diff options
author | Ramy Elgammal <ramelg01@e129512.arm.com> | 2023-01-11 18:48:04 +0000 |
---|---|---|
committer | Ramy Elgammal <ramy.elgammal@arm.com> | 2023-01-25 10:24:23 +0000 |
commit | 002e6530f6218b00a28aef9be8b21efb08cf3602 (patch) | |
tree | f3e2f9d064b985ffe283512825b34cdd59f29f50 /src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp | |
parent | cc2877368d5e15d9ea89d31c84ec651fc0fffd13 (diff) | |
download | ComputeLibrary-002e6530f6218b00a28aef9be8b21efb08cf3602.tar.gz |
Implement dynamic fusion softmax operator
- Return aux tensorInfo by get_aux_tensors() at runtime to init the aux
tensor with the right size.
- Keep softmax unfusable for this commit
- Hence, added Tensor3D to template writer arguments declaration, for sake of
keeping dynamic fusion softmax componenets' kernels matching their cl
counterparts.
Resolves: COMPMID-5523
Change-Id: I667f39545db925f667036ef448302c79a0330373
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/483924
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: bsgcomp <bsgcomp@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8986
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp')
-rw-r--r-- | src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp | 95 |
1 files changed, 95 insertions, 0 deletions
diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp new file mode 100644 index 0000000000..7864d56d29 --- /dev/null +++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.cpp @@ -0,0 +1,95 @@ +/* + * Copyright (c) 2023 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 "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentLogits1DNorm.h" + +#include "arm_compute/core/CL/CLHelpers.h" +#include "arm_compute/core/Validate.h" +#include "arm_compute/core/utils/misc/ShapeCalculator.h" +#include "arm_compute/dynamic_fusion/sketch/attributes/SoftmaxAttributes.h" +#include "src/core/CL/CLValidate.h" +#include "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateLogits1DNorm.h" + +namespace arm_compute +{ +namespace experimental +{ +namespace dynamic_fusion +{ +Status ClComponentLogits1DNorm::validate( + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes) +{ + ARM_COMPUTE_UNUSED(properties, attributes); + + const ITensorInfo *src = tensors.get_const_tensor(TensorType::ACL_SRC_0); + const ITensorInfo *sum = tensors.get_const_tensor(TensorType::ACL_SRC_1); + const ITensorInfo *dst = tensors.get_const_tensor(TensorType::ACL_DST_0); + + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src); + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(sum); + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(dst); + + // 1. Check validity + // All tensor infos are initialized + ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().total_size() == 0); + ARM_COMPUTE_RETURN_ERROR_ON(sum->tensor_shape().total_size() == 0); + ARM_COMPUTE_RETURN_ERROR_ON(dst->tensor_shape().total_size() == 0); + + // Check for mismatches in shapes and data types + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst, sum); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src, dst); + + ARM_COMPUTE_RETURN_ERROR_ON(attributes.is_log_softmax() && !is_data_type_float(src->data_type())); + + // Device requirements are met + ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src); + + // 2. Check support level + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::F16, DataType::F32); + + return Status{}; +} + +ClComponentLogits1DNorm::ClComponentLogits1DNorm(ComponentId id, + const Properties &properties, + const ArgumentPack<ITensorInfo> &tensors, + const Attributes &attributes) + : IGpuKernelComponent{ id, properties, tensors }, + _component_writer{ std::make_unique<ClTemplateLogits1DNorm>(id, tensors, attributes) } +{ +} + +ClComponentLogits1DNorm::~ClComponentLogits1DNorm() +{ +} + +const IGpuTemplateComponentWriter *ClComponentLogits1DNorm::template_writer() const +{ + return _component_writer.get(); +} +} // namespace dynamic_fusion +} // namespace experimental +} // namespace arm_compute |