From 8be99de82d2ff0634c1289d9b4d1785364a68a44 Mon Sep 17 00:00:00 2001
From: cnugteren <web@cedricnugteren.nl>
Date: Thu, 14 Apr 2016 19:58:26 -0600
Subject: Added support for the SASUM/DASUM/ScASUM/DzASUM routines

---
 src/routines/level1/xasum.cc | 109 +++++++++++++++++++++++++++++++++++++++++++
 src/routines/level1/xnrm2.cc |   1 +
 2 files changed, 110 insertions(+)
 create mode 100644 src/routines/level1/xasum.cc

(limited to 'src/routines')

diff --git a/src/routines/level1/xasum.cc b/src/routines/level1/xasum.cc
new file mode 100644
index 00000000..5799e25a
--- /dev/null
+++ b/src/routines/level1/xasum.cc
@@ -0,0 +1,109 @@
+
+// =================================================================================================
+// 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 Xasum class (see the header for information about the class).
+//
+// =================================================================================================
+
+#include "internal/routines/level1/xasum.h"
+
+#include <string>
+#include <vector>
+
+namespace clblast {
+// =================================================================================================
+
+// Specific implementations to get the memory-type based on a template argument
+template <> const Precision Xasum<float>::precision_ = Precision::kSingle;
+template <> const Precision Xasum<double>::precision_ = Precision::kDouble;
+template <> const Precision Xasum<float2>::precision_ = Precision::kComplexSingle;
+template <> const Precision Xasum<double2>::precision_ = Precision::kComplexDouble;
+
+// =================================================================================================
+
+// Constructor: forwards to base class constructor
+template <typename T>
+Xasum<T>::Xasum(Queue &queue, EventPointer event, const std::string &name):
+    Routine<T>(queue, event, name, {"Xdot"}, precision_) {
+  source_string_ =
+    #include "../../kernels/level1/xasum.opencl"
+  ;
+}
+
+// =================================================================================================
+
+// The main routine
+template <typename T>
+StatusCode Xasum<T>::DoAsum(const size_t n,
+                            const Buffer<T> &asum_buffer, const size_t asum_offset,
+                            const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc) {
+
+  // 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 = TestVectorDot(1, asum_buffer, asum_offset, sizeof(T));
+  if (ErrorIn(status)) { return status; }
+
+  // Retrieves the Xasum kernels from the compiled binary
+  try {
+    auto& program = GetProgramFromCache();
+    auto kernel1 = Kernel(program, "Xasum");
+    auto kernel2 = Kernel(program, "XasumEpilogue");
+
+    // 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, temp_buffer());
+
+    // Event waiting list
+    auto eventWaitList = std::vector<Event>();
+
+    // Launches the main kernel
+    auto global1 = std::vector<size_t>{db_["WGS1"]*temp_size};
+    auto local1 = std::vector<size_t>{db_["WGS1"]};
+    auto kernelEvent = Event();
+    status = RunKernel(kernel1, global1, local1, kernelEvent.pointer());
+    if (ErrorIn(status)) { return status; }
+    eventWaitList.push_back(kernelEvent);
+
+    // Sets the arguments for the epilogue kernel
+    kernel2.SetArgument(0, temp_buffer());
+    kernel2.SetArgument(1, asum_buffer());
+    kernel2.SetArgument(2, static_cast<int>(asum_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, event_, eventWaitList);
+    if (ErrorIn(status)) { return status; }
+
+    // Succesfully finished the computation
+    return StatusCode::kSuccess;
+  } catch (...) { return StatusCode::kInvalidKernel; }
+}
+
+// =================================================================================================
+
+// Compiles the templated class
+template class Xasum<float>;
+template class Xasum<double>;
+template class Xasum<float2>;
+template class Xasum<double2>;
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/routines/level1/xnrm2.cc b/src/routines/level1/xnrm2.cc
index 04e4137c..ceabe586 100644
--- a/src/routines/level1/xnrm2.cc
+++ b/src/routines/level1/xnrm2.cc
@@ -69,6 +69,7 @@ StatusCode Xnrm2<T>::DoNrm2(const size_t n,
     kernel1.SetArgument(2, static_cast<int>(x_offset));
     kernel1.SetArgument(3, static_cast<int>(x_inc));
     kernel1.SetArgument(4, temp_buffer());
+
     // Event waiting list
     auto eventWaitList = std::vector<Event>();
 
-- 
cgit v1.2.3