From 9fb2c61b256ccf66b6a7b6f605008125288d60cf Mon Sep 17 00:00:00 2001
From: Cedric Nugteren <web@cedricnugteren.nl>
Date: Sun, 7 Jan 2018 14:27:15 +0100
Subject: Added API and tests for new GemmStridedBatched routine

---
 src/routines/levelx/xgemmstridedbatched.hpp | 66 +++++++++++++++++++++++++++++
 1 file changed, 66 insertions(+)
 create mode 100644 src/routines/levelx/xgemmstridedbatched.hpp

(limited to 'src/routines/levelx/xgemmstridedbatched.hpp')

diff --git a/src/routines/levelx/xgemmstridedbatched.hpp b/src/routines/levelx/xgemmstridedbatched.hpp
new file mode 100644
index 00000000..0dbbcb10
--- /dev/null
+++ b/src/routines/levelx/xgemmstridedbatched.hpp
@@ -0,0 +1,66 @@
+
+// =================================================================================================
+// 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 XgemmStridedBatched routine. This is a non-blas batched version of GEMM.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_ROUTINES_XGEMMSTRIDEDBATCHED_H_
+#define CLBLAST_ROUTINES_XGEMMSTRIDEDBATCHED_H_
+
+#include <vector>
+
+#include "routine.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class XgemmStridedBatched: public Routine {
+public:
+
+  // Constructor
+  XgemmStridedBatched(Queue &queue, EventPointer event, const std::string &name = "GEMMSTRIDEDBATCHED");
+
+  // Templated-precision implementation of the routine
+  void DoGemmStridedBatched(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<T> &a_buffer, const size_t a_offset, const size_t a_ld, const size_t a_stride,
+                            const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld, const size_t b_stride, const T beta,
+                            const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld, const size_t c_stride,
+                            const size_t batch_count);
+
+  // Indirect version of strided batched GEMM (with pre and post-processing kernels)
+  void BatchedGemmIndirect(const size_t m, const size_t n, const size_t k, const T alpha,
+                           const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld, const size_t a_stride,
+                           const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld, const size_t b_stride, const T beta,
+                           const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld, const size_t c_stride,
+                           const bool a_do_transpose, const bool b_do_transpose, const bool c_do_transpose,
+                           const bool a_conjugate, const bool b_conjugate,
+                           const size_t a_one, const size_t a_two,
+                           const size_t b_one, const size_t b_two,
+                           const size_t c_one, const size_t c_two,
+                           const size_t batch_count);
+
+  // Direct version of strided batched GEMM (no pre and post-processing kernels)
+  void BatchedGemmDirect(const size_t m, const size_t n, const size_t k, const T alpha,
+                         const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld, const size_t a_stride,
+                         const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld, const size_t b_stride, const T beta,
+                         const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld, const size_t c_stride,
+                         const bool a_do_transpose, const bool b_do_transpose, const bool c_do_transpose,
+                         const bool a_conjugate, const bool b_conjugate,
+                         const size_t batch_count);
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_ROUTINES_XGEMMSTRIDEDBATCHED_H_
+#endif
-- 
cgit v1.2.3