diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2016-03-19 11:09:09 +0100 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2016-03-19 11:09:09 +0100 |
commit | d9356954176eba875138e38b5cf0fb131e1ac925 (patch) | |
tree | 7844472ae9025bf0024c303f9de287910b8f2ecd /src/clblast.cc | |
parent | 918797735da29f102f027b9dc19dccc751950c76 (diff) |
Added __declspec(dllexport) to create a DLL on Windows
Diffstat (limited to 'src/clblast.cc')
-rw-r--r-- | src/clblast.cc | 1485 |
1 files changed, 743 insertions, 742 deletions
diff --git a/src/clblast.cc b/src/clblast.cc index 3695aa02..e7f2477f 100644 --- a/src/clblast.cc +++ b/src/clblast.cc @@ -16,6 +16,7 @@ #include <string> #include "clblast.h" +#include "internal/public_api.h" // BLAS level-1 includes #include "internal/routines/level1/xswap.h" @@ -81,22 +82,22 @@ StatusCode Swap(const size_t n, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Swap<float>(const size_t, - cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Swap<double>(const size_t, - cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Swap<float2>(const size_t, - cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Swap<double2>(const size_t, - cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Swap<float>(const size_t, + cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Swap<double>(const size_t, + cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Swap<float2>(const size_t, + cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Swap<double2>(const size_t, + cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Vector scaling: SSCAL/DSCAL/CSCAL/ZSCAL template <typename T> @@ -113,22 +114,22 @@ StatusCode Scal(const size_t n, alpha, Buffer<T>(x_buffer), x_offset, x_inc); } -template StatusCode Scal<float>(const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Scal<double>(const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Scal<float2>(const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Scal<double2>(const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Scal<float>(const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Scal<double>(const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Scal<float2>(const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Scal<double2>(const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Vector copy: SCOPY/DCOPY/CCOPY/ZCOPY template <typename T> @@ -145,22 +146,22 @@ StatusCode Copy(const size_t n, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Copy<float>(const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Copy<double>(const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Copy<float2>(const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Copy<double2>(const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Copy<float>(const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Copy<double>(const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Copy<float2>(const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Copy<double2>(const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Vector-times-constant plus vector: SAXPY/DAXPY/CAXPY/ZAXPY template <typename T> @@ -179,26 +180,26 @@ StatusCode Axpy(const size_t n, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Axpy<float>(const size_t, - const float, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Axpy<double>(const size_t, - const double, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Axpy<float2>(const size_t, - const float2, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Axpy<double2>(const size_t, - const double2, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Axpy<float>(const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Axpy<double>(const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Axpy<float2>(const size_t, + const float2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Axpy<double2>(const size_t, + const double2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Dot product of two vectors: SDOT/DDOT template <typename T> @@ -217,16 +218,16 @@ StatusCode Dot(const size_t n, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Dot<float>(const size_t, - cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Dot<double>(const size_t, - cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Dot<float>(const size_t, + cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Dot<double>(const size_t, + cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Dot product of two complex vectors: CDOTU/ZDOTU template <typename T> @@ -245,16 +246,16 @@ StatusCode Dotu(const size_t n, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Dotu<float2>(const size_t, - cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Dotu<double2>(const size_t, - cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Dotu<float2>(const size_t, + cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Dotu<double2>(const size_t, + cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Dot product of two complex vectors, one conjugated: CDOTC/ZDOTC template <typename T> @@ -273,16 +274,16 @@ StatusCode Dotc(const size_t n, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Dotc<float2>(const size_t, - cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Dotc<double2>(const size_t, - cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Dotc<float2>(const size_t, + cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Dotc<double2>(const size_t, + cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // ================================================================================================= // BLAS level-2 (matrix-vector) routines @@ -311,38 +312,38 @@ StatusCode Gemv(const Layout layout, const Transpose a_transpose, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Gemv<float>(const Layout, const Transpose, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gemv<double>(const Layout, const Transpose, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gemv<float2>(const Layout, const Transpose, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gemv<double2>(const Layout, const Transpose, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemv<float>(const Layout, const Transpose, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemv<double>(const Layout, const Transpose, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemv<float2>(const Layout, const Transpose, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemv<double2>(const Layout, const Transpose, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // General banded matrix-vector multiplication: SGBMV/DGBMV/CGBMV/ZGBMV template <typename T> @@ -367,38 +368,38 @@ StatusCode Gbmv(const Layout layout, const Transpose a_transpose, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Gbmv<float>(const Layout, const Transpose, - const size_t, const size_t, const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gbmv<double>(const Layout, const Transpose, - const size_t, const size_t, const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gbmv<float2>(const Layout, const Transpose, - const size_t, const size_t, const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gbmv<double2>(const Layout, const Transpose, - const size_t, const size_t, const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gbmv<float>(const Layout, const Transpose, + const size_t, const size_t, const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gbmv<double>(const Layout, const Transpose, + const size_t, const size_t, const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gbmv<float2>(const Layout, const Transpose, + const size_t, const size_t, const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gbmv<double2>(const Layout, const Transpose, + const size_t, const size_t, const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Hermitian matrix-vector multiplication: CHEMV/ZHEMV template <typename T> @@ -423,22 +424,22 @@ StatusCode Hemv(const Layout layout, const Triangle triangle, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Hemv<float2>(const Layout, const Triangle, - const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Hemv<double2>(const Layout, const Triangle, - const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hemv<float2>(const Layout, const Triangle, + const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hemv<double2>(const Layout, const Triangle, + const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Hermitian banded matrix-vector multiplication: CHBMV/ZHBMV template <typename T> @@ -463,22 +464,22 @@ StatusCode Hbmv(const Layout layout, const Triangle triangle, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Hbmv<float2>(const Layout, const Triangle, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Hbmv<double2>(const Layout, const Triangle, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hbmv<float2>(const Layout, const Triangle, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hbmv<double2>(const Layout, const Triangle, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Hermitian packed matrix-vector multiplication: CHPMV/ZHPMV template <typename T> @@ -503,22 +504,22 @@ StatusCode Hpmv(const Layout layout, const Triangle triangle, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Hpmv<float2>(const Layout, const Triangle, - const size_t, - const float2, - const cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Hpmv<double2>(const Layout, const Triangle, - const size_t, - const double2, - const cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hpmv<float2>(const Layout, const Triangle, + const size_t, + const float2, + const cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hpmv<double2>(const Layout, const Triangle, + const size_t, + const double2, + const cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Symmetric matrix-vector multiplication: SSYMV/DSYMV template <typename T> @@ -543,22 +544,22 @@ StatusCode Symv(const Layout layout, const Triangle triangle, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Symv<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Symv<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Symv<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Symv<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Symmetric banded matrix-vector multiplication: SSBMV/DSBMV template <typename T> @@ -583,22 +584,22 @@ StatusCode Sbmv(const Layout layout, const Triangle triangle, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Sbmv<float>(const Layout, const Triangle, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Sbmv<double>(const Layout, const Triangle, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Sbmv<float>(const Layout, const Triangle, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Sbmv<double>(const Layout, const Triangle, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Symmetric packed matrix-vector multiplication: SSPMV/DSPMV template <typename T> @@ -623,22 +624,22 @@ StatusCode Spmv(const Layout layout, const Triangle triangle, beta, Buffer<T>(y_buffer), y_offset, y_inc); } -template StatusCode Spmv<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Spmv<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Spmv<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Spmv<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Triangular matrix-vector multiplication: STRMV/DTRMV/CTRMV/ZTRMV template <typename T> @@ -657,26 +658,26 @@ StatusCode Trmv(const Layout layout, const Triangle triangle, const Transpose a_ Buffer<T>(a_buffer), a_offset, a_ld, Buffer<T>(x_buffer), x_offset, x_inc); } -template StatusCode Trmv<float>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trmv<double>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trmv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trmv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmv<float>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmv<double>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Triangular banded matrix-vector multiplication: STBMV/DTBMV/CTBMV/ZTBMV template <typename T> @@ -695,26 +696,26 @@ StatusCode Tbmv(const Layout layout, const Triangle triangle, const Transpose a_ Buffer<T>(a_buffer), a_offset, a_ld, Buffer<T>(x_buffer), x_offset, x_inc); } -template StatusCode Tbmv<float>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tbmv<double>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tbmv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tbmv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbmv<float>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbmv<double>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbmv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbmv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Triangular packed matrix-vector multiplication: STPMV/DTPMV/CTPMV/ZTPMV template <typename T> @@ -733,26 +734,26 @@ StatusCode Tpmv(const Layout layout, const Triangle triangle, const Transpose a_ Buffer<T>(ap_buffer), ap_offset, Buffer<T>(x_buffer), x_offset, x_inc); } -template StatusCode Tpmv<float>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tpmv<double>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tpmv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tpmv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpmv<float>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpmv<double>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpmv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpmv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Solves a triangular system of equations: STRSV/DTRSV/CTRSV/ZTRSV template <typename T> @@ -763,26 +764,26 @@ StatusCode Trsv(const Layout, const Triangle, const Transpose, const Diagonal, cl_command_queue*, cl_event*) { return StatusCode::kNotImplemented; } -template StatusCode Trsv<float>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trsv<double>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trsv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trsv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsv<float>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsv<double>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Solves a banded triangular system of equations: STBSV/DTBSV/CTBSV/ZTBSV template <typename T> @@ -793,26 +794,26 @@ StatusCode Tbsv(const Layout, const Triangle, const Transpose, const Diagonal, cl_command_queue*, cl_event*) { return StatusCode::kNotImplemented; } -template StatusCode Tbsv<float>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tbsv<double>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tbsv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tbsv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbsv<float>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbsv<double>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbsv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tbsv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Solves a packed triangular system of equations: STPSV/DTPSV/CTPSV/ZTPSV template <typename T> @@ -823,26 +824,26 @@ StatusCode Tpsv(const Layout, const Triangle, const Transpose, const Diagonal, cl_command_queue*, cl_event*) { return StatusCode::kNotImplemented; } -template StatusCode Tpsv<float>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tpsv<double>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tpsv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Tpsv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, - const size_t, - const cl_mem, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpsv<float>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpsv<double>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpsv<float2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Tpsv<double2>(const Layout, const Triangle, const Transpose, const Diagonal, + const size_t, + const cl_mem, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // General rank-1 matrix update: SGER/DGER template <typename T> @@ -865,20 +866,20 @@ StatusCode Ger(const Layout layout, Buffer<T>(y_buffer), y_offset, y_inc, Buffer<T>(a_buffer), a_offset, a_ld); } -template StatusCode Ger<float>(const Layout, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Ger<double>(const Layout, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Ger<float>(const Layout, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Ger<double>(const Layout, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // General rank-1 complex matrix update: CGERU/ZGERU template <typename T> @@ -901,20 +902,20 @@ StatusCode Geru(const Layout layout, Buffer<T>(y_buffer), y_offset, y_inc, Buffer<T>(a_buffer), a_offset, a_ld); } -template StatusCode Geru<float2>(const Layout, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Geru<double2>(const Layout, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Geru<float2>(const Layout, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Geru<double2>(const Layout, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // General rank-1 complex conjugated matrix update: CGERC/ZGERC template <typename T> @@ -937,20 +938,20 @@ StatusCode Gerc(const Layout layout, Buffer<T>(y_buffer), y_offset, y_inc, Buffer<T>(a_buffer), a_offset, a_ld); } -template StatusCode Gerc<float2>(const Layout, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gerc<double2>(const Layout, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gerc<float2>(const Layout, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gerc<double2>(const Layout, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Hermitian rank-1 matrix update: CHER/ZHER template <typename T> @@ -971,18 +972,18 @@ StatusCode Her(const Layout layout, const Triangle triangle, Buffer<std::complex<T>>(x_buffer), x_offset, x_inc, Buffer<std::complex<T>>(a_buffer), a_offset, a_ld); } -template StatusCode Her<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Her<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Her<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Her<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Hermitian packed rank-1 matrix update: CHPR/ZHPR template <typename T> @@ -1003,18 +1004,18 @@ StatusCode Hpr(const Layout layout, const Triangle triangle, Buffer<std::complex<T>>(x_buffer), x_offset, x_inc, Buffer<std::complex<T>>(ap_buffer), ap_offset); } -template StatusCode Hpr<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Hpr<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hpr<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hpr<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); // Hermitian rank-2 matrix update: CHER2/ZHER2 template <typename T> @@ -1037,20 +1038,20 @@ StatusCode Her2(const Layout layout, const Triangle triangle, Buffer<T>(y_buffer), y_offset, y_inc, Buffer<T>(a_buffer), a_offset, a_ld); } -template StatusCode Her2<float2>(const Layout, const Triangle, - const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Her2<double2>(const Layout, const Triangle, - const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Her2<float2>(const Layout, const Triangle, + const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Her2<double2>(const Layout, const Triangle, + const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Hermitian packed rank-2 matrix update: CHPR2/ZHPR2 template <typename T> @@ -1073,20 +1074,20 @@ StatusCode Hpr2(const Layout layout, const Triangle triangle, Buffer<T>(y_buffer), y_offset, y_inc, Buffer<T>(ap_buffer), ap_offset); } -template StatusCode Hpr2<float2>(const Layout, const Triangle, - const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Hpr2<double2>(const Layout, const Triangle, - const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hpr2<float2>(const Layout, const Triangle, + const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hpr2<double2>(const Layout, const Triangle, + const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); // Symmetric rank-1 matrix update: SSYR/DSYR template <typename T> @@ -1107,18 +1108,18 @@ StatusCode Syr(const Layout layout, const Triangle triangle, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(a_buffer), a_offset, a_ld); } -template StatusCode Syr<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syr<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Symmetric packed rank-1 matrix update: SSPR/DSPR template <typename T> @@ -1139,18 +1140,18 @@ StatusCode Spr(const Layout layout, const Triangle triangle, Buffer<T>(x_buffer), x_offset, x_inc, Buffer<T>(ap_buffer), ap_offset); } -template StatusCode Spr<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Spr<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Spr<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Spr<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); // Symmetric rank-2 matrix update: SSYR2/DSYR2 template <typename T> @@ -1173,20 +1174,20 @@ StatusCode Syr2(const Layout layout, const Triangle triangle, Buffer<T>(y_buffer), y_offset, y_inc, Buffer<T>(a_buffer), a_offset, a_ld); } -template StatusCode Syr2<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syr2<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr2<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr2<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Symmetric packed rank-2 matrix update: SSPR2/DSPR2 template <typename T> @@ -1209,20 +1210,20 @@ StatusCode Spr2(const Layout layout, const Triangle triangle, Buffer<T>(y_buffer), y_offset, y_inc, Buffer<T>(ap_buffer), ap_offset); } -template StatusCode Spr2<float>(const Layout, const Triangle, - const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Spr2<double>(const Layout, const Triangle, - const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Spr2<float>(const Layout, const Triangle, + const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Spr2<double>(const Layout, const Triangle, + const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, + cl_command_queue*, cl_event*); // ================================================================================================= // BLAS level-3 (matrix-matrix) routines @@ -1251,38 +1252,38 @@ StatusCode Gemm(const Layout layout, const Transpose a_transpose, const Transpos beta, Buffer<T>(c_buffer), c_offset, c_ld); } -template StatusCode Gemm<float>(const Layout, const Transpose, const Transpose, - const size_t, const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gemm<double>(const Layout, const Transpose, const Transpose, - const size_t, const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gemm<float2>(const Layout, const Transpose, const Transpose, - const size_t, const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Gemm<double2>(const Layout, const Transpose, const Transpose, - const size_t, const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemm<float>(const Layout, const Transpose, const Transpose, + const size_t, const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemm<double>(const Layout, const Transpose, const Transpose, + const size_t, const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemm<float2>(const Layout, const Transpose, const Transpose, + const size_t, const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Gemm<double2>(const Layout, const Transpose, const Transpose, + const size_t, const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Symmetric matrix-matrix multiplication: SSYMM/DSYMM/CSYMM/ZSYMM template <typename T> @@ -1307,38 +1308,38 @@ StatusCode Symm(const Layout layout, const Side side, const Triangle triangle, beta, Buffer<T>(c_buffer), c_offset, c_ld); } -template StatusCode Symm<float>(const Layout, const Side, const Triangle, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Symm<double>(const Layout, const Side, const Triangle, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Symm<float2>(const Layout, const Side, const Triangle, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Symm<double2>(const Layout, const Side, const Triangle, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Symm<float>(const Layout, const Side, const Triangle, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Symm<double>(const Layout, const Side, const Triangle, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Symm<float2>(const Layout, const Side, const Triangle, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Symm<double2>(const Layout, const Side, const Triangle, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Hermitian matrix-matrix multiplication: CHEMM/ZHEMM template <typename T> @@ -1363,22 +1364,22 @@ StatusCode Hemm(const Layout layout, const Side side, const Triangle triangle, beta, Buffer<T>(c_buffer), c_offset, c_ld); } -template StatusCode Hemm<float2>(const Layout, const Side, const Triangle, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Hemm<double2>(const Layout, const Side, const Triangle, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hemm<float2>(const Layout, const Side, const Triangle, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Hemm<double2>(const Layout, const Side, const Triangle, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Rank-K update of a symmetric matrix: SSYRK/DSYRK/CSYRK/ZSYRK template <typename T> @@ -1401,34 +1402,34 @@ StatusCode Syrk(const Layout layout, const Triangle triangle, const Transpose a_ beta, Buffer<T>(c_buffer), c_offset, c_ld); } -template StatusCode Syrk<float>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syrk<double>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syrk<float2>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syrk<double2>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syrk<float>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syrk<double>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syrk<float2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syrk<double2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Rank-K update of a hermitian matrix: CHERK/ZHERK template <typename T> @@ -1451,20 +1452,20 @@ StatusCode Herk(const Layout layout, const Triangle triangle, const Transpose a_ beta, Buffer<std::complex<T>>(c_buffer), c_offset, c_ld); } -template StatusCode Herk<float>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Herk<double>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Herk<float>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Herk<double>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Rank-2K update of a symmetric matrix: SSYR2K/DSYR2K/CSYR2K/ZSYR2K template <typename T> @@ -1489,38 +1490,38 @@ StatusCode Syr2k(const Layout layout, const Triangle triangle, const Transpose a beta, Buffer<T>(c_buffer), c_offset, c_ld); } -template StatusCode Syr2k<float>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syr2k<double>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syr2k<float2>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Syr2k<double2>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double2, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr2k<float>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr2k<double>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr2k<float2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Syr2k<double2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Rank-2K update of a hermitian matrix: CHER2K/ZHER2K template <typename T, typename U> @@ -1545,22 +1546,22 @@ StatusCode Her2k(const Layout layout, const Triangle triangle, const Transpose a beta, Buffer<T>(c_buffer), c_offset, c_ld); } -template StatusCode Her2k<float2,float>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const float, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Her2k<double2,double>(const Layout, const Triangle, const Transpose, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - const cl_mem, const size_t, const size_t, - const double, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Her2k<float2,float>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Her2k<double2,double>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, + const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Triangular matrix-matrix multiplication: STRMM/DTRMM/CTRMM/ZTRMM template <typename T> @@ -1581,30 +1582,30 @@ StatusCode Trmm(const Layout layout, const Side side, const Triangle triangle, c Buffer<T>(a_buffer), a_offset, a_ld, Buffer<T>(b_buffer), b_offset, b_ld); } -template StatusCode Trmm<float>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trmm<double>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trmm<float2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trmm<double2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmm<float>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmm<double>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmm<float2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trmm<double2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // Solves a triangular system of equations: STRSM/DTRSM/CTRSM/ZTRSM template <typename T> @@ -1616,30 +1617,30 @@ StatusCode Trsm(const Layout, const Side, const Triangle, const Transpose, const cl_command_queue*, cl_event*) { return StatusCode::kNotImplemented; } -template StatusCode Trsm<float>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const float, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trsm<double>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const double, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trsm<float2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const float2, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); -template StatusCode Trsm<double2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, - const size_t, const size_t, - const double2, - const cl_mem, const size_t, const size_t, - cl_mem, const size_t, const size_t, - cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsm<float>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsm<double>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsm<float2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Trsm<double2>(const Layout, const Side, const Triangle, const Transpose, const Diagonal, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); // ================================================================================================= } // namespace clblast |