From 3704464b68ddd22739b38354de1721a3db4267b5 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Tue, 30 Oct 2018 14:53:25 +0000 Subject: COMPMID-1703: Collapse the 4th dimensions in CLDepthWiseConvolutionLayer3x3Kernel Change-Id: Ie274da79b15c03f86dfedc85bb721b3de34a0bb4 --- src/core/CL/cl_kernels/depthwise_convolution.cl | 121 +++++++++++++-------- .../cl_kernels/depthwise_convolution_quantized.cl | 120 ++++++++++++++------ .../CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp | 22 ++-- 3 files changed, 178 insertions(+), 85 deletions(-) (limited to 'src') diff --git a/src/core/CL/cl_kernels/depthwise_convolution.cl b/src/core/CL/cl_kernels/depthwise_convolution.cl index 97b46c47cf..bfaa92be10 100644 --- a/src/core/CL/cl_kernels/depthwise_convolution.cl +++ b/src/core/CL/cl_kernels/depthwise_convolution.cl @@ -147,12 +147,12 @@ inline float2 convolution3x3( /** This OpenCL kernel computes the depthwise convolution 3x3 * - * @param[in] src_ptr Pointer to the source image. Supported data types: F32 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: 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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image + * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor * @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] dst_ptr Pointer to the destination tensor. Supported data types: F32 @@ -271,12 +271,12 @@ __kernel void depthwise_convolution_3x3( /** This OpenCL kernel is optimized for Bifrost architectures and computes the depthwise convolution 3x3 when both * stride_x and stride_y are equal to 1 * - * @param[in] src_ptr Pointer to the source image. Supported data types: F32 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: 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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image + * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor * @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] dst_ptr Pointer to the destination tensor. Supported data types: F32 @@ -372,12 +372,12 @@ __kernel void depthwise_convolution_3x3_stridex1_stridey1_bifrost_f32( /** This OpenCL kernel is optimized for Bifrost architectures and computes the depthwise convolution 3x3 when both * stride_x and stride_y are equal to 2 * - * @param[in] src_ptr Pointer to the source image. Supported data types: F32 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: 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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image + * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor * @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] dst_ptr Pointer to the destination tensor. Supported data types: F32 @@ -419,8 +419,8 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f32( float2 pixels1 = 0.0f; // Extract channel and linearized batch indices - const int channel = get_global_id(2) % DST_CHANNELS; - const int batch = get_global_id(2) / DST_CHANNELS; + const int channel = get_global_id(2) % DST_CHANNELS; + const int batch = get_global_id(2) / DST_CHANNELS; // Load relevant input and weights data (Accounts depth multiplier when indexing input, OFM = IFM * DEPTH_MULTIPLIER) __global uchar *weights_addr = weights.ptr + get_global_id(0) * weights_step_x + get_global_id(1) * weights_step_y + channel * weights_step_z; __global uchar *src_addr = src.ptr - batch * (DST_CHANNELS / DEPTH_MULTIPLIER) * (DEPTH_MULTIPLIER - 1) * src_step_z - (channel - (channel / DEPTH_MULTIPLIER)) * src_step_z; @@ -753,14 +753,14 @@ inline half4 convolution3x3_f16( /** This OpenCL kernel computes the depthwise convolution 3x3 * - * @param[in] src_ptr Pointer to the source image. Supported data types: F16 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: F16 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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[in] 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) @@ -826,14 +826,14 @@ __kernel void depthwise_convolution_3x3_f16( /** This OpenCL kernel is optimized for Bifrost architectures and computes the 16bit floating point depthwise convolution 3x3 * when both stride_x and stride_y are equal to 1 * - * @param[in] src_ptr Pointer to the source image. Supported data types: F16 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: F16 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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[in] 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) @@ -930,14 +930,14 @@ __kernel void depthwise_convolution_3x3_stridex1_stridey1_bifrost_f16( /** This OpenCL kernel is optimized for Bifrost architectures and computes 16bit floating point the depthwise convolution 3x3 * when both stride_x and stride_y are equal to 2 * - * @param[in] src_ptr Pointer to the source image. Supported data types: F16 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: F16 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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_step_z src_stride_y * 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] 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) @@ -1043,14 +1043,16 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f16( * @note The convolution stride along the width must be passed at compile time using -DCONV_STRIDE_X (e.g. -DCONV_STRIDE_Y=X) * @note The convolution stride along the height must be passed at compile time using -DCONV_STRIDE_Y (e.g. -DCONV_STRIDE_Y=1) * - * @param[in] src_ptr Pointer to the source image. Supported data types: FP32 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: FP32 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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_step_z src_stride_y * number of elements along Z processed per workitem(in bytes) + * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes) + * @param[in] src_step_w src_stride_w * number of elements along W 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] dst_ptr Pointer to the destination tensor. Supported data types: same as 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) @@ -1058,6 +1060,8 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f16( * @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_stride_w Stride of the destination tensor in W dimension (in bytes) + * @param[in] dst_step_w dst_stride_w * number of elements along W 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] weights_ptr Pointer to the weights tensor. Supported data types: QASYMM8 * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes) @@ -1074,8 +1078,8 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f16( * @param[in] biases_offset_first_element_in_bytes (Optional) The offset of the first element in the biases vector */ __kernel void depthwise_convolution_3x3_nhwc( - TENSOR3D_DECLARATION(src), - TENSOR3D_DECLARATION(dst), + TENSOR4D_DECLARATION(src), + TENSOR4D_DECLARATION(dst), TENSOR3D_DECLARATION(weights), #if defined(HAS_BIAS) VECTOR_DECLARATION(biases), @@ -1084,11 +1088,20 @@ __kernel void depthwise_convolution_3x3_nhwc( { int x = get_global_id(0); // channels int y = get_global_id(1); // spatial coordinate x +#if defined(DST_DEPTH) + int z = get_global_id(2) % (int)DST_DEPTH; // spatial coordinate y + int b = get_global_id(2) / (int)DST_DEPTH; // batch +#else /* defined(DST_DEPTH) */ int z = get_global_id(2); // spatial coordinate y +#endif /* defined(DST_DEPTH) */ Vector weights = CONVERT_TO_VECTOR_STRUCT(weights); +#if defined(DST_DEPTH) + __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE + b * src_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE; +#endif /* defined(DST_DEPTH) */ int z_coord = 0; int4 offset = 0; @@ -1158,9 +1171,14 @@ __kernel void depthwise_convolution_3x3_nhwc( acc += bias_values; #endif // defined(HAS_BIAS) - Image dst = CONVERT_TENSOR3D_TO_IMAGE_STRUCT(dst); +#if defined(DST_DEPTH) + __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z + b * dst_stride_w; +#else /* defined(DST_DEPTH) */ + __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z; +#endif /* defined(DST_DEPTH) */ + VSTORE(VEC_SIZE) - (acc, 0, (__global DATA_TYPE *)(dst.ptr)); + (acc, 0, (__global DATA_TYPE *)(dst_addr)); } #endif // defined(CONV_STRIDE_X) && defined(CONV_STRIDE_Y) @@ -1174,14 +1192,16 @@ __kernel void depthwise_convolution_3x3_nhwc( * @note The convolution pad top must be passed at compile time using -DCONV_PAD_TOP (e.g. -DCONV_PAD_TOP=1) * @note The convolution pad top must be passed at compile time using -DCONV_PAD_LEFT (e.g. -DCONV_PAD_LEFT=1) * - * @param[in] src_ptr Pointer to the source image. Supported data types: FP32 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: FP32 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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_step_z src_stride_y * number of elements along Z processed per workitem(in bytes) + * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes) + * @param[in] src_step_w src_stride_w * number of elements along W 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] dst_ptr Pointer to the destination tensor. Supported data types: same as 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) @@ -1189,6 +1209,8 @@ __kernel void depthwise_convolution_3x3_nhwc( * @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_stride_w Stride of the destination tensor in W dimension (in bytes) + * @param[in] dst_step_w dst_stride_w * number of elements along W 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] weights_ptr Pointer to the weights tensor. Supported data types: QASYMM8 * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes) @@ -1205,8 +1227,8 @@ __kernel void depthwise_convolution_3x3_nhwc( * @param[in] biases_offset_first_element_in_bytes (Optional) The offset of the first element in the biases vector */ __kernel void depthwise_convolution_3x3_nhwc_stride1( - TENSOR3D_DECLARATION(src), - TENSOR3D_DECLARATION(dst), + TENSOR4D_DECLARATION(src), + TENSOR4D_DECLARATION(dst), TENSOR3D_DECLARATION(weights), #if defined(HAS_BIAS) VECTOR_DECLARATION(biases), @@ -1215,11 +1237,20 @@ __kernel void depthwise_convolution_3x3_nhwc_stride1( { int x = get_global_id(0); // channels int y = get_global_id(1); // spatial coordinate x +#if defined(DST_DEPTH) + int z = get_global_id(2) % (int)DST_DEPTH; // spatial coordinate y + int b = get_global_id(2) / (int)DST_DEPTH; // batch +#else /* defined(DST_DEPTH) */ int z = get_global_id(2); // spatial coordinate y +#endif /* defined(DST_DEPTH) */ Vector weights = CONVERT_TO_VECTOR_STRUCT(weights); +#if defined(DST_DEPTH) + __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE + b * src_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE; +#endif /* defined(DST_DEPTH) */ int z_coord = 0; int4 offset = 0; @@ -1340,7 +1371,11 @@ __kernel void depthwise_convolution_3x3_nhwc_stride1( acc3 += bias_values; #endif // defined(HAS_BIAS) +#if defined(DST_DEPTH) + __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + (z * NUM_PLANES_PROCESSED) * dst_step_z + b * dst_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + (z * NUM_PLANES_PROCESSED) * dst_step_z; +#endif /* defined(DST_DEPTH) */ VSTORE(VEC_SIZE) (acc0, 0, (__global DATA_TYPE *)(dst_addr + 0 * dst_stride_y)); diff --git a/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl b/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl index 421c8b6aab..5a732b4863 100644 --- a/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl +++ b/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl @@ -87,14 +87,14 @@ /** This function computes the depthwise convolution quantized. * - * @param[in] src_ptr Pointer to the source image. Supported data types: QASYMM8 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: QASYMM8 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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[in] dst_ptr Pointer to the destination tensor. Supported data types: QASYMM8 * @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) @@ -319,14 +319,14 @@ __kernel void depthwise_convolution_3x3_quantized_nchw( #endif /* CONV_STRIDE_X */ /** This function computes the depthwise convolution quantized using dot product when the data layout is NCHW. * - * @param[in] src_ptr Pointer to the source image. Supported data types: QASYMM8 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: QASYMM8 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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[in] dst_ptr Pointer to the destination tensor. Supported data types: QASYMM8 * @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) @@ -644,14 +644,16 @@ __kernel void depthwise_convolution_3x3_quantized_dot8_nchw( * @note The convolution stride along the width must be passed at compile time using -DCONV_STRIDE_X (e.g. -DCONV_STRIDE_Y=X) * @note The convolution stride along the height must be passed at compile time using -DCONV_STRIDE_Y (e.g. -DCONV_STRIDE_Y=1) * - * @param[in] src_ptr Pointer to the source image. Supported data types: QASYMM8 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: QASYMM8 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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_step_z src_stride_y * number of elements along Z processed per workitem(in bytes) + * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes) + * @param[in] src_step_w src_stride_w * number of elements along W 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] 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) @@ -659,6 +661,8 @@ __kernel void depthwise_convolution_3x3_quantized_dot8_nchw( * @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_stride_w Stride of the destination tensor in W dimension (in bytes) + * @param[in] dst_step_w dst_stride_w * number of elements along W 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] weights_ptr Pointer to the weights tensor. Supported data types: same as @p src_ptr * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes) @@ -675,8 +679,8 @@ __kernel void depthwise_convolution_3x3_quantized_dot8_nchw( * @param[in] max_offset Max offset for the input tensor */ __kernel void depthwise_convolution_3x3_quantized_nhwc( - TENSOR3D_DECLARATION(src), - TENSOR3D_DECLARATION(dst), + TENSOR4D_DECLARATION(src), + TENSOR4D_DECLARATION(dst), TENSOR3D_DECLARATION(weights), #if defined(HAS_BIAS) VECTOR_DECLARATION(biases), @@ -685,11 +689,20 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc( { const int x = get_global_id(0); // channels const int y = get_global_id(1); // spatial coordinate x - const int z = get_global_id(2); // spatial coordinate y +#if defined(DST_DEPTH) + int z = get_global_id(2) % (int)DST_DEPTH; // spatial coordinate y + int b = get_global_id(2) / (int)DST_DEPTH; // batch +#else /* defined(DST_DEPTH) */ + int z = get_global_id(2); // spatial coordinate y +#endif /* defined(DST_DEPTH) */ Vector weights = CONVERT_TO_VECTOR_STRUCT(weights); +#if defined(DST_DEPTH) + __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * VEC_SIZE + b * src_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * VEC_SIZE; +#endif /* defined(DST_DEPTH) */ int z_coord = 0; int4 offset = 0; @@ -799,9 +812,14 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc( VEC_UCHAR res = CONVERT_SAT(acc, VEC_UCHAR); res = CLAMP(res, (VEC_UCHAR)0, (VEC_UCHAR)255); - Image dst = CONVERT_TENSOR3D_TO_IMAGE_STRUCT(dst); +#if defined(DST_DEPTH) + __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z + b * dst_stride_w; +#else /* defined(DST_DEPTH) */ + __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z; +#endif /* defined(DST_DEPTH) */ + VSTORE(VEC_SIZE) - (ACTIVATION_FUNC(res), 0, dst.ptr); + (ACTIVATION_FUNC(res), 0, dst_addr); } #endif // defined(CONV_STRIDE_X) && defined(CONV_STRIDE_Y) @@ -815,14 +833,16 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc( * @note The convolution pad top must be passed at compile time using -DCONV_PAD_TOP (e.g. -DCONV_PAD_TOP=1) * @note The convolution pad top must be passed at compile time using -DCONV_PAD_LEFT (e.g. -DCONV_PAD_LEFT=1). * - * @param[in] src_ptr Pointer to the source image. Supported data types: QASYMM8 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: QASYMM8 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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_step_z src_stride_y * number of elements along Z processed per workitem(in bytes) + * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes) + * @param[in] src_step_w src_stride_w * number of elements along W 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] 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) @@ -830,6 +850,8 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc( * @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_stride_w Stride of the destination tensor in W dimension (in bytes) + * @param[in] dst_step_w dst_stride_w * number of elements along W 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] weights_ptr Pointer to the weights tensor. Supported data types: same as @p src_ptr * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes) @@ -847,8 +869,8 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc( */ __kernel void depthwise_convolution_3x3_quantized_nhwc_stride1( - TENSOR3D_DECLARATION(src), - TENSOR3D_DECLARATION(dst), + TENSOR4D_DECLARATION(src), + TENSOR4D_DECLARATION(dst), TENSOR3D_DECLARATION(weights), #if defined(HAS_BIAS) VECTOR_DECLARATION(biases), @@ -857,11 +879,20 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc_stride1( { int x = get_global_id(0); int y = get_global_id(1); - int z = get_global_id(2); +#if defined(DST_DEPTH) + int z = get_global_id(2) % (int)DST_DEPTH; // spatial coordinate y + int b = get_global_id(2) / (int)DST_DEPTH; // batch +#else /* defined(DST_DEPTH) */ + int z = get_global_id(2); // spatial coordinate y +#endif /* defined(DST_DEPTH) */ Vector weights = CONVERT_TO_VECTOR_STRUCT(weights); +#if defined(DST_DEPTH) + __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * VEC_SIZE + b * src_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * VEC_SIZE; +#endif /* defined(DST_DEPTH) */ int z_coord = 0; int4 offset = 0; @@ -1050,7 +1081,11 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc_stride1( res2 = CLAMP(res2, (VEC_UCHAR)0, (VEC_UCHAR)255); res3 = CLAMP(res3, (VEC_UCHAR)0, (VEC_UCHAR)255); +#if defined(DST_DEPTH) + __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + (z * NUM_PLANES_PROCESSED) * dst_step_z + b * dst_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + (z * NUM_PLANES_PROCESSED) * dst_step_z; +#endif /* defined(DST_DEPTH) */ VSTORE(VEC_SIZE) (ACTIVATION_FUNC(res0), 0, dst_addr + 0 * dst_stride_y); @@ -1080,14 +1115,16 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc_stride1( * @note If REAL_MULTIPLIER is passed at compile time (i.e. -DREAL_MULTIPLIER=1.355f), the final quantization is performed using a floating point multiplication. * If not, the quantization will be performed using a fixed point multiplication * - * @param[in] src_ptr Pointer to the source image. Supported data types: QASYMM8 - * @param[in] src_stride_x Stride of the source image in X dimension (in bytes) + * @param[in] src_ptr Pointer to the source tensor. Supported data types: QASYMM8 + * @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 image in Y dimension (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_offset_first_element_in_bytes The offset of the first element in the source image * @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_step_z src_stride_y * number of elements along Z processed per workitem(in bytes) + * @param[in] src_stride_w Stride of the source tensor in W dimension (in bytes) + * @param[in] src_step_w src_stride_w * number of elements along W 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] dst_ptr Pointer to the destination tensor. Supported data types: QASYMM8 * @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) @@ -1095,6 +1132,8 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc_stride1( * @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_stride_w Stride of the destination tensor in W dimension (in bytes) + * @param[in] dst_step_w dst_stride_w * number of elements along W 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] weights_ptr Pointer to the weights tensor. Supported data types: QASYMM8 * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes) @@ -1112,8 +1151,8 @@ __kernel void depthwise_convolution_3x3_quantized_nhwc_stride1( */ __kernel void depthwise_convolution_3x3_quantized_dot8_nhwc_stride1( - TENSOR3D_DECLARATION(src), - TENSOR3D_DECLARATION(dst), + TENSOR4D_DECLARATION(src), + TENSOR4D_DECLARATION(dst), TENSOR3D_DECLARATION(weights), #if defined(HAS_BIAS) VECTOR_DECLARATION(biases), @@ -1122,11 +1161,20 @@ __kernel void depthwise_convolution_3x3_quantized_dot8_nhwc_stride1( { int x = get_global_id(0); int y = get_global_id(1); - int z = get_global_id(2); +#if defined(DST_DEPTH) + int z = get_global_id(2) % (int)DST_DEPTH; // spatial coordinate y + int b = get_global_id(2) / (int)DST_DEPTH; // batch +#else /* defined(DST_DEPTH) */ + int z = get_global_id(2); // spatial coordinate y +#endif /* defined(DST_DEPTH) */ Vector weights = CONVERT_TO_VECTOR_STRUCT(weights); +#if defined(DST_DEPTH) + __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * VEC_SIZE + b * src_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * VEC_SIZE; +#endif /* defined(DST_DEPTH) */ int z_coord = 0; int4 offset = 0; @@ -1249,7 +1297,11 @@ __kernel void depthwise_convolution_3x3_quantized_dot8_nhwc_stride1( res0 = CLAMP(res0, (VEC_UCHAR)0, (VEC_UCHAR)255); res1 = CLAMP(res1, (VEC_UCHAR)0, (VEC_UCHAR)255); +#if defined(DST_DEPTH) + __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z + b * dst_stride_w; +#else /* defined(DST_DEPTH) */ __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z; +#endif /* defined(DST_DEPTH) */ VSTORE(VEC_SIZE) (ACTIVATION_FUNC(res0), 0, dst_addr + 0 * dst_stride_y); diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp index 93d96dad1b..d76f5495f1 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp @@ -245,6 +245,8 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::configure(const ICLTensor *input, build_opts.add_option("-DCONV_STRIDE_X=" + support::cpp11::to_string(conv_stride_x)); build_opts.add_option("-DCONV_STRIDE_Y=" + support::cpp11::to_string(_conv_stride_y)); } + build_opts.add_option_if(_input->info()->tensor_shape().total_size_upper(3) > 1, + "-DDST_DEPTH=" + support::cpp11::to_string(static_cast(std::ceil(_output->info()->dimension(2) / static_cast(_num_planes_processed_per_iteration))))); // Create kernel std::string kernel_name = std::string("depthwise_convolution_3x3") + (is_qasymm ? std::string("_quantized") + ((is_dot8_supported @@ -291,8 +293,12 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::run(const Window &window, cl::Com ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window); - Window win = window; - win.set(Window::DimZ, Window::Dimension(0, std::ceil(_output->info()->dimension(2) / static_cast(_num_planes_processed_per_iteration)), 1)); + // Collapse window + Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ); + const size_t total_batches = _input->info()->tensor_shape().total_size_upper(3); + + Window win = window_collapsed; + win.set(Window::DimZ, Window::Dimension(0, std::ceil(_output->info()->dimension(2) / static_cast(_num_planes_processed_per_iteration)) * total_batches, 1)); // Create input window and adjust Window win_in = win; @@ -301,10 +307,10 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::run(const Window &window, cl::Com ARM_COMPUTE_ERROR_ON((win_in.y().step() < window.y().step()) || (win_in.z().step() < window.z().step())); - Window slice_in = win_in.first_slice_window_3D(); - Window slice_out = win.first_slice_window_3D(); + Window slice_in = win_in.first_slice_window_4D(); + Window slice_out = win.first_slice_window_4D(); - unsigned int idx = 3 * num_arguments_per_3D_tensor(); + unsigned int idx = 2 * num_arguments_per_4D_tensor() + num_arguments_per_3D_tensor(); if(_biases != nullptr) { @@ -321,11 +327,11 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::run(const Window &window, cl::Com do { unsigned int idx = 0; - add_3D_tensor_argument(idx, _input, slice_in); - add_3D_tensor_argument(idx, _output, slice_out); + add_4D_tensor_argument(idx, _input, slice_in); + add_4D_tensor_argument(idx, _output, slice_out); add_3D_tensor_argument(idx, _weights, slice_out); enqueue(queue, *this, slice_out, lws_hint()); } - while(window.slide_window_slice_3D(slice_out) && win_in.slide_window_slice_3D(slice_in)); + while(win.slide_window_slice_4D(slice_out) && win_in.slide_window_slice_4D(slice_in)); } -- cgit v1.2.1