diff options
Diffstat (limited to 'include/internal/routines/level3')
-rw-r--r-- | include/internal/routines/level3/xgemm.h | 46 | ||||
-rw-r--r-- | include/internal/routines/level3/xhemm.h | 58 | ||||
-rw-r--r-- | include/internal/routines/level3/xher2k.h | 48 | ||||
-rw-r--r-- | include/internal/routines/level3/xherk.h | 47 | ||||
-rw-r--r-- | include/internal/routines/level3/xsymm.h | 60 | ||||
-rw-r--r-- | include/internal/routines/level3/xsyr2k.h | 48 | ||||
-rw-r--r-- | include/internal/routines/level3/xsyrk.h | 49 | ||||
-rw-r--r-- | include/internal/routines/level3/xtrmm.h | 58 |
8 files changed, 414 insertions, 0 deletions
diff --git a/include/internal/routines/level3/xgemm.h b/include/internal/routines/level3/xgemm.h new file mode 100644 index 00000000..7ad4fcfb --- /dev/null +++ b/include/internal/routines/level3/xgemm.h @@ -0,0 +1,46 @@ + +// ================================================================================================= +// 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 Xgemm routine. The precision is implemented using a template argument. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XGEMM_H_ +#define CLBLAST_ROUTINES_XGEMM_H_ + +#include "internal/routine.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xgemm: public Routine { + public: + Xgemm(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoGemm(const Layout layout, const Transpose a_transpose, const Transpose b_transpose, + const size_t m, const size_t n, const size_t k, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer &b_buffer, const size_t b_offset, const size_t b_ld, + const T beta, + const Buffer &c_buffer, const size_t c_offset, const size_t c_ld); + + private: + // Static variable to get the precision + const static Precision precision_; +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XGEMM_H_ +#endif diff --git a/include/internal/routines/level3/xhemm.h b/include/internal/routines/level3/xhemm.h new file mode 100644 index 00000000..6cc9d9ec --- /dev/null +++ b/include/internal/routines/level3/xhemm.h @@ -0,0 +1,58 @@ + +// ================================================================================================= +// 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 Xhemm routine. It is based on the generalized matrix multiplication +// routine (Xgemm). The implementation is very similar to the Xsymm routine. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XHEMM_H_ +#define CLBLAST_ROUTINES_XHEMM_H_ + +#include "internal/routines/level3/xgemm.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xhemm: public Xgemm<T> { + public: + + // Uses several variables from the Routine class + using Routine::db_; + using Routine::context_; + + // Uses several helper functions from the Routine class + using Routine::RunKernel; + using Routine::ErrorIn; + using Routine::TestMatrixA; + using Routine::GetProgramFromCache; + + // Uses the regular Xgemm routine + using Xgemm<T>::DoGemm; + + // Constructor + Xhemm(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoHemm(const Layout layout, const Side side, const Triangle triangle, + const size_t m, const size_t n, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer &b_buffer, const size_t b_offset, const size_t b_ld, + const T beta, + const Buffer &c_buffer, const size_t c_offset, const size_t c_ld); +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XHEMM_H_ +#endif diff --git a/include/internal/routines/level3/xher2k.h b/include/internal/routines/level3/xher2k.h new file mode 100644 index 00000000..1836a812 --- /dev/null +++ b/include/internal/routines/level3/xher2k.h @@ -0,0 +1,48 @@ + +// ================================================================================================= +// 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 Xher2k routine. The precision is implemented using the template argument +// 'T', whereas the alpha/beta arguments are of type 'U'. The implementation is very similar to the +// Xsyr2k routine. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XHER2K_H_ +#define CLBLAST_ROUTINES_XHER2K_H_ + +#include "internal/routine.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T, typename U> +class Xher2k: public Routine { + public: + Xher2k(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoHer2k(const Layout layout, const Triangle triangle, const Transpose ab_transpose, + const size_t n, const size_t k, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer &b_buffer, const size_t b_offset, const size_t b_ld, + const U beta, + const Buffer &c_buffer, const size_t c_offset, const size_t c_ld); + + private: + // Static variable to get the precision + const static Precision precision_; +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XHER2K_H_ +#endif diff --git a/include/internal/routines/level3/xherk.h b/include/internal/routines/level3/xherk.h new file mode 100644 index 00000000..9b361254 --- /dev/null +++ b/include/internal/routines/level3/xherk.h @@ -0,0 +1,47 @@ + +// ================================================================================================= +// 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 Xherk routine. The precision is implemented using the template argument +// 'T', whereas the alpha/beta arguments are of type 'U'. The implementation is very similar to the +// Xsyrk routine. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XHERK_H_ +#define CLBLAST_ROUTINES_XHERK_H_ + +#include "internal/routine.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T, typename U> +class Xherk: public Routine { + public: + Xherk(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoHerk(const Layout layout, const Triangle triangle, const Transpose a_transpose, + const size_t n, const size_t k, + const U alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const U beta, + const Buffer &c_buffer, const size_t c_offset, const size_t c_ld); + + private: + // Static variable to get the precision + const static Precision precision_; +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XHERK_H_ +#endif diff --git a/include/internal/routines/level3/xsymm.h b/include/internal/routines/level3/xsymm.h new file mode 100644 index 00000000..2028ceea --- /dev/null +++ b/include/internal/routines/level3/xsymm.h @@ -0,0 +1,60 @@ + +// ================================================================================================= +// 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 Xsymm routine. It is based on the generalized matrix multiplication +// routine (Xgemm). The Xsymm class inherits from the templated class Xgemm, allowing it to call the +// "DoGemm" function directly. The "DoSymm" function first preprocesses the symmetric matrix by +// transforming it into a general matrix, and then calls the regular GEMM code. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XSYMM_H_ +#define CLBLAST_ROUTINES_XSYMM_H_ + +#include "internal/routines/level3/xgemm.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xsymm: public Xgemm<T> { + public: + + // Uses several variables from the Routine class + using Routine::db_; + using Routine::context_; + + // Uses several helper functions from the Routine class + using Routine::RunKernel; + using Routine::ErrorIn; + using Routine::TestMatrixA; + using Routine::GetProgramFromCache; + + // Uses the regular Xgemm routine + using Xgemm<T>::DoGemm; + + // Constructor + Xsymm(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoSymm(const Layout layout, const Side side, const Triangle triangle, + const size_t m, const size_t n, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer &b_buffer, const size_t b_offset, const size_t b_ld, + const T beta, + const Buffer &c_buffer, const size_t c_offset, const size_t c_ld); +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XSYMM_H_ +#endif diff --git a/include/internal/routines/level3/xsyr2k.h b/include/internal/routines/level3/xsyr2k.h new file mode 100644 index 00000000..6259313c --- /dev/null +++ b/include/internal/routines/level3/xsyr2k.h @@ -0,0 +1,48 @@ + +// ================================================================================================= +// 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 Xsyr2k routine. The precision is implemented using a template argument. +// The implementation is very similar to Xsyrk (see header for details), except for the fact that +// the main XgemmUpper/XgemmLower kernel is called twice: C = AB^T + C and C = BA^T + C. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XSYR2K_H_ +#define CLBLAST_ROUTINES_XSYR2K_H_ + +#include "internal/routine.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xsyr2k: public Routine { + public: + Xsyr2k(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoSyr2k(const Layout layout, const Triangle triangle, const Transpose ab_transpose, + const size_t n, const size_t k, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer &b_buffer, const size_t b_offset, const size_t b_ld, + const T beta, + const Buffer &c_buffer, const size_t c_offset, const size_t c_ld); + + private: + // Static variable to get the precision + const static Precision precision_; +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XSYR2K_H_ +#endif diff --git a/include/internal/routines/level3/xsyrk.h b/include/internal/routines/level3/xsyrk.h new file mode 100644 index 00000000..3dab731f --- /dev/null +++ b/include/internal/routines/level3/xsyrk.h @@ -0,0 +1,49 @@ + +// ================================================================================================= +// 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 Xsyrk routine. The precision is implemented using a template argument. +// The implementation is based on the regular Xgemm routine and kernel, but with two main changes: +// 1) The final unpad(transpose) kernel updates only the upper/lower triangular part. +// 2) The main Xgemm kernel masks workgroups not contributing to usefull data. This is only for +// performance reasons, as the actual masking is done later (see the first point). +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XSYRK_H_ +#define CLBLAST_ROUTINES_XSYRK_H_ + +#include "internal/routine.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xsyrk: public Routine { + public: + Xsyrk(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoSyrk(const Layout layout, const Triangle triangle, const Transpose a_transpose, + const size_t n, const size_t k, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const T beta, + const Buffer &c_buffer, const size_t c_offset, const size_t c_ld); + + private: + // Static variable to get the precision + const static Precision precision_; +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XSYRK_H_ +#endif diff --git a/include/internal/routines/level3/xtrmm.h b/include/internal/routines/level3/xtrmm.h new file mode 100644 index 00000000..4f49bebd --- /dev/null +++ b/include/internal/routines/level3/xtrmm.h @@ -0,0 +1,58 @@ + +// ================================================================================================= +// 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 Xtrmm routine. The implementation is based on first transforming the +// upper/lower unit/non-unit triangular matrix into a regular matrix and then calling the GEMM +// routine. Therefore, this class inherits from the Xgemm class. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XTRMM_H_ +#define CLBLAST_ROUTINES_XTRMM_H_ + +#include "internal/routines/level3/xgemm.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xtrmm: public Xgemm<T> { + public: + + // Uses several variables from the Routine class + using Routine::db_; + using Routine::context_; + + // Uses several helper functions from the Routine class + using Routine::RunKernel; + using Routine::ErrorIn; + using Routine::TestMatrixA; + using Routine::GetProgramFromCache; + + // Uses the regular Xgemm routine + using Xgemm<T>::DoGemm; + + // Constructor + Xtrmm(CommandQueue &queue, Event &event); + + // Templated-precision implementation of the routine + StatusCode DoTrmm(const Layout layout, const Side side, const Triangle triangle, + const Transpose a_transpose, const Diagonal diagonal, + const size_t m, const size_t n, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer &b_buffer, const size_t b_offset, const size_t b_ld); +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XTRMM_H_ +#endif |