| R"( |
| |
| /* |
| * Copyright (c) 2016-2018 Arm Limited. |
| * |
| * SPDX-License-Identifier: MIT |
| * |
| * Permission is hereby granted, free of charge, to any person obtaining a copy |
| * of this software and associated documentation files (the "Software"), to |
| * deal in the Software without restriction, including without limitation the |
| * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| * sell copies of the Software, and to permit persons to whom the Software is |
| * furnished to do so, subject to the following conditions: |
| * |
| * The above copyright notice and this permission notice shall be included in all |
| * copies or substantial portions of the Software. |
| * |
| * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| * SOFTWARE. |
| */ |
| /* |
| * Copyright (c) 2016-2020 Arm Limited. |
| * |
| * SPDX-License-Identifier: MIT |
| * |
| * Permission is hereby granted, free of charge, to any person obtaining a copy |
| * of this software and associated documentation files (the "Software"), to |
| * deal in the Software without restriction, including without limitation the |
| * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| * sell copies of the Software, and to permit persons to whom the Software is |
| * furnished to do so, subject to the following conditions: |
| * |
| * The above copyright notice and this permission notice shall be included in all |
| * copies or substantial portions of the Software. |
| * |
| * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| * SOFTWARE. |
| */ |
| #ifndef ARM_COMPUTE_HELPER_H |
| #define ARM_COMPUTE_HELPER_H |
| |
| /* |
| * Copyright (c) 2020 Arm Limited. |
| * |
| * SPDX-License-Identifier: MIT |
| * |
| * Permission is hereby granted, free of charge, to any person obtaining a copy |
| * of this software and associated documentation files (the "Software"), to |
| * deal in the Software without restriction, including without limitation the |
| * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| * sell copies of the Software, and to permit persons to whom the Software is |
| * furnished to do so, subject to the following conditions: |
| * |
| * The above copyright notice and this permission notice shall be included in all |
| * copies or substantial portions of the Software. |
| * |
| * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| * 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 |
| * |
| * @note in case @p STORE_N0 != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty. |
| * |
| * @param[in] N0 The width of the passed in vector. Supported: 1, 2, 3, 4, 8, 16 |
| * @param[in] STORE_N0 The **lower** size of the vectors to store. Supported: [1-16 and <= @p N0 |
| * @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_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0)); |
| |
| #define STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1)); |
| |
| #define STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2)); |
| |
| #define STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3)); |
| |
| #define STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4)); |
| |
| #define STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5)); |
| |
| #define STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6)); |
| |
| #define STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7)); |
| |
| #define STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8)); |
| |
| #define STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9)); |
| |
| #define STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A)); |
| |
| #define STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B)); |
| |
| #define STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C)); |
| |
| #define STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D)); |
| |
| #define STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E)); |
| |
| #define STORE_ROW_PARTIAL_16(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \ |
| VSTORE_PARTIAL(N0, STORE_N0) \ |
| (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F)); |
| /** @} */ // end of groupd STORE_ROW_PARTIAL_n |
| |
| /** Partially store a block of the given size STORE_M0xSTORE_N0 |
| * @name STORE_BLOCK_PARTIAL |
| * |
| * @note The vector width @p N0 is also required for correct partial storing behaviour. |
| * @note in case @p STORE_N0 != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty. |
| * |
| * The data to store is expected to have consecutive names for each row. |
| * E.g., for STORE_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 STORE_M0=3 and Z=zin, the expected z offset names are zin0, zin1 and zin2. |
| * |
| * @param[in] STORE_M0 The number of rows to store. Supported: 1-16 |
| * @param[in] STORE_N0 The lower number of elements of vectors to store. Supported: 1-16 and <= @p N0 |
| * @param[in] N0 The size of each 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_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) |
| #define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) |
| /** Store a block that can be partial in both x and y dimensions |
| * |
| * @note in cases @p PARTIAL_STORE_N0 != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty. |
| * |
| * 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, for non-partial blocks. Supported: 1-16 |
| * @param[in] N0 The size of each vector, for non-partial blocks. 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 |
| * @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] 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, 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. |
| * |
| * @note in case @p N0 or @p PARTIAL_STORE_N0 != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty. |
| * |
| * 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, for non-partial blocks. Supported: 1-16 |
| * @param[in] N0 The size of each vector, for non-partial blocks. 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 |
| * @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported range: [1, @p N0) |
| * @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, 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. |
| * |
| * @note in case @p N0 or @p PARTIAL_STORE_N0 != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty. |
| * |
| * 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, for non-partial blocks. Supported: 1-16 |
| * @param[in] N0 The size of each vector, for non-partial blocks. 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 |
| * @param[in] PARTIAL_STORE_M0 The partial size in y, for partial blocks. Supported range: [1, @p M0) |
| * @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. |
| */ |
| #define STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \ |
| if(!(PARTIAL_COND_Y)) \ |
| { \ |
| STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \ |
| } \ |
| else \ |
| { \ |
| STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \ |
| } |
| /** @} */ // end of group STORE_BLOCK_PARTIAL |
| |
| #if defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0) |
| |
| /** Boundary-aware GEMM block store |
| * @name STORE_BLOCK_BOUNDARY_AWARE |
| * This macro assumes the following schemes to achieve boundary-awareness: |
| * - Overlapping load in Y axis from lhs tensor. This implies lhs has no padding along y dim. |
| * - Non-Overlapping(normal) load from rhs tensor. This imples rhs can have paddings. |
| * - Overlapping load in Y axis from bias tensor. This implies rhs has no padding along y dim. |
| * The macro then ensures that the dst tensor can be stored without any paddings in both x and y dim. |
| * |
| * In the y dimension, we place the partial blocks **at the beginning** while in the x dimension, we place the partial |
| * blocks **at the end**. |
| * Say, the dst tensor is of shape MxN and we have M0 and N0 as the block size, this is how we define "partial blocks"/ |
| * "boundary block" (we use the 2 terms "partial blocks" and "boundary blocks" interchangeably) and its various parameters: |
| * |
| * *--x--> x == 0 x == 1 |
| * | |<------------------------------N-------------------------->| |
| * y |<--------------N0------------->|<----PARTIAL_STORE_N0----->| |
| * | -------------############################################################# |
| * * | | |...............................|...........................| |
| * y == 0 | PAR_..._M0 |......Boundary block in y......|.Boundary block in x and y.| |
| * | | |...............................|...........................| |
| * M --############################################################# |
| * | | | |...........................| |
| * y == 1 | M0 | Non-boundary block |....Boundary block in x....| |
| * | | | |...........................| |
| * |------------############################################################# |
| * |
| * Then @p PARTIAL_STORE_M0 = M % M0 and @p PARTIAL_STORE_N0 = N % N0 |
| * |
| * @note in cases @p PARTIAL_STORE_N0 != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty. |
| * |
| * It automatically detects if a giving M,N,M0,N0 combination can yield partial blocks in either X and Y dimension, |
| * and select corresponding store methods such that the boundary detection logic is only added when needed. |
| * |
| * 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, for non-partial blocks. Supported: 1-16 |
| * @param[in] N0 The size of each vector, for non-partial blocks. 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 |
| * @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] 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, 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, 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, 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, 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 |
| |
| #endif // defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0) |
| /** @} */ // end of group STORE_BLOCK_BOUNDARY_AWARE |
| |
| #if defined(PARTIAL_STORE_M0) |
| /** Compute the start m0 row (LHS, BIAS and DST) in a boundary-aware way so as to avoid padding |
| * @name COMPUTE_M0_START_ROW |
| * If there're any partial blocks in y dimension, they are placed at the beginning of the rows. |
| * This shift amount is added to all rows such that the partial block (at the beginning) overlaps with the subsequent |
| * blocks in the y dimension to avoid any padding. |
| * EG: M0=4, PARTIAL_STORE_M0=1: |
| * | Non-overlapping | +M0_ROW_SHIFT (Overlapping) |
| * block 0 (partial)| start row = 0 | start row = 0 |
| * block 1 (full) | start row = 4 | start row = 1 |
| * block 2 (full) | start row = 8 | start row = 5 |
| * |
| * @param[in] y Global id of current block in y. |
| * @param[in] M0 The number of rows to store, for non-partial blocks. Supported: 1-16 |
| * @param[in] PARTIAL_STORE_M0 The partial size in y, for partial blocks. Supported: [0, @p M0) |
| * @{ |
| */ |
| #define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \ |
| ((uint)(max(0, (int)(y * M0) - (int)((M0 - PARTIAL_STORE_M0) % M0)))) |
| #else // defined(PARTIAL_STORE_M0) |
| #define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \ |
| ((uint)(y * M0)) |
| #endif // defined(PARTIAL_STORE_M0) |
| /** @} */ // end of group COMPUTE_M0_START_ROW |
| |
| /** Store a vector that can only be partial in x. |
| * |
| * @note in case @p vec_size or @p leftover != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty. |
| * |
| * The data to store is expected to end in a 0. |
| * E.g., for basename=c, the expected name is c0. |
| * |
| * @param[in] basename The name of the variable without trailing 0 |
| * @param[in] data_type The data type of the vector |
| * @param[in] ptr The base pointer |
| * @param[in] vec_size The vector size if cond = false. Supported: 1, 2, 3, 4, 8, 16 |
| * @param[in] leftover The vector size if cond = true. Supported range: [1, @p vec_size0) |
| * @param[in] cond Condition to select either vec_size0 or vec_size1 |
| * @{ |
| */ |
| #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, cond) |
| /** @} */ // end of group STORE_VECTOR_SELECT |
| |
| #if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16) |
| #pragma OPENCL EXTENSION cl_khr_fp16 : enable |
| #endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16) |
| |
| #if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8) |
| #pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : enable |
| #endif // defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8) |
| |
| #if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8) |
| #pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : enable |
| #endif // defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8) |
| |
| #if defined(ARM_COMPUTE_DEBUG_ENABLED) && defined(cl_arm_printf) |
| #pragma OPENCL EXTENSION cl_arm_printf : enable |
| #endif // defined(ARM_COMPUTE_DEBUG_ENABLED) && defined(cl_arm_printf) |
| |
| #define GPU_ARCH_MIDGARD 0x100 |
| #define GPU_ARCH_BIFROST 0x200 |
| |
| /** Concatenate two inputs. |
| * |
| * @param[in] a The first input to be concatenated |
| * @param[in] b The second input to be concatenated |
| * |
| * @return The concatenated output |
| */ |
| #define CONCAT(a, b) a##b |
| |
| /** Expand the given vector |
| * |
| * @param[in] x The vector to be expanded |
| * |
| * @return The expanded output |
| */ |
| #define EXPAND(x) x |
| |
| /** Clamp the given value between an upper and lower bound. |
| * |
| * @param[in] x The value to be clamped |
| * @param[in] min_val The lower bound |
| * @param[in] max_val The upper bound |
| * |
| * @return The clamped value. |
| */ |
| #define CLAMP(x, min_val, max_val) min(max(x, min_val), max_val) |
| |
| /** REVn reverses the given vector whose size is n. |
| * @name REVn |
| * |
| * @param[in] x The vector to be reversed |
| * |
| * @return The reversed vector |
| * @{ |
| */ |
| #define REV1(x) ((x)) |
| #define REV2(x) ((x).s10) |
| #define REV3(x) ((x).s210) |
| #define REV4(x) ((x).s3210) |
| #define REV8(x) ((x).s76543210) |
| #define REV16(x) ((x).sFEDCBA9876543210) |
| /** @} */ // end of group REVn |
| |
| /** Reverse the given vector. |
| * @name REVERSE |
| * |
| * @param[in] x The vector to be reversed |
| * @param[in] s The size of the vector |
| * |
| * @return The reversed vector |
| * @{ |
| */ |
| #define REVERSE_STR(x, s) REV##s((x)) |
| #define REVERSE(x, s) REVERSE_STR(x, s) |
| /** @} */ // end of group REVERSE |
| |
| /** Circular-right-shift (rotate-right) the vector of size s by the amount of n. |
| * @name ROTs_n |
| * |
| * @param[in] x The vector to be shifted |
| * |
| * @return The shifted vector |
| * @{ |
| */ |
| #define ROT1_0(x) ((x)) |
| |
| #define ROT2_0(x) ((x)) |
| #define ROT2_1(x) ((x).s10) |
| |
| #define ROT3_0(x) ((x)) |
| #define ROT3_1(x) ((x).s201) |
| #define ROT3_2(x) ((x).s120) |
| |
| #define ROT4_0(x) ((x)) |
| #define ROT4_1(x) ((x).s3012) |
| #define ROT4_2(x) ((x).s2301) |
| #define ROT4_3(x) ((x).s1230) |
| |
| #define ROT8_0(x) ((x)) |
| #define ROT8_1(x) ((x).s70123456) |
| #define ROT8_2(x) ((x).s67012345) |
| #define ROT8_3(x) ((x).s56701234) |
| #define ROT8_4(x) ((x).s45670123) |
| #define ROT8_5(x) ((x).s34567012) |
| #define ROT8_6(x) ((x).s23456701) |
| #define ROT8_7(x) ((x).s12345670) |
| |
| #define ROT16_0(x) ((x)) |
| #define ROT16_1(x) ((x).sF0123456789ABCDE) |
| #define ROT16_2(x) ((x).sEF0123456789ABCD) |
| #define ROT16_3(x) ((x).sDEF0123456789ABC) |
| #define ROT16_4(x) ((x).sCDEF0123456789AB) |
| #define ROT16_5(x) ((x).sBCDEF0123456789A) |
| #define ROT16_6(x) ((x).sABCDEF0123456789) |
| #define ROT16_7(x) ((x).s9ABCDEF012345678) |
| #define ROT16_8(x) ((x).s89ABCDEF01234567) |
| #define ROT16_9(x) ((x).s789ABCDEF0123456) |
| #define ROT16_10(x) ((x).s6789ABCDEF012345) |
| #define ROT16_11(x) ((x).s56789ABCDEF01234) |
| #define ROT16_12(x) ((x).s456789ABCDEF0123) |
| #define ROT16_13(x) ((x).s3456789ABCDEF012) |
| #define ROT16_14(x) ((x).s23456789ABCDEF01) |
| #define ROT16_15(x) ((x).s123456789ABCDEF0) |
| /** @} */ // end of group ROTs_n |
| |
| /** Circular-right-shift (rotate-right) the given vector by the given amount. |
| * @name ROTATE |
| * |
| * @param[in] x The vector to be shifted |
| * @param[in] s The size of the vector |
| * @param[in] n The amount to be shifted |
| * |
| * @return The shifted vector |
| * @{ |
| */ |
| #define ROTATE_STR(x, s, n) ROT##s##_##n(x) |
| #define ROTATE(x, s, n) ROTATE_STR(x, s, n) |
| /** @} */ // end of group ROTATE |
| |
| /** Creates a vector of size n filled with offset values corresponding to the location of each element. |
| * @name V_OFFSn |
| * |
| * @param[in] dt The data type of the output vector |
| * |
| * @return The vector filled with offset values |
| * @{ |
| */ |
| #define V_OFFS1(dt) (dt##1)(0) |
| #define V_OFFS2(dt) (dt##2)(0, 1) |
| #define V_OFFS3(dt) (dt##3)(0, 1, 2) |
| #define V_OFFS4(dt) (dt##4)(0, 1, 2, 3) |
| #define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7) |
| #define V_OFFS16(dt) (dt##16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15) |
| /** @} */ // end of group V_OFFSn |
| |
| /** Create a vector filled with offset values corresponding to the location of each element. |
| * @name VEC_OFFS |
| * |
| * @param[in] dt The data type of the output vector |
| * @param[in] s The size of the output vector |
| * |
| * @return The vector filled with offset values |
| * @{ |
| */ |
| #define VEC_OFFS_STR(dt, s) V_OFFS##s(dt) |
| #define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s) |
| /** @} */ // end of group VEC_OFFS |
| |
| #define VLOAD_STR(size) vload##size |
| #define VLOAD(size) VLOAD_STR(size) |
| |
| #define PIXEL_UNIT4 1 |
| #define PIXEL_UNIT8 2 |
| #define PIXEL_UNIT16 4 |
| |
| /** Utility macro to convert a vector size in pixel unit. |
| * |
| * @name CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT |
| * |
| * @param[in] vec_size Vector size. Only 4,8 and 16 is supported |
| * |
| * @return The pixel unit (number of pixels) |
| * @{ |
| */ |
| #define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) PIXEL_UNIT##vec_size |
| #define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) |
| /** @} */ // end of group CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT |
| |
| #define read_image2d_floatx1(img, x_coord, y_coord) (float4)(read_imagef(img, (int2)(x_coord, y_coord))); |
| #define read_image2d_floatx2(img, x_coord, y_coord) (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord))); |
| #define read_image2d_floatx4(img, x_coord, y_coord) (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord))); |
| |
| #if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16) |
| #define read_image2d_halfx1(img, x_coord, y_coord) (half4)(read_imageh(img, (int2)(x_coord, y_coord))); |
| #define read_image2d_halfx2(img, x_coord, y_coord) (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord))); |
| #define read_image2d_halfx4(img, x_coord, y_coord) (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord))); |
| #endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16) |
| |
| /** Utility macro to read a 2D OpenCL image object. |
| * |
| * @note Coordinates are not normalized |
| * |
| * @param[in] data_type Data type |
| * @param[in] n0 Number of pixel to read. Only 1,2 and 4 is supported |
| * @param[in] img OpenCL image object |
| * @param[in] x_coord The x coordinate for the top-left pixel |
| * @param[in] y_coord The y coordinate for the top-left pixel |
| * |
| * @return Pixels from the 2D OpenCL image object |
| * @{ |
| */ |
| #define READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) read_image2d_##data_type##x##n0(img, x_coord, y_coord) |
| #define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) |
| |
| #define VSTORE_STR(size) vstore##size |
| #define VSTORE(size) VSTORE_STR(size) |
| |
| #define float1 float |
| #define half1 half |
| #define char1 char |
| #define uchar1 uchar |
| #define short1 short |
| #define ushort1 ushort |
| #define int1 int |
| #define uint1 uint |
| #define long1 long |
| #define ulong1 ulong |
| #define double1 double |
| |
| #define vload1(OFFSET, PTR) *(OFFSET + PTR) |
| #define vstore1(DATA, OFFSET, PTR) *(OFFSET + PTR) = DATA |
| |
| /** Extended partial vstore that correctly handles scalar values as well. |
| * Store the **lower** 0 to (n-1)th elements of the given vector while minimising the amount of vstore ops |
| * @name VSTORE_PARTIAL |
| * |
| * @note With this macro, the passed data can be both a vector and a scalar |
| * @note @p store_size needs to be <= @p size |
| * eg 1: Valid |
| * VSTORE_PARTIAL(16, 15) ...; |
| * eg 2: Invalid |
| * VSTORE_PARTIAL(4, 7) ...; |
| * |
| * @param[in] size The width of @p DATA. Supported values: 1(scalar), 2, 3, 4, 8, 16 |
| * @param[in] store_size The number of lower elements to store. Supported values: 1-16, but has to be <= @p size |
| * @{ |
| */ |
| #define VSTORE_PARTIAL_STR(size, store_size) vstore_partial_##size##_##store_size |
| #define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size) |
| |
| #define NO_STORE(data, offs, ptr) \ |
| { \ |
| } |
| |
| // Size == 1 (scalar) |
| #define vstore_partial_1_0 NO_STORE |
| #define vstore_partial_1_1 vstore1 |
| #define vstore_partial_1_2 NO_STORE |
| #define vstore_partial_1_3 NO_STORE |
| #define vstore_partial_1_4 NO_STORE |
| #define vstore_partial_1_5 NO_STORE |
| #define vstore_partial_1_6 NO_STORE |
| #define vstore_partial_1_7 NO_STORE |
| #define vstore_partial_1_8 NO_STORE |
| #define vstore_partial_1_9 NO_STORE |
| #define vstore_partial_1_10 NO_STORE |
| #define vstore_partial_1_11 NO_STORE |
| #define vstore_partial_1_12 NO_STORE |
| #define vstore_partial_1_13 NO_STORE |
| #define vstore_partial_1_14 NO_STORE |
| #define vstore_partial_1_15 NO_STORE |
| #define vstore_partial_1_16 NO_STORE |
| // Size == 2 |
| #define vstore_partial_2_0 NO_STORE |
| #define vstore_partial_2_1 vstore_partial_1 |
| #define vstore_partial_2_2 vstore_partial_2 |
| #define vstore_partial_2_3 NO_STORE |
| #define vstore_partial_2_4 NO_STORE |
| #define vstore_partial_2_5 NO_STORE |
| #define vstore_partial_2_6 NO_STORE |
| #define vstore_partial_2_7 NO_STORE |
| #define vstore_partial_2_8 NO_STORE |
| #define vstore_partial_2_9 NO_STORE |
| #define vstore_partial_2_10 NO_STORE |
| #define vstore_partial_2_11 NO_STORE |
| #define vstore_partial_2_12 NO_STORE |
| #define vstore_partial_2_13 NO_STORE |
| #define vstore_partial_2_14 NO_STORE |
| #define vstore_partial_2_15 NO_STORE |
| #define vstore_partial_2_16 NO_STORE |
| // Size == 3 |
| #define vstore_partial_3_0 NO_STORE |
| #define vstore_partial_3_1 vstore_partial_1 |
| #define vstore_partial_3_2 vstore_partial_2 |
| #define vstore_partial_3_3 vstore_partial_3 |
| #define vstore_partial_3_4 NO_STORE |
| #define vstore_partial_3_5 NO_STORE |
| #define vstore_partial_3_6 NO_STORE |
| #define vstore_partial_3_7 NO_STORE |
| #define vstore_partial_3_8 NO_STORE |
| #define vstore_partial_3_9 NO_STORE |
| #define vstore_partial_3_10 NO_STORE |
| #define vstore_partial_3_11 NO_STORE |
| #define vstore_partial_3_12 NO_STORE |
| #define vstore_partial_3_13 NO_STORE |
| #define vstore_partial_3_14 NO_STORE |
| #define vstore_partial_3_15 NO_STORE |
| #define vstore_partial_3_16 NO_STORE |
| // Size == 4 |
| #define vstore_partial_4_0 NO_STORE |
| #define vstore_partial_4_1 vstore_partial_1 |
| #define vstore_partial_4_2 vstore_partial_2 |
| #define vstore_partial_4_3 vstore_partial_3 |
| #define vstore_partial_4_4 vstore_partial_4 |
| #define vstore_partial_4_5 NO_STORE |
| #define vstore_partial_4_6 NO_STORE |
| #define vstore_partial_4_7 NO_STORE |
| #define vstore_partial_4_8 NO_STORE |
| #define vstore_partial_4_9 NO_STORE |
| #define vstore_partial_4_10 NO_STORE |
| #define vstore_partial_4_11 NO_STORE |
| #define vstore_partial_4_12 NO_STORE |
| #define vstore_partial_4_13 NO_STORE |
| #define vstore_partial_4_14 NO_STORE |
| #define vstore_partial_4_15 NO_STORE |
| #define vstore_partial_4_16 NO_STORE |
| // Size == 8 |
| #define vstore_partial_8_0 NO_STORE |
| #define vstore_partial_8_1 vstore_partial_1 |
| #define vstore_partial_8_2 vstore_partial_2 |
| #define vstore_partial_8_3 vstore_partial_3 |
| #define vstore_partial_8_4 vstore_partial_4 |
| #define vstore_partial_8_5 vstore_partial_5 |
| #define vstore_partial_8_6 vstore_partial_6 |
| #define vstore_partial_8_7 vstore_partial_7 |
| #define vstore_partial_8_8 vstore_partial_8 |
| #define vstore_partial_8_9 NO_STORE |
| #define vstore_partial_8_10 NO_STORE |
| #define vstore_partial_8_11 NO_STORE |
| #define vstore_partial_8_12 NO_STORE |
| #define vstore_partial_8_13 NO_STORE |
| #define vstore_partial_8_14 NO_STORE |
| #define vstore_partial_8_15 NO_STORE |
| #define vstore_partial_8_16 NO_STORE |
| // Size == 16 |
| #define vstore_partial_16_0 NO_STORE |
| #define vstore_partial_16_1 vstore_partial_1 |
| #define vstore_partial_16_2 vstore_partial_2 |
| #define vstore_partial_16_3 vstore_partial_3 |
| #define vstore_partial_16_4 vstore_partial_4 |
| #define vstore_partial_16_5 vstore_partial_5 |
| #define vstore_partial_16_6 vstore_partial_6 |
| #define vstore_partial_16_7 vstore_partial_7 |
| #define vstore_partial_16_8 vstore_partial_8 |
| #define vstore_partial_16_9 vstore_partial_9 |
| #define vstore_partial_16_10 vstore_partial_10 |
| #define vstore_partial_16_11 vstore_partial_11 |
| #define vstore_partial_16_12 vstore_partial_12 |
| #define vstore_partial_16_13 vstore_partial_13 |
| #define vstore_partial_16_14 vstore_partial_14 |
| #define vstore_partial_16_15 vstore_partial_15 |
| #define vstore_partial_16_16 vstore_partial_16 |
| |
| /** Partial vstore. Store the **lower** 0 to (n-1)th elements of the given vector while minimising the amount of vstore ops |
| * @name vstore_partial_n |
| * |
| * @note @p DATA needs to be a vector not a scalar |
| * @note n needs to be <= the vector width of the input variable @p DATA |
| * eg 1: Valid |
| * vstore_partial_15(var:float16, 0, 0xabcd); |
| * eg 2: Invalid |
| * vstore_partial_7(var:float4, 0, 0xabcd); |
| * |
| * @note in cases n == 1, 2, 3, 4, 8, 16, no extra vstore is invoked, thus there's no performance penalty. |
| * |
| * @param[in] DATA The name of the variable |
| * @param[in] OFFSET Offset in n |
| * @param[in] PTR The base pointer |
| * @{ |
| */ |
| #define vstore_partial_1(DATA, OFFSET, PTR) \ |
| vstore1(DATA.s0, OFFSET, PTR); |
| |
| #define vstore_partial_2(DATA, OFFSET, PTR) \ |
| vstore2(DATA.s01, OFFSET, PTR); |
| |
| #define vstore_partial_3(DATA, OFFSET, PTR) \ |
| vstore3(DATA.s012, OFFSET, PTR); |
| |
| #define vstore_partial_4(DATA, OFFSET, PTR) \ |
| vstore4(DATA.s0123, OFFSET, PTR); |
| |
| #define vstore_partial_5(DATA, OFFSET, PTR) \ |
| vstore_partial_4(DATA.s0123, OFFSET, PTR); \ |
| vstore1(DATA.s4, OFFSET, PTR + 4); |
| |
| #define vstore_partial_6(DATA, OFFSET, PTR) \ |
| vstore_partial_4(DATA.s0123, OFFSET, PTR); \ |
| vstore_partial_2(DATA.s45, OFFSET, PTR + 4); |
| |
| #define vstore_partial_7(DATA, OFFSET, PTR) \ |
| vstore_partial_4(DATA.s0123, OFFSET, PTR); \ |
| vstore_partial_3(DATA.s456, OFFSET, PTR + 4); |
| |
| #define vstore_partial_8(DATA, OFFSET, PTR) \ |
| vstore8(DATA.s01234567, OFFSET, PTR); |
| |
| #define vstore_partial_9(DATA, OFFSET, PTR) \ |
| vstore_partial_8(DATA.s01234567, OFFSET, PTR); \ |
| vstore1(DATA.s8, OFFSET, PTR + 8); |
| |
| #define vstore_partial_10(DATA, OFFSET, PTR) \ |
| vstore_partial_8(DATA.s01234567, OFFSET, PTR); \ |
| vstore_partial_2(DATA.s89, OFFSET, PTR + 8); |
| |
| #define vstore_partial_11(DATA, OFFSET, PTR) \ |
| vstore_partial_8(DATA.s01234567, OFFSET, PTR); \ |
| vstore_partial_3(DATA.s89a, OFFSET, PTR + 8); |
| |
| #define vstore_partial_12(DATA, OFFSET, PTR) \ |
| vstore_partial_8(DATA.s01234567, OFFSET, PTR); \ |
| vstore_partial_4(DATA.s89ab, OFFSET, PTR + 8); |
| |
| #define vstore_partial_13(DATA, OFFSET, PTR) \ |
| vstore_partial_8(DATA.s01234567, OFFSET, PTR); \ |
| vstore_partial_5(DATA.s89abcdef, OFFSET, PTR + 8); |
| |
| #define vstore_partial_14(DATA, OFFSET, PTR) \ |
| vstore_partial_8(DATA.s01234567, OFFSET, PTR); \ |
| vstore_partial_6(DATA.s89abcdef, OFFSET, PTR + 8); |
| |
| #define vstore_partial_15(DATA, OFFSET, PTR) \ |
| vstore_partial_8(DATA.s01234567, OFFSET, PTR); \ |
| vstore_partial_7(DATA.s89abcdef, OFFSET, PTR + 8); |
| |
| #define vstore_partial_16(DATA, OFFSET, PTR) \ |
| vstore16(DATA, OFFSET, PTR); |
| /** @} */ // end of groupd vstore_partial_n |
| /** @} */ // end of groupd VSTORE_PARTIAL |
| |
| // Convert built-in functions with _sat modifier are not supported in floating point so we create defines |
| // without _sat to overcome this issue |
| #define convert_float_sat convert_float |
| #define convert_float1_sat convert_float |
| #define convert_float2_sat convert_float2 |
| #define convert_float3_sat convert_float3 |
| #define convert_float4_sat convert_float4 |
| #define convert_float8_sat convert_float8 |
| #define convert_float16_sat convert_float16 |
| #define convert_half_sat convert_float |
| #define convert_half1_sat convert_half |
| #define convert_half2_sat convert_half2 |
| #define convert_half3_sat convert_half3 |
| #define convert_half4_sat convert_half4 |
| #define convert_half8_sat convert_half8 |
| #define convert_half16_sat convert_half16 |
| |
| #define convert_float1 convert_float |
| #define convert_half1 convert_half |
| #define convert_char1 convert_char |
| #define convert_uchar1 convert_uchar |
| #define convert_short1 convert_short |
| #define convert_ushort1 convert_ushort |
| #define convert_int1 convert_int |
| #define convert_uint1 convert_uint |
| #define convert_long1 convert_long |
| #define convert_ulong1 convert_ulong |
| #define convert_double1 convert_double |
| |
| #define convert_char1_sat convert_char_sat |
| #define convert_uchar1_sat convert_uchar_sat |
| #define convert_short1_sat convert_short_sat |
| #define convert_ushort1_sat convert_ushort_sat |
| #define convert_int1_sat convert_int_sat |
| #define convert_uint1_sat convert_uint_sat |
| #define convert_long1_sat convert_long_sat |
| #define convert_ulong1_sat convert_ulong_sat |
| #define convert_double1_sat convert_double_sat |
| |
| #define VEC_DATA_TYPE_STR(type, size) type##size |
| #define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size) |
| |
| #define CONVERT_STR(x, type) (convert_##type((x))) |
| #define CONVERT(x, type) CONVERT_STR(x, type) |
| |
| #define CONVERT_SAT_STR(x, type) (convert_##type##_sat((x))) |
| #define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type) |
| |
| #define CONVERT_SAT_ROUND_STR(x, type, round) (convert_##type##_sat_##round((x))) |
| #define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round) |
| |
| #define select_vec_dt_uchar(size) uchar##size |
| #define select_vec_dt_char(size) char##size |
| #define select_vec_dt_ushort(size) ushort##size |
| #define select_vec_dt_short(size) short##size |
| #define select_vec_dt_half(size) short##size |
| #define select_vec_dt_uint(size) uint##size |
| #define select_vec_dt_int(size) int##size |
| #define select_vec_dt_float(size) int##size |
| #define select_vec_dt_ulong(size) ulong##size |
| #define select_vec_dt_long(size) long##size |
| |
| #define SELECT_VEC_DATA_TYPE_STR(type, size) select_vec_dt_##type(size) |
| #define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size) |
| #define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1) |
| |
| #define sum_reduce_1(x) (x) |
| #define sum_reduce_2(x) ((x).s0) + ((x).s1) |
| #define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2) |
| #define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23) |
| #define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567) |
| #define sum_reduce_16(x) sum_reduce_8((x).s01234567) + sum_reduce_8((x).s89ABCDEF) |
| |
| #define SUM_REDUCE_STR(x, size) sum_reduce_##size(x) |
| #define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size) |
| |
| #define max_reduce_1(x) (x) |
| #define max_reduce_2(x) max(((x).s0), ((x).s1)) |
| #define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2)) |
| #define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23)) |
| #define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567)) |
| #define max_reduce_16(x) max(max_reduce_8((x).s01234567), max_reduce_8((x).s89ABCDEF)) |
| |
| #define MAX_REDUCE_STR(x, size) max_reduce_##size(x) |
| #define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size) |
| |
| #define VECTOR_DECLARATION(name) \ |
| __global uchar *name##_ptr, \ |
| uint name##_stride_x, \ |
| uint name##_step_x, \ |
| uint name##_offset_first_element_in_bytes |
| |
| #define IMAGE_DECLARATION(name) \ |
| __global uchar *name##_ptr, \ |
| uint name##_stride_x, \ |
| uint name##_step_x, \ |
| uint name##_stride_y, \ |
| uint name##_step_y, \ |
| uint name##_offset_first_element_in_bytes |
| |
| #define TENSOR3D_DECLARATION(name) \ |
| __global uchar *name##_ptr, \ |
| uint name##_stride_x, \ |
| uint name##_step_x, \ |
| uint name##_stride_y, \ |
| uint name##_step_y, \ |
| uint name##_stride_z, \ |
| uint name##_step_z, \ |
| uint name##_offset_first_element_in_bytes |
| |
| #define TENSOR4D_DECLARATION(name) \ |
| __global uchar *name##_ptr, \ |
| uint name##_stride_x, \ |
| uint name##_step_x, \ |
| uint name##_stride_y, \ |
| uint name##_step_y, \ |
| uint name##_stride_z, \ |
| uint name##_step_z, \ |
| uint name##_stride_w, \ |
| uint name##_step_w, \ |
| uint name##_offset_first_element_in_bytes |
| |
| #define CONVERT_TO_VECTOR_STRUCT(name) \ |
| update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x) |
| |
| #define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name) \ |
| update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0) |
| |
| #define CONVERT_TO_IMAGE_STRUCT(name) \ |
| update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y) |
| |
| #define CONVERT_TO_IMAGE_STRUCT_NO_STEP(name) \ |
| update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0) |
| |
| #define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \ |
| update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z) |
| |
| #define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \ |
| update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, name##_step_z) |
| |
| #define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \ |
| update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z) |
| |
| #define CONVERT_TO_TENSOR3D_STRUCT(name) \ |
| update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \ |
| name##_stride_z, name##_step_z) |
| |
| #define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \ |
| update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0) |
| |
| #define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size) \ |
| update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \ |
| name##_stride_z, name##_step_z, name##_stride_w, name##_step_w, mod_size) |
| |
| #define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \ |
| update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size) |
| |
| #define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name) \ |
| tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \ |
| name##_stride_z, name##_step_z) |
| |
| /** Structure to hold Vector information */ |
| typedef struct Vector |
| { |
| __global uchar *ptr; /**< Pointer to the starting postion of the buffer */ |
| int offset_first_element_in_bytes; /**< The offset of the first element in the source image */ |
| int stride_x; /**< Stride of the image in X dimension (in bytes) */ |
| } Vector; |
| |
| /** Structure to hold Image information */ |
| typedef struct Image |
| { |
| __global uchar *ptr; /**< Pointer to the starting postion of the buffer */ |
| int offset_first_element_in_bytes; /**< The offset of the first element in the source image */ |
| int stride_x; /**< Stride of the image in X dimension (in bytes) */ |
| int stride_y; /**< Stride of the image in Y dimension (in bytes) */ |
| } Image; |
| |
| /** Structure to hold 3D tensor information */ |
| typedef struct Tensor3D |
| { |
| __global uchar *ptr; /**< Pointer to the starting postion of the buffer */ |
| int offset_first_element_in_bytes; /**< The offset of the first element in the source image */ |
| int stride_x; /**< Stride of the image in X dimension (in bytes) */ |
| int stride_y; /**< Stride of the image in Y dimension (in bytes) */ |
| int stride_z; /**< Stride of the image in Z dimension (in bytes) */ |
| } Tensor3D; |
| |
| /** Structure to hold 4D tensor information */ |
| typedef struct Tensor4D |
| { |
| __global uchar *ptr; /**< Pointer to the starting postion of the buffer */ |
| int offset_first_element_in_bytes; /**< The offset of the first element in the source image */ |
| int stride_x; /**< Stride of the image in X dimension (in bytes) */ |
| int stride_y; /**< Stride of the image in Y dimension (in bytes) */ |
| int stride_z; /**< Stride of the image in Z dimension (in bytes) */ |
| int stride_w; /**< Stride of the image in W dimension (in bytes) */ |
| } Tensor4D; |
| |
| /** Wrap vector information into an Vector structure, and make the pointer point at this workitem's data. |
| * |
| * @param[in] ptr Pointer to the starting postion of the buffer |
| * @param[in] offset_first_element_in_bytes The offset of the first element in the source vector |
| * @param[in] stride_x Stride of the vector in X dimension (in bytes) |
| * @param[in] step_x stride_x * number of elements along X processed per workitem(in bytes) |
| * |
| * @return An image object |
| */ |
| inline Vector update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x) |
| { |
| Vector vector = |
| { |
| .ptr = ptr, |
| .offset_first_element_in_bytes = offset_first_element_in_bytes, |
| .stride_x = stride_x, |
| }; |
| vector.ptr += vector.offset_first_element_in_bytes + get_global_id(0) * step_x; |
| return vector; |
| } |
| |
| /** Wrap image information into an Image structure, and make the pointer point at this workitem's data. |
| * |
| * @param[in] ptr Pointer to the starting postion of the buffer |
| * @param[in] offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] stride_x Stride of the image in X dimension (in bytes) |
| * @param[in] step_x stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] stride_y Stride of the image in Y dimension (in bytes) |
| * @param[in] step_y stride_y * number of elements along Y processed per workitem(in bytes) |
| * |
| * @return An image object |
| */ |
| inline Image update_image_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y) |
| { |
| Image img = |
| { |
| .ptr = ptr, |
| .offset_first_element_in_bytes = offset_first_element_in_bytes, |
| .stride_x = stride_x, |
| .stride_y = stride_y |
| }; |
| img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y; |
| return img; |
| } |
| |
| /** Wrap 3D tensor information into an image structure, and make the pointer point at this workitem's data. |
| * |
| * @param[in] ptr Pointer to the starting postion of the buffer |
| * @param[in] offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] stride_x Stride of the image in X dimension (in bytes) |
| * @param[in] step_x stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] stride_y Stride of the image in Y dimension (in bytes) |
| * @param[in] step_y stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] stride_z Stride of the image in Z dimension (in bytes) |
| * @param[in] step_z stride_z * number of elements along Z processed per workitem(in bytes) |
| * |
| * @return A 3D tensor object |
| */ |
| inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z) |
| { |
| Image img = |
| { |
| .ptr = ptr, |
| .offset_first_element_in_bytes = offset_first_element_in_bytes, |
| .stride_x = stride_x, |
| .stride_y = stride_y |
| }; |
| img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z; |
| return img; |
| } |
| |
| /** Wrap 3D tensor information into an tensor structure, and make the pointer point at this workitem's data. |
| * |
| * @param[in] ptr Pointer to the starting postion of the buffer |
| * @param[in] offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] stride_x Stride of the image in X dimension (in bytes) |
| * @param[in] step_x stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] stride_y Stride of the image in Y dimension (in bytes) |
| * @param[in] step_y stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] stride_z Stride of the image in Z dimension (in bytes) |
| * @param[in] step_z stride_z * number of elements along Z processed per workitem(in bytes) |
| * |
| * @return A 3D tensor object |
| */ |
| inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z) |
| { |
| Tensor3D tensor = |
| { |
| .ptr = ptr, |
| .offset_first_element_in_bytes = offset_first_element_in_bytes, |
| .stride_x = stride_x, |
| .stride_y = stride_y, |
| .stride_z = stride_z |
| }; |
| tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z; |
| return tensor; |
| } |
| |
| /** Wrap 3D tensor information into an tensor structure. |
| * |
| * @param[in] ptr Pointer to the starting postion of the buffer |
| * @param[in] offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] stride_x Stride of the image in X dimension (in bytes) |
| * @param[in] step_x stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] stride_y Stride of the image in Y dimension (in bytes) |
| * @param[in] step_y stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] stride_z Stride of the image in Z dimension (in bytes) |
| * @param[in] step_z stride_z * number of elements along Z processed per workitem(in bytes) |
| * |
| * @return A 3D tensor object |
| */ |
| inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z) |
| { |
| Tensor3D tensor = |
| { |
| .ptr = ptr, |
| .offset_first_element_in_bytes = offset_first_element_in_bytes, |
| .stride_x = stride_x, |
| .stride_y = stride_y, |
| .stride_z = stride_z |
| }; |
| return tensor; |
| } |
| |
| inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z, uint stride_w, |
| uint step_w, |
| uint mod_size) |
| { |
| Tensor4D tensor = |
| { |
| .ptr = ptr, |
| .offset_first_element_in_bytes = offset_first_element_in_bytes, |
| .stride_x = stride_x, |
| .stride_y = stride_y, |
| .stride_z = stride_z, |
| .stride_w = stride_w |
| }; |
| |
| tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w; |
| return tensor; |
| } |
| |
| /** Get the pointer position of a Vector |
| * |
| * @param[in] vec Pointer to the starting position of the buffer |
| * @param[in] x Relative X position |
| */ |
| inline __global const uchar *vector_offset(const Vector *vec, int x) |
| { |
| return vec->ptr + x * vec->stride_x; |
| } |
| |
| /** Get the pointer position of a Image |
| * |
| * @param[in] img Pointer to the starting position of the buffer |
| * @param[in] x Relative X position |
| * @param[in] y Relative Y position |
| */ |
| inline __global uchar *offset(const Image *img, int x, int y) |
| { |
| return img->ptr + x * img->stride_x + y * img->stride_y; |
| } |
| |
| /** Get the pointer position of a Tensor3D |
| * |
| * @param[in] tensor Pointer to the starting position of the buffer |
| * @param[in] x Relative X position |
| * @param[in] y Relative Y position |
| * @param[in] z Relative Z position |
| */ |
| inline __global const uchar *tensor3D_offset(const Tensor3D *tensor, int x, int y, int z) |
| { |
| return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z; |
| } |
| |
| /** Get the pointer position of a Tensor4D |
| * |
| * @param[in] tensor Pointer to the starting position of the buffer |
| * @param[in] x Relative X position |
| * @param[in] y Relative Y position |
| * @param[in] z Relative Z position |
| * @param[in] w Relative W position |
| */ |
| inline __global const uchar *tensor4D_offset(const Tensor4D *tensor, int x, int y, int z, int w) |
| { |
| return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + w * tensor->stride_w; |
| } |
| |
| /** Get the offset for a given linear index of a Tensor3D |
| * |
| * @param[in] tensor Pointer to the starting position of the buffer |
| * @param[in] width Width of the input tensor |
| * @param[in] height Height of the input tensor |
| * @param[in] depth Depth of the input tensor |
| * @param[in] index Linear index |
| */ |
| inline __global const uchar *tensor3D_index2ptr(const Tensor3D *tensor, uint width, uint height, uint depth, uint index) |
| { |
| uint num_elements = width * height; |
| |
| const uint z = index / num_elements; |
| |
| index %= num_elements; |
| |
| const uint y = index / width; |
| |
| index %= width; |
| |
| const uint x = index; |
| |
| return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + tensor->offset_first_element_in_bytes; |
| } |
| |
| #endif // _HELPER_H |
| |
| /** Convert an RGB888 image to RGBX8888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void RGB888_to_RGBA8888_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out = CONVERT_TO_IMAGE_STRUCT(output); |
| |
| // handle 16 pixels every time |
| uchar16 rgb_0 = vload16(0, in.ptr); |
| uchar16 rgb_1 = vload16(0, in.ptr + 16); |
| uchar16 rgb_2 = vload16(0, in.ptr + 32); |
| |
| uchar16 rgba_0 = (uchar16)(rgb_0.s012, 255, rgb_0.s345, 255, rgb_0.s678, 255, rgb_0.s9ab, 255); |
| uchar16 rgba_1 = (uchar16)(rgb_0.scde, 255, rgb_0.sf, rgb_1.s01, 255, rgb_1.s234, 255, rgb_1.s567, 255); |
| uchar16 rgba_2 = (uchar16)(rgb_1.s89a, 255, rgb_1.sbcd, 255, rgb_1.sef, rgb_2.s0, 255, rgb_2.s123, 255); |
| uchar16 rgba_3 = (uchar16)(rgb_2.s456, 255, rgb_2.s789, 255, rgb_2.sabc, 255, rgb_2.sdef, 255); |
| |
| vstore16(rgba_0, 0, out.ptr); |
| vstore16(rgba_1, 0, out.ptr + 16); |
| vstore16(rgba_2, 0, out.ptr + 32); |
| vstore16(rgba_3, 0, out.ptr + 48); |
| } |
| |
| /** Convert an RGB888 image to U8 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: RGB888 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void RGB888_to_U8_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out = CONVERT_TO_IMAGE_STRUCT(output); |
| |
| // handle 16 pixels every time |
| const uchar16 rgb_0 = vload16(0, in.ptr); |
| const uchar16 rgb_1 = vload16(0, in.ptr + 16); |
| const uchar16 rgb_2 = vload16(0, in.ptr + 32); |
| |
| //Resequence values from a sequence of 16 RGB values to sequence of 16 R, 16 G, 16 B values |
| const uchar16 rgb_r = (uchar16)(rgb_0.s0369, rgb_0.scf, rgb_1.s258b, rgb_1.se, rgb_2.s147a, rgb_2.sd); |
| const uchar16 rgb_g = (uchar16)(rgb_0.s147a, rgb_0.sd, rgb_1.s0369, rgb_1.scf, rgb_2.s258b, rgb_2.se); |
| const uchar16 rgb_b = (uchar16)(rgb_0.s258b, rgb_0.se, rgb_1.s147a, rgb_1.sd, rgb_2.s0369, rgb_2.scf); |
| |
| const float16 rgb2u8_red_coef_bt709 = 0.2126f; |
| const float16 rgb2u8_green_coef_bt709 = 0.7152f; |
| const float16 rgb2u8_blue_coef_bt709 = 0.0722f; |
| |
| //Computation of 16 greyscale values in float |
| const float16 greyscale_f_0 = rgb2u8_red_coef_bt709 * convert_float16(rgb_r) + rgb2u8_green_coef_bt709 * convert_float16(rgb_g) + rgb2u8_blue_coef_bt709 * convert_float16(rgb_b); |
| |
| //Convert it to 16 grayscale uchar values |
| const uchar16 greyscale_u8_0 = convert_uchar16_sat_rtz(greyscale_f_0); |
| |
| vstore16(greyscale_u8_0, 0, out.ptr); |
| } |
| |
| /** Convert an RGB888 image to RGBX8888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void RGBA8888_to_RGB888_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out = CONVERT_TO_IMAGE_STRUCT(output); |
| // handle 16 pixels every time |
| uchar16 rgba_0 = vload16(0, in.ptr); |
| uchar16 rgba_1 = vload16(0, in.ptr + 16); |
| uchar16 rgba_2 = vload16(0, in.ptr + 32); |
| uchar16 rgba_3 = vload16(0, in.ptr + 48); |
| |
| uchar16 rgb_0 = (uchar16)(rgba_0.s01245689, rgba_0.sacde, rgba_1.s0124); |
| uchar16 rgb_1 = (uchar16)(rgba_1.s5689acde, rgba_2.s01245689); |
| uchar16 rgb_2 = (uchar16)(rgba_2.sacde, rgba_3.s01245689, rgba_3.sacde); |
| |
| vstore16(rgb_0, 0, out.ptr); |
| vstore16(rgb_1, 0, out.ptr + 16); |
| vstore16(rgb_2, 0, out.ptr + 32); |
| } |
| |
| /** Convert a UYVY422 image to RGB888 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 8), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void UYVY422_to_RGB888_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out = CONVERT_TO_IMAGE_STRUCT(output); |
| |
| // handle 8 pixels every time |
| uchar16 uyvy = vload16(0, in.ptr); |
| |
| uchar8 luma = (uchar8)(uyvy.s1, uyvy.s3, uyvy.s5, uyvy.s7, uyvy.s9, uyvy.sb, uyvy.sd, uyvy.sf); |
| char8 cb = (char8)(uyvy.s0, uyvy.s0, uyvy.s4, uyvy.s4, uyvy.s8, uyvy.s8, uyvy.sc, uyvy.sc) - (char8)(128); |
| char8 cr = (char8)(uyvy.s2, uyvy.s2, uyvy.s6, uyvy.s6, uyvy.sa, uyvy.sa, uyvy.se, uyvy.se) - (char8)(128); |
| |
| float8 red_coef_bt709 = (float8)(1.5748f); |
| float8 green_coef_bt709 = (float8)(-0.1873f); |
| float8 green_coef2_bt709 = (float8)(-0.4681f); |
| float8 blue_coef_bt709 = (float8)(1.8556f); |
| float8 lumav = convert_float8(luma); |
| |
| float8 f_r = red_coef_bt709 * convert_float8(cr); |
| float8 f_g = green_coef_bt709 * convert_float8(cb) + green_coef2_bt709 * convert_float8(cr); |
| float8 f_b = blue_coef_bt709 * convert_float8(cb); |
| |
| f_r += lumav; |
| f_g += lumav; |
| f_b += lumav; |
| |
| uchar8 r_0 = convert_uchar8_sat_rtz(f_r); |
| uchar8 g_0 = convert_uchar8_sat_rtz(f_g); |
| uchar8 b_0 = convert_uchar8_sat_rtz(f_b); |
| |
| uchar16 rgb_0 = (uchar16)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2, b_0.s2, |
| r_0.s3, g_0.s3, b_0.s3, r_0.s4, g_0.s4, b_0.s4, r_0.s5); |
| uchar8 rgb_1 = (uchar8)(g_0.s5, b_0.s5, r_0.s6, g_0.s6, b_0.s6, r_0.s7, g_0.s7, b_0.s7); |
| |
| vstore16(rgb_0, 0, out.ptr); |
| vstore8(rgb_1, 0, out.ptr + 16); |
| } |
| |
| /** Convert a UYVY422 image to RGBX8888 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 8), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void UYVY422_to_RGBA8888_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out = CONVERT_TO_IMAGE_STRUCT(output); |
| |
| // handle 8 pixels every time |
| uchar16 uyvy = vload16(0, in.ptr); |
| |
| uchar8 luma = (uchar8)(uyvy.s1, uyvy.s3, uyvy.s5, uyvy.s7, uyvy.s9, uyvy.sb, uyvy.sd, uyvy.sf); |
| char8 cb = (char8)(uyvy.s0, uyvy.s0, uyvy.s4, uyvy.s4, uyvy.s8, uyvy.s8, uyvy.sc, uyvy.sc) - (char8)(128); |
| char8 cr = (char8)(uyvy.s2, uyvy.s2, uyvy.s6, uyvy.s6, uyvy.sa, uyvy.sa, uyvy.se, uyvy.se) - (char8)(128); |
| |
| float8 red_coef_bt709 = (float8)(1.5748f); |
| float8 green_coef_bt709 = (float8)(-0.1873f); |
| float8 green_coef2_bt709 = (float8)(-0.4681f); |
| float8 blue_coef_bt709 = (float8)(1.8556f); |
| float8 lumav = convert_float8(luma); |
| |
| float8 f_r = red_coef_bt709 * convert_float8(cr); |
| float8 f_g = green_coef_bt709 * convert_float8(cb) + green_coef2_bt709 * convert_float8(cr); |
| float8 f_b = blue_coef_bt709 * convert_float8(cb); |
| |
| f_r += lumav; |
| f_g += lumav; |
| f_b += lumav; |
| |
| uchar8 r_0 = convert_uchar8_sat_rtz(f_r); |
| uchar8 g_0 = convert_uchar8_sat_rtz(f_g); |
| uchar8 b_0 = convert_uchar8_sat_rtz(f_b); |
| |
| uchar16 rgba_0 = (uchar16)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255, |
| r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| uchar16 rgba_1 = (uchar16)(r_0.s4, g_0.s4, b_0.s4, 255, r_0.s5, g_0.s5, b_0.s5, 255, |
| r_0.s6, g_0.s6, b_0.s6, 255, r_0.s7, g_0.s7, b_0.s7, 255); |
| |
| vstore16(rgba_0, 0, out.ptr); |
| vstore16(rgba_1, 0, out.ptr + 16); |
| } |
| |
| /** Convert a YUYV422 image to RGB888 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 8), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void YUYV422_to_RGB888_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out = CONVERT_TO_IMAGE_STRUCT(output); |
| |
| // handle 8 pixels every time |
| uchar16 uyvy = vload16(0, in.ptr); |
| |
| uchar8 luma = (uchar8)(uyvy.s0, uyvy.s2, uyvy.s4, uyvy.s6, uyvy.s8, uyvy.sa, uyvy.sc, uyvy.se); |
| char8 cb = (char8)(uyvy.s1, uyvy.s1, uyvy.s5, uyvy.s5, uyvy.s9, uyvy.s9, uyvy.sd, uyvy.sd) - (char8)(128); |
| char8 cr = (char8)(uyvy.s3, uyvy.s3, uyvy.s7, uyvy.s7, uyvy.sb, uyvy.sb, uyvy.sf, uyvy.sf) - (char8)(128); |
| |
| float8 red_coef_bt709 = (float8)(1.5748f); |
| float8 green_coef_bt709 = (float8)(-0.1873f); |
| float8 green_coef2_bt709 = (float8)(-0.4681f); |
| float8 blue_coef_bt709 = (float8)(1.8556f); |
| float8 lumav = convert_float8(luma); |
| |
| float8 f_r = red_coef_bt709 * convert_float8(cr); |
| float8 f_g = green_coef_bt709 * convert_float8(cb) + green_coef2_bt709 * convert_float8(cr); |
| float8 f_b = blue_coef_bt709 * convert_float8(cb); |
| |
| f_r += lumav; |
| f_g += lumav; |
| f_b += lumav; |
| |
| uchar8 r_0 = convert_uchar8_sat_rtz(f_r); |
| uchar8 g_0 = convert_uchar8_sat_rtz(f_g); |
| uchar8 b_0 = convert_uchar8_sat_rtz(f_b); |
| |
| uchar16 rgb_0 = (uchar16)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2, b_0.s2, |
| r_0.s3, g_0.s3, b_0.s3, r_0.s4, g_0.s4, b_0.s4, r_0.s5); |
| uchar8 rgb_1 = (uchar8)(g_0.s5, b_0.s5, r_0.s6, g_0.s6, b_0.s6, r_0.s7, g_0.s7, b_0.s7); |
| |
| vstore16(rgb_0, 0, out.ptr); |
| vstore8(rgb_1, 0, out.ptr + 16); |
| } |
| |
| /** Convert a YUYV422 image to RGBX8888 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 8), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void YUYV422_to_RGBA8888_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out = CONVERT_TO_IMAGE_STRUCT(output); |
| |
| // handle 8 pixels every time |
| uchar16 uyvy = vload16(0, in.ptr); |
| |
| uchar8 luma = (uchar8)(uyvy.s0, uyvy.s2, uyvy.s4, uyvy.s6, uyvy.s8, uyvy.sa, uyvy.sc, uyvy.se); |
| char8 cb = (char8)(uyvy.s1, uyvy.s1, uyvy.s5, uyvy.s5, uyvy.s9, uyvy.s9, uyvy.sd, uyvy.sd) - (char8)(128); |
| char8 cr = (char8)(uyvy.s3, uyvy.s3, uyvy.s7, uyvy.s7, uyvy.sb, uyvy.sb, uyvy.sf, uyvy.sf) - (char8)(128); |
| |
| float8 red_coef_bt709 = (float8)(1.5748f); |
| float8 green_coef_bt709 = (float8)(-0.1873f); |
| float8 green_coef2_bt709 = (float8)(-0.4681f); |
| float8 blue_coef_bt709 = (float8)(1.8556f); |
| float8 lumav = convert_float8(luma); |
| |
| float8 f_r = red_coef_bt709 * convert_float8(cr); |
| float8 f_g = green_coef_bt709 * convert_float8(cb) + green_coef2_bt709 * convert_float8(cr); |
| float8 f_b = blue_coef_bt709 * convert_float8(cb); |
| |
| f_r += lumav; |
| f_g += lumav; |
| f_b += lumav; |
| |
| uchar8 r_0 = convert_uchar8_sat_rtz(f_r); |
| uchar8 g_0 = convert_uchar8_sat_rtz(f_g); |
| uchar8 b_0 = convert_uchar8_sat_rtz(f_b); |
| |
| uchar16 rgba_0 = (uchar16)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255, |
| r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| uchar16 rgba_1 = (uchar16)(r_0.s4, g_0.s4, b_0.s4, 255, r_0.s5, g_0.s5, b_0.s5, 255, |
| r_0.s6, g_0.s6, b_0.s6, 255, r_0.s7, g_0.s7, b_0.s7, 255); |
| |
| vstore16(rgba_0, 0, out.ptr); |
| vstore16(rgba_1, 0, out.ptr + 16); |
| } |
| |
| /** Convert a RGB image to NV12 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_step_x luma_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_step_y luma_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_offset_first_element_in_bytes The offset of the first element in the destination image luma channel |
| * @param[out] uv_ptr Pointer to the destination uv channel. Supported Format: U8 |
| * @param[in] uv_stride_x Stride of the destination uv channel in X dimension (in bytes) |
| * @param[in] uv_step_x uv_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] uv_step_y uv_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_offset_first_element_in_bytes The offset of the first element in the destination image uv channel |
| * |
| */ |
| __kernel void RGB888_to_NV12_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(luma), |
| IMAGE_DECLARATION(uv)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma); |
| Image out_uv = CONVERT_TO_IMAGE_STRUCT(uv); |
| |
| // handle 4 pixels every time, two lines, each line for 2 pixels |
| // Read 2 pixel of the first line |
| uchar8 rgb_0 = vload8(0, in.ptr); |
| uchar2 r_0 = (uchar2)(rgb_0.s0, rgb_0.s3); |
| uchar2 g_0 = (uchar2)(rgb_0.s1, rgb_0.s4); |
| uchar2 b_0 = (uchar2)(rgb_0.s2, rgb_0.s5); |
| |
| float2 f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_0) + (float2)(0.7152f) * convert_float2(g_0) + (float2)(0.0722f) * convert_float2(b_0); |
| float2 f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_0) - (float2)(0.3854f) * convert_float2(g_0) + (float2)(0.5000f) * convert_float2(b_0); |
| float2 f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_0) - (float2)(0.4542f) * convert_float2(g_0) - (float2)(0.0458f) * convert_float2(b_0); |
| |
| short2 i_y = convert_short2_rtz(f_y); |
| short2 i_u = convert_short2_rtz(f_u) + (short2)(128); |
| short2 i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_0 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_0, 0, out_y.ptr); |
| |
| uchar2 cb_0 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_0 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| |
| // Read 2 pixel of the second line |
| uchar8 rgb_1 = vload8(0, in.ptr + input_stride_y); |
| uchar2 r_1 = (uchar2)(rgb_1.s0, rgb_1.s3); |
| uchar2 g_1 = (uchar2)(rgb_1.s1, rgb_1.s4); |
| uchar2 b_1 = (uchar2)(rgb_1.s2, rgb_1.s5); |
| |
| f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_1) + (float2)(0.7152f) * convert_float2(g_1) + (float2)(0.0722f) * convert_float2(b_1); |
| f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_1) - (float2)(0.3854f) * convert_float2(g_1) + (float2)(0.5000f) * convert_float2(b_1); |
| f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_1) - (float2)(0.4542f) * convert_float2(g_1) - (float2)(0.0458f) * convert_float2(b_1); |
| |
| i_y = convert_short2_rtz(f_y); |
| i_u = convert_short2_rtz(f_u) + (short2)(128); |
| i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_1 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_1, 0, out_y.ptr + luma_stride_y); |
| |
| uchar2 cb_1 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_1 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| uchar2 cbcr = (uchar2)(((cb_0.s0 + cb_0.s1 + cb_1.s0 + cb_1.s1) / 4), |
| ((cr_0.s0 + cr_0.s1 + cr_1.s0 + cr_1.s1) / 4)); |
| |
| vstore2(cbcr, 0, out_uv.ptr); |
| } |
| |
| /* |
| R'= Y' + 0.0000*U + 1.5748*V |
| G'= Y' - 0.1873*U - 0.4681*V |
| B'= Y' + 1.8556*U + 0.0000*V |
| */ |
| |
| /** Convert an NV12 image to RGB888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] rgb_output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] rgb_output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] rgb_output_step_x rgb_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgb_output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] rgb_output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgb_output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void NV12_to_RGB888_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(rgb_output)) |
| { |
| Image in_luma = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_rgb = CONVERT_TO_IMAGE_STRUCT(rgb_output); |
| |
| // handle 8 pixels every time, two lines, each line for 4 pixels |
| uchar4 luma_0 = vload4(0, in_luma.ptr); |
| uchar4 luma_1 = vload4(0, in_luma.ptr + luma_input_stride_y); |
| uchar4 cbcr = vload4(0, in_uv.ptr); |
| char4 cb = (char4)(cbcr.s0, cbcr.s0, cbcr.s2, cbcr.s2) - (char4)(128); |
| char4 cr = (char4)(cbcr.s1, cbcr.s1, cbcr.s3, cbcr.s3) - (char4)(128); |
| |
| float4 temp0 = (float4)(0.0000f) + (float4)(0.0000f) * convert_float4(cb) + (float4)(1.5748f) * convert_float4(cr); |
| float4 temp1 = (float4)(0.0000f) - (float4)(0.1873f) * convert_float4(cb) - (float4)(0.4681f) * convert_float4(cr); |
| float4 temp2 = (float4)(0.0000f) + (float4)(1.8556f) * convert_float4(cb) + (float4)(0.0000f) * convert_float4(cr); |
| |
| float4 f_r = convert_float4(luma_0) + temp0; |
| float4 f_g = convert_float4(luma_0) + temp1; |
| float4 f_b = convert_float4(luma_0) + temp2; |
| |
| uchar4 r_0 = convert_uchar4_sat_rtz(f_r); |
| uchar4 g_0 = convert_uchar4_sat_rtz(f_g); |
| uchar4 b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| uchar8 rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2); |
| uchar4 rgb_1 = (uchar4)(b_0.s2, r_0.s3, g_0.s3, b_0.s3); |
| vstore8(rgb_0, 0, out_rgb.ptr); |
| vstore4(rgb_1, 0, out_rgb.ptr + 8); |
| |
| f_r = convert_float4(luma_1) + temp0; |
| f_g = convert_float4(luma_1) + temp1; |
| f_b = convert_float4(luma_1) + temp2; |
| |
| r_0 = convert_uchar4_sat_rtz(f_r); |
| g_0 = convert_uchar4_sat_rtz(f_g); |
| b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2); |
| rgb_1 = (uchar4)(b_0.s2, r_0.s3, g_0.s3, b_0.s3); |
| vstore8(rgb_0, 0, out_rgb.ptr + rgb_output_stride_y); |
| vstore4(rgb_1, 0, out_rgb.ptr + rgb_output_stride_y + 8); |
| } |
| |
| /** Convert a RGB image to YUV444 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] rgb_input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] rgb_input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] rgb_input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgb_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] rgb_input_step_y rgb_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgb_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination image V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| * |
| */ |
| __kernel void RGB888_to_YUV444_bt709( |
| IMAGE_DECLARATION(rgb_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| // handle 4 pixels every time |
| Image in_rgb = CONVERT_TO_IMAGE_STRUCT(rgb_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // Read 4 pixel |
| uchar16 rgb_0 = vload16(0, in_rgb.ptr); |
| uchar4 r_0 = (uchar4)(rgb_0.s0, rgb_0.s3, rgb_0.s6, rgb_0.s9); |
| uchar4 g_0 = (uchar4)(rgb_0.s1, rgb_0.s4, rgb_0.s7, rgb_0.sa); |
| uchar4 b_0 = (uchar4)(rgb_0.s2, rgb_0.s5, rgb_0.s8, rgb_0.sb); |
| |
| float4 f_y = (float4)(0.0000f) + (float4)(0.2126f) * convert_float4(r_0) + (float4)(0.7152f) * convert_float4(g_0) + (float4)(0.0722f) * convert_float4(b_0); |
| float4 f_u = (float4)(0.0000f) - (float4)(0.1146f) * convert_float4(r_0) - (float4)(0.3854f) * convert_float4(g_0) + (float4)(0.5000f) * convert_float4(b_0); |
| float4 f_v = (float4)(0.0000f) + (float4)(0.5000f) * convert_float4(r_0) - (float4)(0.4542f) * convert_float4(g_0) - (float4)(0.0458f) * convert_float4(b_0); |
| |
| short4 i_y = convert_short4_rtz(f_y); |
| short4 i_u = convert_short4_rtz(f_u) + (short4)(128); |
| short4 i_v = convert_short4_rtz(f_v) + (short4)(128); |
| |
| uchar4 luma_0 = convert_uchar4(max((short4)(0), min(i_y, (short4)(255)))); |
| vstore4(luma_0, 0, out_y.ptr); |
| |
| uchar4 cb_0 = convert_uchar4(max((short4)(0), min(i_u, (short4)(255)))); |
| uchar4 cr_0 = convert_uchar4(max((short4)(0), min(i_v, (short4)(255)))); |
| vstore4(cb_0, 0, out_u.ptr); |
| vstore4(cr_0, 0, out_v.ptr); |
| } |
| |
| /** Convert a RGB image to IYUV using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 2), height ] |
| * No offset. |
| * |
| * @param[in] rgb_input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] rgb_input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] rgb_input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgb_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] rgb_input_step_y rgb_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgb_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| * |
| */ |
| __kernel void RGB888_to_IYUV_bt709( |
| IMAGE_DECLARATION(rgb_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| // handle 4 pixels every time, two lines, each line for 2 pixels |
| Image in_rgb = CONVERT_TO_IMAGE_STRUCT(rgb_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // Read 2 pixel of the first line |
| uchar8 rgb_0 = vload8(0, in_rgb.ptr); |
| uchar2 r_0 = (uchar2)(rgb_0.s0, rgb_0.s3); |
| uchar2 g_0 = (uchar2)(rgb_0.s1, rgb_0.s4); |
| uchar2 b_0 = (uchar2)(rgb_0.s2, rgb_0.s5); |
| |
| float2 f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_0) + (float2)(0.7152f) * convert_float2(g_0) + (float2)(0.0722f) * convert_float2(b_0); |
| float2 f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_0) - (float2)(0.3854f) * convert_float2(g_0) + (float2)(0.5000f) * convert_float2(b_0); |
| float2 f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_0) - (float2)(0.4542f) * convert_float2(g_0) - (float2)(0.0458f) * convert_float2(b_0); |
| |
| short2 i_y = convert_short2_rtz(f_y); |
| short2 i_u = convert_short2_rtz(f_u) + (short2)(128); |
| short2 i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_0 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_0, 0, out_y.ptr); |
| |
| uchar2 cb_0 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_0 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| |
| // Read 2 pixel of the second line |
| uchar8 rgb_1 = vload8(0, in_rgb.ptr + rgb_input_stride_y); |
| uchar2 r_1 = (uchar2)(rgb_1.s0, rgb_1.s3); |
| uchar2 g_1 = (uchar2)(rgb_1.s1, rgb_1.s4); |
| uchar2 b_1 = (uchar2)(rgb_1.s2, rgb_1.s5); |
| |
| f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_1) + (float2)(0.7152f) * convert_float2(g_1) + (float2)(0.0722f) * convert_float2(b_1); |
| f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_1) - (float2)(0.3854f) * convert_float2(g_1) + (float2)(0.5000f) * convert_float2(b_1); |
| f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_1) - (float2)(0.4542f) * convert_float2(g_1) - (float2)(0.0458f) * convert_float2(b_1); |
| |
| i_y = convert_short2_rtz(f_y); |
| i_u = convert_short2_rtz(f_u) + (short2)(128); |
| i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_1 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| |
| uchar2 cb_1 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_1 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| uchar2 cbcr = (uchar2)(((cb_0.s0 + cb_0.s1 + cb_1.s0 + cb_1.s1) / 4), |
| ((cr_0.s0 + cr_0.s1 + cr_1.s0 + cr_1.s1) / 4)); |
| *out_u.ptr = cbcr.x; |
| *out_v.ptr = cbcr.y; |
| } |
| |
| /** Convert a RGBA image to YUV444 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] rgba_input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] rgba_input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] rgba_input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgba_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] rgba_input_step_y rgb_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgba_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination image V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| * |
| */ |
| __kernel void RGBA8888_to_YUV444_bt709( |
| IMAGE_DECLARATION(rgba_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| // handle 4 pixels every time |
| Image in_rgba = CONVERT_TO_IMAGE_STRUCT(rgba_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // Read 4 pixel |
| uchar16 rgb_0 = vload16(0, in_rgba.ptr); |
| uchar4 r_0 = (uchar4)(rgb_0.s0, rgb_0.s4, rgb_0.s8, rgb_0.sc); |
| uchar4 g_0 = (uchar4)(rgb_0.s1, rgb_0.s5, rgb_0.s9, rgb_0.sd); |
| uchar4 b_0 = (uchar4)(rgb_0.s2, rgb_0.s6, rgb_0.sa, rgb_0.se); |
| |
| float4 f_y = (float4)(0.0000f) + (float4)(0.2126f) * convert_float4(r_0) + (float4)(0.7152f) * convert_float4(g_0) + (float4)(0.0722f) * convert_float4(b_0); |
| float4 f_u = (float4)(0.0000f) - (float4)(0.1146f) * convert_float4(r_0) - (float4)(0.3854f) * convert_float4(g_0) + (float4)(0.5000f) * convert_float4(b_0); |
| float4 f_v = (float4)(0.0000f) + (float4)(0.5000f) * convert_float4(r_0) - (float4)(0.4542f) * convert_float4(g_0) - (float4)(0.0458f) * convert_float4(b_0); |
| |
| short4 i_y = convert_short4(f_y); |
| short4 i_u = convert_short4(f_u) + (short4)(128); |
| short4 i_v = convert_short4(f_v) + (short4)(128); |
| |
| uchar4 luma_0 = convert_uchar4_sat(max((short4)(0), min(i_y, (short4)(255)))); |
| vstore4(luma_0, 0, out_y.ptr); |
| |
| uchar4 cb_0 = convert_uchar4_sat(max((short4)(0), min(i_u, (short4)(255)))); |
| uchar4 cr_0 = convert_uchar4_sat(max((short4)(0), min(i_v, (short4)(255)))); |
| vstore4(cb_0, 0, out_u.ptr); |
| vstore4(cr_0, 0, out_v.ptr); |
| } |
| |
| /** Convert a RGBA image to NV12 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 2), height ] |
| * No offset. |
| * |
| * @param[in] input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination image luma channel |
| * @param[out] uv_output_ptr Pointer to the destination uv channel. Supported Format: U8 |
| * @param[in] uv_output_stride_x Stride of the destination uv channel in X dimension (in bytes) |
| * @param[in] uv_output_step_x uv_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_output_stride_y Stride of the destination image uv channel in Y dimension (in bytes) |
| * @param[in] uv_output_step_y uv_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_output_offset_first_element_in_bytes The offset of the first element in the destination image uv channel |
| * |
| */ |
| __kernel void RGBA8888_to_NV12_bt709( |
| IMAGE_DECLARATION(input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(uv_output)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_uv = CONVERT_TO_IMAGE_STRUCT(uv_output); |
| |
| // Read 2 pixel of the first line |
| uchar8 rgb_0 = vload8(0, in.ptr); |
| uchar2 r_0 = (uchar2)(rgb_0.s0, rgb_0.s4); |
| uchar2 g_0 = (uchar2)(rgb_0.s1, rgb_0.s5); |
| uchar2 b_0 = (uchar2)(rgb_0.s2, rgb_0.s6); |
| |
| float2 f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_0) + (float2)(0.7152f) * convert_float2(g_0) + (float2)(0.0722f) * convert_float2(b_0); |
| float2 f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_0) - (float2)(0.3854f) * convert_float2(g_0) + (float2)(0.5000f) * convert_float2(b_0); |
| float2 f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_0) - (float2)(0.4542f) * convert_float2(g_0) - (float2)(0.0458f) * convert_float2(b_0); |
| |
| short2 i_y = convert_short2_rtz(f_y); |
| short2 i_u = convert_short2_rtz(f_u) + (short2)(128); |
| short2 i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_0 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_0, 0, out_y.ptr); |
| |
| uchar2 cb_0 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_0 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| |
| // Read 2 pixel of the second line |
| uchar8 rgb_1 = vload8(0, in.ptr + input_stride_y); |
| uchar2 r_1 = (uchar2)(rgb_1.s0, rgb_1.s4); |
| uchar2 g_1 = (uchar2)(rgb_1.s1, rgb_1.s5); |
| uchar2 b_1 = (uchar2)(rgb_1.s2, rgb_1.s6); |
| |
| f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_1) + (float2)(0.7152f) * convert_float2(g_1) + (float2)(0.0722f) * convert_float2(b_1); |
| f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_1) - (float2)(0.3854f) * convert_float2(g_1) + (float2)(0.5000f) * convert_float2(b_1); |
| f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_1) - (float2)(0.4542f) * convert_float2(g_1) - (float2)(0.0458f) * convert_float2(b_1); |
| |
| i_y = convert_short2_rtz(f_y); |
| i_u = convert_short2_rtz(f_u) + (short2)(128); |
| i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_1 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| |
| uchar2 cb_1 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_1 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| uchar2 cbcr = (uchar2)(((cb_0.s0 + cb_0.s1 + cb_1.s0 + cb_1.s1) / 4), |
| ((cr_0.s0 + cr_0.s1 + cr_1.s0 + cr_1.s1) / 4)); |
| vstore2(cbcr, 0, out_uv.ptr); |
| } |
| |
| /** Convert a RGBA image to IYUV using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 2), height ] |
| * No offset. |
| * |
| * @param[in] rgba_input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] rgba_input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] rgba_input_step_x rgba_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgba_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] rgba_input_step_y rgba_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgba_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| * |
| */ |
| __kernel void RGBA8888_to_IYUV_bt709( |
| IMAGE_DECLARATION(rgba_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| // handle 4 pixels every time, two lines, each line for 2 pixels |
| Image in_rgb = CONVERT_TO_IMAGE_STRUCT(rgba_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // Read 2 pixel of the first line |
| uchar8 rgb_0 = vload8(0, in_rgb.ptr); |
| uchar2 r_0 = (uchar2)(rgb_0.s0, rgb_0.s4); |
| uchar2 g_0 = (uchar2)(rgb_0.s1, rgb_0.s5); |
| uchar2 b_0 = (uchar2)(rgb_0.s2, rgb_0.s6); |
| |
| float2 f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_0) + (float2)(0.7152f) * convert_float2(g_0) + (float2)(0.0722f) * convert_float2(b_0); |
| float2 f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_0) - (float2)(0.3854f) * convert_float2(g_0) + (float2)(0.5000f) * convert_float2(b_0); |
| float2 f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_0) - (float2)(0.4542f) * convert_float2(g_0) - (float2)(0.0458f) * convert_float2(b_0); |
| |
| short2 i_y = convert_short2_rtz(f_y); |
| short2 i_u = convert_short2_rtz(f_u) + (short2)(128); |
| short2 i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_0 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_0, 0, out_y.ptr); |
| |
| uchar2 cb_0 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_0 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| |
| // Read 2 pixel of the second line |
| uchar8 rgb_1 = vload8(0, in_rgb.ptr + rgba_input_stride_y); |
| uchar2 r_1 = (uchar2)(rgb_1.s0, rgb_1.s4); |
| uchar2 g_1 = (uchar2)(rgb_1.s1, rgb_1.s5); |
| uchar2 b_1 = (uchar2)(rgb_1.s2, rgb_1.s6); |
| |
| f_y = (float2)(0.0000f) + (float2)(0.2126f) * convert_float2(r_1) + (float2)(0.7152f) * convert_float2(g_1) + (float2)(0.0722f) * convert_float2(b_1); |
| f_u = (float2)(0.0000f) - (float2)(0.1146f) * convert_float2(r_1) - (float2)(0.3854f) * convert_float2(g_1) + (float2)(0.5000f) * convert_float2(b_1); |
| f_v = (float2)(0.0000f) + (float2)(0.5000f) * convert_float2(r_1) - (float2)(0.4542f) * convert_float2(g_1) - (float2)(0.0458f) * convert_float2(b_1); |
| |
| i_y = convert_short2_rtz(f_y); |
| i_u = convert_short2_rtz(f_u) + (short2)(128); |
| i_v = convert_short2_rtz(f_v) + (short2)(128); |
| |
| uchar2 luma_1 = convert_uchar2(max((short2)(0), min(i_y, (short2)(255)))); |
| vstore2(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| |
| uchar2 cb_1 = convert_uchar2(max((short2)(0), min(i_u, (short2)(255)))); |
| uchar2 cr_1 = convert_uchar2(max((short2)(0), min(i_v, (short2)(255)))); |
| uchar2 cbcr = (uchar2)(((cb_0.s0 + cb_0.s1 + cb_1.s0 + cb_1.s1) / 4), |
| ((cr_0.s0 + cr_0.s1 + cr_1.s0 + cr_1.s1) / 4)); |
| *out_u.ptr = cbcr.x; |
| *out_v.ptr = cbcr.y; |
| } |
| |
| /** Convert an NV12 image to RGB8888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] rgb_output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] rgb_output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] rgb_output_step_x rgb_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgb_output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] rgb_output_step_y rgb_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgb_output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void NV12_to_RGBA8888_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(rgb_output)) |
| { |
| Image in_luma = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_rgb = CONVERT_TO_IMAGE_STRUCT(rgb_output); |
| |
| uchar4 luma_0 = vload4(0, in_luma.ptr); |
| uchar4 luma_1 = vload4(0, in_luma.ptr + luma_input_stride_y); |
| uchar4 cbcr = vload4(0, in_uv.ptr); |
| char4 cb = (char4)(cbcr.s0, cbcr.s0, cbcr.s2, cbcr.s2) - (char4)(128); |
| char4 cr = (char4)(cbcr.s1, cbcr.s1, cbcr.s3, cbcr.s3) - (char4)(128); |
| |
| float4 temp0 = (float4)(0.0000f) + (float4)(0.0000f) * convert_float4(cb) + (float4)(1.5748f) * convert_float4(cr); |
| float4 temp1 = (float4)(0.0000f) - (float4)(0.1873f) * convert_float4(cb) - (float4)(0.4681f) * convert_float4(cr); |
| float4 temp2 = (float4)(0.0000f) + (float4)(1.8556f) * convert_float4(cb) + (float4)(0.0000f) * convert_float4(cr); |
| |
| float4 f_r = convert_float4(luma_0) + temp0; |
| float4 f_g = convert_float4(luma_0) + temp1; |
| float4 f_b = convert_float4(luma_0) + temp2; |
| |
| uchar4 r_0 = convert_uchar4_sat_rtz(f_r); |
| uchar4 g_0 = convert_uchar4_sat_rtz(f_g); |
| uchar4 b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| uchar8 rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255); |
| uchar8 rgb_1 = (uchar8)(r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| vstore8(rgb_0, 0, out_rgb.ptr); |
| vstore8(rgb_1, 0, out_rgb.ptr + 8); |
| |
| f_r = convert_float4(luma_1) + temp0; |
| f_g = convert_float4(luma_1) + temp1; |
| f_b = convert_float4(luma_1) + temp2; |
| |
| r_0 = convert_uchar4_sat_rtz(f_r); |
| g_0 = convert_uchar4_sat_rtz(f_g); |
| b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255); |
| rgb_1 = (uchar8)(r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| vstore8(rgb_0, 0, out_rgb.ptr + rgb_output_stride_y); |
| vstore8(rgb_1, 0, out_rgb.ptr + rgb_output_stride_y + 8); |
| } |
| |
| /** Convert an NV12 image to IYUV |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| */ |
| __kernel void NV12_to_IYUV_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // handle 32 pixels every time, two lines, each line for 16 pixels |
| uchar16 luma_0 = vload16(0, in_y.ptr); |
| uchar16 luma_1 = vload16(0, in_y.ptr + luma_input_stride_y); |
| uchar16 cbcr = vload16(0, in_uv.ptr); |
| uchar8 cb = (uchar8)(cbcr.s0, cbcr.s2, cbcr.s4, cbcr.s6, cbcr.s8, cbcr.sa, cbcr.sc, cbcr.se); |
| uchar8 cr = (uchar8)(cbcr.s1, cbcr.s3, cbcr.s5, cbcr.s7, cbcr.s9, cbcr.sb, cbcr.sd, cbcr.sf); |
| |
| vstore16(luma_0, 0, out_y.ptr); |
| vstore16(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| vstore8(cb, 0, out_u.ptr); |
| vstore8(cr, 0, out_v.ptr); |
| } |
| |
| /** Convert an NV12 image to YUV444 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| */ |
| __kernel void NV12_to_YUV444_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // handle 32 pixels every time, two lines, each line for 16 pixels |
| uchar16 luma_0 = vload16(0, in_y.ptr); |
| uchar16 luma_1 = vload16(0, in_y.ptr + luma_input_stride_y); |
| uchar16 cbcr = vload16(0, in_uv.ptr); |
| uchar16 cb = (uchar16)(cbcr.s0, cbcr.s0, cbcr.s2, cbcr.s2, cbcr.s4, cbcr.s4, cbcr.s6, cbcr.s6, cbcr.s8, cbcr.s8, |
| cbcr.sa, cbcr.sa, cbcr.sc, cbcr.sc, cbcr.se, cbcr.se); |
| uchar16 cr = (uchar16)(cbcr.s1, cbcr.s1, cbcr.s3, cbcr.s3, cbcr.s5, cbcr.s5, cbcr.s7, cbcr.s7, cbcr.s9, cbcr.s9, |
| cbcr.sb, cbcr.sb, cbcr.sd, cbcr.sd, cbcr.sf, cbcr.sf); |
| |
| vstore16(luma_0, 0, out_y.ptr); |
| vstore16(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| vstore16(cb, 0, out_u.ptr); |
| vstore16(cb, 0, out_u.ptr + u_output_stride_y); |
| vstore16(cr, 0, out_v.ptr); |
| vstore16(cr, 0, out_v.ptr + v_output_stride_y); |
| } |
| |
| /** Convert an NV21 image to RGB888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] rgb_output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] rgb_output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] rgb_output_step_x rgb_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgb_output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] rgb_output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgb_output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void NV21_to_RGB888_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(rgb_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_rgb = CONVERT_TO_IMAGE_STRUCT(rgb_output); |
| |
| // handle 8 pixels every time, two lines, each line for 4 pixels |
| uchar4 luma_0 = vload4(0, in_y.ptr); |
| uchar4 luma_1 = vload4(0, in_y.ptr + luma_input_stride_y); |
| uchar4 cbcr = vload4(0, in_uv.ptr); |
| char4 cr = (char4)(cbcr.s0, cbcr.s0, cbcr.s2, cbcr.s2) - (char4)(128); |
| char4 cb = (char4)(cbcr.s1, cbcr.s1, cbcr.s3, cbcr.s3) - (char4)(128); |
| |
| float4 temp0 = (float4)(0.0000f) + (float4)(0.0000f) * convert_float4(cb) + (float4)(1.5748f) * convert_float4(cr); |
| float4 temp1 = (float4)(0.0000f) - (float4)(0.1873f) * convert_float4(cb) - (float4)(0.4681f) * convert_float4(cr); |
| float4 temp2 = (float4)(0.0000f) + (float4)(1.8556f) * convert_float4(cb) + (float4)(0.0000f) * convert_float4(cr); |
| |
| float4 f_r = convert_float4(luma_0) + temp0; |
| float4 f_g = convert_float4(luma_0) + temp1; |
| float4 f_b = convert_float4(luma_0) + temp2; |
| |
| uchar4 r_0 = convert_uchar4_sat_rtz(f_r); |
| uchar4 g_0 = convert_uchar4_sat_rtz(f_g); |
| uchar4 b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| uchar8 rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2); |
| uchar4 rgb_1 = (uchar4)(b_0.s2, r_0.s3, g_0.s3, b_0.s3); |
| vstore8(rgb_0, 0, out_rgb.ptr); |
| vstore4(rgb_1, 0, out_rgb.ptr + 8); |
| |
| f_r = convert_float4(luma_1) + temp0; |
| f_g = convert_float4(luma_1) + temp1; |
| f_b = convert_float4(luma_1) + temp2; |
| |
| r_0 = convert_uchar4_sat_rtz(f_r); |
| g_0 = convert_uchar4_sat_rtz(f_g); |
| b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2); |
| rgb_1 = (uchar4)(b_0.s2, r_0.s3, g_0.s3, b_0.s3); |
| vstore8(rgb_0, 0, out_rgb.ptr + rgb_output_stride_y); |
| vstore4(rgb_1, 0, out_rgb.ptr + rgb_output_stride_y + 8); |
| } |
| |
| /** Convert an NV12 image to RGB8888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] rgba_output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] rgba_output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] rgba_output_step_x rgba_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgba_output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] rgba_output_step_y rgba_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgba_output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void NV21_to_RGBA8888_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(rgba_output)) |
| { |
| Image in_luma = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_rgb = CONVERT_TO_IMAGE_STRUCT(rgba_output); |
| |
| // handle 8 pixels every time, two lines, each line for 4 pixels |
| uchar4 luma_0 = vload4(0, in_luma.ptr); |
| uchar4 luma_1 = vload4(0, in_luma.ptr + luma_input_stride_y); |
| uchar4 cbcr = vload4(0, in_uv.ptr); |
| char4 cr = (char4)(cbcr.s0, cbcr.s0, cbcr.s2, cbcr.s2) - (char4)(128); |
| char4 cb = (char4)(cbcr.s1, cbcr.s1, cbcr.s3, cbcr.s3) - (char4)(128); |
| |
| float4 temp0 = (float4)(0.0000f) + (float4)(0.0000f) * convert_float4(cb) + (float4)(1.5748f) * convert_float4(cr); |
| float4 temp1 = (float4)(0.0000f) - (float4)(0.1873f) * convert_float4(cb) - (float4)(0.4681f) * convert_float4(cr); |
| float4 temp2 = (float4)(0.0000f) + (float4)(1.8556f) * convert_float4(cb) + (float4)(0.0000f) * convert_float4(cr); |
| |
| float4 f_r = convert_float4(luma_0) + temp0; |
| float4 f_g = convert_float4(luma_0) + temp1; |
| float4 f_b = convert_float4(luma_0) + temp2; |
| |
| uchar4 r_0 = convert_uchar4_sat_rtz(f_r); |
| uchar4 g_0 = convert_uchar4_sat_rtz(f_g); |
| uchar4 b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| uchar8 rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255); |
| uchar8 rgb_1 = (uchar8)(r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| vstore8(rgb_0, 0, out_rgb.ptr); |
| vstore8(rgb_1, 0, out_rgb.ptr + 8); |
| |
| f_r = convert_float4(luma_1) + temp0; |
| f_g = convert_float4(luma_1) + temp1; |
| f_b = convert_float4(luma_1) + temp2; |
| |
| r_0 = convert_uchar4_sat_rtz(f_r); |
| g_0 = convert_uchar4_sat_rtz(f_g); |
| b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255); |
| rgb_1 = (uchar8)(r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| vstore8(rgb_0, 0, out_rgb.ptr + rgba_output_stride_y); |
| vstore8(rgb_1, 0, out_rgb.ptr + rgba_output_stride_y + 8); |
| } |
| |
| /** Convert an NV21 image to YUV444 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| */ |
| __kernel void NV21_to_YUV444_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // handle 32 pixels every time, two lines, each line for 16 pixels |
| uchar16 luma_0 = vload16(0, in_y.ptr); |
| uchar16 luma_1 = vload16(0, in_y.ptr + luma_input_stride_y); |
| uchar16 cbcr = vload16(0, in_uv.ptr); |
| uchar16 cr = (uchar16)(cbcr.s0, cbcr.s0, cbcr.s2, cbcr.s2, cbcr.s4, cbcr.s4, cbcr.s6, cbcr.s6, cbcr.s8, cbcr.s8, |
| cbcr.sa, cbcr.sa, cbcr.sc, cbcr.sc, cbcr.se, cbcr.se); |
| uchar16 cb = (uchar16)(cbcr.s1, cbcr.s1, cbcr.s3, cbcr.s3, cbcr.s5, cbcr.s5, cbcr.s7, cbcr.s7, cbcr.s9, cbcr.s9, |
| cbcr.sb, cbcr.sb, cbcr.sd, cbcr.sd, cbcr.sf, cbcr.sf); |
| |
| vstore16(luma_0, 0, out_y.ptr); |
| vstore16(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| vstore16(cb, 0, out_u.ptr); |
| vstore16(cb, 0, out_u.ptr + u_output_stride_y); |
| vstore16(cr, 0, out_v.ptr); |
| vstore16(cr, 0, out_v.ptr + v_output_stride_y); |
| } |
| |
| /** Convert an NV21 image to IYUV |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] uv_input_ptr Pointer to the source uv channel. Supported Format: U8 |
| * @param[in] uv_input_stride_x Stride of the source image uv channel in X dimension (in bytes) |
| * @param[in] uv_input_step_x uv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uv_input_step_y uv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| */ |
| __kernel void NV21_to_IYUV_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(uv_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_uv = CONVERT_TO_IMAGE_STRUCT(uv_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| uchar16 luma_0 = vload16(0, in_y.ptr); |
| uchar16 luma_1 = vload16(0, in_y.ptr + luma_input_stride_y); |
| uchar16 cbcr = vload16(0, in_uv.ptr); |
| uchar8 cr = (uchar8)(cbcr.s0, cbcr.s2, cbcr.s4, cbcr.s6, cbcr.s8, cbcr.sa, cbcr.sc, cbcr.se); |
| uchar8 cb = (uchar8)(cbcr.s1, cbcr.s3, cbcr.s5, cbcr.s7, cbcr.s9, cbcr.sb, cbcr.sd, cbcr.sf); |
| |
| vstore16(luma_0, 0, out_y.ptr); |
| vstore16(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| vstore8(cb, 0, out_u.ptr); |
| vstore8(cr, 0, out_v.ptr); |
| } |
| |
| /** Convert a UYVY image to IYUV using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 8), height ] |
| * No offset. |
| * |
| * @param[in] uyvy_input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] uyvy_input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] uyvy_input_step_x uyvy_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uyvy_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] uyvy_input_step_y uyvy_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uyvy_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| * |
| */ |
| __kernel void UYVY422_to_IYUV_bt709( |
| IMAGE_DECLARATION(uyvy_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| Image in_uyvy = CONVERT_TO_IMAGE_STRUCT(uyvy_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // handle 16 pixels every time, each line 8 pixels |
| uchar16 uyvy = vload16(0, in_uyvy.ptr); |
| uchar8 luma = (uchar8)(uyvy.s1, uyvy.s3, uyvy.s5, uyvy.s7, uyvy.s9, uyvy.sb, uyvy.sd, uyvy.sf); |
| ushort4 cb_0 = (ushort4)(uyvy.s0, uyvy.s4, uyvy.s8, uyvy.sc); |
| ushort4 cr_0 = (ushort4)(uyvy.s2, uyvy.s6, uyvy.sa, uyvy.se); |
| vstore8(luma, 0, out_y.ptr); |
| |
| uyvy = vload16(0, in_uyvy.ptr + uyvy_input_stride_y); |
| luma = (uchar8)(uyvy.s1, uyvy.s3, uyvy.s5, uyvy.s7, uyvy.s9, uyvy.sb, uyvy.sd, uyvy.sf); |
| ushort4 cb_1 = (ushort4)(uyvy.s0, uyvy.s4, uyvy.s8, uyvy.sc); |
| ushort4 cr_1 = (ushort4)(uyvy.s2, uyvy.s6, uyvy.sa, uyvy.se); |
| vstore8(luma, 0, out_y.ptr + luma_output_stride_y); |
| |
| uchar4 cb = convert_uchar4((cb_0 + cb_1) / (ushort4)(2)); |
| uchar4 cr = convert_uchar4((cr_0 + cr_1) / (ushort4)(2)); |
| vstore4(cb, 0, out_u.ptr); |
| vstore4(cr, 0, out_v.ptr); |
| } |
| |
| /** Convert a YUYV image to IYUV using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 8), height ] |
| * No offset. |
| * |
| * @param[in] yuyv_input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] yuyv_input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] yuyv_input_step_x yuyv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] yuyv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] yuyv_input_step_y yuyv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] yuyv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| * |
| */ |
| __kernel void YUYV422_to_IYUV_bt709( |
| IMAGE_DECLARATION(yuyv_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| Image in_yuyv = CONVERT_TO_IMAGE_STRUCT(yuyv_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // handle 16 pixels every time, each line 8 pixels |
| uchar16 yuyv = vload16(0, in_yuyv.ptr); |
| uchar8 luma = (uchar8)(yuyv.s0, yuyv.s2, yuyv.s4, yuyv.s6, yuyv.s8, yuyv.sa, yuyv.sc, yuyv.se); |
| ushort4 cb_0 = (ushort4)(yuyv.s1, yuyv.s5, yuyv.s9, yuyv.sd); |
| ushort4 cr_0 = (ushort4)(yuyv.s3, yuyv.s7, yuyv.sb, yuyv.sf); |
| vstore8(luma, 0, out_y.ptr); |
| |
| yuyv = vload16(0, in_yuyv.ptr + yuyv_input_stride_y); |
| luma = (uchar8)(yuyv.s0, yuyv.s2, yuyv.s4, yuyv.s6, yuyv.s8, yuyv.sa, yuyv.sc, yuyv.se); |
| ushort4 cb_1 = (ushort4)(yuyv.s1, yuyv.s5, yuyv.s9, yuyv.sd); |
| ushort4 cr_1 = (ushort4)(yuyv.s3, yuyv.s7, yuyv.sb, yuyv.sf); |
| vstore8(luma, 0, out_y.ptr + luma_output_stride_y); |
| |
| uchar4 cb = convert_uchar4((cb_0 + cb_1) / (ushort4)(2)); |
| uchar4 cr = convert_uchar4((cr_0 + cr_1) / (ushort4)(2)); |
| vstore4(cb, 0, out_u.ptr); |
| vstore4(cr, 0, out_v.ptr); |
| } |
| |
| /** Convert an IYUV image to RGB888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] u_input_ptr Pointer to the source U channel. Supported Format: U8 |
| * @param[in] u_input_stride_x Stride of the source image U channel in X dimension (in bytes) |
| * @param[in] u_input_step_x u_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] u_input_step_y u_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_input_offset_first_element_in_bytes The offset of the first element in the source U channel |
| * @param[in] v_input_ptr Pointer to the source V channel. Supported Format: U8 |
| * @param[in] v_input_stride_x Stride of the source image V channel in X dimension (in bytes) |
| * @param[in] v_input_step_x v_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_input_stride_y Stride of the source image V channel in Y dimension (in bytes) |
| * @param[in] v_input_step_y v_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_input_offset_first_element_in_bytes The offset of the first element in the source image V channel |
| * @param[out] rgb_output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] rgb_output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] rgb_output_step_x rgb_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgb_output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] rgb_output_step_y rgb_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgb_output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void IYUV_to_RGB888_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(u_input), |
| IMAGE_DECLARATION(v_input), |
| IMAGE_DECLARATION(rgb_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_u = CONVERT_TO_IMAGE_STRUCT(u_input); |
| Image in_v = CONVERT_TO_IMAGE_STRUCT(v_input); |
| Image out_rgb = CONVERT_TO_IMAGE_STRUCT(rgb_output); |
| |
| // handle 8 pixels every time, two lines, each line for 4 pixels |
| uchar4 luma_0 = vload4(0, in_y.ptr); |
| uchar4 luma_1 = vload4(0, in_y.ptr + luma_input_stride_y); |
| uchar4 cbcr = (uchar4)(vload2(0, in_u.ptr), vload2(0, in_v.ptr)); |
| char4 cb = (char4)(cbcr.s0, cbcr.s0, cbcr.s1, cbcr.s1) - (char4)(128); |
| char4 cr = (char4)(cbcr.s2, cbcr.s2, cbcr.s3, cbcr.s3) - (char4)(128); |
| |
| float4 temp0 = (float4)(0.0000f) + (float4)(0.0000f) * convert_float4(cb) + (float4)(1.5748f) * convert_float4(cr); |
| float4 temp1 = (float4)(0.0000f) - (float4)(0.1873f) * convert_float4(cb) - (float4)(0.4681f) * convert_float4(cr); |
| float4 temp2 = (float4)(0.0000f) + (float4)(1.8556f) * convert_float4(cb) + (float4)(0.0000f) * convert_float4(cr); |
| |
| float4 f_r = convert_float4(luma_0) + temp0; |
| float4 f_g = convert_float4(luma_0) + temp1; |
| float4 f_b = convert_float4(luma_0) + temp2; |
| |
| uchar4 r_0 = convert_uchar4_sat_rtz(f_r); |
| uchar4 g_0 = convert_uchar4_sat_rtz(f_g); |
| uchar4 b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| uchar8 rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2); |
| uchar4 rgb_1 = (uchar4)(b_0.s2, r_0.s3, g_0.s3, b_0.s3); |
| vstore8(rgb_0, 0, out_rgb.ptr); |
| vstore4(rgb_1, 0, out_rgb.ptr + 8); |
| |
| f_r = convert_float4(luma_1) + temp0; |
| f_g = convert_float4(luma_1) + temp1; |
| f_b = convert_float4(luma_1) + temp2; |
| |
| r_0 = convert_uchar4_sat_rtz(f_r); |
| g_0 = convert_uchar4_sat_rtz(f_g); |
| b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, r_0.s1, g_0.s1, b_0.s1, r_0.s2, g_0.s2); |
| rgb_1 = (uchar4)(b_0.s2, r_0.s3, g_0.s3, b_0.s3); |
| vstore8(rgb_0, 0, out_rgb.ptr + rgb_output_stride_y); |
| vstore4(rgb_1, 0, out_rgb.ptr + rgb_output_stride_y + 8); |
| } |
| |
| /** Convert an IYUV image to RGB8888 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] u_input_ptr Pointer to the source U channel. Supported Format: U8 |
| * @param[in] u_input_stride_x Stride of the source image U channel in X dimension (in bytes) |
| * @param[in] u_input_step_x u_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] u_input_step_y u_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_input_offset_first_element_in_bytes The offset of the first element in the source U channel |
| * @param[in] v_input_ptr Pointer to the source V channel. Supported Format: U8 |
| * @param[in] v_input_stride_x Stride of the source image V channel in X dimension (in bytes) |
| * @param[in] v_input_step_x v_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_input_stride_y Stride of the source image V channel in Y dimension (in bytes) |
| * @param[in] v_input_step_y v_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_input_offset_first_element_in_bytes The offset of the first element in the source image V channel |
| * @param[out] rgba_output_ptr Pointer to the destination image. Supported Format: U8 |
| * @param[in] rgba_output_stride_x Stride of the destination image in X dimension (in bytes) |
| * @param[in] rgba_output_step_x rgba_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] rgba_output_stride_y Stride of the destination image in Y dimension (in bytes) |
| * @param[in] rgba_output_step_y rgba_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] rgba_output_offset_first_element_in_bytes The offset of the first element in the destination image |
| */ |
| __kernel void IYUV_to_RGBA8888_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(u_input), |
| IMAGE_DECLARATION(v_input), |
| IMAGE_DECLARATION(rgba_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_u = CONVERT_TO_IMAGE_STRUCT(u_input); |
| Image in_v = CONVERT_TO_IMAGE_STRUCT(v_input); |
| Image out_rgb = CONVERT_TO_IMAGE_STRUCT(rgba_output); |
| |
| // handle 8 pixels every time, two lines, each line for 4 pixels |
| uchar4 luma_0 = vload4(0, in_y.ptr); |
| uchar4 luma_1 = vload4(0, in_y.ptr + luma_input_stride_y); |
| uchar4 cbcr = (uchar4)(vload2(0, in_u.ptr), vload2(0, in_v.ptr)); |
| char4 cb = (char4)(cbcr.s0, cbcr.s0, cbcr.s1, cbcr.s1) - (char4)(128); |
| char4 cr = (char4)(cbcr.s2, cbcr.s2, cbcr.s3, cbcr.s3) - (char4)(128); |
| |
| float4 temp0 = (float4)(0.0000f) + (float4)(0.0000f) * convert_float4(cb) + (float4)(1.5748f) * convert_float4(cr); |
| float4 temp1 = (float4)(0.0000f) - (float4)(0.1873f) * convert_float4(cb) - (float4)(0.4681f) * convert_float4(cr); |
| float4 temp2 = (float4)(0.0000f) + (float4)(1.8556f) * convert_float4(cb) + (float4)(0.0000f) * convert_float4(cr); |
| |
| float4 f_r = convert_float4(luma_0) + temp0; |
| float4 f_g = convert_float4(luma_0) + temp1; |
| float4 f_b = convert_float4(luma_0) + temp2; |
| |
| uchar4 r_0 = convert_uchar4_sat_rtz(f_r); |
| uchar4 g_0 = convert_uchar4_sat_rtz(f_g); |
| uchar4 b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| uchar8 rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255); |
| uchar8 rgb_1 = (uchar8)(r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| vstore8(rgb_0, 0, out_rgb.ptr); |
| vstore8(rgb_1, 0, out_rgb.ptr + 8); |
| |
| f_r = convert_float4(luma_1) + temp0; |
| f_g = convert_float4(luma_1) + temp1; |
| f_b = convert_float4(luma_1) + temp2; |
| |
| r_0 = convert_uchar4_sat_rtz(f_r); |
| g_0 = convert_uchar4_sat_rtz(f_g); |
| b_0 = convert_uchar4_sat_rtz(f_b); |
| |
| rgb_0 = (uchar8)(r_0.s0, g_0.s0, b_0.s0, 255, r_0.s1, g_0.s1, b_0.s1, 255); |
| rgb_1 = (uchar8)(r_0.s2, g_0.s2, b_0.s2, 255, r_0.s3, g_0.s3, b_0.s3, 255); |
| vstore8(rgb_0, 0, out_rgb.ptr + rgba_output_stride_y); |
| vstore8(rgb_1, 0, out_rgb.ptr + rgba_output_stride_y + 8); |
| } |
| |
| /** Convert an IYUV image to YUV444 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] u_input_ptr Pointer to the source U channel. Supported Format: U8 |
| * @param[in] u_input_stride_x Stride of the source image U channel in X dimension (in bytes) |
| * @param[in] u_input_step_x u_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] u_input_step_y u_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_input_offset_first_element_in_bytes The offset of the first element in the source U channel |
| * @param[in] v_input_ptr Pointer to the source V channel. Supported Format: U8 |
| * @param[in] v_input_stride_x Stride of the source image V channel in X dimension (in bytes) |
| * @param[in] v_input_step_x v_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_input_stride_y Stride of the source image V channel in Y dimension (in bytes) |
| * @param[in] v_input_step_y v_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_input_offset_first_element_in_bytes The offset of the first element in the source image V channel |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] u_output_ptr Pointer to the destination U channel. Supported Format: U8 |
| * @param[in] u_output_stride_x Stride of the destination U channel in X dimension (in bytes) |
| * @param[in] u_output_step_x u_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] u_output_step_y u_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_output_offset_first_element_in_bytes The offset of the first element in the destination U channel |
| * @param[out] v_output_ptr Pointer to the destination V channel. Supported Format: U8 |
| * @param[in] v_output_stride_x Stride of the destination V channel in X dimension (in bytes) |
| * @param[in] v_output_step_x v_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_output_stride_y Stride of the destination V channel in Y dimension (in bytes) |
| * @param[in] v_output_step_y v_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_output_offset_first_element_in_bytes The offset of the first element in the destination V channel |
| * |
| */ |
| __kernel void IYUV_to_YUV444_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(u_input), |
| IMAGE_DECLARATION(v_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(u_output), |
| IMAGE_DECLARATION(v_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_u = CONVERT_TO_IMAGE_STRUCT(u_input); |
| Image in_v = CONVERT_TO_IMAGE_STRUCT(v_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_u = CONVERT_TO_IMAGE_STRUCT(u_output); |
| Image out_v = CONVERT_TO_IMAGE_STRUCT(v_output); |
| |
| // handle 32 pixels every time, two lines, each line for 16 pixels |
| uchar16 luma_0 = vload16(0, in_y.ptr); |
| uchar16 luma_1 = vload16(0, in_y.ptr + luma_input_stride_y); |
| uchar8 cb_src = vload8(0, in_u.ptr); |
| uchar8 cr_src = vload8(0, in_v.ptr); |
| uchar16 cb = (uchar16)(cb_src.s0, cb_src.s0, cb_src.s1, cb_src.s1, cb_src.s2, cb_src.s2, cb_src.s3, cb_src.s3, |
| cb_src.s4, cb_src.s4, cb_src.s5, cb_src.s5, cb_src.s6, cb_src.s6, cb_src.s7, cb_src.s7); |
| uchar16 cr = (uchar16)(cr_src.s0, cr_src.s0, cr_src.s1, cr_src.s1, cr_src.s2, cr_src.s2, cr_src.s3, cr_src.s3, |
| cr_src.s4, cr_src.s4, cr_src.s5, cr_src.s5, cr_src.s6, cr_src.s6, cr_src.s7, cr_src.s7); |
| |
| vstore16(luma_0, 0, out_y.ptr); |
| vstore16(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| vstore16(cb, 0, out_u.ptr); |
| vstore16(cb, 0, out_u.ptr + u_output_stride_y); |
| vstore16(cr, 0, out_v.ptr); |
| vstore16(cr, 0, out_v.ptr + v_output_stride_y); |
| } |
| |
| /** Convert an IYUV image to NV12 |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 16), height ] |
| * No offset. |
| * |
| * @param[in] luma_input_ptr Pointer to the source luma channel. Supported Format: U8 |
| * @param[in] luma_input_stride_x Stride of the luma image in X dimension (in bytes) |
| * @param[in] luma_input_step_x luma_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_input_stride_y Stride of the source luma channel in Y dimension (in bytes) |
| * @param[in] luma_input_step_y luma_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[in] u_input_ptr Pointer to the source U channel. Supported Format: U8 |
| * @param[in] u_input_stride_x Stride of the source image U channel in X dimension (in bytes) |
| * @param[in] u_input_step_x u_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] u_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] u_input_step_y u_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] u_input_offset_first_element_in_bytes The offset of the first element in the source U channel |
| * @param[in] v_input_ptr Pointer to the source V channel. Supported Format: U8 |
| * @param[in] v_input_stride_x Stride of the source image V channel in X dimension (in bytes) |
| * @param[in] v_input_step_x v_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] v_input_stride_y Stride of the source image V channel in Y dimension (in bytes) |
| * @param[in] v_input_step_y v_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] v_input_offset_first_element_in_bytes The offset of the first element in the source image V channel |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] uv_output_ptr Pointer to the destination UV channel. Supported Format: U8 |
| * @param[in] uv_output_stride_x Stride of the destination UV channel in X dimension (in bytes) |
| * @param[in] uv_output_step_x uv_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_output_stride_y Stride of the destination image U channel in Y dimension (in bytes) |
| * @param[in] uv_output_step_y uv_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_output_offset_first_element_in_bytes The offset of the first element in the destination UV channel |
| * |
| */ |
| __kernel void IYUV_to_NV12_bt709( |
| IMAGE_DECLARATION(luma_input), |
| IMAGE_DECLARATION(u_input), |
| IMAGE_DECLARATION(v_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(uv_output)) |
| { |
| Image in_y = CONVERT_TO_IMAGE_STRUCT(luma_input); |
| Image in_u = CONVERT_TO_IMAGE_STRUCT(u_input); |
| Image in_v = CONVERT_TO_IMAGE_STRUCT(v_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_uv = CONVERT_TO_IMAGE_STRUCT(uv_output); |
| |
| // handle 32 pixels every time, two lines, each line for 16 pixels |
| uchar16 luma_0 = vload16(0, in_y.ptr); |
| uchar16 luma_1 = vload16(0, in_y.ptr + luma_input_stride_y); |
| uchar8 cb = vload8(0, in_u.ptr); |
| uchar8 cr = vload8(0, in_v.ptr); |
| uchar16 cbcr = (uchar16)(cb.s0, cr.s0, cb.s1, cr.s1, cb.s2, cr.s2, cb.s3, cr.s3, cb.s4, cr.s4, cb.s5, cr.s5, cb.s6, |
| cr.s6, cb.s7, cr.s7); |
| |
| vstore16(luma_0, 0, out_y.ptr); |
| vstore16(luma_1, 0, out_y.ptr + luma_output_stride_y); |
| vstore16(cbcr, 0, out_uv.ptr); |
| } |
| |
| /** Convert a YUYV image to NV12 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 8), height ] |
| * No offset. |
| * |
| * @param[in] yuyv_input_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] yuyv_input_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] yuyv_input_step_x yuyv_input_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] yuyv_input_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] yuyv_input_step_y yuyv_input_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] yuyv_input_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_output_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_output_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_output_step_x luma_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_output_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_output_step_y luma_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_output_offset_first_element_in_bytes The offset of the first element in the destination luma channel |
| * @param[out] uv_output_ptr Pointer to the destination UV channel. Supported Format: U8 |
| * @param[in] uv_output_stride_x Stride of the destination UV channel in X dimension (in bytes) |
| * @param[in] uv_output_step_x uv_output_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_output_stride_y Stride of the destination image UV channel in Y dimension (in bytes) |
| * @param[in] uv_output_step_y uv_output_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_output_offset_first_element_in_bytes The offset of the first element in the destination UV channel |
| * |
| */ |
| __kernel void YUYV422_to_NV12_bt709( |
| IMAGE_DECLARATION(yuyv_input), |
| IMAGE_DECLARATION(luma_output), |
| IMAGE_DECLARATION(uv_output)) |
| { |
| Image in_yuyv = CONVERT_TO_IMAGE_STRUCT(yuyv_input); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma_output); |
| Image out_uv = CONVERT_TO_IMAGE_STRUCT(uv_output); |
| |
| // handle 16 pixels every time, each line 8 pixels |
| uchar16 yuyv = vload16(0, in_yuyv.ptr); |
| ushort8 cbcr_0 = (ushort8)(yuyv.s1, yuyv.s3, yuyv.s5, yuyv.s7, yuyv.s9, yuyv.sb, yuyv.sd, yuyv.sf); |
| uchar8 luma = (uchar8)(yuyv.s0, yuyv.s2, yuyv.s4, yuyv.s6, yuyv.s8, yuyv.sa, yuyv.sc, yuyv.se); |
| vstore8(luma, 0, out_y.ptr); |
| |
| yuyv = vload16(0, in_yuyv.ptr + yuyv_input_stride_y); |
| ushort8 cbcr_1 = (ushort8)(yuyv.s1, yuyv.s3, yuyv.s5, yuyv.s7, yuyv.s9, yuyv.sb, yuyv.sd, yuyv.sf); |
| luma = (uchar8)(yuyv.s0, yuyv.s2, yuyv.s4, yuyv.s6, yuyv.s8, yuyv.sa, yuyv.sc, yuyv.se); |
| vstore8(luma, 0, out_y.ptr + luma_output_stride_y); |
| |
| uchar8 cbcr = convert_uchar8((cbcr_0 + cbcr_1) / (ushort8)(2)); |
| vstore8(cbcr, 0, out_uv.ptr); |
| } |
| |
| /** Convert a UYVY image to NV12 using BT709 color space |
| * |
| * Global Workgroup Size [ DIV_CEIL(width, 4), height ] |
| * No offset. |
| * |
| * @param[in] input_uyvy_ptr Pointer to the source image. Supported Format: U8 |
| * @param[in] input_uyvy_stride_x Stride of the source image in X dimension (in bytes) |
| * @param[in] input_uyvy_step_x input_uyvy_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] input_uyvy_stride_y Stride of the source image in Y dimension (in bytes) |
| * @param[in] input_uyvy_step_y input_uyvy_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] input_uyvy_offset_first_element_in_bytes The offset of the first element in the source image |
| * @param[out] luma_ptr Pointer to the destination luma channel. Supported Format: U8 |
| * @param[in] luma_stride_x Stride of the destination luma channel in X dimension (in bytes) |
| * @param[in] luma_step_x luma_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] luma_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] luma_step_y luma_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] luma_offset_first_element_in_bytes The offset of the first element in the destination image luma channel |
| * @param[out] uv_ptr Pointer to the destination uv channel. Supported Format: U8 |
| * @param[in] uv_stride_x Stride of the destination uv channel in X dimension (in bytes) |
| * @param[in] uv_step_x uv_stride_x * number of elements along X processed per workitem(in bytes) |
| * @param[in] uv_stride_y Stride of the destination image luma channel in Y dimension (in bytes) |
| * @param[in] uv_step_y uv_stride_y * number of elements along Y processed per workitem(in bytes) |
| * @param[in] uv_offset_first_element_in_bytes The offset of the first element in the destination image uv channel |
| * |
| */ |
| __kernel void UYVY422_to_NV12_bt709( |
| IMAGE_DECLARATION(input_uyvy), |
| IMAGE_DECLARATION(luma), |
| IMAGE_DECLARATION(uv)) |
| { |
| Image in = CONVERT_TO_IMAGE_STRUCT(input_uyvy); |
| Image out_y = CONVERT_TO_IMAGE_STRUCT(luma); |
| Image out_uv = CONVERT_TO_IMAGE_STRUCT(uv); |
| |
| // handle 16 pixels every time, each line 8 pixels |
| const uchar16 uyvy_t = vload16(0, in.ptr); |
| vstore8(uyvy_t.s13579bdf, 0, out_y.ptr); |
| |
| const uchar16 uyvy_b = vload16(0, in.ptr + input_uyvy_stride_y); |
| vstore8(uyvy_b.s13579bdf, 0, out_y.ptr + luma_stride_y); |
| |
| const ushort8 cbcr_t = (ushort8)(uyvy_t.s0, uyvy_t.s2, uyvy_t.s4, uyvy_t.s6, uyvy_t.s8, uyvy_t.sa, uyvy_t.sc, uyvy_t.se); |
| const ushort8 cbcr_b = (ushort8)(uyvy_b.s0, uyvy_b.s2, uyvy_b.s4, uyvy_b.s6, uyvy_b.s8, uyvy_b.sa, uyvy_b.sc, uyvy_b.se); |
| const uchar8 cbcr = convert_uchar8((cbcr_t + cbcr_b) / (ushort8)(2)); |
| vstore8(cbcr, 0, out_uv.ptr); |
| } |
| |
| )" |