diff options
| -rw-r--r-- | arm_compute/core/CL/kernels/CLL2NormalizeLayerKernel.h | 8 | ||||
| -rw-r--r-- | arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h | 8 | ||||
| -rw-r--r-- | src/core/CL/CLKernelLibrary.cpp | 5 | ||||
| -rw-r--r-- | src/core/CL/cl_kernels/l2_normalize.cl | 59 | ||||
| -rw-r--r-- | src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp | 45 | ||||
| -rw-r--r-- | tests/validation/CL/L2NormalizeLayer.cpp | 4 | ||||
| -rw-r--r-- | tests/validation/reference/L2NormalizeLayer.cpp | 40 |
7 files changed, 134 insertions, 35 deletions
diff --git a/arm_compute/core/CL/kernels/CLL2NormalizeLayerKernel.h b/arm_compute/core/CL/kernels/CLL2NormalizeLayerKernel.h index ae05bcf879..8dd4609250 100644 --- a/arm_compute/core/CL/kernels/CLL2NormalizeLayerKernel.h +++ b/arm_compute/core/CL/kernels/CLL2NormalizeLayerKernel.h @@ -50,24 +50,24 @@ public: /** Set the input and output tensors. * - * @param[in] input Source tensor. Data types supported: F32. Data layouts supported: NCHW/NHWC. + * @param[in] input Source tensor. Data types supported: F16/F32. Data layouts supported: NCHW/NHWC. * @param[in] sum Sum values tensor. Data types supported: same as @p input. * Sum will have the same number of dimensions as input. * @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input. * Output will have the same number of dimensions as input. - * @param[in] axis Axis along which to reduce. Supported reduction axis : 0 + * @param[in] axis Axis along which to reduce. Supported reduction axis : 0, 1, 2 * @param[in] epsilon Lower bound value for the normalization. */ void configure(const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, unsigned int axis, float epsilon); /** Static function to check if given info will lead to a valid configuration of @ref CLL2NormalizeLayerKernel. * - * @param[in] input Source tensor info. Data types supported: F32. Data layouts supported: NCHW/NHWC. + * @param[in] input Source tensor info. Data types supported: F16/F32. Data layouts supported: NCHW/NHWC. * @param[in] sum Sum values tensor info. Data types supported: same as @p input. * Sum will have the same number of dimensions as input. * @param[in] output Destination tensor info. Data types and data layouts supported: Same as @p input. * Output will have the same number of dimensions as input. - * @param[in] axis Axis along which to reduce. Supported reduction axis : 0 + * @param[in] axis Axis along which to reduce. Supported reduction axis : 0, 1, 2 * @param[in] epsilon Lower bound value for the normalization. * * @return a status diff --git a/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h b/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h index a7c47a2327..2cabaee5de 100644 --- a/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h +++ b/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h @@ -53,18 +53,18 @@ public: /** Set the input and output tensors. * - * @param[in] input Source tensor. Data types supported: F32. Data layouts supported: NCHW/NHWC. + * @param[in] input Source tensor. Data types supported: F16/F32. Data layouts supported: NCHW/NHWC. * @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input. - * @param[in] axis Axis along which to reduce. Supported reduction axis : 0, 2 + * @param[in] axis Axis along which to reduce. Supported reduction axis : 0, 1, 2 * @param[in] epsilon (Optional) Lower bound value for the normalization. */ void configure(ICLTensor *input, ICLTensor *output, unsigned int axis, float epsilon = 1e-12f); /** Static function to check if given info will lead to a valid configuration of @ref CLL2NormalizeLayer. * - * @param[in] input Source tensor info. Data types supported: F32. Data layouts supported: NCHW/NHWC. + * @param[in] input Source tensor info. Data types supported: F16/F32. Data layouts supported: NCHW/NHWC. * @param[in] output Destination tensor info. Data types and data layouts supported: Same as @p input. - * @param[in] axis Axis along which to reduce. Supported reduction axis : 0, 2, + * @param[in] axis Axis along which to reduce. Supported reduction axis : 0, 1, 2 * @param[in] epsilon (Optional) Lower bound value for the normalization. * * @return a status diff --git a/src/core/CL/CLKernelLibrary.cpp b/src/core/CL/CLKernelLibrary.cpp index 847236925a..3c2528f358 100644 --- a/src/core/CL/CLKernelLibrary.cpp +++ b/src/core/CL/CLKernelLibrary.cpp @@ -303,8 +303,9 @@ const std::map<std::string, std::string> CLKernelLibrary::_kernel_program_map = { "IYUV_to_RGB888_bt709", "color_convert.cl" }, { "IYUV_to_RGBA8888_bt709", "color_convert.cl" }, { "IYUV_to_YUV444_bt709", "color_convert.cl" }, - { "l2_normalize_nchw", "l2_normalize.cl" }, - { "l2_normalize_nhwc", "l2_normalize.cl" }, + { "l2_normalize_x", "l2_normalize.cl" }, + { "l2_normalize_y", "l2_normalize.cl" }, + { "l2_normalize_z", "l2_normalize.cl" }, { "lktracker_stage0", "optical_flow_pyramid_lk.cl" }, { "lktracker_stage1", "optical_flow_pyramid_lk.cl" }, { "magnitude_phase", "magnitude_phase.cl" }, diff --git a/src/core/CL/cl_kernels/l2_normalize.cl b/src/core/CL/cl_kernels/l2_normalize.cl index d230487030..5f66efbcc4 100644 --- a/src/core/CL/cl_kernels/l2_normalize.cl +++ b/src/core/CL/cl_kernels/l2_normalize.cl @@ -23,7 +23,7 @@ */ #include "helpers.h" -/** This kernel performs l2 normalization. (NCHW) +/** This kernel performs l2 normalization on x-axis * * @note The data type must be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=float * @note The data size must be passed at compile time using -DDATA_SIZE e.g. -DDATA_SIZE=32 @@ -42,7 +42,7 @@ * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor * @param[in] epsilon Epsilon value */ -__kernel void l2_normalize_nchw( +__kernel void l2_normalize_x( VECTOR_DECLARATION(src), VECTOR_DECLARATION(sum), VECTOR_DECLARATION(dst), @@ -60,7 +60,7 @@ __kernel void l2_normalize_nchw( vstore16(in * normalize_value, 0, (__global DATA_TYPE *)dst.ptr); } -/** This kernel performs l2 normalization. (NHWC) +/** This kernel performs l2 normalization on y-axis. * * @note The data type must be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=float * @note The data size must be passed at compile time using -DDATA_SIZE e.g. -DDATA_SIZE=32 @@ -85,7 +85,7 @@ __kernel void l2_normalize_nchw( * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor * @param[in] epsilon Epsilon value */ -__kernel void l2_normalize_nhwc( +__kernel void l2_normalize_y( IMAGE_DECLARATION(src), IMAGE_DECLARATION(sum), IMAGE_DECLARATION(dst), @@ -104,4 +104,55 @@ __kernel void l2_normalize_nhwc( normalize_value = (VEC_DATA_TYPE(DATA_TYPE, 16))rsqrt(fmax(sums, epsilon)); vstore16(in * normalize_value, 0, (__global DATA_TYPE *)dst.ptr); +} +/** This kernel performs l2 normalization on z-axis. + * + * @note The data type must be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=float + * @note The data size must be passed at compile time using -DDATA_SIZE e.g. -DDATA_SIZE=32 + * + * @param[in] src_ptr Pointer to the source tensor. Supported data types: F16/F32 + * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes) + * @param[in] src_step_x src_stride_x * number of elements along Y processed per workitem(in bytes) + * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes) + * @param[in] src_step_y src_stride_y * number of elements along X 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 tensor + * @param[in] sum_ptr Pointer to the source tensor. Supported data types: F16/F32 + * @param[in] sum_stride_x Stride of the source tensor in X dimension (in bytes) + * @param[in] sum_step_x sum_stride_x * number of elements along X processed per workitem(in bytes) + * @param[in] sum_stride_y Stride of the source tensor in Y dimension (in bytes) + * @param[in] sum_step_y sum_stride_y * number of elements along Y processed per workitem(in bytes) + * @param[in] sum_stride_z Stride of the source tensor in Z dimension (in bytes) + * @param[in] sum_step_z sum_stride_z * number of elements along Z processed per workitem(in bytes) + * @param[in] sum_offset_first_element_in_bytes The offset of the first element in the source tensor + * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr + * @param[in] dst_stride_x Stride of the destination tensor 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 tensor 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 Y processed per workitem(in bytes) + * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor + * @param[in] epsilon Epsilon value + */ +__kernel void l2_normalize_z( + TENSOR3D_DECLARATION(src), + TENSOR3D_DECLARATION(sum), + TENSOR3D_DECLARATION(dst), + DATA_TYPE epsilon) +{ + Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src); + Tensor3D sum = CONVERT_TO_TENSOR3D_STRUCT(sum); + Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst); + + VEC_DATA_TYPE(DATA_TYPE, 16) + in = vload16(0, (__global DATA_TYPE *)src.ptr); + VEC_DATA_TYPE(DATA_TYPE, 16) + sums = vload16(0, (__global DATA_TYPE *)sum.ptr); + + VEC_DATA_TYPE(DATA_TYPE, 16) + normalize_value = (VEC_DATA_TYPE(DATA_TYPE, 16))rsqrt(fmax(sums, epsilon)); + + vstore16(in * normalize_value, 0, (__global DATA_TYPE *)dst.ptr); }
\ No newline at end of file diff --git a/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp b/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp index cfd04ef392..97dd919d08 100644 --- a/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp +++ b/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp @@ -109,11 +109,28 @@ void CLL2NormalizeLayerKernel::configure(const ICLTensor *input, const ICLTensor build_opts.emplace(("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration))); // Create kernel - const DataLayout data_layout = input->info()->data_layout(); - _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("l2_normalize_" + lower_string(string_from_data_layout(data_layout)), build_opts)); + std::string kernel_name; + unsigned int idx = 0; + switch(axis) + { + case 0: + kernel_name = "x"; + idx = num_arguments_per_1D_tensor() * 3; + break; + case 1: + kernel_name = "y"; + idx = num_arguments_per_2D_tensor() * 3; + break; + case 2: + kernel_name = "z"; + idx = num_arguments_per_3D_tensor() * 3; + break; + default: + ARM_COMPUTE_ERROR("Not supported"); + } + _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("l2_normalize_" + kernel_name, build_opts)); // Set epsilon argument - unsigned int idx = data_layout == DataLayout::NCHW ? num_arguments_per_1D_tensor() * 3 : num_arguments_per_2D_tensor() * 3; if(input->info()->data_type() == DataType::F32) { _kernel.setArg<cl_uint>(idx, _epsilon); @@ -145,9 +162,9 @@ void CLL2NormalizeLayerKernel::run(const Window &window, cl::CommandQueue &queue Window window_sum(window); - switch(_input->info()->data_layout()) + switch(_axis) { - case DataLayout::NCHW: + case 0: { window_sum.set(Window::DimX, Window::Dimension(0, 0, 0)); Window in_slice = window.first_slice_window_1D(); @@ -163,7 +180,7 @@ void CLL2NormalizeLayerKernel::run(const Window &window, cl::CommandQueue &queue while(window.slide_window_slice_1D(in_slice) && window.slide_window_slice_1D(sum_slice)); } break; - case DataLayout::NHWC: + case 1: { window_sum.set(Window::DimY, Window::Dimension(0, 0, 0)); Window in_slice = window.first_slice_window_2D(); @@ -179,6 +196,22 @@ void CLL2NormalizeLayerKernel::run(const Window &window, cl::CommandQueue &queue while(window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice)); } break; + case 2: + { + window_sum.set(Window::DimZ, Window::Dimension(0, 0, 0)); + Window in_slice = window.first_slice_window_3D(); + Window sum_slice = window_sum.first_slice_window_3D(); + do + { + unsigned int idx = 0; + add_3D_tensor_argument(idx, _input, in_slice); + add_3D_tensor_argument(idx, _sum, sum_slice); + add_3D_tensor_argument(idx, _output, in_slice); + enqueue(queue, *this, in_slice); + } + while(window.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(sum_slice)); + } + break; default: ARM_COMPUTE_ERROR("Not supported"); } diff --git a/tests/validation/CL/L2NormalizeLayer.cpp b/tests/validation/CL/L2NormalizeLayer.cpp index 517ba84069..fae7fd66db 100644 --- a/tests/validation/CL/L2NormalizeLayer.cpp +++ b/tests/validation/CL/L2NormalizeLayer.cpp @@ -46,8 +46,8 @@ namespace constexpr AbsoluteTolerance<float> tolerance_f32(0.00001f); constexpr AbsoluteTolerance<float> tolerance_f16(0.01f); -auto data = concat(combine(framework::dataset::make("DataLayout", { DataLayout::NCHW }), framework::dataset::make("Axis", { 0 })), combine(framework::dataset::make("DataLayout", { DataLayout::NHWC }), - framework::dataset::make("Axis", { 1 }))); +auto data = concat(combine(framework::dataset::make("DataLayout", { DataLayout::NCHW }), framework::dataset::make("Axis", { 0, 1, 2 })), combine(framework::dataset::make("DataLayout", { DataLayout::NHWC }), + framework::dataset::make("Axis", { 0, 1 }))); } // namespace diff --git a/tests/validation/reference/L2NormalizeLayer.cpp b/tests/validation/reference/L2NormalizeLayer.cpp index 26677511e4..fcd6226f07 100644 --- a/tests/validation/reference/L2NormalizeLayer.cpp +++ b/tests/validation/reference/L2NormalizeLayer.cpp @@ -57,24 +57,38 @@ SimpleTensor<T> l2_normalize(const SimpleTensor<T> &src, unsigned int axis, floa SimpleTensor<T> sum = reduction_operation(src, get_output_shape(src.shape(), axis), axis, ReductionOperation::SUM_SQUARE); // Compute reference - const int elems = src.shape()[axis]; - const int upper_dims = src.shape().total_size_upper(axis + 1); + const int upper_dims = src.shape().total_size_upper(axis + 1); + const int lower_dims = src.shape().total_size_lower(axis + 1); + const int lower_dims_sum = sum.shape().total_size_lower(axis + 1); for(int du = 0; du < upper_dims; ++du) { - if(axis == 0) + const T *src_row_ptr = src.data() + du * lower_dims; + T *dst_row_ptr = dst.data() + du * lower_dims; + switch(axis) { - const T *src_row_ptr = src.data() + du * elems; - T *dst_row_ptr = dst.data() + du * elems; - const T normalization_value = sqrt(std::max(sum[du], static_cast<T>(epsilon))); - std::transform(src_row_ptr, src_row_ptr + elems, dst_row_ptr, [normalization_value](T val) + case 0: { - return val / normalization_value; - }); - } - else - { - ARM_COMPUTE_ERROR("Unsupported normalization axis"); + const int elems = src.shape()[0]; + const T normalization_value = sqrt(std::max(sum[du], static_cast<T>(epsilon))); + std::transform(src_row_ptr, src_row_ptr + elems, dst_row_ptr, [normalization_value](T val) + { + return val / normalization_value; + }); + } + break; + case 1: + case 2: + { + for(int ld = 0; ld < lower_dims; ++ld) + { + const T normalization_value = sqrt(std::max(sum[ld % lower_dims_sum + du * lower_dims_sum], static_cast<T>(epsilon))); + dst_row_ptr[ld] = src_row_ptr[ld] / normalization_value; + } + } + break; + default: + ARM_COMPUTE_ERROR("Axis not supported"); } } |