COMPMID-926 Add depth multiplier support to NEON/CL/GLES depthwise convolution

Change-Id: I03f32c62350e5ea43e77bb15fc5a832d83719e3b
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/126657
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

5 files changed, 39 insertions, 20 deletions

diff --git a/src/core/CL/cl_kernels/depthwise_convolution.cl b/src/core/CL/cl_kernels/depthwise_convolution.cl
index 07e67f4f2c..21c28539ef 100644
--- a/src/core/CL/cl_kernels/depthwise_convolution.cl
+++ b/src/core/CL/cl_kernels/depthwise_convolution.cl
@@ -24,6 +24,7 @@
 
 #include "helpers.h"
 
+#if defined(DEPTH_MULTIPLIER)
 #if defined(CONV_STRIDE_X)
 
 #if CONV_STRIDE_X == 1
@@ -192,6 +193,8 @@ __kernel void depthwise_convolution_3x3(
     Vector biases = CONVERT_TO_VECTOR_STRUCT_NO_STEP(biases);
 #endif //defined(HAS_BIAS)
 
+    src.ptr -= (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;
+
     uchar3 offset          = (uchar3)(0, 1, 2) * (uchar3)weights_stride_y;
     float3 weights_values0 = vload3(0, (__global float *)(weights.ptr + offset.s0));
     float3 weights_values1 = vload3(0, (__global float *)(weights.ptr + offset.s1));
@@ -312,7 +315,7 @@ __kernel void depthwise_convolution_3x3_stridex1_stridey1_bifrost_f32(
     float2 pixels3 = 0.0f;
 
     __global uchar *weights_addr = (__global uchar *)weights.ptr;
-    __global uchar *src_addr     = (__global uchar *)offset(&src, 0, 0);
+    __global uchar *src_addr     = src.ptr - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;
 
     // Load the weights
     float3 weights_row0 = vload3(0, (__global float *)(weights_addr + 0 * weights_stride_y));
@@ -407,7 +410,7 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f32(
     float2 pixels1 = 0.0f;
 
     __global uchar *weights_addr = (__global uchar *)weights.ptr;
-    __global uchar *src_addr     = (__global uchar *)offset(&src, 0, 0);
+    __global uchar *src_addr     = src.ptr - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;
 
     // Load the weights
     float3 weights_row0 = vload3(0, (__global float *)(weights_addr + 0 * weights_stride_y));
@@ -446,6 +449,8 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f32(
     vstore2(pixels1, 0, (__global float *)(dst.ptr + 1 * dst_stride_y));
 }
 
+#endif // defined(DEPTH_MULTIPLIER)
+
 #if defined(SRC_WIDTH) && defined(DATA_TYPE)
 /** This kernel reshapes each of the tensor's low three dimensions to single rows.
  *
@@ -501,11 +506,11 @@ __kernel void depthwise_weights_reshape(
 }
 #endif //defined(SRC_WIDTH) && defined(DATA_TYPE)
 
-#if defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(SRC_HEIGHT) && defined(DATA_TYPE) && defined(PAD_VALUE)
+#if defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(SRC_HEIGHT) && defined(DATA_TYPE) && defined(PAD_VALUE) && defined(DEPTH_MULTIPLIER)
 /** This kernel performs a reshaping of the input tensor to a tensor used to perform depthwise convolution using vector to matrix multiplication.
  *
  * @note The data type must be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=float
- * @note The convolution information must be passed at compile time using -DSTRIDE_X, -DSTRIDE_Y, -DPAD_LEFT, -DPAD_TOP, -DPAD_RIGHT, -DPAD_BOTTOM, -DKERNEL_WIDHT, -DKERNEL_HEIGHT, -DSRC_WIDTH, -DSRC_HEIGHT
+ * @note The convolution information must be passed at compile time using -DSTRIDE_X, -DSTRIDE_Y, -DPAD_LEFT, -DPAD_TOP, -DPAD_RIGHT, -DPAD_BOTTOM, -DKERNEL_WIDHT, -DKERNEL_HEIGHT, -DSRC_WIDTH, -DSRC_HEIGHT, -DDEPTH_MULTIPLIER
  *
  * @param[in]  src_ptr                           Pointer to the source tensor. Supported data types: QS8/QS16/F16/F32
  * @param[in]  src_stride_x                      Stride of the source tensor in X dimension (in bytes)
@@ -534,7 +539,7 @@ __kernel void depthwise_im2col(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(d
 
     const int src_x = -PAD_LEFT + src_pixel_linear % max_initial_x;
     const int src_y = -PAD_TOP + src_pixel_linear / max_initial_x * STRIDE_Y;
-    const int src_z = get_global_id(2);
+    const int src_z = get_global_id(2) / DEPTH_MULTIPLIER;
 
     __global uchar *input_ptr      = src_ptr + src_offset_first_element_in_bytes + src_z * src_stride_z;
     __global DATA_TYPE *output_ptr = ((__global DATA_TYPE *)(dst.ptr));
@@ -558,7 +563,7 @@ __kernel void depthwise_im2col(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(d
 #endif // defined(HAS_BIAS)
 }
 
-#endif //defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(DATA_TYPE) && defined(PAD_VALUE)
+#endif //defined(STRIDE_X) && defined(STRIDE_Y) && defined(PAD_LEFT) && defined(PAD_TOP) && defined(PAD_RIGHT) && defined(PAD_BOTTOM) && defined(KERNEL_WIDTH) && defined(KERNEL_HEIGHT) && defined(SRC_WIDTH) && defined(DATA_TYPE) && defined(PAD_VALUE) && defined(DEPTH_MULTIPLIER)
 
 #if defined(CONV_WIDTH) && defined(CONV_HEIGHT) && defined(DATA_TYPE)
 
@@ -597,7 +602,7 @@ __kernel void depthwise_vector_to_tensor(
 
 #endif //defined(CONV_WIDTH) && defined(CONV_HEIGHT) && defined(DATA_TYPE)
 
-#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)
+#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(DEPTH_MULTIPLIER)
 #if defined(CONV_STRIDE_X)
 #if CONV_STRIDE_X == 1
 #define convolution1x3_f16 convolution1x3_stride_1_f16
@@ -716,6 +721,8 @@ inline half4 convolution3x3_f16(
     return pixels;
 }
 
+#if defined(DEPTH_MULTIPLIER)
+
 /** This OpenCL kernel computes the depthwise convolution 3x3
  *
  * @param[in] src_ptr                               Pointer to the source image. Supported data types: F16
@@ -764,6 +771,8 @@ __kernel void depthwise_convolution_3x3_f16(
     Vector biases = CONVERT_TO_VECTOR_STRUCT_NO_STEP(biases);
 #endif //defined(HAS_BIAS)
 
+    src.ptr -= (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;
+
     uchar3 offset         = (uchar3)(0, 1, 2) * (uchar3)weights_stride_y;
     half3 weights_values0 = vload3(0, (__global half *)(weights.ptr + offset.s0));
     half3 weights_values1 = vload3(0, (__global half *)(weights.ptr + offset.s1));
@@ -778,6 +787,7 @@ __kernel void depthwise_convolution_3x3_f16(
 
     vstore4(pixels, 0, (__global half *)dst.ptr);
 }
+#endif // defined(DEPTH_MULTIPLIER)
 #endif // defined(CONV_STRIDE_X)
 
 /** This OpenCL kernel is optimized for Bifrost architectures and computes the 16bit floating point depthwise convolution 3x3
@@ -838,7 +848,7 @@ __kernel void depthwise_convolution_3x3_stridex1_stridey1_bifrost_f16(
     half4 pixels3 = 0.0f;
 
     __global uchar *weights_addr = (__global uchar *)weights.ptr;
-    __global uchar *src_addr     = (__global uchar *)offset(&src, 0, 0);
+    __global uchar *src_addr     = (__global uchar *)offset(&src, 0, 0) - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;
 
     // Load the weights
     half3 weights_row0 = vload3(0, (__global half *)(weights_addr + 0 * weights_stride_y));
@@ -935,7 +945,7 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f16(
     half4 pixels1 = 0.0f;
 
     __global uchar *weights_addr = (__global uchar *)weights.ptr;
-    __global uchar *src_addr     = (__global uchar *)offset(&src, 0, 0);
+    __global uchar *src_addr     = (__global uchar *)offset(&src, 0, 0) - (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;
 
     // Load the weights
     half3 weights_row0 = vload3(0, (__global half *)(weights_addr + 0 * weights_stride_y));
@@ -969,4 +979,4 @@ __kernel void depthwise_convolution_3x3_stridex2_stridey2_bifrost_f16(
     vstore4(pixels0, 0, (__global half *)(dst.ptr + 0 * dst_stride_y));
     vstore4(pixels1, 0, (__global half *)(dst.ptr + 1 * dst_stride_y));
 }
-#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)
+#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(DEPTH_MULTIPLIER)
diff --git a/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl b/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl
index a0c0a8b1fb..ccb3a1ffe2 100644
--- a/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl
+++ b/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl
@@ -126,6 +126,8 @@ __kernel void depthwise_convolution_3x3_quantized_nchw(
     int bias_value = *((__global int *)(vector_offset(&biases, get_global_id(2))));
 #endif //defined(HAS_BIAS)
 
+    src.ptr -= (get_global_id(2) - get_global_id(2) / DEPTH_MULTIPLIER) * src_step_z;
+
     uchar3 w0 = vload3(0, weights.ptr + 0 * weights_stride_y);
     uchar3 w1 = vload3(0, weights.ptr + 1 * weights_stride_y);
     uchar3 w2 = vload3(0, weights.ptr + 2 * weights_stride_y);
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp
index de68ceda11..1997a901fe 100644
--- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp
@@ -50,6 +50,7 @@ BorderSize CLDepthwiseConvolutionLayer3x3NCHWKernel::border_size() const
 }
 
 void CLDepthwiseConvolutionLayer3x3NCHWKernel::configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info,
+                                                         unsigned int        depth_multiplier,
                                                          ActivationLayerInfo act_info)
 {
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
@@ -73,7 +74,7 @@ void CLDepthwiseConvolutionLayer3x3NCHWKernel::configure(const ICLTensor *input,
     }
 
     // Get convolved dimensions
-    const TensorShape output_shape = compute_depthwise_convolution_shape(*input->info(), *weights->info(), conv_info);
+    const TensorShape output_shape = compute_depthwise_convolution_shape(*input->info(), *weights->info(), conv_info, depth_multiplier);
 
     // Output auto inizialitation if not yet initialized
     auto_init_if_empty(*output->info(),
@@ -84,6 +85,7 @@ void CLDepthwiseConvolutionLayer3x3NCHWKernel::configure(const ICLTensor *input,
                        input->info()->quantization_info());
 
     ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape);
+    ARM_COMPUTE_ERROR_ON(output->info()->dimension(2) != weights->info()->dimension(2));
 
     _input         = input;
     _output        = output;
@@ -98,6 +100,7 @@ void CLDepthwiseConvolutionLayer3x3NCHWKernel::configure(const ICLTensor *input,
     // Set build options
     ARM_COMPUTE_ERROR_ON(_conv_stride_x < 1 || _conv_stride_x > 3);
     CLBuildOptions build_opts;
+    build_opts.add_option("-DDEPTH_MULTIPLIER=" + support::cpp11::to_string(depth_multiplier));
     build_opts.add_option("-DCONV_STRIDE_X=" + support::cpp11::to_string(_conv_stride_x));
     build_opts.add_option_if(_biases != nullptr, "-DHAS_BIAS");
 
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp
index d783b9e159..a02b84fba1 100644
--- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp
@@ -41,7 +41,7 @@ using namespace arm_compute::misc::shape_calculator;
 
 namespace
 {
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier,
                           const ActivationLayerInfo &act_info)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8);
@@ -50,6 +50,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights,
                                     && (act_info.activation() != ActivationLayerInfo::ActivationFunction::RELU),
                                     "For QASYMM8 only relu, lower bounded relu and lower-upper bounded relu are supported");
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights);
+    ARM_COMPUTE_RETURN_ERROR_ON(depth_multiplier > 1); // COMPMID-1071 Add depth multiplier support for NHWC
     ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(1) != 3 || weights->dimension(2) != 3);
 
     if(biases != nullptr)
@@ -61,7 +62,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights,
 
     if(output->total_size() != 0)
     {
-        const TensorShape output_shape = compute_depthwise_convolution_shape(*input, *weights, conv_info);
+        const TensorShape output_shape = compute_depthwise_convolution_shape(*input, *weights, conv_info, depth_multiplier);
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
     }
 
@@ -105,12 +106,13 @@ BorderSize CLDepthwiseConvolutionLayer3x3NHWCKernel::border_size() const
 }
 
 void CLDepthwiseConvolutionLayer3x3NHWCKernel::configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info,
+                                                         unsigned int        depth_multiplier,
                                                          ActivationLayerInfo act_info)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
 
     // Get convolved dimensions
-    const TensorShape output_shape = compute_depthwise_convolution_shape(*input->info(), *weights->info(), conv_info);
+    const TensorShape output_shape = compute_depthwise_convolution_shape(*input->info(), *weights->info(), conv_info, depth_multiplier);
 
     // Output auto inizialitation if not yet initialized
     auto_init_if_empty(*output->info(),
@@ -120,7 +122,7 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::configure(const ICLTensor *input,
                        input->info()->fixed_point_position(),
                        input->info()->quantization_info());
 
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(), conv_info, act_info));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(), conv_info, depth_multiplier, act_info));
 
     const unsigned int conv_stride_x = conv_info.stride().first;
     ARM_COMPUTE_ERROR_ON(conv_stride_x < 1 || conv_stride_x > 2);
@@ -208,9 +210,10 @@ void CLDepthwiseConvolutionLayer3x3NHWCKernel::configure(const ICLTensor *input,
 }
 
 Status CLDepthwiseConvolutionLayer3x3NHWCKernel::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
+                                                          unsigned int        depth_multiplier,
                                                           ActivationLayerInfo act_info)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, weights, biases, output, conv_info, act_info));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, weights, biases, output, conv_info, depth_multiplier, act_info));
     ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), weights->clone().get(), output->clone().get(), conv_info).first);
 
     return Status{};
diff --git a/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp b/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp
index a0784dcad6..0aef52f791 100644
--- a/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp
+++ b/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp
@@ -42,13 +42,13 @@ CLDepthwiseIm2ColKernel::CLDepthwiseIm2ColKernel()
 {
 }
 
-void CLDepthwiseIm2ColKernel::configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+void CLDepthwiseIm2ColKernel::configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, unsigned int depth_multiplier)
 {
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
     ARM_COMPUTE_ERROR_ON(is_data_type_quantized_asymmetric(input->info()->data_type()) && has_bias);
-    ARM_COMPUTE_ERROR_ON(input->info()->dimension(2) != output->info()->dimension(2));
+    ARM_COMPUTE_ERROR_ON((input->info()->dimension(2) * depth_multiplier) != output->info()->dimension(2));
     ARM_COMPUTE_ERROR_ON(output->info()->dimension(0) != (kernel_dims.width * kernel_dims.height + ((has_bias) ? 1 : 0)));
 
     _input  = input;
@@ -68,6 +68,7 @@ void CLDepthwiseIm2ColKernel::configure(const ICLTensor *input, ICLTensor *outpu
     build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(input->info()->dimension(1)));
     build_opts.add_option("-DKERNEL_WIDTH=" + support::cpp11::to_string(kernel_dims.width));
     build_opts.add_option("-DKERNEL_HEIGHT=" + support::cpp11::to_string(kernel_dims.height));
+    build_opts.add_option("-DDEPTH_MULTIPLIER=" + support::cpp11::to_string(depth_multiplier));
     build_opts.add_option_if(has_bias, "-DHAS_BIAS");
     build_opts.add_option_if_else(is_data_type_quantized_asymmetric(input->info()->data_type()),
                                   "-DPAD_VALUE=" + support::cpp11::to_string(input->info()->quantization_info().offset),
@@ -85,8 +86,8 @@ void CLDepthwiseIm2ColKernel::configure(const ICLTensor *input, ICLTensor *outpu
     }
 
     // Configure  kernel window
-    Window win = calculate_max_window(*input->info(), Steps());
-    // The CLDepthwiseIm2ColKernel doesn't need padding so update_window_and_padding() can be skipped
+    Window win = calculate_max_window(*output->info(), Steps());
+    // CLDepthwiseIm2ColKernel doesn't need padding so update_window_and_padding() can be skipped
     output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
 
     ICLKernel::configure(win);