From 7068f9900d136312318ff430aef588b14e0c87ad Mon Sep 17 00:00:00 2001 From: Anthony Barbier Date: Thu, 26 Oct 2017 15:23:08 +0100 Subject: COMPMID-631: Merge branches/gles_compute branch Last commit: commit b25c5f68042b0c81bf611d59a1bb8535e1c42497 Author: Xinghang Zhou Date: Wed Oct 25 18:48:10 2017 +0800 Synced validation's tolerances of GCSoftmax from cl side Change-Id: Ibe72054205c1c8721845d679a31af7ed0a7c5cf6 Reviewed-on: http://mpd-gerrit.cambridge.arm.com/93283 Reviewed-by: Anthony Barbier Tested-by: Kaizen --- src/core/GLES_COMPUTE/IGCKernel.cpp | 157 ++++++++++++++++++++++++++++++++++++ 1 file changed, 157 insertions(+) create mode 100644 src/core/GLES_COMPUTE/IGCKernel.cpp (limited to 'src/core/GLES_COMPUTE/IGCKernel.cpp') diff --git a/src/core/GLES_COMPUTE/IGCKernel.cpp b/src/core/GLES_COMPUTE/IGCKernel.cpp new file mode 100644 index 0000000000..154a2c0c66 --- /dev/null +++ b/src/core/GLES_COMPUTE/IGCKernel.cpp @@ -0,0 +1,157 @@ +/* + * Copyright (c) 2017 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 "arm_compute/core/GLES_COMPUTE/IGCKernel.h" + +#include "arm_compute/core/Error.h" +#include "arm_compute/core/GLES_COMPUTE/GCHelpers.h" +#include "arm_compute/core/GLES_COMPUTE/IGCTensor.h" +#include "arm_compute/core/Helpers.h" +#include "arm_compute/core/TensorInfo.h" +#include "arm_compute/core/Utils.h" +#include "arm_compute/core/Validate.h" +#include "arm_compute/core/Window.h" +#include "arm_compute/runtime/GLES_COMPUTE/GCScheduler.h" + +#include +#include + +using namespace arm_compute; + +void arm_compute::enqueue(IGCKernel &kernel, const Window &window, const gles::NDRange &lws) +{ + ARM_COMPUTE_UNUSED(kernel); + + if(kernel.kernel().get_program() == 0) + { + return; + } + + ARM_COMPUTE_ERROR_ON((0 == (window.x().end() - window.x().start())) || (0 == (window.y().end() - window.y().start()))); + + ARM_COMPUTE_ERROR_ON_MSG((((window.x().end() - window.x().start()) % (window.x().step() * lws[0])) != 0), + "window x end =%d, start=%d, step=%d, lws x=%d", window.x().end(), window.x().start(), window.x().step(), lws[0]); + ARM_COMPUTE_ERROR_ON_MSG((((window.y().end() - window.y().start()) % (window.y().step() * lws[1])) != 0), + "window y end =%d, start=%d, step=%d, lws y=%d", window.y().end(), window.y().start(), window.y().step(), lws[1]); + ARM_COMPUTE_ERROR_ON_MSG((((window.z().end() - window.z().start()) % (window.z().step() * lws[2])) != 0), + "window z end =%d, start=%d, step=%d, lws z=%d", window.z().end(), window.z().start(), window.z().step(), lws[2]); + + ARM_COMPUTE_GL_CHECK(glDispatchCompute((window.x().end() - window.x().start()) / (window.x().step() / lws[0]), + (window.y().end() - window.y().start()) / (window.y().step() / lws[1]), + (window.z().end() - window.z().start()) / (window.z().step() / lws[2]))); +} + +IGCKernel::IGCKernel() + : _kernel() +{ +} + +GCKernel &IGCKernel::kernel() +{ + return _kernel; +} + +template +unsigned int IGCKernel::num_arguments_per_tensor() const +{ + return 2 + 2 * dimension_size; +} + +template +void IGCKernel::add_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const BufferParam ¶m, const Window &window) +{ + ARM_COMPUTE_ERROR_ON(tensor == nullptr); + + const ITensorInfo *info = tensor->info(); + const Strides &strides = info->strides_in_bytes(); + + // Calculate offset to the start of the window + unsigned int offset_first_element = info->offset_first_element_in_bytes(); + + for(unsigned int n = 0; n < info->num_dimensions(); ++n) + { + offset_first_element += window[n].start() * strides[n]; + } + + unsigned int idx_start = idx; + + for(unsigned int dimension = 0; dimension < dimension_size; dimension++) + { + _kernel.set_params(idx++, strides[dimension]); + _kernel.set_params(idx++, strides[dimension] * window[dimension].step()); + } + + _kernel.set_params(idx++, offset_first_element); + _kernel.set_params(idx++, param.buffer_data_type_shift); + + ARM_COMPUTE_GL_CHECK(glBindBufferBase(GL_SHADER_STORAGE_BUFFER, param.binding_point, tensor->gc_buffer())); + + ARM_COMPUTE_ERROR_ON_MSG(idx_start + num_arguments_per_tensor() != idx, + "add_%dD_tensor_argument() is supposed to add exactly %d arguments to the kernel", dimension_size, num_arguments_per_tensor()); + ARM_COMPUTE_UNUSED(idx_start); +} + +void IGCKernel::add_1D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window) +{ + add_tensor_argument<1>(idx, tensor, BufferParam(binding_point, 0), window); +} + +void IGCKernel::add_1D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const BufferParam ¶m, const Window &window) +{ + add_tensor_argument<1>(idx, tensor, param, window); +} + +void IGCKernel::add_2D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window) +{ + add_tensor_argument<2>(idx, tensor, BufferParam(binding_point, 0), window); +} + +void IGCKernel::add_2D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const BufferParam ¶m, const Window &window) +{ + add_tensor_argument<2>(idx, tensor, param, window); +} + +void IGCKernel::add_3D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window) +{ + add_tensor_argument<3>(idx, tensor, BufferParam(binding_point, 0), window); +} + +void IGCKernel::add_3D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const BufferParam ¶m, const Window &window) +{ + add_tensor_argument<3>(idx, tensor, param, window); +} + +unsigned int IGCKernel::num_arguments_per_1D_tensor() const +{ + return num_arguments_per_tensor<1>(); +} + +unsigned int IGCKernel::num_arguments_per_2D_tensor() const +{ + return num_arguments_per_tensor<2>(); +} + +unsigned int IGCKernel::num_arguments_per_3D_tensor() const +{ + return num_arguments_per_tensor<3>(); +} -- cgit v1.2.1