COMPMID-3712 Remove OpenCL padding: CLDepthwiseConvolutionLayer3x3NHWCKernel FP16/32
Removed unused N from partial block loading macro
Created utility to assert change in padding
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: Ifdd30c66dbf5f2842c6b2d939000613d5011708e
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4192
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/CL/cl_kernels/load_store_utility.h b/src/core/CL/cl_kernels/load_store_utility.h
index cb833d0..84b6d2e 100644
--- a/src/core/CL/cl_kernels/load_store_utility.h
+++ b/src/core/CL/cl_kernels/load_store_utility.h
@@ -21,6 +21,234 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
+
+/** Store the 0 to (n-1)th rows of the given variables
+ * @name STORE_ROW_n
+ *
+ * @param[in] N0 The width of the passed in vector. Supported: 1, 2, 3, 4, 8, 16
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME The basename of the variables
+ * @param[in] PTR The base pointer
+ * @param[in] STRIDE_Y The stride value in y-axis direction
+ * @param[in] Z The offset in z-axis direction
+ * @{
+ */
+#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
+
+#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
+
+#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
+
+#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
+
+#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
+
+#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
+
+#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
+
+#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
+
+#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
+
+#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
+
+#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
+
+#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
+
+#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
+
+#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
+
+#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
+
+#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
+/** @} */ // end of groupd STORE_ROW_n
+
+/** Convert and store the 0th to (n-1)th rows of the given variables
+ * @name CONVERT_STORE_ROW_n
+ *
+ * @param[in] N0 The size of the vectors
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME The basename of the variables
+ * @param[in] PTR The base pointer
+ * @param[in] STRIDE_Y The stride value in y-axis direction
+ * @param[in] Z The offset in z-axis direction
+ * @{
+ */
+#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
+
+#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
+
+#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
+
+#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
+
+#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
+
+#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
+
+#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
+
+#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
+
+#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
+
+#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
+
+#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
+
+#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
+
+#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
+
+#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
+
+#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
+
+#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ VSTORE(N0) \
+ (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
+
+/** @} */ // end of groupd CONVERT_STORE_ROW_n
+
+/** Store a block of the given size M0xN0
+ * @name STORE_BLOCK
+ *
+ * Supported cases are M0=1,2,3,...,16 and N0=2,3,4,8,16.
+ * The data to store is expected to have consecutive names for each row.
+ * E.g., for M0=3 and basename=c, the expected names are c0, c1 and c2.
+ * The Z offset is expected to have consecutive names.
+ * E.g., for M0=3 and Z=zin, the expected z offset names are zin0, zin1 and zin2.
+ *
+ * @param[in] M0 The number of rows to store
+ * @param[in] N0 The size of each vector
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME The basename of the variables
+ * @param[in] PTR The base pointer
+ * @param[in] STRIDE_Y The stride value in y-axis direction
+ * @param[in] Z The offset in z-axis direction
+ * @{
+ */
+#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+/** @} */ // end of group STORE_BLOCK
+
+/** Convert and store a block of the given size M0xN0
+ * @name CONVERT_STORE_BLOCK
+ *
+ * Supported cases are M0=1,2,3,...,16 and N0=2,3,4,8,16.
+ * The data to store is expected to have consecutive names for each row.
+ * E.g., for M0=3 and basename=c, the expected names are c0, c1 and c2.
+ * The Z offset is expected to have consecutive names.
+ * E.g., for M0=3 and Z=zin, the expected z offset names are zin0, zin1 and zin2.
+ *
+ * @param[in] M0 The number of rows to store
+ * @param[in] N0 The size of each vector
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME The basename of the variables
+ * @param[in] PTR The base pointer
+ * @param[in] STRIDE_Y The stride value in y-axis direction
+ * @param[in] Z The offset in z-axis direction
+ * @{
+ */
+#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+/** @} */ // end of group CONVERT_STORE_BLOCK
+
/** Partially store the 0 to (n-1)th rows of the given variables
* @name STORE_ROW_PARTIAL_n
* Within each row, store the lower @p STORE_N0 elements of vectors of width @p N0
@@ -157,26 +385,25 @@
* @param[in] Z The offset in z-axis direction
* @param[in] PARTIAL_STORE_M0 The partial size in y, for partial blocks. Supported range: [1, @p M0)
* @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported range: [1, @p N0)
- * @param[in] N Total number of columns. Used to detect if current block is at the boundary in x.
* @param[in] PARTIAL_COND_Y Condition on the y axis to perform the partial store Y. True to use PARTIAL_STORE_M0 rather than M0.
* @param[in] PARTIAL_COND_X Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
*/
-#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
- if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y)) \
- { \
- STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
- } \
- else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X)) \
- { \
- STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
- } \
- else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X)) \
- { \
- STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
- } \
- else \
- { \
- STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y)) \
+ { \
+ STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+ } \
+ else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X)) \
+ { \
+ STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+ } \
+ else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X)) \
+ { \
+ STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+ } \
+ else \
+ { \
+ STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
}
/** Store a block that can only be partial in x but not y.
*
@@ -195,17 +422,16 @@
* @param[in] STRIDE_Y The stride value in y-axis direction
* @param[in] Z The offset in z-axis direction
* @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported range: [1, @p N0)
- * @param[in] N Total number of columns. Used to detect if current block is at the boundary in x.
* @param[in] PARTIAL_COND_X Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
*/
-#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, N, PARTIAL_COND_X) \
- if(!(PARTIAL_COND_X)) \
- { \
- STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
- } \
- else \
- { \
- STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
+ if(!(PARTIAL_COND_X)) \
+ { \
+ STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+ } \
+ else \
+ { \
+ STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
}
/** Store a block that can only be partial in y but not x.
*
@@ -286,36 +512,35 @@
* @param[in] Z The offset in z-axis direction
* @param[in] PARTIAL_STORE_M0 The partial size in y, for partial blocks. Supported: [0, @p M0)
* @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported: [0, @p N0)
- * @param[in] N Total number of columns. Used to detect if current block is at the boundary in x.
* @param[in] PARTIAL_COND_Y Condition on the y axis to perform the partial store Y. True to use PARTIAL_STORE_M0 rather than M0.
* @param[in] PARTIAL_COND_X Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
* @{
*/
#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
// Case1: No partial blocks in either x or y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0
// Case2: Partial blocks in y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)
#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0
// Case3: Partial blocks in x
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
- STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, N, PARTIAL_COND_X)
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)
#else // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
// Case4: Partial blocks in both x and y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
- STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X)
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)
#endif // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
#else // defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#endif // defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)
@@ -362,5 +587,5 @@
* @{
*/
#define STORE_VECTOR_SELECT(basename, data_type, ptr, vec_size, leftover, cond) \
- STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, 0, cond)
+ STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, cond)
/** @} */ // end of group STORE_VECTOR_SELECT
\ No newline at end of file