[Beignet] [ANNOUNCE] Beignet 1.0.2 (2015-03-16)
Zhigang Gong
zhigang.gong at linux.intel.com
Mon Mar 16 01:12:59 PDT 2015
Beignet 1.0.2 (2015-03-16)
==========================
Beignet 1.0.2 has been released. It's a bug fix release including some bug fixes, minor new
features and some performance improvements. The highlighted improvements are as below:
1. Added LLVM 3.6 support.
2. Fixed a bug in popcount().
3. Refactor the legalize pass and const expression handling.
4. Fixed a bug when hardware is inaccessible.
5. Enabled multiarch 32bit and 64bit coexisting.
6. Implemented some new device side clCopyImage kernels.
7. Optimize clEnqueueCopyImageToBuffer for 16 aligned buffer.
Git tag: Release_v1.0.2
Gitweb URL: http://cgit.freedesktop.org/beignet
Download: https://01.org/sites/default/files/beignet-1.0.2-source.tar.gz
Official release notes: https://01.org/beignet/downloads/beignet-1.0.2-2015-03-16
md5sum: 437b5d26a947002c9200a9571355b1c6 Beignet-1.0.2-Source.tar.gz
sha1sum: f8377bf02b18af9b51114ea89e7a1460079f10a0 Beignet-1.0.2-Source.tar.gz
sha256sum: 2521fa523123b25c05e42e1c11c89d4ed7f28b335552c82a8883c61c235c8729 Beignet-1.0.2-Source.tar.gz
-----------------------------------------------------------------
Andreas Beckmann (2):
prefer newer llvm versions over 3.3
remove unsafe define -D__$(USER)__
Chuanbo Weng (4):
Add example to show libva buffer sharing with extension clCreateImageFromLibvaIntel.
Add document to describe the detials of libva buffer sharing.
Implement 1D/2D image array related cl_mem_kernel_copy_image in cl way instead of cpu way.
Optimization of clEnqueueCopyImageToBuffer for 16 aligned case.
Guo Yejun (3):
correct the cache line size to be 64
loose the alignment limitation for host_ptr of CL_MEM_USE_HOST_PTR
update utest to loose userptr limitation
Junyan He (6):
Add test case for long bitcast.
Add the check for src and dst span different registers.
Fix bug for bitcast test case because of long type.
Fix a bug of 1d image array test case.
Backend: Fix one bug of printf because of ir reorder.
Correct the bit fields error for indirect address of Gen8
Luo (1):
check the predication in case of endless loop.
Rebecca N. Palmer (3):
Return error, don't crash, on allocation failure
Crash when hardware inaccessible
Enable multiarch (32/64-bit co-installation)
Ruiling Song (14):
GBE: Import constantexpr lower pass from pNaCl
GBE: expand large integer instructions
GBE: Fix a build error against llvm release version
GBE: Fix a bug in legalize pass.
GBE: Load/store should use same address space as before.
GBE: Import PromoteIntegers pass from pNaCl
GBE: We need use exiting block here.
libocl: define NULL to zero
libocl: Directly scalarize built-in with vector input.
GBE: unify element type before insertelement in legalize pass.
GBE: Support unaligned load/store of dword/qword in GenIR.
GBE: Fix fast-math issue under llvm 3.6.
GBE: Only add non-zero offset in gep lowering pass.
GBE: Only emit multiply when immediate is not one.
Yang Rong (6):
Change the KB and MB define to enum.
Use llvm-c's LLVMLinkModules instead of llvm::Linker::LinkModules.
Add llvm3.6 build support.
Remove useless llvm head file FindUsedTypes.h.
Correct the error llvm link msg copy in function genProgramLinkFromLLVM.
Fix llvm3.6 build error.
Zhigang Gong (19):
GBE: fix an ACC register related instruction scheduling bug
GBE: fix popcount bugs.
GBE: add GEN_TYPE_HF to getTypeSize.
Add submodule libva for examples.
update document.
runtime: fix a potential null pointer dereference.
runtime: don't free the host_ptr for a subbuffer.
GBE: fix build error for llvm 3.6.
GBE: fix build error for LLVM 3.4/3.3.
build: use @BEIGNET_INSTALL_DIR@ for the icd file.
GBE: expand constant expressions in constant vector
GBE: remove constant expression handling code in gen writer pass.
GBE: remove the unecessary type check for SEL instructio.
GBE: support compare two bool variables.
GBE: add fastcall support.
Build: use -Bsymbolic to fix conflicts with other LLVM users.
GBE: add a new incompatible compile option -cl-finite-math-only.
Build: set 3.5 as the stable LLVM version for beignet.
Bump version to 1.0.2
--
Thanks,
Zhigang Gong.
More information about the Beignet
mailing list