab5]
  • Enabled media API that works with image accessors for ESIMD. Those accessors do not depend on stateful/stateless mode. [b95b2d88e446]
  • Allowed implicit conversion from std::experimental::simd_mask to ESIMD::simd_mask. [de5b479a2de9]
  • Added host_task image accessor support. [b6faefb08e23]
  • Started to support 64-bit offsets with accessors in stateless mode for ESIMD APIs. [f28ae00bf6a3][d04ebb03c1c8][62955fc20e14]
  • Added __imf_llabs to the imf device libraries. [3c19581f828c]
  • Added explicit conversion of multi_ptr<T> to multi_ptr<const T>. [10d2f5b613f3]
  • Added support for scalar logical operators with group algorithms. [11ac73003056][90406b2ba07c]
  • Implemented device query for 64 bit atomic support in ESIMD emulator. [c40baa6db64b]
  • Added support of 16 bit data for lsc_atomic_update and lsc_slm_atomic_update ESIMD API. [3028d82a75d2]
  • Added initial implementation of sycl_ext_oneapi_root_group extension. There are severe limitations: max_num_work_group_sync query always returns 1 and the implemented barrier has work group scope since all work items in a root group are currently in the same work group. [743c35be2da7]
  • Added support for tf32 type using the unified interface for SYCL Matrix extension. [aba6d85f9f16]
  • Implemented Host Pipes described in sycl_ext_intel_dataflow_pipes extension. [992ef064289f][5bd42eb6a7df]
  • Added support for more math built-ins for bfloat16 together with the extension specification sycl_ext_oneapi_bfloat16_math_functions update [c7759bb8cd0b]
  • Added decorated async_work_group_copy overloads. [d1f67ca228f2]
  • Added initial implementation of the Unified Runtime plugin and routed to it with SYCL_PREFER_UR. [5a239c6a08e1]
  • Added support for accessors to atomic_update ESIMD API. [59a1aa0841da]
  • Enabled pretty-printing of reference objects by GDB. [e3fc16f56c0e]
  • Added Xmethods that help GDB to support reading local_accessors on GPU from SLM. [ddfc5220afaa]
  • Enabled passing local accessors to ESIMD kernel and via invoke_simd() API, enabled usage of get_pointer() and operator[] for accessors in ESIMD kernel. The support of accessor and local_accessor is still limited comparing to SYCL. [21ca00fb1cc5][82ae85d12bf0]
  • Implement unpadding for 2d block load/store ESIMD API. [9be8a306d0eb]
  • Added SYCL 2020 image classes. [9035cdb0898d][1a3e99307e4f]
  • Added interfaces for SYCL 2020 image accessors, only the host accessors interfaces are actually functional. [a58d8e3e88c0]
  • Added XPTI notifications for SYCL 2020 images as well as for the associated accessor classes. [8c535276785d][73d5c04d0305]
  • Started to print device aspects in sycl-ls --verbose. [e32ab1759fed]
  • Enabled group algorithms to be used with tangle_group and opportunistic_group arguments. [29e629e00e83]
  • Implemented info::device::backend_version query. [df743a5c97e1]
  • Added fixed_size_group support to algorithms. [01ac0338bfa9]
  • Added simple abs(int) to imf libdevice. [2a3aaeea5be8]
  • Added ballot_group support to algorithms. [70a35de4eacb]
  • Implemented interoperability support for images for the Level Zero backend. [836ceec36b57]
  • Added marray support for math built-in functions. [a7a6de2e40a2][c274f40588ff][2953d2548791]
  • Enabled inline assembly support in ESIMD kernels. [f5a062dd057f][48abe38c0bee]
  • Enhanced interop queue interface to choose standard or immediate commandlists. [0ff3ec40e241]
  • Enabled double type for atomic_update() ESIMD API. [98c443543c72]
  • Added support for addc and subb operations for ESIMD. [0a098933d9ba]
  • Allowed zero-sized 3D accessors. [5cb8279ddb8e]

Documentation

Improvements

SYCL Compiler

  • Deprecated -fsycl-link-huge-device-code in favor of a new option, -flink-huge-device-code. The new option is identical in functionality but allowed with -fopenmp-targets. [79ea9b7aa52b]
  • Optimized size of produced binaries when device_global is used. [f99335f5e71a]
  • Started to emit an error when PCH is triggered in SYCL mode. [062446584d52]
  • Improved FPGA archive device unbundling with AOCO. [4ed3676f6402]
  • Moved imf abs to a separate device library for Deep Learning. [e4f074a6aabe]
  • Fixed the bug report URL for DPC++. [98b7de88a7d9]
  • Started to link with bfloat16 related device libraries only when they are used. [663042b04b63]
  • Started to properly pass -fsycl-esimd-force-stateless-mem to the host code compilation. [8b85b6400b82]

SYCL Library

  • Combined ADL-S and RPL-S device architectures. [f87be6f17428]
  • Implemented multi_ptr default to be legacy to avoid code break with SYCL 1.2.1. [52edb3798749]
  • Started to use aggregate initialization for group_local_memory arrays according to updated sycl_ext_oneapi_local_memory.[b50440ec76b9]
  • Deprecated ESIMD Emulator. [177680e3e918]
  • Deprecated ext::oneapi::sub_group. [0662e2a81014]
  • Improved error message related to ext_intel_free_memory aspect. [c1bfcaf6d6ed]
  • Removed non-necessary barrier after region copy in the Level Zero backend. [768e61e7966c]
  • Fixed get_info<device::free_memory> to check its aspect. [dfb75675bc0a]
  • Removed the workaround for release of auxiliary buffers. [fda9171fc218]
  • Enabled optimization for read-only buffer accesses from multiple devices in a context. [dddaf7f5d54d]
  • Removed old specialization constants extension and implementation. [e50ae05e28d3][833a9fe0a9cf]
  • Improved is_compatible to check if specific target is defined with -fsycl-targets and change the result. [ef033238cfb5]
  • Improved quotient approximation for host sycl::remquo. [d335fdf4afa2]
  • Allowed accessor constructed with zero-size buffers. [0c0809590f37]
  • Improved resources recycling for the Level Zero backend by cleaning up events on in-order queue wait. [260182a1ad75]
  • Added code_location parameter to the rest of sycl::queue methods which allows to improve error reporting. [fdd609a5c724]
  • Enabled xpti::node_create signal emit for parallel_for that bypasses graph. [4eaaaa963ca2]
  • Improved the accuracy of host sycl::cospi. [37aa84b7afd5]
  • Replaced usage of outdated memset OpenCL API clEnqueueMemsetINTEL with clEnqueueMemFillINTEL. [c5d0e1d4d9be]
  • Added memory pointer to XPTI memory allocation event metadata to allow more informative XPTI memory tracing. [01ffe65da562]
  • Implemented recycling of immediate command lists for queues in a context for the Level Zero backend. [2ddbf8c3b7a5]
  • Optimized ext_oneapi_submit_barrier() for in-order queue for the Level Zero backend. [c21fd476c78a]
  • Change the SYCL_PI_LEVEL_ZERO_USM_RESIDENT default to force device allocations only. [33874f76c59f]
  • Started to report false for aspect::image on all devices. [5cf0f7cf5c21]
  • Removed lambda parameter from "if_architecture_is" according to sycl_ext_oneapi_device_architecture. [71f745bcda7d]
  • Improved error reporting when reqd_work_group_size is not supported by a device. [084c02797084]
  • Adjusted static restriction checks for block_2d APIs. [7902d6b35348]
  • Disallowed local accessor misuse according to SYCL 2020. [d7a5ec048368]
  • Implemented passing address space information to SPIR-V Joint Matrix Load/Store intrinsics. [854ab7eab5e4]
  • Enabled immediate command lists by default for the Level Zero backend. [9e5b2ed1ddc3]
  • Improved SYCL RT performance by removing extra map lookup for eliminated kernel arguments during enqueue of cached kernels. [7a1fa6c38e03]
  • Changed the default to SYCL_PI_LEVEL_ZERO_USM_RESIDENT=2. [85435666d477]
  • Added heuristics to reduce overhead from immediate command-list cleanup in the Level Zero backend. [ce69d6906e30]
  • Renamed cluster_group to fixed_size_group. [62c091a0af17]
  • Added error for invalid uniform arguments in InvokeSIMD. [487c1f8cb040]
  • Reduced overhead from queue profiling by using steady_clock::now() to retrieve host time as piGetDeviceAndHostTimer has large overhead. [070598e433a7]
  • Implemented graceful handling of the case that Level Zero runtime was already unloaded when we do cleanup. [928a919f15d9]
  • Improved error messages for InvokeSIMD. [0c74bbb20e65]
  • Updated native_specialization_constant() for the case when no specialization constants are present. [9eb1184686fb]
  • Optimized re-use of command lists in the Level Zero backend. [24ae0153e8e8]
  • Changed the behavior of sycl::maximum and sycl::minimum to be consistent with std::max and std::min. [77d648d949c3]
  • Switched to sycl::exception instead of sycl::runtime_error in the majority of SYCL device headers. sycl::runtime_error is deprecated in SYCL 2020. [23a836f39af6]

SYCL Tools

  • Added USM calls parameter verification layer to sycl-trace which provides improved diagnostic on accessing invalid device memory under USM. [758bd264bb90]
  • Started to print supported SG sizes in sycl-ls --verbose. [12515177ab03]

Documentation

Bug fixes

SYCL Compiler

  • Fixed uses_aspects to be applied to function declarations and not only function definitions. [5d928976f9e2]
  • Fixed ivdep attribute in template contexts. [008e3743929e]
  • Fixed option restriction for device with Windows. [be3482ff3f0e]
  • Fixed handling unsupported attributes for ESIMD. [8990c5503d47]
  • Fixed optimization option processing for device options. [9a82d283ae2b]
  • Enabled proper behavior of optional kernel features with SYCL_EXTERNAL. [1dce70f413e6]
  • Fixed "SLM init call is supported only in kernels" message for slm_init() inside a ESIMD kernel when compiling with -O0 option. [2056c669b122]
  • Fixed memory leak in Base64::decode. [28113ec69167]
  • Fixed crash when using libraries with -fsycl. [ef0d151e57a4]
  • Added predefines for /MDd to fix certain build issues when using reductions on Windows. [1a602d77b7b2]
  • Fixed compilation issue when kernel name is a class defined with final keyword. [9eeb7825ea2d]
  • Fixed an issue with setting VCSLMSize attribute which was causing JIT compilation failure when using slm_init() in ESIMD kernel. [f24a0bb343b2]
  • Fixed option for undefined symbols for hostdep link to work with gold linker. [554ed2187d32]
  • Fixed directory field of DIFile. [797187de2199]

SYCL Library

  • Fixed the bug in the Level zero backend where zeModuleDestroy is called although ext::oneapi::level_zero::ownership::keep is passed to make_kernel_bundle leading to a double free corruption. [1344fa29e6eb]
  • Addressed specification mismatches for multi_ptr. [ed5941feb363]
  • Fixed corner case when using short or char with exclusive scan. [aa31c1e2deb9]
  • Updated the legacy multi_ptr construction from local_accessors to conform with the construction method used by other accessors. [6314af860c8c]
  • Made get_pointer noexcept to comply with SYCL 2020. [85d84114510c]
  • Started to return the correct sycl::errc in case of invalid sycl::reqd_work_group_size. [d16a1fd39c08]
  • Fixed sycl::remquo truncation error. [43fbde38e26f]
  • Fixed linkage errors when trying to get mem_channel property or check it using has_property(). [9352172a9b81]
  • Fixed dynamic loading of ocloc tool for online_compiler. [1854b263672b]
  • Fixed global memory reporting for Arc. [0a99d87a7f31]
  • Fixed sycl::sub_group to follow by-value semantics according to SYCL 2020. [064f332f156c]
  • Fixed invalid value returned by event.get_info<sycl::info::event::command_execution_status>() for OpenCL backend. [5a91df97973a]
  • Fixed assignment operator for ESIMD globals. [a3a9dd02d420]
  • Fixed the Level Zero backend to not destroy build log on program build failure.[b7ffbc776b37]
  • Fixed error appearing when validation layer is enabled for the Level Zero with ZE_DEBUG=6. [abfdb6306d7c]
  • Allowed group algorithms to accept a function object with an explicit type. [78f09e6f31c6]
  • Fixed sycl::is_compatible() for the case when vector of kernels ids is empty. [23fbba1e6b8b]
  • Fixed multi_ptr ctor for extended address spaces. [a69e5152874e]
  • Fixed trivially_copyable to device_copyable for fill and copy. [9c662e228fa4]
  • Fixed range deductions in reduction parallel_for. [3024161f1b7b]
  • Adjusted multi_ptr deduction guides to comply with SYCL 2020. [3df6fdc58b00]
  • Fixed gather/scatter with accessors when passing scalar for ESIMD backend. [27755824d050]
  • Fixed lost data during implicit conversion in local and host accessors. [7618dffd78ae]
  • Fixed memory leak because of unnecessary kernel retain. [f110fd73f8e7]
  • Fixed the pointer type contained by a read-only accessor to const. [fd4598035b7e]
  • Fixed max_sub_group_size query for devices without sub groups.[98cdf4c3acda]
  • Removed undefined behavior in host-side abs_diff. [a4105101e3c5]
  • Fixed vec::as<vec<bool, N>>(). [72cffff649f5]
  • Fixed segfault when using ZE_DEBUG and an exception is thrown on device selection phase. [05866c17ce87]
  • Removed an unnecessary multi_ptr conversion which had conflict with an existing one resulting in ambiguity on implicit conversion. [aba3007fe71b]
  • Fixed lost data in accessor implicit conversion. [10551294dcc7]
  • Fixed stack smashing which happened in some device info queries on OpenCL backend. [f4577ff62ace]
  • Stopped shipping the Level Zero loader and headers with the DPCPP toolchain installation. [6581bc013a46][ad8b7d447612]
  • Corrected the PVC device id check in the Level Zero backend. [59a23020e1be]
  • Fixed ambiguity for bit_cast. [834df4737a85]
  • Fixed nan/inf handling in sycl::stream when in -fast-math mode. [757e4a72d58d]
  • Allowed for different types in group algorithms reduce, exclusive_scan, and inclusive_scan. [8ad0e1df6e76]
  • Fixed bug in get methods in the config_2d_mem_access class for 2D block stateless load/store API. [7e5f529db655]
  • Fixed compilation error for sycl::fabs builtin in -fast-math mode. [43cb2c0e4867]
  • Allowed host_task deduction tags in placeholder accessors. [21beef77542b]
  • Added iterator operations for zero-dimension accessors to comply with SYCL 2020. [6eb8b4369b64]
  • Fixed rounding issue in __imf_vavgs. [38d05361f06e]
  • Fixed bug in fill operation for zero-dimensional accessor. [5b46e0dbcd2c]
  • Fixed arguments passed to ocloc via -ftarget-compile-fast. [d8bb69e5b166]
  • Implemented implicit conversion for local and host accessors to comply with SYCL 2020. [3d4da8fb5eee]
  • Throw exception when empty accessor calls require() to comply with SYCL 2020. [aa08c7e9664a]
  • Fixed operator~ for sycl::vec<bool, N>. [58e64ce83be6]
  • Fixed sub_group shuffle for vectors of long long and half. [eeba87925d53]
  • Fixed static destruction order issue in OpenCL extension fptr cache. [379a094d8726]
  • Fixed leak of active barriers' events in the Level Zero backend. [86860693375c]
  • Implemented proper queries for aspect::ext_oneapi_srgb. [23978b462fc7]
  • Added vec assignment from scalar and more vec modulus overloads to comply with SYCL 2020. [68e13d41f9c3]
  • Fixed handling of mem_channel buffer property. [909343107452]
  • Fixed operator& and operator[] in local_accessor<const T>. [d5477c87c700]
  • Fixed the third argugment type for select built-in. [528eba8660a9]
  • Fixed native_specialization_constant() API implementation. [41344ed7da50]
  • Moved Level Zero specific collectors to dynamic libraries loaded by request in sycl-trace tool to support the case when Level Zero driver is not available. [4c5250643b3c]
  • Added missing marray relational functions: any, all, bitselect; fixed scalar select relational function; aligned scalar abs integer function with SYCL 2020 and fixed math functions which take multi_ptr argument. [8b0b210a95db]
  • Added operator[] and element_type to swizzle vec to align with SYCL 2020. [9050d3325f2c]
  • Fixed buffer range in atomic_memory_order_acq_rel. [9fb49f757f04]
  • Started to throw for invalid global_work_size query. [61433888d4b5]
  • Fixed the Level Zero backend to report events as submitted, not running, until they are completed. [502e0fdd3f3c]
  • Allowed raw send ESIMD API to use non-standard types like sycl::half. [8022c6a4f6ec]
  • Started to throw for invalid info::kernel::num_args query to comply with SYCL 2020. [93e8ca3ed2eb]
  • Updated group_broadcast to support vec types. [09ae6434f55f]
  • Fixed reductions to avoid implicit atomic64 requirements. [e11b35835722]
  • Added a partial profiling workaround for acc devices because queue profiling is no longer supported for OpenCL version < 2.1 after recent changes regarding command submit profiling info. [097d21c33599]
  • Fixed program build API for the Level Zero backend to retain build-log when program build failed. [7d9f5ac5e5da]
  • Implemented missing std::hash specializations for local_accessor and host_accessor to comply with SYCL 2020. [919246bacb82]
  • Fixed global_work_size kernel query. [d666b95af56a]
  • Fixed incorrect sycl::vec<bool, N> constructor behavior. [45aa49bb3f06]
  • Fixed weak_object for host_accessor and stream. [225b3ecad28d]
  • Fixed incorrect behaviors in some operations using sycl::vec<bool, N>. [61f6ce4579b7]
  • Fixed integration footer for specialization_id. [dd96dfaf12da]
  • Fixed compilation break occurring when bfloat16 constructor is used in a kernel. [bc063ac7a9e1]
  • Fixed crash in subgroup info queries when running on OpenCL backend which doesn't support sub-groups. [2408035da9c9]
  • Added missing support for target::host_task specialised accessor constructor using mode_target Tag. [c01d8824b1aa]
  • Fixed identityless reductions with unwritten reducers. [cb15f33e861a]
  • Fixed dangling pointer issue in xpti tracing. [8a49660d4b44]
  • Fixed PI event leak in memcpy2d device-host fallback. [24ec33c39c85]
  • Fixed weak_object and owner_less for device objects. [b9b85136d492]
  • Added noexcept specifier to vec::byte_size to comply with SYCL 2020. [cb6c0580dc30]
  • Fixed undefined behaviour in vector printf specifier. [d190de65a4e1]
  • Fixed handler::fill so it works even if pattern is not supported by a backend natively. [7f49367d234b]
  • Fixed mechanism to throw exception when placeholder accessor passed to a command. (#10110) [93f477358d74]
  • Fixed empty zero-dimensional accessor access range. [b4d2b2c966b1]
  • Fixed incorrect write back in a case when a sycl::buffer is constructed with a const T* as host data. [d05818637be3][1b2b9ddc7f37]
  • Fixed empty accessor default constructor to not create a placeholder to comply with SYCL 2020. [e753fae9fad7]

Documentation

API/ABI breakages

  • Removed deprecated piclCreateProgramWithSource. [42f560977a0a]
  • Removed deprecated barrier API. [00533e0b1e7e]
  • Removed deprecated interop_task [aff7fbb287d0]
  • Removed deprecated sycl::group_local_memory. [1dc77a73b391]
  • Removed deprecated sycl::detail::bitcast [61f3650c28ed]
  • Removed deprecated piEnqueueNativeKernel [c3781336d0f1]
  • Removed deprecated backend enum values: level_zero, cuda, esimd_cpu, hip. [81900499327a]
  • Removed the workaround for release of auxiliary buffers. [fda9171fc218]
  • Removed old specialization constants extension and implementation. [e50ae05e28d3][833a9fe0a9cf]
  • Removed non-standard RT namespace from sycl namespace. [fd636ef3ab99]
  • Removed deprecated ESIMD APIs. [91e8364164c6]
  • Removed DISABLE_SYCL_INSTRUMENTATION_METADATA macro and _CODELOC* macro usage from API. [f4525e901cd6]
  • Removed non-standard RT namespace from sycl namespace. [b3e0428ffef4]
  • Removed getOSModuleHandle usage. [f73230d8a8ba]
  • Removed deprecated sycldevice triple support. [01d7fc097ec6]
  • Dropped support for sycl_ext_oneapi_extended_atomics extension. [17ea6d10f1d5]
  • Dropped support for sycl_ext_oneapi_group_algorithms extension. [8b9d2106af0d]
  • Removed unneeded backwards compatibility of make_queue and get_native. [c5f150a41b1d][5918e0c208a1]
  • Removed support for binaries generated for triples with -sycldevice environment component. [5fc53134fcfc]
  • Removed lambda parameter from if_architecture_is according to sycl_ext_oneapi_device_architecture. [71f745bcda7d]
  • Updated sycl::exception to be SYCL2020 compliant. [ce26ac08a383]
  • Replaced deprecated sycl::runtime_error with SYCL 2020 compliant sycl::exception in SYCL RT. [f08dc9473d40]
  • Renamed win_proxy_loader to pi_win_proxy_loader. [2648b7c5e1a4]
  • Promoted the return type changes of SYCL relational builtins that changed between SYCL 1.2.1 and SYCL 2020 out from the guard of SYCL2020_CONFORMANT_APIS. [35fe4a21e1cb]
  • Fixed get_pointer to return T* for target::device specialized accessor according to specification. [712cb4e4f75e]
  • Fixed max_work_item_sizes return type from id to range according to SYCL 2020. [8f09d3e603da]
  • Deprecated experimental set_kernel_properties API and use_double_grf/use_large_grf properties were removed. New API provided in the extension sycl_ext_intel_grf_size has to be used. [f363bb272b5a]

Known issues

  • Having MESA OpenCL implementation which provides no devices on a system may cause incorrect device discovery. As a workaround such an OpenCL implementation can be disabled by removing /etc/OpenCL/vendor/mesa.icd.
  • -fsycl-dead-args-optimization can't help eliminate offset of accessor even though it's created with no offset specified
  • SYCL 2020 barriers show worse performance than SYCL 1.2.1 do. [18c80faa]
  • When using fallback assert in separate compilation flow it requires explicit linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero. 7dfaf3bd
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows. [50628db1]
  • The support of accessor and local_accessor for ESIMD is still limited comparing to SYCL.
  • sycl_ext_oneapi_root_group implementation has the following limitations: max_num_work_group_sync query always returns 1 and the implemented barrier has work group scope since all work items in a root group are currently in the same work group.

March'23 release notes

Release notes for commit range ca54ea30..cb91c232

New features

SYCL Compiler

  • Added semi-dynamic SLM allocation - sycl::ext::intel::experimental::esimd::slm_allocator [e7674019]
  • Added support of compilation of the .cu files containing CUDA, and SYCL codes in the same translation unit [31608c24] [f5126d28]
  • Added support of amd-gpu-gfx1034 as an acceptable value for -fsycl-targets [5e86a41d]

SYCL Library

  • Implemented sycl_ext_oneapi_matrix extension using new unified interface [f4a9ef1b]
  • Partially implemented sycl_ext_codeplay_kernel_fusion extension for Linux: Intel CPU and GPU devices are fully supported, except specialization constants, streams, images and reductions [89e82e3f] [5d7e86d4] [ab0c46bd] [b5d5aada] [325bc4e2] [cf245f62]
  • Added default constructor to sycl::local_accessor [589824d7]
  • Added support of SYCL_PI_HIP_MAX_LOCAL_MEM_SIZE environment variable for HIP backend [92f6d688]
  • Aligned sycl::vec aliases with SYCL 2020 [1299e21c] [791ac603] [18f2cdab]
  • Partially implemented sycl_ext_intel_fpga_kernel_interface_properties extension [c5d6b89c]
  • Added support for sycl::ext::oneapi::experimental::if_architecture_is to ESIMD kernels [7a811191]
  • Implemented sycl::opencl::cl_* aliases [e98d732406cc]
  • Implemented annotated_arg headers from sycl_ext_oneapi_annotated_arg extension [1649fe46]
  • Added bfloat16 support to imf libdevice functions [378678a3] [e85b51fc] [25a3813b] [2b00cf95]
  • Added sycl::device::get_info<info::device::aspects> specialization [7869c17a]
  • Added rbegin(), rend(), crbegin() and crend() methods to sycl::accessor, sycl::host_accessor and sycl::local_accessor [baa3d739]
  • Added accessor_ptr alias to sycl::accessor and sycl::local_accessor [6b4cd9d0]
  • Implemented get_multi_ptr method for sycl::accessor and sycl::local_accessor [f4185c75]
  • Added size_type to sycl::accessor and sycl::local_accessor classes [0a17e16c]
  • Implemented accessors implicit conversions [51b7f781]
  • Implemented templated sycl::kernel_bundle::has_kernel functions [50e798f9]
  • Implemented sycl::kernel_bundle::get_kernel function [cf018f16]
  • Added support for basic atomic operations for HIP AMD backend [c3c5e923]
  • Added the ability to set specialization constants in sycl::compile [33cd6b1a]
  • Added the value_type and binary_operation member aliases and the dimensions value to the sycl::reducer class [8fdc68d4]
  • Implemented sycl::ext::intel::math::inv free function [a90d0fd9]
  • Added APIs that emulate the behaviors of CUDA SIMD intrinsic [b0bfdfa7]
  • Added overloads on all binary operators with scalars as the left operand to sycl::marray, and allowed half, float, double in && and || operators [a1787de5]
  • Added support of Intel(R) Xe Matrix Extensions with SIMD8 capability [8316a9c8]
  • Implemented device headers part of sycl::any_device_has and sycl::all_devices_have [81baa50b]
  • Added host_task in sycl::target enum class [a8401952]
  • Added support of sycl::marray to relational, common and some math functions [c3a40760] [c5c7ac2e] [d8715418]
  • Implemented sycl_ext_oneapi_annotated_ptr extension [3250db55]
  • Implemented sycl_ext_oneapi_cuda_tex_cache_read extension [5360825e650b]
  • Introduced an API to get thread ID and sub-device ID for ESIMD extension [dd6145e8]
  • Implemented kernel_bundle::native_specialization_constant() [4a412307]
  • Implemented SYCL 2020 reductions without a known or specified identity [648845fb] [1404ff1d]
  • Implemented basic functionality for the following group types: ballot_group, cluster_group, tangle_group, opportunistic_group as part of sycl_ext_oneapi_non_uniform_groups extension [86894205]
  • Implemented is_group trait and changed the definition of is_group_v to be in line with SYCL 2020 [10e07a38]
  • Implemented atomic_memory_scope_capabilities and atomic_memory_order_capabilities device queries for OpenCL and Level Zero backends [2816c14072f3] [b18e6ea83c62]
  • Implemented atomic_fence_scope_capabilities and atomic_fence_order_capabilities device queries for OpenCL and Level Zero backends [632946ac1cbf]
  • Implemented sycl::ext::intel::experimental::esimd::bfn function from ESIMD extension [25d0475f]
  • Made sycl::ext::oneapi::experimental::joint_reduce work with sycl::sub_group [cb91c232]

Documentation

Improvements

SYCL Compiler

  • Improved sycl::reqd_work_group_size for optional dimensions [b211d0df]
  • Enabled DPC++ build with llvm-mingw toolchain on Windows [1715c36b]
  • Deprecated usage of -sycl-std=1.2.1 [8dd6fdff]
  • Disabled force inlining of kernel call operator for FGPA [90e0e650]

SYCL Library

  • Added diagnostics that the joint_matrix API is only supported by the Intel and CUDA backends [ed596eca]
  • Deprecated sycl::ext::oneapi::leader(group), SYCL 2020 member function group::leader should be used instead [94434e78]
  • Deprecated get_access for sycl::host_accessor in accordance with SYCL 2020 [d145da8f]
  • Deprecated get_size() and get_max_statement_size() member functions in the sycl::stream class, added replacements: size() and get_work_item_buffer_size() [aab0ba71]
  • Deprecated legacy scalar and vector type aliases [e83d0ae0] [016d68b0]
  • Empty platforms are no longer returned [5e8ae211]
  • Started to throw an exception when selected device is not in context [262299d1]
  • Changed signature of sycl::host_accessor::get_pointer in accordance with SYCL 2020 [a7473fd2]
  • Started to ignore placeholder template parameter for accessors in accordance with SYCL 2020 [2b1d3205]
  • Started to throw the correct exception when passing unbound accessor to command [22cefabf]
  • Hid the members of sycl::reducer that are not mentioned in the SYCL 2020 specification and introduced the identity member function [505aa7d1]
  • Improved group_local_memory_for_overwrite (part of sycl_ext_oneapi_local_memory) by default initializing the returned memory, if needed [37a43f93]
  • Augmented sycl::has_known_identity to return true for std::complex and std::plus operators [2e733125]
  • Improved diagnostic when sycl::accessor isn't bound to sycl::handler [3c87f55e]
  • Added identities for arithmetic and integral operations on bool in accordance with the SYCL 2020 specification, added a check for the specification defined identities [091481d8]
  • Renamed raw_send{s}_{load/store} APIs in ESIMD extension [cb280eab]
  • Made sycl::reducer uncopyable and immovable in accordance with SYCL 2020 [7c7efeee]
  • Updated aspects list to be SYCL 2020 compliant [b9122b30]
  • Updated sycl::kernel_bundle::has_kernel(kernel_id, device) to return true if the kernel is valid for a given device in accordance with SYCL 2020 [a950450d]

Documentation

  • Added Explicit SIMD design and implementation notes [59dd3879]
  • Removed sycl_ext_oneapi_nd_range_reductions docs, implementation had been removed in the middle of 2022 [a46efa0a]
  • Clarified default device_global allocation behavior [15edb1bd]
  • Corrected the specified return type for the UUID query [77d25cd8]
  • Clarified the usage of Intel GPU drivers in WSL under the docker [1f39511c]

Bug fixes

SYCL Compiler

  • Fixed -f[no-]sycl-rdc option in clang-cl [e39d0c18]
  • Fixed emission of FPGA annotations [8199de8c]
  • Cleaned up Windows defaultlib linking behaviors [aaae3348]
  • Fixed an issue where the dimensions of the work-group hint were not correctly reversed [ba51bb68]
  • Fixed a crash during generation of debug information with the option -ffile-prefix-map [723b80da]

SYCL Library

  • Fixed operator- for constant accessor iterator [890aa072]
  • Fixed the case when sub-device free memory may exceed root-device free memory [9dc14a29]
  • Fixed sycl::handler::require() to accept a non-placeholder argument [abdc0fe3]
  • Fixed sycl::ext::oneapi::experimental::joint_reduce, it missed the case when work-group size is bigger than size of input data [23b68dbd]
  • Added missing include of sycl/builtins.hpp to sycl/ext/intel/math.hpp [efe1074e]
  • Fixed host compilation issue for lsc_atomic_update for store operation in ESIMD extension [fdf8ac77]
  • Fixed event profiling for sycl::info::event_profiling::command_submit in Level Zero and other backends [71d77972]
  • Fixed sycl::group::get_local_linear_id(), now it uses the correct range to linearize the local ID [c3849635]
  • Fixed compilation issue when sycl::stream::size was called from device code [368b6605]
  • Fixed sycl::host_accessor zero dimension constructors [23f107d4]
  • Fixed atomic memory orders in reduction implementation [2aef0c94]
  • Fixed accessor subscript ambiguity and reference type [8c244de9]
  • Fixed an issue with missing explicit instantiations in sycl/properties/queue_properties.hpp [2864eac4]
  • Fixed return type of sycl::reducer::combine [44a26c1e]
  • Fixed an error when a scalar offset is provided as a parameter to the API [035ef2b2]
  • Fixed return type of identity-less reduction as it used an unexpected value for the deprecated placeholder template parameter in accessor [c64f88ec]
  • Fixed variadic sycl::marray constructor by accepting values that are indirectly convertible to the element type of the marray [becc391a]
  • Fixed sycl::atomic_ref constructor to fix operator ambiguity [00c60624]
  • Fixed handling of host-side memory in 2D memory operations [2dcc5817]
  • Fixed constexpr initialization of sycl::vec for sycl::half [ba0ea790]
  • Fixed default sycl::host_accessor iteration methods [f8ba0e48]
  • Fixed lsc_prefetch_2d in ESIMD extension [10a9c14f]
  • Fixed over-allocation in specialization constants [d58eb783]
  • Fixed sycl::handler::get_specialization_constant segmentation fault that happened in certain cases likely due to strict aliasing violations [86c08b3c]
  • Fixed a regression in ESIMD's atomic_update [1396da2e]
  • Fixed return type of the sycl::accessor::get_pointer and sycl::local_accessor::get_pointer [3fdbfeb6]

Known issues

  • Having MESA OpenCL implementation which provides no devices on a system may cause incorrect device discovery. As a workaround such an OpenCL implementation can be disabled by removing /etc/OpenCL/vendor/mesa.icd.
  • -fsycl-dead-args-optimization can't help eliminate offset of accessor even though it's created with no offset specified
  • SYCL 2020 barriers show worse performance than SYCL 1.2.1 do. [18c80faa]
  • When using fallback assert in separate compilation flow it requires explicit linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero. 7dfaf3bd
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows. [50628db1]

December'22 release notes

Release notes for commit range 0f579bae..6977f1ac

New features

SYCL Compiler

  • Added support for per-object device code compilation under the option -fno-sycl-rdc. This improves compiler performance and reduces memory usage, but can only be used if there are no cross-object dependencies. [f884993d]
  • Added support for per-aspect device code split mode. [9a2c4fe5]
  • Extended support for the large GRF mode to non-ESIMD kernels. [9994934b] [ab2a42c6]
  • Implemented the sycl_ext_oneapi_device_architecture extension. [0e32a28d] [b59d93cc] [5bd5c871] [e5de913f]
  • Implemented the sycl_ext_oneapi_kernel_properties experimental extension. [332e4ee0] [27454de6] [70ee3d56] [430c7221]
  • Added support for generic address space atomic built-ins to CUDA libclc. [d6a8fd1f]

SYCL Library

  • Implemented accessor member functions swap, byte_size, max_size and empty. [f1f907a2]
  • Implemented SYCL 2020 default accessor constructor. [04928f9f]
  • Implemented SYCL 2020 accessor iterators. [5b9fd3ca] [c7b1a000]
  • Changed value_type of read-only accessors to const in accordance with SYCL 2020. [227614ca]
  • Implemented SYCL 2020 multi_ptr and address_space_cast. [8700b768] [483984a9] [4a9e9a0e]
  • Implemented SYCL 2020 has_extension free functions. [7f1a6efd]
  • Implemented SYCL 2020 aspect_selector. [c0a4a565]
  • Implemented new SYCL 2020 style FPGA selectors. [04176516]
  • Implemented SYCL 2020 default async_handler behavior. [cd93d8f6]
  • Implemented SYCL 2020 is_compatible free function. [67f6bba2]
  • Implemented queue shortcut functions with placeholder accessors. [5ee066e4]
  • Added support for creating a kernel bundle with descendent devices of the passed context's members. [a7827795]
  • Implemented non-blocking destruction and deferred release of memory objects without attached host memory. [894ce256]
  • Implemented the sycl_ext_oneapi_queue_priority extension. [cdb09dca]
  • Implemented the sycl_ext_oneapi_user_defined_reductions extension. [8311d790]
  • Implemented the sycl_ext_oneapi_queue_empty extension proposal. [c4932957]
  • Implemented the sycl_ext_oneapi_weak_object extension. [d9484274] [9297f63e]
  • Implemented the sycl_ext_intel_cslice extension. The old behavior that exposed compute slices as sub-sub-devices is now deprecated. For compatibility purposes, it can be brought back via the SYCL_PI_LEVEL_ZERO_EXPOSE_CSLICE_IN_AFFINITY_PARTITIONING environment varible. [5995c618]
  • Implemented the sycl_ext_intel_queue_index extension. [d2ec9645] [7179e830]
  • Implemented the sycl_ext_oneapi_memcpy2d extension. [516d4112]
  • Implemented device ID, memory clock rate and bus width information queries from the sycl_ext_intel_device_info extension. [1d993446] [4f7787c8]
  • Implemented ext::oneapi::experimental::radix_sorter from the sycl_ext_oneapi_group_sort extension proposal. [86ba1809]
  • Implemented a new unified interface for the sycl_ext_oneapi_matrix extension for CUDA. [166bbc36]
  • Added support for sorting over sub-groups. [168767c6]
  • Added C++ API wrappers for the Intel math functions ceil, floor, rint, sqrt, rsqrt and trunc. [1b7582bb]
  • Implemented a SYCL device library for bfloat16 Intel math function utilities. [fc136d69]
  • Added support for range reductions with any number of reduction variables. [572bc503]
  • Added support for reductions with kernels accepting item. [5d5e9f4e]
  • Enabled sub-group masks for 64-bit subgroups. [10d50ed3]
  • Implemented the new non-experimental API for DPAS. [55bf1a0d] [1e7a8eaa]
  • Added 8/16-bit type support to lsc_block_load and lsc_block_store ESIMD API. [f9d8059d]
  • Implemented atomic operation support in the ESIMD emulator. [a6a0dea7]
  • Added various trivial utility functions for the half type. [b4ce7c08]
  • Added type cast functions between half and float/integer types to libdevice. [599b1b96]
  • Implemented the ONEAPI_DEVICE_SELECTOR environment variable that, in addition to supporting SYCL_DEVICE_FILTER syntax, allows to expose GPU sub-devices as SYCL root devices and supports negative filters. SYCL_DEVICE_FILTER is now deprecated. [28d0cd3a] [b21e74ef] [77b6f34f] [6bd5f9ce] [6aefd631]
  • Added the SYCL_PI_LEVEL_ZERO_SINGLE_ROOT_DEVICE_BUFFER_MIGRATION enviornment variable. [bd03e0d3]

Documentation

Improvements

SYCL Compiler

  • Added the InferAddressSpaces pass to the SPIR/SPIR-V compilation pipeline, reducing the size of the generated device code. [a3ae0dd1]
  • Redesigned pointer handling so that it no longer decomposes kernel argument types containing pointers. [3916d3b6] [d55e9c28] [9b025068]
  • Kernel lambda operator is now always inlined in the device code entry point unless -O0 is used. [b91b7327] [2359d948]
  • Improved entry point handling in the sycl-post-link tool. [53d9c7b6]
  • The reqd_work_group_size attribute now works with 1, 2 or 3 operands. [4ff42c3c]
  • Enabled using -fcf-protection option with -fsycl, which results in it being applied only to host code compilation and producing a warning. [b6f61f62]
  • Linux based compiler driver on Windows now pulls in the sycld debug library when msvcrtd is specified as a dependent library. [ebf6c597]
  • Added /Zc:__cplusplus as a default option during host compilation with MSVC. [e7ed8607]
  • Improved the ESIMDOptimizeVecArgCallConv optimization pass to cover more IR patterns. [49264549]
  • Added support for more types in ESIMD lsc functions. [d9e40ecd]
  • Added error diagnostics for using sycl::ext::oneapi::experimental::annotated_arg/ptr as a nested type. [321c733c]
  • The status of bfloat16 support was changed from experimental to supported. [7b47ebb6]

SYCL Library

  • Updated online_compiler with Gen12 GPU support. [adfb1c10]
  • get_kernel_bundle and has_kernel_bundle now check that the kernels are compatible with the devices. [91b15156]
  • Waiting for an event associated with a kernel that uses a stream now also waits for the stream to be flushed. [1db0e81f]
  • Added the requested device type to the message of the exception thrown when no such devices are found. [6b83ad7c]
  • Optimized operator[] of host_accessor. [01e60f76]
  • Improved reduction performance on discrete GPUs. [99bdc823]
  • Added invoke_simd support for functions with void return type. [3fd08509]
  • The Level Zero plugin now creates every event as host-visible by default. [f3d245d6]
  • Added Level Zero plugin support for global work sizes greater than UINT32_MAX as long as they are divisible by some legal work-group size and the resulting quotient does not exceed UINT32_MAX. [62dd13dd]
  • Improved native Level Zero event handling in the immediate command list mode by removing excessive status queries. [179ffa16]
  • Removed an uninitialized buffer migration copy in the Level Zero plugin [b0c34046]
  • Implemented an optimization that reuses discarded Level Zero events in the plugin. [b1533c5e]
  • The host device is now inaccessible. [5b13d5b6]
  • Removed deprecated make_queue API. [9fc1d684]
  • Deprecated group::get_global_range(). [95338719]

Documentation

Bug fixes

SYCL Compiler

  • Fixed a crash when attempting to compile code that uses a function object without a defined call operator as a kernel. [490ee550]
  • Fixed a crash that occurred during compilation of device code with a captured structured binding. [0e455c91]
  • Fixed the work_group_size_hint attribute not being applicable to lambda functions using non-conforming syntax. [c2a0db08]
  • Fixed integration header parameter kind information for annotated types. [b8f35cf3]
  • Fixed an issue with offload dependencies when using -fsycl-force-target. [54777c00]
  • Fixed debug information generation when an integration footer is present. [83febf94]
  • Fixed a __builtin_printf related error when compiling device code with _GLIBCXX_ASSERTIONS=1. [d8fd9bc8]
  • Fixed a compiler error that occurred during archive generation when using -fsycl-link for FPGA. [3159db53]
  • Fixed memory corruption caused by the ESIMDOptimizeVecArgCallConv pass. [86f709c3]
  • Fixed a crash during ESIMD intrinsic generation. [20b1bea5]
  • Fixed libclc function mangling. [e6c4c154] [f32d34f3]
  • Fixed an issue with specifying the offload architecture for AMD and NVIDIA targets. [41898580]
  • Fixed incorrectly issued warnings about CUDA or HIP targets missing from a linked library. [72d9b052]

SYCL Library

  • Fixed an issue where the in-order queue property was not respected when submitting USM commands and host tasks. [067d3b39]
  • Fixed a memory leak when enqueueing a barrier to a discard_events queue. [968f9e76]
  • Fixed a memory leak related to submitting host tasks without memory object dependencies. [c44050a9]
  • Fixed an invalid event error when handling cross-queue no-op dependencies. [6c9a380d]
  • Fixed an error when setting a specialization constant in a command group with no kernel. [b333cee9]
  • Fixed an issue where submitting a kernel that explicitly depends on a host task was a blocking call that waited for the host task. [c44050a9]
  • Removed noexcept from some of usm_allocator member functions to align with the specification. [7b026976]
  • Fixed ext::intel::experimental::atomic_update with the fcmpwr operation. [52923e6d]
  • Fixed memory leak issues when constructing a SYCL kernel/kernel_bundle using interoperability. [b083e054] [a32021ba] [3c22764f]
  • Fixed an error where the native handle returned by get_native from a default constructed event was unusable. [72021733]
  • Fixed an issue where reinterpreting a buffer to a const type changed the corresponding buffer_allocator type to const. [3aabd264]
  • Fixed handler::set_arg with local_accessor. [ef792c67]
  • Added the missing default template argument for sycl::info::device::max_work_item_sizes. [3a4e797d]
  • Fixed an issue where some aspects could be incorrectly reported as unsupported by a device. [f90d2b4c]
  • Fixed return type of scalar versions of relational functions. The fix requires defining SYCL2020_CONFORMANT_APIS macro. [45d516cc]
  • Fixed an issue where device code cache was not used if the compilation was triggered by different paths. [a0254c98] [1c77f9a7]
  • Fixed a use-after-move bug when caching device code built for multiple devices. [97c0c997]
  • Removed the unintended requirement of fp64 support from stream and ESIMD float fmod implementations. [a67807ac] [cc189044]
  • Fixed several complex math operations failing on devices that don't support fp64. [6b24fdc6]
  • Aligned host side float-to-half mantissa rounding with device side. [4cbd459a]
  • Fixed float-to-half conversion of the half minimum subnormal value on host. [514708b7]
  • Fixed marray math function implementation. [73a992b3]
  • Fixed an out-of-bounds write in the group operations implementation. [0fa7542b]
  • Fixed a reduction performance regression caused by using the wrong implementation for the float type. [97725f19]
  • Fixed header deprecation warnings to work properly on Linux. [2cefad19]
  • Fixed deprecation of SYCL 1.2.1 device selectors. [a6222ba6]
  • Fixed multiple issues in GDB xmethods scripts. [e49aa083]
  • Fixed an issue with sycl-prof JSON output. [37a74c7b]
  • Fixed compilation errors on Windows when using the ESIMD API. [3aa48db2] [77e92cea]
  • Fixed invalid calculation in the ESIMD tanh function. [cda6680e]
  • Fixed kernel_bundle errors when using ESIMD emulator devices. [9baa9d92]
  • Fixed an issue where ESIMD emulator was picked by the default selector even in the presence of other devices. [44d7926c]
  • Fixed an error when querying an ESIMD emulator device for sub-group sizes. [acca6089]
  • Fixed invalid behavior of the maximum sub-group size query on some OpenCL systems. [5998d7c7]
  • Fixed an issue where the OpenCL plugin checked whether a program is supported on a device by looking up platform version/extensions rather than device ones. [9f892472]
  • Fixed the result of the free device memory query with the Level Zero backend. [c191fb0a]
  • Fixed an issue with ext_oneapi_barrier not working when using the Level Zero backend. [1f8d90fc]
  • Fixed a hang after submitting a barrier to a Level Zero in-order queue. [dd5a191c]
  • Fixed an issue that occurred when submitting a barrier to a Level Zero queue with no prior submissions. [5c9e5434]
  • Fixed a memory leak when tracking indirect access in the Level Zero plugin. [1b79491d]
  • Fixed an invalid read issue that occurred during Level Zero event release. [c49eeda8]
  • Fixed a synchronization issue when using device scope Level Zero events. [9811ef24]
  • Fixed an issue that occurred when using get_native on a newly constructed Level Zero queue. [5d0d4ef9]
  • Fixed a segmentation fault related to events recycling in immediate command list mode in the Level Zero plugin. [a3e93e07] [5b021a22]
  • Fixed an issue where an invalid maximum of compute units was reported for Level Zero sub-sub-devices. [b9f49196]
  • Fixed a segmentation fault when using Level Zero sub-sub-devices with the immediate command lists mode. [ed3d35c5]
  • Reverted the Level Zero plugin change that preferred using copy engine for memory read/write operations due to functional regressions. [44aa363a]
  • Added the missing fp16 case of FMulKHR libclc function. [4372915c]
  • Fixed several bugs in the barrier implementation in the CUDA and HIP plugins. [1c3d5987] [ce7c594f]
  • Fixed get_native not working for CUDA devices. [998fd91e]
  • Fixed a crash when submitting a kernel with a range of 0 to a CUDA device. [a3958865]
  • Fixed an issue where make_device produced duplicate CUDA devices. [75302c53]
  • Fixed an issue where the HIP backend always reported the fp64 aspect as missing. [cd832bff]

Known issues

  • Having MESA OpenCL implementation which provides no devices on a system may cause incorrect device discovery. As a workaround such an OpenCL implementation can be disabled by removing /etc/OpenCL/vendor/mesa.icd.
  • Compilation may fail on Windows in debug mode if a kernel uses std::array. This happens because debug version of std::array in Microsoft STL C++ headers calls functions that are illegal for the device code. As a workaround the following can be done:
    1. Dump compiler pipeline execution strings by passing -### option to the compiler. The compiler will print the internal execution strings of compilation tools. The actual compilation will not happen.
    2. Modify the (usually) first execution string (it should have -fsycl-is-device option) by adding -D_CONTAINER_DEBUG_LEVEL=0 -D_ITERATOR_DEBUG_LEVEL=0 options to the end of the string. Execute all string one by one.
  • -fsycl-dead-args-optimization can't help eliminate offset of accessor even though it's created with no offset specified
  • SYCL 2020 barriers show worse performance than SYCL 1.2.1 do. [18c80faa]
  • When using fallback assert in separate compilation flow it requires explicit linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero. 7dfaf3bd
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows. [50628db1]

September'22 release notes

Release notes for commit range 4043dda3..0f579bae

New features

SYCL Compiler

  • Added ability to enforce stateless memory accesses for ESIMD. [18111623]
  • Added support for -fsycl-force-target compiler option. [1d95f2ec]
  • Added support for [[intel::max_reinvocation_delay]] loop attribute. [90fa5bb0]
  • Added support for -fsycl-huge-device-code compiler option, which allows linking object files larger than 2GB. [f963062c]
  • Added support for compiling .cu files with SYCL compiler. [e76ad72f]
  • Added support for assert on HIP backend. [ade1870a]
  • Enabled CXX standard library functions for CUDA backend. [1fe92c54]
  • Implemented group collective built-in functions for more integral types. [d4933b6f]

SYCL Library

  • Implemented SYCL 2020 callable device selectors. [64f0db7a]
  • Implemented SYCL 2020 standalone device selectors. [bfc7e984]
  • Added SYCL 2020 property interfaces for local_accessor, usm_allocator, accessor and host_accessor classes. [1136b403] [da7dcf82]
  • Added support for fpga_simulator_selector. [9bef890d]
  • Added support for local_accessor. Deprecated target::local. [e4423ef4]
  • Added support for querying free device memory on Level Zero backend. [0eeef2b3]
  • Added support for querying free device memory on CUDA and HIP backends. [436f0d89]
  • Implemented bfloat16 conversions from/to float for host. [2a383f1c]
  • Added support for ext::oneapi::property::queue::discard_events to Level Zero PI plugin. [13721204]
  • Added lsc_atomic support on ESIMD emulator. [0c051a89]
  • Added dpas support on ESIMD emulator. [3d506a34]
  • Added C++ API for imf libdevice built-ins. [830916a3]
  • Implemented make_queue for CUDA backend. [89460e81]
  • Implemented has_native_event and make_event for CUDA backend. [74369c84]
  • Added support of CUDA XPTI tracing. [0cd04144]
  • Introduced predicates for ESIMD lsc_block_store/load. [f44edce3]
  • Added experimental set_kernel_properties API and use_double_grf property for ESIMD. [9a55da53]
  • Added "eager initialization" mode to Level Zero PI plugin. It might result in an unnecessary work done by the plugin, but ensures the fastest possible execution on hot and reportable paths. [c1459598]
  • Added full support of element wise operations on joint_matrix on CUDA backend including bfloat16 support. [0a1d751b]
  • Implemented group::get_linear_id(int) method [6e83c127]

Documentation

Improvements

SYCL Library

  • Ensured that a correct errc thrown for an unassociated placeholder accessor. [4f9935ad]
  • Removed dependency on OpenCL ICD Loader from the runtime. [90e8b5ef]
  • Added support for ZEBIN format to persistent caching mechanism. [34dcf83d]
  • Added identification mechanism for binaries in newer ZEBIN format. [f4dee549]
  • Switched to use struct information descriptors in accordance with SYCL 2020. Removed some deprecated information queries. [b3cbda57]
  • Updated kernel_device_specific::max_sub_group_size query to match SYCL 2020 spec. Deprecated the old variant. [7842d056]
  • Deprecated SYCL 1.2.1 device selectors. [c058380f]
  • Improved error messages reported for unsupported device partitioning. [1c9ddbaa]
  • Made device and platform default to default_selector_v. [b32dd41a]
  • Deprecated address_space::constant_space. [351b123a]
  • Marked sycl::exception::has_context as noexcept. [ad923c9a]
  • Improved range reductions performance on CPU. [3323da65]
  • Made sycl::exception nothrow copy constructible. [289e33da]
  • Marked has_property methods as noexcept. [417b5a29]
  • Improved sycl::event::get_profiling_info exception message when event is default constructed. [2e86cd459]
  • Added a diagnostic (in form of static_assert) about kernel lambda size mismatch between host and device. [d278c671] [ec179b7a] [f417a88e]
  • Updated pipes class to throw exceptions if used on host. [eab29696]
  • Updated ESIMD Emulator PI plugin to report support for cl_khr_fp64 extension. [398571a5]
  • Updated Level Zero plugin to prefer copy engine for memory read/write operations. [65c3ea29]
  • Optimized some memory transfers. [92d35cd1]
  • Enabled event caching in Level Zero PI plugin. [a41b33c3]
  • Optimized some reductions with parallel_for accepting sycl::range for discrete GPUs. [c22a5d3f]
  • Improved performance of event synchronization on CUDA backend. [c4f326aa]
  • Added ability to use descendent devices of context members within that context. Not supported with OpenCL backend yet. [a0c8c503] [78a483c7]
  • Added support for querying atomic64 device capability with HIP backend. [cb190fc2]
  • Enabled FTZ operations for CUDA/PTX backend via -fcuda-flush-denormals-to-zero. [e8e7ae83]
  • Improved error message about incorrect kernel argument types with CUDA backend. [2542e6a8]
  • Limited allowed argument types for rol/ror ESIMD functions to better represent HW capabilities. [b05f256c]
  • Implemented mem_advise reset and managed memory checks for CUDA backend. [fe18839c]
  • Added concurrent memory check to mem_advise on CUDA backend. [33746d8d]
  • Enabled multiple HIP streams per SYCL queue. [e0c40a9f]
  • Implemented lazy mechanism of setting context for default-constructed events. [ed92c4ca]
  • Improved performance for multi-dimensional accessors with multiple accesses in a kernel. [7c58b9a6]

SYCL Compiler

  • Increased max _Bitint size to 4096 for FPGA target. [db5f72a8] [3f06cad0]
  • Removed deprecation message for [[intel::disable_loop_pipelining]] attribute. [07201f56]
  • Allowed __builtin_assume_aligned to be called from device code. [24937eac]
  • Improved link step performance when per_kernel device code split is used. [84de9d6d]
  • Added support for SYCL_EXTERNAL on device_global variables. [8b958f67]
  • Updated __builtin_intel_fpga_mem to accept more parameters. [231338dc]
  • Updated ivdep attribute to allow safelen = 0. [558b3ba4]
  • Improved linking with sycl.lib on Windows. [404d2816]
  • Implemented more diagnostics about incorrect device_global usages. [12657218]
  • Improved library resolution for libsycl.so. [4ce19d69]
  • Improved diagnostics when linking with mismatched objects. [0e0202ee]
  • Added a warning for floating-point size changes after implicit conversions. [e4f5d55f]
  • Made invoke_simd convert its argument to appropriate types. [038764fd]

Documentation

  • Removed explicit cl namespace references. [433ea5c7]
  • Added a short guideline on using CMake with SYCL compiler. [fa603c3e]

Bug fixes

SYCL Library

  • Fixed a compilation issue where it wasn't possible to pass an initializer list for dependency events vector in queue shortcuts with offset parameter. [f4f83d95]
  • Fixed sycl::get_pointer_device throwing an exception when it passed a descendent device (sub-device) instead of a root device. [26d5d98b]
  • Fixed memory leak happening when kernel bundles are linked. [980677d9]
  • Fixed USM free throwing an exception when it passed a context created for a descendent device. [c49d4944]
  • Fixed accessor's CTAD for g++ host compiler. [57aabe7e]
  • Fixed a compilation issue when using multi-dimensional accessor's subscript operator. [22e3fc56]
  • Fixed "definition with the same mangled name" error happening when used multiple buffer reductions in a kernel. [a0a4d721]
  • Fixed a compilation issue with SYCL math built-ins when GCC < 11.1 is used as a host compiler. [c786894f]
  • Fixed a compilation issue with SYCL math built-ins (such as sycl::modf, for example) not accepting pointers to half. [e2861665]
  • Fixed an issues with reductions when MSVC is used as host compiler. [94c4b80a]
  • Fixed a compilation issue when fully specialized sycl::span is initialized from an array. [2b50820b]
  • Fixed a crash in Level Zero PI plugins caused by specialization constants not being used on device side, but present in a program. [9500875f]
  • Fixed event leak in Level Zero plugin. [6d04aa64]
  • Fixed an issue with sub-sub-devices in Level Zero plugin. [4b1b01bd]
  • Fixed an issue with incorrect half conversion on ESIMD emulator. [6143e55a]
  • Fixed a compilation issue with abs ESIMD function. [c72a85dd]
  • Fixed some warning coming out of SYCL headers when compiled in C++20 mode. [12ac4c36]
  • Fixed a compilation issue when using multiple bitwise shift operations in ESIMD. [40d08c23]
  • Fixed a crash in Level Zero PI plugin which occurs when the runtime tries to reset a command list which does not have a synchronization fence associated with it. [a61ac7a0]
  • Fixed a performance issue with excessive streams synchronization on CUDA backend. [5352b423]
  • Fixed a compilation issue with sycl::get_native<sycl::backend::ext_oneapi_cuda>(sycl::device) free function (#6653). [4d69c297]
  • Fixed synchronization issue for explicit dependencies (depends_on usage) which is blocked by host task or host accessor. [346a6c53]
  • Fixed an issue in Level Zero plugin which could cause barriers to not be correctly applied for an entire queue. [d01371b3]
  • Fixed a synchronization issue on CUDA backend. [e848c15f]
  • Fixed an issue with a context not properly retained with even interop on CUDA backend. [2baf1de5]
  • Fixed accessor so gdb can parse its template parameters correctly. [372cc948]
  • Fixed uses of common macro names in the implementation's header files. [e87adfd2]
  • Fixed a performance regression in Level Zero backend related to command list. [8a4777d0]

SYCL Compiler

  • Fixed cleanup of temporary files produced by unbundling archives. [7ef4b0a3]
  • Fixed optimizing out device_global variables with internal linkage. [5b2cfe21]
  • Fixed an issue when compiling and linking with different optimization levels could cause runtime errors. [0cc7540e]
  • Fixed description of -f[no-]sycl-unnamed-lambda compiler option. [616ecf75]
  • Fixed an issue when building SYCL programs in Debug mode with Windows-Clang.cmake. [084f34c1]
  • Fixed CUDA fat-binaries filename extensions in SYCL toolchain (.o -> .cubin). [2c217cfd]
  • Fixed an issue causing incorrect conversions involving unsigned types in ESIMD. [2a1151b8]
  • Fixed a crash in applications containing a mix of unnamed ESIMD and non-ESIMD kernels. [3d54d7ef]
  • Fixed type casting of image coordinates in built-ins implementation for CUDA backend. [338b4edc]
  • Fixed a linking issue when targeting AMD/HIP. [9829897a]
  • Fixed an issue when op[] was called with typedef argument under gdb. [b7bfe391]

API/ABI breakages

  • Removed deprecated kernel::get_work_group_info [94883470]
  • Removed deprecated get_native class method [282e7744]
  • Removed support for intel::fpga_pipeline attribute [6fe90d20]
  • Added MAJOR_VERSION to the name of the SYCL library on Windows. [c8427d6d]
  • Removed sycl::program class. [85a6833d]
  • Removed ext::oneapi::reduction. [7fa607fb]
  • Removed deprecated address_space enum values. [351b123a]
  • Removed event::get method. [da296ba3]
  • Removed using namespace experimental inside ext::intel. [7ddd8a97]
  • Made intel-specific device info descriptors namespace qualified. [0f4a0f3b]
  • Removed deprecated make_queue API. [26b7762a]
  • Aligned return types of sycl::get_native and interop::get_native_mem functions to be conformance with SYCL 2020 spec. [78275902]
  • Aligned sycl::buffer_allocator interface with SYCL 2020 spec. [78275902]
  • Removed cl namespace from sycl/sycl.hpp header. [3d2b25e1]
  • Dropped support for compiling SYCL in less than C++17 mode. [43e713c3]
  • Many other ABI-breaking changes resulting from internal refactoring

Known issues

  • This release is not backwards compatible with previous releases, which means that existing SYCL applications won't work with the newer runtime without re-compilation.
  • Having MESA OpenCL implementation which provides no devices on a system may cause incorrect device discovery. As a workaround such an OpenCL implementation can be disabled by removing /etc/OpenCL/vendor/mesa.icd.
  • Compilation may fail on Windows in debug mode if a kernel uses std::array. This happens because debug version of std::array in Microsoft STL C++ headers calls functions that are illegal for the device code. As a workaround the following can be done:
    1. Dump compiler pipeline execution strings by passing -### option to the compiler. The compiler will print the internal execution strings of compilation tools. The actual compilation will not happen.
    2. Modify the (usually) first execution string (it should have -fsycl-is-device option) by adding -D_CONTAINER_DEBUG_LEVEL=0 -D_ITERATOR_DEBUG_LEVEL=0 options to the end of the string. Execute all string one by one.
  • -fsycl-dead-args-optimization can't help eliminate offset of accessor even though it's created with no offset specified
  • SYCL 2020 barriers show worse performance than SYCL 1.2.1 do. [18c80faa]
  • When using fallback assert in separate compilation flow it requires explicit linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero. 7dfaf3bd
  • On the following scenario on Level Zero backend:
    1. Kernel A, which uses buffer A, is submitted to queue A.
    2. Kernel B, which uses buffer B, is submitted to queue B.
    3. queueA.wait().
    4. queueB.wait(). DPCPP runtime used to treat unmap/write commands for buffer A/B as host dependencies (i.e. they were waited for prior to enqueueing any command that's dependent on them). This allowed Level Zero plugin to detect that each queue is idle on steps 1/2 and submit the command list right away. This is no longer the case since we started passing these dependencies in an event waitlist and Level Zero plugin attempts to batch these commands, so the execution of kernel B starts only on step 4. The workaround restores the old behavior in this case until this is resolved. [2023e10d] [6c137f87]
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows. [50628db1]

June'22 release notes

Release notes for commit range f34ba2c..4043dda

New features

SYCL Compiler

  • Added -fcuda-prec-sqrt frontend compiler option which enables higher presision version of sqrt in the device code [ebf9ea8]
  • Added support for local memory accessors for the HIP backend. [58508ba]
  • Added initial support of -lname processing when searching for fat static libraries. [35e32d8] [a33f9c8]
  • Added -fsycl-fp32-prec-sqrt flag which enables correctly rounded sycl::sqrt. [5c8b7e7]
  • Added support for [[intel::loop_count()]] attribute. [c536e76]
  • Added support for passing driver options to JIT compiler and linker. [1c93bfe]
  • Added default argument support for work_group_size_hint attribute. [0cff80e]
  • Added support for float and double exchange and compare exchange atomic operations in CUDA libclc. [1d84c99]
  • Added --ffast-math support for CUDA libclc. [0f0c5d1]
  • Added support for software atomics (except for the ones using system scope) for lower sm versions of CUDA architecture. Enabled SYCL_USE_NATIVE_FP_ATOMICS by default. [7bc8447]
  • Added support for the global offset for AMDGPU. [2dc3c06]
  • Added support for asynchronous barrier for CUDA backend sm 80+. [6770421]
  • Added -f[no-]sycl-device-lib-jit-link option to control JIT linking of SYCL device libraries. [dfb37a8] [c946286]
  • Added support for the new FPGA attribute [[intel::fpga_pipeline(N)]] for loop pipelining. [92aadf3]
  • Added assert support for Windows NVPTX. [f29b498]
  • Added support for sycl_ext_oneapi_properties extension. [87f60f6][1984e74][a2583ec][cdf561a][d2982c6][35c2e00]

SYCL Library

  • Added support for Nvidia MMA for bf16, mixed precision int ((u)int8/int32), and mixed precision float (half/float). [5373362]
  • Added a mode for the Level Zero plugin where only last command in each batch yields a host-visible event. Enabled this mode by default. [c6b7b8e]
  • Added an option to query for atomic scope capabilities for the CUDA backend. Updated returns for atomics memory order capabilties. [43a4192]
  • Added support for an experimental Level Zero API for host pointer import into USM. The feature can be enabled using SYCL_USM_HOSTPTR_IMPORT environment variable. [844d7b6]
  • Added support for the wi_element for bf16 type. [9f2b7bd]
  • Added complex support for the reduce and scan group algorithms. [90a4dc7]
  • Added support for SYCL 2020 methods in the group class. [73d59ce]
  • Added SYCL_RT_WARNING_LEVEL environment variable which allows to control amount of warnings and performance hints the runtime library may print. [2741010]
  • Added tanh (for floats/halfs) and exp2 (for halfs) native definitions for CUDA backend. [250c498]
  • Added bf16 builtins for fma, fmin, fmax and fmax on CUDA backend. [62651dd]
  • Added support for USM buffer location properties which allows to specify at what memory location the device usm allocation should be in. [12c988a]
  • Added support for buffer_location property to the sycl::buffer. [9808525]
  • Added single_task support for ESIMD_EMULATOR backend. [2331160]
  • Added support for SVM 1,2,4-elements gather/scatter for ESIMD. [e200720]
  • Added support for bf16 builtins operating on storage types for CUDA backend. [413a9ef]
  • Added support for backend_version device property for CUDA backend. [4b1a4bc]
  • Added support for round-robin submissions to multiple compute CCS for the Level Zero backend. Disabled by default, can be controlled using SYCL_PI_LEVEL_ZERO_USE_COMPUTE_ENGINE. [a836c87]
  • Added support for buffer migration for contexts with multiple devices in the Level Zero plugin. [7baf152]
  • Added mode where the Level Zero plugin uses immediate command-lists instead of standard command-lists. This mode is disabled by default, can be enabled using SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS environment variable. [b9cb1d1]
  • Added support for sycl::get_native(sycl::buffer) for OpenCL and CUDA backends. [8b3c8c4]
  • Added reduction overloads accepting span. [863383b]
  • Added LSC support for ESIMD_EMULATOR backend. [b78bf00]
  • Added half type support for __esimd_convertvector_to/from. [0bfffd6]
  • Added buffer_allocator SYCL 2020 conformant variant. [53430c8]
  • Added support for the USM buffer location property in malloc_shared. [6e89821] [9f61c8e][8c4d9a5]
  • Added support for the USM buffer location property in malloc_host. [2c7caab]
  • Added experimental context and device interoperability support for CUDA. [f0df89a]
  • Added support for memory intrinsics for the ESIMD_EMULATOR plugin. [1a8f501]
  • Added support for named barrier APIs for ESIMD. [1df0038]
  • Added support for DPAS API for ESIMD. [5881938]
  • Added support for LSC memory access APIs for ESIMD. [4bd50e7]
  • Added support for the invoke_simd feature. [4072557][8471ff3][8c7bb45][62afb59][3e1c1bf]
  • Added support for info::device::atomic64 for OpenCL and Level Zero backends. [8feb558]
  • Added support for sycl_ext_oneapi_usm_device_read_only extension [644c614][58c9d3a]
  • Added support for mapping/unmapping operations for ESIMD_EMULATOR plugin. [bc0579a]
  • Added support for make_buffer API for the Level Zero backend. [7c49984]
  • Added interoperability support for HIP backend. [e06d1b5]
  • Added missing +-*/ operations for half. [059efbc]
  • Introduced new environment variable SYCL_PI_CUDA_MAX_LOCAL_MEM_SZ to control the max local memory allowed to be allocated per kernel on CUDA backend. [2e24304]
  • Added ext_intel_global_host_space in accordance with sycl_ext_intel_usm_address_spaces extension. [7a2f44b]
  • Added aspect for bfloat16. [f84fc32]
  • Introduced "Intel math functions" device library with support of type cast util functions for float, double and integer type. [a310952]
  • Added bfloat16 support for joint_matrix [6ac62ab]

Documentation

Tools

  • Implemented property set generation for device globals in the sycl-post-link. Added the --device-globals command-line argument for lowering and generating information about device global variables. [88123c1]
  • Introduced XPTI-based tools for SYCL applications: sycl-trace, sycl-prof, sycl-sanitize. [789d138]
  • Add support for tracing Level Zero API calls using XPTI and updated sycl-trace tool to be able to display both PI and Level Zero calls simultaneously. [fc9cf52]

Improvements

SYCL Compiler

  • Added a diagnostic on attempt to use zero length arrays in the device code [52e8f58]
  • Added support for consuming fat objects containing SPIR-V [c878063]
  • Added support for generating SPRIV based fat objects [1e94ef3]
  • Added support for group collective functions for HIP backend [106882c]
  • Added a diagnostic on attempt to use -fsycl and -static-libstdc++ together. This combination is not supported due to the runtime dependence with libsycl.so [bb0055c]
  • Added support for atomic loads and stores with various memory orders and scopes [e15ac50] [6b2635e]
  • Improved performance of accessing memory pointed by sycl::accessor for FPGA device [fbab374]
  • Added support for setting CUDA installation directory using CUDA_PATH environment variable [b0c145a]
  • Improved deferred diagnostics for usages within function templates in device code. [d0efca5]
  • Added support for sycl_special_class attribute to mark SYCL classes/struct that need the additional compiler handling. [8ba9c79]
  • Improved driver to do device section checking only when offloading is enabled. [3742b93]
  • Allowed calls to constant expression function pointers in device code. [e84c952]
  • Disabled the passing code coverage and profiling options to device compilation.
  • Added clang support of code location information for kernels. [96d2e17]
  • Disallowed explicit casts between mismatching address spaces. [1cee960]
  • Added support of [[sycl::device_has]] attribute on kernel. [aa2162c]
  • Added a warning on explicit cast from default address space to named. [9adb25b]
  • Added a warning for converting 'c' input to 'c++' in SYCL mode. [5b62ee0]
  • Silenced unknown attribute warnings on host compilation. [2d359df]
  • Added a diagnostic on attemp to use accessor::operator[] in ESIMD code. [9d7a651]
  • Expanded driver's ability to discover fat static archives after /link option on Windows. [271ef40]
  • Added support for saving user specified names for lambda captures in kernel lambda object for FPGA target. [af29982] [5ffb2ee]
  • Adjusted the compilation when preprocessing is requested to allow for the device compilation to fail and continue to perform the preprocessing steps. [7f2e99c]
  • Added the ability to detect a kernel size mismatch in the case when the host and device compilers are different. [ef90e6a]
  • Improved handling of specialization constants by backends. [e62b5aa]
  • Improved support of -mlong-double options. [6083920]
  • Improved -save-temps to allow optimization when performing a SYCL device compilation. [05fe5ae]
  • Removed warning diagnostic on host compilation when using __attribute__((sycl_device)). [49e595e]
  • Introduceed multiple streams in each queue for CUDA backend to improve concurrent execution. [dd41845]
  • Improved compiler to collect information for optimization record only if optmization record is saved by user (i.e. -fsave-optimization-record or -opt-record-file is passed). [cb94c80]
  • Improved [[intel::max_concurrency()]] attribute support. [b6aa4cc]
  • Added the new kernel_arg_exclusive_ptr metadata which guarantees that the kernel pointer argument, or pointers that derive from it, will not be dereferenced outside current invocation of the kernel. [e03c4ed]
  • Added a warning for the case when invalid subgroup size is used on kernel for CUDA backend. [6dab69f]
  • Improved deprecation messaging for options. [f0b65a1]
  • Improved diagnostic behavior for -fsanitize with -fsycl. [9397cbc]

SYCL Library

  • Added support for sycl::ctz API [d5eb769]
  • Improved the diagnostic for unresolved symbols in the device code for Level Zero backend [33cfb9f]
  • Added several arithmetic operations for sycl::ext::oneapi::experimental::wi_element [fe2e35e]
  • Added support for sycl::queue::submit_barrier for HIP backend [53f1cce]
  • Added sycl::property_list APIs to sycl::stream [4b39793]
  • Defined sycl::access::decorated in the SYCL headers. [d183807]
  • Improved performance by allowing batching for wait with a barrier commands for Level Zero backend [bd1ed6a]
  • Disabled image support for CUDA backend since the support is not complete [90c8f05]
  • Avoided JITing unnecessary device code when using sycl::handler::set_specialization_constant [5746906]
  • Updated image accessor constructor to make it possible to use const references in parallel_for. [3070b95]
  • Relax the mutex lock duration in queue finish for the Level Zero backend to allow working with the queue from other threads. [8573935]
  • Added XPTI instrumentation for USM allocations. [7df3923]
  • Extended XPTI information with buffer constructor data. [fc0d28a]
  • Added error handling for sycl::event::get_profiling_info(). [ec74a5c]
  • Eliminated recursion and duplicated dependencies in leaf buffers handling in the scheduler. [6f40376]
  • Improved runtime to emit program build logs when SYCL_RT_WARNING_LEVEL is set to 2 or higher. [008519a]
  • Improved the error message at command execution failure. [930ecbf]
  • Improved runtime to build program for root device only and re-use the binary for sub-devices to avoid "duplicate" builds. [13a7455]
  • Improved sycl::kernel::get_kernel_bundle performance. [4817b3f]
  • Changed USM pooling parameters for the Level Zero backend to boost performance. [57f8a44][b000db8]
  • Exposed value_type and min_capacity from SYCL pipes extension class. [e1619fa]
  • Improved thread-safety of the Level Zero plugin by guarding access to the PI objects. [3321141] [a37c10b] [bd80f34] [8f97fe2]
  • Optimized half builtins for fma, fmin, fmax and fmax on CUDA backend. [62651dd]
  • Improved runtime to redirect warning from using SYCL_DEVICE_FILTER with sycl-ls to std::cerr. [70593d6]
  • Use new SPIR-V group operations within uniform control flow instead of non-uniform operations in SYCL headers. [9b84dd8]
  • Enabled online linking of the device libraries. [9fcab29]
  • Improved esimd-verifier logic for detecting valid SYCL calls. [eaf8b42]
  • Extended XPTI information with the kernel info. [4b9eef3]
  • Improved error message for exceeding CUDA grid limits. [ed877c2]
  • Added overload for sycl::select(a, b, c) where c is a bool. [7ae8fd3]
  • Fixed batching related thresholds to improve performance. [c6313bd]
  • Added always_inline for libdevice functions to enable which allows an underlying runtime to do inlining. [dfc87cc]
  • Improved performance by caching the result of zeKernelGetName in the Level Zero plugin. [40cece3]
  • Updated the experimental latency control API to use property list and made the template argument approach is deprecated. [273034a]
  • Renewed and synced the queue::parallel_for() with SYCL2020 . [e59fb89]
  • Improved runtime to ignore CUDA prefetch hint if not supported on the system and emit an optional warning in this case depending on warning level set using SYCL_RT_WARNING_LEVEL. [082929a]
  • Enabled pooling of small USM allocations for the Level Zero backend to improve performance. [6244efe]
  • Added managed memory check to enqueue prefetch and made it to ignore the prefetch hint and emit a warning if the memory provided is not managed. [0fe322c]
  • Enabled device code instrumentation by default. [53fc8e4]
  • Improveed performance of queue::wait() on CUDA backend. [8b85a3c]
  • Optimized host event wait. [f3a0970]
  • Improved default selector to filter devices based on available device images. [0e67db8]
  • Enabled caching of native OpenCL and Level Zero executable binaries. [f0283fc]
  • Deprecated sycl::ext::intel::ctz extension function,sycl::ctz from the core specification must be used instead. [50435a6]
  • Deprecated cl::sycl::atomic in SYCL 2020 mode. [52fad5a]
  • Deprecated ext_intel_host_device_space which is replaced by ext_intel_global_host_space. [7a2f44b]
  • Added CUDA-specific USM memory advice values. [e2e71a9]

Tools

  • Added an option --enable-esimd-emulator to enable esimd emulator build using configure.py. [ddbcbb1]
  • Added an ability to build plugins separately. [c16412b]
  • Added --enable-all-llvm-targets switch to configure.py. [5e6642a]
  • Added PI tracing support for loadOsLibrary. [647249c]

Documentation

  • Clarified the interaction between the sycl_ext_oneapi_invoke_simd extension and SYCL_EXTERNAL functions. [ac3e816]
  • Removed extensions specifications which were adopted to SYCL 2020. Please refer to extensions/removed/README for the list of removed extensions. [ae7e3d8] [57c9017]
  • Structured sycl/doc directory:
  • Updated CompileTimeProperties design documentation. [2359e81][f6420c7]
  • Clarified which SPIR-V decorations the sycl-post-link tool generates for each device global variable. [3c3b485]
  • Updated the design for device global variables for variables that are "shadowed" in an unnamed namespace. [e211d73]
  • Clarified the specification that device global with SYCL_EXTERNAL is allowed. [0d95d6f]
  • Add an overview README for the extensions directory. [d54708a]
  • Clarified the C++ attribute [[sycl_detail::uses_aspects()]] usage in the documentation. [27cc930]
  • Added a template document to use when creating new SYCL extension specifications. [4bd6d20]
  • Added new rule for naming C++ identifiers in the SYCL project. [1445528]
  • Updated GettingStartedGuide to recommend cuda 11.6. [0b456ce]
  • Added ESIMD_EMULATOR to SYCL_DEVICE_FILTER description. [526ad0c]
  • Clarified availability of get_property(). [a2c5e90]
  • Deprecated extended atomics extension. [7581741]
  • Improved GettingStartedGuide for non-standard HIP installations. [09c3b46]
  • Added description of ESIMD_EMULATOR backend to sycl_ext_intel_esimd/README. [b8e6d23]

Bug fixes

SYCL Compiler

  • Fixed a crash happened if an overloaded new operator is used in a recursive function in the device code [2085978] [2085978]
  • Fixed wrong address space of event_t which could lead to builtins like barriers work incorrectly when using HIP backend. [22532c2]
  • Fixed an issue with certain macros being unavailable when using a custom host compiler. [652417b]
  • Fixed an issue with device code linking when one of the targets is not spir64 based. [1f8874f]
  • Disabled part of SimplifyCFG optimizations in SYCL mode which was resuling in invalid optimizations in some cases. [8b29220]
  • Silenced "unknown attribute" warning emitted during host part of full -fsycl compilation when it saw [[intel::device_indirectly_callable]] attribute. [718c0b1]
  • Removed incorrect assertion for use of -fopenmp-new-driver for multiple inputs. [6e0f6d1]
  • Fixed the issue caused by using the nvvm_reflect function in the nvptx backend with -O0. [537e51b]
  • Fixed a regression in cases where function pointers were captured as kernel arguments. [b19e2e4]
  • Fixed the libclc remangler to clone functions rather than aliasing to enable DPC++ for CUDA with -O0. [7b2fb02]
  • Fixed the error "Explicit load/store type does not match pointee type of pointer operand" caused by incorrect address space. [e688fa5]
  • Fixed incorrect diagnostic for __spirv calls when the reqd_sub_group_size attribute is applied on a sycl kernel. [1df7b59]
  • Fixed alignment of emulated specialization constants. [0cec3c6]
  • Fixed the group collective implementation for AMDGCN. Fixed the shuffleUp and shuffleDown functions for the AMDGCN builtins and SYCL headers. [d99e957]
  • Removed llvm.nvvm.suq.depth instruction which was causing CUDA_ERROR_NOT_FOUND or CUDA_ERROR_NOT_SUPPORTED errors if present in the fatbin. [ec29322]

SYCL Library

  • Fixed alignment of kernel local arguments in the CUDA backend. [ebb1281]
  • Fixed a crash which could happen when bulding program for multiple devices. [64c2d35]
  • Fixed max constant value query for the HIP backend. [1e55cf3]
  • Fixed ambiguity error with sycl::oneapi::experimental::this_nd_item. [8aad52dd6027]
  • Fixed a performance issue caused by unnecessary command batching in the Level Zero plugin. [4d031a4]
  • Fixed an issue with sycl::get_pointer_device API working incorrectly for CUDA and HIP backends [8fa17b4]
  • Fixed an issue which might result in JITing for only one device while context is accosiated with multiple devices for Level Zero backend [7068457]
  • Fixed namespace ambiguity in this_id, this_item, and this_group. [19369b6]
  • Workarounded two bugs in the Level Zero driver related to static linking extension. [2930a94]
  • Fixed return type of get_nativesycl::backend::opencl(event) from cl_event to vector<cl_event>. [a2189c6]
  • Modified Level Zero plugin support for copy engines to address scenario when main copy engine is not available. [478a576]
  • Fixed support for query of USM capabilities. [5941394]
  • Fixed memory leak in the USM prefetch functionality. [5d4573f]
  • Fixed host device local accessor alignment. [08b14da]
  • Fixed sycl::errc values for exceptions per SYCL 2020. [270e78d]
  • Fixed bug with constexpr_recurse usage. [bd15de9]
  • Fixed max_work_group_size and reqd_work_group_size attribute arguments check. [7f37250]
  • Fixed iterator debug level mistmatch error on Widnows when building programs with /MDd when libsycl-fallback-cassert.obj is involved. [93b573a]
  • Fixed get_native() for sycl::event per requirements of the specification. [8878962]
  • Fixed device enumeration for the next platforms when current platform doesn't have devices. [0272ec2]
  • Fixed thread-safety issue in the scheduler which can appear if command gets cleaned up by another thread while adding a host accessor. [62ca43a]
  • Fixed reported device name for HIP backend. [68b089f]
  • Fixed SYCL_PROGRAM_COMPILE_OPTIONS and SYCL_PROGRAM_LINK_OPTIONS to override compile and link options respectively. [7a8fa1a]
  • Fixed incorrect handling of queue indexing for Level Zero backend. [1f4c9df]
  • Fixed memory leak in the reductions that require additional resources (such as buffers) [9aefea0]
  • Defined get_property/has_property in the queue for property::queue::in_order. [ca9fea6]
  • Fixed memory leak in the scheduler for run_on_host_intel commands. [86cf56a]
  • Fixed thread-safety issue caused by parallel access to the command list cache in the Level Zero plugin. [1f531c0]
  • Fixed device code outlining for static local variables to avoid invalid device code generation. [66e207e]
  • Fixed dynamic batching in the Level Zero plugin. [d6f115c]
  • Fixed unsigned long warning in fallback cstring on Windows. [5fbe02b]
  • Fixed sync of host task vs kernel for in-order queue. [c7ba937]
  • Fixed include dependency in fpga_lsu.hpp and pipes.hpp headers. [ac6a4f5]
  • Fixed kernel timestamp calculation in the Level Zero plugin. [c228f12] [7efb3e6] [76a3898]
  • Fixed usage of copy-engines in the Level Zero interoperability queue. [847f8b6]
  • Fixed kernel execution hang under large memory consumption by workarounding a bug in the Level Zero runtime. [b831bd0]
  • Fixed the Level Zero plugin to honor property::queue::enable_profiling. [fb27c65]
  • Fixed memory leak which existed when program build failed for the Level Zero backend. [beb7277]
  • Fixed buffer creation from rvalue iterator. [a905a27]
  • Fixed queue::device_has() to private. [279ef0d]
  • Fixed crash for case when a device image has no kernels. [279ef89]
  • Fixed dependency between host/device actions for unbundled FPGA specific archives. [9699575]
  • Fixed PI CUDA plugin to avoid linking against libsycl which may cause issues on some systems. [6821e66]
  • Fixed support of bfloat16 for CUDA. [5231fe4]
  • Fixed interop_handle::get_native_mem so that it can work with accessors that use non-empty accessor_property_list. [5452a5d]
  • Fixed sub-device count calculation for numa partitioning. [ae284f1]
  • Fixed SYCL_ENABLE_PLUGINS to enable both the OpenCL and the Level Zero PI plugins if it is unset. [ff384bb]
  • Fixed BDF format on PCI query for the Level Zero backend. [ad6253e]
  • Fixed sycl::queue XPTI instrumentation. [ec57cd7]
  • Fixed interoperability return type for sycl::buffer to std::vector<cl_mem> per SYCL 2020. [8b3c8c4]
  • Fixed SYCL_DUMP_IMAGES handling to also dump when spec constants are on. [211ccda]
  • Fixed failure in case of using zero-size local accessor on some backends. [1292532]
  • Fixed flaky bug which might appear in multi-threaded applications with simultaneous access to the cache of device lib programs. [92cfd53]
  • Fixed compfail issue with -ffast-math on CUDA backend. [90ac3ee]
  • Fixed make_queue interoperability API for the Level Zero to accept device argument to properly associate queue with the right device. [29a5369]
  • Fixed invalid handler issue by updating OpenCL ICD loader from community.
  • Fixed "undefined symbol" error for ldexpf, hypotf, frexpf on SYCL GPU device when using 3rd-party math headers instead of MSVC math headers on Windows.[476a351]
  • Fixed memory leak for interop events created from native handle. [3c1d342]
  • Fixed alignment of the memory returned from USM allocation functions. [3114f02]
  • Fixed Vendor ID for AMD devices. [a1b42aa]
  • Fixed sporadic failure of in-order queue due to non-closed batch on the Level Zero backend. [e8bff05]
  • Fixed possible deadlock in case of having dependent events from different queues in a multi-threaded application. [4c619e9]
  • Fixed issue with delivery of assert message before abort'ing [6a32706]
  • Fixed default value for the Alignment template parameter of the usm_allocator. [3a91cec]
  • Fixed API to get maximum width/height/depth of an image for the Level Zero backend. [9ecc74b]
  • Fixed runtime for CUDA backend to ignore usm mem advise and warn if device doesn't meet requirements. [ccaaa99]

Tools

  • Fixed sycl-post-link tool to properly handle the offset in specialization constant descriptors. [12d7c1f]
  • Fixed sycl-post-link tool to properly handle the padding at the end of composite types. [12d7c1f]
  • Fixed translation of Vector[Extract/Insert]Dynamic instructions in llvm-spirv [bf43d7b]
  • Fixed unconditional debug info generation for libsycl_profiler_collector.so [45784cd]
  • Fixed sycl-post-link failure caused by incorrect removal of llvm.used in the case when specialization constant has 2+ users. [80e9148]

Documentation

API/ABI breakages

  • Removed deprecated API from ESIMD headers. [ec0385d]
  • Renamed wi_slice to wi_data. [f364e18]
  • Renamed nbarrier_* API to named_barrier_* for ESIMD. [5023657]
  • Moved a part of ESIMD APIs outside of experimental namespace. [c557d78][b2ee289]
  • Moved bfloat16 from intel namespace to oneapi namespace. [5231fe4]

Known issues

  • Having MESA OpenCL implementation which provides no devices on a system may cause incorrect device discovery. As a workaround such an OpenCL implementation can be disabled by removing /etc/OpenCL/vendor/mesa.icd.
  • Compilation may fail on Windows in debug mode if a kernel uses std::array. This happens because debug version of std::array in Microsoft STL C++ headers calls functions that are illegal for the device code. As a workaround the following can be done:
    1. Dump compiler pipeline execution strings by passing -### option to the compiler. The compiler will print the internal execution strings of compilation tools. The actual compilation will not happen.
    2. Modify the (usually) first execution string (it should have -fsycl-is-device option) by adding -D_CONTAINER_DEBUG_LEVEL=0 -D_ITERATOR_DEBUG_LEVEL=0 options to the end of the string. Execute all string one by one.
  • -fsycl-dead-args-optimization can't help eliminate offset of accessor even though it's created with no offset specified
  • SYCL 2020 barriers show worse performance than SYCL 1.2.1 do [18c80fa]
  • When using fallback assert in separate compilation flow it requires explicit linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero[7dfaf3bd]
  • On the following scenario on Level Zero backend:
    1. Kernel A, which uses buffer A, is submitted to queue A.
    2. Kernel B, which uses buffer B, is submitted to queue B.
    3. queueA.wait().
    4. queueB.wait(). DPCPP runtime used to treat unmap/write commands for buffer A/B as host dependencies (i.e. they were waited for prior to enqueueing any command that's dependent on them). This allowed Level Zero plugin to detect that each queue is idle on steps 1/2 and submit the command list right away. This is no longer the case since we started passing these dependencies in an event waitlist and Level Zero plugin attempts to batch these commands, so the execution of kernel B starts only on step 4. The workaround restores the old behavior in this case until this is resolved [2023e10d][6c137f87].
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::program/sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows [50628db]

December'21 release notes

Release notes for commit range 23ca0c2..27f59d8

New features

SYCL Compiler

  • Added support for -fgpu-inline-threshold which allows controlling inline threshold of the SYCL device code [5f7b607]
  • Added experimental support for CUDA backend on Windows [8aa3513]
  • Added support for experimental option -fsycl-max-parallel-link-jobs=<N> which can be used specify how many processes the compiler can use for linking the device code [c2221f0]

SYCL Library

  • Added support for default context extension on Linux [315593d]
  • Added experimental support for group sorting algorithm [932ae56]
  • Added support for sub-group mask extension [78a3e77]
  • Added sycl::ext::intel::experimental::esimd::simd_mask as a replaced for sycl::ext::intel::experimental::esimd::mask_type_t to represent Gen predicates [01351f1]
  • Added stripped PDBs for SYCL libraries [6e5dd483]
  • Added support ESIMD emulator backend [f4ad3c11]
  • Added support for FPGA DSP control extension [790aa8ba]
  • Implemented discard_events extension [9542e288]
  • Extended XPTI notifications with information about SYCL memory management [a068b154] [8f9d0d2b]
  • Added support for sycl::device::info::ext_intel_gpu_hw_threads_per_eu [2e798df8]
  • Added sycl::ext::oneapi::no_offset property for sycl::ext::oneapi::accessor_property_list [308e5ad5a378]
  • Implemented SYCL 2020 property traits [3b17da60b590]
  • Implemented MAX_WORK_GROUP_QUERY extension [2fdf94006672]
  • Added function wrapper for function-level loop fusion attribute ([[intel::loop_fuse(N)]] and [[intel::loop_fuse_independent(N)]]) for FPGA [e8ac5a076a20]
  • Added querying number of registers for kernel with ext_codeplay_num_regs kernel info query [97d33b7815e8]
  • Added backend macros SYCL_BACKEND_OPENCL and SYCL_EXT_ONEAPI_* for CUDA, Level Zero, ESIMD emulator, HIP [2b0ebab376dc]
  • Added support for sycl::ext::intel::experimental::esimd_ballot function [0bbb091c1baa]
  • Added initial support for Tensor Cores matrix extension [711ba58c30a8]

Documentation

Improvements

SYCL Compiler

  • Added diagnostics on attempt to pass an incorrect value to -fsycl-device-code-split [631fd69]
  • Improved output of -fsycl-help[2404d02]
  • Allowed ::printf builtin for CUDA backend only [0c55d3a6]
  • Implemented nextafter for sycl::half on CUDA backend [53c32682]
  • Added atomics with scopes and memory orders for CUDA backend [2ebde5f6] [00f43b34]
  • Added support for missing mathematical builtins in CUDA backend [789ec8ba] [f0747747] [390e1055]
  • Added diagnostic for non-forward declarable kernel name types [653bae90]
  • Added group_ballot intrinsic for CUDA backend [0680e5c5]
  • Added support for device side assert for CUDA backend [5a87b8cd]
  • Turned on -fsycl-dead-args-optimization by default [5983dfd8]
  • Improved compilation time by removing free function queries calls detection [e4791d11b]
  • Reduced memory consumption of device code linking [62668203]
  • Improved UX of sycl::ext::oneapi::experimental::printf by allowing format string to reside in a non-constant address space [2d62e51d]
  • Improved barrier and sync instructions to use full mask when targeting NVPTX [5ce99b89]
  • Added match for default SPIR device architecture with host architecture i.e. x86_64 matches spir64 and i686 matches spir [f4d01cd4437b]
  • Set default device code split mode to off for FPGA [bea72e65a954]
  • Improved diagnostic for invalid SYCL kernel names [455dce89e3f0] [df1ff7ae3a43]
  • Made Xsycl-target-frontend= to accept device tripple aliases [7fa05699b5e9]
  • Improved diagnostic messages for -fsycl-libspirv-path [c54c6051df4c]
  • Made implied default device to force emulation for FPGA builds [074944e44f5a]
  • Added support for sycl::ext::oneapi::sub_group::get_local_id for HIP backend [7a9335d258f1]
  • Added a diagnostic of indirect implicit capture of this for kernel lambda [dce4c6a68220]

SYCL Library

  • Updated joint matrix queries to report if unsigned int variants of mad matrix instruction are supported [dd7ebce]
  • Reduced overhead of device code assert implementation [b94f23a] [58ac74e]
  • Added a diagnostic on attempt to call sycl::get_kernel_id with an invalid kernel [9dd1ea3]
  • Reduced overhead on kernel submission for CUDA backend [b79ae69]
  • Reduced overhead on kernel submission in backend independent part of runtime [e292aa5]
  • Aligned Level-Zero Interoperability API with SYCL 2020 specification [dd7f82c] [e662166]
  • Made sycl::half default constructor constexpr [d32a444]
  • Changed CUDA and HIP backends to report each device in a separate platform [8dddb11]
  • Added initial support for SYCL2020 exceptions [15e0ab1]
  • Added SYCL 2020 sycl::target::device enumeration value [f710886]
  • Added a diagnostic on attempt to print std::byte using sycl::stream [dd5e094]
  • Added possibility to specify ownership of ze_module_handle_t when creating a sycl::kernel_bundle from it [e3c9c92]
  • Improve performance of sycl::nd_item::get_group_range() [0cd7b7e]
  • Deprecated sycl::target::global_buffer
  • Made device_num which can be passed to SYCL_DEVICE_FILTER unique [7aa5be0]
  • Added a diagnostic on using mutually exclusive sycl::handler methods [6f620a4]
  • Added support for std::byte to sycl::vec class [8fa04fe]
  • Added sycl::make_kernel interoperability support for Level-Zero backend [98896fd]
  • Optimized work with events in the Level Zero backend [973aee9]
  • Added support for sycl::ext::oneapi::experimental::matrix::wi_slice and sycl::ext::oneapi::experimental::matrix::joint_matrix_fill [97127ebf] [cbad428d]
  • Enabled code location information when NDEBUG is not defined in XPTI notifications [e9f2d640] [9ca7ceac]
  • Added a diagnostic on attempt to pass a command group function object to sycl::queue::single_task [2614d4d8]
  • Enlarged the maximum batch size to 64 for Level Zero backend to improve performance [596f6938]
  • Reduced kernel submission overhead for CUDA backend [35729a7f]
  • Improved translation of Level Zero error codes [6699a5d0], [5d9a04bac]
  • Added support for an arbitrary number of elements to sycl::ext::intel::experimental::esimd::simd::copy_from/to methods [2bdc4c42]
  • Added HIP support to sycl::ext::oneapi::filter_selector [7224cb2a], [b7cee067]
  • Added support for batching copy commands for Level Zero backend [4c3e6992]
  • Reduced sycl::queue::submit overhead by enabling post-enqueue execution graph cleanup [6fd6098c]
  • Added support for classes implicitly converted from sycl::item in sycl::handler::parallel_for parameter to align with the SYCL 2020 specification [34b93bf8]
  • Removed direct initialization constructor from sycl::ext::intel::experimental::bfloat16 class [81154ece]
  • Added sycl::vec and sycl::marray support to sycl::known_identity type trait [8fefb257]
  • Added minimal support for the generic space address space to match sycl::atomic_ref class definition in specification [e99f298e]
  • Improved cache of command-lists in the context to be per-device for Level Zero backend [ca457d96]
  • Extended group algorithms to support broadened types [3205368f67e0]
  • Added support for alignement flags in sycl::ext::intel::experimental::esimd::simd::copy_from/copy_to operations [27f5c12674fb]
  • Made sycl::ext::oneapi::atomic_ref available in sycl namespace [2cdcbedaf0a6]
  • Renamed cuda and hip enum values and namespaces to ext_oneapu_cuda and ext_oneapi_hip to align with SYCL 2020 specification [97f916e596fc]
  • Improved performance of kernel submission process [535ad1efc865]
  • Eliminated build of unwanted kernels when creating one with make_kernel [53ea8b9bc1e9]
  • Removed duplicate devices on submission to kernel_bundle API functions [c222497c52b5]
  • Deprecated sycl::aspects::int64_base_atomics and sycl::aspects::int64_extended_atomics [554b79cf7c17]
  • Made backend specific headers be included implicitly [bc8a00ac79ff]
  • Removed program class and related API [e7cc7b08ba5d]
  • Excluded current working directory from DLL search path when looking for runtime dependencies [0a65cb4b6e2f]
  • Enabled persistent device code cache for kernel bundles [810d67a21646]
  • Removed SYCL 1.2.1-style OpenCL interoperability API [bbafe08fd490]
  • Added a diagnostic on attempt to pass incorrect local size for HIP backend [c56499c2a72b]
  • Added a diagnostic on attempt to create an object of sycl::bufferwith non-device copyable underlying date type [61f1ae6a9ef8]
  • Added a diagnostic(exception) on attempt to pass an empty sycl::device list to sycl::link [e95c1849a8fb]
  • Disabled fallback assert mechanism for objects of sycl::kernelcreated using interoperability API [f8014f57360c]
  • Added a diagnostic(exception) on attempt to pass a sycl::kernel_bundle to sycl::handler::use_kernel_bundle when kernel bundle and handler are bound to different contexts [49eb2d7967d7]

Tools

  • Improved sycl-ls device indexing to handle new backends [0cee18e]

Documentation

  • Updated Level-Zero backend extension to align with the SYCL 2020 specification [8bbebf5]
  • Updated ITT instrumentation documentation [9c0508b]
  • Extended FPGA lsu and SYCL_INTEL_data_flow_pipes extensions to with latency control feature [5ab3cd3a]
  • Updated OpenCL CPU and FPGA emulator runtimes configuration on Windows to use OpenCL ICD registry records instead of OCL_ICD_FILENAMES. That is done to use the latest OpenCL ICD loader which ignores OCL_ICD_FILENAMES configuration in the administrative console [92bcb419]
  • Improved group sorting spec [f1c3506b]
  • Clarified restrictions on device global variables [589d48844edb]
  • Deprecated group algorithms and subgroup extensions as it's superseded by SYCL2020 [7dc82742b5ae] [d2a4f502b558]
  • Removed SubGroupAlgorithms extension [cd5417b13a27]
  • Added proposal to add query for number of hardware threads per execution unit [5b771a9c4eca]
  • Updated the experimental matrix interface to match new AMX/DPAS JIT implementation [6495575258a8]
  • Updated test section of the contribution guide [145b9e782258]
  • Updated CUDA documentation with Windows support details [1cf024ac32a3]
  • Added FPGA properties to device global specification [fd2bd6e1eac7]
  • Added clarifications in environment variables documentation [d5ba0dbb4912]

Bug fixes

SYCL Compiler

  • Fixed a problem which might affect headers searching [aad223f]
  • Fixed a problem which prevented submitting kernels from a static library when using CUDA backend [665a0a2]
  • Fixed backwards compatibility for libraries in archive format [53308b1]
  • Fixed crash on attempt to compile a kernel which uses templated global variable [226679a0]
  • Fixed a possible hang caused by recursive calls to memcpy or memset on CPU device [cb5e8ae54]
  • Allowed to have SYCL headers included in extern C++ block [fcd7e317]
  • Disabled the warning when fsycl-host-compiler is used [756c2e8f]
  • Fixed handling of specializations constants when compiling for AOT and SPIR-V images [f82ddf41]
  • Fixed nan builtin for doubles in libclc [cc0d67f0]
  • Fixed out-of-bound behavior for none addressing mode on CUDA [fceb10e5]
  • Fixed memory leak which happened during device code compilation [bac0a25c]
  • Fixed compiler crash when specialization constant is value dependent [6dcc988eadaa]
  • Updated the device triple population logic to also incorporate usage of -fsycl-link-targets [e877e3bf8798]
  • Fixed -Xsycl option triple check with multiple -fsycl-targets to only use the last value of the option [c623223a78e0]
  • Fixed a bug with default value generation for specialization constant [4f5fa0eb59f2]

SYCL Library

  • Fixed sycl::is_device_copyable type trait which could conflict with range rounding optimization [2d28cd4]
  • Aligned sycl::id and sycl::range implementations with the specification. The conversion operator from sycl::id to sycl::range has been deprecated and will be removed in future versions [560a214]
  • Fixed a problem with some non-users kernels showing up in kernel bundles [49e1e74] [4112cbc]
  • Fixed sycl::event::get_wait_list() which could return incorrect result [02852c5] [5bb3ab9e]
  • Fixed conversion of zero dimension sycl::accessor with sycl::access::address_space::global_device_space to an object of sycl::atomic type [ce7725d]
  • Fixed possible problem which could happen if there is no active CUDA context during a kernel submission in CUDA backend [0d3cc99]
  • Fixed memory leak which could happen in case of unsuccessful sycl::stream construction [4ceba5b]
  • Fixed a problem which could lead to memory corruption during graph cleanup [42c6f44]
  • Fixed a problem with sycl::accessor::get_range returning incorrect results in case of 3 dimensional image accessor [266515a]
  • Fixed sycl::ext::oneapi::sub_group_mask::insert_bits [f88a19e]
  • Fixed sycl::ext::intel::experimental::esimd::simd_mask behavior for sub-group sizes less than 32 [c855fd17]
  • Fixed a problem with sycl::handler::fill for HIP backend [cee76d9]
  • Fixed a problem resulting in an exception when trying to allocate a big sycl::image on the host [3061bc7]
  • Fixed behavior of sycl::handler::parallel_for which takes a sycl::kernel to ignore a sycl::kernel_bundle which may be set earlier [0b97344]
  • Made sycl::kernel_bundle::get_native const [91fef67]
  • Fixed logic, arithmetic and compare operators for sycl::ext::intel::experimental::esimd::simd_view with scalar values [7e11e48] [b88c455]
  • Aligned sycl::device::create_sub_devices exceptions with SYCL 2020 specification [d2252e6]
  • Added a diagnostic on attempt to submit a kernel which is missing from a sycl::kernel_bundle set for a handler [7dbb6fb]
  • Disabled fallback assert for CUDA, HIP and FPGA devices [b0411f8] [09ece34c]
  • Fixed a race which could happen when submitting a host task to the in-order sycl::queue [20d9346]
  • Fixed memory leak when constructing a sycl::queue using interoperability API with Level Zero [299f5068]
  • Fixed a problem with releasing USM memory while there are not finished kernels that may access this memory for GPU devices of OpenCL backend [c4a72908]
  • Fixed a hang which could happen when cross queues dependencies are used [23e180bc]
  • Fixed a memory leak in ext::intel::experimental::online_compiler [2af6ccdb]
  • Fixed alignment of sycl::vec class on windows platform [826c5693]
  • Fixed an issue with incorrect arguments setting in case of multiple device images [5ca3628b]
  • Made accessor::get_pointer always return base pointer, even for accessors with offsets [59fcb82d]
  • Fixed problems with selection of the correct binary [a346c081]
  • Fixed sycl::span type deduction [f5d08c50]
  • Made initialization of host task thread pool thread-safe [2caf555c]
  • Fixed infinite loop when sycl::handler::parallel_for range exceeds INT_MAX [64720d87], [fd0b1082]
  • Fixed platform query in USM allocation info for HIP backend [2bc3c92a]
  • Fixed dangling pointer in sycl::device_event [8a3bd1c5]
  • Widen (u)int8/16 to (u)int32 and half to float in sycl::group_broadcast to workaround missing support of (u)int8/16 and half on OpenCL CPU [1f3f9b9b]
  • Fixed an issue with sycl::make_kernel [13443a96]
  • Fixed a memory leak of specialization constant buffer [ae711ab9]
  • Fixed a memory leak in in-memory program cache [0b8cff452158]
  • Fixed a problem with invalid sycl::kernel_boundle being associated with a sycl::kernel created using sycl::make_kernel [7431afae7eb3]
  • Disabled use of std::byte when __HAS_STD_BYTE=0 definition supplied on MSVC [347b114579ed]
  • Removed half class from global namespace [c9128e68ef39]
  • Defined missing feature test macros [504ac975d68e]
  • Fixed a problem with using sycl::multi_ptr with std::byte as an underlying type [cb388aae4086]
  • Fixed an issue with sycl::queue::[memcpy|memset] APIs breaking dependency chain when used for USM memory and zero size is specified [1a08b7e24864]
  • Fixed device aspect check in sycl::has_kernel_bundle [88cc5247ce8c]

API/ABI breakages

  • The following classes have been removed:
    • sycl::vector_class
    • sycl::string_class
    • sycl::function_class
    • sycl::mutex_class
    • sycl::unique_ptr_class
    • sycl::shared_ptr_class
    • sycl::weak_ptr_class
    • sycl::hash_class
    • sycl::exception_ptr_class. Replacement: analogs from STL
  • Support for the following attributes have been removed:
    • intelfpga::num_simd_work_items
    • intelfpga::max_work_group_size
    • intelfpga::max_global_work_dim
    • intelfpga::no_global_work_offset
    • intelfpga::ivdep
    • intelfpga::ii
    • intelfpga::max_concurrency
    • intelfpga::loop_coalesce
    • intelfpga::disable_loop_pipelining
    • intelfpga::max_interleaving
    • intelfpga::speculated_iterations
    • intelfpga::doublepump
    • intelfpga::singlepump
    • intelfpga::memory
    • intelfpga::register
    • intelfpga::bankwidth
    • intelfpga::numbanks
    • intelfpga::private_copies
    • intelfpga::merge
    • intelfpga::max_replicates
    • intelfpga::simple_dual_port
    • intelfpga::bank_bits
    • intelfpga::force_pow2_depth
    • intelfpga::scheduler_target_fmax_mhz Replacement: the same attributes, but in intel:: namespace
  • Removed sycl::ONEAPI and sycl::INTEL Replacement: sycl::ext::oneapi and sycl::ext::intel
  • The patch [8a3bd1c5] should not break SYCL library ABI but can affect device-code ABI
  • Removed program class and related API [e7cc7b08ba5d]
  • Removed SYCL 1.2.1-style OpenCL interoperability API [bbafe08fd490]
  • Removed half class from global namespace [c9128e68ef39]

Known issues

  • [new] Having MESA OpenCL implementation which provides no devices on a system may cause incorrect device discovery. As a workaround such an OpenCL implementation can be disabled by removing /etc/OpenCL/vendor/mesa.icd.
  • [new] Compilation may fail on Windows in debug mode if a kernel uses std::array. This happens because debug version of std::array in Microsoft STL C++ headers calls functions that are illegal for the device code. As a workaround the following can be done:
    1. Dump compiler pipeline execution strings by passing -### option to the compiler. The compiler will print the internal execution strings of compilation tools. The actual compilation will not happen.
    2. Modify the (usually) first execution string (it should have -fsycl-is-device option) by adding -D_CONTAINER_DEBUG_LEVEL=0 -D_ITERATOR_DEBUG_LEVEL=0 options to the end of the string. Execute all string one by one.
  • [new] -fsycl-dead-args-optimization can't help eliminate offset of accessor even though it's created with no offset specified
  • [new] cuMemPrefetchAsync has issues on Windows. Hence, using sycl::queue::prefetch API on Windows might lead to failure [0c33048e4926]
  • SYCL 2020 barriers show worse performance than SYCL 1.2.1 do [18c80fa]
  • When using fallback assert in separate compilation flow it requires explicit linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • Performance may be impacted by JIT-ing an extra 'copier' kernel and due running the 'copier' kernel and host-task after each kernel which uses assert
  • Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero[7dfaf3bd]
  • On the following scenario on Level Zero backend:
    1. Kernel A, which uses buffer A, is submitted to queue A.
    2. Kernel B, which uses buffer B, is submitted to queue B.
    3. queueA.wait().
    4. queueB.wait(). DPCPP runtime used to treat unmap/write commands for buffer A/B as host dependencies (i.e. they were waited for prior to enqueueing any command that's dependent on them). This allowed Level Zero plugin to detect that each queue is idle on steps 1/2 and submit the command list right away. This is no longer the case since we started passing these dependencies in an event waitlist and Level Zero plugin attempts to batch these commands, so the execution of kernel B starts only on step 4. The workaround restores the old behavior in this case until this is resolved [2023e10d][6c137f87].
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::program/sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows [50628db]

September'21 release notes

Release notes for commit range 4fc5ebe..bd68232

New features

SYCL Compiler

  • Added support for HIP backend [06fdabb5][da138cc][de6e848c][104b45ee][2e08d0e][bfbd5af][c19294f]

SYCL Library

  • Added sRGBA support [e488327][191efdd]
  • Added a preview feature implementation for the DPC++ experimental matrix extension [7f218531] [a95f46d]
  • Added support for SYCL 2020 exceptions [5c0f748][eef07606][5af8c43d]
  • Added support for sycl_ext_intel_bf16_conversion extension [8075463]
  • Added support for fallback implementation of assert feature [56c9ec4]
  • Added support SYCL 2020 sycl::logical_and and sycl::logical_or operators [6c077a0]

Documentation

Improvements

SYCL Compiler

  • Added default device triple spir64 when the compiler encounters any incoming object/libraries that have been built with the spir64 target. -fno-sycl-link-spirv can be used for disabling this behaviour [1342360]
  • Added support for non-uniform IMul and FMul operation for ptx-nvidiacl [98a339d]
  • Added splitting modules capability when compiling for NVPTX and AMDGCN [c1324e6]
  • Added -fsycl-footer-path=<path> command-line option to set path where to store integration footer [155acd1]
  • Improved read only accessor handling - added readonly attribute to the associated pointer to memory [3661685]
  • Improved the output project name generation. If the output value has one of .a .o .out .lib .obj .exe extension, the output project directory name will omit extension. For any other extension the output project directory will keep this extension [d8237a6e]
  • Improved handling of default device with AOCX archive [e3a579f]
  • Added support for NVPTX device printf [4af2eb51]
  • Added support for non-const private statics in ESIMD kernels [bc51fe01]
  • Improved diagnostic generation when incorrect accessor format is used [a2922141]
  • Allowed passing -Xsycl-target-backend and -Xsycl-target-link when default target is used [d37b832]
  • Disabled kernel function propagation up the call tree to callee when in SYCL 2020 mode [2667e3e]

SYCL Library

  • Improved information passed to XPTI subscribers [2af0599] [66770f0]
  • Added event interoperability to Level Zero plugin [ef33c57]
  • Enabled blitter engine for in-order queues in Level Zero plugin [904967e]
  • Removed deprecation warning for SYCL 1.2.1 barriers [18c80fa]
  • Moved free function queries to experimental namespace sycl::ext::oneapi::experimental [63ba1ce]
  • Added info query for device::info::atomic_memory_order_capabilities and context::info::atomic_memory_order_capabilities [9b04f41f]
  • Improved performance of generic shuffles [fb08adf]
  • Renamed ONEAPI/INTEL namespace to ext::oneapi/intel [d703f57] [ea4b8a9] [e9d308e]
  • Added Level Zero interoperability which allows to specify ownership of queue [4614ee4] [6cf48fa]
  • Added support for reqd_work_group_size attribute in CUDA plugin [a8fe4a5]
  • Introduced SYCL_CACHE_DIR environment variable which allows to specify a directory for persistent cache [4011775]
  • Added version of parallel_for accepting range and a reduction variable [d1556e4]
  • Added verbosity to some errors handling [84ee39a]
  • Added SYCL 2020 sycl::errc_for API [02756e3f]
  • Added SYCL 2020 byte_size method for sycl::buffer and sycl::vec classes. get_size was deprecated [282d1dec]
  • Added support for USM pointers for sycl::joint_exclusive_scan and sycl::joint_inclusive_scan [2de0f92a]
  • Added copy and move constructors for sycl::ext::intel::experimental::esimd::simd_view [daae1475]
  • Optimized memory allocation when sub-devices are used in Level Zero plugin [6504ba0a]
  • Added constexpr constructors for vec and marray classes [e7cd86b9][449721b2]
  • Optimized kernel cache [c16705a5]
  • Added caching of device properties in Level Zero plugin [a50f45b8]
  • Optimized Cuda plugin work with small kernels [07189af0]
  • Optimized submission of kernels [441dc3b2][33432df6]
  • Aligned implementation of sycl_ext_oneapi_local_memory extension document with updated document [b3db5e5]
  • Improved sycl::accessor initialization performance on device [a10199d]
  • Added support sycl::get_kernel_ids and cache for sycl::kernel_id objects [491ec6d]
  • Deprecated ::half since it should not be available in global namespace, sycl::half can be used instead [6ff9cf7]
  • Deprecated sycl::interop_handler, sycl::handler::interop_task, sycl::handler::run_on_host_intel, sycl::kernel::get_work_group_info and sycl::spec_constant APIs [5120763]
  • Made sycl::marray trivially copyable and device copyable [6e02880] [b5b66737]
  • Made Level Zero interoperability API SYCL 2020 compliant for sycl::platform, sycl::device and sycl::context [c696415]
  • Deprecated unstable keys of SYCL_DEVICE_ALLOWLIST [b27c57c]
  • Added predefined vendor macro SYCL_IMPLEMENTATION_ONEAPI and SYCL_IMPLEMENTATION_INTEL [6d34ebf]
  • Deprecated sycl::ext::intel::online_compiler, sycl::ext::intel::experimental::online_compiler can be used instead [7fb56cf]
  • Deprecated global_device_space and global_host_space values of sycl::access::address_space enumeration, ext_intel_global_device_space ext_intel_host_device_space can be used instead [7fb56cf]
  • Deprecated sycl::handler::barrier and sycl::queue::submit_barrier, sycl::handler::ext_oneapi_barrier and sycl::queue::ext_oneapi_submit_barrier can be used instead [7fb56cf]
  • Removed sycl::handler::codeplay_host_task API [9a0ea9a]

Tools

  • Added support for ROCm devices in get_device_count_by_type [03155e7]

Documentation

  • Extended group sort algorithms extension with interfaces to scratchpad memory [f57091d]
  • Updated several extension documents to follow SYCL 2020 extension rules [7fb56cf]

Bug fixes

SYCL Compiler

  • Fixed emission of integration header with type aliases [e3cfa19]
  • Fixed compilation for AMD GPU with -fsycl-dead-args-optimization [5ed48b4]
  • Removed faulty implementations for atomic loads and stores for acquire, release and seq_cst memory orders libclc for NVPTX [4876443]
  • Fixed having two specialization for the specialization_id, one of which was invalid [f71a1d5]
  • Fixed context destruction in HIP plugin [6042d3a]
  • Changed queue::mem_advise and handler::mem_advise to take int instead of pi_mem_advice [af2bf96]
  • Prevented passing of -fopenmp-simd to device compilation when used along with -fsycl [226ed8b]
  • Fixed generation of the integration header when non-base-ascii chars are used in the kernel name [91f50477]
  • Fixed a problem which could lead to picking up incorrect kernel at runtime in some situations when unnamed lambda feature is used [27c632e]
  • Fixed suggested target triple in the warning message [7cc89fa]
  • Fixed identity for multiplication on CUDA backend [a6447ca4]
  • Fixed a problem with dependency file generation [fd6d948] [1d5b2cb]
  • Fixed builtins address space type for CUDA backend [1e3136e4]
  • Fixed a problem which could lead to incorrect user header to be picked up [c23fe4b]

SYCL Library

  • Added assign operator to specializations of sycl::ext::oneapi::atomic_ref [c6bc5a6]
  • Fixed the way managed memory is freed in CUDA plugin [e825916]
  • Changed names of some SYCL internal enumerations to avoid possible conflicts with user macro [1419415]
  • Fixed status which was returned for host events by event::get_info<info::event::command_execution_status>() call [09715f6]
  • Fixed memory ordering used for barriers [73455a1]
  • Fixed several places in CUDA and HIP plugins where bool was used instead of uint32_t [764b6ff]
  • Fixed event pool memory leak in Level Zero plugin [0e95e5a]
  • Removed redundant memcpy call for copying struct using fpga_reg [a5d290d5]
  • Fixed an issue where the native memory object passed to interoperability memory object constructor was ignored on devices without host unified memory [da196784a]
  • Fixed a bug in simd::replicate_w API [d36480da]
  • Fixed group operations for (u)int8/16 types [6a055ec2]
  • Fixed a problem with non-native specialization constants being undefined if they are not explicitly updated to non-default values [3d96e1d]
  • Fixed a crash which could happen when a default constructed event is passed to sycl::handler::depends_on[2fe7dd3]
  • Fixed sycl::link which was returning a kernel bundle in the incorrect state [6d98beb]
  • Fixed missing dependency for host tasks in in-order queues [739487c]
  • Fixed a memory leak in Level Zero plugin [a5b221f]
  • Fixed a problem with copying buffers when an offset is specified in CUDA backend [0abdfd6]
  • Fixed a problem which was preventing passing asynchronous exceptions produced in host tasks [f823d61]
  • Fixed a crash which could happen when submitting tasks to multiple queues from multiple threads [4ccfd21]
  • Fixed a possible multiple definitions problem which could happen when using fsycl-device-code-split option [d21082f]
  • Fixed several problem in configuration file processing [2a35df0]
  • Fixed imbalance in events release and retain in Level Zero plugin [6117a1b]
  • Fixed a problem which could lead to leaks of full paths to source files from the build environment [6bbfe42b3]

API/ABI breakages

  • Removed intel::reqd_work_group_size attribute, sycl::reqd_work_group_size can be used instead [c583a20]

Known issues

  • [new] SYCL 2020 barriers show worse performance than SYCL 1.2.1 do [18c80fa]
  • [new] When using fallback assert in separate compilation flow requires explicit, linking against lib/libsycl-fallback-cassert.o or lib/libsycl-fallback-cassert.spv
  • [new] Performance may be impacted by JIT-ing an extra 'copier' kernel and due running the 'copier' kernel and host-task after each kernel which uses assert
  • [new] Driver issue. When a two-step AOT build is used and there's at least a single call to devicelib function from within kernel, the device binary image gets corrupted
  • [new] Limit alignment of allocation requests at 64KB which is the only alignment supported by Level Zero[7dfaf3bd]
  • [new] On the following scenario on Level Zero backend:
    1. Kernel A, which uses buffer A, is submitted to queue A.
    2. Kernel B, which uses buffer B, is submitted to queue B.
    3. queueA.wait().
    4. queueB.wait(). DPCPP runtime used to treat unmap/write commands for buffer A/B as host dependencies (i.e. they were waited for prior to enqueueing any command that's dependent on them). This allowed Level Zero plugin to detect that each queue is idle on steps 1/2 and submit the command list right away. This is no longer the case since we started passing these dependencies in an event waitlist and Level Zero plugin attempts to batch these commands, so the execution of kernel B starts only on step 4. The workaround restores the old behavior in this case until this is resolved [2023e10d][6c137f87].
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::program/sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows [50628db]

July'21 release notes

Release notes for commit range 6a49170027fb..962909fe9e78

New features

  • Implemented SYCL 2020 specialization constants [07b27965] [ba3d657] [bd8dcf4] [d15b841]
  • Provided SYCL 2020 function objects [24a2ad89]
  • Added support for ITT notification in SYCL Runtime [a7b8daf] [8d3921e3]
  • Implemented SYCL 2020 sub_group algorithms [e8caf6c3]
  • Implemented SYCL 2020 sycl::handler::host_task method [75e5a269]
  • Implemented SYCL 2020 sycl::sub_group class [19dcac79]
  • Added support for AMD GPU devices [ec612228]
  • Implemented SYCL 2020 sycl::is_device_copyable type trait [44c1cbcd]
  • Implemented SYCL 2020 USM features [1df6873d]
  • Implemented support for Device UUID from Intel's Extensions for Device Information [25aee287]
  • Implemented SYCL 2020 sycl::atomic_fence [dcd59547]
  • Implemented intel::loop_count_max, intel::loop_count_max, intel::loop_count_avg attributes that allow to specify number of loop iterations for FPGA [f74b4ef]
  • Implemented generation of compiler report for kernel arguments [201f902]
  • Implemented SYCL 2020 [[reqd_sub_group_size]] attribute [347e41c]
  • Implemented support for [[intel::named_sub_group_size(primary)]] attribute from sub-group extension [347e41c]
  • Implemented SYCL 2020 interoperability API [e6733e4]
  • Added group sorting algorithm extension specification [edaee9b]
  • Added initial draft for querying of free device memory in LevelZero backend extension [fa428bf]
  • Added InvokeSIMD and Uniform extensions [72e1611]
  • Added Matrix Programming Extension for DPC++ document [ace4c733]
  • Implemented SYCL 2020 sycl::span [9356d53]
  • Added device-if extension [4fb95fc]
  • Added a programming guide for multi-tile and multi-card under Level Zero backend [d581178a]
  • Implemented SYCL 2020 sycl::bit_cast [d4b66bd]

Improvements

SYCL Compiler

  • Use opencl-aot instead of aoc when AOT flow for FPGA Emulator is triggered [3a99558]
  • Allowed for using an external host compiler [6f0ad1a]
  • Cleaned up preprocessing output when -fsycl option is passed [3a18db6]
  • Allowed kernel names in anonymous namespace [e47dbad]
  • Set default value of -sycl-std to 2020 for SYCL enabled compilations [680adc0]
  • Added implication of -fPIC compilation for wrapped object when using -shared [1754934]
  • Added a diagnostic for -fsycl and -ffreestanding as non-supported combination [a36c6720]
  • [ESIMD] Renamed simd::format to simd::bit_cast_view [653dede1]
  • Allowed [[sycl::work_group_size_hint]] to accept constant expr args [ef8e4019]
  • Deprecated the old-style SYCL attributes according to the SYCL 2020 spec [001bbd42]
  • Deprecated [[intel::reqd_work_group_size]] attribute spelling, please use [[sycl::reqd_work_group_size]] instead [8ef7eacc]
  • Enabled native FP atomics by default. Defining the SYCL_USE_NATIVE_FP_ATOMICS macro explicitly is no longer required - it is now automatically defined for hardware targets with "native" support for atomic functions. [0bbb68ee]
  • Switched to ignoring -O0 option for device code when compiling for FPGA with hardware [7d94edf4]
  • Allowed for known aliases to be used for -fsycl-targets. Passing *-unknown-unknown-sycldevice components of the SYCL target triple is no longer necessary. [9778952a]
  • [ESIMD] Added support for half type in ESIMD intrinsics [d5958ebf]
  • Implemented sycl::kernel::get_kernel_bundle method [69a68a6d]
  • Added a diagnostic in case of timing issues for FPGA AOT [c69a3115]
  • Added support for C memcpy usages in the device code [76051ccf]
  • [ESIMD] Added support for vectorizing scalar function [3fc66cc]
  • Disabled vectorization and loop transformation passes because loop unrolling in "SYCL optimization mode" used default heuristic, which is tuned the code for CPU and might not have been profitable for other devices [ff6929e6]

SYCL Library

  • Added an exception throw if no matched device is found when SYCL_DEVICE_FILTER is set regardless of device_selector used [ef4e6dd]
  • Changed event status update to complete without waiting when run on CUDA devices [be7c1cb]
  • Improved performance when executing with dynamic batching on Level Zero backend [fa382d6]
  • Introduced pooling for USM and buffer allocations in Level Zero backend [4cffedd]
  • Added support for vectors with length of 3 and 16 elements in sub-group load and store operations [4e6452d]
  • Added interop types for images for Level Zero and OpenCL backends [a58cfef]
  • Improved plugins discovery - continue discovering even if a plugin fails to load [8c07803]
  • Implemented queries for IEEE rounded sqrt/div in Level Zero backend [91b35c4]
  • Added SYCL 2020 interop_handle::get_backend() method [041ca27]
  • [ESIMD] Deprecated block_load/block_store and simd::copy_from/simd::copy_to [5c41ed6]
  • Allowed for const and volatile pointer in sub-group load operation [50edee4]
  • Replaced use of interop<> with SYCL 2020 backend_return_t<> in interop_handle [d08c21a]
  • [ESIMD] Moved ESIMD APIs to sycl::ext::intel::experimental::esimd namespace [92da579]
  • Added global offset support for Level Zero backend [9ca2f911]
  • [ESIMD] Changed simd::replicate API by adding suffixes into the names to reflect the order of template arguments [e45408ad]
  • Introduced SYCL_REDUCTION_DETERMINISTIC macro which forces reduction algorithms to produce stable results [a3fc51a4]
  • Improved SYCL_DEVICE_ALLOWLIST format [9216b49d]
  • Added SYCL_DISABLE_PARALLEL_FOR_RANGE_ROUNDING macro to disable range rounding [5c4275ac]
  • Disabled range rounding by default when compiling for FPGA [5c4275ac]
  • Deprecated sycl::buffer::get_count(), please use sycl::buffer::size() instead [baf2ed9d]
  • Implemented sycl::group_barrier free function [48363902]
  • Added support of SYCL_INTEL_enqueue_barrier extension for CUDA backend [2e978482]
  • Deprecated has_extension method of sycl::device and sycl::platform classes, please use has method with aspects APIs instead [51c747da]
  • Deprecated sycl::*_class types, please use STL classes instead [51c747da]
  • Deprecated sycl::ndrange with an offset [51c747da]
  • Deprecated barrier and mem_fence methods of sycl::nd_item class, please use sycl::group_barrier() and sycl::atomic_fence() free functions instead [51c747da]
  • Deprecated sycl::byte, please use std::byte instead [51c747da]
  • Deprecated sycl::info::device::max_constant_buffer_size and sycl::info::device::max_constant_args [51c747da]
  • Deprecated sycl::ext::intel::fpga_reg taking non-trivially copyable structs [b4c322a8]
  • Added support for sycl::property::queue::cuda::use_default_stream queue property [08330525]
  • Switched to using atomic version of reductions if sycl::aspect::atomic64 is available for a target [544fb7c8]
  • Added support for sycl::aspect::fp16 for CUDA backend [db20bab3]
  • Deprecated sycl::aspect::usm_system_allocator, please use sycl::aspect::usm_system_allocations instead [000cc82d]
  • Optimized sycl::queue::wait to wait for batch of events rather than waiting for each event individually [7fe72dba]
  • Deprecated sycl::ONEAPI::atomic_fence, please use sycl::atomic_fence instead [dcd59547]
  • Added constexpr constructor for sycl::half type [5759e2a1]
  • Added support for more than 4Gb device allocations in Level Zero backend [fb1808b8]

Documentation

  • Updated sub-group algoritms extension to use marray instead of vec [98715ae]
  • Updated data flow pipes extension to be based on SYCL 2020 [f22f2e0]
  • Updated ESIMD documentation reflecting recent API changes [1e0bd1ed]
  • Updated devicelib extension document with scalnbn, abs and div (and their variants) as supported [febfb5a]
  • Addressed renaming of TBB dll to tbb12.dll in the install script [25433ba]

Bug fixes

SYCL Compiler

  • Fixed crash which could happen in corner cases when null attribute created [cec6469]
  • Fixed crash when lowering __sycl_alocateLocalMemory [4960e71]
  • Fixed workflow for multi-file compilation in AOT mode [a0099a5]
  • Fixed problem with unbundling from object for device archives for FPGA [25ea6e1]
  • Stopped implying defaultlib msvcrt for Linux based driver on Windows [d3dc212d]
  • Fixed handling of [[intel::max_global_work_dim()]] in case of redeclarations [9b615928]
  • Fixed incorrect diagnostics in the presence of OpenMP [cbec0b5f]
  • Fixed an issue with incorrect output project report when using -o option with FPGA AOT enabled [18ac1723]
  • Removed restriction that was preventing from applying [[intel::use_stall_enable_clusters]] attribute to ANY function [15da879d]
  • Fixed bugs with recursion in SYCL kernels - diagnostics won't be emitted on using recursion in a discarded branch and in constexpr context [9a9a018c]
  • Fixed handling of intel::use_stall_enable_clusters attribute [06e4ebc7]

SYCL Library

  • Fixed build issue when CUDA 11 is used [f7224f1]
  • Fixed caching of sub-devices in Level Zero backend[4c34f93]
  • Fixed requesting of USM memory allocation info on CUDA [691f842]
  • Fixed joint_matrix_mad behaviour to return A*B+C instead of assigning the result to C [ea59c2b]
  • Workaround an issue in Level Zero backend when event isn't waited upon its completion but is queried for its status in an infinite loop [bfef316]
  • Fixed persistent cache key comparison (esp. when there is \0 symbol) [3e9ed1d]
  • Fixed a build issue when sycl::kernel::get_native is used [eb17836]
  • Fixed collisions of helper functions and SPIR-V operations when building with -O0 or -O1 [9f2fd98] [c2d6cfa]
  • [OpenCL] Fixed false-positive assertion trigger when allocation alignment is expected [3351916ad]
  • Aligned behavior of empty command groups with SYCL 2020 [1cf697bd]
  • Fixed build options handling when they come from different sources [67411472]
  • Fixed host task CUDA native memory handle [e9cf124b6]
  • Fixed a memory leak which could happen if a command submission fails [67eac4bd]
  • Fixed support for math functions floor/rndd/rndu/rndz/rnde in ESIMD mode [de694dd8]
  • Fixed memory allocations for multi-device contexts on Level Zero [f83c9356a]
  • Renamed sycl::property::no_init property to sycl::property::no_init in accordance to final SYCL 2020 specification, the old spelling is deprecated [ad46b641]
  • Use local size specified in [[sycl::reqd_work_group_size]] if no local size explicitly passed [0a54bef2]
  • Disabled persistent device code caching by default since it doesn't reliably identify driver version change [48f6bc9e]
  • [ESIMD] Fixed a bug in simd_view::operator-- [ccc97e23]
  • Fixed a memory leak for host USM allocations [c18c3456]
  • Fixed possible crashes that could happen when sycl::free is called while there are still running kernels [c74f05d6]

API/ABI breakages

  • None

Known issues

  • [new] The compiler generates a temporary source file which is used during host compilation. This source file will appear to be a source dependency and could break build environments (such as Bazel) which closely keeps track of the generated files during a compilation. Build environments such as these will need to be configured in the DPC++ space to expect an additional intermediate file to be part of the compilation flow.
  • User-defined functions with the name and signature matching those of any OpenCL C built-in function (i.e. an exact match of arguments, return type doesn't matter) can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::program/sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows [50628db]

May'21 release notes

Release notes for commit range 2ffafb95f887..6a49170027fb

New features

  • [ESIMD] Allowed ESIMD and regular SYCL kernels to coexist in the same translation unit and in the same program. The -fsycl-explicit-simd option is no longer required for compiling ESIMD code and was deprecated. DPCPP RT implicitly appends -vc-codegen compile option for ESIMD images.
  • [ESIMD] Added indirect read and write methods to ESIMD class [8208427]
  • Provided sycl::ONEAPI::has_known_identity type trait to determine if reduction interface supports user-defined type at compile-time [0c7bd24] [060fd50]
  • Added support for multiple reduction items [c042f9e]
  • Added support for +=, *=, |=, ^=, &= operations for custom type reducers [b249099]
  • Added SYCL 2020 sycl::kernel_bundle support [5af118a] [dcfb6b1] [ae45333] [8335e17]
  • Added sycl/sycl.hpp entry header in compliance with SYCL 2020 [5edb228] [24d179c]
  • Added __LIBSYCL_[MAJOR|MINOR|PATCH]_VERSION macros, see PreprocessorMacros for more information [9f3a74c]
  • Added support for SYCL 2020 reductions with read_write access mode to reduction variables [733d5e3]
  • Added support for SYCL 2020 reductions with sycl::property::reduction::initialize_to_identity property [3473c1a]
  • Implemented zero argument version of sycl::buffer::reinterpret() for SYCL 2020 [c0c3c80]
  • Added an initial AOT implementation of the experimental matrix extension on the CPU device to target AMX hardware. Base features are supported [35db973]
  • Added support for sycl_ext_oneapi_local_memory extension [5a66fcb] [9a734f6]
  • Documented Level Zero backend [8994e6d]

Improvements

SYCL Compiler

  • Added support for math built-ins: fmax, fmin, isinf, isfinite, isnormal, fpclassify [1040b94]
  • The FPGA initiation interval attribute spelling [[intel::ii]] is deprecated. The new spelling is [[intel::initiation_interval]]. In addition, [[intel::initiation_interval]] may now be used as a function attribute, formerly its use was limited to statement attribute [b04e6a0]
  • Added support for function attribute [[intel::disable_loop_pipelining]] and [[intel::max_concurrency(n)]] [7324b3e]
  • Enabled -fsycl-id-queries-fit-in-int by default [f27bb01]
  • Added support for stdlib functions: abs, labs, llabs, div, ldiv, lldiv [86716c5] [2e9d33c]
  • Enabled range rounding for ESIMD kernels [25b482b] [bb20b7b]
  • Improved diagnostics on invalid kernel names [0c0f4c5]
  • Improved compilation time by combining device code compilation and integration header generation into one step [f110dd4]
  • Added support for sycl::queue::mem_advise for the CUDA backend [2b56ac9]

SYCL Library

  • Specialized atomic fetch_add, fetch_min and fetch_max for floating-point types [37a9a2a] [59ceaf4]
  • Added support for accessors to array types [7ed4f58]
  • Added sub-group information queries on CUDA [c36fa65]
  • Added support for sycl::queue::barrier in Level Zero plugin [7c31f90]
  • Improved runtime memory usage in Level Zero plugin [c9d71d4] [2ce2ca6] [46e3c64]
  • Added Level Zero interoperability with specifying of ownership [41221e2]
  • Improved runtime memory usage when using USM [461fa02]
  • Provided facility for user to control execution range rounding [f6ac45f]
  • Ensured correct access mode in sycl::handler::copy() method [b489479]
  • Disallowed for atomic accessors in sycl::handler::copy() method [14437db]
  • Provided move-assignability of usm_allocator class [05a805e]
  • Improved performance of copying data during native memory object creation on devices without host unified memory [ad8c9d1]
  • [ESIMD] Added implicit set up of fence before barrier as required by hardware [692228c]
  • Allowed for using of interoperability program constructor with multi-device context [c7f7674]
  • Allowed trace of Level Zero calls only with SYCL_PI_TRACE=-1 [ea73219]
  • Added throw of feature_not_supported when when upon attempt to create program using create_program_with_source with Level Zero or CUDA [ba77e3a]
  • Added support for inline cl namespace in debugger [8e441d4]
  • Added support for build with GCC 7 [d8fea22]
  • Added in-memory caching of programs built with custom build options [86b0e8d] [e152b0d]
  • Improved range rounding heuristics [7efb692]
  • Added get_backend methods to SYCL classes [ee7e99f]
  • Added sycl::sub_group::load and sycl::sub_group::store versions that take raw pointers [248f550]
  • Enabled caching of devices in sycl::device interoperability constructors [d3aeb4a]
  • Added a warning on using SYCL 1.2.1 OpenCL interoperability API when compiling in SYCL 2020 mode. It can be suppressed by defining SYCL2020_DISABLE_DEPRECATION_WARNINGS [a249316]
  • Added support for blitter engine in Level Zero plugin. Some memory operations are submitted to a Level Zero copy queue now [11ba5b5]
  • Improved sycl::INTEL::lsu::load and sycl::INTEL::lsu::store to take sycl::multi_ptr [697469f]
  • Added a diagnostic on attempt to compile a SYCL application without dynamic C++ RT on Windows [d4180f4]
  • Added support for Queue Order Properties extension for Level Zero [50005c7]
  • Improved plugin discovery mechanism - if a plugin fails to initialize others will be discovered anyway [d513074]
  • Added support for sycl::info::partition_affinity_domain::numa in Level Zero plugin [2ba8e05]

Documentation

  • Updated TBB paths in GetStartedGuide [a9acb70]
  • Aligned linked allocation document with recent changes [22b9d01]
  • Updated GetStartedGuide for building with libcxx [d3a74c3]
  • Updated table of contents in GetStartedGuide [0f401bf]
  • Filled in address spaces handling section in design documentation [f782c2a]
  • Improved design document for program cache [ed4b4c4]
  • Updated compiler options description [e56e576]
  • Updated SYCL_INTEL_sub_group extension document to use automatic instead of auto [c4d08f5]

Bug fixes

SYCL Compiler

  • Suppressed link time warning on Windows that incorrectly diagnosed conflicting section names while linking device binaries [8e6a3ec]
  • Disabled code coverage for device compilations [12a0b11]
  • Fixed an issue when unbundling a fat static archive and targeting non-FPGA device [90c79c7]
  • Addressed inconsistencies when performing compilations by using the target triple for FPGA (spir64_fpga-unknown-unknown-sycldevice) vs using -fintelfpga [c9a65fc]
  • Fixed generation of the output report folder when performing FPGA AOT compilations from a previously generated AOCR archive [eab4791]
  • Addressed issues dealing with improper settings when performing preprocessing when offloading is enabled [d03de03]
  • Fixed issue when using -fsycl-device-only on Windows when specifying an output file with /o [d1d6c5d]
  • Fixed inlining functions called from an ESIMD kernel, which broke code generation in the Intel GPU vector back-end [65b459d]
  • Fixed JIT crash on ESIMD kernels compiled with -fsycl-id-queries-fit-in-int [ad86c34]
  • Fixed compiler crash on ESIMD kernels calling external functions with gpu::simd arguments [dfaaaed]
  • Fixed issue with generating preprocessed output when using -fsycl-device-only [3d2225a]

SYCL Library

  • Fixed race-condition happening on application exit [8eb00d7] [c9c1de9]
  • Fixed faulty behaviour that happened when accessing a buffer in different contexts using discard_* access mode [f75b439]
  • Fixed support for SYCL_PROGRAM_LINK_OPTIONS and SYCL_PROGRAM_COMPILE_OPTIONS environment variables when compiling/linking through sycl::program class [9d74846]
  • Fixed deadlock in Level Zero plugin when batching enabled [645db17]
  • Fixed possible stack overflow in Level Zero plugin [ec6fbe1]
  • Fixed issues with empty wait list in Level Zero plugin [d8c8e08]
  • Added missing double3 and double4 support in geometric function cross() [b8afff4]
  • Fixed issue when using std::vector<bool> & argument for sycl::buffer::set_final_data() method [084d83a, 2a751bd]
  • Fixed support for long long in sycl::vec::convert() on Windows [5b49cd3]
  • Aligned local and image accessor with specification by allowing for property list in their constructor [88fab25]
  • Fixed support for offset in parallel_for for host device [1958715]
  • Added missing constructors for sycl::buffer class [bdfad9e]
  • Fixed coordinate conversion for sampler class on host device [cd6529f]
  • Fixed support for local accessors in debugger [fdacb75]
  • Fixed dropping of kernel attributes when execution range rounding is used [496f9a0] [677a7ea]
  • Added support for interoperability tasks that use get_mem() methods with Level Zero plugin [149f08d]
  • Fixed sub-device caching in the Level Zero plugin [0b18b49]
  • Fixed get_native methods to retain reference counter in case of OpenCL backend [ee7e99f]
  • Fixed sporadic failure happening due to illegal destruction of events before they have been signaled [2a76b2a]
  • Resolved a pinned host memory specific performance regression on CUDA that was introduced with the host unified behavior dependent logic [3be63ab]
  • Fixed illegal accesses that could happen when an application that uses host tasks exits without waiting for host tasks completion [552a521]
  • Fixed sycl::event::get_info queries that were working incorrectly when called on event without an encapsulated native handle [5d5a792]
  • Fixed compilation error with using multidimensional subscript for sycl::accessor with atomic access mode [0bfd34e]
  • Fixed a crash that happened when an accessor passed to a reduction was destroyed immediately after [b80f13e]
  • Fixed sycl::device::get_info with sycl::info::device::max_mem_alloc_size which was returning incorrect value in case of Level Zero backend [8dbaa53]

API/ABI breakages

  • None

Known issues

  • GlobalWorkOffset is not supported by Level Zero backend [6f9e9a76]
  • User-defined functions with the same name and signature (exact match of arguments, return type doesn't matter) as of an OpenCL C built-in function, can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using sycl::program/sycl::kernel_bundle API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Printing internal defines isn't supported on Windows [50628db]

January'21 release notes

Release notes for commit range 5eebd1e4bfce..2ffafb95f887

New features

SYCL Compiler

  • Added template parameter support for work_group_size attributes [ca9d816234c6] [289230c0b52c]
  • Introduced auto mode for device code split feature. This mode is the default one [184d258b902a]
  • Enabled support for multiple AOCX device binaries for FPGA [6ea38f0f1f7a]

SYCL Library

Documentation

Improvements

SYCL Compiler

  • Updated ESIMD support for split barrier, slm load and dp4a [7f5013901874]
  • Disabled swapping dimension values of WorkGroupAttr for semantic checks [7ef84b6fb60c]
  • Enabled clang-offload-deps tool usage for generating device dependencies [08a1c00028e5, 237675ba5a16, 84d2eb53e774, c8455167c2fa]
  • Switched to linking only needed symbols from fat static libraries on device side [5c3e5385b35d]
  • Made Intel attributes consistent with clang attributes [aeb4de7e4fa5]

SYCL Library

  • Added filtering of partition properties to exclude non-SYCL ones [7f7006410c1d]
  • Fixed memory leaks of global object in the Level-Zero plugins [f58c568]
  • Fixed excessive memory allocation for device handles in the OpenCL plugin [8cfcdb20b06e]
  • Added support for syc::ext::oneapi::property::buffer::use_pinned_host_memory buffery property in LevelZero plugin [0b9a7493a2f1]
  • Improved performance of kernels which do not use work-item free functions on the host device [b83a1a8028a8]
  • Switched to retaining events in LevelZero event wait list until command has completed [c55eb20c2f8c]
  • Improved group scan/reduce algorithms performance for CUDA [bd0edfee2908]
  • Enabled sub-group loads and stores in CUDA [cfce9965c366]

Documentation

  • Added notice of the -o build configuration option in GetStartedGuide [665e4062a5ff]
  • Added requirement to reference E2E tests for the pull-request in the commit message [965b63c38ca7]

Bug fixes

SYCL Compiler

  • Eliminated emission of llvm.used global variables to output module(s) in the post-link tool [9b07d114f5c2] [b96e647c82b2]
  • Fixed crash on scalar fptoui LLVM instruction [d7ca5dafcb3b]
  • Fixed detection of free function calls [6cfc3add16e3]
  • Put constant initializer list data in non-generic address space [0f2cf4ddc2fd]

SYCL Library

  • Fixed memory leak in Level-Zero plugin [d98b61a7bd28] [051e1407a195] [45e1b143f0a9]
  • (ESIMD only) Fixed build error "no member named cl_int2" [d7f1b53dec62]
  • Disabled execution range rounding for ESIMD kernels [5500262254c3]
  • Fixed argument printing when tracing plugin interface calls [30db15fac6f6]
  • Fixed as() function for swizzle vector [70e003acbbfe]
  • Fixed synchronization of sub-group and group functions on CUDA [2b6f2cd7ba0a]

API/ABI breakages

  • None

Known issues

  • GlobalWorkOffset is not supported by Level Zero backend [6f9e9a76]
  • The code with function pointer is hanging on Level Zero [d384295e]
  • User-defined functions with the same name and signature (exact match of arguments, return type doesn't matter) as of an OpenCL C built-in function, can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • On Windows, DPC++ compiler enforces using dynamic C++ runtime for application linked with SYCL library by:
    • linking with msvcrt[d].dll when -fsycl switch is used;
    • emitting an error on attempts to compile a program with static C++ RT using -fsycl and /MT or /MTd. That protects you from complicated runtime errors caused by C++ objects crossing sycl[d].dll boundary and not always handled properly by different versions of C++ RT used on app and sycl[d].dll sides.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Employing read sampler for image accessor may result in sporadic issues with Level Zero plugin/backend [2c50c03]
  • Printing internal defines isn't supported on Windows [50628db]
  • Group algorithms for MUL/AND/OR/XOR cannot be enabled for group scope due to SPIR-V limitations, and are not enabled for sub-group scope yet as the SPIR-V version isn't automatically raised from 1.1 to 1.3 [96da39e]
  • We cannot run Dead Argument Elimination for ESIMD since the pointers to SPIR kernel functions are saved in !genx.kernels metadata [cf10351]

December'20 release notes

Release notes for commit range 5d7e0925..5eebd1e4bfce

New features

SYCL Compiler

  • Allow for multiple build options for opencl-aot [5e5703f58449]

SYCL Library

  • Implement SYCL_INTEL_mem_channel_property extension [2f1f3167b7c6]
  • Add marray class as defined by SYCL 2020 provisional [5eebd1e4bfce]
  • Implement dynamic batch size adjusting when using Level-Zero plugin [c70b0477aa8a, cf0d0538d162]
  • Add online compilation API interface [70ac47d23264]

Documentation

Improvements

SYCL Compiler

  • (ESIMD only) Remove wrapping of buffer objects into images which caused problems like incorrect work of scatter/gather of 1- and 2-byte values [d2d20d6c4556]
  • Rename FPGA kernel attribute [[intel::stall_enable]] to [[intel::use_stall_enable_clusters]] [dab9debebe70]
  • Add template parameter support for [[intel::max_global_work_dim]] and [[intel::no_global_work_offset]] attributes [bd8fcc7dee34, a5fde5a924ac]
  • Remove partial-link path when dealing with fat static archives [f1aa7f4d8b79]
  • Remove unused device library function definitions from linked program [e9423ffdec92]
  • Don't dump IR and dot files by default in the LowerWGScope [9d0e3525ba04]
  • Support LLVM floating-point intrinsics in llvm-spirv and frontend [a5065ab85101]
  • Add template parameter support for [[intel::no_global_work_offset()]] attribute [a5fde5a924ac]
  • Improve group size selection by adjusting parallel_for execution range global size [74a68b7da4e7]
  • Add clang support for FPGA loop fusion function attributes [23926b0645ad]
  • Reword some compiler diagnostics [50b81c3cd6e9]

SYCL Library

  • Eliminate performance overhead on devices without host unified memory support [a4f092417ef9]
  • Optimize discard_write access mode for host accessor [6733c8b0efde]
  • Add support for composite specialization constants [c62860fd6b86, d4251e3c55e7, 3ec4594a5a06]
  • Enhance PI tracing with printing output arguments [19f5ad67f30a]
  • Introduce pi_map_flags in lieu of cl_map_flags [f0e7606a6198]
  • Implement robust error handling in LevelZero plugin [65c719ddfc23]
  • Add new device descriptors as SYCL extensions [51ac08c35294]
  • Remove redundant dependencies for in-order queues [632722165db2]

Documentation

  • Add information on AOT to GetStartedGuide [71942fbb3655]
  • Add notice on alignemnt checks in ABI policy [4326b9563575]
  • Updated design of specialization contants on using of POD types [81963d1ec055]
  • Document linked allocation commands [929a764a5ec4]
  • Improve ESIMD documentation rendering [079597d28f1f]
  • Improved device library documentation [f24e2a9ce464]

Bug fixes

SYCL Compiler

  • Do not customize optimizations for non-SPIR targets [cb069fed6712]
  • Fix address space assertion with templates [8905a8cec9a9]
  • Link libm-fp64 device library by default [ac93d6fe3d9d]
  • Add support for specialization constants' typenames declared in namespaces [f64f835b4313]
  • Fix loosing OpenMP device binary when program uses both OpenMP and SYCL offloading models [eb89f5eaab37]

SYCL Library

  • Add missing interoperability API to construct SYCL classes with Level-Zero handles [10b4e8a6fc19]
  • Fix several builtins implementation for host device [8b82c671ab12, 786708914fd4]
  • Fix possible hang upon application finish if streams were used [bd5893ae01b1]
  • Fix failure when employing interoperability host task on queue constructed with reused context [9cff6c9b6127]
  • Fix "instantiation after specialization" warnings [56b9a1dfb92f, eadce94f8ad0]
  • Support copying of stream by value within a kernel without loss of output information [8d37cbacc9b8]
  • Fix handling of big and/or non-uniform work-groups in reduction kernels. The solution may change when reduction kernels precompilation/query approach is implemented [78e2599bc499]
  • Fix memory leak in event pool in Level Zero plugin [68fc7808a50e]
  • Fixed issue with finalizing context of Level Zero plugin [6cfa921856f5]
  • Fix backend selection for SYCL_DEVICE_FILTER=* case [c54da157f5d5]
  • Restore AccessorImplHost layout [a08eeb475679]

Documentation

  • Updated source checkout instruction for Windows in GetStartedGuide [9cde15210d70]

API/ABI breakages

Known issues

  • GlobalWorkOffset is not supported by Level Zero backend [6f9e9a76]
  • The code with function pointer is hanging on Level Zero [d384295e]
  • User-defined functions with the same name and signature (exact match of arguments, return type doesn't matter) as of an OpenCL C built-in function, can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • On Windows, DPC++ compiler enforces using dynamic C++ runtime for application linked with SYCL library by:
    • linking with msvcrt[d].dll when -fsycl switch is used;
    • emitting an error on attempts to compile a program with static C++ RT using -fsycl and /MT or /MTd. That protects you from complicated runtime errors caused by C++ objects crossing sycl[d].dll boundary and not always handled properly by different versions of C++ RT used on app and sycl[d].dll sides.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Employing read sampler for image accessor may result in sporadic issues with Level Zero plugin/backend [2c50c03]
  • Printing internal defines isn't supported on Windows [50628db]
  • Group algorithms for MUL/AND/OR/XOR cannot be enabled for group scope due to SPIR-V limitations, and are not enabled for sub-group scope yet as the SPIR-V version isn't automatically raised from 1.1 to 1.3 [96da39e]
  • We cannot run Dead Argument Elimination for ESIMD since the pointers to SPIR kernel functions are saved in !genx.kernels metadata [cf10351]

November'20 release notes

Release notes for commit range c9d50752..5d7e0925

New features

  • Implemented support for new loop attribute(intel::nofusion) for FPGA [68ab67ad]
  • Implemented support for new FPGA function attribute stall_enable [8fbf4bbe]
  • Implemented accessor-based gather/scatter and scalar mem access for ESIMD feature [0aac708a]
  • Implemented support for dot_product API [6cc97d2a]
  • Implemented ONEAPI::filter_selector that accepts one or more filters for device selection [174fd168]
  • Introduced SYCL_DEVICE_FILTER environment variable allowing to filter available devices [14e227c4], [ccdf8475]
  • Implemented accessor_properties extension [f7d073d1]
  • Implemented SYCL_INTEL_device_specific_kernel_queries [24ae95b3]
  • Implemented support for group algorithms on CUDA backend [909459ba]
  • Implemented support for sub_group extension in CUDA backend [baed6a5b], [551d7067], [f189e413]
  • Implemented support for USM extension in CUDA plugin [da8929e0]
  • Implemented support for cl_intel_create_buffer_with_properties extension [b8a7b012]
  • Implemented support for sycl::info::device::host_unified_memory [08066b24]
  • Added clang support for FPGA kernel attribute scheduler_target_fmax_mhz [20013e23]

Improvements

SYCL Compiler

  • Enabled USM address space by default for the FPGA hardware [7896819d]
  • Added emitting of a warning when size of kernel arguments exceeds 2kB for all devices [e00ab746], [4960fc90]
  • Changed default SYCL standard version to 2020 [67acf814]
  • Added diagnostics when the translator encounters an unknown or unsupported LLVM instruction [5a28d4e5]
  • Added diagnostic for attempt to pass a pointer to variable length array as kernel argument [538c4c9c]
  • Improved FPGA AOT help output with -fsycl-help [dc8a0593]
  • Made /MD the default option of compiler on Windows and made driver generate error if /MT was passed to it. SYCL library is designed in such a way that STL objects must cross the sycl.dll boundary, which is guaranteed to work safe on Windows only if the runtime in the app using sycl.dll and in sycl.dll is the same and is dynamic [d31184e1], [8735bb81], [0092d4da]
  • Enforced C++ for C source files when compiling in SYCL mode [adc2ac72]
  • Added use of template parameter in [[intelfpga::num_simd_work_items()]] attribute [678911a8]
  • Added new spellings for SYCL FPGA attributes [5949228d], [b1cf776e9]
  • All definitions used for compiler needs were marked with underscore prefix [51d3c205]
  • Disabled diagnostic about use of functions with raw pointer in kernels [b4a3f03f]
  • Improved diagnostics for invalid SYCL kernel names [cb5ddb49], [89fd4284]

SYCL Library

  • Removed deprecated spelling ([[cl::intel_reqd_sub_group_size()]]) of IntelReqdSubGroupSize attribute [9dda36fe]
  • Added support for USM shared memory allocator for Level Zero backend [db5037ca]
  • Added support for a context with multiple device in Level Zero backend [129ee442]
  • Added diagnostics for deprecated environment variables: SYCL_BE and SYCL_DEVICE_TYPE [6242160b]
  • Made spec_constant default constructor public and available on host [53d909e2]
  • Added constraints instead of static asserts and is_native_function_object() for group algorithms [97bec247]
  • Added support for L0 loader validation layer [4c6cda3f]
  • Added multi-device and multi-platform support for SYCL_DEVICE_ALLOWLIST [dbf31c3c]
  • Removed two-input sub-group shuffles [ef969c14]
  • Enabled inspecting values wrapped into private_memory by evaluating operator() from GDB [31c23ddc]
  • Changed buffer allocation in the Level Zero plugin to use host shared memory for integrated GPUs [2ae1bc9e]
  • Implemented queue::parallel_for() accepting reduction [ffdadc2e]
  • Improved performance of float atomic_ref [0b7dacf1]
  • Made CUDA backend try to find a better block size using cuOccupancyMaxPotentialBlockSize function from the CUDA driver [4fabfd16a]
  • Supported GroupBroadcast with 32-bit id to cover broadcast algorithm with the sub_group class [6e3f2440]

Documentation

  • Added specification for SPV_INTEL_variable_length_array extension [9e4c51c4]
  • Added specification for accessor_properties and buffer_location extensions [f90614c5]
  • Moved specification for Unified Shared Memory to Khronos specification [a7ffe039]
  • Added documentation for filter_selector [c3f5cfba]
  • Updated C-CXX-StandardLibrary extension specification [0b6f8cd8]
  • Added ESIMD extension introduction document [c36a1411]
  • Added specialization constants extension introduction document [d88ef3b6]
  • Added specialization constant feature design doc [15cac431]
  • Documented kernel-program caching mechanism [5947cde81]
  • Added the SYCL_INTEL_mem_channel_property extension specification [5cf8088c]
  • Provided detailed description for guaranteed sub-group sizes[542c32ae]
  • Documented test-related processes [ff90e06d]
  • Added code examples for all SYCL FPGA loop attributes [6b958205]

Bug fixes

SYCL Compiler

  • Fixed crash of compiler on invalid kernel type argument. [0c220ca5e]
  • Made clang search for the full executable name as opposed to just the case name for the AOT tools (aoc, ocloc, opencl-aot) to avoid directory calls [78a86da3], [244e874b]
  • Linked SYCL device libraries by default as not all backends support SPIRV online linkage [9dd18ca8]
  • Fixed assertion when /P option is used on windows[a21d7ef4]
  • Fixed crash when array of pointers is passed to kernel[1fc0e4f84]
  • Fixed issues with use of type from std namespace in kernel type names [dd7fec83]
  • Fixed debug information missed for work-item built-in translation [9c06d429]
  • Added warnings emission which had been suppressed for SYCL headers [e6eed1a7]
  • Fixed optimization disabling option for gen to use -cl-opt-disable [ba4e567fe]
  • Emulated "funnel shift left" which was not supported in the OpenCL ExtInst set on SPIRV translator side [97d7eec5]
  • Fixed build issue when TBB and libstdc++ 10.X were used caused by including std C++ headers in integration header file [63369132]
  • Fixed processing for partial link step with static archives by passing linker specific arguments there [3ab8cc82]
  • Enabled -reuse-exe support for Windows [43f2d4ba]
  • Fixed missing dependency file when compiling for -fintelfpga and using a named dependency output file [df5f1ab67]

SYCL Library

  • Fixed build log preserving for L0 plugin [638b71b1]
  • Added missing math APIs for devicelib [e438bc814]
  • Enabled async_work_group_copy for scalar and vector bool types [bb78d2cb]
  • Aligned image class constructors with the SYCL specification [049ae996]
  • Removed half type alias causing name conflicts with CUDA headers [c00c1fa3]
  • Fixed explicit copy operation for host device [f20fd4de]
  • Made stream flush operation non-blocking [e7492fb2]
  • Fixed image arguments order when passing to PI routines [70d6f87b]
  • Fixed circular dependency between the device_impl and the platform_impl causing handler leak [255f304f]
  • Fixed work-group size selection in reductions [2ae49f57e]
  • Fixed compilation errors when built with --std=c++20 [ecd0adbb]
  • Fixed treating internal allocations of host memory as read only for memory objects created with const pointer, causing double free issue [8b5506255]
  • Fixed a problem in Level Zero plugin with kernels and programs destruction while they can be used [b9bf9f5f]
  • Fixed wrong exception raised by ALLOWLIST mechanism [d81081f7]
  • Fixed reporting supported device partitioning in Level Zero [766367be]
  • Aligned get_infoinfo::device::version() with the SYCL spec [4644e639]
  • Set default work group size to {1, 1, 1} to fix out-of-memory crashes on some configurations [4d76de43]

Documentation

  • Fixed path to FPGA device selector [ca33f7f7]
  • Renamed LEVEL0 environment variable to LEVEL_ZERO in documents and code comments following source code change [2c3908b4]
  • Clarified --system-ocl key in GetStartedGuide.md [e31b94e5]

API/ABI breakages

  • Implemented accessor_properties extension for accessor class [f7d073d1]

Known issues

  • GlobalWorkOffset is not supported by Level Zero backend [6f9e9a76]
  • The code with function pointer is hanging on Level Zero [d384295e]
  • If an application uses std::* math function in the kernel code the -fsycl-device-lib=libm-fp64 option should be passed to the compiler.
  • User-defined functions with the same name and signature (exact match of arguments, return type doesn't matter) as of an OpenCL C built-in function, can lead to Undefined Behavior.
  • A DPC++ system that has FPGAs installed does not support multi-process execution. Creating a context opens the device associated with the context and places a lock on it for that process. No other process may use that device. Some queries about the device through device.get_info<>() also open up the device and lock it to that process since the runtime needs to query the actual device to obtain that information.
  • On Windows, DPC++ compiler enforces using dynamic C++ runtime for application linked with SYCL library by:
    • linking with msvcrt[d].dll when -fsycl switch is used;
    • emitting an error on attempts to compile a program with static C++ RT using -fsycl and /MT or /MTd. That protects you from complicated runtime errors caused by C++ objects crossing sycl[d].dll boundary and not always handled properly by different versions of C++ RT used on app and sycl[d].dll sides.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • The SYCL library doesn't guarantee stable API/ABI, so applications compiled with older version of the SYCL library may not work with new one. The workaround is to rebuild the application. ABI policy guide
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Employing read sampler for image accessor may result in sporadic issues with Level Zero plugin/backend [2c50c03]
  • Printing internal defines isn't supported on Windows [50628db]
  • Group algorithms for MUL/AND/OR/XOR cannot be enabled for group scope due to SPIR-V limitations, and are not enabled for sub-group scope yet as the SPIR-V version isn't automatically raised from 1.1 to 1.3 [96da39e]
  • We cannot run Dead Argument Elimination for ESIMD since the pointers to SPIR kernel functions are saved in !genx.kernels metadata [cf10351]

September'20 release notes

Release notes for commit range 5976ff0..1fc0e4f

New features

Improvements

SYCL Compiler

  • Assigned the source location of the kernel caller function to the artificial initialization code generated in the kernel body. It enables profiling tools to meaningfully attribute the initialization code [6744364]
  • Provided compile-time warning if size of kernel arguments exceeds 2KiB in GPU AOT mode [e00ab74]
  • Changed default SYCL standard to SYCL-2020 [67acf81]
  • Removed deprecated [[cl::intel_reqd_sub_group_size(N)]] attribute [9dda36f]
  • Enable USM address spaces by default for the FPGA hardware [7896819]
  • Assume SYCL device functions are convergent [047e2ec]
  • Added Dead Argument Elimination optimization [b0d98dc] [f53ede9]
  • Simplified the error checking of arrays by only visiting once [c709986]
  • Stop emitting kernel arguments metadata [f658918]
  • Enabled -f[no-]sycl-early-optimizations on Windows [e1e3658]
  • Mutable kernel functions are now explicitly forbidden in SYCL 2020 [1dbc358]
  • Moved hardware targeted extensions to INTEL namespace [3084982]
  • Added support for union types as kernel parameters [5adfd79]
  • Renamed -fsycl-std-optimizations to -fsycl-early-optimizations [077a507]
  • Added support for -f[no-]sycl-id-queries-fit-in-int. Enabling this will make compiler define _SYCL_ID_QUERIES_FIT_IN_INT_ macro which will signal runtime to emit __builtin_assume() for execution range less than INT_MAX limitation [3e4da3c]
  • Enabled template trail for kernel diagnostics [c767edc]
  • Disabled createIndVarSimplifyPass for SPIR target in SYCL mode [76ffef7]
  • Run Dead Argument Elimination when LLVM optimizations are applied as well [cf10351]

SYCL Library

  • Eliminated circular dependency between event and queue classes [31843cc]
  • Added ONEAPI::filter_selector [174fd168f18]
  • Added CPU-agnostic code path to the host device runtime (validated on AArch64 systems) [2f632f8]
  • Added support for bool2, bool3, bool4, bool8, bool16 [4dfd500]
  • Allowed for creating lots of host accessors [b206293]
  • Improved execution graph traversal [f2eaa23]
  • Improved SYCL_PI_TRACE [4d468f1]
  • Added implementation for SYCL_INTEL_free_function_queries [b6d7792]
  • Allowed for building program for multiple devices within single context (esp. for FPGA devices) [2f64227]
  • Cache devices and platforms [d392b51]
  • Reuse devices and platforms in Level Zero PI plugin [43ba606]
  • Added group algorithms for MUL/OR/XOR/AND operations [96da39e]
  • Moved general language extensions to ONEAPI namespace [a73369d]
  • Added CMake option SYCL_DISABLE_STL_ASSERTIONS to disable assertions [ec2ec99]
  • Implemented USM fill operation as defined in SYCL-2020 provisional [4993646]
  • Added runtime support for device code argument elimination [63ac3d3]
  • Imporved implementation of stream class when used in FPGA device code [13e8dae]
  • Imporved error reporting in Level Zero plugin [257658c]
  • Improved kernel demangling in graph printing [62192a6]
  • Improved error handling in parallel_for [7c73c11]
  • Fixed segfault in interop constructors of context, device, platform classes [c4c3494]

Documentation

Bug fixes

SYCL Compiler

  • Fixed crash when array of pointers is a kernel argument [1fc0e4f]
  • Allowed for -P -fsycl to be used on Windows when offloading [a21d7ef]
  • Fixed looking for tools (e.g. aoc, ocloc, opencl-aot) with full name on Windows (i.e. with .exe suffix) [78a86da]
  • Eliminated compiler crash if invalid declaration is used as kernel argument [0c220ca]
  • Switch SPIRV debug info to legacy mode to support old OpenCL RTs [500a0b8]
  • Disabled vectorizers in SYCL device code when early optimizations are enabled [20921b1]
  • Fixed crash when kernel argument is a multi-dimensional array [36f6ab6]
  • Fixed cl::sycl::INTELlsu::load() method to return value instead of reference [82e5323]
  • Disabled "early" optimizations for Intel FPGA by default [f8902b8]
  • Fixed regression on unused non-USM pointers inside struct type kernel arguments [926eb32]
  • Fixed NULL-pointer dereference in some cases [bdc2b85]
  • Adjusted AUX targets with lang options [43862a3]

SYCL Library

  • Eliminated circular dependency between command group and stream buffers, which caused memory leaking [841e1e7]
  • Added early exit from enqueue process when trying to enqueue blocked commands. This eliminated hang in host-task when used along with multiple buffers [bc8f0a4]
  • Fixed overflow when casting glbal memory size in Level Zero plugin [82893b2]
  • Fixed waiting for events on Level Zero [e503662]
  • Added missing constructors and propety methods for context, program and sampler[30b8acc]
  • Fixed printing types of variables by GDB in some cases [93e1387]
  • Aligned cl::sycl::handler::require API with the SYCL specification [68c275c]
  • Fixed undefined behaviour in memory management intrinsics [4ff2eee]
  • Fixed race condition when using sampler in parallel [34f0c10]
  • Fixed race condition in ProgramManager class, which lead to hang [e6fd911]
  • Fixed thread-safety issue, which took place when using stream class [4688cb3]
  • Unified usm queue's memcpy/memset methods behavior for corner cases [7b7bab6]
  • Enabled USM indirect access for interoperability kernels [ebf5c4e]

API/ABI breakages

  • Added missing constructors and propety methods for context, program and sampler[30b8acc]

Known issues

  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • The SYCL library doesn't guarantee stable API/ABI, so applications compiled with older version of the SYCL library may not work with new one. The workaround is to rebuild the application. ABI policy guide
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 and user specifies -std=c++14 or /std:c++14.
  • Employing read sampler for image accessor may result in sporadic issues with Level Zero plugin/backend [2c50c03]
  • Printing internal defines isn't supported on Windows [50628db]
  • Group algorithms for MUL/AND/OR/XOR cannot be enabled for group scope due to SPIR-V limitations, and are not enabled for sub-group scope yet as the SPIR-V version isn't automatically raised from 1.1 to 1.3 [96da39e]
  • We cannot run Dead Argument Elimination for ESIMD since the pointers to SPIR kernel functions are saved in !genx.kernels metadata [cf10351]

August'20 release notes

Release notes for the commit range 75b3dc2..5976ff0

New features

Improvements

SYCL Compiler

  • [CUDA BE] Removed unnecessary memory fence in the sycl::group::barrier implementation which should improve performance [e2fc1b8]
  • [CUDA BE] Added support for the SYCL builtins from relational, geometric, common and math categories [d4e7929] [d9bad0b] [0c9c9c0] [99957c5]
  • Added support for C array as a kernel parameter [00e7308]
  • [CUDA BE] Added support for kernel offset [c7bb288]
  • [CUDA BE] Added support for sycl::half type [8444189][8f39763]
  • Added support for SYCL kernel inheritance and nested arrays [0b2de9e]
  • Added a diagnostic on attempt to use const static data members that are not const-initialized [bde1085]
  • Added support for a set of standard library functions for AOT compilation [2bd5dab]
  • Allowed use of function declarators with empty parentheses [a4f2182]
  • The fallback implementation of standard library functions is now linked to the device code, only if such functions are used in kernels only [9a8864c]
  • Added support for recursive function calls in a constexpr context [06f667a]
  • Added a diagnostic on attempt to capture this as a kernel parameter [1b9f026]
  • Added [[intel::reqd_sub_group_size()]] attribute as a replacement for [[cl::reqd_sub_group_size()]] which is now deprecated [b2da2c8]
  • Added propagation of attributes from transitive calls to the kernel[5c91609]
  • Changed the driver to pass corresponding device specific options when -g or -O0 is passed [31eb425]
  • The sycl::usm_allocator has been improved. Now it has equality operators and can be used with std::allocate_shared. Disallowed usage with device allocations [ce915ef]
  • Added support for lambda functions passed to reductions [115c1a0]
  • Enabled standard optimization pipeline for the device code by default. The new compiler flag can be used to disable optimizations at compile time: -fno-sycl-std-optimizations [5976ff0]

SYCL Library

  • Added support for braced-init-list or a number as range for sycl::queue::parallel_for family functions [17299ee]
  • Finished implementation of parallel_for simplification extension [af792cb]
  • Added 64-bit type support for to load and store methods of sycl::intel::sub_group [fe8d852]
  • [CUDA BE] Do not enable event profiling if it's not requested by passing sycl::property::queue::enable_profiling property [bbe8457]
  • Sub-group support has been aligned with the latest changes to the extension document [bea6aa2]
  • [CUDA BE] Optimized waiting for event completion by synchronizing with latest event for a queue [d7ee359]
  • Finished implementation of the Host task with interop capabilities extension [f088e38]
  • Added builtins for one-element sycl::vec for host device [073a36b]
  • [L0 BE] Added support for specialization constants [be4e641]
  • Improved diagnostic on attempt to submit a kernel with local size which doesn't math value specified in the sycl::intel::reqd_work_group_size attribute for the kernel [03ef819]
  • [CUDA BE] Changed active context to be persistent [296fa1a]
  • [CUDA BE] Changed default gpu architecture for device code to SM_50 [800e452]
  • Added a diagnostic on attempt to create a device accessor from zero-sized buffer [80b2110]
  • Changed default backend to Level Zero [11ef88c]
  • Improved performance of the SYCL graph cleanup [c099e47]
  • [L0 BE] Added support for sycl::sampler [f3b8cdf]
  • Added support for TriviallyCopyable types to the sycl::intel::sub_group::shuffle [d3c7b20]
  • Implemented range simplification for queue Shortcuts [4009b8b]
  • Changed sycl::accessor::operator[] to return const reference when access mode is sycl::access::mode::read_only [03db009]
  • Exceptions thrown in a host task are now will be returned as asynchronous exceptions [280b93c]
  • Fixed sycl::buffer constructor which takes a contiguous container to enable copy back on destruction.
  • Added support for user-defined sub-group reductions [728429a]
  • The sycl::backend::level0 has been renamed to sycl::backend::level_zero [215f591]
  • Extended sycl::broadcast to support TriviallyCopyable types [df6d715]
  • Implemented get_native and make_* functions for Level Zero allowing to query native handles of SYCL objects and to create SYCL objects by providing a native handle: platform, device, queue, program. The feature is described in the SYCL 2020 provisional specification [a51c333]
  • Added support for sycl::intel::atomic_ref from sycl_ext_oneapi_extended_atomics extension

Documentation

Bug fixes

SYCL Compiler

  • Fixed the diagnostic on cl::reqd_sub_group_size attribute mismatches [75b3dc2]
  • Fixed the issue with empty input for -foffload-static-lib option [8c8137f]
  • Fixed a problem with template instantiation during integration header generation [4ba61d0]
  • Fixed a problem which could happen when using a command lines with large numbers of files [87b94d5]
  • Fixed a crash when a kernel object field is an array of structures [b00fb7c]
  • Fixed issue which could prevent using of structures with constant-sized arrays as a kernel parameter [a4a7950]
  • Fixed a bug in the pass for lowering hierarchical parallelism code (SYCLLowerWGScope). Transformation was generating the code where work items hit the barrier in the loop different number of times which is illegal [a4a7950]
  • Fixed crash on attempt to use objects of sycl::experimental::spec_constant in the struct [d5a7f20]

SYCL Library

  • Fixed problem with waiting on the same events several times which could happen when using USM [9bf602c]
  • Fixed a memory leak of sycl::event objects happened when using USM specific sycl::queue methods [a285b9d]
  • Fixed problem which could lead to a crash or deadlock when using sycl::handler::codeplay_host_task extension [e911de7]
  • Workarounded the problem which happened when an application uses long kernel names [b1b8510]
  • Fixed race which could happen when submitting the same kernel from multiple threads [95d3ec6]
  • [CUDA BE] Fixed a memory leak related to unreleased events [d0a148a]
  • [CUDA BE] Fixed diagnostic on attempt to fetch profiling info for commands which profiling is not enabled for [76bf2ed]
  • [L0 BE] Fixed memory leaks of device objects [eae48f6][6acb812][39e77733]
  • [CUDA BE] Fixed a problem with that several operations were not profiled if required [a420e7a]
  • Fixed a possible race which could happen when an application builds an object of the sycl::program or submits kernels from multiple threads [363ad5f]
  • Fixed a memory leak of queue and context handles, which happened when backend is not OpenCL [9ddca50]
  • [CUDA BE] Fixed 3 dimensional buffer device to device copy [d917446]
  • Fixed one of the sycl::queue constructors which was ignoring sycl::property::queue::enable_profiling property [7863c0b]
  • Fixed endless-loop in sycl::intel::reduction for the data types not having fast atomics in case of local size is 1 [e6b6ae7]
  • Fixed a compilation error which happened when using sycl::interop_handle::get_native_mem method with an object of sycl::accessor created for host target [280b93c]
  • Fixed a deadlock which could happen when multiple threads try to build a program simultaneously
  • Aligned sycl::handler::set_arg with the SYCL specification [a6465c9]
  • Fixed an issue which could lead to "No kernel named was found" exception when using sycl::handler::set_arg method [a08674e]
  • Fixed sycl::device::get_info<cl::sycl::info::device::sub_group_sizes> which was return incorrect data [e65841b]

API/ABI breakages

  • The memory_manager API has changed
  • Layout of internal classes for sycl::sampler and sycl::stream have been changed

Known issues

  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • The SYCL library doesn't guarantee stable API/ABI, so applications compiled with older version of the SYCL library may not work with new one. The workaround is to rebuild the application. ABI policy guide
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 The workaround is to enable -std=c++17 for the failing MSVC version.

June'20 release notes

Release notes for the commit range ba404be..24726df

New features

Improvements

SYCL Compiler

  • Now when -fintelfpga option is passed, the dependency file is created in the temporary files location instead of input source file location [7df381a]
  • Made -std=c++17 the default for DPC++ [3192ee7]
  • Added support for kernel name types templated using enumerations [e7020a1][f9226d2][125b05c][07e8d8f]
  • Added a diagnostic on attempt to use host built-in functions inside device code [2a4c1c8]
  • Added diagnostics on attempt to use sycl::accessor created for unsupported types in the device code [6da42a0]
  • Aligned sizeof(long double) between host and device code [87e6240]
  • The pragma spelling for SYCL-specific attributes except cl::reqd_work_group_size are rejected now [8fe2846]
  • Added template parameter support for cl::intel_reqd_sub_group_size attribute [0ae9729]
  • Added support for more math builtins for PTX target [9370549]
  • Added support for struct members and pointers in intelfpga::ivdep attribute [358ec04]
  • Added support for all builtins from integer and shared categories for PTX target [f0a4fe2]
  • Improved handling of linker inputs for static lib processing [ed2846f]
  • Dependency files are not generated by default when compiling using -fsycl -fintelfpga options now [24726df]

SYCL Library

  • Updated the implementation to align with changes in SubGroup extension [9d4c284]
  • sycl::ordered_queue class has been removed [875347a]
  • Added support of rounding modes for floating and integer types in sycl::vec::convert [096d0a0]
  • Added support for USM vars and placeholder accessors passed to reduction version of sycl::handler::parallel_for [94cb022][2e73da7]
  • Added support of sycl::intel::sub_group::load/store which take sycl::multi_ptr with sycl::access::address_space::local_space [0f5b55b]
  • Added a diagnostic on attempt to recompile an AOT compiled program using sycl::program API [b031186]
  • Started using custom CUDA context by default as it shows better performance results [9d45ead]
  • Prevented NVIDIA OpenCL platform to be selected by a SYCL application [7146426]
  • Adjusted the diagnostic message on attempt to use local size which is greater than global size to be more informative [894c10d]
  • Added a cache for PI plugins, so subsequent calls for sycl::device creation should be cheaper [03dd60d]
  • A SYCL program will be aborted now if program linking is requested when using Level Zero plugin. This is done because L0 doesn't support program linking [d4a5b71]
  • Added a diagnostic on attempt to use sycl::program::set_spec_constant when the program is already in compiled or linked state [e2e3d3d]
  • Improved sycl::stream class implementation on the device side in order to reduce local memory consumption [b838f0e]

Documentation

Bug fixes

SYCL Compiler

  • Fixed device code compile options passing which could lead to CL_INVALID_COMPILER_OPTIONS error [57bad9e]
  • Fixed a problem with creating a queue for FPGA device as a global inline variable [357e9c8]
  • Fixed an issue with that functions which are marked with SYCL_EXTERNAL are not participate in attribute propogation and conflicting attributes checking [0098eab]
  • Fixed an issue which could lead to problems when a kernel name contains a CVR qualified type [62e2f3b]
  • Fixed file processing when using -fsycl-link, now the generated object file can be linked by a non-SYCL enabled compiler/linker [2623abe]

SYCL Library

  • Fixed an issue with map/unmap events which caused problems with read only buffer accessors in CUDA backend [bf1b5b6]
  • Fixed errors happened when using sycl::handler::copy with const void*, void* or a sycl::accessor for a type with const qualifier [ddc0c9d]
  • Fixed an issue with copying memory to itself during sycl::buffer copyback [4bf22cc]
  • Fixed a possible deadlock which could happen when simultaneously submitting and waiting for kernels from multiple threads on Windows [ebace77]
  • Fixed a problem which caused device with a negative score to be still selected [7146426][855d214]
  • Fixed memleak which happened when using sycl::program::get_kernel [ccefc93]
  • Fixed memory copy being wrongly asynchronous which could cause data races on CUDA backend [4f0a3df]
  • Fixed a race which could happen when waiting for the same event from multiple threads [5737ad9]
  • Fixed errors which happened when using half or double types in reduction version of sycl::handler::parallel_for
  • Fixed sycl::device::get_info<sycl::info::device::mem_base_addr_align> query which was returning incorrect result for CUDA plugin [a6d03f3]
  • Fixed incorrect behavior of a sycl::buffer created with non-writable host data(e.g. const int *) on CUDA backend [49b6223]
  • A bunch of fixes to reduction version of sycl::handler::parallel_for:
    • Enabled operator*, operator+, operator|, operator&, operator^= for corresponding transparent functors used in reduction
    • Fixed the case when reduction object is passed as an R-value
    • Allowed identity-less constructors for reductions with transparent functors
    • Replaced some auto declarations with Reduction::result_type and added intermediate assignments/casts to avoid type ambiguities caused by using sycl::half type, and which may also be caused by custom/user types as well
    • Fixed compile time known identity values for MIN and MAX reductions

API/ABI breakages

  • All functions related to sycl::ordered_queue have been removed
  • Removed symbols corresponding to sycl::info::kernel_sub_group::max_sub_group_size_for_ndrange and sycl::info::kernel_sub_group::sub_group_count_for_ndrange queries

Known issues

  • [new] If there is an attribute cl::intel_reqd_sub_group_size with the same value for kernel and function called from the kernel there still can be compilation error.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • The SYCL library doesn't guarantee stable API/ABI, so applications compiled with older version of the SYCL library may not work with new one. The workaround is to rebuild the application. ABI policy guide
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 The workaround is to enable -std=c++17 for the failing MSVC version.

May'20 release notes

Release notes for the commit range ba404be..67d3d9e

New features

  • Implemented reduction extension for sycl::handler::parallel_for accepting a sycl::nd_range object [bb73d926] [04a360a] [05625f1]
  • XPTI instrumentation has been added to the runtime to capture semantic and execution trace information for constructing the task graphs for offline graph and performance analysis [9bf81eb] [cece82e]

Improvements

SYCL Frontend and driver changes

  • Added a diagnostic for implicit declaration of kernel function type [20e8cde]
  • Added a diagnostic on attempt to set C compilation together with -fsycl [4cf1610]
  • Improved diagnostics on attempt to use not supported on device types (such as __int128, zero length arrays) to catch cases when they are used as typedef or auto [808c5c8]
  • Reduced possibility to load incorrect version of the OpenCL headers by reording default include paths.
  • Treat .lo file extension as a static archive [82b93be]
  • Added a diagnostic on attempt to use long double in the device code [62f841d]
  • Improved handling of AOCX based archives on Windows [f117aa4]
  • Removed object format designator used for Windows device binaries [0052d08]
  • Improved -Xsycl-target option parsing [f62de21]
  • -g and -O0 options are now imply -g and -cl-opt-disable for the device compilation [779a601]

SYCL headers and runtime

  • Improved handling of host accessors with read-only type of access. Now they do not trigger redundant memory copy operation [de85c35e]
  • Default selector doesn't select devices of accelerator type anymore [b217bc5]
  • Added more user friendly diagnostics on errors during kernel submission [be42ff7]
  • Implemented get_native for CUDA backend which allows querying native handles of SYCL objects: sycl::queue, sycl::event, sycl::context, sycl::device [2d71c8e]
  • Added support for 0-dim sycl::accessor in sycl::handler::copy [5666107] [aedd449]
  • Added support for various rounding modes for non-host devices in sycl::vec::convert method [7e3cca4]
  • Added support for more image channel types for half4 data type on the host device [9b1d8b8]
  • Changed logic of SYCL_BE environment variable which now forces the SYCL RT to consider only devices of the specified backend during the device selection [937fec1]
  • libsycl.so library is now versioned [4370630]
  • [fd14167]
  • Added a diagnostic on attempt to construct a sycl::queue passing a sycl::context which is not bound to a passed sycl::device[9e79d31]

Misc

  • opencl-aot tool is now included in the sycl-toolchain target [ccc0c27]
  • [cece82e]
  • Added sycl-ls utility for listing devices discovered/selected by SYCL RT [cc0c33b]

Documentation

Bug fixes

SYCL Frontend and driver changes

  • Fixed bug in hierarchical parallelism implementation related to using a private address of the parallel_for_work_group lambda object by all work items in the work group [16f64b8]
  • Fixed a problem with errors that are reported when -fsycl-device-only option is used [7f924a8]
  • Fixed a crash which happened when a specialization constant is referenced twice [ce020c9]
  • Fixed a possible filename collision when -save-temps is used [99fa86f]
  • Fixed a hang which could happen on CUDA backend when sycl::group::async_work_group_copy is used [f836604]

SYCL headers and runtime

  • Ignore exceptions that could happen during sycl objects destruction to avoid terminates [cbd8a72]
  • Fixed an issue with dependency vector invalidation in some cases [2551b19]
  • Resolved a conflict with min/max macro that can be defined by windows.h [e7d4537]
  • Fixed an issue with using wrong CUDA context when creating an event [6788713]
  • Fixed a crash which happened when no plugin is available [d15de0b]
  • Fixed float to half-type conversion for small numbers [3a9a1a22]
  • Fixed sycl::intel::sub_group::broadcast which was incorrectly mapped to SPIRV intrinsic [24471bc]
  • Fixed an issue when a sub-buffer was accessing incorrect memory [4c4054b]
  • Fixed race which could happen when sycl::program API is used from multiple threads [e88a611]
  • [686e32b]
  • Make cl::sycl::event::get method to be constant [03208c0]
  • Fixed a crash which could happen when host accessors are created in multiple threads [1d13f84]
  • Fixed a problem with sycl::program::set_spec_constant which set a value for all sycl::program objects rather than for an object it was called for only [c22e34b]

Misc

  • Fixed an issue with gdb xmethod not working when run command is issued multiple times without terminating the session [042d981]
  • Fixed an issue with platform selection in opencl-aot tool that was caused by an identical supported platform name for CPU and GPU.

Known issues

  • [new] A crash can happen in a multithreaded application if two threads call an API which implies waiting for an event. No known workaround.
  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • The SYCL library doesn't guarantee stable API/ABI, so applications compiled with older version of the SYCL library may not work with new one. The workaround is to rebuild the application. ABI policy guide
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 The workaround is to enable -std=c++17 for the failing MSVC version.

March'20 release notes

Release notes for the commit range e8f1f29..ba404be

New features

  • Initial CUDA backend support [7a9a425]
  • [SYCL][FPGA] Implement IO pipes interface [c900248]
  • Added the implementation of GroupAlgorithms extension [8bfa107]
  • Added a partial implementation of sub group algorithms extension [017af4e]
  • New attributes for Intel FPGA devices: intelfpga::force_pow2_depth, intelfpga::loop_coalesce, intelfpga::speculated_iterations, intelfpga::disable_loop_pipelining, intelfpga::max_interleaving [73dd705][a5b9804]
  • Added support for intel::reqd_work_group_size attribute [8eb588d]
  • Added support for specialization constants feature which is based on SYCL Specialization Constant proposal [29abe37]

Improvements

SYCL Frontend and driver changes

  • Added a diagnostic on attempt to declare or use non-const static variable inside device code [7743e86] [1853516]
  • Relaxed requirements for kernel types even more. Now by default they should have trivial copy constructor and trivial destructor [17aac3c]
  • Changed std::numeric_limits<sycl::half> to constexpr functions [85d7a5e]
  • Added a diagnostic on attempt to use zero length arrays inside device code [e6ce614]
  • Added support for math functions 'fabs' and 'ceil' in device code [f41309b]
  • Added a diagnostic (warning) on attempt to append new device object to an archive which already contains an AOT-compiled device object [9d348eb]
  • Added a diagnostic on attempt to use functions which have no definition in the TU and are not marked with SYCL_EXTERNAL macro inside device code [a3b340b]
  • Added a diagnostic on attempt to use thread local storage inside device code [eb373c4]
  • Removed arch designator from the default output file name when compiling with -fsycl-link option. Now an output file has just a flat name based on the first input file [dc729a7]
  • The SYCL headers were moved from lib/clang/11.0.0/include to include/sycl to support mixed compilers [39501f6]
  • Added support for the GCC style inline assembly in the device code [6f4e007]
  • Improved fat static library support: the driver now consider for offloading static libraries which are passed on the command line as well as libraries passed as part of the linker options. This effectively negates the need to use -foffload-static-lib and -foffload-whole-static-lib options which are deprecated now.
  • The SYCL_EXTERNAL macro is now allowed to be used with class member functions [3baec18]
  • Set aux-target-cpu for the device compilation which sets AVX and other necessary macro based on a target [f953fda]

SYCL headers and runtime

  • Changed sycl::context and sycl::queue constructors to be explicit to avoid unintended conversions [c220eb8][3b6799a]
  • Added a diagnostic on setting SYCL_DEVICE_TYPE environment variable to an incorrect value [0125496]
  • Improved error codes which are encoded in the SYCL exceptions [04ee17c]
  • Removed functions that use float type in the fallback library for fp64 complex [6ccd84a0]
  • Added support for RESTRICT_WRITE_ACCESS_TO_CONSTANT_PTR macro which allows to enable diagnostic on writing to a raw pointer obtained from a sycl::constant_ptr object [c9ed5b2]
  • Added support for USM extension for CUDA backend [498d56c]

Documentation

Bug fixes

SYCL Frontend and driver changes

  • Fixed a problem with compiler not being able to find a dependency file when compiling AOT to an object for FPGA [7b58b01]
  • Fixed a problem with host object not being added to the partial link step when compiling from source and using -foffload-static-lib option [1a951cb]
  • Reversed reqd_work_group_size attribute to match SYCL behavior [1da6fbe]
  • Fixed dependency output location when /Fo<dir> is given [2b6f4f4]
  • Fixed a crash which happened when no kernel name is passed to the sycl::handler::parallel_for [fadaa59]

SYCL headers and runtime

  • Fixed sycl::queue::wait() which was not waiting for event associated with USM operation [850fb9f]
  • Fixed problem with reporting wrong error message on the second attempt to build program if the first attempt failed [9a34a11]
  • Fixed an issue which could happen when sycl::event::wait is called from multiple threads [3da5473]
  • Aligned sub_group::store signature between host and device [b3a9426]
  • Fixed sycl::program::get_compile_options and sycl::program::get_build_options to return correct values [03326f7]
  • Fixed sycl::multi_ptr's methods that were incorrectly enabled/disabled on device/host [401d174]
  • Fixed incorrect dependency handling when creating sub-buffers which could lead to data races [45e39bd]
  • Reversed reported max work-group size for a device to align with work-group size reversing before kernels launch [72b7dee]
  • Fixed incorrect handling of kernels that use hierarchical parallelism when -O0 option is passed to the clang [fd8ae8a]
  • Changed names of SYCL internal variables to avoid conflict with commonly used macros: SUCCESS, BLOCKED and FAILED [0f7e361]
  • Fixed a bug when a host device was always included in the device list returned by sycl::device::get_devices [6cf590f]
  • Fixed a problem with passing sycl::vec object to sycl::group::async_work_group_copy [20aa83e]
  • Fixed behavior of sycl::malloc_shared to return nullptr for the allocation size of zero or less byte, and the behavior of sycl::free functions to ignore the deallocation request from nullptr [d596593]
  • Fixed a possible problem with selecting work-group size which is bigger than max allowed work-group [b48f08f]
  • Fixed an issue which causes errors when using sub-buffers [5d1d716]
  • Changed the implementation of the buffer constructor from a pair of iterators. Now, data is not written back to the host on destruction of the buffer unless the buffer has a valid non-null pointer specified via the member function set_final_data [fb72758]
  • Fixed a problem with incorrect acceptance of a lambda which takes an argument of the sycl::id type in the sycl::handler::parallel_for version which takes a sycl::ndrange object [0408899]
  • Resolved circular dependency between sycl::event and sycl::queue [8c71dcb]

Known issues

  • The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • The SYCL library doesn't guarantee stable API/ABI, so applications compiled with older version of the SYCL library may not work with new one. The workaround is to rebuild the application.
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behavior
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 The workaround is to enable -std=c++17 for the failing MSVC version.

Prerequisites

Linux

Windows

Please, see the runtime installation guide here

February'20 release notes

Release notes for commit e8f1f29

New features

  • Added __builtin_intel_fpga_mem for the FPGA SYCL device. The built-in is used to indicate the characteristics of the load-store unit (LSU) to be used when de-referencing the pointer [1e33c01]
  • Added support for the intelfpga::no_global_work_offset attribute [8bed533] [5a9058b]
  • Added basic xmethod implementation for sycl::accessor::operator[] to make it callable from gdb command line [d6be8ff]
  • Added device libraries for standard math like std::cos and std::complex type [7abd9d5]

Improvements

SYCL Frontend and driver changes

  • Added support for passing a non-type template parameter to the loop_unroll attribute [8d7a32a]
  • Relaxed the standard layout requirement for kernel arguments. Now by default they should be trivially copyable. The -sycl-std=1.2.1 driver option turns standard layout requirement "on" [3adb4a5]
  • Added diagnostic on using __float128 type in the device code [659efdf]
  • The intelfpga::max_private_copies has been renamed to intelfpga::private_copies [97a199f]
  • Prevented duplication of error diagnostics for -fsycl [3a0b62e]
  • Added diagnostic when attempting to use existing FPGA static library with additional device code [6431be6]
  • Added support for non-type template parameters for FPGA memory attributes [ffcad03]

SYCL headers and runtime

  • The SYCL_DEVICE_WHITE_LIST control was renamed to SYCL_DEVICE_ALLOWLIST [4df18fa]
  • Added a leaf limit to the execution graph to avoid leaf bloat in applications that have an overwhelming number of command groups that can be executed in parallel [7c293e2]
  • Added get_range() method to the image and local accessors to align with the SYCL specification [8ed5566]
  • Added a diagnostic on attempt to create a context from devices that are bound to different platforms [8f354f7]
  • An ordered queue can now be created by passing sycl::property::queue::in_order property to the sycl::queue constructor [c855520]
  • Added a diagnostic on attempt to create an accessor with an unsupported type [306624e]
  • Made host device return nullptr for bad USM allocation functions (huge, 0, etc) [2a000d9]
  • Added templated forms of USM allocation functions [42cf5bf]
  • Added support for APIs that query properties of USM pointers [926e38e]
  • Added cleanup of finished command nodes of the execution graph in the situations when the wait for a command is called implicitly or explicitly [438dc49]
  • Added 2 sycl::queue constructors accepting sycl::context and sycl::device arguments [c81c1c5]

Documentation

  • Added documentation for simplification of the sycl::queue functions [478b7c0]
  • Added documentation for reqd_work_group_size extension [c2c416a]
  • The FAQ document was introduced [e42b40e]
  • Added Ordered Queue Property proposal [9fa878f]
  • Added device code split options documentation to the user's manual [1355aa6]
  • Added documentation for ExtendedAtomics extension [4445462]
  • Removed old Ordered Queue proposal and make a note of deprecation [e8f1f29]

Bug fixes

SYCL Frontend and driver changes

  • Fixed variable sharing passed by value to parallel_for_work_group [d8ea63a]
  • Do not produce an error if some restricted feature (e.g. exceptions) is used by a function only referenced in unevaluated contexts [5eae571]
  • Fixed problem with not cleaning up temporary files when device code split feature is enabled [d86ee2f]
  • Fixed issue with emitting wrong 'typename' keywords in the integration header [c19372e]
  • The SYCL target image registration functions have been renamed to avoid conflicts with the OpenMP registration functions [82fd970]
  • Avoid using std::memcpy in the device code [f39f47e]
  • Fixed -save-temps option when used along with -fsycl [f7f4699]
  • Fixed link steps triggering for libraries specified through -foffload-static-lib when no source or object file is provided [360b25b]
  • Fixed output options behavior for -fsycl-link on Windows [67b24d46]

SYCL headers and runtime

  • Fixed final result saturation in the host implementation of sycl::mad_sat [54dddb4]
  • Fixed a crash when a non-nullptr_t 0x0 value is passed to the sycl::buffer::set_final_data method [6a0e279]
  • Fixed an issue with copying sub-buffer between different contexts [0867a38]
  • Resolved the problem when local accessor is a temporary object [1eed329]
  • Fixed an issue with the event not being retained when a memory object is constructed using interoperability constructors [0aabe7e]
  • Fixed compilation of kernels which use sycl::stream for FPGA device [c4dbaa2]
  • Fixed execution graph cleanup on memory object destruction [7a75b54]

Known issues

  • [new] Defining a SUCCESS or FAILED can break SYCL headers because such names are used internally
  • [new] The format of the object files produced by the compiler can change between versions. The workaround is to rebuild the application.
  • [new] The SYCL library doesn't guarantee stable API/ABI, so applications compiled with older version of the SYCL library may not work with new one. The workaround is to rebuild the application.
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behaviour
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 The workaround is to enable -std=c++17 for the failing MSVC version.

Prerequisites

Linux

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2020.10.3.0.04 is the recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 20.06.15619 is the recommended OpenCL GPU RT prerequisite for the SYCL compiler.
  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2020.10.3.0.04 is the recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 26.20.100.7870 is the recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Please, see the runtime installation guide here

December'19 release notes

Release notes for commit 78d80a1cc628af76f09c53673ada906a3d2f0131

New features

  • New attributes for Intel FPGA devices : num_simd_work_items, bank_bits, max_work_group_size, max_global_work_dim [61d60b6] [7becb9d] [9053642] [6441851]
  • The infrastructure for device standard C/C++ libraries has been introduced. assert can now be used in device code [0039ee0]
  • Add opencl-aot tool for AoT compilation for Intel CPU devices [400460b]
  • Added support for cl::sycl::intel::experimental::printf builtin [78d80a1]
  • Implemented device code split feature: compiler can now be instructed to split a single device code module into multiple via the -fsycl-device-code-split option. [9d7dba6] [cc93bc4] [5491486] [a339d4c] [9095749]

Improvements

SYCL Frontend and driver changes

  • Allowed applying cl::reqd_work_group_size and cl::intel_reqd_sub_group_size attributes to a lambda function [b06fc66]
  • Allowed applying SYCL_EXTERNAL to functions with raw pointer arguments or return value if -Wno-error=sycl-strict option is passed [2840458]
  • Added support for FPGA device libraries in fat static libraries [d39ab73]
  • Removed limitation of virtual types in SYCL [98754ff]

SYCL headers and runtime

  • Added support for platform name and platform version in the device allowed list [7f2f668]
  • Made kernels and programs caches thread-safe [118dc82]
  • It's now possible to omit cl namespace specifier when using SYCL API. sycl::buffer can be used instead cl::sycl::buffer [67e4655]
  • Diagnostic on attempt to do several actions in one command group has been implemented [9f8ae50]
  • queue::submit now throws synchronous exceptions [6a83d14]
  • Enqueue pending tasks after host accessor destructor [80d17b2]
  • Implemented more "flat" kernel submission cl::sycl::queue methods [c5318c5]

Documentation

  • Added support for generation of SYCL documentation with Doxygen [de418d6]
  • Design document which describes design of C/C++ standard library support has been added

Bug fixes

SYCL Frontend and driver changes

  • Fixed problem which prevented attaching multiple attributes to a SYCL kernel [b77e5b7]

SYCL headers and runtime

  • Fixed a possible deadlock with host accessor' creation from multiple threads [d1c6dbe]
  • Compatibility issues that can happen when the SYCL library and SYCL application are compiled with different version of standard headers are fixed. [d854643]
  • Fixed compile error for handler::copy with -fsycl-unnamed-lambda [e73d2ce]
  • Fixed error which happened when autogenerated (-fsycl-unnamed-lambda) kernel name contained cl::sycl::half type [514fc0b]
  • Fixed crash on submitting a kernel with zero local size on the host device [b6806ea]
  • Made vec::load and vec::store member functions templated to align with the SYCL specification [4bd76de]
  • Made exceptions that are thrown in the copy back routine during SYCL memory object destructor asynchronous [2d6dcd0]
  • Fixed creation of host accessor to sub-buffer leading to crash in some scenarios [f607520]
  • Fixed build error on Windows caused by RuntimeLibrary value inconsistency for LLVM and apps/tests linked with it [f9296b6]

Known issues

  • [new] The size of object file is increased compared to the older compiler releases due to the recent fat object layout changes in the open source LLVM.
  • Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behaviour
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 For MSVC version having the error the workaround is to use -std=c++17 switch.

Prerequisites

Linux

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2020.9.1.0.18_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 19.48.14977 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Windows

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2020.9.1.0.18_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 26.20.100.7463 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Please, see the runtime installation guide here

November'19 release notes

Release notes for commit e0a62df4e20eaf4bdff5c7dd46cbde566fbaee90

New features

Improvements

SYCL Frontend and driver changes

  • Added more diagnostics on incorrect usage of SYCL options [f8a8785]
  • Added diagnostic on capturing values by reference in a kernel lambda [0e7639a]
  • Removed --sycl option which was an alias to -fsycl-device-only
  • Added support for using integral template parameter as an argument to intelfpga::ivdep, intelfpga::ii and intelfpga::max_concurrency attributes [6a283ae]
  • Improved diagnostic for using incorrect triple for AOT [6af65e1]

SYCL headers and runtime

  • cl::sycl::pipe class was moved to cl::sycl::intel namespace [23af059]
  • Added new query info::device::kernel_kernel_pipe_support which indicates whether device supports SYCL_INTEL_data_flow_pipes extension [8acf36e]
  • Added support for copying accessor with arbitrary dimensions in handler::copy [ff30897]
  • Cache result of compilation and kernel creation of programs created via cl::sycl::program::build_with_kernel_type without build options [8c2e09a] [4ba499c] [d442364]
  • Added diagnostic (static_assert) for dimensions mismatch when passing cl::sycl::vec objects to SYCL builtins [eb3ac32]
  • Added diagnostic (exception) on attempt to create invalid sub-buffer [0bd58df]
  • Added single_task and parallel_for methods to cl::sycl::ordered_queue::single_task. These provide an alternative way to launch kernels [1d2f7ce]
  • Added forms of USM functions for allocations that take a queue reference [f57c2b6]
  • Added C++ deduction guides to
    • cl::sycl::range
    • cl::sycl::id
    • cl::sycl::nd_range
    • cl::sycl::vec
    • cl::sycl::multi_ptr
    • cl::sycl::buffer [644013e] [3546a78] [728b904] [dba1c20]
  • Added a new buffer constructor which takes a contiguous container as an argument [dba1c20]
  • Added support of 1 byte type to load and store method of cl::sycl::intel::sub_group class [c8cffcc]
  • Improved error reporting for kernel enqueue process [d27cff2]
  • Instance of memory objects are now preserved on host after host accessor destruction [16ae15a]
  • Added support for SYCL_DEVICE_WHITE_LIST control which can be used to specify devices that are visible to the SYCL implementation [4ad5263]

Documentation

Bug fixes

SYCL Frontend and driver changes

  • Fixed a problem with option duplication during offload [1edd217]
  • Modified the driver to unbundle files with no extension like objects [84992a5]
  • Fixed AOT compilation for FPGA target on Windows [6b789f9] [525da51]
  • Fixed FPGA AOT compilation when using FPGA AOCX device archives [ec738f2] [16c530a]
  • Fixed problem causing abort when -fsycl and -lstdc++ passed simultaneously [e437fd4]
  • Fixed -o option for FPGA AOT on Windows [ddd24a3]
  • Default -std=c++ setting for device compilations was removed to avoid __cplusplus differences between host and device compilation. [29cabdf]
  • Suppressed warning about cdecl calling convention [0d2dab4]
  • Fixed problem when two different lambdas with the same signature get the same mangled name [d6aa11b]
  • Removed the error being issued for pseudo-destructor expressions [3c9006e]
  • Removed the device linking step for cases with no valid objects being provided from the device compilation [25bbe42]

SYCL headers and runtime

  • Implemented a workaround for the issue with conversion of one element swizzle to a scalar [f752698]
  • Fixed a UX problem with a sycl::stream mixing output of several work-items in some cases [377b3fa]
  • Fixed problem with linker options getting passed to the bundler instead of the linker when -foffload-static-lib is supplied [e437fd4]
  • Fixed problem with using wrong offset inside cl::sycl::handler::copy [5c8e81f]
  • Fixed get_count and get_size methods of cl::sycl::accessor class that used to return incorrect values for non-local accessor created with range [9dd68c5]
  • Fixed issue with asynchronous exceptions not being associated with the returned event [6c512d3]
  • Fixed issue with host accessor not always providing exclusive access to a memory object [16ae15a]
  • Fixed issue with ignoring cl::sycl::property::buffer::use_host_ptr property [16ae15a]

Known issues

  • [new] cl::sycl::program constructor creates unnecessary context which is bound to the device which is chosen by cl::sycl::default_selector. This can lead to unexpected behavior, for example, process may hang if default selector chooses Intel FPGA emulator device but target device is Intel OpenCL CPU one
  • [new] Using cl::sycl::program API to refer to a kernel defined in another translation unit leads to undefined behaviour
  • [new] -fsycl-unnamed-lambda doesn't work with kernel names that contain cl::sycl::half type
  • The addition of the static keyword on an array in the presence of Intel FPGA memory attributes results in the empty kernel after translation
  • A loop's attribute in device code may be lost during compilation
  • Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0 For MSVC version having the error the workaround is to use -std=c++17 switch.

Prerequisites

Linux

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2019.9.12.0.12_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 19.48.14977 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Windows

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2019.9.12.0.12_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 26.20.100.7463 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Please, see the runtime installation guide here

October'19 release notes

Release notes for commit 918b285d8dede6ab0561fccc622f71cb858849a6

New features

  • cl::sycl::queue::mem_advise method was implemented [4828db5]
  • cl::sycl::handler::memcpy and cl::sycl::handler::memset methods that operate on USM pointer were implemented [d9e8467]
  • Implemented ordered queue extension
  • Implemented support for half type in sub-group collectives: broadcast, reduce, inclusive_scan and exclusive_scan [0c78bc8]
  • Added cl::sycl::intel::ctz built-in. [6a96b3c]
  • Added support for SYCL_EXTERNAL macro.
  • Added support for passing device function pointers to a kernel [dc9db24]
  • Added support for USM on host device [5b0952c]
  • Enabled C++11 attribute spelling for clang loop_unroll attribute [2f1e243]
  • Added full support of images on host device
  • Added support for profiling info on host device [6c03c4f]
  • cl::sycl::handler::prefetch is implemented [feeacc1]
  • SYCL sub-buffers is mapped to OpenCL sub-buffers

Improvements

SYCL Frontend and driver changes

  • Added Intel FPGA Command line interface support for Windows [55ebcae]
  • Added support for one-step compilation from source with -fsycl-link [55ebcae]
  • Enabled additional aoc options for dependency files input and output report [55ebcae]
  • Suppressed warning "_declspec attribute 'dllexport' is not supported" when run with -fsycl. Emit error when import function is called in the sycl kernel. [b10bdbb]
  • Changed -fsycl-device-only to override -fsycl option [d429243]
  • Added user-friendly diagnostic for unsupported math built-in functions usage in kernel [0476352]
  • The linking stage is now skipped if -fsycl-device-only option is passed [93178d1]
  • When unbundling static libraries on Windows, do not extract the host section as it is not being used. This fixes possible disk usage issues when working with fat static libraries [93ab97e]
  • Passing -fsycl-help with -### option now prints the actual call to tool being made. [8b8bfa9]
  • Allow for -gN to override default setting with -fintelfpga [3b20615]
  • Update sub-group reduce/scan syntax [cd8194d]
  • Prevent libraries from being considered for unbundling on Windows [3438a48]
  • Improved Windows behaviors for calling lib.exe when creating an archive for Intel FPGA AOT [e7afcb1]

SYCL headers and runtime

  • Removed suppression of exceptions thrown by async_handler from cl::sycl::queue destructor [61574d8]
  • Added the support for output operator for half data types [6a2cd90]
  • Improved efficiency of stream output of cl::sycl::h_item for Intel FPGA device [80e97a0]
  • Added support for std::numeric_limits<cl::sycl::half> [6edca52]
  • Marked barrier flags as constexpr to avoid its extra runtime translation [5635959]
  • Added support for unary plus and minus for cl::sycl::vec class
  • Reversed mapping of SYCL range/ID dimensions to OpenCL, to provide expected performance through unit stride dimension. The highest dimension in SYCL (e.g. r2 in cl::sycl::range<3> R(r0,r1,r2)) now maps to the lowest dimension in OpenCL (e.g. an enqueue of size_t[3] cl_R = {r2,r1,r0}). The same applies to range and ID queries, in kernels defined through OpenCL interop. [40aa3f9]
  • Added support for constructing cl::sycl::image without host ptr but with pitch provided [d1931fd]
  • Added sycld library on Windows which is compiled using /MDd option. This library should be used when SYCL application is compiled with /MDd option to avoid ABI issues [71a75c0]
  • Added driver and runtime support for AOT-compiled images for multiple devices. This handles the case when the device code is AOT-compiled for multiple targets [0d4eb49] [bcf38cf]

Documentation

Bug fixes

SYCL Frontend and driver changes

  • Fixed problem with using aliases as kernel names [a784071]
  • Fixed address space in generation of annotate attribute for static vars and global Intel FPGA annotation [800c8c0]
  • Suppressed emitting errors for TLS declarations [ddc1a7f]
  • Suppressed device code link warnings that happen during linking fat and non-fat object files [b38a8e0]
  • Fixed pointer width on 64-bit version of Windows [63e2b19]
  • Fixed integration header generation when kernel name type is defined in cl, sycl or detail namespaces [5d22a8e]
  • Fixed problem with incorrect generation of output filename caused by processing of libraries in SYCL device toolchain [d3d9d2c]
  • Fixed problem with generation of depfile information for Intel FPGA AOT compilation [fbe951f]
  • Fixed generation of help message in case of -fsycl-help=get option passed [8b8bfa9]
  • Improved use of /Fo on Windows in offload situations so intermediate temporary files are not renamed [6984794]
  • Resolved problem with unnamed lambdas having the same name [f4d182f]
  • Fixed -fsycl-add-targets option to support multiple triple:binary arguments and to emit diagnostics for invalid target triples [21fa901]
  • Fixed AOT compilation for GEN devices [cd2dd9b]

SYCL headers and runtime

  • Fixed problem with using 32 bits integer type as underlying type of cl::sycl::vec class when 64 bits integer types must be used on Windows [b4998f2]
  • cl::sycl::aligned_alloc* now returns nullptr in case of error [9266cd5]
  • Fixed bug in conversion from float to half in the host version of cl::sycl::half type [6a2cd90]
  • Corrected automatic/rte mode conversion of cl::sycl::vec::convert method [6a2cd90]
  • Fixed memory leak related to incorrectly destroying command group objects [d7b5c0d]
  • Fixed layout and alignment of objects of 3 elements cl::sycl::vec type, now they occupy memory for 4 elements underneath [32f0cd5] [8f7f4a0]
  • Fixed problem with reporting the same asynchronous exceptions multiple times [9040739]
  • Fixed a bug with a wrong success code being returned for non-blocking pipes, that was resulting in incorrect array data passing through a pipe. [3339c45]
  • Fixed problem with calling atomic_load for float types in cl::sycl::atomic::load. Now it bitcasts float value to integer one then call atomic_load. [f4b7b17]
  • Fixed crash in case incorrect local size is passed. Now an exception is thrown in such cases. [1865c79]
  • cl::sycl::vec types aliases are now aligned with the SYCL specification.
  • Fixed cl::sycl::rotate method to correctly handle over-sized shift widths [d2e6a26]
  • Changed underlying address space of cl::sycl::constant_ptr from constant to global to avoid casts between constant and generic address spaces [38c2960]
  • Aligned cl::sycl::range class with the SYCL specification by removing its default constructor [d3b6a49]
  • Fixed several thread safety problems in cl::sycl::queue class [349a0d3]
  • Fixed compare_exchange_strong to properly update expected inout parameter [627a137]
  • Fixed issue with host version of cl::sycl::sub_sat function [7865dfc]
  • Fixed initialization of cl::sycl::h_item object when cl::sycl::handler::parallel_for method with flexible range is used [ab3e71e]
  • Fixed host version of cl::sycl::mul_hi built-in to correctly handle negative arguments [8a3b7d9]
  • Fix host memory deallocation size of SYCL memory objects [866d634]
  • Fixed bug preventing from passing structure containing accessor to a kernel on some devices [1d72965]
  • Fixed bug preventing using types from "inline" namespace as kernel names [28d5931]
  • Fixed bug when placeholder accessor behaved like a host accessor fetching memory to be available on the host and blocking further operations on the accessed memory object [d8505ad]
  • Rectified precision issue with the float to half conversion [2de1379]
  • Fixed cl::sycl::buffer::reinterpret method which was working incorrectly with sub-buffers [7b2f630] [916c32d] [60b6e3f]
  • Fixed problem with allocating USM memory on the host [01869a0]
  • Fixed compilation issues of built-in functions. [6bcf548]

Known issues

  • [new] The addition of the static keyword on an array in the presence of Intel FPGA memory attributes results in the empty kernel after translation.
  • [new] A loop's attribute in device code may be lost during compilation.
  • [new] Linkage errors with the following message: error LNK2005: "bool const std::_Is_integral<bool>" (??$_Is_integral@_N@std@@3_NB) already defined can happen when a SYCL application is built using MS Visual Studio 2019 version below 16.3.0.

Prerequisites

Linux

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2019.9.11.0.1106_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 19.43.14583 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Windows

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2019.9.11.0.1106_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 100.7372 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Please, see the runtime installation guide here

September'19 release notes

Release notes for commit d4efd2ae3a708fc995e61b7da9c7419dac900372

New features

Improvements

  • clCreateCommandQueue or clCreateCommandQueueWithProperties is used depending on version of OpenCL implementation. [3511e3d]
  • Added support querying kernel for compile_sub_group_size. [bb7cb34] This resolves #367
  • If device image is in SPIRV format and SPIRV is not supported by OpenCL implementation exception is thrown. [09e328f]
  • Added support for USM pointer to cl::sycl::handler::set_arg method. [df410a5]
  • Added support for -fsycl-help=arg compiler option which can be used to emit help message from corresponding offline compiler. [0e44dd2]
  • Added -reuse-exe compiler option which can be used to avoid recompilaton of device code (SPIRV) if it has not been changed from the previous compilation and all options are the same. This work for Intel FPGA AOT compilation. [2934fd8]
  • SYCL math builtins now work with regular pointers. [24fa42b]
  • Made SYCL specific options available when using clang-cl. [f5a7522]
  • USM now works on pre-context basis, so now it's possible to have two cl::sycl::context objects with USM feature enabled. [e339962]
  • "-fms-compatibility" and "-fdelayed-template-parsing" are not passed implicitly when -fsycl is used on MSVC. [9c3d98d]
  • Implemented cl::sycl::vec::convert method, host device now supports all rounding modes while other devices support automatic rounding mode. [fe3bbf9]
  • cl::sycl::image now uses internal aligned_allocator instead of standard one. [d7380f5]
  • cl::sycl::program class now throws exception containing build log in case of build failure.

Bug fixes

  • Fixed issue that prevented from using lambda functions in kernels. [1b83ae9]
  • Fixed crash happening when cl::sycl::event methods's are called for manually constructed (not obtained from cl::sycl::queue::submit) cl::sycl::event. [90333c3]
  • Fixed problem with functions called from inializer list are being considered as device code. [071b581]
  • Asynchronous error handler is now called in cl::sycl::queue destructor as the SYCL specification requires. [0182f72]
  • Fixed race condition which could happen if multiple host accessors to the same buffer are created simultaneously. [0c61c8c]
  • Fixed problem preventing from using template type with template parameters as a kernel name. [45194f7]
  • Fixed bug with offset passed to cl::sycl::handler::parallel_for methods was ignored in case of host device or when item is an argument of a lambda. [0caeeae]
  • Fixed crash which happened when -fsycl was used with no source file provided. [c291fd3]
  • Resolved problem with using bool type as a kernel name. [07b4f09]
  • Fixed crash which could happen if sycl objects are used during global objects destruction. [fff31fa]
  • Fixed incorrect behavior of host version of cl::sycl::abs_diff function in case if x - y < 0 and x and y are unsigned types. [35fc029]
  • Aligned the type of exception being thrown if no device of requested type is available with the SYCL specification. Now it throws cl::sycl::runtime_error. [9d5faab]
  • cl::sycl::accessor can now be created from cl::sycl::buffer which was constructed with non-default allocator. [8535b24]
  • Programs that failed to build (JIT compile) are not cached anymore. [d4efd2a]
  • Partial initialization of a constant static array now works correctly. [4e52d44]
  • Fixed casting from derived class to base in case of multiple inheritance. [76e223c]
  • Fixed compilation of relational operations with pointer. [4541a7f]
  • Fixed representation of long long int in cl::sycl::vec class on Windows. [336cb82]
  • cl::sycl::stream class has gained support for printing nan and inf values [b63a96f] This resolves #500
  • Fixed bug that prevented from using aliases as non-type template parameters (such as int) and full template specialization in kernel names. [a784071]
  • Crash when using stl allocator with cl::sycl::buffer and cl::sycl::image classes is fixed. [d7380f5]

Prerequisites

Linux

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2019.8.8.0.0822_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 19.34.13890 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Windows

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2019.9.9.0.0901 is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 100.7158 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Please, see the runtime installation guide here

August'19 release notes

Release notes for commit c557eb740d55e828fcf74b28d2b686c928e45318.

New features

  • Support for image accessor has been landed.
  • Added support for unnamed lambda kernels, so parallel_for works without specifying "Kernel Name" type. This can be enabled by passing -fsycl-unnamed-lambda option.
  • Kernel to kernel blocking and non-blocking pipe feature is implemented.
  • Added support for Unified Shared Memory (USM).

Improvements

  • Now OpenCL 1.2 clCreateSampler sampler is used for all version of OpenCL implementation.
  • Added Intel FPGA specific command line interfaces for ahead of time compilation.
  • Intel FPGA memory attributes are now supported for static variables.
  • Hierarchical parallelism is improved to pass conformance tests.
  • private_memory class has been implemented.
  • sycl.lib is automatically linked with -fsycl switch on Windows.
  • Added support for Windows getOSModuleHandle.
  • Functions with variadic arguments doesn't trigger compiler diagnostic about calling convention.
  • Added experimental support for building and linking SYCL runtime with libc++ library instead of libstdc++.
  • Adjusted array and vec classes so they are more efficient for Intel FPGA devices.
  • Exception will be thrown on attempt to create image accessor for the device which doesn't support images.
  • Use llvm-objcopy for merging device and host objects instead of doing partial linking.
  • Check if online compiler is available before building the program.
  • clCreateProgramWithILKHR is used if OpenCL implementation supports cl_khr_il_program extension.
  • Reuse the pointer provided by the user in the buffer constructor (even if use_host_ptr property isn't specified) if its alignment is sufficient.
  • -Ldir now can be used to find libraries with -foffload-static-lib
  • max_concurrency Intel FPGA loop attribute now accepts zero.
  • Ignore incorrectly used Intel FPGA loop attribute emitting user friendly warning instead of compiler error.
  • Added depends_on methods of handler class which can be used to provide additional dependency for a command group.
  • SYCL implementation can now be built on Windows using Visual Studio 2017 or higher.

Bug fixes

  • Fixed assertion failure when use -save-temps option along with -fsycl.
  • Cached JITed programs are now released during destruction of context they are bound to, earlier release happened during a process shutdown.
  • Fixed get_linear_id method of group class, now it calculate according to row-major ordering.
  • Removed printing of error messages to stderr before throwing an exception.
  • Explicit copy API of the handler class is asynchronous again.
  • fill method of handler class now takes element size into account.
  • Fixed bug with ignored several Intel FPGA loop attributes in case of argument is 1.
  • Fixed bug which prevented Intel FPGA loop attributes work with infinite loops.
  • Fixed problem which caused invalid/redundant memory copy operations to be generated.
  • The commands created by Scheduler now cleaned up on destruction of corresponding SYCL memory objects (buffer, image).
  • 1 dimensional sub buffer is passed as cl_mem obtained by calling clCreateSubBuffer when kernel is built from OpenCL C source.
  • Now copy of entire memory is performed when data is needed in new context even if user requests accesses to only part of the memory.
  • Fixed problem with one element vec objects relation/logical operation which was working like scalar.
  • Type casting and conditional operator (ternary 'if') with pointers now working correctly.
  • The problem with calling inlined kernel from multiple TUs is fixed.
  • Fixed compiler warnings for Intel FPGA attributes on host compilation.
  • Fixed bug with passing values of vec<#, half> type to the kernel.
  • Fixed buffer constructor which takes host data as shared_ptr. Now it increments shared_ptr reference counter and reuses provided memory if possible.
  • Fixed a bug with nd_item.barrier not respecting fence_space flag

Prerequisites

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support version 2019.8.7.0.0725_rel is recommended OpenCL CPU RT prerequisite for the SYCL compiler
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 19.29.13530 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

July'19 release notes

Release notes for commit 64c0262c0f0b9e1b7b2e2dcef57542a3fe3bdb97.

New features

  • cl::sycl::stream class support has been added.
  • New attributes for Intel FPGA device are added: merge, max_replicates and simple_dual_port.
  • Initial support for new Plugin Interface (PI) layer is added to SYCL runtime library. This feature simplifies porting SYCL implementation to non-OpenCL APIs.
  • New address space handling rules are implemented in the SYCL device compiler. Raw pointers are allocated in generic address space by default and address space inference is supposed to be done by LLVM pass. Old compiler behavior can be recovered by enabling DISABLE_INFER_AS environment variable.
  • Add basic implementation of hierarchical parallelism API.
  • Add new clang built-in function __unique_stable_name. SYCL compiler may use this built-in function to auto-generate SYCL kernel name for lambdas.

Improvements

  • SYCL integration header is excluded from the dependency list.
  • Raw pointers capturing added to the SYCL device front-end compiler. This capability is required for Unified Shared Memory feature implementation.
  • SYCL device compiler enabled support for OpenCL types like event, sampler, images to simplify compilation of the SYCL code to SPIR-V format. CXXReflower pass used to make "SPIR-V friendly LLVM IR" has been removed.
  • Intel FPGA loop attributes were renamed to avoid potential name conflicts.
  • Old scheduler has been removed.
  • sampler type support is added to the set_arg methods.
  • Internal SYCL device compiler design documentation was improved and updated. Development process documentation has been updated with more details.
  • Initial support for image class (w/o accessor support).
  • Static variables are allocated in global address space now.
  • Made sub-group methods constant to enable more use cases.
  • Added -fsycl-link option to generate fat object "linkable" as regular host object.
  • Enable set_final_data method with shared_ptr parameter.
  • Enable using of the copy method with shared_ptr with const T.

Bug fixes

  • Fixed argument size calculation for zero-dimensional accessor.
  • Removed incorrect source correlation from kernel instructions leading to incorrect profiling and debug information.
  • A number of issues were fixed breaking build of the compiler on Windows
  • Fixed unaligned access in load and store methods of the vector class.
  • global_mem_cache_type values were aligned with the latest revision of the SYCL specification.
  • Stubs for C++ standard headers were removed. This should fix compilation of and with SYCL device compiler.
  • Unscoped enums were removed from global namespace to avoid conflicts with user defined symbols.
  • Explicit copy API of the handler class is blocking i.e. data is copied once the command group has completed execution.
  • Renamed cl::sycl::group::get_linear to cl::sycl::group::get_linear_id.
  • SYCL kernel constructor from OpenCL handle now retains OpenCL object during SYCL object lifetime.
  • Fixed forward declaration compilation inside a SYCL kernel.
  • Fixed code generation for 3-element boolean vectors.

Prerequisites

  • Experimental Intel(R) CPU Runtime for OpenCL(TM) Applications with SYCL support is available now and recommended OpenCL CPU RT prerequisite for the SYCL compiler.
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 19.25.13237 is recommended OpenCL GPU RT prerequisite for the SYCL compiler.

Known issues

  • New address space handling approach might degrade compilation time (especially for GPU device).
  • Some tests might fail on CPU device with the first experimental CPU runtime due to new address space handling by the SYCL compiler. The workaround for this kind of issues while we wait for CPU runtime update is to set DISABLE_INFER_AS environment variable during compilation. See #277 for more details.

June'19 release notes

The release notes contain information about changes that were done after previous release notes and up to commit d404d1c6767524c21b9c5d05f11b89510abc0ab9.

New Features

  • New FPGA loop attributes supported:
    • intelfpga::ivdep(Safelen)
    • intelfpga::ii(Interval)
    • intelfpga::max_concurrency(NThreads)

Improvements

  • The new scheduler is implemented with the following improvements:
    • optimize host memory allocation for buffer objects.
    • optimize data transfer between host and device by using Map/Unmap instead of Read/Write for 1D accessors.
    • simultaneous read from a buffer is allowed: execution of two kernels reading from the same buffer are not serialized anymore.
  • Memory attribute intelfpga::max_concurrency was renamed to intelfpga::max_private_copies to avoid name conflict with fresh added loop attribute
  • Added support for const values and local accessors in handler::set_arg method.

Bug Fixes

  • The new scheduler is implemented with the following bug fixes:
    • host accessor now blocks subsequent operations(except RAR) with the buffer it provides accesses to until it is destroyed.
    • OpenCL buffers now released on the buffer destruction.
  • accessor::operator[] like methods now take into account offset.
  • Non-SYCL compilation(without -fsycl) was fixed, such application should work on host device, but fail on OpenCL devices.
  • Several warnings were cleaned up.
  • buffer constructor was fixed to support const type as template parameter.
  • event::get_profiling_info now waits for event to be completed.
  • Removed non-const overload of item::operator[] as it's not present in SYCL specification.
  • Compiling multiple objects when using -fsycl-link-targets now creates proper final .spv binary.
  • Fixed bug with crash in sampler destructor when sampler object is created using enumerations.
  • Fixed handler::set_arg, so now it works correctly with kernels created using program constructor which takes cl_program or program::build_with_source.
  • Now lgamma_r builtin works correctly when application is built without specifying -fsycl option to the compiler.

Prerequisites

  • Experimental Intel® CPU Runtime for OpenCL™ Applications with SYCL support is available now and recommended OpenCL CPU RT prerequisite for the SYCL compiler.
  • The Intel(R) Graphics Compute Runtime for OpenCL(TM) version 19.21.13045 is recomended OpenCL GPU RT prerequisite for the SYCL compiler.

Known issues

  • Performance regressions can happen due to additional math for calculation of offset that were added to the accessor::operator[] like methods.
  • Applications can hang at exit when running on OpenCL CPU device because some OpenCL handles allocated inside SYCL(e.g. cl_command_queue) are not released.

May'19 release notes

New Features

  • Added support for half type.
  • Implemented sampler class.

Improvements

  • Implemented several methods of buffer class:
    • buffer::has_property and buffer::get_property
    • buffer::get_access with range and offset classes
    • buffer::get_allocator as well as overall support for custom allocators.
  • Implemented broadcasting vec::operator=.
  • Added support for creating a sub-buffer from a SYCL buffer.
  • Added diagnostic about capturing class static variable in kernel code.
  • Added support for discard_write access::mode.
  • Now SYCL buffer allocates 64 bytes aligned memory.
  • Added support for case when object of accessor class is wrapped by some class.
  • Added support for const void specialization of multi_ptr class.
  • Implemented the following groups of SYCL built-in functions:
    • integers
    • geometric
  • Support for variadic templates in SYCL kernel names.

Bug Fixes

  • Disabled several methods of buffer class that were available for incompatible number of dimensions.
  • Added initialization of range field in buffer interoperability constructor.
  • Fixed buffer constructor with iterators, now the data is written back to the input iterator if it's not const iterator.
  • Fixed the problem which didn't allow using buffer::set_final_data and creating of host accessor from buffer created using interoperability constructor.
  • Now program::get_*_options returns options correctly if the program is created with interoperability constructor.
  • Fixed linking multiple programs compiled using compile_with_kernel_type.
  • Fixed initialization of device list in program class interoperability constructor.
  • Aligned vec class with the SYCL Specification, changed argument to multi_ptr with const type.
  • Fixed queue profiling device information query to work with OpenCL1.2.