summaryrefslogtreecommitdiff
path: root/src/routines
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2017-08-12 15:59:14 +0200
committerCedric Nugteren <web@cedricnugteren.nl>2017-08-12 15:59:14 +0200
commit0a63621579fbb8cb54d0ded2ba73d47ac4e08006 (patch)
treea784c4b4e3477d8b9764414222c5b8d2a9e8aecb /src/routines
parent844e68853e14e5cf5ba67a8b376bd59d27383290 (diff)
Moved functions from the header to the .cpp file to prevent compiling the same code multiple times
Diffstat (limited to 'src/routines')
-rw-r--r--src/routines/common.cpp75
-rw-r--r--src/routines/common.hpp25
2 files changed, 77 insertions, 23 deletions
diff --git a/src/routines/common.cpp b/src/routines/common.cpp
index c995dc12..5b178e53 100644
--- a/src/routines/common.cpp
+++ b/src/routines/common.cpp
@@ -73,4 +73,79 @@ void RunKernel(Kernel &kernel, Queue &queue, const Device &device,
}
// =================================================================================================
+
+// Sets all elements of a matrix to a constant value
+template <typename T>
+void FillMatrix(Queue &queue, const Device &device,
+ const Program &program, const Databases &,
+ EventPointer event, const std::vector<Event> &waitForEvents,
+ const size_t m, const size_t n, const size_t ld, const size_t offset,
+ const Buffer<T> &dest,
+ const T constant_value) {
+ auto kernel = Kernel(program, "FillMatrix");
+ kernel.SetArgument(0, static_cast<int>(m));
+ kernel.SetArgument(1, static_cast<int>(n));
+ kernel.SetArgument(2, static_cast<int>(ld));
+ kernel.SetArgument(3, static_cast<int>(offset));
+ kernel.SetArgument(4, dest());
+ kernel.SetArgument(5, GetRealArg(constant_value));
+ auto local = std::vector<size_t>{8, 8};
+ auto global = std::vector<size_t>{Ceil(m, 8), Ceil(n, 8)};
+ RunKernel(kernel, queue, device, global, local, event, waitForEvents);
+}
+
+// Compiles the above function
+template void FillMatrix<half>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const size_t, const Buffer<half>&, const half);
+template void FillMatrix<float>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const size_t, const Buffer<float>&, const float);
+template void FillMatrix<double>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const size_t, const Buffer<double>&, const double);
+template void FillMatrix<float2>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const size_t, const Buffer<float2>&, const float2);
+template void FillMatrix<double2>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const size_t, const Buffer<double2>&, const double2);
+
+// Sets all elements of a vector to a constant value
+template <typename T>
+void FillVector(Queue &queue, const Device &device,
+ const Program &program, const Databases &,
+ EventPointer event, const std::vector<Event> &waitForEvents,
+ const size_t n, const size_t inc, const size_t offset,
+ const Buffer<T> &dest,
+ const T constant_value) {
+ auto kernel = Kernel(program, "FillVector");
+ kernel.SetArgument(0, static_cast<int>(n));
+ kernel.SetArgument(1, static_cast<int>(inc));
+ kernel.SetArgument(2, static_cast<int>(offset));
+ kernel.SetArgument(3, dest());
+ kernel.SetArgument(4, GetRealArg(constant_value));
+ auto local = std::vector<size_t>{64};
+ auto global = std::vector<size_t>{Ceil(n, 64)};
+ RunKernel(kernel, queue, device, global, local, event, waitForEvents);
+}
+
+// Compiles the above function
+template void FillVector<half>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const Buffer<half>&, const half);
+template void FillVector<float>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const Buffer<float>&, const float);
+template void FillVector<double>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const Buffer<double>&, const double);
+template void FillVector<float2>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const Buffer<float2>&, const float2);
+template void FillVector<double2>(Queue&, const Device&, const Program&, const Databases&,
+ EventPointer, const std::vector<Event>&, const size_t, const size_t,
+ const size_t, const Buffer<double2>&, const double2);
+
+// =================================================================================================
} // namespace clblast
diff --git a/src/routines/common.hpp b/src/routines/common.hpp
index 28a43da5..84ccd9d2 100644
--- a/src/routines/common.hpp
+++ b/src/routines/common.hpp
@@ -40,18 +40,7 @@ void FillMatrix(Queue &queue, const Device &device,
EventPointer event, const std::vector<Event> &waitForEvents,
const size_t m, const size_t n, const size_t ld, const size_t offset,
const Buffer<T> &dest,
- const T constant_value) {
- auto kernel = Kernel(program, "FillMatrix");
- kernel.SetArgument(0, static_cast<int>(m));
- kernel.SetArgument(1, static_cast<int>(n));
- kernel.SetArgument(2, static_cast<int>(ld));
- kernel.SetArgument(3, static_cast<int>(offset));
- kernel.SetArgument(4, dest());
- kernel.SetArgument(5, GetRealArg(constant_value));
- auto local = std::vector<size_t>{8, 8};
- auto global = std::vector<size_t>{Ceil(m, 8), Ceil(n, 8)};
- RunKernel(kernel, queue, device, global, local, event, waitForEvents);
-}
+ const T constant_value);
// Sets all elements of a vector to a constant value
template <typename T>
@@ -60,17 +49,7 @@ void FillVector(Queue &queue, const Device &device,
EventPointer event, const std::vector<Event> &waitForEvents,
const size_t n, const size_t inc, const size_t offset,
const Buffer<T> &dest,
- const T constant_value) {
- auto kernel = Kernel(program, "FillVector");
- kernel.SetArgument(0, static_cast<int>(n));
- kernel.SetArgument(1, static_cast<int>(inc));
- kernel.SetArgument(2, static_cast<int>(offset));
- kernel.SetArgument(3, dest());
- kernel.SetArgument(4, GetRealArg(constant_value));
- auto local = std::vector<size_t>{64};
- auto global = std::vector<size_t>{Ceil(n, 64)};
- RunKernel(kernel, queue, device, global, local, event, waitForEvents);
-}
+ const T constant_value);
// =================================================================================================