Added convgemm skeleton, test infrastructure, and first reference implementation

2 files changed, 116 insertions, 0 deletions

diff --git a/src/routines/levelx/xconvgemm.cpp b/src/routines/levelx/xconvgemm.cpp
new file mode 100644
index 00000000..2676dbda
--- /dev/null
+++ b/src/routines/levelx/xconvgemm.cpp
@@ -0,0 +1,68 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+//   Cedric Nugteren <www.cedricnugteren.nl>
+//
+// This file implements the Xconvgemm class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "routines/levelx/xconvgemm.hpp"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xconvgemm<T>::Xconvgemm(Queue &queue, EventPointer event, const std::string &name):
+    Routine(queue, event, name, {"Copy"}, PrecisionValue<T>(), {}, {
+#include "../../kernels/levelx/im2col.opencl"
+    }) {
+}
+
+// =================================================================================================
+
+template <typename T>
+void Xconvgemm<T>::DoConvgemm(const size_t channels, const size_t height, const size_t width,
+                              const size_t kernel_h, const size_t kernel_w, const size_t pad_h,
+                              const size_t pad_w, const size_t stride_h, const size_t stride_w,
+                              const size_t dilation_h, const size_t dilation_w,
+                              const size_t num_kernels, const size_t batch_count,
+                              const Buffer<T> &im_buffer, const size_t im_offset,
+                              const Buffer<T> &kernel_buffer, const size_t kernel_offset,
+                              const Buffer<T> &result_buffer, const size_t result_offset) {
+
+  // Makes sure all dimensions are larger than zero
+  if ((channels == 0) || (height == 0) || (width == 0) || (num_kernels == 0) || (batch_count == 0)) {
+    throw BLASError(StatusCode::kInvalidDimension);
+  }
+
+  // Sets the output height and width
+  const auto size_h = height + 2 * pad_h;
+  const auto padding_h = dilation_h * (kernel_h - 1) + 1;
+  const auto output_h = (size_h >= padding_h) ? (size_h - padding_h) / stride_h + 1 : 1;
+  const auto size_w = width + 2 * pad_w;
+  const auto padding_w = dilation_w * (kernel_w - 1) + 1;
+  const auto output_w = (size_w >= padding_w) ? (size_w - padding_w) / stride_w + 1 : 1;
+
+  throw BLASError(StatusCode::kNotImplemented);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xconvgemm<half>;
+template class Xconvgemm<float>;
+template class Xconvgemm<double>;
+template class Xconvgemm<float2>;
+template class Xconvgemm<double2>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/levelx/xconvgemm.hpp b/src/routines/levelx/xconvgemm.hpp
new file mode 100644
index 00000000..01795ea8
--- /dev/null
+++ b/src/routines/levelx/xconvgemm.hpp
@@ -0,0 +1,48 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+//   Cedric Nugteren <www.cedricnugteren.nl>
+//
+// This file implements the Xconvgemm routine. The precision is implemented as a template argument.
+// This implements batched convolution of a 4D input 'image' tensor, a 3D input 'kernel' matrix,
+// resulting in a 4D output 'result' tensor.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XCONVGEMM_H_
+#define CLBLAST_ROUTINES_XCONVGEMM_H_
+
+#include "routine.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class Xconvgemm: public Routine {
+ public:
+
+  // Constructor
+  Xconvgemm(Queue &queue, EventPointer event, const std::string &name = "CONVGEMM");
+
+  // Templated-precision implementation of the routine
+  void DoConvgemm(const size_t channels, const size_t height, const size_t width,
+                  const size_t kernel_h, const size_t kernel_w,
+                  const size_t pad_h, const size_t pad_w,
+                  const size_t stride_h, const size_t stride_w,
+                  const size_t dilation_h, const size_t dilation_w,
+                  const size_t num_kernels, const size_t batch_count,
+                  const Buffer<T> &im_buffer, const size_t im_offset,
+                  const Buffer<T> &kernel_buffer, const size_t kernel_offset,
+                  const Buffer<T> &result_buffer, const size_t result_offset);
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XCONVGEMM_H_
+#endif