projects / contrib / beignet.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: d622727) | patch
GBE: Merge successive load/store together for better performance.
authorRuiling Song <ruiling.song@intel.com>
Thu, 8 May 2014 02:18:22 +0000 (10:18 +0800)
committerZhigang Gong <zhigang.gong@intel.com>
Fri, 9 May 2014 04:28:10 +0000 (12:28 +0800)
commitc5e3f688271f9cd733b5ccba1830c290cbdeac64
tree92d28f9acb377bf569f765eab531e9ca20fb0fe3tree | snapshot
parentd6227270310a9fd4601a042899701312a8d4823ecommit | diff
GBE: Merge successive load/store together for better performance.

Gen support at most 4 DWORD read/write in one single instruction.
So we merge successive read/write for less instruction and better performance.
This improves about 10% for LuxMark medium scene.

Signed-off-by: Ruiling Song <ruiling.song@intel.com>
Reviewed-by: Zhigang Gong <zhigang.gong@linux.intel.com>
backend/src/CMakeLists.txt diff | blob | history
backend/src/llvm/llvm_gen_backend.hpp diff | blob | history
backend/src/llvm/llvm_loadstore_optimization.cpp [new file with mode: 0644] blob
backend/src/llvm/llvm_to_gen.cpp diff | blob | history
Domain: SCM / Uncategorized;