/* * Copyright (c) 2017-2019 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 "gemm_helpers.h" #include "repeat.h" #if defined(M0) && defined(K0) && defined(V0) && defined(DATA_TYPE) && defined(SRC_WIDTH) #define INC2 (VEC_DATA_TYPE(uint, 2))(0, 1) #define INC3 (VEC_DATA_TYPE(uint, 3))(0, 1, 2) #define INC4 (VEC_DATA_TYPE(uint, 4))(0, 1, 2, 3) #define INC8 (VEC_DATA_TYPE(uint, 8))(0, 1, 2, 3, 4, 5, 6, 7) #define INC16 (VEC_DATA_TYPE(uint, 16))(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15) #define CONCAT_INC(K0) INC##K0 #define INC(K0) CONCAT_INC(K0) #if(SRC_WIDTH % K0) #define BOUNDARY_CONDITION_X(x, a) \ ({ \ a = select(0, a, CONVERT(((x * (VEC_DATA_TYPE(uint, K0))K0 + INC(K0)) < (VEC_DATA_TYPE(uint, K0))SRC_WIDTH), VEC_DATA_TYPE(DATA_TYPE, K0))); \ }) #else // (SRC_WIDTH % K0) #define BOUNDARY_CONDITION_X(x, a) \ ({}) #endif // (SRC_WIDTH % K0) /** This OpenCL kernel reshapes the lhs input matrix. The kernel splits the input matrix in blocks of size M0xK0 and stores each one (not transposed) in * the output matrix unrolling the values. * * @note The data type must be passed at compile time using -DDATA_TYPE (e.g. -DDATA_TYPE=float) * @note The width of the input tensor must be passed at compile time using -DSRC_WIDTH (e.g. -DSRC_WIDTH=16) * @note The block's dimensions (M0 and K0) must be passed at compile time using -DM0 and -DK0 (e.g. -DM0=2, -DK0=2). * @note The number of M0xK0 vertical blocks to store on the same output row must be passed at compile time using -DV0 (e.g. -DV0=2) * @note Only the following values for M0, K0 and V0 are supported: * M0: 2,3,4,5,6,7,8 * K0: 2,3,4,8,16 * V0: greater than 0 * @note In case the input has to be reinterpreted as a 3D tensor (e.g. input of convolution layer 1x1), the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# HEIGHT_GEMM3D: The height of the input in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the input in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * @note If the M0xK0 blocks have to be interleaved, the option -DINTERLEAVE must passed at compile time. * * @param[in] src_ptr Pointer to the source LHS tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 * @param[in] src_stride_x Stride of the source LHS tensor in X dimension (in bytes) * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src_stride_y Stride of the source LHS tensor in Y dimension (in bytes) * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src_stride_z Stride of the source LHS tensor in Z dimension (in bytes) * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source LHS tensor * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_INPUT_AS_3D) */ __kernel void gemm_reshape_lhs_matrix_nt(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(dst) #if defined(REINTERPRET_INPUT_AS_3D) , uint cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D ) { // Block size #define BLOCK_SIZE ((M0) * (K0)) // Output offset X #if defined(INTERLEAVE) #define OUTPUT_OFFSET_X (K0) #else // defined(INTERLEAVE) #define OUTPUT_OFFSET_X (BLOCK_SIZE) #endif // defined(INTERLEAVE) // Output step X #if defined(INTERLEAVE) #define OUTPUT_STEP_X (K0) * (V0) #else // Do not interleave #define OUTPUT_STEP_X (K0) #endif // defined(INTERLEAVE) // Compute source and destination addresses uint x = get_global_id(0); uint y = get_global_id(1); uint z = get_global_id(2); // ------------------ Compute input/output addresses --------------------------- // Compute the input address __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)K0 * sizeof(DATA_TYPE) + y * (uint)M0 * src_stride_y; // Compute the output address __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)BLOCK_SIZE * (uint)V0 * sizeof(DATA_TYPE)) + ((y / (uint)V0) * (uint)dst_stride_y) + ((y % V0) * (uint)OUTPUT_OFFSET_X * sizeof(DATA_TYPE)); // Create variables: uint zin0=0, zin1=0, zin2=0...zin(M0-1)=0; REPEAT_VAR_INIT_TO_CONST(M0, uint, zin, 0); #if defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply src_stride_z by DEPTH_GEMM3D input_ptr += z * (uint)src_stride_z * DEPTH_GEMM3D; // The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zin, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, cross_plane_pad, src_stride_y); #else // defined(REINTERPRET_INPUT_AS_3D) input_ptr += z * (uint)src_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM output_ptr += z * (uint)dst_stride_z; // ---------------------------Load input values -------------------------------- // Load values from the LHS matrix LOAD_BLOCK(M0, K0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin); BOUNDARY_CONDITION_X(x, a0); #if M0 > 1 BOUNDARY_CONDITION_X(x, a1); #endif // M0 > 1 #if M0 > 2 BOUNDARY_CONDITION_X(x, a2); #endif // M0 > 2 #if M0 > 3 BOUNDARY_CONDITION_X(x, a3); #endif // M0 > 3 #if M0 > 4 BOUNDARY_CONDITION_X(x, a4); #endif // M0 > 4 #if M0 > 5 BOUNDARY_CONDITION_X(x, a5); #endif // M0 > 5 #if M0 > 6 BOUNDARY_CONDITION_X(x, a6); #endif // M0 > 6 #if M0 > 7 BOUNDARY_CONDITION_X(x, a7); #endif // M0 > 7 // ---------------------------Store output values ------------------------------ REPEAT_VAR_INIT_TO_CONST(16, uint, zout, 0); STORE_BLOCK(M0, K0, DATA_TYPE, a, output_ptr, OUTPUT_STEP_X * sizeof(DATA_TYPE), zout); #undef BLOCK_SIZE #undef OUTPUT_OFFSET_X #undef OUTPUT_STEP_X } #if M0 == 2 #define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, M0) \ res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i); \ VSTORE(M0) \ (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \ }) #elif M0 == 3 // M0 == 3 #define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, M0) \ res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i, a2.s##i); \ VSTORE(M0) \ (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \ }) #elif M0 == 4 // M0 == 4 #define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, M0) \ res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \ VSTORE(M0) \ (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \ }) #elif M0 == 5 // M0 == 5 #define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, 4) \ res0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \ DATA_TYPE res1 = a4.s##i; \ VSTORE(4) \ (res0, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \ *((__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE)) + 4) = res1; \ }) #elif M0 == 6 // M0 == 6 #define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, 4) \ res0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \ VEC_DATA_TYPE(DATA_TYPE, 2) \ res1 = (VEC_DATA_TYPE(DATA_TYPE, 2))(a4.s##i, a5.s##i); \ VSTORE(4) \ (res0, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \ VSTORE(2) \ (res1, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE)) + 4); \ }) #elif M0 == 7 // M0 == 7 #define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, 4) \ res0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s##i, a1.s##i, a2.s##i, a3.s##i); \ VEC_DATA_TYPE(DATA_TYPE, 3) \ res1 = (VEC_DATA_TYPE(DATA_TYPE, 3))(a4.s##i, a5.s##i, a6.s##i); \ VSTORE(4) \ (res0, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \ VSTORE(3) \ (res1, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE)) + 4); \ }) #elif M0 == 8 // M0 == 8 #define TRANSPOSE_COLUMN_AND_STORE(output_ptr, output_step_x, i) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, M0) \ res = (VEC_DATA_TYPE(DATA_TYPE, M0))(a0.s##i, a1.s##i, a2.s##i, a3.s##i, a4.s##i, a5.s##i, a6.s##i, a7.s##i); \ VSTORE(M0) \ (res, 0, (__global DATA_TYPE *)(output_ptr + 0x##i * output_step_x * sizeof(DATA_TYPE))); \ }) #else // M0 not supported #error "M0 value not supported" #endif // N0 conditions /** This OpenCL kernel reshapes the lhs input matrix. The kernel splits the input matrix in blocks of size M0xK0 and stores each one (transposed) in * the output matrix unrolling the values. * * @note The data type must be passed at compile time using -DDATA_TYPE (e.g. -DDATA_TYPE=float) * @note The width of the input tensor must be passed at compile time using -DSRC_WIDTH (e.g. -DSRC_WIDTH=16) * @note The block's dimensions (M0 and K0) must be passed at compile time using -DM0 and -DK0 (e.g. -DM0=2, -DK0=2). * @note The number of M0xK0 vertical blocks to store on the same output row must be passed at compile time using -DV0 (e.g. -DV0=2) * @note Only the following values for M0, K0 and V0 are supported: * M0: 2,3,4,5,6,7,8 * K0: 2,3,4,8,16 * V0: greater than 0 * @note In case the input has to be reinterpreted as a 3D tensor (e.g. input of convolution layer 1x1), the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# HEIGHT_GEMM3D: The height of the input in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the input in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * @note If the M0xK0 blocks have to be interleaved, the option -DINTERLEAVE must passed at compile time. * * @param[in] src_ptr Pointer to the source LHS tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 * @param[in] src_stride_x Stride of the source LHS tensor in X dimension (in bytes) * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src_stride_y Stride of the source LHS tensor in Y dimension (in bytes) * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src_stride_z Stride of the source LHS tensor in Z dimension (in bytes) * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source LHS tensor * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_INPUT_AS_3D) */ __kernel void gemm_reshape_lhs_matrix_t(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(dst) #if defined(REINTERPRET_INPUT_AS_3D) , uint cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D ) { // Block size #define BLOCK_SIZE ((M0) * (K0)) // Output offset X #if defined(INTERLEAVE) #define OUTPUT_OFFSET_X (M0) #else // defined(INTERLEAVE) #define OUTPUT_OFFSET_X (BLOCK_SIZE) #endif // defined(INTERLEAVE) // Output step X #if defined(INTERLEAVE) #define OUTPUT_STEP_X (M0) * (V0) #else // Do not interleave #define OUTPUT_STEP_X (M0) #endif // defined(INTERLEAVE) // Compute source and destination addresses uint x = get_global_id(0); uint y = get_global_id(1); uint z = get_global_id(2); // ------------------ Compute input/output addresses --------------------------- // Compute the input address __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)K0 * sizeof(DATA_TYPE) + y * (uint)M0 * src_stride_y; // Compute the output address __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)BLOCK_SIZE * (uint)V0 * sizeof(DATA_TYPE)) + ((y / (uint)V0) * (uint)dst_stride_y) + ((y % V0) * (uint)OUTPUT_OFFSET_X * sizeof(DATA_TYPE)); // Create variables: uint zin0=0, zin1=0, zin2=0...zin(M0-1)=0; REPEAT_VAR_INIT_TO_CONST(M0, uint, zin, 0); #if defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply src_stride_z by DEPTH_GEMM3D input_ptr += z * (uint)src_stride_z * DEPTH_GEMM3D; // The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zin, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, cross_plane_pad, src_stride_y); #else // defined(REINTERPRET_INPUT_AS_3D) input_ptr += z * (uint)src_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM output_ptr += z * (uint)dst_stride_z; // ---------------------------Load input values -------------------------------- // Load values from the LHS matrix LOAD_BLOCK(M0, K0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin); BOUNDARY_CONDITION_X(x, a0); #if M0 > 1 BOUNDARY_CONDITION_X(x, a1); #endif // M0 > 1 #if M0 > 2 BOUNDARY_CONDITION_X(x, a2); #endif // M0 > 2 #if M0 > 3 BOUNDARY_CONDITION_X(x, a3); #endif // M0 > 3 #if M0 > 4 BOUNDARY_CONDITION_X(x, a4); #endif // M0 > 4 #if M0 > 5 BOUNDARY_CONDITION_X(x, a5); #endif // M0 > 5 #if M0 > 6 BOUNDARY_CONDITION_X(x, a6); #endif // M0 > 6 #if M0 > 7 BOUNDARY_CONDITION_X(x, a7); #endif // M0 > 7 // ---------------------------Transpose and store block ----------------------- TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 0); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 1); #if K0 > 2 TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 2); #endif // K0 > 2 #if K0 > 3 TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 3); #endif // K0 > 3 #if K0 > 4 TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 4); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 5); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 6); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 7); #endif // K0 > 4 #if K0 > 8 TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 8); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, 9); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, A); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, B); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, C); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, D); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, E); TRANSPOSE_COLUMN_AND_STORE(output_ptr, OUTPUT_STEP_X, F); #endif // K0 > 8 #undef BLOCK_SIZE #undef OUTPUT_OFFSET_X #undef OUTPUT_STEP_X } #endif // defined(M0) && defined(K0) && defined(V0) && defined(DATA_TYPE) && defined(SRC_WIDTH) #if defined(K0) && defined(N0) && defined(H0) && defined(DATA_TYPE) && defined(SRC_HEIGHT) /** This OpenCL kernel reshapes the rhs input matrix. The kernel splits the input matrix in blocks of size K0xN0 and stores each one (not transposed) in * the output matrix unrolling the values. * * @note The data type must be passed at compile time using -DDATA_TYPE (e.g. -DDATA_TYPE=float) * @note The height of the input tensor must be passed at compile time using -DSRC_HEIGHT (e.g. -DSRC_HEIGHT=16) * @note The block's dimensions (K0 and N0) must be passed at compile time using -DK0 and -DN0 (e.g. -DK0=2, -DN0=2). * @note The number of K0xN0 vertical blocks to store on the same output row must be passed at compile time using -DH0 (e.g. -DH0=2) * @note If the K0xN0 blocks have to be interleaved, the option -DINTERLEAVE must passed at compile time. * @note Only the following values for K0, N0 and H0 are supported: * N0: 2,3,4,8,16 * K0: 1,2,3,4,8,16 * H0: greater than 0 * * @param[in] src_ptr Pointer to the source RHS tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 * @param[in] src_stride_x Stride of the source RHS tensor in X dimension (in bytes) * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src_stride_y Stride of the source RHS tensor in Y dimension (in bytes) * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src_stride_z Stride of the source RHS tensor in Z dimension (in bytes) * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source RHS tensor * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix */ __kernel void gemm_reshape_rhs_matrix_nt(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(dst)) { // Block size #define BLOCK_SIZE ((K0) * (N0)) // Output offset X #if defined(INTERLEAVE) #define OUTPUT_OFFSET_X (N0) #else // defined(INTERLEAVE) #define OUTPUT_OFFSET_X (BLOCK_SIZE) #endif // defined(INTERLEAVE) // Output step X #if defined(INTERLEAVE) #define OUTPUT_STEP_X (N0) * (H0) #else // Do not interleave #define OUTPUT_STEP_X (N0) #endif // defined(INTERLEAVE) // Compute source and destination addresses uint x = get_global_id(0); uint y = get_global_id(1); uint z = get_global_id(2); // ------------------ Compute input/output addresses --------------------------- // Compute the input address __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)N0 * sizeof(DATA_TYPE) + y * (uint)K0 * src_stride_y + z * (uint)src_stride_z; // Compute the output address __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (y * (uint)BLOCK_SIZE * (uint)H0 * sizeof(DATA_TYPE)) + ((x % (uint)H0) * (uint)OUTPUT_OFFSET_X * sizeof(DATA_TYPE)) + (( x / (uint)H0) * (uint)dst_stride_y) + z * (uint)dst_stride_z; // ---------------------------Load input values -------------------------------- REPEAT_VAR_INIT_TO_CONST(K0, VEC_DATA_TYPE(DATA_TYPE, N0), a, 0); ////uint a0=0, a1=0, a2=0...a(M0-1)=0; // Load values from the RHS matrix a0 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 0 * src_stride_y)); #if K0 > 1 if(y * (uint)K0 + 1 < SRC_HEIGHT) { a1 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 1 * src_stride_y)); } #endif // K0 > 1 #if K0 > 2 if(y * (uint)K0 + 2 < SRC_HEIGHT) { a2 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 2 * src_stride_y)); } #endif // K0 > 2 #if K0 > 3 if(y * (uint)K0 + 3 < SRC_HEIGHT) { a3 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 3 * src_stride_y)); } #endif // K0 > 3 #if K0 > 4 if(y * (uint)K0 + 4 < SRC_HEIGHT) { a4 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 4 * src_stride_y)); } if(y * (uint)K0 + 5 < SRC_HEIGHT) { a5 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 5 * src_stride_y)); } if(y * (uint)K0 + 6 < SRC_HEIGHT) { a6 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 6 * src_stride_y)); } if(y * (uint)K0 + 7 < SRC_HEIGHT) { a7 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 7 * src_stride_y)); } #endif // K0 > 4 #if K0 > 8 if(y * (uint)K0 + 8 < SRC_HEIGHT) { a8 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 8 * src_stride_y)); } if(y * (uint)K0 + 9 < SRC_HEIGHT) { a9 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 9 * src_stride_y)); } if(y * (uint)K0 + 10 < SRC_HEIGHT) { aA = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 10 * src_stride_y)); } if(y * (uint)K0 + 11 < SRC_HEIGHT) { aB = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 11 * src_stride_y)); } if(y * (uint)K0 + 12 < SRC_HEIGHT) { aC = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 12 * src_stride_y)); } if(y * (uint)K0 + 13 < SRC_HEIGHT) { aD = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 13 * src_stride_y)); } if(y * (uint)K0 + 14 < SRC_HEIGHT) { aE = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 14 * src_stride_y)); } if(y * (uint)K0 + 15 < SRC_HEIGHT) { aF = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 15 * src_stride_y)); } #endif // K0 > 8 // ---------------------------Store output values ------------------------------ REPEAT_VAR_INIT_TO_CONST(16, uint, zout, 0); STORE_BLOCK(K0, N0, DATA_TYPE, a, output_ptr, OUTPUT_STEP_X * sizeof(DATA_TYPE), zout); #undef BLOCK_SIZE #undef OUTPUT_OFFSET_X #undef OUTPUT_STEP_X } #if defined(TRANSPOSE) /** This OpenCL kernel reshapes the rhs input matrix. The kernel splits the input matrix in blocks of size K0xN0 and stores each one (transposed) in * the output matrix unrolling the values. * * @note The data type must be passed at compile time using -DDATA_TYPE (e.g. -DDATA_TYPE=float) * @note The height of the input tensor must be passed at compile time using -DSRC_HEIGHT (e.g. -DSRC_HEIGHT=16) * @note The block's dimensions (K0 and N0) must be passed at compile time using -DK0 and -DN0 (e.g. -DK0=2, -DN0=2). * @note The number of K0xN0 vertical blocks to store on the same output row must be passed at compile time using -DH0 (e.g. -DH0=2) * @note If the K0xN0 blocks have to be interleaved, the option -DINTERLEAVE must passed at compile time. * @note The option -DTRANSPOSE must passed at compile time. * @note Only the following values for K0, N0 and H0 are supported: * N0: 2,3,4,8,16 * K0: 2,3,4,8,16 * H0: greater than 0 * * @param[in] src_ptr Pointer to the source RHS tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 * @param[in] src_stride_x Stride of the source RHS tensor in X dimension (in bytes) * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src_stride_y Stride of the source RHS tensor in Y dimension (in bytes) * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src_stride_z Stride of the source RHS tensor in Z dimension (in bytes) * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source RHS tensor * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix */ __kernel void gemm_reshape_rhs_matrix_t(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(dst)) { // Block size #define BLOCK_SIZE ((K0) * (N0)) // Output offset X #if defined(INTERLEAVE) #define OUTPUT_OFFSET_X (K0) #else // defined(INTERLEAVE) #define OUTPUT_OFFSET_X (BLOCK_SIZE) #endif // defined(INTERLEAVE) // Output step X #if defined(INTERLEAVE) #define OUTPUT_STEP_X (K0) * (H0) #else // Do not interleave #define OUTPUT_STEP_X (K0) #endif // defined(INTERLEAVE) // Compute source and destination addresses uint x = get_global_id(0); uint y = get_global_id(1); uint z = get_global_id(2); // ------------------ Compute input/output addresses --------------------------- // Compute the input address __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * (uint)N0 * sizeof(DATA_TYPE) + y * (uint)K0 * src_stride_y + z * (uint)src_stride_z; // Compute the output address __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (y * (uint)BLOCK_SIZE * (uint)H0 * sizeof(DATA_TYPE)) + ((x % H0) * (uint)OUTPUT_OFFSET_X * sizeof(DATA_TYPE)) + ((x / (uint)H0) * (uint)dst_stride_y) + z * (uint)dst_stride_z; // ---------------------------Load input values -------------------------------- REPEAT_VAR_INIT_TO_CONST(K0, VEC_DATA_TYPE(DATA_TYPE, N0), a, 0); //VEC_DATA_TYPE(DATA_TYPE, N0) a0=0, a1=0, ... a(K0-1)=0; // Load values from the RHS matrix a0 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 0 * src_stride_y)); if(y * (uint)K0 + 1 < SRC_HEIGHT) { a1 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 1 * src_stride_y)); } #if K0 > 2 if(y * (uint)K0 + 2 < SRC_HEIGHT) { a2 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 2 * src_stride_y)); } #endif // K0 > 2 #if K0 > 3 if(y * (uint)K0 + 3 < SRC_HEIGHT) { a3 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 3 * src_stride_y)); } #endif // K0 > 3 #if K0 > 4 if(y * (uint)K0 + 4 < SRC_HEIGHT) { a4 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 4 * src_stride_y)); } if(y * (uint)K0 + 5 < SRC_HEIGHT) { a5 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 5 * src_stride_y)); } if(y * (uint)K0 + 6 < SRC_HEIGHT) { a6 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 6 * src_stride_y)); } if(y * (uint)K0 + 7 < SRC_HEIGHT) { a7 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 7 * src_stride_y)); } #endif // K0 > 4 #if K0 > 8 if(y * (uint)K0 + 8 < SRC_HEIGHT) { a8 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 8 * src_stride_y)); } if(y * (uint)K0 + 9 < SRC_HEIGHT) { a9 = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 9 * src_stride_y)); } if(y * (uint)K0 + 10 < SRC_HEIGHT) { aA = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 10 * src_stride_y)); } if(y * (uint)K0 + 11 < SRC_HEIGHT) { aB = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 11 * src_stride_y)); } if(y * (uint)K0 + 12 < SRC_HEIGHT) { aC = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 12 * src_stride_y)); } if(y * (uint)K0 + 13 < SRC_HEIGHT) { aD = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 13 * src_stride_y)); } if(y * (uint)K0 + 14 < SRC_HEIGHT) { aE = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 14 * src_stride_y)); } if(y * (uint)K0 + 15 < SRC_HEIGHT) { aF = VLOAD(N0)(0, (__global DATA_TYPE *)(input_ptr + 15 * src_stride_y)); } #endif // K0 > 8 // ---------------------------Transpose the block ------------------------------ REPEAT_VAR_INIT_TO_CONST(N0, VEC_DATA_TYPE(DATA_TYPE, K0), res, 0); //VEC_DATA_TYPE(DATA_TYPE, K0) res0=0, res1=0, res2=0,... res(N0-1)=0; #if K0 == 2 // This part computes the following transpositions: // 2x2 -> 2x2 // 2x4 -> 4x2 // 2x8 -> 8x2 // 2x16 -> 16x2 res0 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s0, a1.s0); res1 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s1, a1.s1); #if N0 > 2 res2 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s2, a1.s2); #endif // N0 > 2 #if N0 > 3 res3 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s3, a1.s3); #endif // N0 > 3 #if N0 > 4 res4 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s4, a1.s4); res5 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s5, a1.s5); res6 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s6, a1.s6); res7 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s7, a1.s7); #endif // N0 > 4 #if N0 > 8 res8 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s8, a1.s8); res9 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s9, a1.s9); resA = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sA, a1.sA); resB = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sB, a1.sB); resC = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sC, a1.sC); resD = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sD, a1.sD); resE = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sE, a1.sE); resF = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sF, a1.sF); #endif // N0 > 8 #elif K0 == 3 // K0 == 2 // This part computes the following transpositions: // 3x2 -> 2x3 // 3x4 -> 4x3 // 3x8 -> 8x3 // 3x16 -> 16x3 res0 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s0, a1.s0, a2.s0); res1 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s1, a1.s1, a2.s1); #if N0 > 2 res2 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s2, a1.s2, a2.s2); #endif // N0 > 2 #if N0 > 3 res3 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s3, a1.s3, a2.s3); #endif // N0 > 3 #if N0 > 4 res4 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s4, a1.s4, a2.s4); res5 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s5, a1.s5, a2.s5); res6 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s6, a1.s6, a2.s6); res7 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s7, a1.s7, a2.s7); #endif // N0 > 4 #if N0 > 8 res8 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s8, a1.s8, a2.s8); res9 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s9, a1.s9, a2.s9); resA = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sA, a1.sA, a2.sA); resB = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sB, a1.sB, a2.sB); resC = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sC, a1.sC, a2.sC); resD = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sD, a1.sD, a2.sD); resE = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sE, a1.sE, a2.sE); resF = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sF, a1.sF, a2.sF); #endif // N0 > 8 #elif K0 == 4 // K0 == 4 // This part computes the following transpositions: // 4x2 -> 2x4 // 4x4 -> 4x4 // 4x8 -> 8x4 // 4x16 -> 16x4 res0 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s0, a1.s0, a2.s0, a3.s0); res1 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s1, a1.s1, a2.s1, a3.s1); #if N0 > 2 res2 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s2, a1.s2, a2.s2, a3.s2); #endif // N0 > 2 #if N0 > 3 res3 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s3, a1.s3, a2.s3, a3.s3); #endif // N0 > 3 #if N0 > 4 res4 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s4, a1.s4, a2.s4, a3.s4); res5 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s5, a1.s5, a2.s5, a3.s5); res6 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s6, a1.s6, a2.s6, a3.s6); res7 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s7, a1.s7, a2.s7, a3.s7); #endif // N0 > 4 #if N0 > 8 res8 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s8, a1.s8, a2.s8, a3.s8); res9 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s9, a1.s9, a2.s9, a3.s9); resA = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sA, a1.sA, a2.sA, a3.sA); resB = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sB, a1.sB, a2.sB, a3.sB); resC = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sC, a1.sC, a2.sC, a3.sC); resD = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sD, a1.sD, a2.sD, a3.sD); resE = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sE, a1.sE, a2.sE, a3.sE); resF = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sF, a1.sF, a2.sF, a3.sF); #endif // N0 > 8 #elif K0 == 8 // K0 == 8 // This part computes the following transpositions: // 8x2 -> 2x8 // 8x4 -> 4x8 // 8x8 -> 8x8 // 8x16 -> 16x8 res0 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s0, a1.s0, a2.s0, a3.s0, a4.s0, a5.s0, a6.s0, a7.s0); res1 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s1, a1.s1, a2.s1, a3.s1, a4.s1, a5.s1, a6.s1, a7.s1); #if N0 > 2 res2 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s2, a1.s2, a2.s2, a3.s2, a4.s2, a5.s2, a6.s2, a7.s2); #endif // N0 > 2 #if N0 > 3 res3 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s3, a1.s3, a2.s3, a3.s3, a4.s3, a5.s3, a6.s3, a7.s3); #endif // N0 > 3 #if N0 > 4 res4 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s4, a1.s4, a2.s4, a3.s4, a4.s4, a5.s4, a6.s4, a7.s4); res5 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s5, a1.s5, a2.s5, a3.s5, a4.s5, a5.s5, a6.s5, a7.s5); res6 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s6, a1.s6, a2.s6, a3.s6, a4.s6, a5.s6, a6.s6, a7.s6); res7 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s7, a1.s7, a2.s7, a3.s7, a4.s7, a5.s7, a6.s7, a7.s7); #endif // N0 > 4 #if N0 > 8 res8 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s8, a1.s8, a2.s8, a3.s8, a4.s8, a5.s8, a6.s8, a7.s8); res9 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s9, a1.s9, a2.s9, a3.s9, a4.s9, a5.s9, a6.s9, a7.s9); resA = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sA, a1.sA, a2.sA, a3.sA, a4.sA, a5.sA, a6.sA, a7.sA); resB = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sB, a1.sB, a2.sB, a3.sB, a4.sB, a5.sB, a6.sB, a7.sB); resC = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sC, a1.sC, a2.sC, a3.sC, a4.sC, a5.sC, a6.sC, a7.sC); resD = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sD, a1.sD, a2.sD, a3.sD, a4.sD, a5.sD, a6.sD, a7.sD); resE = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sE, a1.sE, a2.sE, a3.sE, a4.sE, a5.sE, a6.sE, a7.sE); resF = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sF, a1.sF, a2.sF, a3.sF, a4.sF, a5.sF, a6.sF, a7.sF); #endif // N0 > 8 #elif K0 == 16 // K0 == 16 // This part computes the following transpositions: // 16x2 -> 2x16 // 16x4 -> 4x16 // 16x8 -> 8x16 // 16x16 -> 16x16 res0 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s0, a1.s0, a2.s0, a3.s0, a4.s0, a5.s0, a6.s0, a7.s0, a8.s0, a9.s0, aA.s0, aB.s0, aC.s0, aD.s0, aE.s0, aF.s0); res1 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s1, a1.s1, a2.s1, a3.s1, a4.s1, a5.s1, a6.s1, a7.s1, a8.s1, a9.s1, aA.s1, aB.s1, aC.s1, aD.s1, aE.s1, aF.s1); #if N0 > 2 res2 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s2, a1.s2, a2.s2, a3.s2, a4.s2, a5.s2, a6.s2, a7.s2, a8.s2, a9.s2, aA.s2, aB.s2, aC.s2, aD.s2, aE.s2, aF.s2); #endif // N0 > 2 #if N0 > 3 res3 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s3, a1.s3, a2.s3, a3.s3, a4.s3, a5.s3, a6.s3, a7.s3, a8.s3, a9.s3, aA.s3, aB.s3, aC.s3, aD.s3, aE.s3, aF.s3); #endif // N0 > 3 #if N0 > 4 res4 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s4, a1.s4, a2.s4, a3.s4, a4.s4, a5.s4, a6.s4, a7.s4, a8.s4, a9.s4, aA.s4, aB.s4, aC.s4, aD.s4, aE.s4, aF.s4); res5 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s5, a1.s5, a2.s5, a3.s5, a4.s5, a5.s5, a6.s5, a7.s5, a8.s5, a9.s5, aA.s5, aB.s5, aC.s5, aD.s5, aE.s5, aF.s5); res6 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s6, a1.s6, a2.s6, a3.s6, a4.s6, a5.s6, a6.s6, a7.s6, a8.s6, a9.s6, aA.s6, aB.s6, aC.s6, aD.s6, aE.s6, aF.s6); res7 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s7, a1.s7, a2.s7, a3.s7, a4.s7, a5.s7, a6.s7, a7.s7, a8.s7, a9.s7, aA.s7, aB.s7, aC.s7, aD.s7, aE.s7, aF.s7); #endif // N0 > 4 #if N0 > 8 res8 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s8, a1.s8, a2.s8, a3.s8, a4.s8, a5.s8, a6.s8, a7.s8, a8.s8, a9.s8, aA.s8, aB.s8, aC.s8, aD.s8, aE.s8, aF.s8); res9 = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.s9, a1.s9, a2.s9, a3.s9, a4.s9, a5.s9, a6.s9, a7.s9, a8.s9, a9.s9, aA.s9, aB.s9, aC.s9, aD.s9, aE.s9, aF.s9); resA = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sA, a1.sA, a2.sA, a3.sA, a4.sA, a5.sA, a6.sA, a7.sA, a8.sA, a9.sA, aA.sA, aB.sA, aC.sA, aD.sA, aE.sA, aF.sA); resB = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sB, a1.sB, a2.sB, a3.sB, a4.sB, a5.sB, a6.sB, a7.sB, a8.sB, a9.sB, aA.sB, aB.sB, aC.sB, aD.sB, aE.sB, aF.sB); resC = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sC, a1.sC, a2.sC, a3.sC, a4.sC, a5.sC, a6.sC, a7.sC, a8.sC, a9.sC, aA.sC, aB.sC, aC.sC, aD.sC, aE.sC, aF.sC); resD = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sD, a1.sD, a2.sD, a3.sD, a4.sD, a5.sD, a6.sD, a7.sD, a8.sD, a9.sD, aA.sD, aB.sD, aC.sD, aD.sD, aE.sD, aF.sD); resE = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sE, a1.sE, a2.sE, a3.sE, a4.sE, a5.sE, a6.sE, a7.sE, a8.sE, a9.sE, aA.sE, aB.sE, aC.sE, aD.sE, aE.sE, aF.sE); resF = (VEC_DATA_TYPE(DATA_TYPE, K0))(a0.sF, a1.sF, a2.sF, a3.sF, a4.sF, a5.sF, a6.sF, a7.sF, a8.sF, a9.sF, aA.sF, aB.sF, aC.sF, aD.sF, aE.sF, aF.sF); #endif // N0 > 8 #else // N0 == 16 #error "Not supported N0 value" #endif // N0 > 2 // ---------------------------Store the output values ------------------------------ REPEAT_VAR_INIT_TO_CONST(16, uint, zout, 0); STORE_BLOCK(N0, K0, DATA_TYPE, res, output_ptr, OUTPUT_STEP_X * sizeof(DATA_TYPE), zout); #undef BLOCK_SIZE #undef OUTPUT_OFFSET_X #undef OUTPUT_STEP_X } #endif // defined(TRANSPOSE) #endif // defined(K0) && defined(N0) && defined(H0) && defined(DATA_TYPE) && defined(SRC_HEIGHT) #if defined(M0) && defined(N0) && defined(K0) && defined(H0) && defined(DATA_TYPE) && defined(M) && defined(N) && defined(K) #define CONCAT(a, b) a##b #define ARM_DOT1(a, b, c) \ ({ \ c = fma(a, b, c); \ }) #define ARM_DOT2(a, b, c) \ ({ \ c = fma(a.s0, b.s0, c); \ c = fma(a.s1, b.s1, c); \ }) #define ARM_DOT3(a, b, c) \ ({ \ ARM_DOT2(a, b, c); \ c = fma((a.s2), (b.s2), c); \ }) #define ARM_DOT4(a, b, c) \ ({ \ ARM_DOT3(a, b, c); \ c = fma((a.s3), (b.s3), c); \ }) #define ARM_DOT8(a, b, c) \ ({ \ ARM_DOT4((a.lo), (b.lo), c); \ ARM_DOT4((a.hi), (b.hi), c); \ }) #define ARM_DOT16(a, b, c) \ ({ \ ARM_DOT8((a.lo), (b.lo), c); \ ARM_DOT8((a.hi), (b.hi), c); \ }) #if N0 == 2 #define ARM_DOT_K0XN0(k0, a, b, c) \ ({ \ CONCAT(ARM_DOT, k0) \ ((a), (b##0), (c.s0)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##1), (c.s1)); \ }) #elif N0 == 3 // N0 == 3 #define ARM_DOT_K0XN0(k0, a, b, c) \ ({ \ CONCAT(ARM_DOT, k0) \ ((a), (b##0), (c.s0)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##1), (c.s1)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##2), (c.s2)); \ }) #elif N0 == 4 // N0 == 4 #define ARM_DOT_K0XN0(k0, a, b, c) \ ({ \ CONCAT(ARM_DOT, k0) \ ((a), (b##0), (c.s0)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##1), (c.s1)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##2), (c.s2)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##3), (c.s3)); \ }) #elif N0 == 8 // N0 == 8 #define ARM_DOT_K0XN0(k0, a, b, c) \ ({ \ CONCAT(ARM_DOT, k0) \ ((a), (b##0), (c.s0)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##1), (c.s1)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##2), (c.s2)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##3), (c.s3)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##4), (c.s4)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##5), (c.s5)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##6), (c.s6)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##7), (c.s7)); \ }) #elif N0 == 16 // N0 == 16 #define ARM_DOT_K0XN0(k0, a, b, c) \ ({ \ CONCAT(ARM_DOT, k0) \ ((a), (b##0), (c.s0)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##1), (c.s1)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##2), (c.s2)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##3), (c.s3)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##4), (c.s4)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##5), (c.s5)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##6), (c.s6)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##7), (c.s7)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##8), (c.s8)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##9), (c.s9)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##A), (c.sA)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##B), (c.sB)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##C), (c.sC)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##D), (c.sD)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##E), (c.sE)); \ CONCAT(ARM_DOT, k0) \ ((a), (b##F), (c.sF)); \ }) #else // N0 not supported #error "N0 value not supported" #endif // N0 conditions /** This OpenCL kernel computes the matrix multiplication between 2 matrices. * The LHS matrix is NOT reshaped * The RHS is reshaped with @ref CLGEMMReshapeRHSMatrixKernel and the block K0xN0 is transposed * * @note If the first two dimensions of NDRange have been dispatched with "dummy_work_items" support, the option -DDUMMY_WORK_ITEMS must be passed at compile time. * @note The GEMM's dimensions (M,N and K) must be passed at compile time using -DM, -DN and and -DK (e.g. -DM=52, -DN=30 and -DK=90) * @note The number of columns of LHS matrix must be passed at compile time using -DK (e.g. -DK=64) * @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=8, -DK0=4). * @note The number of M0 rows to process must be passed at compile time using -DM0 (e.g. -DM0=2) * @note The number of K0xN0 horizontal blocks stored on the same output row of the reshaped RHS matrix must be passed at compile time using -DH0 (e.g. -DH0=2) * @note If the K0xN0 blocks in the reshaped RHS matrix have been interleaved, the option -DRHS_INTERLEAVE must passed at compile time. * @note Only the following configurations of M0, N0 and K0 are currently supported: * - M0 = 1, 2, 3, 4, 5, 6, 7, 8 * - N0 = 2, 3, 4, 8, 16 * - K0 = 2, 3, 4, 8, 16 * - H0 >= 1 * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns LHS matrix * * @param[in] lhs_ptr Pointer to the LHS reshaped matrix. Supported data type: F16/F32 * @param[in] lhs_stride_x Stride of the LHS reshaped matrix in X dimension (in bytes) * @param[in] lhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] lhs_stride_y Stride of the LHS reshaped matrix in Y dimension (in bytes) * @param[in] lhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS reshaped matrix * @param[in] rhs_ptr Pointer to the RHS reshaped matrix. Supported data type: same as @p lhs_ptr * @param[in] rhs_stride_x Stride of the RHS reshaped matrix in X dimension (in bytes) * @param[in] rhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] rhs_stride_y Stride of the RHS reshaped matrix in Y dimension (in bytes) * @param[in] rhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS reshaped matrix * @param[in] bias_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] bias_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] bias_step_x (Optional) bias_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] bias_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] bias_step_y (Optional) bias_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] bias_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data type: same as @p lhs_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] lhs_stride_z Stride of the LHS reshaped matrix in Z dimension (in bytes) * @param[in] rhs_stride_z Stride of the RHS reshaped matrix in Z dimension (in bytes) * @param[in] bias_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] lhs_cross_plane_pad (Optional) Bottom paddings for LHS matrix in unit of elements (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings for the output matrix in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_reshaped_only_rhs_t(IMAGE_DECLARATION(lhs), IMAGE_DECLARATION(rhs), #if defined(BETA) IMAGE_DECLARATION(bias), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint lhs_stride_z, uint rhs_stride_z, #if defined(BETA) uint bias_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint lhs_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { // Block size #define RHS_BLOCK_SIZE ((K0) * (N0)) // RHS offset and step X #if defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (K0) #define RHS_STEP_X ((K0) * (H0)) #define RHS_STEP_LOOP (1) #else // defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (RHS_BLOCK_SIZE) #define RHS_STEP_X (K0) #define RHS_STEP_LOOP (H0) #endif // defined(RHS_INTERLEAVE) uint x = get_global_id(0); uint y = get_global_id(1); uint z = get_global_id(2); #if defined(DUMMY_WORK_ITEMS) if((x * N0 >= N) || (y * M0 >= M)) { return; } #endif // defined(DUMMY_WORK_ITEMS) // Compute LHS matrix address uint lhs_offset = lhs_offset_first_element_in_bytes + y * M0 * (uint)lhs_stride_y; // Compute RHS matrix address uint rhs_offset = rhs_offset_first_element_in_bytes + (x % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (x / (uint)H0) * rhs_stride_y; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 rhs_offset += (z % MATRIX_B_DEPTH) * rhs_stride_z; #else // defined(MATRIX_B_DEPTH) rhs_offset += z * rhs_stride_z; #endif // defined(MATRIX_B_DEPTH) REPEAT_VAR_INIT_TO_CONST(8, uint, zlhs, 0); //uint zlhs0=0,zlhs1=0,zlhs2=0,... zlhs7=0; REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0); #if defined(REINTERPRET_INPUT_AS_3D) // The plane (zlhs) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zlhs, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, lhs_cross_plane_pad, lhs_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply lhs_stride_z by DEPTH_GEMM3D lhs_offset += z * lhs_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM lhs_offset += z * lhs_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) // Initialize the accumulators REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), c, 0); //VEC_DATA_TYPE(DATA_TYPE, N0) c0=0,c1=0,c2=0,... c(M0-1)=0; int i = 0; for(; i <= (K - K0); i += K0) { // Supported cases (M0, K0): // 1,2 - 1,3 - 1,4 - 1,8 - 1,16 // 2,2 - 2,3 - 2,4 - 2,8 - 2,16 // 3,2 - 3,3 - 3,4 - 3,8 - 3,16 // 4,2 - 4,3 - 4,4 - 4,8 - 4,16 // 5,2 - 5,3 - 5,4 - 5,8 - 5,16 // 6,2 - 6,3 - 6,4 - 6,8 - 6,16 // 7,2 - 7,3 - 7,4 - 7,8 - 7,16 // 8,2 - 8,3 - 8,4 - 8,8 - 8,16 // Load values from LHS matrix LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs); // Load values from RHS matrix LOAD_BLOCK(N0, K0, DATA_TYPE, b, rhs_ptr, rhs_offset, RHS_STEP_X * sizeof(DATA_TYPE), zero); // Accumulate ARM_DOT_K0XN0(K0, a0, b, c0); #if M0 > 1 ARM_DOT_K0XN0(K0, a1, b, c1); #endif // M0 > 1 #if M0 > 2 ARM_DOT_K0XN0(K0, a2, b, c2); #endif // M0 > 2 #if M0 > 3 ARM_DOT_K0XN0(K0, a3, b, c3); #endif // M0 > 3 #if M0 > 4 ARM_DOT_K0XN0(K0, a4, b, c4); #endif // M0 > 4 #if M0 > 5 ARM_DOT_K0XN0(K0, a5, b, c5); #endif // M0 > 5 #if M0 > 6 ARM_DOT_K0XN0(K0, a6, b, c6); #endif // M0 > 6 #if M0 > 7 ARM_DOT_K0XN0(K0, a7, b, c7); #endif // M0 > 7 lhs_offset += K0 * sizeof(DATA_TYPE); rhs_offset += (N0 * RHS_STEP_X * RHS_STEP_LOOP) * sizeof(DATA_TYPE); } // Left-over accumulations for(; i < K; ++i) { // Load values from LHS matrix LOAD_BLOCK(M0, 1, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs); // Load values from RHS matrix LOAD_BLOCK(N0, 1, DATA_TYPE, b, rhs_ptr, rhs_offset, RHS_STEP_X * sizeof(DATA_TYPE), zero); // Accumulate ARM_DOT_K0XN0(1, a0, b, c0); #if M0 > 1 ARM_DOT_K0XN0(1, a1, b, c1); #endif // M0 > 1 #if M0 > 2 ARM_DOT_K0XN0(1, a2, b, c2); #endif // M0 > 2 #if M0 > 3 ARM_DOT_K0XN0(1, a3, b, c3); #endif // M0 > 3 #if M0 > 4 ARM_DOT_K0XN0(1, a4, b, c4); #endif // M0 > 4 #if M0 > 5 ARM_DOT_K0XN0(1, a5, b, c5); #endif // M0 > 5 #if M0 > 6 ARM_DOT_K0XN0(1, a6, b, c6); #endif // M0 > 6 #if M0 > 7 ARM_DOT_K0XN0(1, a7, b, c7); #endif // M0 > 7 lhs_offset += sizeof(DATA_TYPE); rhs_offset += sizeof(DATA_TYPE); } __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)M0 * dst_stride_y); REPEAT_VAR_INIT_TO_CONST(8, uint, zout, 0); //uint zout0=0,zout1=0,zout2=0,... zout7=0; #if defined(REINTERPRET_OUTPUT_AS_3D) // The plane (zout) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zout, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) #if defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)); LOAD_BLOCK(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(M0, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * bias_stride_y) + get_global_id( 2) * bias_stride_z; LOAD_BLOCK(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(M0, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(M0, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store output block STORE_BLOCK(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout); #undef RHS_BLOCK_SIZE #undef RHS_OFFSET_X #undef RHS_STEP_X } #define VFMA(a, b, c) \ ({ \ c = fma(a, b, c); \ }) #if M0 == 1 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ }) #elif M0 == 2 // M0 == 2 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ }) #elif M0 == 3 // M0 == 3 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ }) #elif M0 == 4 // M0 == 4 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ }) #elif M0 == 5 // M0 == 5 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ }) #elif M0 == 6 // M0 == 6 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \ }) #elif M0 == 7 // M0 == 7 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \ }) #elif M0 == 8 // M0 == 8 #define LD_RHS_VFMA_M0xN0(i, a, c) \ ({ \ VEC_DATA_TYPE(DATA_TYPE, N0) \ b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0x##i * RHS_STEP_X * sizeof(DATA_TYPE))); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##7).s##i), b, (c##7)); \ }) #else // M0 not supported #error "M0 not supported" #endif // M0 not supported /** This OpenCL kernel computes the matrix multiplication between 2 matrices. * The LHS matrix is NOT reshaped * The RHS is reshaped with @ref CLGEMMReshapeRHSMatrixKernel and the block K0xN0 is NOT transposed * * @note If the first two dimensions of NDRange have been dispatched with "dummy_work_items" support, the option -DDUMMY_WORK_ITEMS must be passed at compile time. * @note The GEMM's dimensions (M,N and K) must be passed at compile time using -DM, -DN and and -DK (e.g. -DM=52, -DN=30 and -DK=90). * @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=8, -DK0=4). * @note The number of M0 rows to process must be passed at compile time using -DM0 (e.g. -DM0=2) * @note The number of K0xN0 horizontal blocks stored on the same output row of the reshaped RHS matrix must be passed at compile time using -DH0 (e.g. -DH0=2) * @note If the K0xN0 blocks in the reshaped RHS matrix have been interleaved, the option -DRHS_INTERLEAVE must passed at compile time. * @note Only the following configurations of M0, N0 and K0 are currently supported: * - M0 = 1, 2, 3, 4, 5, 6, 7, 8 * - N0 = 2, 3, 4, 8, 16 * - K0 = 2, 3, 4, 8, 16 * - H0 >= 1 * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns LHS matrix * * @param[in] lhs_ptr Pointer to the LHS reshaped matrix. Supported data type: F16/F32 * @param[in] lhs_stride_x Stride of the LHS reshaped matrix in X dimension (in bytes) * @param[in] lhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] lhs_stride_y Stride of the LHS reshaped matrix in Y dimension (in bytes) * @param[in] lhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS reshaped matrix * @param[in] rhs_ptr Pointer to the RHS reshaped matrix. Supported data type: same as @p lhs_ptr * @param[in] rhs_stride_x Stride of the RHS reshaped matrix in X dimension (in bytes) * @param[in] rhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] rhs_stride_y Stride of the RHS reshaped matrix in Y dimension (in bytes) * @param[in] rhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS reshaped matrix * @param[in] bias_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] bias_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] bias_step_x (Optional) bias_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] bias_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] bias_step_y (Optional) bias_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] bias_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data type: same as @p lhs_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] lhs_stride_z Stride of the LHS reshaped matrix in Z dimension (in bytes) * @param[in] rhs_stride_z Stride of the RHS reshaped matrix in Z dimension (in bytes) * @param[in] bias_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] lhs_cross_plane_pad (Optional) Bottom paddings for LHS matrix in unit of elements (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings for the output matrix in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_reshaped_only_rhs_nt(IMAGE_DECLARATION(lhs), IMAGE_DECLARATION(rhs), #if defined(BETA) IMAGE_DECLARATION(bias), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint lhs_stride_z, uint rhs_stride_z, #if defined(BETA) uint bias_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint lhs_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { // Block size #define RHS_BLOCK_SIZE ((K0) * (N0)) // RHS offset and step X #if defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (N0) #define RHS_STEP_X ((N0) * (H0)) #define RHS_STEP_LOOP (1) #else // defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (RHS_BLOCK_SIZE) #define RHS_STEP_X (N0) #define RHS_STEP_LOOP (H0) #endif // defined(RHS_INTERLEAVE) uint x = get_global_id(0); uint y = get_global_id(1); uint z = get_global_id(2); #if defined(DUMMY_WORK_ITEMS) if((x * N0 >= N) || (y * M0 >= M)) { return; } #endif // defined(DUMMY_WORK_ITEMS) // Compute LHS matrix address uint lhs_offset = lhs_offset_first_element_in_bytes + y * M0 * (uint)lhs_stride_y; // Compute RHS matrix address uint rhs_offset = rhs_offset_first_element_in_bytes + (x % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (x / (uint)H0) * rhs_stride_y; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 rhs_offset += (z % MATRIX_B_DEPTH) * rhs_stride_z; #else // defined(MATRIX_B_DEPTH) rhs_offset += z * rhs_stride_z; #endif // defined(MATRIX_B_DEPTH) REPEAT_VAR_INIT_TO_CONST(8, uint, zin, 0); //uint zin0=0,zin1=0,zin2=0,... zin7=0; REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0); //uint zero0=0,zero1=0,zero2=0,... zero7=0; #if defined(REINTERPRET_INPUT_AS_3D) // The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zin, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, lhs_cross_plane_pad, lhs_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply lhs_stride_z by DEPTH_GEMM3D lhs_offset += z * lhs_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM lhs_offset += z * lhs_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) // Initialize the accumulators REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), c, 0); //VEC_DATA_TYPE(DATA_TYPE, N0) c0=0,c1=0,c2=0,... c(N0-1)=0; int i = 0; for(; i <= (K - K0); i += K0) { // Supported cases (M0, K0): // 1,2 - 1,3 - 1,4 - 1,8 - 1,16 // 2,2 - 2,3 - 2,4 - 2,8 - 2,16 // 3,2 - 3,3 - 3,4 - 3,8 - 3,16 // 4,2 - 4,3 - 4,4 - 4,8 - 4,16 // 5,2 - 5,3 - 5,4 - 5,8 - 5,16 // 6,2 - 6,3 - 6,4 - 6,8 - 6,16 // 7,2 - 7,3 - 7,4 - 7,8 - 7,16 // 8,2 - 8,3 - 8,4 - 8,8 - 8,16 // Load values from LHS matrix LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zin); LD_RHS_VFMA_M0xN0(0, a, c); LD_RHS_VFMA_M0xN0(1, a, c); #if K0 > 2 LD_RHS_VFMA_M0xN0(2, a, c); #endif // K0 > 2 #if K0 > 3 LD_RHS_VFMA_M0xN0(3, a, c); #endif // K0 > 3 #if K0 > 4 LD_RHS_VFMA_M0xN0(4, a, c); LD_RHS_VFMA_M0xN0(5, a, c); LD_RHS_VFMA_M0xN0(6, a, c); LD_RHS_VFMA_M0xN0(7, a, c); #endif // K0 > 4 #if K0 > 8 LD_RHS_VFMA_M0xN0(8, a, c); LD_RHS_VFMA_M0xN0(9, a, c); LD_RHS_VFMA_M0xN0(A, a, c); LD_RHS_VFMA_M0xN0(B, a, c); LD_RHS_VFMA_M0xN0(C, a, c); LD_RHS_VFMA_M0xN0(D, a, c); LD_RHS_VFMA_M0xN0(E, a, c); LD_RHS_VFMA_M0xN0(F, a, c); #endif // K0 > 8 lhs_offset += K0 * sizeof(DATA_TYPE); rhs_offset += K0 * RHS_STEP_X * RHS_STEP_LOOP * sizeof(DATA_TYPE); } // Left-over accumulations for(; i < K; ++i) { // Load values from LHS matrix VEC_DATA_TYPE(DATA_TYPE, 2) a0 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 0 * lhs_stride_y + zin0)); #if M0 > 1 VEC_DATA_TYPE(DATA_TYPE, 2) a1 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 1 * lhs_stride_y + zin1)); #endif // M0 > 1 #if M0 > 2 VEC_DATA_TYPE(DATA_TYPE, 2) a2 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 2 * lhs_stride_y + zin2)); #endif // M0 > 2 #if M0 > 3 VEC_DATA_TYPE(DATA_TYPE, 2) a3 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 3 * lhs_stride_y + zin3)); #endif // M0 > 3 #if M0 > 4 VEC_DATA_TYPE(DATA_TYPE, 2) a4 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 4 * lhs_stride_y + zin4)); #endif // M0 > 4 #if M0 > 5 VEC_DATA_TYPE(DATA_TYPE, 2) a5 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 5 * lhs_stride_y + zin5)); #endif // M0 > 5 #if M0 > 6 VEC_DATA_TYPE(DATA_TYPE, 2) a6 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 6 * lhs_stride_y + zin6)); #endif // M0 > 6 #if M0 > 7 VEC_DATA_TYPE(DATA_TYPE, 2) a7 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zin7)); #endif // M0 > 7 LD_RHS_VFMA_M0xN0(0, a, c); lhs_offset += sizeof(DATA_TYPE); rhs_offset += RHS_STEP_X * sizeof(DATA_TYPE); } __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)M0 * dst_stride_y); REPEAT_VAR_INIT_TO_CONST(8, uint, zout, 0); //uint zout0=0,zout1=0,zout2=0,... zout7=0; #if defined(REINTERPRET_OUTPUT_AS_3D) // The plane (zout) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zout, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) #if defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)); LOAD_BLOCK(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(M0, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * bias_stride_y) + get_global_id( 2) * bias_stride_z; LOAD_BLOCK(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(M0, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(M0, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store output block STORE_BLOCK(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout); #undef RHS_BLOCK_SIZE #undef RHS_OFFSET_X #undef RHS_STEP_X } #endif // defined(M0) && defined(N0) && defined(K0) && defined(H0) && defined(DATA_TYPE) && defined(M) && defined(N) && defined(K) #if defined(M0) && defined(N0) && defined(K0) && defined(V0) && defined(H0) && defined(DATA_TYPE) && defined(M) && defined(N) #if K0 == 2 #define ARM_DOT_K0(a, b, c) \ ({ \ c = fma(a.s0, b.s0, c); \ c = fma(a.s1, b.s1, c); \ }) #elif K0 == 3 // K0 == 3 #define ARM_DOT_K0(a, b, c) \ ({ \ c = fma(a.s0, b.s0, c); \ c = fma(a.s1, b.s1, c); \ c = fma(a.s2, b.s2, c); \ }) #elif K0 == 4 // K0 == 4 #define ARM_DOT_K0(a, b, c) \ ({ \ c = fma(a.s0, b.s0, c); \ c = fma(a.s1, b.s1, c); \ c = fma(a.s2, b.s2, c); \ c = fma(a.s3, b.s3, c); \ }) #elif K0 == 8 // K0 == 8 #define ARM_DOT_K0(a, b, c) \ ({ \ c = fma(a.s0, b.s0, c); \ c = fma(a.s1, b.s1, c); \ c = fma(a.s2, b.s2, c); \ c = fma(a.s3, b.s3, c); \ c = fma(a.s4, b.s4, c); \ c = fma(a.s5, b.s5, c); \ c = fma(a.s6, b.s6, c); \ c = fma(a.s7, b.s7, c); \ }) #elif K0 == 16 // K0 == 16 #define ARM_DOT_K0(a, b, c) \ ({ \ c = fma(a.s0, b.s0, c); \ c = fma(a.s1, b.s1, c); \ c = fma(a.s2, b.s2, c); \ c = fma(a.s3, b.s3, c); \ c = fma(a.s4, b.s4, c); \ c = fma(a.s5, b.s5, c); \ c = fma(a.s6, b.s6, c); \ c = fma(a.s7, b.s7, c); \ c = fma(a.s8, b.s8, c); \ c = fma(a.s9, b.s9, c); \ c = fma(a.sA, b.sA, c); \ c = fma(a.sB, b.sB, c); \ c = fma(a.sC, b.sC, c); \ c = fma(a.sD, b.sD, c); \ c = fma(a.sE, b.sE, c); \ c = fma(a.sF, b.sF, c); \ }) #else // K0 not supported #error "K0 value not supported" #endif // K0 conditions #if N0 == 2 #define ARM_DOT_K0XN0(a, b, c) \ ({ \ ARM_DOT_K0((a), (b##0), (c.s0)); \ ARM_DOT_K0((a), (b##1), (c.s1)); \ }) #elif N0 == 3 // N0 == 3 #define ARM_DOT_K0XN0(a, b, c) \ ({ \ ARM_DOT_K0((a), (b##0), (c.s0)); \ ARM_DOT_K0((a), (b##1), (c.s1)); \ ARM_DOT_K0((a), (b##2), (c.s2)); \ }) #elif N0 == 4 // N0 == 4 #define ARM_DOT_K0XN0(a, b, c) \ ({ \ ARM_DOT_K0((a), (b##0), (c.s0)); \ ARM_DOT_K0((a), (b##1), (c.s1)); \ ARM_DOT_K0((a), (b##2), (c.s2)); \ ARM_DOT_K0((a), (b##3), (c.s3)); \ }) #elif N0 == 8 // N0 == 8 #define ARM_DOT_K0XN0(a, b, c) \ ({ \ ARM_DOT_K0((a), (b##0), (c.s0)); \ ARM_DOT_K0((a), (b##1), (c.s1)); \ ARM_DOT_K0((a), (b##2), (c.s2)); \ ARM_DOT_K0((a), (b##3), (c.s3)); \ ARM_DOT_K0((a), (b##4), (c.s4)); \ ARM_DOT_K0((a), (b##5), (c.s5)); \ ARM_DOT_K0((a), (b##6), (c.s6)); \ ARM_DOT_K0((a), (b##7), (c.s7)); \ }) #elif N0 == 16 // N0 == 16 #define ARM_DOT_K0XN0(a, b, c) \ ({ \ ARM_DOT_K0((a), (b##0), (c.s0)); \ ARM_DOT_K0((a), (b##1), (c.s1)); \ ARM_DOT_K0((a), (b##2), (c.s2)); \ ARM_DOT_K0((a), (b##3), (c.s3)); \ ARM_DOT_K0((a), (b##4), (c.s4)); \ ARM_DOT_K0((a), (b##5), (c.s5)); \ ARM_DOT_K0((a), (b##6), (c.s6)); \ ARM_DOT_K0((a), (b##7), (c.s7)); \ ARM_DOT_K0((a), (b##8), (c.s8)); \ ARM_DOT_K0((a), (b##9), (c.s9)); \ ARM_DOT_K0((a), (b##A), (c.sA)); \ ARM_DOT_K0((a), (b##B), (c.sB)); \ ARM_DOT_K0((a), (b##C), (c.sC)); \ ARM_DOT_K0((a), (b##D), (c.sD)); \ ARM_DOT_K0((a), (b##E), (c.sE)); \ ARM_DOT_K0((a), (b##F), (c.sF)); \ }) #else // N0 not supported #error "N0 value not supported" #endif // N0 conditions /** This OpenCL kernel computes the matrix multiplication between 2 matrices. * The LHS matrix must be reshaped with @ref CLGEMMReshapeLHSMatrixKernel and the M0xK0 must be NOT transposed * The RHS matrix must be reshaped with @ref CLGEMMReshapeRHSMatrixKernel and the K0xN0 must be transposed * * @note If the first two dimensions of NDRange have been dispatched with "dummy_work_items" support, the option -DDUMMY_WORK_ITEMS must be passed at compile time. * @note The GEMM's dimensions M and N must be passed at compile time using -DM and -DN (e.g. -DM=52 and -DN=90). * @note The block's dimensions used for reshaping the LHS matrix and the RHS matrix (M0, N0 and K0) must be passed at compile time using -DM0, -DN0 and -DK0 (e.g. -DM0=4, -DN0=8, -DK0=4). * @note The number of M0xK0 vertical blocks stored on the same output row of the reshaped LHS matrix must be passed at compile time using -DV0 (e.g. -DV0=2) * @note The number of K0xN0 horizontal blocks stored on the same output row of the reshaped RHS matrix must be passed at compile time using -DH0 (e.g. -DH0=2) * @note If the M0xK0 blocks in the reshaped LHS matrix have been interleaved, the option -DLHS_INTERLEAVE must passed at compile time. * @note If the K0xN0 blocks in the reshaped RHS matrix have been interleaved, the option -DRHS_INTERLEAVE must passed at compile time. * @note Only the following configurations of M0, N0 and K0 are currently supported: * - M0 = 2, 3, 4, 5, 6, 7, 8 * - N0 = 2, 3, 4, 8, 16 * - K0 = 2, 3, 4, 8, 16 * - V0 >= 1 * - H0 >= 1 * * @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 * @note In case the output has to be reinterpreted as a 3D tensor (e.g. output of convolution layer), the following information must be passed at compile time: * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns LHS matrix NOT reshaped * * @param[in] lhs_ptr Pointer to the LHS reshaped matrix. Supported data type: F16/F32 * @param[in] lhs_stride_x Stride of the LHS reshaped matrix in X dimension (in bytes) * @param[in] lhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] lhs_stride_y Stride of the LHS reshaped matrix in Y dimension (in bytes) * @param[in] lhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS reshaped matrix * @param[in] rhs_ptr Pointer to the RHS reshaped matrix. Supported data type: same as @p lhs_ptr * @param[in] rhs_stride_x Stride of the RHS reshaped matrix in X dimension (in bytes) * @param[in] rhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] rhs_stride_y Stride of the RHS reshaped matrix in Y dimension (in bytes) * @param[in] rhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS reshaped matrix * @param[in] bias_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] bias_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] bias_step_x (Optional) bias_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] bias_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] bias_step_y (Optional) bias_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] bias_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data type: same as @p lhs_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] k Number of columns in LHS matrix and rows in RHS matrix not reshaped. * @param[in] lhs_stride_z Stride of the LHS reshaped matrix in Z dimension (in bytes) * @param[in] rhs_stride_z Stride of the RHS reshaped matrix in Z dimension (in bytes) * @param[in] bias_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_reshaped_lhs_nt_rhs_t(IMAGE_DECLARATION(lhs), IMAGE_DECLARATION(rhs), #if defined(BETA) IMAGE_DECLARATION(bias), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint k, uint lhs_stride_z, uint rhs_stride_z, #if defined(BETA) uint bias_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { // Block size #define LHS_BLOCK_SIZE ((K0) * (M0)) #if defined(LHS_INTERLEAVE) #define LHS_OFFSET_X (K0) #define LHS_STEP_X ((K0) * (V0)) #define LHS_STEP_LOOP (1) #else // defined(INTERLEAVE) #define LHS_OFFSET_X (LHS_BLOCK_SIZE) #define LHS_STEP_X (K0) #define LHS_STEP_LOOP (V0) #endif // defined(INTERLEAVE) // Block size #define RHS_BLOCK_SIZE ((K0) * (N0)) // RHS offset and step X #if defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (K0) #define RHS_STEP_X ((K0) * (H0)) #define RHS_STEP_LOOP (1) #else // defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (RHS_BLOCK_SIZE) #define RHS_STEP_X (K0) #define RHS_STEP_LOOP (H0) #endif // defined(RHS_INTERLEAVE) #if defined(DUMMY_WORK_ITEMS) if((get_global_id(0) * N0 >= N) || (get_global_id(1) * M0 >= M)) { return; } #endif // defined(DUMMY_WORK_ITEMS) // Compute LHS matrix address __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (get_global_id(1) % V0) * (uint)LHS_OFFSET_X * sizeof(DATA_TYPE) + (get_global_id(1) / V0) * (uint)lhs_stride_y + (get_global_id(2) * lhs_stride_z); // Compute RHS matrix address __global uchar *rhs_addr = rhs_ptr + rhs_offset_first_element_in_bytes + (get_global_id(0) % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (get_global_id(0) / (uint)H0) * rhs_stride_y; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 rhs_addr += (get_global_id(2) % MATRIX_B_DEPTH) * rhs_stride_z; #else // defined(MATRIX_B_DEPTH) rhs_addr += get_global_id(2) * rhs_stride_z; #endif // defined(MATRIX_B_DEPTH) // Initialize the accumulators REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), c, 0); //VEC_DATA_TYPE(DATA_TYPE, N0) c0=0,c1=0,c2=0,... c(M0-1)=0; REPEAT_VAR_INIT_TO_CONST(M0, uint, zlhs, 0); //uint zlhs0=0,zlhs1=0,zlhs2=0,... zlhs7=0; REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0); for(int i = 0; i < k; i += K0) { // Supported cases (M0, K0): // 1,2 - 1,3 - 1,4 - 1,8 - 1,16 // 2,2 - 2,3 - 2,4 - 2,8 - 2,16 // 3,2 - 3,3 - 3,4 - 3,8 - 3,16 // 4,2 - 4,3 - 4,4 - 4,8 - 4,16 // 5,2 - 5,3 - 5,4 - 5,8 - 5,16 // 6,2 - 6,3 - 6,4 - 6,8 - 6,16 // 7,2 - 7,3 - 7,4 - 7,8 - 7,16 // 8,2 - 8,3 - 8,4 - 8,8 - 8,16 // Load values from LHS matrix LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_addr, 0, LHS_STEP_X * sizeof(DATA_TYPE), zlhs); // Load values from RHS matrix LOAD_BLOCK(N0, K0, DATA_TYPE, b, rhs_addr, 0, RHS_STEP_X * sizeof(DATA_TYPE), zero); // Accumulate ARM_DOT_K0XN0(a0, b, c0); #if M0 > 1 ARM_DOT_K0XN0(a1, b, c1); #endif // M0 > 1 #if M0 > 2 ARM_DOT_K0XN0(a2, b, c2); #endif // M0 > 2 #if M0 > 3 ARM_DOT_K0XN0(a3, b, c3); #endif // M0 > 3 #if M0 > 4 ARM_DOT_K0XN0(a4, b, c4); #endif // M0 > 4 #if M0 > 5 ARM_DOT_K0XN0(a5, b, c5); #endif // M0 > 5 #if M0 > 6 ARM_DOT_K0XN0(a6, b, c6); #endif // M0 > 6 #if M0 > 7 ARM_DOT_K0XN0(a7, b, c7); #endif // M0 > 7 lhs_addr += (M0 * LHS_STEP_X * LHS_STEP_LOOP) * sizeof(DATA_TYPE); rhs_addr += (N0 * RHS_STEP_X * RHS_STEP_LOOP) * sizeof(DATA_TYPE); } __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * dst_stride_y); REPEAT_VAR_INIT_TO_CONST(M0, uint, zout, 0); #if defined(REINTERPRET_OUTPUT_AS_3D) // The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zout, get_global_id(1), HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += get_global_id(2) * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += get_global_id(2) * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) #if defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)); LOAD_BLOCK(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(M0, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * bias_stride_y) + get_global_id( 2) * bias_stride_z; LOAD_BLOCK(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(M0, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(M0, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store output block STORE_BLOCK(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout); #undef LHS_BLOCK_SIZE #undef LHS_OFFSET_X #undef LHS_STEP_X #undef RHS_BLOCK_SIZE #undef RHS_OFFSET_X #undef RHS_STEP_X } #if defined(LHS_TRANSPOSE) #define VTYPE(TYPE, SIZE) VEC_DATA_TYPE(TYPE, SIZE) #if GPU_ARCH == GPU_ARCH_MIDGARD #define ARM_VFMA(SIZE, a, b, c) c += (a) * (b); #else // GPU_ARCH == GPU_ARCH_MIDGARD #define ARM_VFMA_1(a, b, c) \ ({ \ c = fma((a), (b), (c)); \ }) #define ARM_VFMA_2(a, b, c) \ ({ \ (c).s0 = fma((a).s0, (b).s0, (c).s0); \ (c).s1 = fma((a).s1, (b).s1, (c).s1); \ }) #define ARM_VFMA_3(a, b, c) \ ({ \ ARM_VFMA_2(a, b, c); \ (c).s2 = fma((a).s2, (b).s2, (c).s2); \ }) #define ARM_VFMA_4(a, b, c) \ ({ \ ARM_VFMA_3(a, b, c); \ (c).s3 = fma((a).s3, (b).s3, (c).s3); \ }) #define ARM_VFMA_8(a, b, c) \ ({ \ ARM_VFMA_4(a, b, c); \ (c).s4 = fma((a).s4, (b).s4, (c).s4); \ (c).s5 = fma((a).s5, (b).s5, (c).s5); \ (c).s6 = fma((a).s6, (b).s6, (c).s6); \ (c).s7 = fma((a).s7, (b).s7, (c).s7); \ }) #define ARM_VFMA_16(a, b, c) \ ({ \ ARM_VFMA_8(a, b, c); \ (c).s8 = fma((a).s8, (b).s8, (c).s8); \ (c).s9 = fma((a).s9, (b).s9, (c).s9); \ (c).sA = fma((a).sA, (b).sA, (c).sA); \ (c).sB = fma((a).sB, (b).sB, (c).sB); \ (c).sC = fma((a).sC, (b).sC, (c).sC); \ (c).sD = fma((a).sD, (b).sD, (c).sD); \ (c).sE = fma((a).sE, (b).sE, (c).sE); \ (c).sF = fma((a).sF, (b).sF, (c).sF); \ }) // Factory macro for the vector FMA #define ARM_VFMA(SIZE, a, b, c) ARM_VFMA_##SIZE((a), (b), (c)) #endif // GPU_ARCH == GPU_ARCH_MIDGARD #define ARM_VVM_T_NT_1xN0x1(N0, TYPE, a, b, C) \ ({ \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a), b, (C##0)); \ }) #define ARM_VVM_T_NT_2xN0x1(N0, TYPE, a, b, C) \ ({ \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s0), b, (C##0)); \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s1), b, (C##1)); \ }) #define ARM_VVM_T_NT_3xN0x1(N0, TYPE, a, b, C) \ ({ \ ARM_VVM_T_NT_2xN0x1(N0, TYPE, a, b, C); \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s2), b, (C##2)); \ }) #define ARM_VVM_T_NT_4xN0x1(N0, TYPE, a, b, C) \ ({ \ ARM_VVM_T_NT_3xN0x1(N0, TYPE, a, b, C); \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s3), b, (C##3)); \ }) #define ARM_VVM_T_NT_8xN0x1(N0, TYPE, a, b, C) \ ({ \ ARM_VVM_T_NT_4xN0x1(N0, TYPE, a, b, C); \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s4), b, (C##4)); \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s5), b, (C##5)); \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s6), b, (C##6)); \ ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s7), b, (C##7)); \ }) // Factory macro for the column-vector (transposed) by row-vector (not transposed) multiplication. K0 = 1 // a is the column-vector (transposed) // b is the row-vector (not transposed) // C is the output matrix // Lower case is a vector (a, b) // Upper case is a matrix (C) #define ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, a, b, C) ARM_VVM_T_NT_##M0##xN0x1(N0, TYPE, a, b, C) #define ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, A, B, C) \ ({ \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##0), (B##0), C); \ }) #define ARM_MM_T_NT_M0xN0x2(M0, N0, TYPE, A, B, C) \ ({ \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, A, B, C); \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##1), (B##1), C); \ }) #define ARM_MM_T_NT_M0xN0x3(M0, N0, TYPE, A, B, C) \ ({ \ ARM_MM_T_NT_M0xN0x2(M0, N0, TYPE, A, B, C); \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##2), (B##2), C); \ }) #define ARM_MM_T_NT_M0xN0x4(M0, N0, TYPE, A, B, C) \ ({ \ ARM_MM_T_NT_M0xN0x3(M0, N0, TYPE, A, B, C); \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##3), (B##3), C); \ }) #define ARM_MM_T_NT_M0xN0x8(M0, N0, TYPE, A, B, C) \ ({ \ ARM_MM_T_NT_M0xN0x4(M0, N0, TYPE, A, B, C); \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##4), (B##4), C); \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##5), (B##5), C); \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##6), (B##6), C); \ ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##7), (B##7), C); \ }) #define ARM_MM_T_NT_M0xN0x16(M0, N0, TYPE, A, B, C) \ ({ \ ARM_MM_T_NT_M0xN0x8(M0, N0, TYPE, A, B, C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##8), (B##8), C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##9), (B##9), C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##A), (B##A), C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##B), (B##B), C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##C), (B##C), C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##D), (B##D), C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##E), (B##E), C); \ ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##F), (B##F), C); \ }) // Factory macro for the matrix (transposed) by matrix (not transposed) multiplication. // The dimensions for this matrix multiplications are defined through M0, N0 and K0 // The dimensions supported are: // M0: 1, 2, 3, 4, 8 // N0: 1, 2, 3, 4, 8, 16 // K0: 1, 2, 3, 4, 8, 16 // This macro calls the vector-by-matrix macro K0 times // A, B and C are matrices #define ARM_MM_T_NT(M0, N0, K0, TYPE, A, B, C) CONCAT(ARM_MM_T_NT_M0xN0x, K0) \ (M0, N0, TYPE, A, B, C) /** This OpenCL kernel computes the matrix multiplication between 2 matrices. * The LHS matrix must be reshaped with @ref CLGEMMReshapeLHSMatrixKernel and the M0xK0 must be transposed * The RHS matrix must be reshaped with @ref CLGEMMReshapeRHSMatrixKernel and the K0xN0 must be NOT transposed * * @note LHS_TRANSPOSE should be passed at compile time in order to compile this OpenCL kernel (e.g. -DLHS_TRANSPOSE). * @note If the first two dimensions of NDRange have been dispatched with "dummy_work_items" support, the option -DDUMMY_WORK_ITEMS must be passed at compile time. * @note The GEMM's dimensions M and N must be passed at compile time using -DM and -DN (e.g. -DM=52 and -DN=90). * @note The block's dimensions used for reshaping the LHS matrix and the RHS matrix (M0, N0 and K0) must be passed at compile time using -DM0, -DN0 and -DK0 (e.g. -DM0=4, -DN0=8, -DK0=4). * @note The number of M0xK0 vertical blocks stored on the same output row of the reshaped LHS matrix must be passed at compile time using -DV0 (e.g. -DV0=2) * @note The number of K0xN0 horizontal blocks stored on the same output row of the reshaped RHS matrix must be passed at compile time using -DH0 (e.g. -DH0=2) * @note If the M0xK0 blocks in the reshaped LHS matrix have been interleaved, the option -DLHS_INTERLEAVE must passed at compile time. * @note If the K0xN0 blocks in the reshaped RHS matrix have been interleaved, the option -DRHS_INTERLEAVE must passed at compile time. * @note Only the following configurations of M0, N0 and K0 are currently supported: * - M0 = 2, 3, 4, 8 * - N0 = 2, 3, 4, 8, 16 * - K0 = 2, 3, 4, 8, 16 * - V0 >= 1 * - H0 >= 1 * * @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 * @note In case the output has to be reinterpreted as a 3D tensor (e.g. output of convolution layer), the following information must be passed at compile time: * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns LHS matrix NOT reshaped * * @param[in] lhs_ptr Pointer to the LHS reshaped matrix. Supported data type: F16/F32 * @param[in] lhs_stride_x Stride of the LHS reshaped matrix in X dimension (in bytes) * @param[in] lhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] lhs_stride_y Stride of the LHS reshaped matrix in Y dimension (in bytes) * @param[in] lhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS reshaped matrix * @param[in] rhs_ptr Pointer to the RHS reshaped matrix. Supported data type: same as @p lhs_ptr * @param[in] rhs_stride_x Stride of the RHS reshaped matrix in X dimension (in bytes) * @param[in] rhs_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] rhs_stride_y Stride of the RHS reshaped matrix in Y dimension (in bytes) * @param[in] rhs_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS reshaped matrix * @param[in] bias_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] bias_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] bias_step_x (Optional) bias_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] bias_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] bias_step_y (Optional) bias_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] bias_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data type: same as @p lhs_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] k Number of columns in LHS matrix and rows in RHS matrix not reshaped. * @param[in] lhs_stride_z Stride of the LHS reshaped matrix in Z dimension (in bytes) * @param[in] rhs_stride_z Stride of the RHS reshaped matrix in Z dimension (in bytes) * @param[in] bias_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_reshaped_lhs_t_rhs_nt(IMAGE_DECLARATION(lhs), IMAGE_DECLARATION(rhs), #if defined(BETA) IMAGE_DECLARATION(bias), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint k, uint lhs_stride_z, uint rhs_stride_z, #if defined(BETA) uint bias_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { // Block size #define LHS_BLOCK_SIZE ((K0) * (M0)) #if defined(LHS_INTERLEAVE) #define LHS_OFFSET_X (M0) #define LHS_STEP_X ((M0) * (V0)) #define LHS_STEP_LOOP (1) #else // defined(INTERLEAVE) #define LHS_OFFSET_X (LHS_BLOCK_SIZE) #define LHS_STEP_X (M0) #define LHS_STEP_LOOP (V0) #endif // defined(INTERLEAVE) // Block size #define RHS_BLOCK_SIZE ((K0) * (N0)) // RHS offset and step X #if defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (N0) #define RHS_STEP_X ((N0) * (H0)) #define RHS_STEP_LOOP (1) #else // defined(RHS_INTERLEAVE) #define RHS_OFFSET_X (RHS_BLOCK_SIZE) #define RHS_STEP_X (N0) #define RHS_STEP_LOOP (H0) #endif // defined(RHS_INTERLEAVE) const uint x = get_global_id(0); const uint y = get_global_id(1); const uint z = get_global_id(2); #if defined(DUMMY_WORK_ITEMS) if((x * N0 >= N) || (y * M0 >= M)) { return; } #endif // defined(DUMMY_WORK_ITEMS) // Compute LHS matrix address __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (y % V0) * (uint)LHS_OFFSET_X * sizeof(DATA_TYPE) + (y / V0) * (uint)lhs_stride_y + (z * lhs_stride_z); // Compute RHS matrix address __global uchar *rhs_addr = rhs_ptr + rhs_offset_first_element_in_bytes + (x % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (x / (uint)H0) * rhs_stride_y; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 rhs_addr += (z % MATRIX_B_DEPTH) * rhs_stride_z; #else // defined(MATRIX_B_DEPTH) rhs_addr += z * rhs_stride_z; #endif // defined(MATRIX_B_DEPTH) // Initialize the accumulators REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), c, 0); //VEC_DATA_TYPE(DATA_TYPE, N0) c0=0,c1=0,c2=0,... c(M0-1)=0; REPEAT_VAR_INIT_TO_CONST(K0, uint, zlhs, 0); //uint zlhs0=0,zlhs1=0,zlhs2=0,... zlhs7=0; REPEAT_VAR_INIT_TO_CONST(M0, uint, zero, 0); for(int i = 0; i < k; i += K0) { // Supported cases (K0, M0): // 1,2 - 2,2 - 3,2 - 4,2 - 5,2 - 6,2 - 7,2 - 8,2 // 1,3 - 2,3 - 3,3 - 4,3 - 5,3 - 6,3 - 7,3 - 8,3 // 1,4 - 2,4 - 3,4 - 4,4 - 5,4 - 6,4 - 7,4 - 8,4 // 1,8 - 2,8 - 3,8 - 4,8 - 5,8 - 6,8 - 7,8 - 8,8 // 1,16 - 2,16 - 3,16 - 4,16 - 5,16 - 6,16 - 7,16 - 8,16 // Load values from LHS matrix LOAD_BLOCK(K0, M0, DATA_TYPE, a, lhs_addr, 0, LHS_STEP_X * sizeof(DATA_TYPE), zlhs); // Load values from RHS matrix LOAD_BLOCK(K0, N0, DATA_TYPE, b, rhs_addr, 0, RHS_STEP_X * sizeof(DATA_TYPE), zlhs); // Perform the partial matrix multiplication ARM_MM_T_NT(M0, N0, K0, DATA_TYPE, a, b, c); lhs_addr += (K0 * LHS_STEP_X * LHS_STEP_LOOP) * sizeof(DATA_TYPE); rhs_addr += (K0 * RHS_STEP_X * RHS_STEP_LOOP) * sizeof(DATA_TYPE); } __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)M0 * dst_stride_y); REPEAT_VAR_INIT_TO_CONST(M0, uint, zout, 0); #if defined(REINTERPRET_OUTPUT_AS_3D) // The plane (zin) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zout, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) #if defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)); LOAD_BLOCK(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(M0, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)M0 * bias_stride_y) + z * bias_stride_z; LOAD_BLOCK(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(M0, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(M0, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store output block STORE_BLOCK(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout); #undef LHS_BLOCK_SIZE #undef LHS_OFFSET_X #undef LHS_STEP_X #undef RHS_BLOCK_SIZE #undef RHS_OFFSET_X #undef RHS_STEP_X } #endif // defined(LHS_TRANSPOSE) #endif // defined(M0) && defined(N0) && defined(K0) && defined(V0) && defined(H0) && defined(K) && defined(DATA_TYPE) #if defined(M0) && defined(N0) && defined(K0) && defined(K) && defined(DATA_TYPE) #define VFMA(a, b, c) \ ({ \ c = fma(a, b, c); \ }) #if M0 == 1 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ }) #elif M0 == 2 // M0 == 2 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ }) #elif M0 == 3 // M0 == 3 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ }) #elif M0 == 4 // M0 == 4 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ }) #elif M0 == 5 // M0 == 5 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ }) #elif M0 == 6 // M0 == 6 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \ }) #elif M0 == 7 // M0 == 7 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \ }) #elif M0 == 8 // M0 == 8 #define RHS_VFMA_M0xN0(i, a, b, c) \ ({ \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \ VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##7).s##i), b, (c##7)); \ }) #else // M0 not supported #error "M0 not supported" #endif // M0 not supported /** This OpenCL kernel computes the matrix multiplication between 2 matrices. * The LHS matrix is NOT reshaped * The RHS matrix is NOT reshaped * * @note If the first two dimensions of NDRange have been dispatched with "dummy_work_items" support, the option -DDUMMY_WORK_ITEMS must be passed at compile time. * @note The GEMM's dimensions (M,N and K) must be passed at compile time using -DM, -DN and and -DK (e.g. -DM=52, -DN=30 and -DK=90) * @note The number of columns of LHS matrix must be passed at compile time using -DK (e.g. -DK=64) * @note The number of M0 rows to process must be passed at compile time using -DM0 (e.g. -DM0=2) * @note The number of K0 partial accumulations must be passed at compile time using -DK0 (e.g., -DK0=2) * @note The number of N0 columns to process must be passed at compile time using -DN0 (e.g. -DN0=2) * @note Only the following configurations of M0, N0 and K0 are currently supported: * - M0 = 1, 2, 3, 4, 5, 6, 7, 8 * - N0 = 2, 3, 4, 8, 16 * - K0 = 2, 3, 4, 8, 16 * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns LHS matrix * * @param[in] lhs_ptr Pointer to the LHS matrix. Supported data type: F16/F32 * @param[in] lhs_stride_x Stride of the LHS matrix in X dimension (in bytes) * @param[in] lhs_step_x lhs_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] lhs_stride_y Stride of the LHS matrix in Y dimension (in bytes) * @param[in] lhs_step_y lhs_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS matrix * @param[in] rhs_ptr Pointer to the RHS matrix. Supported data type: same as @p lhs_ptr * @param[in] rhs_stride_x Stride of the RHS matrix in X dimension (in bytes) * @param[in] rhs_step_x rhs_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] rhs_stride_y Stride of the RHS matrix in Y dimension (in bytes) * @param[in] rhs_step_y rhs_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS matrix * @param[in] bias_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] bias_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] bias_step_x (Optional) bias_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] bias_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] bias_step_y (Optional) bias_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] bias_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data type: same as @p lhs_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] lhs_stride_z Stride of the LHS matrix in Z dimension (in bytes) * @param[in] rhs_stride_z Stride of the RHS matrix in Z dimension (in bytes) * @param[in] bias_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] lhs_cross_plane_pad (Optional) Bottom paddings for LHS matrix in unit of elements (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings for the output matrix in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_native(IMAGE_DECLARATION(lhs), IMAGE_DECLARATION(rhs), #if defined(BETA) IMAGE_DECLARATION(bias), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint lhs_stride_z, uint rhs_stride_z, #if defined(BETA) uint bias_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint lhs_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { // Block size #define RHS_BLOCK_SIZE ((K0) * (N0)) // RHS offset and step X #define RHS_OFFSET_X (RHS_BLOCK_SIZE) uint x = get_global_id(0); uint y = get_global_id(1); uint z = get_global_id(2); #if defined(DUMMY_WORK_ITEMS) if((x * N0 >= N) || (y * M0 >= M)) { return; } #endif // defined(DUMMY_WORK_ITEMS) // Compute LHS matrix address uint lhs_offset = lhs_offset_first_element_in_bytes + y * M0 * (uint)lhs_stride_y; // Compute RHS matrix address uint rhs_offset = rhs_offset_first_element_in_bytes + x * N0 * sizeof(DATA_TYPE); #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 rhs_offset += (z % MATRIX_B_DEPTH) * rhs_stride_z; #else // defined(MATRIX_B_DEPTH) rhs_offset += z * rhs_stride_z; #endif // defined(MATRIX_B_DEPTH) REPEAT_VAR_INIT_TO_CONST(M0, uint, zlhs, 0); REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0); #if defined(REINTERPRET_INPUT_AS_3D) // The plane (zlhs) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zlhs, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, lhs_cross_plane_pad, lhs_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply lhs_stride_z by DEPTH_GEMM3D lhs_offset += z * lhs_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM lhs_offset += z * lhs_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) // Initialize the accumulators REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), c, 0); //VEC_DATA_TYPE(DATA_TYPE, N0) c0=0,c1=0,c2=0,... c(M0-1)=0; int i = 0; for(; i <= (K - K0); i += K0) { // Supported cases (M0, K0): // 1,2 - 1,3 - 1,4 - 1,8 - 1,16 // 2,2 - 2,3 - 2,4 - 2,8 - 2,16 // 3,2 - 3,3 - 3,4 - 3,8 - 3,16 // 4,2 - 4,3 - 4,4 - 4,8 - 4,16 // 5,2 - 5,3 - 5,4 - 5,8 - 5,16 // 6,2 - 6,3 - 6,4 - 6,8 - 6,16 // 7,2 - 7,3 - 7,4 - 7,8 - 7,16 // 8,2 - 8,3 - 8,4 - 8,8 - 8,16 // Load values from LHS matrix LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs); // Load values from RHS matrix LOAD_BLOCK(K0, N0, DATA_TYPE, b, rhs_ptr, rhs_offset, rhs_stride_y, zero); RHS_VFMA_M0xN0(0, a, b0, c); RHS_VFMA_M0xN0(1, a, b1, c); #if K0 > 2 RHS_VFMA_M0xN0(2, a, b2, c); #endif // K0 > 2 #if K0 > 3 RHS_VFMA_M0xN0(3, a, b3, c); #endif // K0 > 3 #if K0 > 4 RHS_VFMA_M0xN0(4, a, b4, c); RHS_VFMA_M0xN0(5, a, b5, c); RHS_VFMA_M0xN0(6, a, b6, c); RHS_VFMA_M0xN0(7, a, b7, c); #endif // K0 > 4 #if K0 > 8 RHS_VFMA_M0xN0(8, a, b8, c); RHS_VFMA_M0xN0(9, a, b9, c); RHS_VFMA_M0xN0(A, a, b10, c); RHS_VFMA_M0xN0(B, a, b11, c); RHS_VFMA_M0xN0(C, a, b12, c); RHS_VFMA_M0xN0(D, a, b13, c); RHS_VFMA_M0xN0(E, a, b14, c); RHS_VFMA_M0xN0(F, a, b15, c); #endif // K0 > 8 lhs_offset += K0 * sizeof(DATA_TYPE); rhs_offset += K0 * rhs_stride_y; } // Left-over accumulations for(; i < K; ++i) { // Load values from LHS matrix VEC_DATA_TYPE(DATA_TYPE, 2) a0 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 0 * lhs_stride_y + zlhs0)); #if M0 > 1 VEC_DATA_TYPE(DATA_TYPE, 2) a1 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 1 * lhs_stride_y + zlhs1)); #endif // M0 > 1 #if M0 > 2 VEC_DATA_TYPE(DATA_TYPE, 2) a2 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 2 * lhs_stride_y + zlhs2)); #endif // M0 > 2 #if M0 > 3 VEC_DATA_TYPE(DATA_TYPE, 2) a3 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 3 * lhs_stride_y + zlhs3)); #endif // M0 > 3 #if M0 > 4 VEC_DATA_TYPE(DATA_TYPE, 2) a4 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 4 * lhs_stride_y + zlhs4)); #endif // M0 > 4 #if M0 > 5 VEC_DATA_TYPE(DATA_TYPE, 2) a5 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 5 * lhs_stride_y + zlhs5)); #endif // M0 > 5 #if M0 > 6 VEC_DATA_TYPE(DATA_TYPE, 2) a6 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 6 * lhs_stride_y + zlhs6)); #endif // M0 > 6 #if M0 > 7 VEC_DATA_TYPE(DATA_TYPE, 2) a7 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zlhs7)); #endif // M0 > 7 VEC_DATA_TYPE(DATA_TYPE, N0) b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0 * rhs_stride_y)); RHS_VFMA_M0xN0(0, a, b, c); lhs_offset += sizeof(DATA_TYPE); rhs_offset += rhs_stride_y; } __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)M0 * dst_stride_y); REPEAT_VAR_INIT_TO_CONST(M0, uint, zout, 0); #if defined(REINTERPRET_OUTPUT_AS_3D) // The plane (zout) is calculated dividing M (y * M0) by HEIGHT_GEMM3D CALCULATE_Z_OFFSET(M0, uint, zout, y, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) #if defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)); LOAD_BLOCK(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(M0, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * bias_stride_y) + get_global_id( 2) * bias_stride_z; LOAD_BLOCK(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(M0, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(M0, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store output block STORE_BLOCK(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout); #undef RHS_BLOCK_SIZE #undef RHS_OFFSET_X #undef RHS_STEP_X } #endif // defined(M0) && defined(N0) && defined(K0) && defined(K) && defined(DATA_TYPE) #if defined(COLS_B) && defined(MULT_TRANSPOSE1XW_WIDTH) && defined(MULT_INTERLEAVE4X4_HEIGHT) /** This OpenCL kernel is optimised for Midgard. It computes the matrix multiplication between matrix A reshaped (src0) and matrix B reshaped (src1) * * @note The number of columns of matrix B and the optional alpha's value need to be passed at compile time using -DCOLS_B and -DALPHA * @note The multiplication factor for the transposition width (mult_transpose1xW_width) must be passed at compile time using -DMULT_TRANSPOSE1XW_WIDTH (e.g. -DMULT_TRANSPOSE1XW_WIDTH=2) * @note The multiplication factor for the height of the 4x4 interleaved block must be passed at compile time using -DMULT_INTERLEAVE4X4_HEIGHT (e.g. -DMULT_INTERLEAVE4X4_HEIGHT=2) * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the output has to be reinterpreted as a 3D tensor (e.g. output of convolution layer), the following information must be passed at compile time: * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F32 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_interleaved_transposed_f32(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_OUTPUT_AS_3D) , uint cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH; int y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT; int z = get_global_id(2); // Offset const int offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4; const int offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 4; // src_addr_a = address of matrix A // src_addr_b = address of matrix B int src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes; int src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src1_addr_in_bytes += z * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) __global float *src_addr_a = (__global float *)(src0_ptr + src0_addr_in_bytes); __global float *src_addr_b = (__global float *)(src1_ptr + src1_addr_in_bytes); // Compute end row address for matrix B __global float *src_end_addr_b = src_addr_b + COLS_B; src_addr_a += offset_row_a; src_addr_b += offset_row_b; // Reset accumulators float4 c0 = 0.0f; float4 c1 = 0.0f; float4 c2 = 0.0f; float4 c3 = 0.0f; for(; src_addr_b <= (src_end_addr_b - (int)(8 * MULT_TRANSPOSE1XW_WIDTH)); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH) { // Load values from matrix A (interleaved) and matrix B (transposed) float4 a0 = vload4(0, src_addr_a); float4 b0 = vload4(0, src_addr_b); c0 += (float4)a0.s0 * b0; c1 += (float4)a0.s1 * b0; c2 += (float4)a0.s2 * b0; c3 += (float4)a0.s3 * b0; // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a + 4 * MULT_INTERLEAVE4X4_HEIGHT); b0 = vload4(0, src_addr_b + 4 * MULT_TRANSPOSE1XW_WIDTH); c0 += (float4)a0.s0 * b0; c1 += (float4)a0.s1 * b0; c2 += (float4)a0.s2 * b0; c3 += (float4)a0.s3 * b0; } for(; src_addr_b < src_end_addr_b; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH) { // Load values from matrix A (interleaved) and matrix B (transposed) float4 a0 = vload4(0, src_addr_a); float4 b0 = vload4(0, src_addr_b); c0 += (float4)a0.s0 * b0; c1 += (float4)a0.s1 * b0; c2 += (float4)a0.s2 * b0; c3 += (float4)a0.s3 * b0; } // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(4, float, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(4, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)4 * sizeof(float)); LOAD_BLOCK(1, 4, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, float, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(4, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)4 * sizeof(float)) + (get_global_id(1) * (uint)4 * src2_stride_y) + get_global_id( 2) * src2_stride_z; LOAD_BLOCK(4, 4, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(4, float, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(4, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(4, ACTIVATION_TYPE, float, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store 4x4 block vstore4(c0, 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0)); vstore4(c1, 0, (__global float *)(dst_addr + 1 * dst_stride_y + zout.s1)); vstore4(c2, 0, (__global float *)(dst_addr + 2 * dst_stride_y + zout.s2)); vstore4(c3, 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3)); } /** This OpenCL kernel is optimized for Bifrost and tt computes the matrix multiplication between matrix A reshaped (src0) and matrix B reshaped (src1) * * @note The number of columns of matrix B and the optional alpha's value need to be passed at compile time using -DCOLS_B and -DALPHA * @note The multiplication factor for the transposition width (mult_transpose1xW_width) must be passed at compile time using -DMULT_TRANSPOSE1XW_WIDTH (e.g. -DMULT_TRANSPOSE1XW_WIDTH=2) * @note The multiplication factor for the height of the 4x4 interleaved block must be passed at compile time using -DMULT_INTERLEAVE4X4_HEIGHT (e.g. -DMULT_INTERLEAVE4X4_HEIGHT=2) * @note The multiplication factor for the height of the 4x4 interleaved block must be passed at compile time using -DMULT_INTERLEAVE4X4_HEIGHT (e.g. -DMULT_INTERLEAVE4X4_HEIGHT=2) * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the output has to be reinterpreted as a 3D tensor (e.g. output of convolution layer), the following information must be passed at compile time: * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F32 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_interleaved_transposed_f32_bifrost(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_OUTPUT_AS_3D) , uint cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH; int y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT; int z = get_global_id(2); // Offset const int offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4; const int offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 4; // src_addr_a = address of matrix A // src_addr_b = address of matrix B int src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes; int src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src1_addr_in_bytes += z * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) __global float *src_addr_a = (__global float *)(src0_ptr + src0_addr_in_bytes); __global float *src_addr_b = (__global float *)(src1_ptr + src1_addr_in_bytes); src_addr_a += offset_row_a; src_addr_b += offset_row_b; // Reset accumulators float4 c0 = 0.0f; float4 c1 = 0.0f; float4 c2 = 0.0f; float4 c3 = 0.0f; #define COLS_MTX_B (COLS_B / (4 * MULT_TRANSPOSE1XW_WIDTH)) int i = 0; for(; i <= (int)(COLS_MTX_B - 4); i += 4) { // Load values from matrix A (interleaved) and matrix B (transposed) float4 a0 = vload4(0, src_addr_a); float4 b0 = vload4(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH; c0.s0 = fma(a0.s0, b0.s0, c0.s0); c0.s1 = fma(a0.s0, b0.s1, c0.s1); c0.s2 = fma(a0.s0, b0.s2, c0.s2); c0.s3 = fma(a0.s0, b0.s3, c0.s3); c1.s0 = fma(a0.s1, b0.s0, c1.s0); c1.s1 = fma(a0.s1, b0.s1, c1.s1); c1.s2 = fma(a0.s1, b0.s2, c1.s2); c1.s3 = fma(a0.s1, b0.s3, c1.s3); c2.s0 = fma(a0.s2, b0.s0, c2.s0); c2.s1 = fma(a0.s2, b0.s1, c2.s1); c2.s2 = fma(a0.s2, b0.s2, c2.s2); c2.s3 = fma(a0.s2, b0.s3, c2.s3); c3.s0 = fma(a0.s3, b0.s0, c3.s0); c3.s1 = fma(a0.s3, b0.s1, c3.s1); c3.s2 = fma(a0.s3, b0.s2, c3.s2); c3.s3 = fma(a0.s3, b0.s3, c3.s3); // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a); b0 = vload4(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH; c0.s0 = fma(a0.s0, b0.s0, c0.s0); c0.s1 = fma(a0.s0, b0.s1, c0.s1); c0.s2 = fma(a0.s0, b0.s2, c0.s2); c0.s3 = fma(a0.s0, b0.s3, c0.s3); c1.s0 = fma(a0.s1, b0.s0, c1.s0); c1.s1 = fma(a0.s1, b0.s1, c1.s1); c1.s2 = fma(a0.s1, b0.s2, c1.s2); c1.s3 = fma(a0.s1, b0.s3, c1.s3); c2.s0 = fma(a0.s2, b0.s0, c2.s0); c2.s1 = fma(a0.s2, b0.s1, c2.s1); c2.s2 = fma(a0.s2, b0.s2, c2.s2); c2.s3 = fma(a0.s2, b0.s3, c2.s3); c3.s0 = fma(a0.s3, b0.s0, c3.s0); c3.s1 = fma(a0.s3, b0.s1, c3.s1); c3.s2 = fma(a0.s3, b0.s2, c3.s2); c3.s3 = fma(a0.s3, b0.s3, c3.s3); // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a); b0 = vload4(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH; c0.s0 = fma(a0.s0, b0.s0, c0.s0); c0.s1 = fma(a0.s0, b0.s1, c0.s1); c0.s2 = fma(a0.s0, b0.s2, c0.s2); c0.s3 = fma(a0.s0, b0.s3, c0.s3); c1.s0 = fma(a0.s1, b0.s0, c1.s0); c1.s1 = fma(a0.s1, b0.s1, c1.s1); c1.s2 = fma(a0.s1, b0.s2, c1.s2); c1.s3 = fma(a0.s1, b0.s3, c1.s3); c2.s0 = fma(a0.s2, b0.s0, c2.s0); c2.s1 = fma(a0.s2, b0.s1, c2.s1); c2.s2 = fma(a0.s2, b0.s2, c2.s2); c2.s3 = fma(a0.s2, b0.s3, c2.s3); c3.s0 = fma(a0.s3, b0.s0, c3.s0); c3.s1 = fma(a0.s3, b0.s1, c3.s1); c3.s2 = fma(a0.s3, b0.s2, c3.s2); c3.s3 = fma(a0.s3, b0.s3, c3.s3); // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a); b0 = vload4(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH; c0.s0 = fma(a0.s0, b0.s0, c0.s0); c0.s1 = fma(a0.s0, b0.s1, c0.s1); c0.s2 = fma(a0.s0, b0.s2, c0.s2); c0.s3 = fma(a0.s0, b0.s3, c0.s3); c1.s0 = fma(a0.s1, b0.s0, c1.s0); c1.s1 = fma(a0.s1, b0.s1, c1.s1); c1.s2 = fma(a0.s1, b0.s2, c1.s2); c1.s3 = fma(a0.s1, b0.s3, c1.s3); c2.s0 = fma(a0.s2, b0.s0, c2.s0); c2.s1 = fma(a0.s2, b0.s1, c2.s1); c2.s2 = fma(a0.s2, b0.s2, c2.s2); c2.s3 = fma(a0.s2, b0.s3, c2.s3); c3.s0 = fma(a0.s3, b0.s0, c3.s0); c3.s1 = fma(a0.s3, b0.s1, c3.s1); c3.s2 = fma(a0.s3, b0.s2, c3.s2); c3.s3 = fma(a0.s3, b0.s3, c3.s3); } for(; i < (int)(COLS_MTX_B); ++i) { // Load values from matrix A (interleaved) and matrix B (transposed) float4 a0 = vload4(0, src_addr_a); float4 b0 = vload4(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 4 * MULT_TRANSPOSE1XW_WIDTH; c0.s0 = fma(a0.s0, b0.s0, c0.s0); c0.s1 = fma(a0.s0, b0.s1, c0.s1); c0.s2 = fma(a0.s0, b0.s2, c0.s2); c0.s3 = fma(a0.s0, b0.s3, c0.s3); c1.s0 = fma(a0.s1, b0.s0, c1.s0); c1.s1 = fma(a0.s1, b0.s1, c1.s1); c1.s2 = fma(a0.s1, b0.s2, c1.s2); c1.s3 = fma(a0.s1, b0.s3, c1.s3); c2.s0 = fma(a0.s2, b0.s0, c2.s0); c2.s1 = fma(a0.s2, b0.s1, c2.s1); c2.s2 = fma(a0.s2, b0.s2, c2.s2); c2.s3 = fma(a0.s2, b0.s3, c2.s3); c3.s0 = fma(a0.s3, b0.s0, c3.s0); c3.s1 = fma(a0.s3, b0.s1, c3.s1); c3.s2 = fma(a0.s3, b0.s2, c3.s2); c3.s3 = fma(a0.s3, b0.s3, c3.s3); } // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(4, float, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(4, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)4 * sizeof(float)); LOAD_BLOCK(1, 4, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, float, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(4, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)4 * sizeof(float)) + (get_global_id(1) * (uint)4 * src2_stride_y) + get_global_id( 2) * src2_stride_z; LOAD_BLOCK(4, 4, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(4, float, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(4, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(4, ACTIVATION_TYPE, float, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store 4x4 block vstore4(c0, 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0)); vstore4(c1, 0, (__global float *)(dst_addr + 1 * dst_stride_y + zout.s1)); vstore4(c2, 0, (__global float *)(dst_addr + 2 * dst_stride_y + zout.s2)); vstore4(c3, 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3)); } // Undefine local defines #undef COLS_MTX_B #if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) /** This OpenCL kernel computes the matrix multiplication between matrix A reshaped (src0) and matrix B reshaped (src1) * * @note The number of columns of matrix B and the optional alpha's value need to be passed at compile time using -DCOLS_B and -DALPHA * @note The multiplication factor for the transposition width (mult_transpose1xW_width) must be passed at compile time using -DMULT_TRANSPOSE1XW_WIDTH (e.g. -DMULT_TRANSPOSE1XW_WIDTH=2) * @note The multiplication factor for the height of the 4x4 interleaved block must be passed at compile time using -DMULT_INTERLEAVE4X4_HEIGHT (e.g. -DMULT_INTERLEAVE4X4_HEIGHT=2) * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the output has to be reinterpreted as a 3D tensor (e.g. output of convolution layer), the following information must be passed at compile time: * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F16 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_interleaved_transposed_f16(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_OUTPUT_AS_3D) , uint cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH; int y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT; int z = get_global_id(2); // Offset const int offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4; const int offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 8; // src_addr_a = address of matrix A // src_addr_b = address of matrix B int src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes; int src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src1_addr_in_bytes += z * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) __global half *src_addr_a = (__global half *)(src0_ptr + src0_addr_in_bytes); __global half *src_addr_b = (__global half *)(src1_ptr + src1_addr_in_bytes); // Compute end row address for matrix B __global half *src_end_addr_b = src_addr_b + COLS_B; src_addr_a += offset_row_a; src_addr_b += offset_row_b; // Reset accumulators half8 c0 = 0.0f; half8 c1 = 0.0f; half8 c2 = 0.0f; half8 c3 = 0.0f; for(; src_addr_b <= (src_end_addr_b - (int)(16 * MULT_TRANSPOSE1XW_WIDTH)); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 16 * MULT_TRANSPOSE1XW_WIDTH) { // Load values from matrix A (interleaved) and matrix B (transposed) half4 a0 = vload4(0, src_addr_a); half8 b0 = vload8(0, src_addr_b); c0 += (half8)a0.s0 * b0; c1 += (half8)a0.s1 * b0; c2 += (half8)a0.s2 * b0; c3 += (half8)a0.s3 * b0; // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a + 4 * MULT_INTERLEAVE4X4_HEIGHT); b0 = vload8(0, src_addr_b + 8 * MULT_TRANSPOSE1XW_WIDTH); c0 += (half8)a0.s0 * b0; c1 += (half8)a0.s1 * b0; c2 += (half8)a0.s2 * b0; c3 += (half8)a0.s3 * b0; } for(; src_addr_b < src_end_addr_b; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH) { // Load values from matrix A (interleaved) and matrix B (transposed) half4 a0 = vload4(0, src_addr_a); half8 b0 = vload8(0, src_addr_b); c0 += (half8)a0.s0 * b0; c1 += (half8)a0.s1 * b0; c2 += (half8)a0.s2 * b0; c3 += (half8)a0.s3 * b0; } // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(4, half, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(4, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)); LOAD_BLOCK(1, 8, half, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, half, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(4, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)) + (get_global_id(1) * (uint)4 * src2_stride_y) + get_global_id( 2) * src2_stride_z; LOAD_BLOCK(4, 8, half, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(4, half, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(4, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(4, ACTIVATION_TYPE, half, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store 4x8 block vstore8(c0, 0, (__global half *)(dst_addr + 0 * dst_stride_y + zout.s0)); vstore8(c1, 0, (__global half *)(dst_addr + 1 * dst_stride_y + zout.s1)); vstore8(c2, 0, (__global half *)(dst_addr + 2 * dst_stride_y + zout.s2)); vstore8(c3, 0, (__global half *)(dst_addr + 3 * dst_stride_y + zout.s3)); } /** This OpenCL kernel computes the matrix multiplication between matrix A reshaped (src0) and matrix B reshaped (src1) while accumulating the result in a 32 floating point variable. * * @note The number of columns of matrix B and the optional alpha's value need to be passed at compile time using -DCOLS_B and -DALPHA * @note The multiplication factor for the transposition width (mult_transpose1xW_width) must be passed at compile time using -DMULT_TRANSPOSE1XW_WIDTH (e.g. -DMULT_TRANSPOSE1XW_WIDTH=2) * @note The multiplication factor for the height of the 4x4 interleaved block must be passed at compile time using -DMULT_INTERLEAVE4X4_HEIGHT (e.g. -DMULT_INTERLEAVE4X4_HEIGHT=2) * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the output has to be reinterpreted as a 3D tensor (e.g. output of convolution layer), the following information must be passed at compile time: * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F16 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_interleaved_transposed_f16_acc32(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_OUTPUT_AS_3D) , uint cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH; int y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT; int z = get_global_id(2); // Offset const int offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4; const int offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 8; // src_addr_a = address of matrix A // src_addr_b = address of matrix B int src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes; int src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src1_addr_in_bytes += z * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) __global half *src_addr_a = (__global half *)(src0_ptr + src0_addr_in_bytes); __global half *src_addr_b = (__global half *)(src1_ptr + src1_addr_in_bytes); // Compute end row address for matrix B __global half *src_end_addr_b = src_addr_b + COLS_B; src_addr_a += offset_row_a; src_addr_b += offset_row_b; // Reset accumulators float8 c0 = 0.0f; float8 c1 = 0.0f; float8 c2 = 0.0f; float8 c3 = 0.0f; for(; src_addr_b <= (src_end_addr_b - (int)(16 * MULT_TRANSPOSE1XW_WIDTH)); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 16 * MULT_TRANSPOSE1XW_WIDTH) { // Load values from matrix A (interleaved) and matrix B (transposed) float4 a0 = convert_float4(vload4(0, src_addr_a)); float8 b0 = convert_float8(vload8(0, src_addr_b)); c0 += (float8)a0.s0 * b0; c1 += (float8)a0.s1 * b0; c2 += (float8)a0.s2 * b0; c3 += (float8)a0.s3 * b0; // Load values from matrix A (interleaved) and matrix B (transposed) a0 = convert_float4(vload4(0, src_addr_a + 4 * MULT_INTERLEAVE4X4_HEIGHT)); b0 = convert_float8(vload8(0, src_addr_b + 8 * MULT_TRANSPOSE1XW_WIDTH)); c0 += (float8)a0.s0 * b0; c1 += (float8)a0.s1 * b0; c2 += (float8)a0.s2 * b0; c3 += (float8)a0.s3 * b0; } for(; src_addr_b < src_end_addr_b; src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT, src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH) { // Load values from matrix A (interleaved) and matrix B (transposed) float4 a0 = convert_float4(vload4(0, src_addr_a)); float8 b0 = convert_float8(vload8(0, src_addr_b)); c0 += (float8)a0.s0 * b0; c1 += (float8)a0.s1 * b0; c2 += (float8)a0.s2 * b0; c3 += (float8)a0.s3 * b0; } // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(4, float, c, ALPHA); #endif // defined(ALPHA) #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(4, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)); LOAD_BLOCK(1, 8, half, bias, src2_addr, 0, src2_stride_y, zero); float8 bias_f0 = convert_float8(bias0); #ifndef UNIT_BETA SCALE_BLOCK(1, float, bias_f, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(4, c, bias_f0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)) + (get_global_id(1) * (uint)4 * src2_stride_y) + get_global_id( 2) * src2_stride_z; LOAD_BLOCK(4, 8, half, bias, src2_addr, 0, src2_stride_y, zero); float8 bias_f0 = convert_float8(bias0); float8 bias_f1 = convert_float8(bias1); float8 bias_f2 = convert_float8(bias2); float8 bias_f3 = convert_float8(bias3); #ifndef UNIT_BETA SCALE_BLOCK(4, float, bias_f, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(4, c, bias_f); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) half8 c_h0 = convert_half8(c0); half8 c_h1 = convert_half8(c1); half8 c_h2 = convert_half8(c2); half8 c_h3 = convert_half8(c3); #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(4, ACTIVATION_TYPE, half, c_h, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store 4x8 block vstore8(c_h0, 0, (__global half *)(dst_addr + 0 * dst_stride_y + zout.s0)); vstore8(c_h1, 0, (__global half *)(dst_addr + 1 * dst_stride_y + zout.s1)); vstore8(c_h2, 0, (__global half *)(dst_addr + 2 * dst_stride_y + zout.s2)); vstore8(c_h3, 0, (__global half *)(dst_addr + 3 * dst_stride_y + zout.s3)); } /** This OpenCL kernel optimized for Bifrost architectures computes the matrix multiplication between matrix A reshaped (src0) and matrix B reshaped (src1) * * @note The number of columns of matrix B and the optional alpha's value need to be passed at compile time using -DCOLS_B and -DALPHA * @note The multiplication factor for the transposition width (mult_transpose1xW_width) must be passed at compile time using -DMULT_TRANSPOSE1XW_WIDTH (e.g. -DMULT_TRANSPOSE1XW_WIDTH=2) * @note The multiplication factor for the height of the 4x4 interleaved block must be passed at compile time using -DMULT_INTERLEAVE4X4_HEIGHT (e.g. -DMULT_INTERLEAVE4X4_HEIGHT=2) * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the output has to be reinterpreted as a 3D tensor (e.g. output of convolution layer), the following information must be passed at compile time: * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F16 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_interleaved_transposed_f16_bifrost(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_OUTPUT_AS_3D) , uint cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int x = get_global_id(0) / MULT_TRANSPOSE1XW_WIDTH; int y = get_global_id(1) / MULT_INTERLEAVE4X4_HEIGHT; int z = get_global_id(2); // Offset const int offset_row_a = (get_global_id(1) % MULT_INTERLEAVE4X4_HEIGHT) * 4; const int offset_row_b = (get_global_id(0) % MULT_TRANSPOSE1XW_WIDTH) * 8; // src_addr_a = address of matrix A // src_addr_b = address of matrix B int src0_addr_in_bytes = z * src0_stride_z + y * src0_stride_y + src0_offset_first_element_in_bytes; int src1_addr_in_bytes = x * src1_stride_y + src1_offset_first_element_in_bytes; #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src1_addr_in_bytes += (z % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src1_addr_in_bytes += z * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) __global half *src_addr_a = (__global half *)(src0_ptr + src0_addr_in_bytes); __global half *src_addr_b = (__global half *)(src1_ptr + src1_addr_in_bytes); // Compute end row address for matrix B __global half *src_end_addr_b = src_addr_b + COLS_B; src_addr_a += offset_row_a; src_addr_b += offset_row_b; // Reset accumulators half8 c0 = 0.0f; half8 c1 = 0.0f; half8 c2 = 0.0f; half8 c3 = 0.0f; #define COLS_MTX_B (COLS_B / (8 * MULT_TRANSPOSE1XW_WIDTH)) int i = 0; for(; i <= (int)(COLS_MTX_B - 4); i += 4) { #if MULT_INTERLEAVE4X4_HEIGHT == 1 // Load values from matrix A (interleaved) and matrix B (transposed) half8 a0 = vload8(0, src_addr_a); half8 b0 = vload8(0, src_addr_b); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s0, b0, c0); c1 = fma((half8)a0.s1, b0, c1); c2 = fma((half8)a0.s2, b0, c2); c3 = fma((half8)a0.s3, b0, c3); // Load values from matrix B (transposed) b0 = vload8(0, src_addr_b); src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s4, b0, c0); c1 = fma((half8)a0.s5, b0, c1); c2 = fma((half8)a0.s6, b0, c2); c3 = fma((half8)a0.s7, b0, c3); // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload8(0, src_addr_a); b0 = vload8(0, src_addr_b); src_addr_a += 8 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s0, b0, c0); c1 = fma((half8)a0.s1, b0, c1); c2 = fma((half8)a0.s2, b0, c2); c3 = fma((half8)a0.s3, b0, c3); // Load values from matrix B (transposed) b0 = vload8(0, src_addr_b); src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s4, b0, c0); c1 = fma((half8)a0.s5, b0, c1); c2 = fma((half8)a0.s6, b0, c2); c3 = fma((half8)a0.s7, b0, c3); #else // MULT_INTERLEAVE4X4_HEIGHT == 1 // Load values from matrix A (interleaved) and matrix B (transposed) half4 a0 = vload4(0, src_addr_a); half8 b0 = vload8(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s0, b0, c0); c1 = fma((half8)a0.s1, b0, c1); c2 = fma((half8)a0.s2, b0, c2); c3 = fma((half8)a0.s3, b0, c3); // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a); b0 = vload8(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s0, b0, c0); c1 = fma((half8)a0.s1, b0, c1); c2 = fma((half8)a0.s2, b0, c2); c3 = fma((half8)a0.s3, b0, c3); // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a); b0 = vload8(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s0, b0, c0); c1 = fma((half8)a0.s1, b0, c1); c2 = fma((half8)a0.s2, b0, c2); c3 = fma((half8)a0.s3, b0, c3); // Load values from matrix A (interleaved) and matrix B (transposed) a0 = vload4(0, src_addr_a); b0 = vload8(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s0, b0, c0); c1 = fma((half8)a0.s1, b0, c1); c2 = fma((half8)a0.s2, b0, c2); c3 = fma((half8)a0.s3, b0, c3); #endif // MULT_INTERLEAVE4X4_HEIGHT == 1 } for(; i < (int)(COLS_MTX_B); ++i) { // Load values from matrix A (interleaved) and matrix B (transposed) half4 a0 = vload4(0, src_addr_a); half8 b0 = vload8(0, src_addr_b); src_addr_a += 4 * MULT_INTERLEAVE4X4_HEIGHT; src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH; c0 = fma((half8)a0.s0, b0, c0); c1 = fma((half8)a0.s1, b0, c1); c2 = fma((half8)a0.s2, b0, c2); c3 = fma((half8)a0.s3, b0, c3); } // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * 4)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(4, half, c, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(4, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)); LOAD_BLOCK(1, 8, half, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, half, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(4, c, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)) + (get_global_id(1) * (uint)4 * src2_stride_y) + get_global_id( 2) * src2_stride_z; LOAD_BLOCK(4, 8, half, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(4, half, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(4, c, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(4, ACTIVATION_TYPE, half, c, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store 4x8 block vstore8(c0, 0, (__global half *)(dst_addr + 0 * dst_stride_y + zout.s0)); vstore8(c1, 0, (__global half *)(dst_addr + 1 * dst_stride_y + zout.s1)); vstore8(c2, 0, (__global half *)(dst_addr + 2 * dst_stride_y + zout.s2)); vstore8(c3, 0, (__global half *)(dst_addr + 3 * dst_stride_y + zout.s3)); } // Undefine local defines #undef COLS_MTX_B #endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) #endif // defined(COLS_B) && defined(MULT_TRANSPOSE1XW_WIDTH) && defined(MULT_INTERLEAVE4X4_HEIGHT) #if defined(COLS_A) && defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) && (NUM_ELEMS_PROCESSED_PER_THREAD_Y) #if defined(DATA_TYPE) #define VECTOR_TYPE VEC_DATA_TYPE(DATA_TYPE, NUM_ELEMS_PROCESSED_PER_THREAD_X) /** This OpenCL kernel computes the matrix by matrix multiplication between the matrix A (src0) and matrix B (src1) in case both matrices have not been reshaped. * * @note This OpenCL kernel works with floating point data types (F16/F32) * @note The floating point data type must be passed at compile time using -DDATA_TYPE (e.g. -DDATA_TYPE=float) * @note The number of elements processed along the x and y directions must be passed at compile time using -DNUM_ELEMS_PROCESSED_PER_THREAD_X and -DNUM_ELEMS_PROCESSED_PER_THREAD_Y * @note The number of matrix A columns and the optional alpha's value need to be passed at compile time using -DCOLS_A and -DALPHA * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F16/F32 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] src_cross_plane_pad (Optional) Bottom paddings in unit of elements for the input tensor (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings in unit of elements for the output tensor (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_floating_point(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint src_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X; // Compute starting address for matrix A and Matrix B int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes)); // Update address for the matrix A src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y; // Update address for the matrix B src_addr.s1 += idx * sizeof(DATA_TYPE); #if defined(REINTERPRET_INPUT_AS_3D) // Since we load a 2D input tile from a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zin) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D uint4 zin = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zin = min(DEPTH_GEMM3D - 1, zin); // Add offset due to the cross plane paddings zin *= (src_cross_plane_pad * src0_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply src0_stride_z by DEPTH_GEMM3D src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM src_addr.s0 += get_global_id(2) * src0_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src_addr.s1 += get_global_id(2) * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) int end_row_vec_a = src_addr.s0 + (COLS_A * sizeof(DATA_TYPE)); VECTOR_TYPE acc0 = 0.0f; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 VECTOR_TYPE acc1 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 VECTOR_TYPE acc2 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 VECTOR_TYPE acc3 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 for(; src_addr.s0 <= (end_row_vec_a - 2 * (int)sizeof(DATA_TYPE)); src_addr += (int2)(2 * sizeof(DATA_TYPE), 2 * src1_stride_y)) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 2, DATA_TYPE, a, src0_ptr, src_addr.s0, src0_stride_y, zin.s); #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A VEC_DATA_TYPE(DATA_TYPE, 2) a0 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 VEC_DATA_TYPE(DATA_TYPE, 2) a1 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 VEC_DATA_TYPE(DATA_TYPE, 2) a2 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 VEC_DATA_TYPE(DATA_TYPE, 2) a3 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B VECTOR_TYPE b0 = VLOAD(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, (__global DATA_TYPE *)(src1_ptr + src_addr.s1)); VECTOR_TYPE b1 = VLOAD(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, (__global DATA_TYPE *)(src1_ptr + src_addr.s1 + src1_stride_y)); // Accumulate acc0 += b0 * (VECTOR_TYPE)a0.s0; acc0 += b1 * (VECTOR_TYPE)a0.s1; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 += b0 * (VECTOR_TYPE)a1.s0; acc1 += b1 * (VECTOR_TYPE)a1.s1; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 += b0 * (VECTOR_TYPE)a2.s0; acc2 += b1 * (VECTOR_TYPE)a2.s1; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 += b0 * (VECTOR_TYPE)a3.s0; acc3 += b1 * (VECTOR_TYPE)a3.s1; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 } for(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(sizeof(DATA_TYPE), src1_stride_y)) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A DATA_TYPE a0 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 DATA_TYPE a1 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 DATA_TYPE a2 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 DATA_TYPE a3 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A DATA_TYPE a0 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 DATA_TYPE a1 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 DATA_TYPE a2 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 DATA_TYPE a3 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B VECTOR_TYPE b0 = VLOAD(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, (__global DATA_TYPE *)(src1_ptr + src_addr.s1)); // Accumulate acc0 += b0 * (VECTOR_TYPE)a0; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 += b0 * (VECTOR_TYPE)a1; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 += b0 * (VECTOR_TYPE)a2; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 += b0 * (VECTOR_TYPE)a3; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 } int z = get_global_id(2); // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (dst_cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, DATA_TYPE, acc, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)NUM_ELEMS_PROCESSED_PER_THREAD_X * sizeof(DATA_TYPE)); LOAD_BLOCK(1, NUM_ELEMS_PROCESSED_PER_THREAD_X, DATA_TYPE, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias[broadcasted] ADD_BLOCK_BROADCAST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)NUM_ELEMS_PROCESSED_PER_THREAD_X * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)NUM_ELEMS_PROCESSED_PER_THREAD_Y * src2_stride_y) + get_global_id(2) * src2_stride_z; LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, NUM_ELEMS_PROCESSED_PER_THREAD_X, DATA_TYPE, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, DATA_TYPE, bias, BETA); #endif // UNIT_BIAS // c = c + bias ADD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, ACTIVATION_TYPE, DATA_TYPE, acc, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store output block STORE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, NUM_ELEMS_PROCESSED_PER_THREAD_X, DATA_TYPE, acc, dst_addr, dst_stride_y, zout.s); } #endif // defined(DATA_TYPE) /** This OpenCL kernel computes the matrix by matrix multiplication between the matrix A (src0) and matrix B (src1) in case both matrices have not been reshaped * * @note This OpenCL kernel works with the 32-bit floating point data type (float) and uses the fma units. * @note The number of elements processed along the x and y directions must be passed at compile time using -DNUM_ELEMS_PROCESSED_PER_THREAD_X and -DNUM_ELEMS_PROCESSED_PER_THREAD_Y. * This kernel optimally uses -DNUM_ELEMS_PROCESSED_PER_THREAD_X=4. * @note The number of matrix A columns must be passed at compile time using -DCOLS_A. * @note The optional value of scalar alpha is passed at compile time using -DALPHA=alpha * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F32 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] src_cross_plane_pad (Optional) Bottom paddings in unit of elements for the input tensor (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_floating_point_f32_bifrost(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint src_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X; // Compute starting address for matrix A and matrix B int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes)); // Update address for matrix A src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y; // Update address for matrix B src_addr.s1 += idx * sizeof(float); #if defined(REINTERPRET_INPUT_AS_3D) // Since we load a 2D input tile from a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zin) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D uint4 zin = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zin = min(DEPTH_GEMM3D - 1, zin); // Add offset due to the cross plane paddings zin *= (src_cross_plane_pad * src0_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply src0_stride_z by DEPTH_GEMM3D src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM src_addr.s0 += get_global_id(2) * src0_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src_addr.s1 += get_global_id(2) * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) // Initialize accumulators float4 acc0 = 0.0f; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float4 acc1 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float4 acc2 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float4 acc3 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 // A and B src indices get incremented at the same time. int i = 0; for(; i <= ((int)COLS_A - 4); i += 4) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A and matrix B LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 4, float, a, src0_ptr, src_addr.s0, src0_stride_y, zin.s); #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A and matrix B float4 a0 = vload4(0, (__global float *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float4 a1 = vload4(0, (__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float4 a2 = vload4(0, (__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float4 a3 = vload4(0, (__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) float4 b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Multiply and accumulate acc0.s0 = fma(a0.s0, b0.s0, acc0.s0); acc0.s1 = fma(a0.s0, b0.s1, acc0.s1); acc0.s2 = fma(a0.s0, b0.s2, acc0.s2); acc0.s3 = fma(a0.s0, b0.s3, acc0.s3); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1.s0 = fma(a1.s0, b0.s0, acc1.s0); acc1.s1 = fma(a1.s0, b0.s1, acc1.s1); acc1.s2 = fma(a1.s0, b0.s2, acc1.s2); acc1.s3 = fma(a1.s0, b0.s3, acc1.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2.s0 = fma(a2.s0, b0.s0, acc2.s0); acc2.s1 = fma(a2.s0, b0.s1, acc2.s1); acc2.s2 = fma(a2.s0, b0.s2, acc2.s2); acc2.s3 = fma(a2.s0, b0.s3, acc2.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3.s0 = fma(a3.s0, b0.s0, acc3.s0); acc3.s1 = fma(a3.s0, b0.s1, acc3.s1); acc3.s2 = fma(a3.s0, b0.s2, acc3.s2); acc3.s3 = fma(a3.s0, b0.s3, acc3.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 // Load values from matrix A and matrix B b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Multiply and accumulate acc0.s0 = fma(a0.s1, b0.s0, acc0.s0); acc0.s1 = fma(a0.s1, b0.s1, acc0.s1); acc0.s2 = fma(a0.s1, b0.s2, acc0.s2); acc0.s3 = fma(a0.s1, b0.s3, acc0.s3); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1.s0 = fma(a1.s1, b0.s0, acc1.s0); acc1.s1 = fma(a1.s1, b0.s1, acc1.s1); acc1.s2 = fma(a1.s1, b0.s2, acc1.s2); acc1.s3 = fma(a1.s1, b0.s3, acc1.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2.s0 = fma(a2.s1, b0.s0, acc2.s0); acc2.s1 = fma(a2.s1, b0.s1, acc2.s1); acc2.s2 = fma(a2.s1, b0.s2, acc2.s2); acc2.s3 = fma(a2.s1, b0.s3, acc2.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3.s0 = fma(a3.s1, b0.s0, acc3.s0); acc3.s1 = fma(a3.s1, b0.s1, acc3.s1); acc3.s2 = fma(a3.s1, b0.s2, acc3.s2); acc3.s3 = fma(a3.s1, b0.s3, acc3.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 // Load values from matrix A and matrix B b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Multiply and accumulate acc0.s0 = fma(a0.s2, b0.s0, acc0.s0); acc0.s1 = fma(a0.s2, b0.s1, acc0.s1); acc0.s2 = fma(a0.s2, b0.s2, acc0.s2); acc0.s3 = fma(a0.s2, b0.s3, acc0.s3); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1.s0 = fma(a1.s2, b0.s0, acc1.s0); acc1.s1 = fma(a1.s2, b0.s1, acc1.s1); acc1.s2 = fma(a1.s2, b0.s2, acc1.s2); acc1.s3 = fma(a1.s2, b0.s3, acc1.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2.s0 = fma(a2.s2, b0.s0, acc2.s0); acc2.s1 = fma(a2.s2, b0.s1, acc2.s1); acc2.s2 = fma(a2.s2, b0.s2, acc2.s2); acc2.s3 = fma(a2.s2, b0.s3, acc2.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3.s0 = fma(a3.s2, b0.s0, acc3.s0); acc3.s1 = fma(a3.s2, b0.s1, acc3.s1); acc3.s2 = fma(a3.s2, b0.s2, acc3.s2); acc3.s3 = fma(a3.s2, b0.s3, acc3.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 // Load values from matrix A and matrix B b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Multiply and accumulate acc0.s0 = fma(a0.s3, b0.s0, acc0.s0); acc0.s1 = fma(a0.s3, b0.s1, acc0.s1); acc0.s2 = fma(a0.s3, b0.s2, acc0.s2); acc0.s3 = fma(a0.s3, b0.s3, acc0.s3); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1.s0 = fma(a1.s3, b0.s0, acc1.s0); acc1.s1 = fma(a1.s3, b0.s1, acc1.s1); acc1.s2 = fma(a1.s3, b0.s2, acc1.s2); acc1.s3 = fma(a1.s3, b0.s3, acc1.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2.s0 = fma(a2.s3, b0.s0, acc2.s0); acc2.s1 = fma(a2.s3, b0.s1, acc2.s1); acc2.s2 = fma(a2.s3, b0.s2, acc2.s2); acc2.s3 = fma(a2.s3, b0.s3, acc2.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3.s0 = fma(a3.s3, b0.s0, acc3.s0); acc3.s1 = fma(a3.s3, b0.s1, acc3.s1); acc3.s2 = fma(a3.s3, b0.s2, acc3.s2); acc3.s3 = fma(a3.s3, b0.s3, acc3.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 src_addr.s0 += 4 * sizeof(float); } for(; i < (int)COLS_A; ++i) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A float a0 = *((__global float *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float a1 = *((__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float a2 = *((__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float a3 = *((__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A float a0 = *((__global float *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float a1 = *((__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float a2 = *((__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float a3 = *((__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B float4 b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Multiply and accumulate acc0.s0 = fma(a0, b0.s0, acc0.s0); acc0.s1 = fma(a0, b0.s1, acc0.s1); acc0.s2 = fma(a0, b0.s2, acc0.s2); acc0.s3 = fma(a0, b0.s3, acc0.s3); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1.s0 = fma(a1, b0.s0, acc1.s0); acc1.s1 = fma(a1, b0.s1, acc1.s1); acc1.s2 = fma(a1, b0.s2, acc1.s2); acc1.s3 = fma(a1, b0.s3, acc1.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2.s0 = fma(a2, b0.s0, acc2.s0); acc2.s1 = fma(a2, b0.s1, acc2.s1); acc2.s2 = fma(a2, b0.s2, acc2.s2); acc2.s3 = fma(a2, b0.s3, acc2.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3.s0 = fma(a3, b0.s0, acc3.s0); acc3.s1 = fma(a3, b0.s1, acc3.s1); acc3.s2 = fma(a3, b0.s2, acc3.s2); acc3.s3 = fma(a3, b0.s3, acc3.s3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 src_addr.s0 += sizeof(float); } int z = get_global_id(2); // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (dst_cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, float, acc, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)4 * sizeof(float)); LOAD_BLOCK(1, 4, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, float, bias, BETA); #endif // UNIT_BIAS // acc = acc + bias[broadcasted] ADD_BLOCK_BROADCAST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)4 * sizeof(float)) + (get_global_id(1) * (uint)NUM_ELEMS_PROCESSED_PER_THREAD_Y * src2_stride_y) + get_global_id(2) * src2_stride_z; LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 4, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, float, bias, BETA); #endif // UNIT_BIAS // acc = acc + bias ADD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, ACTIVATION_TYPE, float, acc, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store the output block vstore4(acc0, 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 vstore4(acc1, 0, (__global float *)(dst_addr + 1 * dst_stride_y + zout.s1)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 vstore4(acc2, 0, (__global float *)(dst_addr + 2 * dst_stride_y + zout.s2)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 vstore4(acc3, 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 } /** This OpenCL kernel computes the matrix by matrix multiplication between the matrix A (src0) and matrix B (src1) in case both matrices have not been reshaped * * @note This OpenCL kernel works with the 32-bit floating point data type (float) and uses the fma units. * This OpenCL kernel is optimized for Bifrost when the number of matrix B columns is less or equal to 1000. * @note The number of elements processed along the x and y directions must be passed at compile time using -DNUM_ELEMS_PROCESSED_PER_THREAD_X and -DNUM_ELEMS_PROCESSED_PER_THREAD_Y. * This kernel optimally uses -DNUM_ELEMS_PROCESSED_PER_THREAD_X=2. * @note The number of matrix A columns must be passed at compile time using -DCOLS_A. * @note The optional value of scalar alpha is passed at compile time using -DALPHA=alpha if alpha!=1.0f. * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F32 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] src_cross_plane_pad (Optional) Bottom paddings in unit of elements for the input tensor (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_floating_point_f32_bifrost_1000(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint src_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { // Requires 2 NUM_ELEMS_PROCESSED_PER_THREAD_X, C vect2, A vect4, B (2 vload2) // to fix for NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 int idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X; // Compute starting address for matrix A and Matrix B int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes)); // Update address for the matrix A src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y; // Update address for the matrix B src_addr.s1 += idx * sizeof(float); #if defined(REINTERPRET_INPUT_AS_3D) // Since we load a 2D input tile from a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zin) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D uint4 zin = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zin = min(DEPTH_GEMM3D - 1, zin); // Add offset due to the cross plane paddings zin *= (src_cross_plane_pad * src0_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply src0_stride_z by DEPTH_GEMM3D src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM src_addr.s0 += get_global_id(2) * src0_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src_addr.s1 += get_global_id(2) * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) // Initialize accumulators float2 acc0 = 0.0f; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float2 acc1 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float2 acc2 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float2 acc3 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 // A and B src indices get incremented at the same time. int i = 0; for(; i <= ((int)COLS_A - 8); i += 8) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A float8 a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + zin.s0)); #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A float8 a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0)); #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B float2 b0 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; float2 b1 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; float2 b2 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; float2 b3 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; float2 b4 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; float2 b5 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; float2 b6 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; float2 b7 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Multiply and accumulate acc0.s0 = fma(a0.s0, b0.s0, acc0.s0); acc0.s0 = fma(a0.s1, b1.s0, acc0.s0); acc0.s0 = fma(a0.s2, b2.s0, acc0.s0); acc0.s0 = fma(a0.s3, b3.s0, acc0.s0); acc0.s0 = fma(a0.s4, b4.s0, acc0.s0); acc0.s0 = fma(a0.s5, b5.s0, acc0.s0); acc0.s0 = fma(a0.s6, b6.s0, acc0.s0); acc0.s0 = fma(a0.s7, b7.s0, acc0.s0); acc0.s1 = fma(a0.s0, b0.s1, acc0.s1); acc0.s1 = fma(a0.s1, b1.s1, acc0.s1); acc0.s1 = fma(a0.s2, b2.s1, acc0.s1); acc0.s1 = fma(a0.s3, b3.s1, acc0.s1); acc0.s1 = fma(a0.s4, b4.s1, acc0.s1); acc0.s1 = fma(a0.s5, b5.s1, acc0.s1); acc0.s1 = fma(a0.s6, b6.s1, acc0.s1); acc0.s1 = fma(a0.s7, b7.s1, acc0.s1); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if defined(REINTERPRET_INPUT_AS_3D) a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1)); #else // defined(REINTERPRET_INPUT_AS_3D) a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // defined(REINTERPRET_INPUT_AS_3D) acc1.s0 = fma(a0.s0, b0.s0, acc1.s0); acc1.s0 = fma(a0.s1, b1.s0, acc1.s0); acc1.s0 = fma(a0.s2, b2.s0, acc1.s0); acc1.s0 = fma(a0.s3, b3.s0, acc1.s0); acc1.s0 = fma(a0.s4, b4.s0, acc1.s0); acc1.s0 = fma(a0.s5, b5.s0, acc1.s0); acc1.s0 = fma(a0.s6, b6.s0, acc1.s0); acc1.s0 = fma(a0.s7, b7.s0, acc1.s0); acc1.s1 = fma(a0.s0, b0.s1, acc1.s1); acc1.s1 = fma(a0.s1, b1.s1, acc1.s1); acc1.s1 = fma(a0.s2, b2.s1, acc1.s1); acc1.s1 = fma(a0.s3, b3.s1, acc1.s1); acc1.s1 = fma(a0.s4, b4.s1, acc1.s1); acc1.s1 = fma(a0.s5, b5.s1, acc1.s1); acc1.s1 = fma(a0.s6, b6.s1, acc1.s1); acc1.s1 = fma(a0.s7, b7.s1, acc1.s1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if defined(REINTERPRET_INPUT_AS_3D) a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2)); #else // defined(REINTERPRET_INPUT_AS_3D) a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // defined(REINTERPRET_INPUT_AS_3D) acc2.s0 = fma(a0.s0, b0.s0, acc2.s0); acc2.s0 = fma(a0.s1, b1.s0, acc2.s0); acc2.s0 = fma(a0.s2, b2.s0, acc2.s0); acc2.s0 = fma(a0.s3, b3.s0, acc2.s0); acc2.s0 = fma(a0.s4, b4.s0, acc2.s0); acc2.s0 = fma(a0.s5, b5.s0, acc2.s0); acc2.s0 = fma(a0.s6, b6.s0, acc2.s0); acc2.s0 = fma(a0.s7, b7.s0, acc2.s0); acc2.s1 = fma(a0.s0, b0.s1, acc2.s1); acc2.s1 = fma(a0.s1, b1.s1, acc2.s1); acc2.s1 = fma(a0.s2, b2.s1, acc2.s1); acc2.s1 = fma(a0.s3, b3.s1, acc2.s1); acc2.s1 = fma(a0.s4, b4.s1, acc2.s1); acc2.s1 = fma(a0.s5, b5.s1, acc2.s1); acc2.s1 = fma(a0.s6, b6.s1, acc2.s1); acc2.s1 = fma(a0.s7, b7.s1, acc2.s1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #if defined(REINTERPRET_INPUT_AS_3D) a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3)); #else // defined(REINTERPRET_INPUT_AS_3D) a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // defined(REINTERPRET_INPUT_AS_3D) acc3.s0 = fma(a0.s0, b0.s0, acc3.s0); acc3.s0 = fma(a0.s1, b1.s0, acc3.s0); acc3.s0 = fma(a0.s2, b2.s0, acc3.s0); acc3.s0 = fma(a0.s3, b3.s0, acc3.s0); acc3.s0 = fma(a0.s4, b4.s0, acc3.s0); acc3.s0 = fma(a0.s5, b5.s0, acc3.s0); acc3.s0 = fma(a0.s6, b6.s0, acc3.s0); acc3.s0 = fma(a0.s7, b7.s0, acc3.s0); acc3.s1 = fma(a0.s0, b0.s1, acc3.s1); acc3.s1 = fma(a0.s1, b1.s1, acc3.s1); acc3.s1 = fma(a0.s2, b2.s1, acc3.s1); acc3.s1 = fma(a0.s3, b3.s1, acc3.s1); acc3.s1 = fma(a0.s4, b4.s1, acc3.s1); acc3.s1 = fma(a0.s5, b5.s1, acc3.s1); acc3.s1 = fma(a0.s6, b6.s1, acc3.s1); acc3.s1 = fma(a0.s7, b7.s1, acc3.s1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 src_addr.s0 += sizeof(float) * 8; } // float size increment for(; i < (int)COLS_A; ++i) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A float a0 = *((__global float *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float a1 = *((__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float a2 = *((__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float a3 = *((__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A float a0 = *((__global float *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float a1 = *((__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float a2 = *((__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float a3 = *((__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B float2 b0 = vload2(0, (__global float *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Multiply and accumulate acc0.s0 = fma(a0, b0.s0, acc0.s0); acc0.s1 = fma(a0, b0.s1, acc0.s1); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1.s0 = fma(a1, b0.s0, acc1.s0); acc1.s1 = fma(a1, b0.s1, acc1.s1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2.s0 = fma(a2, b0.s0, acc2.s0); acc2.s1 = fma(a2, b0.s1, acc2.s1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3.s0 = fma(a3, b0.s0, acc3.s0); acc3.s1 = fma(a3, b0.s1, acc3.s1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 src_addr.s0 += sizeof(float); } int z = get_global_id(2); // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (dst_cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, float, acc, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)2 * sizeof(float)); LOAD_BLOCK(1, 2, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, float, bias, BETA); #endif // UNIT_BIAS // acc = acc + bias[broadcasted] ADD_BLOCK_BROADCAST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)2 * sizeof(float)) + (get_global_id(1) * (uint)NUM_ELEMS_PROCESSED_PER_THREAD_Y * src2_stride_y) + get_global_id(2) * src2_stride_z; LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 2, float, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, float, bias, BETA); #endif // UNIT_BIAS // acc = acc + bias ADD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, ACTIVATION_TYPE, float, acc, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store the output block vstore2(acc0, 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 vstore2(acc1, 0, (__global float *)(dst_addr + 1 * dst_stride_y + zout.s1)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 vstore2(acc2, 0, (__global float *)(dst_addr + 2 * dst_stride_y + zout.s2)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 vstore2(acc3, 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 } #if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) /** This OpenCL kernel computes the matrix by matrix multiplication between the matrix A (src0) and matrix B (src1) in case both matrices have not beed reshaped * * @note This OpenCL kernel works with the 16-bit floating point data type (half) and accumulating the result in a 32 floating point variable. * @note The number of elements processed along the x and y directions must be passed at compile time using -DNUM_ELEMS_PROCESSED_PER_THREAD_X and -DNUM_ELEMS_PROCESSED_PER_THREAD_Y. * This kernel optimally uses -DNUM_ELEMS_PROCESSED_PER_THREAD_X=4. * @note The number of matrix A columns must be passed at compile time using -DCOLS_A. * @note The optional value of scalar alpha is passed at compile time using -DALPHA=alpha * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F16 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] src_cross_plane_pad (Optional) Bottom paddings in unit of elements for the input tensor (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_floating_point_f16_bifrost_acc32(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint src_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X; // Compute starting address for matrix A and Matrix B int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes)); // Update address for the matrix A src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y; // Update address for the matrix B src_addr.s1 += idx * sizeof(half); #if defined(REINTERPRET_INPUT_AS_3D) // Since we load a 2D input tile from a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zin) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D uint4 zin = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zin = min(DEPTH_GEMM3D - 1, zin); // Add offset due to the cross plane paddings zin *= (src_cross_plane_pad * src0_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply src0_stride_z by DEPTH_GEMM3D src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM src_addr.s0 += get_global_id(2) * src0_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src_addr.s1 += get_global_id(2) * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) float8 acc0 = 0.0h; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float8 acc1 = 0.0h; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float8 acc2 = 0.0h; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float8 acc3 = 0.0h; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 int i = 0; for(; i <= ((int)COLS_A - 4); i += 4) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 4, half, a, src0_ptr, src_addr.s0, src0_stride_y, zin.s); #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A half4 a0 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half4 a1 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half4 a2 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half4 a3 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B float8 b0 = convert_float8(vload8(0, (__global half *)(src1_ptr + src_addr.s1))); src_addr.s1 += src1_stride_y; // Accumulate acc0 = fma(b0, (float8)a0.s0, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (float8)a1.s0, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (float8)a2.s0, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (float8)a3.s0, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 b0 = convert_float8(vload8(0, (__global half *)(src1_ptr + src_addr.s1))); src_addr.s1 += src1_stride_y; acc0 = fma(b0, (float8)a0.s1, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (float8)a1.s1, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (float8)a2.s1, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (float8)a3.s1, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 b0 = convert_float8(vload8(0, (__global half *)(src1_ptr + src_addr.s1))); src_addr.s1 += src1_stride_y; acc0 = fma(b0, (float8)a0.s2, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (float8)a1.s2, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (float8)a2.s2, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (float8)a3.s2, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 b0 = convert_float8(vload8(0, (__global half *)(src1_ptr + src_addr.s1))); src_addr.s1 += src1_stride_y; acc0 = fma(b0, (float8)a0.s3, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (float8)a1.s3, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (float8)a2.s3, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (float8)a3.s3, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 src_addr.s0 += 4 * sizeof(half); } for(; i < (int)COLS_A; ++i) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A half a0 = *((__global half *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half a1 = *((__global half *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half a2 = *((__global half *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half a3 = *((__global half *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A half a0 = *((__global half *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half a1 = *((__global half *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half a2 = *((__global half *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half a3 = *((__global half *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B float8 b0 = convert_float8(vload8(0, (__global half *)(src1_ptr + src_addr.s1))); src_addr += (int2)(sizeof(half), src1_stride_y); // Accumulate acc0 = fma(b0, (float8)a0, acc0); // b0 * (half8)a0; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (float8)a1, acc1); // b0 * (half8)a1; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (float8)a2, acc2); // b0 * (half8)a2; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (float8)a3, acc3); // b0 * (half8)a3; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 } int z = get_global_id(2); // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (dst_cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, float, acc, ALPHA); #endif // defined(ALPHA) #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)); LOAD_BLOCK(1, 8, half, bias, src2_addr, 0, src2_stride_y, zero); float8 bias_f0 = convert_float8(bias0); #ifndef UNIT_BETA SCALE_BLOCK(1, float, bias_f, BETA); #endif // UNIT_BIAS // acc = acc + bias[broadcasted] ADD_BLOCK_BROADCAST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias_f0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)) + (get_global_id(1) * (uint)NUM_ELEMS_PROCESSED_PER_THREAD_Y * src2_stride_y) + get_global_id(2) * src2_stride_z; LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 8, half, bias, src2_addr, 0, src2_stride_y, zero); float8 bias_f0 = convert_float8(bias0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 float8 bias_f1 = convert_float8(bias1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 float8 bias_f2 = convert_float8(bias2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 float8 bias_f3 = convert_float8(bias3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #ifndef UNIT_BETA SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, float, bias_f, BETA); #endif // UNIT_BIAS // acc = acc + bias ADD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias_f); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) half8 acc_h0 = convert_half8(acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half8 acc_h1 = convert_half8(acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half8 acc_h2 = convert_half8(acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half8 acc_h3 = convert_half8(acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, ACTIVATION_TYPE, half, acc_h, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store the output block STORE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 8, half, acc_h, dst_addr, dst_stride_y, zout.s); } /** This OpenCL kernel computes the matrix by matrix multiplication between the matrix A (src0) and matrix B (src1) in case both matrices have not beed reshaped * * @note This OpenCL kernel works with the 16-bit floating point data type (half) and uses the fma units. * @note The number of elements processed along the x and y directions must be passed at compile time using -DNUM_ELEMS_PROCESSED_PER_THREAD_X and -DNUM_ELEMS_PROCESSED_PER_THREAD_Y. * This kernel optimally uses -DNUM_ELEMS_PROCESSED_PER_THREAD_X=4. * @note The number of matrix A columns must be passed at compile time using -DCOLS_A. * @note The optional value of scalar alpha is passed at compile time using -DALPHA=alpha * @note In case the matrix B has 3 dimensions and the matrix A more than 3, in order to avoid out-of-bounds reads, the number of channels of matrix B must be passed at compile time using MATRIX_B_DEPTH (e.g. -DMATRIX_B_DEPTH=16) * This case can happen when GEMM is used to perform the element-wise multiplication through a batched matrix multiplication (2D Winograd) and we have multiple inputs (e.g. a = [K, M, 16, Batches], b = [N, K, 16]) * * @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 * @note In case the input or output have to be reinterpreted as a 3D tensor, the following information must be passed at compile time: * -# REINTERPRET_INPUT_AS_3D: To reinterpret the input as 3D * -# REINTERPRET_OUTPUT_AS_3D: To reinterpret the output as 3D * -# HEIGHT_GEMM3D: The height of the output in case it has to be reinterpreted as a 3D tensor. * -# DEPTH_GEMM3D: The depth of the output in case it has to be reinterpreted as a 3D tensor * (HEIGHT_GEMM3D * DEPTH_GEMM3D) = columns matrix A NOT reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F16 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src2_ptr (Optional) Pointer to the bias matrix. Supported data type: same as @p lhs_ptr * @param[in] src2_stride_x (Optional) Stride of the bias matrix in X dimension (in bytes) * @param[in] src2_step_x (Optional) src2_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src2_stride_y (Optional) Stride of the bias matrix in Y dimension (in bytes) * @param[in] src2_step_y (Optional) src2_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src2_offset_first_element_in_bytes (Optional) The offset of the first element in the bias matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix * @param[in] src0_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src2_stride_z (Optional) Stride of the bias matrix in Z dimension (in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] src_cross_plane_pad (Optional) Bottom paddings in unit of elements for the input tensor (only if defined REINTERPRET_INPUT_AS_3D) * @param[in] dst_cross_plane_pad (Optional) Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D) */ __kernel void gemm_mm_floating_point_f16_bifrost(IMAGE_DECLARATION(src0), IMAGE_DECLARATION(src1), #if defined(BETA) IMAGE_DECLARATION(src2), #endif // defined(BETA) IMAGE_DECLARATION(dst), uint src0_stride_z, uint src1_stride_z, #if defined(BETA) uint src2_stride_z, #endif //defined(BETA) uint dst_stride_z #if defined(REINTERPRET_INPUT_AS_3D) , uint src_cross_plane_pad #endif // REINTERPRET_INPUT_AS_3D #if defined(REINTERPRET_OUTPUT_AS_3D) , uint dst_cross_plane_pad #endif // REINTERPRET_OUTPUT_AS_3D ) { int idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X; // Compute starting address for matrix A and Matrix B int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes)); // Update address for the matrix A src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y; // Update address for the matrix B src_addr.s1 += idx * sizeof(half); #if defined(REINTERPRET_INPUT_AS_3D) // Since we load a 2D input tile from a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zin) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D uint4 zin = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zin = min(DEPTH_GEMM3D - 1, zin); // Add offset due to the cross plane paddings zin *= (src_cross_plane_pad * src0_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply src0_stride_z by DEPTH_GEMM3D src_addr.s0 += get_global_id(2) * src0_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_INPUT_AS_3D) // Add offset for batched GEMM src_addr.s0 += get_global_id(2) * src0_stride_z; #endif // defined(REINTERPRET_INPUT_AS_3D) #if defined(MATRIX_B_DEPTH) // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 src_addr.s1 += (get_global_id(2) % MATRIX_B_DEPTH) * src1_stride_z; #else // defined(MATRIX_B_DEPTH) src_addr.s1 += get_global_id(2) * src1_stride_z; #endif // defined(MATRIX_B_DEPTH) half8 acc0 = 0.0h; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half8 acc1 = 0.0h; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half8 acc2 = 0.0h; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half8 acc3 = 0.0h; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 int i = 0; for(; i <= ((int)COLS_A - 4); i += 4) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 4, half, a, src0_ptr, src_addr.s0, src0_stride_y, zin.s); #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A half4 a0 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half4 a1 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half4 a2 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half4 a3 = vload4(0, (__global half *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B half8 b0 = vload8(0, (__global half *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; // Accumulate acc0 = fma(b0, (half8)a0.s0, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (half8)a1.s0, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (half8)a2.s0, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (half8)a3.s0, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 b0 = vload8(0, (__global half *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; acc0 = fma(b0, (half8)a0.s1, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (half8)a1.s1, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (half8)a2.s1, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (half8)a3.s1, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 b0 = vload8(0, (__global half *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; acc0 = fma(b0, (half8)a0.s2, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (half8)a1.s2, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (half8)a2.s2, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (half8)a3.s2, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 b0 = vload8(0, (__global half *)(src1_ptr + src_addr.s1)); src_addr.s1 += src1_stride_y; acc0 = fma(b0, (half8)a0.s3, acc0); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (half8)a1.s3, acc1); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (half8)a2.s3, acc2); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (half8)a3.s3, acc3); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 src_addr.s0 += 4 * sizeof(half); } for(; i < (int)COLS_A; ++i) { #if defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A half a0 = *((__global half *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y + zin.s0)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half a1 = *((__global half *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y + zin.s1)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half a2 = *((__global half *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y + zin.s2)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half a3 = *((__global half *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y + zin.s3)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #else // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix A half a0 = *((__global half *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 half a1 = *((__global half *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 half a2 = *((__global half *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 half a3 = *((__global half *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y)); #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 #endif // defined(REINTERPRET_INPUT_AS_3D) // Load values from matrix B half8 b0 = vload8(0, (__global half *)(src1_ptr + src_addr.s1)); src_addr += (int2)(sizeof(half), src1_stride_y); // Accumulate acc0 = fma(b0, (half8)a0, acc0); // b0 * (half8)a0; #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 acc1 = fma(b0, (half8)a1, acc1); // b0 * (half8)a1; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 acc2 = fma(b0, (half8)a2, acc2); // b0 * (half8)a2; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 acc3 = fma(b0, (half8)a3, acc3); // b0 * (half8)a3; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 } int z = get_global_id(2); // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); // Compute dst address __global uchar *dst_addr = offset(&dst, 0, 0); uint4 zout = 0; #if defined(REINTERPRET_OUTPUT_AS_3D) // Since we store a 2D output tile in a 3D tensor, we need to check when the plane changes across the z dimension // in order to take into account the presence of possible cross plane paddings // // | | // | plane0 | // | | // |__________________| // |******************| // | cross_plane_pad | // |******************| // | | // | plane1 | // | | // |__________________| // The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D zout = ((uint4)(0, 1, 2, 3) + (uint4)(get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y)) / (uint4)HEIGHT_GEMM3D; zout = min(DEPTH_GEMM3D - 1, zout); // Add offset due to the cross plane paddings zout *= (dst_cross_plane_pad * dst_stride_y); // Add offset for batched GEMM. The batches will be in the fourth dimension and for this reason we // multiply dst_stride_z by DEPTH_GEMM3D dst_addr += z * dst_stride_z * DEPTH_GEMM3D; #else // defined(REINTERPRET_OUTPUT_AS_3D) // Add offset for batched GEMM dst_addr += z * dst_stride_z; #endif // defined(REINTERPRET_OUTPUT_AS_3D) // Multiply by the weight of matrix-matrix product and store the result #if defined(ALPHA) SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, half, acc, ALPHA); #endif // defined(ALPHA) // Add beta*bias #if defined(BETA) REPEAT_VAR_INIT_TO_CONST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, uint, zero, 0); #if defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)); LOAD_BLOCK(1, 8, half, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(1, half, bias, BETA); #endif // UNIT_BIAS // acc = acc + bias[broadcasted] ADD_BLOCK_BROADCAST(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias0); #else // defined(BROADCAST_BIAS) __global uchar *src2_addr = src2_ptr + src2_offset_first_element_in_bytes + (get_global_id(0) * (uint)8 * sizeof(half)) + (get_global_id(1) * (uint)NUM_ELEMS_PROCESSED_PER_THREAD_Y * src2_stride_y) + get_global_id(2) * src2_stride_z; LOAD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 8, half, bias, src2_addr, 0, src2_stride_y, zero); #ifndef UNIT_BETA SCALE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, half, bias, BETA); #endif // UNIT_BIAS // acc = acc + bias ADD_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, acc, bias); #endif // defined(BROADCAST_BIAS) #endif // defined(BETA) #if defined(ACTIVATION_TYPE) ACTIVATION_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, ACTIVATION_TYPE, half, acc, A_VAL, B_VAL); #endif // defined(ACTIVATION_TYPE) // Store the output block STORE_BLOCK(NUM_ELEMS_PROCESSED_PER_THREAD_Y, 8, half, acc, dst_addr, dst_stride_y, zout.s); } #endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) #endif // defined(COLS_A) && defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) && (NUM_ELEMS_PROCESSED_PER_THREAD_Y) #if defined(BETA) /** This OpenCL kernel performs the in-place matrix addition between 2 matrices taking into account that the second matrix might be weighted by a scalar value beta: * * @note The beta's value need to be passed at compile time using -DBETA * * @param[in] src_ptr Pointer to the source matrix. Supported data types: F32 * @param[in] src_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] src_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix */ __kernel void gemm_ma_f32(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(dst)) { // Compute source and destination addresses Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src); Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst); // Load values from A x B float4 alpha_ab = vload4(0, (__global float *)dst.ptr); // Load values from Matrix C float4 c = vload4(0, (__global float *)src.ptr); // Computes alpha * axb + beta * c float4 out = alpha_ab + (float4)BETA * c; // Store final result in axb matrix vstore4(out, 0, (__global float *)dst.ptr); } #if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) /** This OpenCL kernel performs the in-place matrix addition between 2 matrices taking into account that the second matrix might be weighted by a scalar value beta: * * @note The beta's value need to be passed at compile time using -DBETA * * @param[in] src_ptr Pointer to the source matrix. Supported data types: F16 * @param[in] src_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] src_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix */ __kernel void gemm_ma_f16(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(dst)) { // Compute source and destination addresses Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src); Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst); // Load values from A x B half8 alpha_ab = vload8(0, (__global half *)dst.ptr); // Load values from Matrix C half8 c = vload8(0, (__global half *)src.ptr); // Computes alpha * axb + beta * c half8 out = alpha_ab + (half8)BETA * c; // Store final result in axb matrix vstore8(out, 0, (__global half *)dst.ptr); } #endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) #endif // defined(BETA) #if defined(WIDTH_VECTOR_A) /** This OpenCL kernel computes the vector by matrix multiplication between each row of A (src0) and matrix B (src1) used for locally connected layer * * @note The width of A need to be passed at compile time using -DWIDTH_VECTOR_A * * @note The input A and matrix B must not be reshaped * * @param[in] src0_ptr Pointer to the source matrix. Supported data types: F32 * @param[in] src0_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src0_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src0_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src0_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src0_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[in] src1_ptr Pointer to the source matrix. Supported data types: same as @p src0_ptr * @param[in] src1_stride_x Stride of the source matrix in X dimension (in bytes) * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] src1_stride_y Stride of the source matrix in Y dimension (in bytes) * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] src1_stride_z Stride of the source matrix in Z dimension (in bytes) * @param[in] src1_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source matrix * @param[out] dst_ptr Pointer to the destination matrix Supported data types: same as @p src0_ptr * @param[in] dst_stride_x Stride of the destination matrix in X dimension (in bytes) * @param[in] dst_step_x dst_gx_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] dst_stride_y Stride of the destination matrix in Y dimension (in bytes) * @param[in] dst_step_y dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination matrix */ __kernel void gemm_lc_vm_f32(IMAGE_DECLARATION(src0), TENSOR3D_DECLARATION(src1), IMAGE_DECLARATION(dst)) { int idx = get_global_id(0) * 4; int idy = get_global_id(1); // Compute the address for the vector A and matrix B int2 src_addr = ((int2)(src0_offset_first_element_in_bytes + src0_stride_y * idy, src1_offset_first_element_in_bytes + src1_stride_z * idy)); src_addr.s1 += idx * sizeof(float); int end_row_vec_a = src_addr.s0 + (WIDTH_VECTOR_A * sizeof(float)); float4 acc = 0.0f; for(; src_addr.s0 <= (end_row_vec_a - 2 * (int)sizeof(float)); src_addr += (int2)(2 * sizeof(float), 2 * src1_stride_y)) { float2 a0 = vload2(0, (__global float *)(src0_ptr + src_addr.s0)); float4 b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); float4 b1 = vload4(0, (__global float *)(src1_ptr + src_addr.s1 + src1_stride_y)); acc += b0 * (float4)a0.s0; acc += b1 * (float4)a0.s1; } for(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(sizeof(float), src1_stride_y)) { float a0 = *((__global float *)(src0_ptr + src_addr.s0)); float4 b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); acc += b0 * (float4)a0; } // Compute destination address Image dst = CONVERT_TO_IMAGE_STRUCT(dst); vstore4(acc, 0, (__global float *)(offset(&dst, 0, 0))); } #endif // defined(WIDTH_VECTOR_A) /** This kernel accumulates each row with the biases vector. * * @note The data type must be passed at compile time using -DDATA_TYPE e.g. -DDATA_TYPE=short. * @note The vector size must be passed at compile time using -DVECTOR_SIZE e.g. -DVECTOR_SIZE=16. * * @param[in, out] accum_ptr Pointer to the accumulate tensor. Supported data type: U8/S8/U16/S16/F16/U32/S32/F32 * @param[in] accum_stride_x Stride of the accmulate tensor in X dimension (in bytes) * @param[in] accum_step_x accum_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] accum_stride_y Stride of the accumlulate tensor in Y dimension (in bytes) * @param[in] accum_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) * @param[in] accum_offset_first_element_in_bytes The offset of the first element in the accumulate tensor * @param[in] biases_ptr Pointer to the biases vector. Same as @p accum_ptr * @param[in] biases_stride_x Stride of the destination tensor in X dimension (in bytes) * @param[in] biases_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) * @param[in] biases_offset_first_element_in_bytes The offset of the first element in the destination tensor */ #if defined(DATA_TYPE) && defined(VECTOR_SIZE) __kernel void gemm_accumulate_biases( IMAGE_DECLARATION(accum), VECTOR_DECLARATION(biases)) { Image accum = CONVERT_TO_IMAGE_STRUCT(accum); Vector biases = CONVERT_TO_VECTOR_STRUCT(biases); // Vector size, e.g. number of vector elements. VEC_DATA_TYPE(DATA_TYPE, VECTOR_SIZE) accum_value = VLOAD(VECTOR_SIZE)(0, (__global DATA_TYPE *)accum.ptr); VEC_DATA_TYPE(DATA_TYPE, VECTOR_SIZE) biases_value = VLOAD(VECTOR_SIZE)(0, (__global DATA_TYPE *)biases.ptr); accum_value = biases_value + accum_value; // Store result in the accumulate buffer VSTORE(VECTOR_SIZE) (accum_value, 0, (__global DATA_TYPE *)accum.ptr); } #endif // defined(DATA_TYPE) && defined(VECTOR_SIZE)