/* * Copyright (c) 2022 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. */ #include "activation_float_helpers.h" #include "helpers.h" #include "tile_helpers.h" #if defined(GEMMLOWP_MM_RESHAPED_ONLY_RHS_MMUL) /** This OpenCL kernel computes the matrix multiplication between 2 matrices using the MMUL extension: * * The LHS matrix is NOT reshaped * The RHS is reshaped with @ref ClGemmMatrixMultiplyReshapedOnlyRhsKernel and the block K0xN0 is transposed * * @note The block's dimensions used for reshaping the RHS matrix (N0 and K0) must be passed at compile time using -DN0 and -DK0 (e.g. -DN0=1, -DK0=1). * @note The number of M0 rows to process must be passed at compile time using -DM0 (e.g. -DM0=1) * @note The number of output columns processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_N0 (e.g., -DMMUL_N0=4) * @note The number of output rows processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_M0 (e.g., -DMMUL_M0=4) * @note The number of lhs columns (or rhs rows) processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_K0 (e.g., -DMMUL_K0=16) * @note Only the following configurations of M0, N0 and K0 are currently supported: * - M0 = 1, 2, 4 * - N0 = 1, 4, 8 * - K0 = 4 * * @note If the activation type were passed at compile time through -DACTIVATION_TYPE (e.g. -DACTIVATION_TYPE=RELU), A, B variables, required by some activation functions, should be passed at compile time as well using -DA_VAL= and -DB_VAL= respectively. * The activation function is performed after the bias addition * * @param[in] lhs_ptr Pointer to the LHS tensor. Supported data types: QASYMM8/QASYMM8_SIGNED * @param[in] lhs_stride_y Stride of the LHS tensor in Y dimension (in bytes) * @param[in] lhs_stride_z Stride of the LHS tensor in Z dimension (in bytes) * @param[in] lhs_w The size of the width dimension of the LHS tensor * @param[in] lhs_h The size of the height dimension of the LHS tensor * @param[in] lhs_n The size of the depth dimension of the LHS tensor * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS tensor * @param[in] rhs_ptr Pointer to the RHS reshaped tensor. Supported data type: same as @p lhs_ptr * @param[in] rhs_stride_y Stride of the RHS tensor in Y dimension (in bytes) * @param[in] rhs_stride_z Stride of the RHS tensor in Z dimension (in bytes) * @param[in] rhs_w The size of the width dimension of the RHS tensor * @param[in] rhs_h The size of the height dimension of the RHS tensor * @param[in] rhs_n The size of the depth dimension of the RHS tensor * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS tensor * @param[in] bia_ptr (Optional) Pointer to the bias tensor. Supported data type: S32 * @param[in] bia_stride_y (Optional) Stride of the bias tensor in Y dimension (in bytes) * @param[in] bia_stride_z (Optional) Stride of the bias tensor in Z dimension (in bytes) * @param[in] bia_w (Optional) The size of the width dimension of the bias tensor * @param[in] bia_h (Optional) The size of the height dimension of the bias tensor * @param[in] bia_n (Optional) The size of the depth dimension of the bias tensor * @param[in] bia_offset_first_element_in_bytes (Optional) The offset of the first element in the bias tensor * @param[out] dst_ptr Pointer to the destination tensor. Supported data type: same as @p lhs_ptr or S32 * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_w The size of the width dimension of the destination tensor * @param[in] dst_h The size of the height dimension of the destination tensor * @param[in] dst_n The size of the depth dimension of the destination tensor * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor * @param[in] M Number of rows in LHS matrix not reshaped * @param[in] N Number of columns in RHS matrix not reshaped * @param[in] K Number of columns in LHS matrix and rows in RHS matrix not reshaped * @param[in] sum_col_ptr (Optional) Pointer to the source tensor. Supported data type: S32 * @param[in] sum_col_stride_x (Optional) Stride of the source tensor in X dimension (in bytes) * @param[in] sum_col_step_x (Optional) sum_col_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] sum_col_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes) * @param[in] sum_col_step_y (Optional) sum_col_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] sum_col_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor * @param[in] sum_row_ptr (Optional) Pointer to the source tensor. Supported data type: S32 * @param[in] sum_row_stride_x (Optional) Stride of the source tensor in X dimension (in bytes) * @param[in] sum_row_step_x (Optional) sum_row_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] sum_row_stride_y (Optional) Stride of the source tensor in Y dimension (in bytes) * @param[in] sum_row_step_y (Optional) sum_row_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] sum_row_offset_first_element_in_bytes (Optional) The offset of the first element in the source tensor */ __kernel void gemmlowp_mm_reshaped_only_rhs_mmul( TENSOR3D_T(lhs, BUFFER), TENSOR3D_T(rhs, BUFFER), #if defined(ADD_BIAS) TENSOR3D_T(bia, BUFFER), #endif // defined(ADD_BIAS) TENSOR3D_T(dst, BUFFER), const int M, const int N, const int K #if defined(A_OFFSET) , TENSOR3D_T(sum_col, BUFFER) #endif // defined(A_OFFSET) #if defined(B_OFFSET) , TENSOR3D_T(sum_row, BUFFER) #endif // defined(B_OFFSET) ) { #define MMUL_BLOCK_SIZE (MMUL_N0 * MMUL_M0) #define VEC_SIZE 4 // For int8 types input to mmul instruction is a length 4 vector uint x0 = get_global_id(0); uint y0 = get_global_id(1); uint z = get_global_id(2); // Get block ID and thread ID within the block uint block_id = (x0 / MMUL_BLOCK_SIZE); uint thread_id = (x0 % MMUL_BLOCK_SIZE); // Coordinate within a block uint block_x = thread_id % MMUL_N0; uint block_y = (thread_id / MMUL_M0); // Starting destination coordinates uint dst_x = min(block_x * N0 + block_id * MMUL_N0 * N0, (uint)(N - 1)); uint dst_y = min(block_y * M0 + y0 * M0 * MMUL_M0, (uint)(M - M0)); uint lhs_x = VEC_SIZE * block_x; uint lhs_y = dst_y; uint rhs_x = VEC_SIZE * N0 * block_y; uint rhs_y = 4 * block_id + block_x; // Compute LHS/RHS/DST matrix address lhs_offset_first_element_in_bytes += lhs_x * sizeof(DATA_TYPE) + lhs_y * lhs_stride_y + z * lhs_stride_z; rhs_offset_first_element_in_bytes += rhs_x * sizeof(DATA_TYPE) + rhs_y * rhs_stride_y + z * rhs_stride_z; dst_offset_first_element_in_bytes += dst_x * sizeof(OUT_DATA_TYPE) + dst_y * dst_stride_y + z * dst_stride_z; TILE(ACC_DATA_TYPE, M0, N0, c); LOOP_UNROLLING(int, i, 0, 1, M0, { c[i].v = 0; }) for(int k = 0; k <= K - MMUL_K0; k += MMUL_K0) { TILE(DATA_TYPE, M0, VEC_SIZE, a); T_LOAD(DATA_TYPE, M0, VEC_SIZE, BUFFER, lhs, 0, 0, 1, lhs_stride_y, a); TILE(DATA_TYPE, N0, VEC_SIZE, b); T_LOAD(DATA_TYPE, N0, VEC_SIZE, BUFFER, rhs, 0, 0, 1, VEC_SIZE, b); LOOP_UNROLLING(int, m0, 0, 1, M0, { LOOP_UNROLLING(int, n0, 0, 1, N0, { VEC_TYPE vec_a = (VEC_TYPE)(a[m0].s[0], a[m0].s[1], a[m0].s[2], a[m0].s[3]); VEC_TYPE vec_b = (VEC_TYPE)(b[n0].s[0], b[n0].s[1], b[n0].s[2], b[n0].s[3]); c[m0].s[n0] = arm_matrix_multiply(vec_a, vec_b, c[m0].s[n0]); }) }) lhs_offset_first_element_in_bytes += MMUL_K0 * sizeof(DATA_TYPE); rhs_offset_first_element_in_bytes += MMUL_K0 * N0 * sizeof(DATA_TYPE); } if(block_x * N0 + block_id * MMUL_N0 * N0 >= N) { return; } if(block_y * M0 + y0 * M0 * MMUL_M0 >= M) { return; } #if defined(FUSED_OUTPUT_STAGE_FIXED_POINT) TILE(int, M0, N0, offset_s32); LOOP_UNROLLING(int, i, 0, 1, M0, { offset_s32[i].v = (VEC_DATA_TYPE(int, N0))K_OFFSET; }) #if defined(A_OFFSET) TILE(int, 1, N0, a_offset_s32); T_LOAD(int, 1, N0, BUFFER, sum_col, dst_x, z, 1, sum_col_stride_z, a_offset_s32); a_offset_s32[0].v *= A_OFFSET; T_ELTWISE_BROADCAST_ADD_X(int, M0, N0, offset_s32, a_offset_s32, offset_s32); #endif // defined(A_OFFSET) #if defined(B_OFFSET) TILE(int, M0, 1, b_offset_s32); T_LOAD(int, M0, 1, BUFFER, sum_row, dst_y, z * M, 1, 4, b_offset_s32); LOOP_UNROLLING(int, m0, 0, 1, M0, { offset_s32[m0].v += b_offset_s32[m0].v *B_OFFSET; }) #endif // defined(B_OFFSET) #if defined(ADD_BIAS) #if defined(BROADCAST_BIAS) bia_offset_first_element_in_bytes += dst_x * sizeof(ACC_DATA_TYPE) + z * bia_stride_y; TILE(int, M0, N0, bias); T_LOAD(int, M0, N0, BUFFER, bia, dst_x, dst_y, 1, 1, bias); T_ADD(ACC_DATA_TYPE, M0, N0, offset_s32, bias, offset_s32); #else // defined(BROADCAST_BIAS) bia_offset_first_element_in_bytes += dst_x * sizeof(ACC_DATA_TYPE); TILE(int, 1, N0, bias); if(dst_x + N0 <= N || N0_LEFTOVER == 0) { bias[0].v = VLOAD(N0)(0, (ACC_DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes)); } else { VLOAD_PARTIAL(N0, N0_LEFTOVER) (bias[0].v, 0, (ACC_DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes)); } T_ELTWISE_BROADCAST_ADD_X(int, M0, N0, offset_s32, bias, offset_s32); #endif // defined(BROADCAST_BIAS) #endif // defined(ADD_BIAS) T_ADD(ACC_DATA_TYPE, M0, N0, c, offset_s32, c); TILE(OUT_DATA_TYPE, M0, N0, c_lp); T_QUANTIZE8(ACC_DATA_TYPE, OUT_DATA_TYPE, PER_TENSOR, M0, N0, RESULT_OFFSET, RESULT_SHIFT, RESULT_MULTIPLIER, c, 0, 0, c_lp); #if defined(MIN_BOUND) LOOP_UNROLLING(int, i, 0, 1, M0, { c_lp[i].v = max(c_lp[i].v, (VEC_DATA_TYPE(OUT_DATA_TYPE, N0))MIN_BOUND); }) #endif // defined(MIN_BOUND) #if defined(MAX_BOUND) LOOP_UNROLLING(int, i, 0, 1, M0, { c_lp[i].v = min(c_lp[i].v, (VEC_DATA_TYPE(OUT_DATA_TYPE, N0))MAX_BOUND); }) #endif // defined(MAX_BOUND) T_ACTIVATION(DATA_TYPE, M0, N0, ACTIVATION_TYPE, A_VAL, B_VAL, c, c); if(dst_x + N0 <= N || N0_LEFTOVER == 0) { LOOP_UNROLLING(int, m0, 0, 1, M0, { if(dst_y + m0 < M || M0_LEFTOVER == 0) { VSTORE(N0) (c_lp[m0].v, 0, (__global OUT_DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y)); } }) } else { LOOP_UNROLLING(int, m0, 0, 1, M0, { if(dst_y + m0 < M || M0_LEFTOVER == 0) { VSTORE_PARTIAL(N0, N0_LEFTOVER) (c_lp[m0].v, 0, (__global OUT_DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y)); } }) } #else // FUSED_OUTPUT_STAGE_FIXED_POINT // Store if(dst_x + N0 <= N || N0_LEFTOVER == 0) { LOOP_UNROLLING(int, m0, 0, 1, M0, { if(dst_y + m0 < M || M0_LEFTOVER == 0) { VSTORE(N0) (c[m0].v, 0, (__global OUT_DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y)); } }) } else { LOOP_UNROLLING(int, m0, 0, 1, M0, { if(dst_y + m0 < M || M0_LEFTOVER == 0) { VSTORE_PARTIAL(N0, N0_LEFTOVER) (c[m0].v, 0, (__global OUT_DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y)); } }) } #endif // FUSED_OUTPUT_STAGE_FIXED_POINT } #endif // defined(GEMMLOWP_MM_RESHAPED_ONLY_RHS_MMUL)