<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40"> <head> <meta http-equiv="Content-Type" content="text/html; charset=us-ascii"> <meta name="Generator" content="Microsoft Word 14 (filtered medium)"> <style></style> </head> <body lang="ZH-CN" link="blue" vlink="purple" style="text-justify-trim:punctuation"> <div class="WordSection1"> Release 0.3 (2013-10-22)<o:p></o:p> =========================<o:p></o:p> <o:p> </o:p> What's new?<o:p></o:p> ===========<o:p></o:p> <o:p> </o:p> In this release, the major improvement of this version is to add many new <o:p></o:p> features and fixs. There are a bunch of new features/enhancement/bug<o:p></o:p> fixes in this release. Here is a short list:<o:p></o:p> <o:p> </o:p> 1. Implemented all builtin functions.<o:p></o:p> 2. Support Long/ulong.<o:p></o:p> 3. Support register spilling/filling.<o:p></o:p> 4. Support event.<o:p></o:p> 5. Support profiling.<o:p></o:p> 6. Experimantal integration with libva driver(use OpenCL to do video post processing for libva).<o:p></o:p> 7. Implement more OpenCL APIs.<o:p></o:p> clEnqueueCopyImage/ImageToBuffer/BufferToImage/...<o:p></o:p> 8. Implement load/store the binary from/to program.<o:p></o:p> 9. Fix some random hang bugs.<o:p></o:p> 10. Other bug fixes.<o:p></o:p> <o:p> </o:p> Where can I get it?<o:p></o:p> ===================<o:p></o:p> <o:p> </o:p> git address: git://anongit.freedesktop.org/beignet<o:p></o:p> git tag: Release_v0.3<o:p></o:p> tar ball: http://cgit.freedesktop.org/beignet/snapshot/Release_v0.3.tar.gz<o:p></o:p> mail list: http://lists.freedesktop.org/mailman/listinfo/beignet<o:p></o:p> <o:p> </o:p> What is it?<o:p></o:p> ===========<o:p></o:p> Beignet is an open source implementation of the OpenCL specification - a generic<o:p></o:p> compute oriented API. This code base contains the code to run OpenCL programs on<o:p></o:p> Intel GPUs which basically defines and implements the OpenCL host functions<o:p></o:p> required to initialize the device, create the command queues, the kernels and<o:p></o:p> the programs and run them on the GPU. The code base also contains the compiler<o:p></o:p> part of the stack which is included in `backend/`. For more specific information<o:p></o:p> about the compiler, please refer to `backend/README.md`<o:p></o:p> <o:p> </o:p> Visit wiki: http://wiki.freedesktop.org/www/Software/Beignet/ for more information<o:p></o:p> about the project.<o:p></o:p> <o:p> </o:p> What changed in detail since version 0.3.<o:p></o:p> =======================<o:p></o:p> Boqun Feng (1):<o:p></o:p> GBE: define python interpreter by cmake variable<o:p></o:p> <o:p> </o:p> Chuanbo Weng (1):<o:p></o:p> Add a test case that trigger a known bug.<o:p></o:p> <o:p> </o:p> Homer Hsing (92):<o:p></o:p> support built-in functions "mul24", "mad24"<o:p></o:p> test cases for "mul24", "mad24"<o:p></o:p> support built-in functions "degrees" and "radians"<o:p></o:p> test built-in functions "degrees" and "radians"<o:p></o:p> support const indexed global constant array<o:p></o:p> test const-indexed global constant array<o:p></o:p> support built-in function "bitselect"<o:p></o:p> test built-in function "bitselect"<o:p></o:p> improve clCreateContext conformance<o:p></o:p> support clGetImageInfo<o:p></o:p> test API function "clGetImageInfo"<o:p></o:p> support built-in function "smoothstep"<o:p></o:p> test function "smoothstep"<o:p></o:p> add built-in function "mad_sat"<o:p></o:p> test function "mad_sat"<o:p></o:p> built-in function "sign"<o:p></o:p> test built-in function "sign"<o:p></o:p> fix vectorial built-in functions "min, max, clamp"<o:p></o:p> add built-in function "frexp"<o:p></o:p> test built-in function "frexp"<o:p></o:p> add built-in function "nextafter"<o:p></o:p> test built-in function "nextafter"<o:p></o:p> add builtin function "modf"<o:p></o:p> test builtin function "modf"<o:p></o:p> add built-in function "remquo"<o:p></o:p> test built-in function "remquo"<o:p></o:p> add address_space modifier to builtin functions' pointer<o:p></o:p> add builtin function "shuffle"<o:p></o:p> test builtin function "shuffle"<o:p></o:p> revise built-in function "shuffle"<o:p></o:p> add address space qualifier to "remquo"<o:p></o:p> add address space qualifier to "modf"<o:p></o:p> add built-in function "islessgreater"<o:p></o:p> add built-in function "isordered", "isunordered"<o:p></o:p> add built-in function "shuffle2"<o:p></o:p> test built-in function "shuffle2"<o:p></o:p> test if register allocation and 64-bit reading are fixed<o:p></o:p> support 64bit-integer reading(writing)<o:p></o:p> support 64bit-integer addition, subtraction<o:p></o:p> support 64bit-integer immediate value<o:p></o:p> support 64bit-integer AND(&), OR(|), XOR(^) arithmetic<o:p></o:p> test 64bit-integer immediate value, and "and", "or", "xor" arithmetic<o:p></o:p> keep address space qualifier of pointers<o:p></o:p> support 64bit-integer selection operator "?:"<o:p></o:p> test 64bit-integer selection operator<o:p></o:p> no "div by zero" in smoothstep test case<o:p></o:p> Define temporary reg as dest reg of instruction<o:p></o:p> support converting shorter int to 64bit int<o:p></o:p> support 64bit-integer shifting<o:p></o:p> test 64bit-integer shifting<o:p></o:p> support 64bit-integer multiplication<o:p></o:p> support 64bit-integer comparing<o:p></o:p> test 64bit-integer comparing<o:p></o:p> support built-in function mad_sat(int) and mad_sat(uint)<o:p></o:p> add empty 64bit-integer version built-in functions<o:p></o:p> add 64bit version of "upsample"<o:p></o:p> test 64bit version of "upsample"<o:p></o:p> fix a typo<o:p></o:p> GBE: skip instruction pattern match for 64 bit sel_cmp.<o:p></o:p> enable unsigned 64bit version of "abs_diff"<o:p></o:p> enable signed 64-bit version of "abs_diff"<o:p></o:p> improve built-in function "sinpi"<o:p></o:p> add built-in function "tgamma"<o:p></o:p> add built-in function "lgamma", "lgamma_r"<o:p></o:p> add 64-bit version of "bitselect"<o:p></o:p> fix scalar type built-in function "select"<o:p></o:p> Add scalar version of "convert_*(*)"<o:p></o:p> add 64-bit version of "shuffle", "shuffle2"<o:p></o:p> fix 8-bit version of "clz"<o:p></o:p> add 64-bit version of "clz"<o:p></o:p> add 64-bit version of "rotate"<o:p></o:p> fix 32-bit signed version of "sub_sat"<o:p></o:p> add same type "convert_*(*)"<o:p></o:p> fix GPU data type for 16-bit moving<o:p></o:p> fix 64-bit "clz" if parameter is "long4" or "ulong4"<o:p></o:p> add built-in function "atan2"<o:p></o:p> support converting 64-bit integer to shorter integer<o:p></o:p> add 64-bit version of "hadd"<o:p></o:p> add built-in function "atan2pi"<o:p></o:p> support converting 64-bit integer to 32-bit float<o:p></o:p> add 64-bit version of "rhadd"<o:p></o:p> fix scalarizing of llvm phi node<o:p></o:p> 64-bit-int: allocate flag register by RA<o:p></o:p> fix 64bit writing<o:p></o:p> add 64-bit version of "mul_hi"<o:p></o:p> add 64-bit version of "mad_sat"<o:p></o:p> support 64-bit version "add_sat"<o:p></o:p> add 64-bit version of "sub_sat"<o:p></o:p> support 64-bit division and remainder<o:p></o:p> fix isnan (builtin function)<o:p></o:p> saturated conversion of native GPU data type, larger to narrower<o:p></o:p> support LLVM 3.4<o:p></o:p> <o:p> </o:p> Junyan He (16):<o:p></o:p> Improve the clEnqueueMapBuffer and clCreateBuffer API<o:p></o:p> Add the support for clSetMemObjectDestructorCallback API<o:p></o:p> Improve the clGetMemObjectInfo API, add more info option<o:p></o:p> Add the PCH support when building the source.<o:p></o:p> Add the serialization support for backend<o:p></o:p> Add one tool program to build and serial the program.<o:p></o:p> Implement the clCreateProgramWithBinary to deseralize the binary.<o:p></o:p> Add a test case for binary load.<o:p></o:p> Add the virtual dctr function of Serialization to kill warning.<o:p></o:p> Add the string format support for gbe_bin_generater<o:p></o:p> Add the internal used kernels for buffer copy<o:p></o:p> Implement the clEnqueueCopyBuffer API using internal binary kernel<o:p></o:p> Add the test case for clEnqueueCopyBuffer<o:p></o:p> Delete the redundant intel_batchbuffer_t init in intel_gpgpu_new<o:p></o:p> Implement the CL api for clGetEventProfilingInfo<o:p></o:p> Using the PIPE_CONTROL to implement get time stamp in gen backend<o:p></o:p> <o:p> </o:p> Lu Guanqun (9):<o:p></o:p> fix left shift warnings in utests<o:p></o:p> fix left shift warning<o:p></o:p> fix warning when egl is not there<o:p></o:p> rename ulong to ulong64 to avoid the conflicts in <sys/types.h><o:p></o:p> list all available utests' names<o:p></o:p> add a space to make the error more readable<o:p></o:p> we should check the 'err' parameter<o:p></o:p> fix the missing assignment for offset<o:p></o:p> refactor the api of intel_driver_share_buffer<o:p></o:p> <o:p> </o:p> Ruiling Song (23):<o:p></o:p> Fix a bug in stack calculation.<o:p></o:p> enable scratch memory allocation and read/write<o:p></o:p> Implement spill/unspill<o:p></o:p> Fix a re-schedule issue of scratch write<o:p></o:p> Skip spill/unspill instruction when trying to do spill.<o:p></o:p> GBE: Clear Flag register to fix a gpu hang.<o:p></o:p> utests: Add a unit test for non-aligned group size.<o:p></o:p> Fix utest compiler_group_size4 error.<o:p></o:p> utest: memset the output buffer to fix random fail.<o:p></o:p> GBE: Enable DWord scatter gather message for constant cache read.<o:p></o:p> Change constant unit test to cover 4 byte data type.<o:p></o:p> Implement constant buffer based on constant cache.<o:p></o:p> Fix non-4byte program global constant issue.<o:p></o:p> change constant test case to cover short/long type.<o:p></o:p> GBE: Support composite type constant.<o:p></o:p> utests: add more constant test cases for composite type.<o:p></o:p> GBE: Fix a constant bug which over-write memory.<o:p></o:p> utests: put compiler_vector_inc into known issue list.<o:p></o:p> GBE: Support local variable inside kernel function.<o:p></o:p> GBE: Update program binary format.<o:p></o:p> GBE: Inline all function calls.<o:p></o:p> GBE: Skip non-kernel functions in backend passes.<o:p></o:p> utests: add test cases for function call.<o:p></o:p> <o:p> </o:p> Simon Richter (4):<o:p></o:p> Fix OpenCL C version format<o:p></o:p> Use access() instead of fopen() to search for PCH<o:p></o:p> Add generated header and PCH to gitignore<o:p></o:p> ICD dispatch table must be first<o:p></o:p> <o:p> </o:p> Yang Rong (44):<o:p></o:p> Add build clang option fno-builtin to disable intrinsics.<o:p></o:p> Add the empty functions of cl_enqueueXXX.<o:p></o:p> Add a struct and a function to handle all implemented enqueue api.<o:p></o:p> Add some functions to support event in intel gpgpu.<o:p></o:p> Add function cl_command_queue_flush to flush a command<o:p></o:p> Add openCL event support.<o:p></o:p> Add event unit test.<o:p></o:p> Add bool move imm support.<o:p></o:p> Add a load bool imm test case.<o:p></o:p> Fix event pthread_mutex_lock dead lock.<o:p></o:p> Fix unit test compiler_load_bool_imm error.<o:p></o:p> Implement async and prefetch built-in.<o:p></o:p> Add async copy and async stride copy test case.<o:p></o:p> Add pfn_notify support in clCreateContext.<o:p></o:p> Add clEnqueueMapBuffer and clEnqueueMapImage non-blocking map support.<o:p></o:p> Change event test case to cover clEnqueueMapBuffer.<o:p></o:p> Correct event type' typo.<o:p></o:p> Fix atomic_xchg float type error.<o:p></o:p> Add clEnqueueReadBufferRect api.<o:p></o:p> Add clEnqueueWriteBufferRect api.<o:p></o:p> Add clEnqueueCopyBufferRect api.<o:p></o:p> Add api clEnqueueCopyImage.<o:p></o:p> Implement api clEnqueueTask and clEnqueueNativeKernel.<o:p></o:p> Implement api clEnqueueCopyImageToBuffer.<o:p></o:p> Implement api clEnqueueCopyBufferToImage.<o:p></o:p> Fix cl_mem_kernel_copy_image typo.<o:p></o:p> Remove non-used data in clEnqueueMapImage to fix, and fix a clGetEventInfo bug.<o:p></o:p> Refine and fix some event bugs.<o:p></o:p> Implement clEnqueueMarker and clEnqueueBarrier.<o:p></o:p> Fix store undef value assert.<o:p></o:p> Unmap the cl_mem in driver when application map a cl_mem and release without unmap.<o:p></o:p> Fix clEnqueueMapImage error.<o:p></o:p> Remove global offset need divide by local size restriction.<o:p></o:p> Change optimize level to -O2, to avoid loopunswitch opt.<o:p></o:p> Remove blocking asserts in clEnqueueXXX apis.<o:p></o:p> Add some preprocessor macros __IMAGE_SUPPORT__ and __FAST_RELAXED_MATH__ define.<o:p></o:p> Implement api clCreateKernelsInProgram.<o:p></o:p> Fix a vector argument deallocate assert.<o:p></o:p> Refine vector register deallocate.<o:p></o:p> Change -O3 to -O2 again because my previous change's typo.<o:p></o:p> Fix a read64/write64 schedule bug.<o:p></o:p> Add type long/ulong/double's async copy.<o:p></o:p> Remove newValueProxy from scalarize pass to genWriter pass.<o:p></o:p> Add test case for newValueProxy of InsertElementInst.<o:p></o:p> <o:p> </o:p> Yi Sun (9):<o:p></o:p> utest: add built-in test case for get_global_id.<o:p></o:p> utest: Add test for built-in function get_local_size.<o:p></o:p> utest: Add test for built-in function get_local_id.<o:p></o:p> utests: Add a test case for built-in functions get_num_groups.<o:p></o:p> Improve the accuracy of built-in function asin.<o:p></o:p> Handle boundary and illegal values.<o:p></o:p> utest: Add test case for function acos/acosh/asin/asinh.<o:p></o:p> Utests_run: Add known issue cases support.<o:p></o:p> utest.cpp: run the cases with issue seperately.<o:p></o:p> <o:p> </o:p> Zhigang Gong (56):<o:p></o:p> CL: Refine the version string handling.<o:p></o:p> utest: Query the device driver version and the open cl c version.<o:p></o:p> Frexp support global memory directly<o:p></o:p> Implement a pyton script to auto generate those builtin vector functions.<o:p></o:p> Split the thounsands autogenerated code out from ocl_stdlib header file.<o:p></o:p> check whether python is installed.<o:p></o:p> Add misc builtin vector functions.<o:p></o:p> Fix the indention handling in vector builtin function generator.<o:p></o:p> Enable islessgreater/isordered/isunordered builtin vector functions.<o:p></o:p> Added memory space parameters support at the autogeneration script.<o:p></o:p> Need to define local to __local.<o:p></o:p> GBE: refactor double support.<o:p></o:p> GBE: enable double vector load/store support.<o:p></o:p> GBE: fix insntruction scheduling related bugs in read64/write64.<o:p></o:p> GBE: Fix one bug in instruction scheduling.<o:p></o:p> Utests: enable long/ulong in vector load/store test case.<o:p></o:p> GBE: Fixed a bug and release 2 or 3 simdWidth register space.<o:p></o:p> Driver: Fix the incorrect size of surface 1.<o:p></o:p> GBE: set temporary address register for read64 to U64.<o:p></o:p> GBE: I64CMP should be treated as CMP in reg allocation and insn scheduling.<o:p></o:p> GBE: fix an illegal instruction.<o:p></o:p> Utests: enable long/ulong for abs_diff test case.<o:p></o:p> GBE: disable cl_khr_fp64.<o:p></o:p> CL: Refactor cl_mem's implementation.<o:p></o:p> Runtime: fix the incorrect platform info size (conformance).<o:p></o:p> GBE: don't use flag register as src 1 for xor instruction.<o:p></o:p> GBE: add some macros for atom_xxx builtin functions.<o:p></o:p> GBE: null register could be used as src1.<o:p></o:p> Runtime: clEnqueueMapImage also need to maintain the mapped images.<o:p></o:p> Runtime: vendor specified information is required for CL_DEVICE_VERSION/OPENCL_C_VERSION.<o:p></o:p> Runtime: initialize single fp mode correctly.<o:p></o:p> GBE: We should set no predication/mask for EOT preparation.<o:p></o:p> Runtime: fix the max group size for GT2.<o:p></o:p> GBE: Support builtin vector functions for select() autogeneration.<o:p></o:p> Runtime: fix the incorrect global mem size.<o:p></o:p> Utests: Enable bool_cross_basic_block.<o:p></o:p> GBE: silent the compilation warning when generate the pch file.<o:p></o:p> Runtime: Only return the format allowed in the spec.<o:p></o:p> CL: Enalbe gl sharing with new egl extension.<o:p></o:p> Runtime: disable some unecessary image formats.<o:p></o:p> Runtime: fix a bug when set sampler value.<o:p></o:p> Runtime: enable border color state support.<o:p></o:p> GBE: check the correct register for whether coord z exists.<o:p></o:p> GBE: fixed the broken 3d image support.<o:p></o:p> Runtime/driver : implement 3D image support.<o:p></o:p> Utests: refine the previous fake 3D test cases.<o:p></o:p> Runtime: prepare for CL_MEM_USE_HOST_PTR for image support.<o:p></o:p> Runtime: Implement CL_MEM_USE_HOST_PTR flag for image.<o:p></o:p> GBE: fixed the store3 bug.<o:p></o:p> Refine cmake script file.<o:p></o:p> clCopyImage: fix up all the surface type to int type.<o:p></o:p> GBE/Runtime: implement workaround for IVB sampler bug<o:p></o:p> GBE: Fix the out-of-box checking for normalized coord clamping.<o:p></o:p> GBE: refact the curbe register payload allocation.<o:p></o:p> GBE: Refine the curbe entry allocation for sampler/image information.<o:p></o:p> GBE: sampler_t should always be a const int.<o:p></o:p> <o:p> </o:p> Zou Nan hai (2):<o:p></o:p> Flush the queue after enqueue.<o:p></o:p> use r112 as source of EOT message<o:p></o:p> <o:p> </o:p> Thanks,<o:p></o:p> Yang Rong<o:p></o:p> </div> </body> </html>