COMPMID-417 - Adding support for rectangular kernels

Change-Id: I4dde0929bc689c83582b95856dd0253228125df2
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/78994
Reviewed-by: Moritz Pflanzer <moritz.pflanzer@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

7 files changed, 47 insertions, 46 deletions

diff --git a/src/core/CL/kernels/CLPoolingLayerKernel.cpp b/src/core/CL/kernels/CLPoolingLayerKernel.cpp
index 7648025caa..15f26e3985 100644
--- a/src/core/CL/kernels/CLPoolingLayerKernel.cpp
+++ b/src/core/CL/kernels/CLPoolingLayerKernel.cpp
@@ -52,16 +52,15 @@ BorderSize CLPoolingLayerKernel::border_size() const
 
 void CLPoolingLayerKernel::configure(const ICLTensor *input, ICLTensor *output, const PoolingLayerInfo &pool_info)
 {
-    int                   pool_pad_x      = 0;
-    int                   pool_pad_y      = 0;
-    int                   pool_stride_x   = 0;
-    int                   pool_stride_y   = 0;
-    unsigned int          pooled_w        = 0;
-    unsigned int          pooled_h        = 0;
-    const PoolingType     pool_type       = pool_info.pool_type();
-    const int             pool_size       = pool_info.pool_size();
-    const PadStrideInfo   pad_stride_info = pool_info.pad_stride_info();
-    DimensionRoundingType pool_round      = pad_stride_info.round();
+    int                 pool_pad_x      = 0;
+    int                 pool_pad_y      = 0;
+    int                 pool_stride_x   = 0;
+    int                 pool_stride_y   = 0;
+    unsigned int        pooled_w        = 0;
+    unsigned int        pooled_h        = 0;
+    const PoolingType   pool_type       = pool_info.pool_type();
+    const int           pool_size       = pool_info.pool_size();
+    const PadStrideInfo pad_stride_info = pool_info.pad_stride_info();
     std::tie(pool_pad_x, pool_pad_y)       = pad_stride_info.pad();
     std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
 
@@ -78,9 +77,8 @@ void CLPoolingLayerKernel::configure(const ICLTensor *input, ICLTensor *output,
     std::tie(pooled_w, pooled_h) = scaled_dimensions(input->info()->dimension(0),
                                                      input->info()->dimension(1),
                                                      pool_size,
-                                                     pool_stride_x, pool_stride_y,
-                                                     pool_pad_x, pool_pad_y,
-                                                     pool_round);
+                                                     pool_size,
+                                                     pool_info.pad_stride_info());
     ARM_COMPUTE_UNUSED(pooled_w);
     ARM_COMPUTE_UNUSED(pooled_h);
     ARM_COMPUTE_ERROR_ON((output->info()->dimension(0) != pooled_w) || (output->info()->dimension(1) != pooled_h));
diff --git a/src/core/NEON/kernels/NEPoolingLayerKernel.cpp b/src/core/NEON/kernels/NEPoolingLayerKernel.cpp
index 8991e9b9ee..578b01419d 100644
--- a/src/core/NEON/kernels/NEPoolingLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEPoolingLayerKernel.cpp
@@ -81,16 +81,15 @@ BorderSize NEPoolingLayerKernel::border_size() const
 
 void NEPoolingLayerKernel::configure(const ITensor *input, ITensor *output, const PoolingLayerInfo &pool_info)
 {
-    int                   pool_pad_x      = 0;
-    int                   pool_pad_y      = 0;
-    int                   pool_stride_x   = 0;
-    int                   pool_stride_y   = 0;
-    unsigned int          pooled_w        = 0;
-    unsigned int          pooled_h        = 0;
-    PoolingType           pool_type       = pool_info.pool_type();
-    int                   pool_size       = pool_info.pool_size();
-    const PadStrideInfo   pad_stride_info = pool_info.pad_stride_info();
-    DimensionRoundingType pool_round      = pad_stride_info.round();
+    int                 pool_pad_x      = 0;
+    int                 pool_pad_y      = 0;
+    int                 pool_stride_x   = 0;
+    int                 pool_stride_y   = 0;
+    unsigned int        pooled_w        = 0;
+    unsigned int        pooled_h        = 0;
+    PoolingType         pool_type       = pool_info.pool_type();
+    int                 pool_size       = pool_info.pool_size();
+    const PadStrideInfo pad_stride_info = pool_info.pad_stride_info();
     std::tie(pool_pad_x, pool_pad_y)       = pad_stride_info.pad();
     std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
 
@@ -109,8 +108,7 @@ void NEPoolingLayerKernel::configure(const ITensor *input, ITensor *output, cons
 
     // Check output dimensions
     std::tie(pooled_w, pooled_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1),
-                                                     pool_size, pool_stride_x, pool_stride_y,
-                                                     pool_pad_x, pool_pad_y, pool_round);
+                                                     pool_size, pool_size, pool_info.pad_stride_info());
     ARM_COMPUTE_UNUSED(pooled_w);
     ARM_COMPUTE_UNUSED(pooled_h);
     ARM_COMPUTE_ERROR_ON((output->info()->dimension(0) != pooled_w) || (output->info()->dimension(1) != pooled_h));
diff --git a/src/core/Utils.cpp b/src/core/Utils.cpp
index bf005c12f6..f6230c0199 100644
--- a/src/core/Utils.cpp
+++ b/src/core/Utils.cpp
@@ -233,22 +233,25 @@ std::string arm_compute::lower_string(const std::string &val)
     return res;
 }
 
-const std::pair<unsigned int, unsigned int> arm_compute::scaled_dimensions(unsigned int width, unsigned int height, unsigned int kernel_size,
-                                                                           unsigned int stride_x, unsigned int stride_y,
-                                                                           unsigned int pad_x, unsigned int pad_y,
-                                                                           DimensionRoundingType round_type)
+const std::pair<unsigned int, unsigned int> arm_compute::scaled_dimensions(unsigned int width, unsigned int height,
+                                                                           unsigned int kernel_width, unsigned int kernel_height,
+                                                                           const PadStrideInfo &pad_stride_info)
 {
-    unsigned int w = 0;
-    unsigned int h = 0;
-    switch(round_type)
+    const unsigned int pad_x    = pad_stride_info.pad().first;
+    const unsigned int pad_y    = pad_stride_info.pad().second;
+    const unsigned int stride_x = pad_stride_info.stride().first;
+    const unsigned int stride_y = pad_stride_info.stride().second;
+    unsigned int       w        = 0;
+    unsigned int       h        = 0;
+    switch(pad_stride_info.round())
     {
         case DimensionRoundingType::FLOOR:
-            w = static_cast<unsigned int>(std::floor((static_cast<float>(width + 2 * pad_x - kernel_size) / stride_x) + 1));
-            h = static_cast<unsigned int>(std::floor((static_cast<float>(height + 2 * pad_y - kernel_size) / stride_y) + 1));
+            w = static_cast<unsigned int>(std::floor((static_cast<float>(width + 2 * pad_x - kernel_width) / stride_x) + 1));
+            h = static_cast<unsigned int>(std::floor((static_cast<float>(height + 2 * pad_y - kernel_height) / stride_y) + 1));
             break;
         case DimensionRoundingType::CEIL:
-            w = static_cast<unsigned int>(std::ceil((static_cast<float>(width + 2 * pad_x - kernel_size) / stride_x) + 1));
-            h = static_cast<unsigned int>(std::ceil((static_cast<float>(height + 2 * pad_y - kernel_size) / stride_y) + 1));
+            w = static_cast<unsigned int>(std::ceil((static_cast<float>(width + 2 * pad_x - kernel_width) / stride_x) + 1));
+            h = static_cast<unsigned int>(std::ceil((static_cast<float>(height + 2 * pad_y - kernel_height) / stride_y) + 1));
             break;
         default:
             ARM_COMPUTE_ERROR("Unsupported rounding type");
diff --git a/src/runtime/CL/functions/CLConvolutionLayer.cpp b/src/runtime/CL/functions/CLConvolutionLayer.cpp
index f0bbc3514f..933d92bef7 100644
--- a/src/runtime/CL/functions/CLConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLConvolutionLayer.cpp
@@ -127,9 +127,10 @@ void CLConvolutionLayer::configure(const ICLTensor *input, const ICLTensor *weig
     unsigned int conv_w = 0;
     unsigned int conv_h = 0;
 
-    const unsigned int kernel_width = _are_weights_reshaped ? weights_info.kernel_size() : weights->info()->dimension(0);
-    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), kernel_width,
-                                                 stride_x, stride_y, pad_x, pad_y, conv_info.round());
+    const unsigned int kernel_width  = _are_weights_reshaped ? weights_info.kernel_size().first : weights->info()->dimension(0);
+    const unsigned int kernel_height = _are_weights_reshaped ? weights_info.kernel_size().second : weights->info()->dimension(1);
+    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), kernel_width, kernel_height,
+                                                 conv_info);
     ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(0) != conv_w) || (output->info()->dimension(1) != conv_h), "Output shape does not match the expected one");
 
     // Check if its a "fully connected" convolution
diff --git a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
index 263fb51987..0e6d23e0d8 100644
--- a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
+++ b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
@@ -68,8 +68,8 @@ void CLLocallyConnectedLayer::configure(const ICLTensor *input, const ICLTensor
     // Get convolved dimensions
     unsigned int conv_w = 0;
     unsigned int conv_h = 0;
-    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), weights->info()->dimension(0),
-                                                 stride_x, stride_y, pad_x, pad_y, conv_info.round());
+    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), weights->info()->dimension(0), weights->info()->dimension(1),
+                                                 conv_info);
 
     ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(0) != conv_w) || (output->info()->dimension(1) != conv_h), "Output shape does not match the expected one");
     ARM_COMPUTE_ERROR_ON_MSG(weights->info()->dimension(4) != (conv_w * conv_h), "Weights shape does not match the expected one");
diff --git a/src/runtime/NEON/functions/NEConvolutionLayer.cpp b/src/runtime/NEON/functions/NEConvolutionLayer.cpp
index 82c33d54bb..b38d6617d5 100644
--- a/src/runtime/NEON/functions/NEConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEConvolutionLayer.cpp
@@ -131,9 +131,10 @@ void NEConvolutionLayer::configure(const ITensor *input, const ITensor *weights,
     unsigned int conv_w = 0;
     unsigned int conv_h = 0;
 
-    const unsigned int kernel_width = (_are_weights_reshaped) ? weights_info.kernel_size() : weights->info()->dimension(0);
-    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), kernel_width,
-                                                 stride_x, stride_y, pad_x, pad_y, conv_info.round());
+    const unsigned int kernel_width  = (_are_weights_reshaped) ? weights_info.kernel_size().first : weights->info()->dimension(0);
+    const unsigned int kernel_height = (_are_weights_reshaped) ? weights_info.kernel_size().second : weights->info()->dimension(1);
+    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), kernel_width, kernel_height,
+                                                 conv_info);
     ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(0) != conv_w) || (output->info()->dimension(1) != conv_h), "Output shape does not match the expected one");
 
     // Check if its a "fully connected" convolution
diff --git a/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp b/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
index 85d7ba3650..3b095b4688 100644
--- a/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
@@ -68,8 +68,8 @@ void NELocallyConnectedLayer::configure(const ITensor *input, const ITensor *wei
     // Get convolved dimensions
     unsigned int conv_w = 0;
     unsigned int conv_h = 0;
-    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), weights->info()->dimension(0),
-                                                 stride_x, stride_y, pad_x, pad_y, conv_info.round());
+    std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), weights->info()->dimension(0), weights->info()->dimension(1),
+                                                 conv_info);
 
     ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(0) != conv_w) || (output->info()->dimension(1) != conv_h), "Output shape does not match the expected one");
     ARM_COMPUTE_ERROR_ON_MSG(weights->info()->dimension(4) != (conv_w * conv_h), "Weights shape does not match the expected one");