From a2e726d3bd4294f1eae1735f6ba23105dccc6b10 Mon Sep 17 00:00:00 2001
From: CNugteren <web@cedricnugteren.nl>
Date: Mon, 14 Sep 2015 16:57:00 +0200
Subject: Added xDOT/xDOTU/xDOTC dot-product routines

---
 src/routines/level1/xdot.cc | 115 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 115 insertions(+)
 create mode 100644 src/routines/level1/xdot.cc

(limited to 'src/routines/level1/xdot.cc')

diff --git a/src/routines/level1/xdot.cc b/src/routines/level1/xdot.cc
new file mode 100644
index 00000000..a0c1e756
--- /dev/null
+++ b/src/routines/level1/xdot.cc
@@ -0,0 +1,115 @@
+
+// =================================================================================================
+// 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 Xdot class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level1/xdot.h"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Specific implementations to get the memory-type based on a template argument
+template <> const Precision Xdot<float>::precision_ = Precision::kSingle;
+template <> const Precision Xdot<double>::precision_ = Precision::kDouble;
+template <> const Precision Xdot<float2>::precision_ = Precision::kComplexSingle;
+template <> const Precision Xdot<double2>::precision_ = Precision::kComplexDouble;
+
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xdot<T>::Xdot(Queue &queue, Event &event, const std::string &name):
+    Routine<T>(queue, event, name, {"Xdot"}, precision_) {
+  source_string_ =
+    #include "../../kernels/level1/xdot.opencl"
+  ;
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xdot<T>::DoDot(const size_t n,
+                          const Buffer<T> &dot_buffer, const size_t dot_offset,
+                          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 bool do_conjugate) {
+
+  // Makes sure all dimensions are larger than zero
+  if (n == 0) { return StatusCode::kInvalidDimension; }
+
+  // Tests the vectors for validity
+  auto status = TestVectorX(n, 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; }
+  status = TestVectorDot(1, dot_buffer, dot_offset, sizeof(T));
+  if (ErrorIn(status)) { return status; }
+
+  // Retrieves the Xdot kernels from the compiled binary
+  try {
+    auto& program = GetProgramFromCache();
+    auto kernel1 = Kernel(program, "Xdot");
+    auto kernel2 = Kernel(program, "XdotEpilogue");
+
+    // Creates the buffer for intermediate values
+    auto temp_size = 2*db_["WGS2"];
+    auto temp_buffer = Buffer<T>(context_, temp_size);
+
+    // Sets the kernel arguments
+    kernel1.SetArgument(0, static_cast<int>(n));
+    kernel1.SetArgument(1, x_buffer());
+    kernel1.SetArgument(2, static_cast<int>(x_offset));
+    kernel1.SetArgument(3, static_cast<int>(x_inc));
+    kernel1.SetArgument(4, y_buffer());
+    kernel1.SetArgument(5, static_cast<int>(y_offset));
+    kernel1.SetArgument(6, static_cast<int>(y_inc));
+    kernel1.SetArgument(7, temp_buffer());
+    kernel1.SetArgument(8, static_cast<int>(do_conjugate));
+
+    // Launches the main kernel
+    auto global1 = std::vector<size_t>{db_["WGS1"]*temp_size};
+    auto local1 = std::vector<size_t>{db_["WGS1"]};
+    status = RunKernel(kernel1, global1, local1);
+    if (ErrorIn(status)) { return status; }
+
+    // Sets the arguments for the epilogue kernel
+    kernel2.SetArgument(0, temp_buffer());
+    kernel2.SetArgument(1, dot_buffer());
+    kernel2.SetArgument(2, static_cast<int>(dot_offset));
+
+    // Launches the epilogue kernel
+    auto global2 = std::vector<size_t>{db_["WGS2"]};
+    auto local2 = std::vector<size_t>{db_["WGS2"]};
+    status = RunKernel(kernel2, global2, local2);
+    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 Xdot<float>;
+template class Xdot<double>;
+template class Xdot<float2>;
+template class Xdot<double2>;
+
+// =================================================================================================
+} // namespace clblast
-- 
cgit v1.2.3