summaryrefslogtreecommitdiff
path: root/src/routines
diff options
context:
space:
mode:
Diffstat (limited to 'src/routines')
-rw-r--r--src/routines/level1/xdotu.cc1
-rw-r--r--src/routines/level2/xger.cc112
-rw-r--r--src/routines/level2/xgerc.cc53
-rw-r--r--src/routines/level2/xgeru.cc52
-rw-r--r--src/routines/level2/xher.cc122
-rw-r--r--src/routines/level2/xhpr.cc50
-rw-r--r--src/routines/level2/xspr.cc50
-rw-r--r--src/routines/level2/xsyr.cc50
8 files changed, 489 insertions, 1 deletions
diff --git a/src/routines/level1/xdotu.cc b/src/routines/level1/xdotu.cc
index 0b1bd2a8..28d9b730 100644
--- a/src/routines/level1/xdotu.cc
+++ b/src/routines/level1/xdotu.cc
@@ -14,7 +14,6 @@
#include "internal/routines/level1/xdotu.h"
#include <string>
-#include <vector>
namespace clblast {
// =================================================================================================
diff --git a/src/routines/level2/xger.cc b/src/routines/level2/xger.cc
new file mode 100644
index 00000000..0953c8bb
--- /dev/null
+++ b/src/routines/level2/xger.cc
@@ -0,0 +1,112 @@
+
+// =================================================================================================
+// 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 Xger class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xger.h"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Specific implementations to get the memory-type based on a template argument
+template <> const Precision Xger<float>::precision_ = Precision::kSingle;
+template <> const Precision Xger<double>::precision_ = Precision::kDouble;
+template <> const Precision Xger<float2>::precision_ = Precision::kComplexSingle;
+template <> const Precision Xger<double2>::precision_ = Precision::kComplexDouble;
+
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xger<T>::Xger(Queue &queue, Event &event, const std::string &name):
+ Routine<T>(queue, event, name, {"Xger"}, precision_) {
+ source_string_ =
+ #include "../../kernels/level2/level2.opencl"
+ #include "../../kernels/level2/xger.opencl"
+ ;
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xger<T>::DoGer(const Layout layout,
+ const size_t m, const size_t n,
+ const T alpha,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc,
+ const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld) {
+
+ // Makes sure all dimensions are larger than zero
+ if (m == 0 || n == 0) { return StatusCode::kInvalidDimension; }
+
+ // Computes whether or not the matrix has an alternative layout (row or column-major).
+ const auto a_is_rowmajor = (layout == Layout::kRowMajor);
+ const auto a_one = (a_is_rowmajor) ? n : m;
+ const auto a_two = (a_is_rowmajor) ? m : n;
+
+ // Tests the matrix and the vectors for validity
+ auto status = TestMatrixA(a_one, a_two, a_buffer, a_offset, a_ld, sizeof(T));
+ if (ErrorIn(status)) { return status; }
+ status = TestVectorX(m, x_buffer, x_offset, x_inc, sizeof(T));
+ if (ErrorIn(status)) { return status; }
+ status = TestVectorY(n, y_buffer, y_offset, y_inc, sizeof(T));
+ if (ErrorIn(status)) { return status; }
+
+ // Retrieves the Xgemv kernel from the compiled binary
+ try {
+ auto& program = GetProgramFromCache();
+ auto kernel = Kernel(program, "Xger");
+
+ // Sets the kernel arguments
+ kernel.SetArgument(0, static_cast<int>(a_one));
+ kernel.SetArgument(1, static_cast<int>(a_two));
+ kernel.SetArgument(2, alpha);
+ kernel.SetArgument(3, x_buffer());
+ kernel.SetArgument(4, static_cast<int>(x_offset));
+ kernel.SetArgument(5, static_cast<int>(x_inc));
+ kernel.SetArgument(6, y_buffer());
+ kernel.SetArgument(7, static_cast<int>(y_offset));
+ kernel.SetArgument(8, static_cast<int>(y_inc));
+ kernel.SetArgument(9, a_buffer());
+ kernel.SetArgument(10, static_cast<int>(a_offset));
+ kernel.SetArgument(11, static_cast<int>(a_ld));
+ kernel.SetArgument(12, static_cast<int>(a_is_rowmajor));
+
+ // Launches the kernel
+ auto a_one_ceiled = CeilDiv(Ceil(a_one, db_["WGS1"]), db_["WPT"]);
+ auto a_two_ceiled = CeilDiv(Ceil(a_two, db_["WGS2"]), db_["WPT"]);
+ auto global = std::vector<size_t>{a_one_ceiled, a_two_ceiled};
+ auto local = std::vector<size_t>{db_["WGS1"], db_["WGS2"]};
+ status = RunKernel(kernel, global, local);
+ if (ErrorIn(status)) { return status; }
+
+ // Waits for all kernels to finish
+ queue_.Finish();
+
+ // Succesfully finished the computation
+ return StatusCode::kSuccess;
+ } catch (...) { return StatusCode::kInvalidKernel; }
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xger<float>;
+template class Xger<double>;
+template class Xger<float2>;
+template class Xger<double2>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xgerc.cc b/src/routines/level2/xgerc.cc
new file mode 100644
index 00000000..09408898
--- /dev/null
+++ b/src/routines/level2/xgerc.cc
@@ -0,0 +1,53 @@
+
+// =================================================================================================
+// 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 Xgerc class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xgerc.h"
+
+#include <string>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xgerc<T>::Xgerc(Queue &queue, Event &event, const std::string &name):
+ Xger<T>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xgerc<T>::DoGerc(const Layout layout,
+ const size_t m, const size_t n,
+ const T alpha,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc,
+ const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld) {
+
+ // Regular Ger operation on complex data, plus conjugation in the kernel guarded by the
+ // ROUTINE_GERC guard.
+ return DoGer(layout, m, n, alpha,
+ x_buffer, x_offset, x_inc,
+ y_buffer, y_offset, y_inc,
+ a_buffer, a_offset, a_ld);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xgerc<float2>;
+template class Xgerc<double2>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xgeru.cc b/src/routines/level2/xgeru.cc
new file mode 100644
index 00000000..36fd9d0a
--- /dev/null
+++ b/src/routines/level2/xgeru.cc
@@ -0,0 +1,52 @@
+
+// =================================================================================================
+// 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 Xgeru class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xgeru.h"
+
+#include <string>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xgeru<T>::Xgeru(Queue &queue, Event &event, const std::string &name):
+ Xger<T>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xgeru<T>::DoGeru(const Layout layout,
+ const size_t m, const size_t n,
+ const T alpha,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc,
+ const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld) {
+
+ // Regular Ger operation on complex data
+ return DoGer(layout, m, n, alpha,
+ x_buffer, x_offset, x_inc,
+ y_buffer, y_offset, y_inc,
+ a_buffer, a_offset, a_ld);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xgeru<float2>;
+template class Xgeru<double2>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xher.cc b/src/routines/level2/xher.cc
new file mode 100644
index 00000000..5eca44b0
--- /dev/null
+++ b/src/routines/level2/xher.cc
@@ -0,0 +1,122 @@
+
+// =================================================================================================
+// 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 Xher class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xher.h"
+
+#include <string>
+
+namespace clblast {
+// =================================================================================================
+
+// Specific implementations to get the memory-type based on a template argument
+template <> const Precision Xher<float, float>::precision_ = Precision::kSingle;
+template <> const Precision Xher<double, double>::precision_ = Precision::kDouble;
+template <> const Precision Xher<float2, float>::precision_ = Precision::kComplexSingle;
+template <> const Precision Xher<double2, double>::precision_ = Precision::kComplexDouble;
+
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T, typename U>
+Xher<T,U>::Xher(Queue &queue, Event &event, const std::string &name):
+ Routine<T>(queue, event, name, {"Xger"}, precision_) {
+ source_string_ =
+ #include "../../kernels/level2/level2.opencl"
+ #include "../../kernels/level2/xher.opencl"
+ ;
+}
+
+// =================================================================================================
+
+// Specializations to compute alpha of type 'T'
+template <> float2 Xher<float2,float>::GetAlpha(const float alpha) { return float2{alpha, 0.0f}; }
+template <> double2 Xher<double2,double>::GetAlpha(const double alpha) { return double2{alpha, 0.0}; }
+template <> float Xher<float,float>::GetAlpha(const float alpha) { return alpha; }
+template <> double Xher<double,double>::GetAlpha(const double alpha) { return alpha; }
+
+// =================================================================================================
+
+// The main routine
+template <typename T, typename U>
+StatusCode Xher<T,U>::DoHer(const Layout layout, const Triangle triangle,
+ const size_t n,
+ const U alpha,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
+ const bool packed) {
+
+ // Makes sure the dimensions are larger than zero
+ if (n == 0) { return StatusCode::kInvalidDimension; }
+
+ // The data is either in the upper or lower triangle
+ const auto is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) ||
+ (triangle == Triangle::kLower && layout == Layout::kRowMajor));
+ const auto is_rowmajor = (layout == Layout::kRowMajor);
+
+ // Creates a matching version of alpha
+ const auto matching_alpha = GetAlpha(alpha);
+
+ // Tests the matrix and the vectors for validity
+ auto status = StatusCode::kSuccess;
+ if (packed) { status = TestMatrixAP(n, a_buffer, a_offset, sizeof(T)); }
+ else { status = TestMatrixA(n, n, a_buffer, a_offset, a_ld, sizeof(T)); }
+ if (ErrorIn(status)) { return status; }
+ status = TestVectorX(n, x_buffer, x_offset, x_inc, sizeof(T));
+ if (ErrorIn(status)) { return status; }
+
+ // If alpha is zero an update is not required
+ if (alpha == U{0}) { return StatusCode::kSuccess; }
+
+ // Retrieves the Xgemv kernel from the compiled binary
+ try {
+ auto& program = GetProgramFromCache();
+ auto kernel = Kernel(program, "Xher");
+
+ // Sets the kernel arguments
+ kernel.SetArgument(0, static_cast<int>(n));
+ kernel.SetArgument(1, matching_alpha);
+ kernel.SetArgument(2, x_buffer());
+ kernel.SetArgument(3, static_cast<int>(x_offset));
+ kernel.SetArgument(4, static_cast<int>(x_inc));
+ kernel.SetArgument(5, a_buffer());
+ kernel.SetArgument(6, static_cast<int>(a_offset));
+ kernel.SetArgument(7, static_cast<int>(a_ld));
+ kernel.SetArgument(8, static_cast<int>(is_upper));
+ kernel.SetArgument(9, static_cast<int>(is_rowmajor));
+
+ // Launches the kernel
+ auto global_one = CeilDiv(Ceil(n, db_["WGS1"]), db_["WPT"]);
+ auto global_two = CeilDiv(Ceil(n, db_["WGS2"]), db_["WPT"]);
+ auto global = std::vector<size_t>{global_one, global_two};
+ auto local = std::vector<size_t>{db_["WGS1"], db_["WGS2"]};
+ status = RunKernel(kernel, global, local);
+ if (ErrorIn(status)) { return status; }
+
+ // Waits for all kernels to finish
+ queue_.Finish();
+
+ // Succesfully finished the computation
+ return StatusCode::kSuccess;
+ } catch (...) { return StatusCode::kInvalidKernel; }
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xher<float, float>;
+template class Xher<double, double>;
+template class Xher<float2, float>;
+template class Xher<double2, double>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xhpr.cc b/src/routines/level2/xhpr.cc
new file mode 100644
index 00000000..abe00669
--- /dev/null
+++ b/src/routines/level2/xhpr.cc
@@ -0,0 +1,50 @@
+
+// =================================================================================================
+// 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 Xhpr class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xhpr.h"
+
+#include <string>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T, typename U>
+Xhpr<T,U>::Xhpr(Queue &queue, Event &event, const std::string &name):
+ Xher<T,U>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T, typename U>
+StatusCode Xhpr<T,U>::DoHpr(const Layout layout, const Triangle triangle,
+ const size_t n,
+ const U alpha,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const Buffer<T> &ap_buffer, const size_t ap_offset) {
+
+ //
+ return DoHer(layout, triangle, n, alpha,
+ x_buffer, x_offset, x_inc,
+ ap_buffer, ap_offset, n, true);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xhpr<float2, float>;
+template class Xhpr<double2, double>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xspr.cc b/src/routines/level2/xspr.cc
new file mode 100644
index 00000000..5159ad50
--- /dev/null
+++ b/src/routines/level2/xspr.cc
@@ -0,0 +1,50 @@
+
+// =================================================================================================
+// 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 Xspr class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xspr.h"
+
+#include <string>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xspr<T>::Xspr(Queue &queue, Event &event, const std::string &name):
+ Xher<T,T>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xspr<T>::DoSpr(const Layout layout, const Triangle triangle,
+ const size_t n,
+ const T alpha,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const Buffer<T> &ap_buffer, const size_t ap_offset) {
+
+ //
+ return DoHer(layout, triangle, n, alpha,
+ x_buffer, x_offset, x_inc,
+ ap_buffer, ap_offset, n, true);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xspr<float>;
+template class Xspr<double>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level2/xsyr.cc b/src/routines/level2/xsyr.cc
new file mode 100644
index 00000000..755fde0d
--- /dev/null
+++ b/src/routines/level2/xsyr.cc
@@ -0,0 +1,50 @@
+
+// =================================================================================================
+// 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 Xsyr class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level2/xsyr.h"
+
+#include <string>
+
+namespace clblast {
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xsyr<T>::Xsyr(Queue &queue, Event &event, const std::string &name):
+ Xher<T,T>(queue, event, name) {
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xsyr<T>::DoSyr(const Layout layout, const Triangle triangle,
+ const size_t n,
+ const T alpha,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld) {
+
+ //
+ return DoHer(layout, triangle, n, alpha,
+ x_buffer, x_offset, x_inc,
+ a_buffer, a_offset, a_ld);
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xsyr<float>;
+template class Xsyr<double>;
+
+// =================================================================================================
+} // namespace clblast
generated by cgit v1.2.3 (git 2.39.1) at 2024-08-03 18:08:04 +0000