aboutsummaryrefslogtreecommitdiff
path: root/src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
diff options
context:
space:
mode:
authorJakub Sujak <jakub.sujak@arm.com>2023-08-24 14:01:20 +0100
committerJakub Sujak <jakub.sujak@arm.com>2023-09-04 14:41:16 +0000
commit0d27b2ee8d811d66693555ac1e7be44d93e662e2 (patch)
tree8b62a464a8bb9cd46702c8b5a60f3a97e3821b41 /src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
parent7ff03b67ba7ce669223f4d807e18fa3efa2f729b (diff)
downloadComputeLibrary-0d27b2ee8d811d66693555ac1e7be44d93e662e2.tar.gz
Remove legacy PostOps code
PostOps was the experimental interface for Dynamic Fusion. It is now replaced by the new Dynamic Fusion interface with code generation using the Compute Kernel Writer. Resolves: COMPMID-6190 Change-Id: I813b48facef2fd6f3aee332588886b4f9b3d33d8 Signed-off-by: Jakub Sujak <jakub.sujak@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10219 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: SiCong Li <sicong.li@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp')
-rw-r--r--src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp55
1 files changed, 5 insertions, 50 deletions
diff --git a/src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp b/src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
index f14a6f1900..d72d29ea1e 100644
--- a/src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
+++ b/src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.cpp
@@ -23,7 +23,6 @@
*/
#include "src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.h"
-#include "arm_compute/core/utils/ActivationFunctionUtils.h"
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/ICLTensor.h"
@@ -31,11 +30,11 @@
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/core/utils/ActivationFunctionUtils.h"
#include "arm_compute/core/utils/StringUtils.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "src/core/CL/CLUtils.h"
#include "src/core/CL/CLValidate.h"
-#include "src/core/experimental/PostOpUtils.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "src/core/utils/helpers/float_ops.h"
@@ -53,25 +52,6 @@ namespace
{
using ElementsProcessed = Steps;
-const auto post_op_utils = experimental::PostOpCLKernelUtils(
-{
- // PostOp sequence -> {Kernel Postfix, PostOp Slots}
- { {}, { "", {} } },
- { { experimental::PostOpType::Activation }, { "", { 1 } } },
-
- { { experimental::PostOpType::Eltwise_Add }, { "_post_act_eltwise_op_act", { 2 } } },
- { { experimental::PostOpType::Eltwise_PRelu }, { "_post_act_eltwise_op_act", { 2 } } },
-
- { { experimental::PostOpType::Activation, experimental::PostOpType::Eltwise_Add }, { "_post_act_eltwise_op_act", { 1, 2 } } },
- { { experimental::PostOpType::Activation, experimental::PostOpType::Eltwise_PRelu }, { "_post_act_eltwise_op_act", { 1, 2 } } },
-
- { { experimental::PostOpType::Eltwise_Add, experimental::PostOpType::Activation }, { "_post_act_eltwise_op_act", { 2, 3 } } },
- { { experimental::PostOpType::Eltwise_PRelu, experimental::PostOpType::Activation }, { "_post_act_eltwise_op_act", { 2, 3 } } },
-
- { { experimental::PostOpType::Activation, experimental::PostOpType::Eltwise_Add, experimental::PostOpType::Activation }, { "_post_act_eltwise_op_act", { 1, 2, 3 } } },
- { { experimental::PostOpType::Activation, experimental::PostOpType::Eltwise_PRelu, experimental::PostOpType::Activation }, { "_post_act_eltwise_op_act", { 1, 2, 3 } } }
-});
-
Status validate_arguments(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, float alpha, float beta, const GEMMLHSMatrixInfo &lhs_info,
const GEMMRHSMatrixInfo &rhs_info,
const GEMMKernelInfo &gemm_info)
@@ -95,7 +75,6 @@ Status validate_arguments(const ITensorInfo *src0, const ITensorInfo *src1, cons
"Bias addition only supported with broadcast mode in case the input or dst has to be reinterpreted as 3D");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.fp_mixed_precision && (src0->data_type() == DataType::F32), "Mixed precision only supported for F16 data type");
ARM_COMPUTE_RETURN_ON_ERROR(gemm::validate_image2d_support_on_rhs(*src1, rhs_info));
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(!post_op_utils.is_post_op_sequence_supported(gemm_info.post_ops), "The sequence of Post Ops is not supported");
const unsigned int m = gemm_info.m;
const unsigned int n = gemm_info.n;
@@ -139,7 +118,6 @@ Status validate_arguments(const ITensorInfo *src0, const ITensorInfo *src1, cons
const TensorInfo tensor_info_dst = dst->clone()->set_tensor_shape(misc::shape_calculator::compute_mm_shape(*src0, *src1, gemm_info));
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(dst, &tensor_info_dst);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src0, dst);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(!post_op_utils.are_post_op_shapes_compliant(dst, gemm_info.post_ops), "The Post Op shapes are not compliant");
}
return Status{};
@@ -202,7 +180,6 @@ void ClGemmMatrixMultiplyReshapedKernel::configure(const CLCompileContext &compi
_use_dummy_work_items = preferred_dummy_work_items_support(CLKernelLibrary::get().get_device());
_add_bias = src2 != nullptr;
_export_to_cl_image = rhs_info.export_to_cl_image;
- _num_post_op_args = gemm_info.post_ops.total_num_arguments();
// Check if we need to slide the matrix B
const unsigned int num_dimensions_src0 = src0->num_dimensions();
@@ -260,23 +237,14 @@ void ClGemmMatrixMultiplyReshapedKernel::configure(const CLCompileContext &compi
build_opts.add_option("-DH0=" + support::cpp11::to_string(rhs_info.h0));
build_opts.add_option("-DPARTIAL_STORE_M0=" + support::cpp11::to_string(partial_store_m0));
build_opts.add_option("-DPARTIAL_STORE_N0=" + support::cpp11::to_string(partial_store_n0));
- // If post_ops are used, then we disable the use of gemm_info.activation_info
- if(gemm_info.post_ops.size() > 0)
- {
- post_op_utils.set_post_ops_cl_build_options(build_opts, gemm_info.post_ops);
- }
- else
- {
- build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(gemm_info.activation_info.activation())));
- build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.a()));
- build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.b()));
- }
+ build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(gemm_info.activation_info.activation())));
+ build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.a()));
+ build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.b()));
std::string kernel_name("gemm_mm_reshaped_");
kernel_name += lhs_info.transpose ? "lhs_t_" : "lhs_nt_";
kernel_name += rhs_info.transpose ? "rhs_t" : "rhs_nt";
kernel_name += rhs_info.export_to_cl_image ? "_texture" : "";
- post_op_utils.set_post_ops_cl_kernel_name(kernel_name, gemm_info.post_ops);
// A macro guard to compile ONLY the kernel of interest
build_opts.add_option("-D" + upper_string(kernel_name));
@@ -395,13 +363,6 @@ void ClGemmMatrixMultiplyReshapedKernel::run_op(ITensorPack &tensors, const Wind
// dst buffer
add_2D_tensor_argument(idx, dst, slice);
- // post op argument buffers
- for(size_t i = 0; i < _num_post_op_args; ++i)
- {
- const auto post_op_arg = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(experimental::get_post_op_arg_type(i)));
- add_2D_tensor_argument(idx, post_op_arg, slice);
- }
-
// LHS stride_z
_kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src0->info()->strides_in_bytes()[2]));
@@ -417,12 +378,6 @@ void ClGemmMatrixMultiplyReshapedKernel::run_op(ITensorPack &tensors, const Wind
// dst stride_z
_kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(dst->info()->strides_in_bytes()[2]));
- // post op argument stride_z
- for(size_t i = 0; i < _num_post_op_args; ++i)
- {
- const auto post_op_arg = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(experimental::get_post_op_arg_type(i)));
- _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(post_op_arg->info()->strides_in_bytes()[2]));
- }
// Cross-plan padding (if _reinterpret_output_as_3d = true)
if(_reinterpret_output_as_3d)
{