Mismatches in dynamically fused direct conv2d + add kernel
Resolves: COMPMID-5269
Change-Id: I4372ea4365d14ead79153e4b08b690a1e20ab0b7
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7531
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/CL/cl_kernels/tile_helpers.h b/src/core/CL/cl_kernels/tile_helpers.h
index eab06aa..ec57022 100644
--- a/src/core/CL/cl_kernels/tile_helpers.h
+++ b/src/core/CL/cl_kernels/tile_helpers.h
@@ -1030,6 +1030,26 @@
}) \
})
+/** Element-wise addition between two tiles (LHS and RHS)
+ *
+ * @note Performs: LHS + RHS = DST
+ * @note Both tiles must have same data type
+ *
+ * @param[in] DST_DATA_TYPE DST data type
+ * @param[in] M0 Number of LHS rows
+ * @param[in] N0 Number of LHS columns
+ * @param[in] lhs LHS tile
+ * @param[in] rhs RHS tile
+ * @param[out] dst DST tile
+ */
+#define T_ADD(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) \
+ ({ \
+ LOOP_UNROLLING(int, _m0, 0, 1, M0, \
+ { \
+ dst[_m0].v = CONVERT(lhs[_m0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)) + CONVERT(rhs[_m0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)); \
+ }) \
+ })
+
/** Matrix multiplication
*
* @note Performs: LHS X RHS + DST = DST