summaryrefslogtreecommitdiff
path: root/src
diff options
context:
space:
mode:
Diffstat (limited to 'src')
-rw-r--r--src/clblast.cpp37
-rw-r--r--src/clblast_c.cpp67
-rw-r--r--src/clblast_netlib_c.cpp90
-rw-r--r--src/routines/levelx/xim2col.cpp55
-rw-r--r--src/routines/levelx/xim2col.hpp44
5 files changed, 293 insertions, 0 deletions
diff --git a/src/clblast.cpp b/src/clblast.cpp
index ca401066..9089b17c 100644
--- a/src/clblast.cpp
+++ b/src/clblast.cpp
@@ -72,6 +72,7 @@
// Level-x includes (non-BLAS)
#include "routines/levelx/xomatcopy.hpp"
+#include "routines/levelx/xim2col.hpp"
#include "routines/levelx/xaxpybatched.hpp"
#include "routines/levelx/xgemmbatched.hpp"
@@ -2212,6 +2213,42 @@ template StatusCode PUBLIC_API Omatcopy<half>(const Layout, const Transpose,
cl_mem, const size_t, const size_t,
cl_command_queue*, cl_event*);
+// Im2col function (non-BLAS function): SIM2COL/DIM2COL/CIM2COL/ZIM2COL/HIM2COL
+template <typename T>
+StatusCode Im2col(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 cl_mem im_buffer, const size_t im_offset,
+ cl_mem col_buffer, const size_t col_offset,
+ cl_command_queue* queue, cl_event* event) {
+ try {
+ auto queue_cpp = Queue(*queue);
+ auto routine = Xim2col<T>(queue_cpp, event);
+ routine.DoIm2col(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ Buffer<T>(im_buffer), im_offset,
+ Buffer<T>(col_buffer), col_offset);
+ return StatusCode::kSuccess;
+ } catch (...) { return DispatchException(); }
+}
+template StatusCode PUBLIC_API Im2col<float>(const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t,
+ const cl_mem, const size_t,
+ cl_mem, const size_t,
+ cl_command_queue*, cl_event*);
+template StatusCode PUBLIC_API Im2col<double>(const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t,
+ const cl_mem, const size_t,
+ cl_mem, const size_t,
+ cl_command_queue*, cl_event*);
+template StatusCode PUBLIC_API Im2col<float2>(const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t,
+ const cl_mem, const size_t,
+ cl_mem, const size_t,
+ cl_command_queue*, cl_event*);
+template StatusCode PUBLIC_API Im2col<double2>(const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t,
+ const cl_mem, const size_t,
+ cl_mem, const size_t,
+ cl_command_queue*, cl_event*);
+template StatusCode PUBLIC_API Im2col<half>(const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t, const size_t,
+ const cl_mem, const size_t,
+ cl_mem, const size_t,
+ cl_command_queue*, cl_event*);
+
// Batched version of AXPY: SAXPYBATCHED/DAXPYBATCHED/CAXPYBATCHED/ZAXPYBATCHED/HAXPYBATCHED
template <typename T>
StatusCode AxpyBatched(const size_t n,
diff --git a/src/clblast_c.cpp b/src/clblast_c.cpp
index d2656274..24697779 100644
--- a/src/clblast_c.cpp
+++ b/src/clblast_c.cpp
@@ -3515,6 +3515,73 @@ CLBlastStatusCode CLBlastHomatcopy(const CLBlastLayout layout, const CLBlastTran
} catch (...) { return static_cast<CLBlastStatusCode>(clblast::DispatchExceptionForC()); }
}
+// IM2COL
+CLBlastStatusCode CLBlastSim2col(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 cl_mem im_buffer, const size_t im_offset,
+ cl_mem col_buffer, const size_t col_offset,
+ cl_command_queue* queue, cl_event* event) {
+ try {
+ return static_cast<CLBlastStatusCode>(
+ clblast::Im2col<float>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer, im_offset,
+ col_buffer, col_offset,
+ queue, event)
+ );
+ } catch (...) { return static_cast<CLBlastStatusCode>(clblast::DispatchExceptionForC()); }
+}
+CLBlastStatusCode CLBlastDim2col(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 cl_mem im_buffer, const size_t im_offset,
+ cl_mem col_buffer, const size_t col_offset,
+ cl_command_queue* queue, cl_event* event) {
+ try {
+ return static_cast<CLBlastStatusCode>(
+ clblast::Im2col<double>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer, im_offset,
+ col_buffer, col_offset,
+ queue, event)
+ );
+ } catch (...) { return static_cast<CLBlastStatusCode>(clblast::DispatchExceptionForC()); }
+}
+CLBlastStatusCode CLBlastCim2col(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 cl_mem im_buffer, const size_t im_offset,
+ cl_mem col_buffer, const size_t col_offset,
+ cl_command_queue* queue, cl_event* event) {
+ try {
+ return static_cast<CLBlastStatusCode>(
+ clblast::Im2col<float2>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer, im_offset,
+ col_buffer, col_offset,
+ queue, event)
+ );
+ } catch (...) { return static_cast<CLBlastStatusCode>(clblast::DispatchExceptionForC()); }
+}
+CLBlastStatusCode CLBlastZim2col(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 cl_mem im_buffer, const size_t im_offset,
+ cl_mem col_buffer, const size_t col_offset,
+ cl_command_queue* queue, cl_event* event) {
+ try {
+ return static_cast<CLBlastStatusCode>(
+ clblast::Im2col<double2>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer, im_offset,
+ col_buffer, col_offset,
+ queue, event)
+ );
+ } catch (...) { return static_cast<CLBlastStatusCode>(clblast::DispatchExceptionForC()); }
+}
+CLBlastStatusCode CLBlastHim2col(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 cl_mem im_buffer, const size_t im_offset,
+ cl_mem col_buffer, const size_t col_offset,
+ cl_command_queue* queue, cl_event* event) {
+ try {
+ return static_cast<CLBlastStatusCode>(
+ clblast::Im2col<half>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer, im_offset,
+ col_buffer, col_offset,
+ queue, event)
+ );
+ } catch (...) { return static_cast<CLBlastStatusCode>(clblast::DispatchExceptionForC()); }
+}
+
// AXPY
CLBlastStatusCode CLBlastSaxpyBatched(const size_t n,
const float *alphas,
diff --git a/src/clblast_netlib_c.cpp b/src/clblast_netlib_c.cpp
index d3b9b5e6..7859dddf 100644
--- a/src/clblast_netlib_c.cpp
+++ b/src/clblast_netlib_c.cpp
@@ -4735,4 +4735,94 @@ void cblas_zomatcopy(const CLBlastLayout layout, const CLBlastTranspose a_transp
b_buffer.Read(queue, b_size, reinterpret_cast<double2*>(b));
}
+// IM2COL
+void cblas_sim2col(const int channels, const int height, const int width, const int kernel_h, const int kernel_w, const int pad_h, const int pad_w, const int stride_h, const int stride_w, const int dilation_h, const int dilation_w,
+ const float* im,
+ float* col) {
+ auto device = get_device();
+ auto context = clblast::Context(device);
+ auto queue = clblast::Queue(context, device);
+ const auto im_size = height * width * channels;
+ const auto col_size = height * width * channels;
+ auto im_buffer = clblast::Buffer<float>(context, im_size);
+ auto col_buffer = clblast::Buffer<float>(context, col_size);
+ im_buffer.Write(queue, im_size, reinterpret_cast<const float*>(im));
+ col_buffer.Write(queue, col_size, reinterpret_cast<float*>(col));
+ auto queue_cl = queue();
+ auto s = clblast::Im2col<float>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer(), 0,
+ col_buffer(), 0,
+ &queue_cl);
+ if (s != clblast::StatusCode::kSuccess) {
+ throw std::runtime_error("CLBlast returned with error code " + clblast::ToString(s));
+ }
+ col_buffer.Read(queue, col_size, reinterpret_cast<float*>(col));
+}
+void cblas_dim2col(const int channels, const int height, const int width, const int kernel_h, const int kernel_w, const int pad_h, const int pad_w, const int stride_h, const int stride_w, const int dilation_h, const int dilation_w,
+ const double* im,
+ double* col) {
+ auto device = get_device();
+ auto context = clblast::Context(device);
+ auto queue = clblast::Queue(context, device);
+ const auto im_size = height * width * channels;
+ const auto col_size = height * width * channels;
+ auto im_buffer = clblast::Buffer<double>(context, im_size);
+ auto col_buffer = clblast::Buffer<double>(context, col_size);
+ im_buffer.Write(queue, im_size, reinterpret_cast<const double*>(im));
+ col_buffer.Write(queue, col_size, reinterpret_cast<double*>(col));
+ auto queue_cl = queue();
+ auto s = clblast::Im2col<double>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer(), 0,
+ col_buffer(), 0,
+ &queue_cl);
+ if (s != clblast::StatusCode::kSuccess) {
+ throw std::runtime_error("CLBlast returned with error code " + clblast::ToString(s));
+ }
+ col_buffer.Read(queue, col_size, reinterpret_cast<double*>(col));
+}
+void cblas_cim2col(const int channels, const int height, const int width, const int kernel_h, const int kernel_w, const int pad_h, const int pad_w, const int stride_h, const int stride_w, const int dilation_h, const int dilation_w,
+ const void* im,
+ void* col) {
+ auto device = get_device();
+ auto context = clblast::Context(device);
+ auto queue = clblast::Queue(context, device);
+ const auto im_size = height * width * channels;
+ const auto col_size = height * width * channels;
+ auto im_buffer = clblast::Buffer<float2>(context, im_size);
+ auto col_buffer = clblast::Buffer<float2>(context, col_size);
+ im_buffer.Write(queue, im_size, reinterpret_cast<const float2*>(im));
+ col_buffer.Write(queue, col_size, reinterpret_cast<float2*>(col));
+ auto queue_cl = queue();
+ auto s = clblast::Im2col<float2>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer(), 0,
+ col_buffer(), 0,
+ &queue_cl);
+ if (s != clblast::StatusCode::kSuccess) {
+ throw std::runtime_error("CLBlast returned with error code " + clblast::ToString(s));
+ }
+ col_buffer.Read(queue, col_size, reinterpret_cast<float2*>(col));
+}
+void cblas_zim2col(const int channels, const int height, const int width, const int kernel_h, const int kernel_w, const int pad_h, const int pad_w, const int stride_h, const int stride_w, const int dilation_h, const int dilation_w,
+ const void* im,
+ void* col) {
+ auto device = get_device();
+ auto context = clblast::Context(device);
+ auto queue = clblast::Queue(context, device);
+ const auto im_size = height * width * channels;
+ const auto col_size = height * width * channels;
+ auto im_buffer = clblast::Buffer<double2>(context, im_size);
+ auto col_buffer = clblast::Buffer<double2>(context, col_size);
+ im_buffer.Write(queue, im_size, reinterpret_cast<const double2*>(im));
+ col_buffer.Write(queue, col_size, reinterpret_cast<double2*>(col));
+ auto queue_cl = queue();
+ auto s = clblast::Im2col<double2>(channels, height, width, kernel_h, kernel_w, pad_h, pad_w, stride_h, stride_w, dilation_h, dilation_w,
+ im_buffer(), 0,
+ col_buffer(), 0,
+ &queue_cl);
+ if (s != clblast::StatusCode::kSuccess) {
+ throw std::runtime_error("CLBlast returned with error code " + clblast::ToString(s));
+ }
+ col_buffer.Read(queue, col_size, reinterpret_cast<double2*>(col));
+}
+
// =================================================================================================
diff --git a/src/routines/levelx/xim2col.cpp b/src/routines/levelx/xim2col.cpp
new file mode 100644
index 00000000..150220d6
--- /dev/null
+++ b/src/routines/levelx/xim2col.cpp
@@ -0,0 +1,55 @@
+
+// =================================================================================================
+// 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 Xim2col class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "routines/levelx/xim2col.hpp"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xim2col<T>::Xim2col(Queue &queue, EventPointer event, const std::string &name):
+ Routine(queue, event, name, {}, PrecisionValue<T>(), {}, {
+#include "../../kernels/level3/level3.opencl"
+ }) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+void Xim2col<T>::DoIm2col(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 Buffer<T> &im_buffer, const size_t im_offset,
+ const Buffer<T> &col_buffer, const size_t col_offset) {
+
+ // Makes sure all dimensions are larger than zero
+ if ((channels == 0) || (height == 0) || (width == 0)) { throw BLASError(StatusCode::kInvalidDimension); }
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xim2col<half>;
+template class Xim2col<float>;
+template class Xim2col<double>;
+template class Xim2col<float2>;
+template class Xim2col<double2>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/levelx/xim2col.hpp b/src/routines/levelx/xim2col.hpp
new file mode 100644
index 00000000..4448b54e
--- /dev/null
+++ b/src/routines/levelx/xim2col.hpp
@@ -0,0 +1,44 @@
+
+// =================================================================================================
+// 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 Xim2col routine. The precision is implemented using a template argument.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XIM2COL_H_
+#define CLBLAST_ROUTINES_XIM2COL_H_
+
+#include "routine.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class Xim2col: public Routine {
+ public:
+
+ // Constructor
+ Xim2col(Queue &queue, EventPointer event, const std::string &name = "IM2COL");
+
+ // Templated-precision implementation of the routine
+ void DoIm2col(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 Buffer<T> &im_buffer, const size_t im_offset,
+ const Buffer<T> &col_buffer, const size_t col_offset);
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XIM2COL_H_
+#endif
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-12 21:48:35 +0000