aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/cl_kernels/flatten.cl
diff options
context:
space:
mode:
authorGian Marco Iodice <gianmarco.iodice@arm.com>2018-06-28 16:29:29 +0100
committerAnthony Barbier <anthony.barbier@arm.com>2018-11-02 16:54:54 +0000
commit215b4ea6c9dee480a22070d5873b0b8cb52531a0 (patch)
tree398e552c4d01c0b84d03a873098a9183ba8f82e4 /src/core/CL/cl_kernels/flatten.cl
parentad486e21e5870f41774f30825c270762e08ae71e (diff)
downloadComputeLibrary-215b4ea6c9dee480a22070d5873b0b8cb52531a0.tar.gz
COMPMID-1277 - Optimizing CLIm2ColKernel for NHWC.
This patch includes: - Im2Col optimizations for NHWC using a new data layout - Refactoring of CLIm2ColKernel adding validation method and auto-init - Removed im2col_reduced from CLIm2ColKernel and created a new kernel CLFlattenLayerKernel Change-Id: I1620640b6796baa268324b33ae92cdd8de53e27c Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/141241 Tested-by: Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Diffstat (limited to 'src/core/CL/cl_kernels/flatten.cl')
-rw-r--r--src/core/CL/cl_kernels/flatten.cl57
1 files changed, 57 insertions, 0 deletions
diff --git a/src/core/CL/cl_kernels/flatten.cl b/src/core/CL/cl_kernels/flatten.cl
new file mode 100644
index 0000000000..df0f9c4886
--- /dev/null
+++ b/src/core/CL/cl_kernels/flatten.cl
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2018 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 "helpers.h"
+
+#if defined(DATA_TYPE) && defined(SRC_WIDTH) && defined(SRC_HEIGHT)
+
+/** This opencl kernel flattens the first 3 dimensions of the input tensor
+ *
+ * @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=float
+ * @note The width and height of the input tensor must be passed at compile time using -DSRC_WIDTH and -DSRC_HEIGHT. e.g. -DSRC_WIDTH=24, -DSRC_HEIGHT=24
+ *
+ * @param[in] src_ptr Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/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 X 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 Y 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 Y processed per workitem(in bytes)
+ * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor
+ * @param[out] dst_ptr Pointer to the destination tensor. 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_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ */
+__kernel void flatten(
+ TENSOR3D_DECLARATION(src),
+ VECTOR_DECLARATION(dst))
+{
+ Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
+
+ __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (get_global_id(0) + get_global_id(1) * (int)SRC_WIDTH + get_global_id(2) * (int)(SRC_WIDTH * SRC_HEIGHT)) * sizeof(
+ DATA_TYPE);
+
+ *((__global DATA_TYPE *)output_ptr) = *((__global DATA_TYPE *)src.ptr);
+}
+#endif // defined(DATA_TYPE) && defined(SRC_WIDTH) && defined(SRC_HEIGHT) \ No newline at end of file