Add Gemm MMUL Reshaped Only Rhs Support for FP32/FP16
This patch introduces a GEMM routine that is optimized for Arm(R) Mali(TM)-G715 and Arm(R) Mali(TM)-G615
Resolves: COMPMID-5216
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Change-Id: I2e5d7806f5904347185bb3e250f73d73d6669dba
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7914
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/CL/CLHelpers.cpp b/src/core/CL/CLHelpers.cpp
index 5172a77..94675d6 100644
--- a/src/core/CL/CLHelpers.cpp
+++ b/src/core/CL/CLHelpers.cpp
@@ -491,4 +491,8 @@
}
}
+bool arm_matrix_multiply_supported(const cl::Device &device)
+{
+ return device_supports_extension(device, "cl_arm_matrix_multiply");
+}
} // namespace arm_compute
diff --git a/src/core/CL/cl_kernels/common/gemm_reshaped_only_rhs_mmul.cl b/src/core/CL/cl_kernels/common/gemm_reshaped_only_rhs_mmul.cl
new file mode 100644
index 0000000..8919023
--- /dev/null
+++ b/src/core/CL/cl_kernels/common/gemm_reshaped_only_rhs_mmul.cl
@@ -0,0 +1,528 @@
+/*
+ * Copyright (c) 2022 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "activation_float_helpers.h"
+#include "helpers.h"
+#include "tile_helpers.h"
+
+#if defined(GEMM_MM_RESHAPED_ONLY_RHS_NT_MMUL)
+/** This OpenCL kernel computes the matrix multiplication between 2 matrices using the MMUL extension:
+ *
+ * The LHS matrix is NOT reshaped
+ * The RHS is reshaped with @ref ClGemmMatrixMultiplyReshapedOnlyRhsKernel and the block K0xN0 is NOT transposed
+ *
+ * @note The block's dimensions used for reshaping the RHS matrix (N0 and K0) must be passed at compile time using -DN0 and -DK0 (e.g. -DN0=8, -DK0=4).
+ * @note The number of M0 rows to process must be passed at compile time using -DM0 (e.g. -DM0=2)
+ * @note The number of output columns processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_N0 (e.g., -DMMUL_N0=2)
+ * @note The number of output rows processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_M0 (e.g., -DMMUL_M0=2)
+ * @note The number of lhs columns (or rhs rows) processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_K0 (e.g., -DMMUL_K0=2)
+ * @note Only the following configurations of M0, N0 and K0 are currently supported:
+ * - M0 > 0
+ * - N0 = 1, 2, 3, 4, 8, 16
+ * - K0 = 1
+ *
+ * @note If the activation type were passed at compile time through -DACTIVATION_TYPE (e.g. -DACTIVATION_TYPE=RELU), A, B variables, required by some activation functions, should be passed at compile time as well using -DA_VAL= and -DB_VAL= respectively.
+ * The activation function is performed after the bias addition
+ *
+ * @param[in] lhs_ptr Pointer to the LHS tensor. Supported data types: F16/F32
+ * @param[in] lhs_stride_y Stride of the LHS tensor in Y dimension (in bytes)
+ * @param[in] lhs_stride_z Stride of the LHS tensor in Z dimension (in bytes)
+ * @param[in] lhs_w The size of the width dimension of the LHS tensor
+ * @param[in] lhs_h The size of the height dimension of the LHS tensor
+ * @param[in] lhs_n The size of the depth dimension of the LHS tensor
+ * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS tensor
+ * @param[in] rhs_ptr Pointer to the RHS reshaped tensor. Supported data type: same as @p lhs_ptr
+ * @param[in] rhs_stride_y Stride of the RHS tensor in Y dimension (in bytes)
+ * @param[in] rhs_stride_z Stride of the RHS tensor in Z dimension (in bytes)
+ * @param[in] rhs_w The size of the width dimension of the RHS tensor
+ * @param[in] rhs_h The size of the height dimension of the RHS tensor
+ * @param[in] rhs_n The size of the depth dimension of the RHS tensor
+ * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS tensor
+ * @param[in] bia_ptr (Optional) Pointer to the bias tensor. Supported data type: same as @p lhs_ptr
+ * @param[in] bia_stride_y (Optional) Stride of the bias tensor in Y dimension (in bytes)
+ * @param[in] bia_stride_z (Optional) Stride of the bias tensor in Z dimension (in bytes)
+ * @param[in] bia_w (Optional) The size of the width dimension of the bias tensor
+ * @param[in] bia_h (Optional) The size of the height dimension of the bias tensor
+ * @param[in] bia_n (Optional) The size of the depth dimension of the bias tensor
+ * @param[in] bia_offset_first_element_in_bytes (Optional) The offset of the first element in the bias tensor
+ * @param[out] dst_ptr Pointer to the destination tensor. Supported data type: same as @p lhs_ptr
+ * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
+ * @param[in] dst_w The size of the width dimension of the destination tensor
+ * @param[in] dst_h The size of the height dimension of the destination tensor
+ * @param[in] dst_n The size of the depth dimension of the destination tensor
+ * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ * @param[in] M Number of rows in LHS matrix not reshaped
+ * @param[in] N Number of columns in RHS matrix not reshaped
+ * @param[in] K Number of columns in LHS matrix and rows in RHS matrix not reshaped
+ */
+__kernel void gemm_mm_reshaped_only_rhs_nt_mmul(
+ TENSOR3D_T(lhs, BUFFER),
+ TENSOR3D_T(rhs, BUFFER),
+#if defined(BETA)
+ TENSOR3D_T(bia, BUFFER),
+#endif // defined(BETA)
+ TENSOR3D_T(dst, BUFFER),
+ const int M,
+ const int N,
+ const int K)
+{
+#define MMUL_BLOCK_SIZE (MMUL_N0 * MMUL_K0)
+
+ uint x0 = get_global_id(0); // (N / N0) * MMUL_K0
+ uint y0 = get_global_id(1); // (M / M0) / MMUL_M0
+ uint z = get_global_id(2); // Batch
+
+ // Get block ID and thread ID within the block
+ uint block_id = (x0 / MMUL_BLOCK_SIZE);
+ uint thread_id = (x0 % MMUL_BLOCK_SIZE);
+
+ // Coordinate within a block
+ uint block_x = thread_id % MMUL_N0;
+ uint block_y = (thread_id / MMUL_M0);
+
+ // Starting destination coordinates
+ uint dst_x = min(block_x * N0 + block_id * MMUL_N0 * N0, (uint)(N - 1));
+ uint dst_y = min(block_y * M0 + y0 * M0 * MMUL_M0, (uint)(M - M0));
+
+ // Note: We need to clamp dst_x and dst_y because we always need to execute a complete MMUL block! Only after the matrix multiplication
+ // part can we exit the kernel if it is out-of-bound. Remember, we have a cooperative matrix multiplication. Therefore, we need a full block to get the correct results
+
+ // Starting LHS coordinates
+ uint lhs_x = block_x;
+ uint lhs_y = dst_y;
+
+ // Starting RHS coordinates
+ uint rhs_x = block_y * N0 * MMUL_N0 + block_x * N0;
+ uint rhs_y = block_id;
+
+ // Compute LHS/RHS/DST matrix address
+ lhs_offset_first_element_in_bytes += lhs_x * sizeof(DATA_TYPE) + lhs_y * lhs_stride_y + z * lhs_stride_z;
+ rhs_offset_first_element_in_bytes += rhs_x * sizeof(DATA_TYPE) + rhs_y * rhs_stride_y + z * rhs_stride_z;
+ dst_offset_first_element_in_bytes += dst_x * sizeof(DATA_TYPE) + dst_y * dst_stride_y + z * dst_stride_z;
+
+ // Note: If RHS derives from the weights of convolution 2d layer, RHS will always be 2D and rhs_stride_z will always be equal to 0 for
+ // not sliding the tensor
+
+ // Initialize the accumulators
+ // MMUL extension accumulate the result in F32 for both F32 and F16
+ TILE(float, M0, N0, c_f32);
+
+#if !defined(HALF_PRECISION)
+#define c c_f32
+#endif // !defined(HALF_PRECISION)
+
+ LOOP_UNROLLING(int, i, 0, 1, M0,
+ {
+ c_f32[i].v = 0;
+ })
+
+ for(int k = 0; k <= K - MMUL_K0; k += MMUL_K0)
+ {
+ TILE(DATA_TYPE, M0, 1, a);
+ TILE(DATA_TYPE, 1, N0, b);
+
+ // Load tile from the lhs/rhs tensors
+ T_LOAD(DATA_TYPE, M0, 1, BUFFER, lhs, 0, 0, 1, lhs_stride_y, a);
+ T_LOAD(DATA_TYPE, 1, N0, BUFFER, rhs, 0, 0, 1, 0, b);
+
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ LOOP_UNROLLING(int, n0, 0, 1, N0,
+ {
+ c_f32[m0].s[n0] = arm_matrix_multiply(a[m0].s[0], b[0].s[n0], c_f32[m0].s[n0]);
+ })
+ })
+
+ lhs_offset_first_element_in_bytes += MMUL_K0 * sizeof(DATA_TYPE);
+ rhs_offset_first_element_in_bytes += MMUL_K0 * MMUL_N0 * N0 * sizeof(DATA_TYPE);
+ }
+
+ if(block_x * N0 + block_id * MMUL_N0 * N0 >= N)
+ {
+ return;
+ }
+
+ if(block_y * M0 + y0 * M0 * MMUL_M0 >= M)
+ {
+ return;
+ }
+
+#if defined(HALF_PRECISION)
+ TILE(DATA_TYPE, M0, N0, c);
+
+ // Conversion required for the half precision
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ LOOP_UNROLLING(int, n0, 0, 1, N0,
+ {
+ c[m0].s[n0] = c_f32[m0].s[n0];
+ })
+ })
+#endif // defined(HALF_PRECISION)
+
+ // Multiply by the weight of matrix-matrix product and store the result
+#if defined(ALPHA)
+ T_SCALE_CONSTANT(DATA_TYPE, M0, N0, c, (DATA_TYPE)ALPHA, c);
+#endif // defined(ALPHA)
+
+ // Add beta*bias
+#if defined(BETA)
+#if defined(BROADCAST_BIAS)
+ bia_offset_first_element_in_bytes += dst_x * sizeof(DATA_TYPE);
+
+ TILE(DATA_TYPE, 1, N0, bias0);
+
+ if(dst_x + N0 <= N || N0_LEFTOVER == 0)
+ {
+ bias0[0].v = VLOAD(N0)(0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes));
+ }
+ else
+ {
+ VLOAD_PARTIAL(N0, N0_LEFTOVER)
+ (bias0[0].v, 0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes));
+ }
+
+#ifndef UNIT_BETA
+ T_SCALE_CONSTANT(DATA_TYPE, 1, N0, bias0, (DATA_TYPE)BETA, bias0);
+#endif // UNIT_BIAS
+
+ // c = c + bias[broadcasted]
+ T_ELTWISE_BROADCAST_X(V_ADD, DATA_TYPE, M0, N0, c, bias0, c);
+#else // defined(BROADCAST_BIAS)
+ TILE(DATA_TYPE, M0, N0, bias0);
+
+ bia_offset_first_element_in_bytes += dst_x * sizeof(DATA_TYPE) + dst_y * bia_stride_y + z * bia_stride_z;
+
+ if(dst_x + N0 <= N || N0_LEFTOVER == 0)
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ bias0[m0].v = VLOAD(N0)(0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes + m0 * bia_stride_y));
+ }
+ })
+ }
+ else
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ VLOAD_PARTIAL(N0, N0_LEFTOVER)
+ (bias0[m0].v, 0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes + m0 * bia_stride_y));
+ }
+ })
+ }
+
+#ifndef UNIT_BETA
+ T_SCALE_CONSTANT(DATA_TYPE, M0, N0, bias0, (DATA_TYPE)BETA, bias0);
+#endif // UNIT_BIAS
+
+ // c = c + bias
+ T_ADD(DATA_TYPE, M0, N0, c, bias0, c);
+ // c = c + bias
+#endif // defined(BROADCAST_BIAS)
+#endif // defined(BETA)
+
+ T_ACTIVATION(DATA_TYPE, M0, N0, ACTIVATION_TYPE, A_VAL, B_VAL, c, c);
+
+ // Store
+ if(dst_x + N0 <= N || N0_LEFTOVER == 0)
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ VSTORE(N0)
+ (c[m0].v, 0, (__global DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y));
+ }
+ })
+ }
+ else
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ VSTORE_PARTIAL(N0, N0_LEFTOVER)
+ (c[m0].v, 0, (__global DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y));
+ }
+ })
+ }
+
+#undef RHS_BLOCK_SIZE
+#undef RHS_OFFSET_X
+#undef RHS_STEP_X
+}
+#endif // defined(GEMM_MM_RESHAPED_ONLY_RHS_MMUL)
+
+#if defined(GEMM_MM_RESHAPED_ONLY_RHS_NT_MMUL_TEXTURE)
+/** This OpenCL kernel computes the matrix multiplication between 2 matrices using the MMUL extension and the OpenCL image for RHS:
+ *
+ * The LHS matrix is NOT reshaped
+ * The RHS is reshaped with @ref ClGemmMatrixMultiplyReshapedOnlyRhsKernel and the block K0xN0 is NOT transposed
+ *
+ * @note The block's dimensions used for reshaping the RHS matrix (N0 and K0) must be passed at compile time using -DN0 and -DK0 (e.g. -DN0=8, -DK0=4).
+ * @note The number of M0 rows to process must be passed at compile time using -DM0 (e.g. -DM0=2)
+ * @note The number of output columns processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_N0 (e.g., -DMMUL_N0=2)
+ * @note The number of output rows processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_M0 (e.g., -DMMUL_M0=2)
+ * @note The number of lhs columns (or rhs rows) processed by the the cooperative mmul extension must be passed at compile time using -DMMUL_K0 (e.g., -DMMUL_K0=2)
+ * @note Only the following configurations of M0, N0 and K0 are currently supported:
+ * - M0 > 0
+ * - N0 = 1, 2, 3, 4, 8, 16
+ * - K0 = 1
+ *
+ * @note If the activation type were passed at compile time through -DACTIVATION_TYPE (e.g. -DACTIVATION_TYPE=RELU), A, B variables, required by some activation functions, should be passed at compile time as well using -DA_VAL= and -DB_VAL= respectively.
+ * The activation function is performed after the bias addition
+ *
+ * @param[in] lhs_ptr Pointer to the LHS tensor. Supported data types: F16/F32
+ * @param[in] lhs_stride_y Stride of the LHS tensor in Y dimension (in bytes)
+ * @param[in] lhs_stride_z Stride of the LHS tensor in Z dimension (in bytes)
+ * @param[in] lhs_w The size of the width dimension of the LHS tensor
+ * @param[in] lhs_h The size of the height dimension of the LHS tensor
+ * @param[in] lhs_n The size of the depth dimension of the LHS tensor
+ * @param[in] lhs_offset_first_element_in_bytes The offset of the first element in the LHS tensor
+ * @param[in] rhs_ptr Pointer to the RHS reshaped tensor. Supported data type: same as @p lhs_ptr
+ * @param[in] rhs_stride_y Stride of the RHS tensor in Y dimension (in bytes)
+ * @param[in] rhs_stride_z Stride of the RHS tensor in Z dimension (in bytes)
+ * @param[in] rhs_w The size of the width dimension of the RHS tensor
+ * @param[in] rhs_h The size of the height dimension of the RHS tensor
+ * @param[in] rhs_n The size of the depth dimension of the RHS tensor
+ * @param[in] rhs_offset_first_element_in_bytes The offset of the first element in the RHS tensor
+ * @param[in] bia_ptr (Optional) Pointer to the bias tensor. Supported data type: same as @p lhs_ptr
+ * @param[in] bia_stride_y (Optional) Stride of the bias tensor in Y dimension (in bytes)
+ * @param[in] bia_stride_z (Optional) Stride of the bias tensor in Z dimension (in bytes)
+ * @param[in] bia_w (Optional) The size of the width dimension of the bias tensor
+ * @param[in] bia_h (Optional) The size of the height dimension of the bias tensor
+ * @param[in] bia_n (Optional) The size of the depth dimension of the bias tensor
+ * @param[in] bia_offset_first_element_in_bytes (Optional) The offset of the first element in the bias tensor
+ * @param[out] dst_ptr Pointer to the destination tensor. Supported data type: same as @p lhs_ptr
+ * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)
+ * @param[in] dst_w The size of the width dimension of the destination tensor
+ * @param[in] dst_h The size of the height dimension of the destination tensor
+ * @param[in] dst_n The size of the depth dimension of the destination tensor
+ * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ * @param[in] M Number of rows in LHS matrix not reshaped
+ * @param[in] N Number of columns in RHS matrix not reshaped
+ * @param[in] K Number of columns in LHS matrix and rows in RHS matrix not reshaped
+ */
+__kernel void gemm_mm_reshaped_only_rhs_nt_mmul_texture(
+ TENSOR3D_T(lhs, BUFFER),
+ TENSOR3D_T(rhs, IMAGE),
+#if defined(BETA)
+ TENSOR3D_T(bia, BUFFER),
+#endif // defined(BETA)
+ TENSOR3D_T(dst, BUFFER),
+ const int M,
+ const int N,
+ const int K)
+{
+#define MMUL_BLOCK_SIZE (MMUL_N0 * MMUL_K0)
+
+ uint x0 = get_global_id(0); // (N / N0) * MMUL_K0
+ uint y0 = get_global_id(1); // (M / M0) / MMUL_M0
+ uint z = get_global_id(2); // Batch
+
+ // Get block ID and thread ID within the block
+ uint block_id = (x0 / MMUL_BLOCK_SIZE);
+ uint thread_id = (x0 % MMUL_BLOCK_SIZE);
+
+ // Coordinate within a block
+ uint block_x = thread_id % MMUL_N0;
+ uint block_y = (thread_id / MMUL_M0);
+
+ // Starting destination coordinates
+ uint dst_x = min(block_x * N0 + block_id * MMUL_N0 * N0, (uint)(N - 1));
+ uint dst_y = min(block_y * M0 + y0 * M0 * MMUL_M0, (uint)(M - M0));
+
+ // Note: We need to clamp dst_x and dst_y because we always need to execute a complete MMUL block! Only after the matrix multiplication
+ // part can we exit the kernel if it is out-of-bound. Remember, we have a cooperative matrix multiplication. Therefore, we need a full block to get the correct results
+
+ // Starting LHS coordinates
+ uint lhs_x = block_x;
+ uint lhs_y = dst_y;
+
+ // Starting RHS coordinates
+ uint rhs_x = block_y * N0 * MMUL_N0 + block_x * N0;
+ uint rhs_y = block_id + z * rhs_h;
+
+ // Compute LHS/RHS/DST matrix address
+ lhs_offset_first_element_in_bytes += lhs_x * sizeof(DATA_TYPE) + lhs_y * lhs_stride_y + z * lhs_stride_z;
+ dst_offset_first_element_in_bytes += dst_x * sizeof(DATA_TYPE) + dst_y * dst_stride_y + z * dst_stride_z;
+
+ // Initialize the accumulators
+ // MMUL extension accumulate the result in F32 for both F32 and F16
+ TILE(float, M0, N0, c_f32);
+
+#if !defined(HALF_PRECISION)
+#define c c_f32
+#endif // !defined(HALF_PRECISION)
+
+ LOOP_UNROLLING(int, i, 0, 1, M0,
+ {
+ c_f32[i].v = 0;
+ })
+
+ for(int k = 0; k <= K - MMUL_K0; k += MMUL_K0)
+ {
+ TILE(DATA_TYPE, M0, 1, a);
+ TILE(DATA_TYPE, 1, N0, b);
+
+ // Load tile from the lhs/rhs tensors
+ T_LOAD(DATA_TYPE, M0, 1, BUFFER, lhs, 0, 0, 1, lhs_stride_y, a);
+ T_LOAD(DATA_TYPE, 1, N0, IMAGE, rhs, rhs_x, rhs_y, 1, rhs_stride_y, b);
+
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ LOOP_UNROLLING(int, n0, 0, 1, N0,
+ {
+ c_f32[m0].s[n0] = arm_matrix_multiply(a[m0].s[0], b[0].s[n0], c_f32[m0].s[n0]);
+ })
+ })
+
+ lhs_offset_first_element_in_bytes += MMUL_K0 * sizeof(DATA_TYPE);
+ rhs_x += MMUL_K0 * MMUL_N0 * N0;
+ }
+
+ if(block_x * N0 + block_id * MMUL_N0 * N0 >= N)
+ {
+ return;
+ }
+
+ if(block_y * M0 + y0 * M0 * MMUL_M0 >= M)
+ {
+ return;
+ }
+
+#if defined(HALF_PRECISION)
+ TILE(DATA_TYPE, M0, N0, c);
+
+ // Conversion required for the half precision
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ LOOP_UNROLLING(int, n0, 0, 1, N0,
+ {
+ c[m0].s[n0] = c_f32[m0].s[n0];
+ })
+ })
+#endif // defined(HALF_PRECISION)
+
+ // Multiply by the weight of matrix-matrix product and store the result
+#if defined(ALPHA)
+ T_SCALE_CONSTANT(DATA_TYPE, M0, N0, c, (DATA_TYPE)ALPHA, c);
+#endif // defined(ALPHA)
+
+ // Add beta*bias
+#if defined(BETA)
+#if defined(BROADCAST_BIAS)
+ bia_offset_first_element_in_bytes += dst_x * sizeof(DATA_TYPE);
+
+ TILE(DATA_TYPE, 1, N0, bias0);
+
+ if(dst_x + N0 <= N || N0_LEFTOVER == 0)
+ {
+ bias0[0].v = VLOAD(N0)(0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes));
+ }
+ else
+ {
+ VLOAD_PARTIAL(N0, N0_LEFTOVER)
+ (bias0[0].v, 0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes));
+ }
+
+#ifndef UNIT_BETA
+ T_SCALE_CONSTANT(DATA_TYPE, 1, N0, bias0, (DATA_TYPE)BETA, bias0);
+#endif // UNIT_BIAS
+
+ // c = c + bias[broadcasted]
+ T_ELTWISE_BROADCAST_X(V_ADD, DATA_TYPE, M0, N0, c, bias0, c);
+#else // defined(BROADCAST_BIAS)
+ TILE(DATA_TYPE, M0, N0, bias0);
+
+ bia_offset_first_element_in_bytes += dst_x * sizeof(DATA_TYPE) + dst_y * bia_stride_y + z * bia_stride_z;
+
+ if(dst_x + N0 <= N || N0_LEFTOVER == 0)
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ bias0[m0].v = VLOAD(N0)(0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes + m0 * bia_stride_y));
+ }
+ })
+ }
+ else
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ VLOAD_PARTIAL(N0, N0_LEFTOVER)
+ (bias0[m0].v, 0, (DATA_TYPE *)(bia_ptr + bia_offset_first_element_in_bytes + m0 * bia_stride_y));
+ }
+ })
+ }
+
+#ifndef UNIT_BETA
+ T_SCALE_CONSTANT(DATA_TYPE, M0, N0, bias0, (DATA_TYPE)BETA, bias0);
+#endif // UNIT_BIAS
+
+ // c = c + bias
+ T_ADD(DATA_TYPE, M0, N0, c, bias0, c);
+ // c = c + bias
+#endif // defined(BROADCAST_BIAS)
+#endif // defined(BETA)
+
+ T_ACTIVATION(DATA_TYPE, M0, N0, ACTIVATION_TYPE, A_VAL, B_VAL, c, c);
+
+ // Store
+ if(dst_x + N0 <= N || N0_LEFTOVER == 0)
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ VSTORE(N0)
+ (c[m0].v, 0, (__global DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y));
+ }
+ })
+ }
+ else
+ {
+ LOOP_UNROLLING(int, m0, 0, 1, M0,
+ {
+ if(dst_y + m0 < M || M0_LEFTOVER == 0)
+ {
+ VSTORE_PARTIAL(N0, N0_LEFTOVER)
+ (c[m0].v, 0, (__global DATA_TYPE *)(dst_ptr + dst_offset_first_element_in_bytes + m0 * dst_stride_y));
+ }
+ })
+ }
+
+#undef RHS_BLOCK_SIZE
+#undef RHS_OFFSET_X
+#undef RHS_STEP_X
+}
+#endif // defined(GEMM_MM_RESHAPED_ONLY_RHS_MMUL_TEXTURE)
\ No newline at end of file
diff --git a/src/core/CL/cl_kernels/tile_helpers.h b/src/core/CL/cl_kernels/tile_helpers.h
index 0ce343e..4b6144a 100644
--- a/src/core/CL/cl_kernels/tile_helpers.h
+++ b/src/core/CL/cl_kernels/tile_helpers.h
@@ -970,8 +970,8 @@
#define ACT_OP_QUANTIZED(op, DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) op##_op_quantized(DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x)
#define ACTIVATION_QUANTIZED(op, DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) ACT_OP_QUANTIZED(op, DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x)
-#define T_ADD(A_VAL, B_VAL) ((A_VAL) + (B_VAL))
-#define T_DIV(A_VAL, B_VAL) ((A_VAL) / (B_VAL))
+#define V_ADD(A_VAL, B_VAL) ((A_VAL) + (B_VAL))
+#define V_DIV(A_VAL, B_VAL) ((A_VAL) / (B_VAL))
/** Element-wise activation for quantized types
*
@@ -995,6 +995,25 @@
}) \
})
+/** Element-wise addition between two tiles
+ *
+ * @note Performs: LHS + RHS = DST
+ *
+ * @param[in] DATA_TYPE LHS/RHS/DST data type
+ * @param[in] M0 Number of LHS rows
+ * @param[in] N0 Number of LHS columns
+ * @param[in] lhs LHS tile
+ * @param[in] rhs Constant RHS tile
+ * @param[out] dst DST tile
+ */
+#define T_ADD(DATA_TYPE, M0, N0, lhs, rhs, dst) \
+ ({ \
+ LOOP_UNROLLING(int, _m0, 0, 1, M0, \
+ { \
+ dst[_m0].v = lhs[_m0].v + rhs[_m0].v; \
+ }) \
+ })
+
/** Element-wise addition with a constant value
*
* @note Performs: LHS + constant = DST
@@ -1010,15 +1029,31 @@
({ \
LOOP_UNROLLING(int, _m0, 0, 1, M0, \
{ \
- LOOP_UNROLLING(int, _n0, 0, 1, N0, \
- { \
- dst[_m0].s[_n0] = lhs[_m0].s[_n0] + rhs_constant; \
- }) \
+ dst[_m0].v = lhs[_m0].v + (DATA_TYPE)rhs_constant; \
}) \
})
-#define T_ELTWISE_BROADCAST_ADD_X(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(T_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
-#define T_ELTWISE_BROADCAST_DIV_X(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(T_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+#define T_ELTWISE_BROADCAST_ADD_X(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+#define T_ELTWISE_BROADCAST_DIV_X(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+
+/** Element-wise scale with a constant value
+ *
+ * @note Performs: LHS * constant = DST
+ *
+ * @param[in] DATA_TYPE LHS/RHS/DST data type
+ * @param[in] M0 Number of LHS rows
+ * @param[in] N0 Number of LHS columns
+ * @param[in] lhs LHS tile
+ * @param[in] rhs_constant Constant value
+ * @param[out] dst DST tile
+ */
+#define T_SCALE_CONSTANT(DATA_TYPE, M0, N0, lhs, rhs_constant, dst) \
+ ({ \
+ LOOP_UNROLLING(int, _m0, 0, 1, M0, \
+ { \
+ dst[_m0].v = lhs[_m0].v * (DATA_TYPE)rhs_constant; \
+ }) \
+ })
/** Element-wise operation with RHS broadcasted (RHS has the X dimension only)
*
@@ -1041,8 +1076,8 @@
}) \
})
-#define T_ELTWISE_ADD(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(T_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
-#define T_ELTWISE_DIV(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(T_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+#define T_ELTWISE_ADD(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+#define T_ELTWISE_DIV(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
/** Element-wise operation between two tiles (LHS and RHS)
*
diff --git a/src/core/GPUTarget.cpp b/src/core/GPUTarget.cpp
index 5984c88..e74abf6 100644
--- a/src/core/GPUTarget.cpp
+++ b/src/core/GPUTarget.cpp
@@ -47,6 +47,14 @@
{
return arm_compute::GPUTarget::G57;
}
+ else if(version.find("G715") != std::string::npos)
+ {
+ return arm_compute::GPUTarget::G715;
+ }
+ else if(version.find("G615") != std::string::npos)
+ {
+ return arm_compute::GPUTarget::G615;
+ }
else
{
return arm_compute::GPUTarget::UNKNOWN;
@@ -141,7 +149,9 @@
{ GPUTarget::G77, "g77" },
{ GPUTarget::G78, "g78" },
{ GPUTarget::G710, "g710" },
- { GPUTarget::G57, "g57" }
+ { GPUTarget::G57, "g57" },
+ { GPUTarget::G715, "g715" },
+ { GPUTarget::G615, "g615" }
};
return gpu_target_map[target];