summaryrefslogtreecommitdiff
path: root/src/routines/level2/xgemv.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/routines/level2/xgemv.cpp')
-rw-r--r--src/routines/level2/xgemv.cpp127
1 files changed, 58 insertions, 69 deletions
diff --git a/src/routines/level2/xgemv.cpp b/src/routines/level2/xgemv.cpp
index 4e32ba41..7b4c2e8f 100644
--- a/src/routines/level2/xgemv.cpp
+++ b/src/routines/level2/xgemv.cpp
@@ -22,52 +22,51 @@ namespace clblast {
// Constructor: forwards to base class constructor
template <typename T>
Xgemv<T>::Xgemv(Queue &queue, EventPointer event, const std::string &name):
- Routine(queue, event, name, {"Pad", "Xgemv", "XgemvFast", "XgemvFastRot"}, PrecisionValue<T>()) {
- source_string_ =
+ Routine(queue, event, name, {"Pad", "Xgemv", "XgemvFast", "XgemvFastRot"}, PrecisionValue<T>(), {}, {
#include "../../kernels/level2/xgemv.opencl"
#include "../../kernels/level2/xgemv_fast.opencl"
- ;
+ }) {
}
// =================================================================================================
// The main routine
template <typename T>
-StatusCode Xgemv<T>::DoGemv(const Layout layout, const Transpose a_transpose,
- const size_t m, const size_t n,
- const T alpha,
- const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
- const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
- const T beta,
- const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc) {
+void Xgemv<T>::DoGemv(const Layout layout, const Transpose a_transpose,
+ const size_t m, const size_t n,
+ const T alpha,
+ const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const T beta,
+ const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc) {
// Performs the matrix-vector multiplication
- return MatVec(layout, a_transpose,
- m, n, alpha,
- a_buffer, a_offset, a_ld,
- x_buffer, x_offset, x_inc, beta,
- y_buffer, y_offset, y_inc,
- true, true,
- 0, false, 0, 0); // N/A for this routine
+ MatVec(layout, a_transpose,
+ m, n, alpha,
+ a_buffer, a_offset, a_ld,
+ x_buffer, x_offset, x_inc, beta,
+ y_buffer, y_offset, y_inc,
+ true, true,
+ 0, false, 0, 0); // N/A for this routine
}
// =================================================================================================
// The generic implementation, also suited for other (non general) matrix-vector multiplications
template <typename T>
-StatusCode Xgemv<T>::MatVec(const Layout layout, const Transpose a_transpose,
- const size_t m, const size_t n,
- const T alpha,
- const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
- const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
- const T beta,
- const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc,
- bool fast_kernel, bool fast_kernel_rot,
- const size_t parameter, const bool packed,
- const size_t kl, const size_t ku) {
+void Xgemv<T>::MatVec(const Layout layout, const Transpose a_transpose,
+ const size_t m, const size_t n,
+ const T alpha,
+ const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
+ const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc,
+ const T beta,
+ const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc,
+ bool fast_kernel, bool fast_kernel_rot,
+ const size_t parameter, const bool packed,
+ const size_t kl, const size_t ku) {
// Makes sure all dimensions are larger than zero
- if (m == 0 || n == 0) { return StatusCode::kInvalidDimension; }
+ if (m == 0 || n == 0) { throw BLASError(StatusCode::kInvalidDimension); }
// Computes whether or not the matrix has an alternative layout (row or column-major).
auto a_altlayout = (layout == Layout::kRowMajor);
@@ -91,14 +90,10 @@ StatusCode Xgemv<T>::MatVec(const Layout layout, const Transpose a_transpose,
auto a_conjugate = (a_transpose == Transpose::kConjugate);
// Tests the matrix and the vectors for validity
- auto status = StatusCode::kSuccess;
- if (packed) { status = TestMatrixAP(n, a_buffer, a_offset); }
- else { status = TestMatrixA(a_one, a_two, a_buffer, a_offset, a_ld); }
- if (ErrorIn(status)) { return status; }
- status = TestVectorX(n_real, x_buffer, x_offset, x_inc);
- if (ErrorIn(status)) { return status; }
- status = TestVectorY(m_real, y_buffer, y_offset, y_inc);
- if (ErrorIn(status)) { return status; }
+ if (packed) { TestMatrixAP(n, a_buffer, a_offset); }
+ else { TestMatrixA(a_one, a_two, a_buffer, a_offset, a_ld); }
+ TestVectorX(n_real, x_buffer, x_offset, x_inc);
+ TestVectorY(m_real, y_buffer, y_offset, y_inc);
// Determines whether or not the fast-version can be used
fast_kernel = fast_kernel && (a_offset == 0) && (a_rotated == 0) && (a_conjugate == 0) &&
@@ -127,39 +122,33 @@ StatusCode Xgemv<T>::MatVec(const Layout layout, const Transpose a_transpose,
}
// Retrieves the Xgemv kernel from the compiled binary
- try {
- const auto program = GetProgramFromCache(context_, PrecisionValue<T>(), routine_name_);
- auto kernel = Kernel(program, kernel_name);
-
- // Sets the kernel arguments
- kernel.SetArgument(0, static_cast<int>(m_real));
- kernel.SetArgument(1, static_cast<int>(n_real));
- kernel.SetArgument(2, GetRealArg(alpha));
- kernel.SetArgument(3, GetRealArg(beta));
- kernel.SetArgument(4, static_cast<int>(a_rotated));
- kernel.SetArgument(5, a_buffer());
- kernel.SetArgument(6, static_cast<int>(a_offset));
- kernel.SetArgument(7, static_cast<int>(a_ld));
- kernel.SetArgument(8, x_buffer());
- kernel.SetArgument(9, static_cast<int>(x_offset));
- kernel.SetArgument(10, static_cast<int>(x_inc));
- kernel.SetArgument(11, y_buffer());
- kernel.SetArgument(12, static_cast<int>(y_offset));
- kernel.SetArgument(13, static_cast<int>(y_inc));
- kernel.SetArgument(14, static_cast<int>(a_conjugate));
- kernel.SetArgument(15, static_cast<int>(parameter)); // extra parameter used for symm/herm
- kernel.SetArgument(16, static_cast<int>(kl)); // only used for banded matrices
- kernel.SetArgument(17, static_cast<int>(ku)); // only used for banded matrices
-
- // Launches the kernel
- auto global = std::vector<size_t>{global_size};
- auto local = std::vector<size_t>{local_size};
- status = RunKernel(kernel, queue_, device_, global, local, event_);
- if (ErrorIn(status)) { return status; }
-
- // Succesfully finished the computation
- return StatusCode::kSuccess;
- } catch (...) { return StatusCode::kInvalidKernel; }
+ const auto program = GetProgramFromCache(context_, PrecisionValue<T>(), routine_name_);
+ auto kernel = Kernel(program, kernel_name);
+
+ // Sets the kernel arguments
+ kernel.SetArgument(0, static_cast<int>(m_real));
+ kernel.SetArgument(1, static_cast<int>(n_real));
+ kernel.SetArgument(2, GetRealArg(alpha));
+ kernel.SetArgument(3, GetRealArg(beta));
+ kernel.SetArgument(4, static_cast<int>(a_rotated));
+ kernel.SetArgument(5, a_buffer());
+ kernel.SetArgument(6, static_cast<int>(a_offset));
+ kernel.SetArgument(7, static_cast<int>(a_ld));
+ kernel.SetArgument(8, x_buffer());
+ kernel.SetArgument(9, static_cast<int>(x_offset));
+ kernel.SetArgument(10, static_cast<int>(x_inc));
+ kernel.SetArgument(11, y_buffer());
+ kernel.SetArgument(12, static_cast<int>(y_offset));
+ kernel.SetArgument(13, static_cast<int>(y_inc));
+ kernel.SetArgument(14, static_cast<int>(a_conjugate));
+ kernel.SetArgument(15, static_cast<int>(parameter)); // extra parameter used for symm/herm
+ kernel.SetArgument(16, static_cast<int>(kl)); // only used for banded matrices
+ kernel.SetArgument(17, static_cast<int>(ku)); // only used for banded matrices
+
+ // Launches the kernel
+ auto global = std::vector<size_t>{global_size};
+ auto local = std::vector<size_t>{local_size};
+ RunKernel(kernel, queue_, device_, global, local, event_);
}
// =================================================================================================
generated by cgit v1.2.3 (git 2.39.1) at 2024-08-03 05:53:23 +0000