diff options
Diffstat (limited to 'src/core/CL/cl_kernels/gemm.cl')
| -rw-r--r-- | src/core/CL/cl_kernels/gemm.cl | 835 |
1 files changed, 749 insertions, 86 deletions
diff --git a/src/core/CL/cl_kernels/gemm.cl b/src/core/CL/cl_kernels/gemm.cl index b0b8b2c6b0..3c017325b1 100644 --- a/src/core/CL/cl_kernels/gemm.cl +++ b/src/core/CL/cl_kernels/gemm.cl @@ -1293,94 +1293,423 @@ __kernel void gemm_mm_reshaped_only_rhs_t(IMAGE_DECLARATION(lhs), #undef RHS_STEP_X } +#if defined(OPENCL_IMAGE_SUPPORT) +/** This OpenCL kernel computes the matrix multiplication between 2 matrices. The RHS matrix is stored in OpenCL image + * The LHS matrix is NOT reshaped + * The RHS is reshaped with @ref CLGEMMReshapeRHSMatrixKernel and the block K0xN0 is transposed + * + * @note -DOPENCL_IMAGE_SUPPORT must be passed at compile time in order to compile this OpenCL kernel + * @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 height of the RHS matrix, defined before creating the OpenCL image object from the OpenCL buffer, should be passed at compile time using -DRHS_HEIGHT=<value> (e.g. -DRHS_HEIGHT=32) + * Since we cannot create a 3d image from a buffer, the third dimension could be collapsed with the second dimension so RHS_HEIGHT + * could be different from the value returned by get_image_height(rhs_img). + * @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 = 4, 8, 16 + * - K0 = 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 matrix. Supported data type: F32 + * @param[in] lhs_stride_x Stride of the LHS 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 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 matrix + * @param[in] rhs_img The RHS reshaped matrix as OpenCL image object. Supported data type: same as @p lhs_ptr + * @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 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_texture(IMAGE_DECLARATION(lhs), + __read_only image2d_t rhs_img, +#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 + ) +{ + // Pixel unit +#define PIXEL_UNIT CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(K0) + +#define LEFTOVER_K (K % K0) + + // Block size +#define RHS_BLOCK_SIZE (PIXEL_UNIT * (N0)) + + // RHS offset and step X +#if defined(RHS_INTERLEAVE) +#define RHS_OFFSET_X (PIXEL_UNIT) +#define RHS_STEP_X (PIXEL_UNIT * (H0)) +#define RHS_STEP_LOOP (1) +#else // defined(RHS_INTERLEAVE) +#define RHS_OFFSET_X (RHS_BLOCK_SIZE) +#define RHS_STEP_X PIXEL_UNIT +#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; + +#if defined(MATRIX_B_DEPTH) + // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 + const uint z_rhs = (get_global_id(2) % MATRIX_B_DEPTH); +#else // defined(MATRIX_B_DEPTH) + const uint z_rhs = get_global_id(2); +#endif // defined(MATRIX_B_DEPTH) + + // Compute RHS matrix coordinates + uint x_rhs = (get_global_id(0) % H0) * (uint)RHS_OFFSET_X; + const uint y_rhs = (get_global_id(0) / (uint)H0) + z_rhs * RHS_HEIGHT; + + 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); + + int i = 0; + for(; i <= (K - K0); i += K0) + { + // 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 stored in a cl_image + REPEAT_VAR_INIT_TO_CONST(N0, VEC_DATA_TYPE(DATA_TYPE, K0), b, 0); + LOAD_TEXTURE2D(N0, PIXEL_UNIT, DATA_TYPE, b, rhs_img, x_rhs, y_rhs, RHS_STEP_X, 0); + + // 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); + x_rhs += N0 * RHS_STEP_X * RHS_STEP_LOOP; + } + +#if LEFTOVER_K != 0 + // Note: We cannot read out-of-bound elements from the RHS matrix because + // the RHS width is always multiple of K0. This is not be true for the LHS matrix + + union UNION_VEC_TYPE + { + DATA_TYPE s[K0]; + VEC_DATA_TYPE(DATA_TYPE, K0) + v; + }; + + union UNION_VEC_TYPE a0 = {.v = 0 }; +#if M0 > 1 + union UNION_VEC_TYPE a1 = {.v = 0 }; +#endif // M0 > 1 +#if M0 > 2 + union UNION_VEC_TYPE a2 = {.v = 0 }; +#endif // M0 > 2 +#if M0 > 3 + union UNION_VEC_TYPE a3 = {.v = 0 }; +#endif // M0 > 3 +#if M0 > 4 + union UNION_VEC_TYPE a4 = {.v = 0 }; +#endif // M0 > 4 +#if M0 > 5 + union UNION_VEC_TYPE a5 = {.v = 0 }; +#endif // M0 > 5 +#if M0 > 6 + union UNION_VEC_TYPE a6 = {.v = 0 }; +#endif // M0 > 6 +#if M0 > 7 + union UNION_VEC_TYPE a7 = {.v = 0 }; +#endif // M0 > 7 + + REPEAT_VAR_INIT_TO_CONST(N0, VEC_DATA_TYPE(DATA_TYPE, K0), b, 0); + + // Load from RHS matrix + LOAD_TEXTURE2D(N0, PIXEL_UNIT, DATA_TYPE, b, rhs_img, x_rhs, y_rhs, RHS_STEP_X, 0); + + // Load from LHS matrix + for(int k = 0; k < LEFTOVER_K; ++k) + { + a0.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 0 * lhs_stride_y + zlhs0); +#if M0 > 1 + a1.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 1 * lhs_stride_y + zlhs1); +#endif // M0 > 1 +#if M0 > 2 + a2.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 2 * lhs_stride_y + zlhs2); +#endif // M0 > 2 +#if M0 > 3 + a3.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 3 * lhs_stride_y + zlhs3); +#endif // M0 > 3 +#if M0 > 4 + a4.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 4 * lhs_stride_y + zlhs4); +#endif // M0 > 4 +#if M0 > 5 + a5.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 5 * lhs_stride_y + zlhs5); +#endif // M0 > 5 +#if M0 > 6 + a6.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 6 * lhs_stride_y + zlhs6); +#endif // M0 > 6 +#if M0 > 7 + a7.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zlhs7); +#endif // M0 > 7 + + lhs_offset += sizeof(DATA_TYPE); + } + + // Accumulate + ARM_DOT_K0XN0(K0, a0.v, b, c0); +#if M0 > 1 + ARM_DOT_K0XN0(K0, a1.v, b, c1); +#endif // M0 > 1 +#if M0 > 2 + ARM_DOT_K0XN0(K0, a2.v, b, c2); +#endif // M0 > 2 +#if M0 > 3 + ARM_DOT_K0XN0(K0, a3.v, b, c3); +#endif // M0 > 3 +#if M0 > 4 + ARM_DOT_K0XN0(K0, a4.v, b, c4); +#endif // M0 > 4 +#if M0 > 5 + ARM_DOT_K0XN0(K0, a5.v, b, c5); +#endif // M0 > 5 +#if M0 > 6 + ARM_DOT_K0XN0(K0, a6.v, b, c6); +#endif // M0 > 6 +#if M0 > 7 + ARM_DOT_K0XN0(K0, a7.v, b, c7); +#endif // M0 > 7 + +#endif // LEFTOVER_K != 0 + + __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); //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 +#undef LEFTOVER_K +#undef PIXEL_UNIT +} +#endif // defined(OPENCL_IMAGE_SUPPORT) + #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)); \ +#define 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 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)); \ +#define 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 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)); \ +#define 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 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)); \ +#define 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 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)); \ +#define 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 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)); \ +#define 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 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)); \ +#define 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 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)); \ +#define 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" @@ -1539,29 +1868,48 @@ __kernel void gemm_mm_reshaped_only_rhs_nt(IMAGE_DECLARATION(lhs), // 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); + VEC_DATA_TYPE(DATA_TYPE, N0) + b0; + + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(0, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 1 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(1, a, b0, c); #if K0 > 2 - LD_RHS_VFMA_M0xN0(2, a, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 2 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(2, a, b0, c); #endif // K0 > 2 #if K0 > 3 - LD_RHS_VFMA_M0xN0(3, a, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 3 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(3, a, b0, 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); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 4 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(4, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 5 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(5, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 6 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(6, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 7 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(7, a, b0, 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); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 8 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(8, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 9 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(9, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 10 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(A, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 11 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(B, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 12 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(C, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 13 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(D, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 14 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(E, a, b0, c); + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 15 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(F, a, b0, c); #endif // K0 > 8 lhs_offset += K0 * sizeof(DATA_TYPE); @@ -1603,7 +1951,11 @@ __kernel void gemm_mm_reshaped_only_rhs_nt(IMAGE_DECLARATION(lhs), a7 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zin7)); #endif // M0 > 7 - LD_RHS_VFMA_M0xN0(0, a, c); + VEC_DATA_TYPE(DATA_TYPE, N0) + b0; + + b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0 * RHS_STEP_X * sizeof(DATA_TYPE))); + VFMA_M0xN0(0, a, b0, c); lhs_offset += sizeof(DATA_TYPE); rhs_offset += RHS_STEP_X * sizeof(DATA_TYPE); @@ -1674,6 +2026,312 @@ __kernel void gemm_mm_reshaped_only_rhs_nt(IMAGE_DECLARATION(lhs), #undef RHS_OFFSET_X #undef RHS_STEP_X } + +#if defined(OPENCL_IMAGE_SUPPORT) +/** 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 -DOPENCL_IMAGE_SUPPORT must be passed at compile time in order to compile this OpenCL kernel + * @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 height of the RHS matrix, defined before creating the OpenCL image object from the OpenCL buffer, should be passed at compile time using -DRHS_HEIGHT=<value> (e.g. -DRHS_HEIGHT=32) + * Since we cannot create a 3d image from a buffer, the third dimension could be collapsed with the second dimension so RHS_HEIGHT + * could be different from the value returned by get_image_height(rhs_img). + * @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 = 4, 8, 16 + * - K0 = 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 matrix. Supported data type: F32 + * @param[in] lhs_stride_x Stride of the LHS 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 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 matrix + * @param[in] rhs_img The RHS reshaped matrix as OpenCL image object. Supported data type: same as @p lhs_ptr + * @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 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_texture(IMAGE_DECLARATION(lhs), + __read_only image2d_t rhs_img, +#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 + ) +{ + // Pixel unit +#define PIXEL_UNIT CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(N0) + + // Block size +#define RHS_BLOCK_SIZE ((K0) * (PIXEL_UNIT)) + + // RHS offset and step X +#if defined(RHS_INTERLEAVE) +#define RHS_OFFSET_X (PIXEL_UNIT) +#define RHS_STEP_X ((PIXEL_UNIT) * (H0)) +#else // defined(RHS_INTERLEAVE) +#define RHS_OFFSET_X (RHS_BLOCK_SIZE) +#define RHS_STEP_X (PIXEL_UNIT) +#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; + +#if defined(MATRIX_B_DEPTH) + // Do not slide matrix B if the matrix B has 3 dimensions and matrix A more than 3 + const uint z_rhs = (z % MATRIX_B_DEPTH); +#else // defined(MATRIX_B_DEPTH) + const uint z_rhs = z; +#endif // defined(MATRIX_B_DEPTH) + + // Compute RHS matrix coordinates + uint x_rhs = (x % H0) * (uint)RHS_OFFSET_X; + const uint y_rhs = (x / (uint)H0) + z_rhs * RHS_HEIGHT; + + REPEAT_VAR_INIT_TO_CONST(8, uint, zin, 0); + REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 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); + + int i = 0; + for(; i <= (K - K0); i += K0) + { + // Load values from LHS matrix + LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zin); + + VEC_DATA_TYPE(DATA_TYPE, N0) + b0; + + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 0 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(0, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 1 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(1, a, b0, c); +#if K0 > 2 + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 2 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(2, a, b0, c); +#endif // K0 > 2 +#if K0 > 3 + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 3 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(3, a, b0, c); +#endif // K0 > 3 +#if K0 > 4 + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 4 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(4, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 5 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(5, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 6 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(6, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 7 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(7, a, b0, c); +#endif // K0 > 4 +#if K0 > 8 + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 8 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(8, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 9 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(9, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 10 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(A, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 11 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(B, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 12 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(C, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 13 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(D, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 14 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(E, a, b0, c); + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 15 * RHS_STEP_X), (y_rhs)); + VFMA_M0xN0(F, a, b0, c); +#endif // K0 > 8 + + lhs_offset += K0 * sizeof(DATA_TYPE); + x_rhs += K0 * RHS_STEP_X * RHS_STEP_LOOP; + } + + // 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 + + VEC_DATA_TYPE(DATA_TYPE, N0) + b0; + b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 0 * RHS_STEP_X), (y_rhs)); + + VFMA_M0xN0(0, a, b0, c); + + lhs_offset += sizeof(DATA_TYPE); + x_rhs += RHS_STEP_X; + } + + __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(OPENCL_IMAGE_SUPPORT) #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(DATA_TYPE_ACCUMULATOR) && defined(M) && defined(N) @@ -2129,6 +2787,9 @@ __kernel void gemm_mm_reshaped_lhs_nt_rhs_t(IMAGE_DECLARATION(lhs), * @note The F16 computation also supports mixed precision through the option -DMIXED_PRECISION passed at compile time. If enabled, DATA_TYPE_ACCUMULATOR should be set to float * @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 -DK (e.g. -DM=52, -DN=90 and -DK=24). + * @note The height of the RHS matrix, defined before creating the OpenCL image object from the OpenCL buffer, should be passed at compile time using -DRHS_HEIGHT=<value> (e.g. -DRHS_HEIGHT=32) + * Since we cannot create a 3d image from a buffer, the third dimension could be collapsed with the second dimension so RHS_HEIGHT + * could be different from the value returned by get_image_height(rhs_img). * @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) @@ -2871,9 +3532,11 @@ __kernel void gemm_mm_reshaped_lhs_t_rhs_nt(IMAGE_DECLARATION(lhs), * * @note -DOPENCL_IMAGE_SUPPORT must be passed at compile time in order to compile this OpenCL kernel * @note LHS_TRANSPOSE should be passed at compile time in order to compile this OpenCL kernel (e.g. -DLHS_TRANSPOSE). - * @note The height of the RHS matrix should be passed at compile time using -DRHS_HEIGHT=<value> (e.g. -DRHS_HEIGHT=32) * @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 -DK (e.g. -DM=52, -DN=90 and -DK=24). + * @note The height of the RHS matrix, defined before creating the OpenCL image object from the OpenCL buffer, should be passed at compile time using -DRHS_HEIGHT=<value> (e.g. -DRHS_HEIGHT=32) + * Since we cannot create a 3d image from a buffer, the third dimension could be collapsed with the second dimension so RHS_HEIGHT + * could be different from the value returned by get_image_height(rhs_img). * @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) |