Merge pull request #11 from CNugteren/level3_routines_2

Added level-3 routines

7 files changed, 273 insertions, 4 deletions

diff --git a/include/clblast.h b/include/clblast.h
index da504a0b..80ea1707 100644
--- a/include/clblast.h
+++ b/include/clblast.h
@@ -75,6 +75,7 @@ enum class Layout { kRowMajor, kColMajor };
 enum class Transpose { kNo, kYes, kConjugate };
 enum class Side { kLeft, kRight };
 enum class Triangle { kUpper, kLower };
+enum class Diagonal { kUnit, kNonUnit };
 
 // Precision scoped enum (values in bits)
 enum class Precision { kHalf = 16, kSingle = 32, kDouble = 64,
@@ -95,7 +96,7 @@ StatusCode Axpy(const size_t n, const T alpha,
 
 // Templated-precision generalized matrix-vector multiplication: SGEMV/DGEMV/CGEMV/ZGEMV
 template <typename T>
-StatusCode Gemv(const Layout layout, const Transpose transpose_a,
+StatusCode Gemv(const Layout layout, const Transpose a_transpose,
                 const size_t m, const size_t n,
                 const T alpha,
                 const cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
@@ -109,7 +110,7 @@ StatusCode Gemv(const Layout layout, const Transpose transpose_a,
 
 // Templated-precision generalized matrix-matrix multiplication: SGEMM/DGEMM/CGEMM/ZGEMM
 template <typename T>
-StatusCode Gemm(const Layout layout, const Transpose transpose_a, const Transpose transpose_b,
+StatusCode Gemm(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 cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
@@ -129,9 +130,30 @@ StatusCode Symm(const Layout layout, const Side side, const Triangle triangle,
                 cl_mem c_buffer, const size_t c_offset, const size_t c_ld,
                 cl_command_queue* queue, cl_event* event);
 
+// Templated-precision hermitian matrix-matrix multiplication: CHEMM/ZHEMM
+template <typename T>
+StatusCode Hemm(const Layout layout, const Side side, const Triangle triangle,
+                const size_t m, const size_t n,
+                const T alpha,
+                const cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
+                const cl_mem b_buffer, const size_t b_offset, const size_t b_ld,
+                const T beta,
+                cl_mem c_buffer, const size_t c_offset, const size_t c_ld,
+                cl_command_queue* queue, cl_event* event);
+
 // Templated-precision rank-K update of a symmetric matrix: SSYRK/DSYRK/CSYRK/ZSYRK
 template <typename T>
-StatusCode Syrk(const Layout layout, const Triangle triangle, const Transpose transpose_a,
+StatusCode Syrk(const Layout layout, const Triangle triangle, const Transpose a_transpose,
+                const size_t n, const size_t k,
+                const T alpha,
+                const cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
+                const T beta,
+                cl_mem c_buffer, const size_t c_offset, const size_t c_ld,
+                cl_command_queue* queue, cl_event* event);
+
+// Templated-precision rank-K update of a hermitian matrix: CHERK/ZHERK
+template <typename T>
+StatusCode Herk(const Layout layout, const Triangle triangle, const Transpose a_transpose,
                 const size_t n, const size_t k,
                 const T alpha,
                 const cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
@@ -141,7 +163,7 @@ StatusCode Syrk(const Layout layout, const Triangle triangle, const Transpose tr
 
 // Templated-precision rank-2K update of a symmetric matrix: SSYR2K/DSYR2K/CSYR2K/ZSYR2K
 template <typename T>
-StatusCode Syr2k(const Layout layout, const Triangle triangle, const Transpose transpose_ab,
+StatusCode Syr2k(const Layout layout, const Triangle triangle, const Transpose ab_transpose,
                  const size_t n, const size_t k,
                  const T alpha,
                  const cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
@@ -150,6 +172,39 @@ StatusCode Syr2k(const Layout layout, const Triangle triangle, const Transpose t
                  cl_mem c_buffer, const size_t c_offset, const size_t c_ld,
                  cl_command_queue* queue, cl_event* event);
 
+// Templated-precision rank-2K update of a hermitian matrix: CHER2K/ZHER2K
+template <typename T, typename U>
+StatusCode Her2k(const Layout layout, const Triangle triangle, const Transpose ab_transpose,
+                 const size_t n, const size_t k,
+                 const T alpha,
+                 const cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
+                 const cl_mem b_buffer, const size_t b_offset, const size_t b_ld,
+                 const U beta,
+                 cl_mem c_buffer, const size_t c_offset, const size_t c_ld,
+                 cl_command_queue* queue, cl_event* event);
+
+// Templated-precision triangular matrix-matrix multiplication: STRMM/DTRMM/CTRMM/ZTRMM
+template <typename T>
+StatusCode Trmm(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 cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
+                cl_mem b_buffer, const size_t b_offset, const size_t b_ld,
+                cl_command_queue* queue, cl_event* event);
+
+// Templated-precision matrix equation solver: STRSM/DTRSM/CTRSM/ZTRSM
+/*
+template <typename T>
+StatusCode Trsm(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 cl_mem a_buffer, const size_t a_offset, const size_t a_ld,
+                cl_mem b_buffer, const size_t b_offset, const size_t b_ld,
+                cl_command_queue* queue, cl_event* event);
+*/
+
 // =================================================================================================
 } // namespace clblast
 
diff --git a/include/internal/routine.h b/include/internal/routine.h
index d2cee52c..acc9a9c8 100644
--- a/include/internal/routine.h
+++ b/include/internal/routine.h
@@ -93,6 +93,7 @@ class Routine {
                                     const Buffer &dest,
                                     const bool do_transpose, const bool do_conjugate,
                                     const bool pad, const bool upper, const bool lower,
+                                    const bool diagonal_imag_zero,
                                     const Program &program);
   
   // Queries the cache and retrieve either a matching program or a boolean whether a match exists.
diff --git a/include/internal/routines/xhemm.h b/include/internal/routines/xhemm.h
new file mode 100644
index 00000000..1b1a0dfa
--- /dev/null
+++ b/include/internal/routines/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/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/xher2k.h b/include/internal/routines/xher2k.h
new file mode 100644
index 00000000..1836a812
--- /dev/null
+++ b/include/internal/routines/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/xherk.h b/include/internal/routines/xherk.h
new file mode 100644
index 00000000..9b361254
--- /dev/null
+++ b/include/internal/routines/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/xtrmm.h b/include/internal/routines/xtrmm.h
new file mode 100644
index 00000000..af9f0266
--- /dev/null
+++ b/include/internal/routines/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/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
diff --git a/include/internal/utilities.h b/include/internal/utilities.h
index 93cd509e..60d70eae 100644
--- a/include/internal/utilities.h
+++ b/include/internal/utilities.h
@@ -46,6 +46,7 @@ constexpr auto kArgATransp = "transA";
 constexpr auto kArgBTransp = "transB";
 constexpr auto kArgSide = "side";
 constexpr auto kArgTriangle = "triangle";
+constexpr auto kArgDiagonal = "diagonal";
 constexpr auto kArgXInc = "incx";
 constexpr auto kArgYInc = "incy";
 constexpr auto kArgXOffset = "offx";
@@ -93,6 +94,7 @@ struct Arguments {
   Transpose b_transpose = Transpose::kNo;
   Side side = Side::kLeft;
   Triangle triangle = Triangle::kUpper;
+  Diagonal diagonal = Diagonal::kUnit;
   size_t x_inc = 1;
   size_t y_inc = 1;
   size_t x_offset = 0;