| 41ab8faaf0d90238d42d8e2bbb7177467c10b4f6 |
|
08-Sep-2016 |
Miao Wang <miaowang@google.com> |
Implement multi-thread CPU GEMM for BLAS Intrinsics
- Multi-thread GEMM utilizes existing RS thread pool on top of
Eigen.
- Large matrix-matrix multiplication is decomposed into multiple
tiled matrix-matrix multiplications. Each thread iterates on
the unfinished works.
- The tiling applies to ONLY ONE dimension of each input matrix,
and whether to tile X or Y depends on the transpose of the matrix.
- The performance increase is proportional to the number of
available CPU cores, for sufficiently large matrices.
Test: CTS test (rsblas) pass on Angler, Fugu and new devices.
Performance test with RsBlasBenchmark and RsNeuralNet demo
on Anger, Ryu, Seed, Shamu, Volantis, Fugu and new devices,
showing roughly 70%(Volantix 2 core) ~ 400+%(Angler 8 core) perf gain.
Change-Id: If96f4119fd34d5d9d98a2542801495e7ffe577ae
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 462de21ac2e1773b99aedee012adb374e476ae36 |
|
16-Nov-2016 |
Chih-Hung Hsieh <chh@google.com> |
Fix google-build-using-namespace warnings in cpu_ref.
* Remove "using namespace ..." statements.
* Replace them with using declarations of the required names.
* Enclose the C++ methods and static and extern "C" functions with
namespace android and renderscript.
* Keep global C++ functions as-is and add using declarations for them.
Bug: 32670901
Test: build with WITH_TIDY=1
Change-Id: I818de466e8786a6c4f9ce0cd8e0fe027f34d7fad
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| e4f999b761180a227864d97b172a42ca1d8c0df3 |
|
03-Feb-2016 |
Miao Wang <miaowang@google.com> |
Switch "transpose" for Matrix A & B, after gemmlowp change.
Change-Id: I26fcaebcca828388ef6fe53c6e9e4db8e60dd4d9
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 9195e5188cb0e72d874512de01e7e58f1f47e0b7 |
|
15-Sep-2015 |
Miao Wang <miaowang@google.com> |
Update IntrinsicBLAS call to gemmlowp after rebase.
Change-Id: Id084ac7b53ea0b3c61311b4f4c78312f397b7c5f
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 223231fe99c9c958de4a1c8723aff88cb667de52 |
|
17-Jul-2015 |
Miao Wang <miaowang@google.com> |
Update eight_bit_int_gemm call after gemmlowp rebase and provide
non-optimal path for armv7 without NEON.
- gemmlowp will handle the optimal path for x86, NEON and aarch64
Change-Id: I67ce4c1e5b3195017a3d46895a8ce096682bc172
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| e941f18202b9c9883ff81c63710f7faec5c988e4 |
|
15-Jul-2015 |
Miao Wang <miaowang@google.com> |
Making libRSSupport able to optionally bundle libblas(V8) through dlopen
and dlsym.
Change-Id: I3ade3ad2802f3b8e5fc5661319b98a6212e6d8a2
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 99d0e8130f5b4bb83d1a68d96496fa558e35193a |
|
07-Jul-2015 |
Miao Wang <miaowang@google.com> |
Update the BNNM cpu reference implementation with NEON friendly
gemmlowp.
Change-Id: I5bcfd0fa988d8075e70272f277d7d7fab93d5fea
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 06deda3751a4a7358a7c7e03fbf1e4325fafb807 |
|
30-Jun-2015 |
Miao Wang <miaowang@google.com> |
update the offset type for BLAS.BNNM
bug: 22184114
Change-Id: I6ec212f8d5feb46fc9d0f97862b206978af1675b
(cherry picked from commit 22cb808b0dfc9bd514d2e19b302a97f8455b5731)
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| c060f1435e7b9405f3be8974417fa6f410f03753 |
|
14-May-2015 |
Stephen Hines <srhines@google.com> |
Use "override" instead of "virtual" when replacing methods.
Bug: 20306487
Change-Id: Ic83cb04cac153a7556f5d516e8f5ec88b5527b6f
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| e64c3d560186f976f9b8923e0f2c6ac3080913a2 |
|
07-May-2015 |
Miao Wang <miaowang@google.com> |
remove dead code (ALOGE) in rsCpuIntrinsicBLAS.cpp
bug: 21028875
Change-Id: Ia2d85a265f6e4a2617373f99b5c7bdc3810a7f24
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 08ef7b7f7977e9c991d8ba94a63860edcb88a3d9 |
|
30-Apr-2015 |
Miao Wang <miaowang@google.com> |
fix the CHER, CHPR, ZHER, ZHPR crash due to incorrect param order.
Change-Id: If91cbf969c75e01afc6d93b204bc8167180c9ef9
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| b75ba0fc7469d0bb4c1a6679664a846b3741792e |
|
27-Apr-2015 |
Miao Wang <miaowang@google.com> |
fix RsBlas_xgemv and RsBlas_xgbmv crash. (typo)
Change-Id: Ia948afa2bc4af22f99323618738d5eb7d415ca97
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 2b999883f2f390ee43ed18317d77c810a0c6657b |
|
13-Apr-2015 |
Tim Murray <timmurray@google.com> |
Rename BGEMM to BNNM. Modify layout of eight-bit GEMM-like intrinsic storage.
Change-Id: If4b1267dfd42d6dd65bedf20c0b674479eefab35
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| aff744561bea3c8a7a7d59c0cb8cd9438f6dcd1c |
|
31-Mar-2015 |
Tim Murray <timmurray@google.com> |
Add eight-bit GEMM-like intrinsic.
Change-Id: I9b920900b4cb8b27e2ab27386d05f4175142d6b2
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|
| 64c682b65cd04ac83b51251b40dca14423df351a |
|
09-Jan-2015 |
Tim Murray <timmurray@google.com> |
Add BLAS to supported intrinsics.
Change-Id: I8e776b2ffdbac09a73924035eee2eca0a12facb3
/frameworks/rs/cpu_ref/rsCpuIntrinsicBLAS.cpp
|