Gitiles
Code Review Sign In
review.mlplatform.org / ml / ComputeLibrary / 7485d5a62685cb745ab50e970adb722cb71557ac / . / src / core / CL / cl_kernels / gemm.cl
blob: f75161ca0a4edc19f26ec3a5a4087244f2eac044 [file] [log] [blame]
/*
* Copyright (c) 2017-2018 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include "helpers.h"
#if defined(TRANSPOSE_W) && defined(MULT_TRANSPOSE1XW_WIDTH)
#if ELEMENT_SIZE == 1
#define DATA_TYPE uchar
#elif ELEMENT_SIZE == 2
#define DATA_TYPE ushort
#elif ELEMENT_SIZE == 4
#define DATA_TYPE uint
#else // ELEMENT_SIZE == 1
#error "Element size not supported"
#endif // ELEMENT_SIZE
/** This OpenCL kernel computes the "vector" 1xW transposition of input matrix
*
* @note The transposition width must be passed at compile time using -DTRANSPOSE_W (i.e. -DTRANSPOSE_W)
* @note The multiplication factor for the transposition width (mult_transpose1xW_width) must be passed at compile time using -DMULT_TRANSPOSE1XW_WIDTH (i.e. -DMULT_TRANSPOSE1XW_WIDTH=2)
*
* @param[in] src_ptr Pointer to the source matrix. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/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 source 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 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_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_transpose1xW(TENSOR3D_DECLARATION(src),
TENSOR3D_DECLARATION(dst))
{
uint x = get_global_id(0);
uint y = get_global_id(1);
uint z = get_global_id(2);
// Compute address for Matrix B - source
Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
// Compute address for Matrix B transposed - destination. X and Y are swapped
uint dst_addr_in_bytes = dst_offset_first_element_in_bytes + y * TRANSPOSE_W * sizeof(DATA_TYPE) * MULT_TRANSPOSE1XW_WIDTH + (x / MULT_TRANSPOSE1XW_WIDTH) * dst_stride_y +
(x % MULT_TRANSPOSE1XW_WIDTH) * TRANSPOSE_W * sizeof(DATA_TYPE);
// Add offset for batched GEMM
dst_addr_in_bytes += z * dst_stride_z;
VEC_DATA_TYPE(DATA_TYPE, TRANSPOSE_W)
b0 = VLOAD(TRANSPOSE_W)(0, (__global DATA_TYPE *)src.ptr);
VSTORE(TRANSPOSE_W)
(b0, 0, (__global DATA_TYPE *)(dst_ptr + dst_addr_in_bytes));
}
#endif // defined(TRANSPOSE_W) && defined(MULT_TRANSPOSE1XW_WIDTH)
#if defined(MULT_INTERLEAVE4X4_HEIGHT) && defined(DATA_TYPE)
/** This OpenCL kernel reshapes the input matrix transposing each 4x4 block and interleaving the values
*
* @note The data type must be passed at compile time using -DDATA_TYPE (i.e. -DDATA_TYPE=float)
* @note The multiplication factor for the height of the 4x4 interleaved block must be passed at compile time using -DMULT_INTERLEAVE4X4_HEIGHT (i.e. -DMULT_INTERLEAVE4X4_HEIGHT=2)
*
* @param[in] src_ptr Pointer to the source matrix. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/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 source 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 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_interleave4x4(TENSOR3D_DECLARATION(src),
TENSOR3D_DECLARATION(dst))
{
// Compute source and destination addresses
uint x = get_global_id(0);
uint y = get_global_id(1);
uint z = get_global_id(2);
// Compute address for source tensor
Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
// Compute address for Matrix B transposed - destination. X and Y are swapped
uint dst_addr_in_bytes = dst_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * 16 * MULT_INTERLEAVE4X4_HEIGHT + (y / MULT_INTERLEAVE4X4_HEIGHT) * dst_stride_y +
(y % MULT_INTERLEAVE4X4_HEIGHT) * 4 * sizeof(DATA_TYPE);
// Add offset for batched GEMM
dst_addr_in_bytes += z * dst_stride_z;
__global uchar *input_ptr = src.ptr;
// Load values from Matrix A
VEC_DATA_TYPE(DATA_TYPE, 4)
a0 = vload4(0, (__global DATA_TYPE *)(input_ptr + 0 * src_stride_y));
VEC_DATA_TYPE(DATA_TYPE, 4)
a1 = vload4(0, (__global DATA_TYPE *)(input_ptr + 1 * src_stride_y));
VEC_DATA_TYPE(DATA_TYPE, 4)
a2 = vload4(0, (__global DATA_TYPE *)(input_ptr + 2 * src_stride_y));
VEC_DATA_TYPE(DATA_TYPE, 4)
a3 = vload4(0, (__global DATA_TYPE *)(input_ptr + 3 * src_stride_y));
VEC_DATA_TYPE(DATA_TYPE, 4)
val0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s0, a1.s0, a2.s0, a3.s0);
vstore4(val0, 0, ((__global DATA_TYPE *)(dst_ptr + dst_addr_in_bytes) + 0 * MULT_INTERLEAVE4X4_HEIGHT));
val0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s1, a1.s1, a2.s1, a3.s1);
vstore4(val0, 0, ((__global DATA_TYPE *)(dst_ptr + dst_addr_in_bytes) + 4 * MULT_INTERLEAVE4X4_HEIGHT));
val0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s2, a1.s2, a2.s2, a3.s2);
vstore4(val0, 0, ((__global DATA_TYPE *)(dst_ptr + dst_addr_in_bytes) + 8 * MULT_INTERLEAVE4X4_HEIGHT));
val0 = (VEC_DATA_TYPE(DATA_TYPE, 4))(a0.s3, a1.s3, a2.s3, a3.s3);
vstore4(val0, 0, ((__global DATA_TYPE *)(dst_ptr + dst_addr_in_bytes) + 12 * MULT_INTERLEAVE4X4_HEIGHT));
}
#endif // defined(MULT_INTERLEAVE4X4_HEIGHT) && 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 (src0) and matrix B (src1)
* Matrix A and matrix B must be reshaped respectively with @ref gemm_interleave4x4_32bit and @ref gemm_transpose1x4 before running the matrix multiplication
*
* @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 (i.e. -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 (i.e. -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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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[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] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
* @param[in] pad_bottom 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),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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 c00 = 0.0f;
float4 c10 = 0.0f;
float4 c20 = 0.0f;
float4 c30 = 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);
c00 += (float4)a0.s0 * b0;
c10 += (float4)a0.s1 * b0;
c20 += (float4)a0.s2 * b0;
c30 += (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);
c00 += (float4)a0.s0 * b0;
c10 += (float4)a0.s1 * b0;
c20 += (float4)a0.s2 * b0;
c30 += (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);
c00 += (float4)a0.s0 * b0;
c10 += (float4)a0.s1 * b0;
c20 += (float4)a0.s2 * b0;
c30 += (float4)a0.s3 * b0;
}
// Compute destination address
Image dst = CONVERT_TO_IMAGE_STRUCT(dst);
#if defined(ALPHA)
// Multiply by the weight of matrix product
c00 = c00 * (float4)ALPHA;
c10 = c10 * (float4)ALPHA;
c20 = c20 * (float4)ALPHA;
c30 = c30 * (float4)ALPHA;
#endif // defined(ALPHA)
// Compute dst address
__global uchar *dst_addr = offset(&dst, 0, 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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store 4x4 block
vstore4(c00, 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0));
vstore4(c10, 0, (__global float *)(dst_addr + 1 * dst_stride_y + zout.s1));
vstore4(c20, 0, (__global float *)(dst_addr + 2 * dst_stride_y + zout.s2));
vstore4(c30, 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3));
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store 4x4 block
vstore4(c00, 0, (__global float *)(dst_addr + 0 * dst_stride_y));
vstore4(c10, 0, (__global float *)(dst_addr + 1 * dst_stride_y));
vstore4(c20, 0, (__global float *)(dst_addr + 2 * dst_stride_y));
vstore4(c30, 0, (__global float *)(dst_addr + 3 * dst_stride_y));
#endif // defined(REINTERPRET_OUTPUT_AS_3D)
}
/** This OpenCL kernel is optimized for Bifrost. It computes the matrix multiplication between matrix A (src0) and matrix B (src1)
* Matrix A and matrix B must be reshaped respectively with @ref gemm_interleave4x4_32bit and @ref gemm_transpose1x4 before running the matrix multiplication
*
* @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 (i.e. -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 (i.e. -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 (i.e. -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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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[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] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
* @param[in] pad_bottom 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),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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
float c00 = 0.0f;
float c01 = 0.0f;
float c02 = 0.0f;
float c03 = 0.0f;
float c10 = 0.0f;
float c11 = 0.0f;
float c12 = 0.0f;
float c13 = 0.0f;
float c20 = 0.0f;
float c21 = 0.0f;
float c22 = 0.0f;
float c23 = 0.0f;
float c30 = 0.0f;
float c31 = 0.0f;
float c32 = 0.0f;
float c33 = 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;
c00 = fma(a0.s0, b0.s0, c00);
c01 = fma(a0.s0, b0.s1, c01);
c02 = fma(a0.s0, b0.s2, c02);
c03 = fma(a0.s0, b0.s3, c03);
c10 = fma(a0.s1, b0.s0, c10);
c11 = fma(a0.s1, b0.s1, c11);
c12 = fma(a0.s1, b0.s2, c12);
c13 = fma(a0.s1, b0.s3, c13);
c20 = fma(a0.s2, b0.s0, c20);
c21 = fma(a0.s2, b0.s1, c21);
c22 = fma(a0.s2, b0.s2, c22);
c23 = fma(a0.s2, b0.s3, c23);
c30 = fma(a0.s3, b0.s0, c30);
c31 = fma(a0.s3, b0.s1, c31);
c32 = fma(a0.s3, b0.s2, c32);
c33 = fma(a0.s3, b0.s3, c33);
// 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;
c00 = fma(a0.s0, b0.s0, c00);
c01 = fma(a0.s0, b0.s1, c01);
c02 = fma(a0.s0, b0.s2, c02);
c03 = fma(a0.s0, b0.s3, c03);
c10 = fma(a0.s1, b0.s0, c10);
c11 = fma(a0.s1, b0.s1, c11);
c12 = fma(a0.s1, b0.s2, c12);
c13 = fma(a0.s1, b0.s3, c13);
c20 = fma(a0.s2, b0.s0, c20);
c21 = fma(a0.s2, b0.s1, c21);
c22 = fma(a0.s2, b0.s2, c22);
c23 = fma(a0.s2, b0.s3, c23);
c30 = fma(a0.s3, b0.s0, c30);
c31 = fma(a0.s3, b0.s1, c31);
c32 = fma(a0.s3, b0.s2, c32);
c33 = fma(a0.s3, b0.s3, c33);
// 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;
c00 = fma(a0.s0, b0.s0, c00);
c01 = fma(a0.s0, b0.s1, c01);
c02 = fma(a0.s0, b0.s2, c02);
c03 = fma(a0.s0, b0.s3, c03);
c10 = fma(a0.s1, b0.s0, c10);
c11 = fma(a0.s1, b0.s1, c11);
c12 = fma(a0.s1, b0.s2, c12);
c13 = fma(a0.s1, b0.s3, c13);
c20 = fma(a0.s2, b0.s0, c20);
c21 = fma(a0.s2, b0.s1, c21);
c22 = fma(a0.s2, b0.s2, c22);
c23 = fma(a0.s2, b0.s3, c23);
c30 = fma(a0.s3, b0.s0, c30);
c31 = fma(a0.s3, b0.s1, c31);
c32 = fma(a0.s3, b0.s2, c32);
c33 = fma(a0.s3, b0.s3, c33);
// 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;
c00 = fma(a0.s0, b0.s0, c00);
c01 = fma(a0.s0, b0.s1, c01);
c02 = fma(a0.s0, b0.s2, c02);
c03 = fma(a0.s0, b0.s3, c03);
c10 = fma(a0.s1, b0.s0, c10);
c11 = fma(a0.s1, b0.s1, c11);
c12 = fma(a0.s1, b0.s2, c12);
c13 = fma(a0.s1, b0.s3, c13);
c20 = fma(a0.s2, b0.s0, c20);
c21 = fma(a0.s2, b0.s1, c21);
c22 = fma(a0.s2, b0.s2, c22);
c23 = fma(a0.s2, b0.s3, c23);
c30 = fma(a0.s3, b0.s0, c30);
c31 = fma(a0.s3, b0.s1, c31);
c32 = fma(a0.s3, b0.s2, c32);
c33 = fma(a0.s3, b0.s3, c33);
}
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;
c00 = fma(a0.s0, b0.s0, c00);
c01 = fma(a0.s0, b0.s1, c01);
c02 = fma(a0.s0, b0.s2, c02);
c03 = fma(a0.s0, b0.s3, c03);
c10 = fma(a0.s1, b0.s0, c10);
c11 = fma(a0.s1, b0.s1, c11);
c12 = fma(a0.s1, b0.s2, c12);
c13 = fma(a0.s1, b0.s3, c13);
c20 = fma(a0.s2, b0.s0, c20);
c21 = fma(a0.s2, b0.s1, c21);
c22 = fma(a0.s2, b0.s2, c22);
c23 = fma(a0.s2, b0.s3, c23);
c30 = fma(a0.s3, b0.s0, c30);
c31 = fma(a0.s3, b0.s1, c31);
c32 = fma(a0.s3, b0.s2, c32);
c33 = fma(a0.s3, b0.s3, c33);
}
// Compute destination address
Image dst = CONVERT_TO_IMAGE_STRUCT(dst);
#if defined(ALPHA)
// Multiply by the weight of matrix product
c00 = c00 * ALPHA;
c01 = c01 * ALPHA;
c02 = c02 * ALPHA;
c03 = c03 * ALPHA;
c10 = c10 * ALPHA;
c11 = c11 * ALPHA;
c12 = c12 * ALPHA;
c13 = c13 * ALPHA;
c20 = c20 * ALPHA;
c21 = c21 * ALPHA;
c22 = c22 * ALPHA;
c23 = c23 * ALPHA;
c30 = c30 * ALPHA;
c31 = c31 * ALPHA;
c32 = c32 * ALPHA;
c33 = c33 * ALPHA;
#endif // defined(ALPHA)
// Compute dst address
__global uchar *dst_addr = offset(&dst, 0, 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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store 4x4 block
vstore4((float4)(c00, c01, c02, c03), 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0));
vstore4((float4)(c10, c11, c12, c13), 0, (__global float *)(dst_addr + 1 * dst_stride_y + zout.s1));
vstore4((float4)(c20, c21, c22, c23), 0, (__global float *)(dst_addr + 2 * dst_stride_y + zout.s2));
vstore4((float4)(c30, c31, c32, c33), 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3));
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store 4x4 block
vstore4((float4)(c00, c01, c02, c03), 0, (__global float *)(dst_addr + 0 * dst_stride_y));
vstore4((float4)(c10, c11, c12, c13), 0, (__global float *)(dst_addr + 1 * dst_stride_y));
vstore4((float4)(c20, c21, c22, c23), 0, (__global float *)(dst_addr + 2 * dst_stride_y));
vstore4((float4)(c30, c31, c32, c33), 0, (__global float *)(dst_addr + 3 * dst_stride_y));
#endif // defined(REINTERPRET_OUTPUT_AS_3D)
}
// Undefine local defines
#undef COLS_MTX_B
#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)
/** This OpenCL kernel computes the matrix multiplication between matrix A (src0) and matrix B (src1)
* Matrix A and matrix B must be reshaped respectively with @ref gemm_interleave4x4_16bit and @ref gemm_transpose1x8 before running the matrix multiplication
*
* @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 (i.e. -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 (i.e. -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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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[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] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
* @param[in] pad_bottom 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),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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 c00 = 0.0f;
half8 c10 = 0.0f;
half8 c20 = 0.0f;
half8 c30 = 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);
c00 += (half8)a0.s0 * b0;
c10 += (half8)a0.s1 * b0;
c20 += (half8)a0.s2 * b0;
c30 += (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);
c00 += (half8)a0.s0 * b0;
c10 += (half8)a0.s1 * b0;
c20 += (half8)a0.s2 * b0;
c30 += (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);
c00 += (half8)a0.s0 * b0;
c10 += (half8)a0.s1 * b0;
c20 += (half8)a0.s2 * b0;
c30 += (half8)a0.s3 * b0;
}
// Compute destination address
Image dst = CONVERT_TO_IMAGE_STRUCT(dst);
#if defined(ALPHA)
// Multiply by the weight of matrix product
c00 = c00 * (half8)ALPHA;
c10 = c10 * (half8)ALPHA;
c20 = c20 * (half8)ALPHA;
c30 = c30 * (half8)ALPHA;
#endif // defined(ALPHA)
// Compute dst address
__global uchar *dst_addr = offset(&dst, 0, 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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store 4x8 block
vstore8(c00, 0, (__global half *)(dst_addr + 0 * dst_stride_y + zout.s0));
vstore8(c10, 0, (__global half *)(dst_addr + 1 * dst_stride_y + zout.s1));
vstore8(c20, 0, (__global half *)(dst_addr + 2 * dst_stride_y + zout.s2));
vstore8(c30, 0, (__global half *)(dst_addr + 3 * dst_stride_y + zout.s3));
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store 4x8 block
vstore8(c00, 0, (__global half *)(dst_addr + 0 * dst_stride_y));
vstore8(c10, 0, (__global half *)(dst_addr + 1 * dst_stride_y));
vstore8(c20, 0, (__global half *)(dst_addr + 2 * dst_stride_y));
vstore8(c30, 0, (__global half *)(dst_addr + 3 * dst_stride_y));
#endif // defined(REINTERPRET_OUTPUT_AS_3D)
}
/** This OpenCL kernel optimized for Bifrost architectures computes the matrix multiplication between matrix A (src0) and matrix B (src1)
* Matrix A and matrix B must be reshaped respectively with @ref gemm_interleave4x4_16bit and @ref gemm_transpose1x8 before running the matrix multiplication
*
* @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 (i.e. -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 (i.e. -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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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[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] pad_bottom 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),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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 c00 = 0.0f;
half8 c10 = 0.0f;
half8 c20 = 0.0f;
half8 c30 = 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;
c00 = fma((half8)a0.s0, b0, c00);
c10 = fma((half8)a0.s1, b0, c10);
c20 = fma((half8)a0.s2, b0, c20);
c30 = fma((half8)a0.s3, b0, c30);
// Load values from matrix B (transposed)
b0 = vload8(0, src_addr_b);
src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;
c00 = fma((half8)a0.s4, b0, c00);
c10 = fma((half8)a0.s5, b0, c10);
c20 = fma((half8)a0.s6, b0, c20);
c30 = fma((half8)a0.s7, b0, c30);
// 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;
c00 = fma((half8)a0.s0, b0, c00);
c10 = fma((half8)a0.s1, b0, c10);
c20 = fma((half8)a0.s2, b0, c20);
c30 = fma((half8)a0.s3, b0, c30);
// Load values from matrix B (transposed)
b0 = vload8(0, src_addr_b);
src_addr_b += 8 * MULT_TRANSPOSE1XW_WIDTH;
c00 = fma((half8)a0.s4, b0, c00);
c10 = fma((half8)a0.s5, b0, c10);
c20 = fma((half8)a0.s6, b0, c20);
c30 = fma((half8)a0.s7, b0, c30);
#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;
c00 = fma((half8)a0.s0, b0, c00);
c10 = fma((half8)a0.s1, b0, c10);
c20 = fma((half8)a0.s2, b0, c20);
c30 = fma((half8)a0.s3, b0, c30);
// 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;
c00 = fma((half8)a0.s0, b0, c00);
c10 = fma((half8)a0.s1, b0, c10);
c20 = fma((half8)a0.s2, b0, c20);
c30 = fma((half8)a0.s3, b0, c30);
// 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;
c00 = fma((half8)a0.s0, b0, c00);
c10 = fma((half8)a0.s1, b0, c10);
c20 = fma((half8)a0.s2, b0, c20);
c30 = fma((half8)a0.s3, b0, c30);
// 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;
c00 = fma((half8)a0.s0, b0, c00);
c10 = fma((half8)a0.s1, b0, c10);
c20 = fma((half8)a0.s2, b0, c20);
c30 = fma((half8)a0.s3, b0, c30);
#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;
c00 = fma((half8)a0.s0, b0, c00);
c10 = fma((half8)a0.s1, b0, c10);
c20 = fma((half8)a0.s2, b0, c20);
c30 = fma((half8)a0.s3, b0, c30);
}
// Compute destination address
Image dst = CONVERT_TO_IMAGE_STRUCT(dst);
#if defined(ALPHA)
// Multiply by the weight of matrix product
c00 = c00 * (half8)ALPHA;
c10 = c10 * (half8)ALPHA;
c20 = c20 * (half8)ALPHA;
c30 = c30 * (half8)ALPHA;
#endif // defined(ALPHA)
// Compute dst address
__global uchar *dst_addr = offset(&dst, 0, 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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * 4) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store 4x8 block
vstore8(c00, 0, (__global half *)(dst_addr + 0 * dst_stride_y + zout.s0));
vstore8(c10, 0, (__global half *)(dst_addr + 1 * dst_stride_y + zout.s1));
vstore8(c20, 0, (__global half *)(dst_addr + 2 * dst_stride_y + zout.s2));
vstore8(c30, 0, (__global half *)(dst_addr + 3 * dst_stride_y + zout.s3));
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store 4x8 block
vstore8(c00, 0, (__global half *)(dst_addr + 0 * dst_stride_y));
vstore8(c10, 0, (__global half *)(dst_addr + 1 * dst_stride_y));
vstore8(c20, 0, (__global half *)(dst_addr + 2 * dst_stride_y));
vstore8(c30, 0, (__global half *)(dst_addr + 3 * dst_stride_y));
#endif // defined(REINTERPRET_OUTPUT_AS_3D)
}
// 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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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/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[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] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
* @param[in] pad_bottom Bottom paddings in unit of elements (only if defined REINTERPRET_OUTPUT_AS_3D)
*/
__kernel void gemm_mm_floating_point(IMAGE_DECLARATION(src0),
IMAGE_DECLARATION(src1),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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);
// Add offset for batched GEMM
src_addr.s0 += get_global_id(2) * src0_stride_z;
#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))
{
// 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
// 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))
{
// 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
// 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
}
// Compute destination address
Image dst = CONVERT_TO_IMAGE_STRUCT(dst);
// Compute dst address
__global uchar *dst_addr = offset(&dst, 0, 0);
// Multiply by the weight of matrix-matrix product and store the result
#if defined(ALPHA)
acc0 = acc0 * (VECTOR_TYPE)ALPHA;
#endif // defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
acc1 = acc1 * (VECTOR_TYPE)ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
acc2 = acc2 * (VECTOR_TYPE)ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
acc3 = acc3 * (VECTOR_TYPE)ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
int z = get_global_id(2);
#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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store output block
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc0, 0, (__global DATA_TYPE *)(dst_addr + 0 * dst_stride_y + zout.s0));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc1, 0, (__global DATA_TYPE *)(dst_addr + 1 * dst_stride_y + zout.s1));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc2, 0, (__global DATA_TYPE *)(dst_addr + 2 * dst_stride_y + zout.s2));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc3, 0, (__global DATA_TYPE *)(dst_addr + 3 * dst_stride_y + zout.s3));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store output block
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc0, 0, (__global DATA_TYPE *)(dst_addr + 0 * dst_stride_y));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc1, 0, (__global DATA_TYPE *)(dst_addr + 1 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc2, 0, (__global DATA_TYPE *)(dst_addr + 2 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
VSTORE(NUM_ELEMS_PROCESSED_PER_THREAD_X)
(acc3, 0, (__global DATA_TYPE *)(dst_addr + 3 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#endif // defined(REINTERPRET_OUTPUT_AS_3D)
}
#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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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/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[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] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
* @param[in] pad_bottom 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),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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);
// Add offset for batched GEMM
src_addr.s0 += get_global_id(2) * src0_stride_z;
#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
float acc00 = 0.0f;
float acc01 = 0.0f;
float acc02 = 0.0f;
float acc03 = 0.0f;
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
float acc10 = 0.0f;
float acc11 = 0.0f;
float acc12 = 0.0f;
float acc13 = 0.0f;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
float acc20 = 0.0f;
float acc21 = 0.0f;
float acc22 = 0.0f;
float acc23 = 0.0f;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
float acc30 = 0.0f;
float acc31 = 0.0f;
float acc32 = 0.0f;
float acc33 = 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)
{
// 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
float4 b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1));
src_addr.s1 += src1_stride_y;
// Multiply and accumulate
acc00 = fma(a0.s0, b0.s0, acc00);
acc01 = fma(a0.s0, b0.s1, acc01);
acc02 = fma(a0.s0, b0.s2, acc02);
acc03 = fma(a0.s0, b0.s3, acc03);
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
acc10 = fma(a1.s0, b0.s0, acc10);
acc11 = fma(a1.s0, b0.s1, acc11);
acc12 = fma(a1.s0, b0.s2, acc12);
acc13 = fma(a1.s0, b0.s3, acc13);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
acc20 = fma(a2.s0, b0.s0, acc20);
acc21 = fma(a2.s0, b0.s1, acc21);
acc22 = fma(a2.s0, b0.s2, acc22);
acc23 = fma(a2.s0, b0.s3, acc23);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
acc30 = fma(a3.s0, b0.s0, acc30);
acc31 = fma(a3.s0, b0.s1, acc31);
acc32 = fma(a3.s0, b0.s2, acc32);
acc33 = fma(a3.s0, b0.s3, acc33);
#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
acc00 = fma(a0.s1, b0.s0, acc00);
acc01 = fma(a0.s1, b0.s1, acc01);
acc02 = fma(a0.s1, b0.s2, acc02);
acc03 = fma(a0.s1, b0.s3, acc03);
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
acc10 = fma(a1.s1, b0.s0, acc10);
acc11 = fma(a1.s1, b0.s1, acc11);
acc12 = fma(a1.s1, b0.s2, acc12);
acc13 = fma(a1.s1, b0.s3, acc13);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
acc20 = fma(a2.s1, b0.s0, acc20);
acc21 = fma(a2.s1, b0.s1, acc21);
acc22 = fma(a2.s1, b0.s2, acc22);
acc23 = fma(a2.s1, b0.s3, acc23);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
acc30 = fma(a3.s1, b0.s0, acc30);
acc31 = fma(a3.s1, b0.s1, acc31);
acc32 = fma(a3.s1, b0.s2, acc32);
acc33 = fma(a3.s1, b0.s3, acc33);
#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
acc00 = fma(a0.s2, b0.s0, acc00);
acc01 = fma(a0.s2, b0.s1, acc01);
acc02 = fma(a0.s2, b0.s2, acc02);
acc03 = fma(a0.s2, b0.s3, acc03);
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
acc10 = fma(a1.s2, b0.s0, acc10);
acc11 = fma(a1.s2, b0.s1, acc11);
acc12 = fma(a1.s2, b0.s2, acc12);
acc13 = fma(a1.s2, b0.s3, acc13);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
acc20 = fma(a2.s2, b0.s0, acc20);
acc21 = fma(a2.s2, b0.s1, acc21);
acc22 = fma(a2.s2, b0.s2, acc22);
acc23 = fma(a2.s2, b0.s3, acc23);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
acc30 = fma(a3.s2, b0.s0, acc30);
acc31 = fma(a3.s2, b0.s1, acc31);
acc32 = fma(a3.s2, b0.s2, acc32);
acc33 = fma(a3.s2, b0.s3, acc33);
#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
acc00 = fma(a0.s3, b0.s0, acc00);
acc01 = fma(a0.s3, b0.s1, acc01);
acc02 = fma(a0.s3, b0.s2, acc02);
acc03 = fma(a0.s3, b0.s3, acc03);
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
acc10 = fma(a1.s3, b0.s0, acc10);
acc11 = fma(a1.s3, b0.s1, acc11);
acc12 = fma(a1.s3, b0.s2, acc12);
acc13 = fma(a1.s3, b0.s3, acc13);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
acc20 = fma(a2.s3, b0.s0, acc20);
acc21 = fma(a2.s3, b0.s1, acc21);
acc22 = fma(a2.s3, b0.s2, acc22);
acc23 = fma(a2.s3, b0.s3, acc23);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
acc30 = fma(a3.s3, b0.s0, acc30);
acc31 = fma(a3.s3, b0.s1, acc31);
acc32 = fma(a3.s3, b0.s2, acc32);
acc33 = fma(a3.s3, b0.s3, acc33);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
src_addr.s0 += 4 * sizeof(float);
}
for(; i < (int)COLS_A; ++i)
{
// Load values from matrix A
float a0 = *((__global float *)(src0_ptr + src_addr.s0));
#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
// 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
acc00 = fma(a0, b0.s0, acc00);
acc01 = fma(a0, b0.s1, acc01);
acc02 = fma(a0, b0.s2, acc02);
acc03 = fma(a0, b0.s3, acc03);
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
acc10 = fma(a1, b0.s0, acc10);
acc11 = fma(a1, b0.s1, acc11);
acc12 = fma(a1, b0.s2, acc12);
acc13 = fma(a1, b0.s3, acc13);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
acc20 = fma(a2, b0.s0, acc20);
acc21 = fma(a2, b0.s1, acc21);
acc22 = fma(a2, b0.s2, acc22);
acc23 = fma(a2, b0.s3, acc23);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
acc30 = fma(a3, b0.s0, acc30);
acc31 = fma(a3, b0.s1, acc31);
acc32 = fma(a3, b0.s2, acc32);
acc33 = fma(a3, b0.s3, acc33);
#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);
// Multiply by the weight of matrix-matrix product and store the result
#if defined(ALPHA)
acc00 = acc00 * ALPHA;
acc01 = acc01 * ALPHA;
acc02 = acc02 * ALPHA;
acc03 = acc03 * ALPHA;
#endif // defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
acc10 = acc10 * ALPHA;
acc11 = acc11 * ALPHA;
acc12 = acc12 * ALPHA;
acc13 = acc13 * ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
acc20 = acc20 * ALPHA;
acc21 = acc21 * ALPHA;
acc22 = acc22 * ALPHA;
acc23 = acc23 * ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
acc30 = acc30 * ALPHA;
acc31 = acc31 * ALPHA;
acc32 = acc32 * ALPHA;
acc33 = acc33 * ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
// Compute dst address
__global uchar *dst_addr = offset(&dst, 0, 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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store the output block
vstore4((float4)(acc00, acc01, acc02, acc03), 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
vstore4((float4)(acc10, acc11, acc12, acc13), 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((float4)(acc20, acc21, acc22, acc23), 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((float4)(acc30, acc31, acc32, acc33), 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store the output block
vstore4((float4)(acc00, acc01, acc02, acc03), 0, (__global float *)(dst_addr + 0 * dst_stride_y));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
vstore4((float4)(acc10, acc11, acc12, acc13), 0, (__global float *)(dst_addr + 1 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
vstore4((float4)(acc20, acc21, acc22, acc23), 0, (__global float *)(dst_addr + 2 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
vstore4((float4)(acc30, acc31, acc32, acc33), 0, (__global float *)(dst_addr + 3 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#endif // defined(REINTERPRET_OUTPUT_AS_3D)
}
/** 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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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/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[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] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
* @param[in] pad_bottom 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),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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);
// Add offset for batched GEMM
src_addr.s0 += get_global_id(2) * src0_stride_z;
#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
float acc00 = 0.0f;
float acc01 = 0.0f;
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
float acc10 = 0.0f;
float acc11 = 0.0f;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
float acc20 = 0.0f;
float acc21 = 0.0f;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
float acc30 = 0.0f;
float acc31 = 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)
{
// Load values from matrix A
float8 a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0));
// 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
acc00 = fma(a0.s0, b0.s0, acc00);
acc00 = fma(a0.s1, b1.s0, acc00);
acc00 = fma(a0.s2, b2.s0, acc00);
acc00 = fma(a0.s3, b3.s0, acc00);
acc00 = fma(a0.s4, b4.s0, acc00);
acc00 = fma(a0.s5, b5.s0, acc00);
acc00 = fma(a0.s6, b6.s0, acc00);
acc00 = fma(a0.s7, b7.s0, acc00);
acc01 = fma(a0.s0, b0.s1, acc01);
acc01 = fma(a0.s1, b1.s1, acc01);
acc01 = fma(a0.s2, b2.s1, acc01);
acc01 = fma(a0.s3, b3.s1, acc01);
acc01 = fma(a0.s4, b4.s1, acc01);
acc01 = fma(a0.s5, b5.s1, acc01);
acc01 = fma(a0.s6, b6.s1, acc01);
acc01 = fma(a0.s7, b7.s1, acc01);
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));
acc10 = fma(a0.s0, b0.s0, acc10);
acc10 = fma(a0.s1, b1.s0, acc10);
acc10 = fma(a0.s2, b2.s0, acc10);
acc10 = fma(a0.s3, b3.s0, acc10);
acc10 = fma(a0.s4, b4.s0, acc10);
acc10 = fma(a0.s5, b5.s0, acc10);
acc10 = fma(a0.s6, b6.s0, acc10);
acc10 = fma(a0.s7, b7.s0, acc10);
acc11 = fma(a0.s0, b0.s1, acc11);
acc11 = fma(a0.s1, b1.s1, acc11);
acc11 = fma(a0.s2, b2.s1, acc11);
acc11 = fma(a0.s3, b3.s1, acc11);
acc11 = fma(a0.s4, b4.s1, acc11);
acc11 = fma(a0.s5, b5.s1, acc11);
acc11 = fma(a0.s6, b6.s1, acc11);
acc11 = fma(a0.s7, b7.s1, acc11);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));
acc20 = fma(a0.s0, b0.s0, acc20);
acc20 = fma(a0.s1, b1.s0, acc20);
acc20 = fma(a0.s2, b2.s0, acc20);
acc20 = fma(a0.s3, b3.s0, acc20);
acc20 = fma(a0.s4, b4.s0, acc20);
acc20 = fma(a0.s5, b5.s0, acc20);
acc20 = fma(a0.s6, b6.s0, acc20);
acc20 = fma(a0.s7, b7.s0, acc20);
acc21 = fma(a0.s0, b0.s1, acc21);
acc21 = fma(a0.s1, b1.s1, acc21);
acc21 = fma(a0.s2, b2.s1, acc21);
acc21 = fma(a0.s3, b3.s1, acc21);
acc21 = fma(a0.s4, b4.s1, acc21);
acc21 = fma(a0.s5, b5.s1, acc21);
acc21 = fma(a0.s6, b6.s1, acc21);
acc21 = fma(a0.s7, b7.s1, acc21);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
a0 = vload8(0, (__global float *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));
acc30 = fma(a0.s0, b0.s0, acc30);
acc30 = fma(a0.s1, b1.s0, acc30);
acc30 = fma(a0.s2, b2.s0, acc30);
acc30 = fma(a0.s3, b3.s0, acc30);
acc30 = fma(a0.s4, b4.s0, acc30);
acc30 = fma(a0.s5, b5.s0, acc30);
acc30 = fma(a0.s6, b6.s0, acc30);
acc30 = fma(a0.s7, b7.s0, acc30);
acc31 = fma(a0.s0, b0.s1, acc31);
acc31 = fma(a0.s1, b1.s1, acc31);
acc31 = fma(a0.s2, b2.s1, acc31);
acc31 = fma(a0.s3, b3.s1, acc31);
acc31 = fma(a0.s4, b4.s1, acc31);
acc31 = fma(a0.s5, b5.s1, acc31);
acc31 = fma(a0.s6, b6.s1, acc31);
acc31 = fma(a0.s7, b7.s1, acc31);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
src_addr.s0 += sizeof(float) * 8;
}
// float size increment
for(; i < (int)COLS_A; ++i)
{
// 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
// 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
acc00 = fma(a0, b0.s0, acc00);
acc01 = fma(a0, b0.s1, acc01);
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
acc10 = fma(a1, b0.s0, acc10);
acc11 = fma(a1, b0.s1, acc11);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
acc20 = fma(a2, b0.s0, acc20);
acc21 = fma(a2, b0.s1, acc21);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
acc30 = fma(a3, b0.s0, acc30);
acc31 = fma(a3, b0.s1, acc31);
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
src_addr.s0 += sizeof(float);
}
// Multiply by the weight of matrix-matrix product and store the result
#if defined(ALPHA)
acc00 = acc00 * ALPHA;
acc01 = acc01 * ALPHA;
#endif // defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
acc10 = acc10 * ALPHA;
acc11 = acc11 * ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
acc20 = acc20 * ALPHA;
acc21 = acc21 * ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
acc30 = acc30 * ALPHA;
acc31 = acc31 * ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
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);
#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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store the output block
vstore2((float2)(acc00, acc01), 0, (__global float *)(dst_addr + 0 * dst_stride_y + zout.s0));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
vstore2((float2)(acc10, acc11), 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((float2)(acc20, acc21), 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((float2)(acc30, acc31), 0, (__global float *)(dst_addr + 3 * dst_stride_y + zout.s3));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store the output block
vstore2((float2)(acc00, acc01), 0, (__global float *)(dst_addr + 0 * dst_stride_y));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
vstore2((float2)(acc10, acc11), 0, (__global float *)(dst_addr + 1 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
vstore2((float2)(acc20, acc21), 0, (__global float *)(dst_addr + 2 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
vstore2((float2)(acc30, acc31), 0, (__global float *)(dst_addr + 3 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#endif // defined(REINTERPRET_OUTPUT_AS_3D)
}
#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 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 (i.e. -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 (i.e. a = [K, M, 16, Batches], b = [N, K, 16])
*
* @note In case the output has to be reinterpreted as a 3D tensor (i.e. 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[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] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
* @param[in] pad_bottom 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),
IMAGE_DECLARATION(dst),
uint src0_stride_z,
uint src1_stride_z,
uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
,
uint pad_bottom
#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);
// Add offset for batched GEMM
src_addr.s0 += get_global_id(2) * src0_stride_z;
#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)
{
// 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
// 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)
{
// 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
// 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
}
// Multiply by the weight of matrix-matrix product and store the result
#if defined(ALPHA)
acc0 = acc0 * (half8)ALPHA;
#endif // defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
acc1 = acc1 * (half8)ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
acc2 = acc2 * (half8)ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2 && defined(ALPHA)
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
acc3 = acc3 * (half8)ALPHA;
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 && defined(ALPHA)
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);
#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 bottom paddings
//
// | |
// | plane0 |
// | |
// |_____________|
// |*************|
// | pad_bottom |
// |*************|
// | |
// | plane1 |
// | |
// |_____________|
// The plane (zout) is calculated dividing M (get_global_id(1) * NUM_ELEMS_PROCESSED_PER_THREAD_Y) by HEIGHT_GEMM3D
uint4 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 bottom paddings
zout *= (pad_bottom * 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;
// Store the output block
vstore8(acc0, 0, (__global half *)(dst_addr + 0 * dst_stride_y + zout.s0));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
vstore8(acc1, 0, (__global half *)(dst_addr + 1 * dst_stride_y + zout.s1));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
vstore8(acc2, 0, (__global half *)(dst_addr + 2 * dst_stride_y + zout.s2));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
vstore8(acc3, 0, (__global half *)(dst_addr + 3 * dst_stride_y + zout.s3));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#else // defined(REINTERPRET_OUTPUT_AS_3D)
// Add offset for batched GEMM
dst_addr += z * dst_stride_z;
// Store the output block
vstore8(acc0, 0, (__global half *)(dst_addr + 0 * dst_stride_y));
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
vstore8(acc1, 0, (__global half *)(dst_addr + 1 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
vstore8(acc2, 0, (__global half *)(dst_addr + 2 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2
#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
vstore8(acc3, 0, (__global half *)(dst_addr + 3 * dst_stride_y));
#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3
#endif // REINTERPRET_OUTPUT_AS_3D
}
#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, i.e. 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)