diff options
| author | Miao Wang <miaowang@google.com> | 2015-06-29 17:29:43 -0700 |
|---|---|---|
| committer | Miao Wang <miaowang@google.com> | 2015-06-30 16:28:46 -0700 |
| commit | 22cb808b0dfc9bd514d2e19b302a97f8455b5731 (patch) | |
| tree | 2aa46f542560006777eb5ca68a338e7295d35559 | |
| parent | 7427a2546ecfa904cb4e580981f0419c628fc416 (diff) | |
| download | android_frameworks_rs-22cb808b0dfc9bd514d2e19b302a97f8455b5731.tar.gz android_frameworks_rs-22cb808b0dfc9bd514d2e19b302a97f8455b5731.tar.bz2 android_frameworks_rs-22cb808b0dfc9bd514d2e19b302a97f8455b5731.zip | |
update the offset type for BLAS.BNNM
bug: 22184114
Change-Id: I6ec212f8d5feb46fc9d0f97862b206978af1675b
| -rw-r--r-- | cpu_ref/rsCpuIntrinsicBLAS.cpp | 16 | ||||
| -rw-r--r-- | rsDefines.h | 8 |
2 files changed, 12 insertions, 12 deletions
diff --git a/cpu_ref/rsCpuIntrinsicBLAS.cpp b/cpu_ref/rsCpuIntrinsicBLAS.cpp index 30344544..a7705b4e 100644 --- a/cpu_ref/rsCpuIntrinsicBLAS.cpp +++ b/cpu_ref/rsCpuIntrinsicBLAS.cpp @@ -47,10 +47,10 @@ protected: uint8_t c_offset = 0; static void kernelBNNM(size_t m, size_t n, size_t k, - const uint8_t* a, uint32_t a_offset, size_t lda, - const uint8_t* b, uint32_t b_offset, size_t ldb, - uint8_t* c, uint32_t c_offset, size_t ldc, - uint32_t c_mult_int); + const uint8_t* a, uint8_t a_offset, size_t lda, + const uint8_t* b, uint8_t b_offset, size_t ldb, + uint8_t* c, int32_t c_offset, size_t ldc, + int32_t c_mult_int); @@ -653,10 +653,10 @@ void RsdCpuScriptIntrinsicBLAS::invokeForEach(uint32_t slot, } void RsdCpuScriptIntrinsicBLAS::kernelBNNM(size_t m, size_t n, size_t k, - const uint8_t* a, uint32_t a_offset, size_t lda, - const uint8_t* b, uint32_t b_offset, size_t ldb, - uint8_t* c, uint32_t c_offset, size_t ldc, - uint32_t c_mult_int) { + const uint8_t* a, uint8_t a_offset, size_t lda, + const uint8_t* b, uint8_t b_offset, size_t ldb, + uint8_t* c, int32_t c_offset, size_t ldc, + int32_t c_mult_int) { // Calculations are done in 1.10.21 fixed-point format for the final output, // just before there's a shift down to drop the fractional parts. The output // values are gated to 0 to 255 to fit in a byte, but the 10-bit format diff --git a/rsDefines.h b/rsDefines.h index e00ddc19..4edda7af 100644 --- a/rsDefines.h +++ b/rsDefines.h @@ -460,10 +460,10 @@ typedef struct { int incY; int KL; int KU; - uint32_t a_offset; - uint32_t b_offset; - uint32_t c_offset; - uint32_t c_mult_int; + uint8_t a_offset; + uint8_t b_offset; + int32_t c_offset; + int32_t c_mult_int; } RsBlasCall; enum RsGlobalProperty { |