diff options
| author | Pablo Tello <pablo.tello@arm.com> | 2017-06-28 17:27:56 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-09-17 14:15:39 +0100 |
| commit | 221f38176b0d4dbc212441779d9bbac3cc0eecfa (patch) | |
| tree | e838d673b35c5b40d4b484a3645cc7ae3c9d3ecc /scripts | |
| parent | 6410fb2a14427713251f5d97144ac5d4f17c988c (diff) | |
| download | ComputeLibrary-221f38176b0d4dbc212441779d9bbac3cc0eecfa.tar.gz | |
COMPMID-421: Fixed FP16 support in Neon GEMM.
Fixed GEMM FP16 problem with matrices that are not multiple of 32.
Added a new test suite NEON/GEMM/Float16/SmallGEMM.
Implemented FP16 function to multiply vector by a matrix.
Change-Id: Ie6c692885a48d0206bd6fe748332fa83bc286d67
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/79118
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Moritz Pflanzer <moritz.pflanzer@arm.com>
Diffstat (limited to 'scripts')
| -rwxr-xr-x | scripts/check_clang-tidy.py | 1 | ||||
| -rw-r--r-- | scripts/clang-tidy.h | 25 |
2 files changed, 26 insertions, 0 deletions
diff --git a/scripts/check_clang-tidy.py b/scripts/check_clang-tidy.py index a376d0b898..6ab1747482 100755 --- a/scripts/check_clang-tidy.py +++ b/scripts/check_clang-tidy.py @@ -39,6 +39,7 @@ if __name__ == "__main__": ("Validation.cpp" in line and "parameter 'expected_labels' is unused" in line) or ("Reference.cpp" in line and "parameter 'rois' is unused" in line) or ("ReferenceCPP.cpp" in line and "parameter 'rois' is unused" in line) or + ("NEGEMMMatrixMultiplyKernel.cpp" in line and "do not use C-style cast to convert between unrelated types" in line) or "3rdparty" in line): continue diff --git a/scripts/clang-tidy.h b/scripts/clang-tidy.h index 32b0f6955e..cbc0d07cd6 100644 --- a/scripts/clang-tidy.h +++ b/scripts/clang-tidy.h @@ -1,5 +1,30 @@ #include <arm_neon.h> +inline float16x8_t vmulq_lane_f16 (float16x8_t, float16x4_t, const int) +{ + return vdupq_n_f16(0); +} + +inline float16x4_t vmul_f16 (float16x4_t, float16x4_t) +{ + return vdup_n_u16(0); +} + +inline float16x4_t vadd_f16 (float16x4_t, float16x4_t) +{ + return vdup_n_u16(0); +} + +inline float16x4_t vmul_lane_f16 (float16x4_t, float16x4_t, const int) +{ + return vdup_n_u16(0); +} + +inline float16x4_t vmul_n_f16 (float16x4_t, float16_t) +{ + return vdup_n_u16(0); +} + inline float16x8_t vcvtq_f16_u16(uint16x8_t) { return vdupq_n_f16(0); |