commit db00668890e1aba956e02fa02e1383b54dfd1435
author steniu01 <steven.niu@arm.com> Wed Aug 09 16:26:22 2017 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:35:24 2018 +0000
tree e20cc07d9bc9eb4bf613213007a2351f5d4eec60
parent ff6ab352f4f6715b7028a39d8722759d19d2524b

COMPMID-478 Implemnt CL direct convolution 5x5

Change-Id: I4b975aff310cda9964d8c5dcee182d5d5c82741b
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/83474
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

src/core/CL/cl_kernels/direct_convolution3x3.cl

diff --git a/src/core/CL/cl_kernels/direct_convolution3x3.cl b/src/core/CL/cl_kernels/direct_convolution3x3.cl
index 4da7c39..28da544 100644
--- a/src/core/CL/cl_kernels/direct_convolution3x3.cl
+++ b/src/core/CL/cl_kernels/direct_convolution3x3.cl
@@ -50,8 +50,8 @@
 
 #define CONVOLUTION1x3_STRIDE1(acc, src_row_ptr, weights_row_ptr)                                                                                  \
     ({                                                                                                                                             \
-        VEC_DATA_TYPE(DATA_TYPE, 4)                                                                                                                \
-        weights_values0 = vload4(0, weights_row_ptr);                                                                                              \
+        VEC_DATA_TYPE(DATA_TYPE, 3)                                                                                                                \
+        weights_values0 = vload3(0, weights_row_ptr);                                                                                              \
         VEC_DATA_TYPE(DATA_TYPE, 8)                                                                                                                \
         src0 = vload8(0, src_row_ptr);                                                                                                             \
         VEC_DATA_TYPE(DATA_TYPE, 2)                                                                                                                \
@@ -64,8 +64,8 @@
 
 #define CONVOLUTION1x3_STRIDE2(acc, src_row_ptr, weights_row_ptr)                                                                               \
     ({                                                                                                                                          \
-        VEC_DATA_TYPE(DATA_TYPE, 4)                                                                                                             \
-        weights_values0 = vload4(0, weights_row_ptr);                                                                                           \
+        VEC_DATA_TYPE(DATA_TYPE, 3)                                                                                                             \
+        weights_values0 = vload3(0, weights_row_ptr);                                                                                           \
         VEC_DATA_TYPE(DATA_TYPE, 16)                                                                                                            \
         src0           = vload16(0, src_row_ptr);                                                                                               \
         DATA_TYPE src1 = *(src_row_ptr + 16);                                                                                                   \
@@ -152,4 +152,4 @@
 
     vstore8(CONVERT_SAT(pixels0, VEC_DATA_TYPE(DATA_TYPE, 8)), 0, (__global DATA_TYPE *)dst.ptr);
 }
-#endif // defined(DATA_TYPE) && defined(STRIDE_X) && defined(WEIGHTS_DEPTH)
\ No newline at end of file
+#endif // defined(DATA_TYPE) && defined(STRIDE_X) && defined(WEIGHTS_DEPTH)