Added TRMV/TBMV/TPMV routines

3 files changed, 375 insertions, 0 deletions

diff --git a/test/routines/level2/xtbmv.h b/test/routines/level2/xtbmv.h
new file mode 100644
index 00000000..dbdddb65
--- /dev/null
+++ b/test/routines/level2/xtbmv.h
@@ -0,0 +1,125 @@
+
+// =================================================================================================
+// 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 a class with static methods to describe the Xtbmv routine. Examples of
+// such 'descriptions' are how to calculate the size a of buffer or how to run the routine. These
+// static methods are used by the correctness tester and the performance tester.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_TEST_ROUTINES_XTBMV_H_
+#define CLBLAST_TEST_ROUTINES_XTBMV_H_
+
+#include <vector>
+#include <string>
+
+#include "wrapper_clblas.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TestXtbmv {
+ public:
+
+  // The BLAS level: 1, 2, or 3
+  static size_t BLASLevel() { return 2; }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() {
+    return {kArgN, kArgKL,
+            kArgLayout, kArgTriangle, kArgATransp, kArgDiagonal,
+            kArgALeadDim, kArgXInc,
+            kArgAOffset, kArgXOffset};
+  }
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &args) {
+    return args.n * args.x_inc + args.x_offset;
+  }
+  static size_t GetSizeA(const Arguments<T> &args) {
+    return args.n * args.a_ld + args.a_offset;
+  }
+
+  // Describes how to set the sizes of all the buffers
+  static void SetSizes(Arguments<T> &args) {
+    args.a_size = GetSizeA(args);
+    args.x_size = GetSizeX(args);
+  }
+
+  // Describes what the default values of the leading dimensions of the matrices are
+  static size_t DefaultLDA(const Arguments<T> &args) { return args.n; }
+  static size_t DefaultLDB(const Arguments<T> &) { return 1; } // N/A for this routine
+  static size_t DefaultLDC(const Arguments<T> &) { return 1; } // N/A for this routine
+
+  // Describes which transpose options are relevant for this routine
+  using Transposes = std::vector<Transpose>;
+  static Transposes GetATransposes(const Transposes &all) { return all; }
+  static Transposes GetBTransposes(const Transposes &) { return {}; } // N/A for this routine
+
+  // Describes how to run the CLBlast routine
+  static StatusCode RunRoutine(const Arguments<T> &args, const Buffers<T> &buffers, Queue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto status = Tbmv<T>(args.layout, args.triangle, args.a_transpose, args.diagonal,
+                          args.n, args.kl,
+                          buffers.a_mat(), args.a_offset, args.a_ld,
+                          buffers.x_vec(), args.x_offset, args.x_inc,
+                          &queue_plain, &event);
+    clWaitForEvents(1, &event);
+    return status;
+  }
+
+  // Describes how to run the clBLAS routine (for correctness/performance comparison)
+  static StatusCode RunReference(const Arguments<T> &args, const Buffers<T> &buffers, Queue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto status = clblasXtbmv<T>(static_cast<clblasOrder>(args.layout),
+                                 static_cast<clblasUplo>(args.triangle),
+                                 static_cast<clblasTranspose>(args.a_transpose),
+                                 static_cast<clblasDiag>(args.diagonal),
+                                 args.n, args.kl,
+                                 buffers.a_mat(), args.a_offset, args.a_ld,
+                                 buffers.x_vec(), args.x_offset, args.x_inc,
+                                 1, &queue_plain, 0, nullptr, &event);
+    clWaitForEvents(1, &event);
+    return static_cast<StatusCode>(status);
+  }
+
+  // Describes how to download the results of the computation (more importantly: which buffer)
+  static std::vector<T> DownloadResult(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
+    std::vector<T> result(args.x_size, static_cast<T>(0));
+    buffers.x_vec.Read(queue, args.x_size, result);
+    return result;
+  }
+
+  // Describes how to compute the indices of the result buffer
+  static size_t ResultID1(const Arguments<T> &args) {
+    return args.n;
+  }
+  static size_t ResultID2(const Arguments<T> &) { return 1; } // N/A for this routine
+  static size_t GetResultIndex(const Arguments<T> &args, const size_t id1, const size_t) {
+    return id1*args.x_inc + args.x_offset;
+  }
+
+  // Describes how to compute performance metrics
+  static size_t GetFlops(const Arguments<T> &args) {
+    return 2 * args.n * args.n;
+  }
+  static size_t GetBytes(const Arguments<T> &args) {
+    return ((args.kl+args.kl+1)*args.n + 2*args.n + args.n) * sizeof(T);
+  }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_TEST_ROUTINES_XTBMV_H_
+#endif
diff --git a/test/routines/level2/xtpmv.h b/test/routines/level2/xtpmv.h
new file mode 100644
index 00000000..4425765e
--- /dev/null
+++ b/test/routines/level2/xtpmv.h
@@ -0,0 +1,125 @@
+
+// =================================================================================================
+// 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 a class with static methods to describe the Xtpmv routine. Examples of
+// such 'descriptions' are how to calculate the size a of buffer or how to run the routine. These
+// static methods are used by the correctness tester and the performance tester.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_TEST_ROUTINES_XTPMV_H_
+#define CLBLAST_TEST_ROUTINES_XTPMV_H_
+
+#include <vector>
+#include <string>
+
+#include "wrapper_clblas.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TestXtpmv {
+ public:
+
+  // The BLAS level: 1, 2, or 3
+  static size_t BLASLevel() { return 2; }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() {
+    return {kArgN,
+            kArgLayout, kArgTriangle, kArgATransp, kArgDiagonal,
+            kArgXInc,
+            kArgAPOffset, kArgXOffset};
+  }
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &args) {
+    return args.n * args.x_inc + args.x_offset;
+  }
+  static size_t GetSizeAP(const Arguments<T> &args) {
+    return ((args.n*(args.n+1)) / 2) + args.ap_offset;
+  }
+
+  // Describes how to set the sizes of all the buffers
+  static void SetSizes(Arguments<T> &args) {
+    args.ap_size = GetSizeAP(args);
+    args.x_size = GetSizeX(args);
+  }
+
+  // Describes what the default values of the leading dimensions of the matrices are
+  static size_t DefaultLDA(const Arguments<T> &args) { return args.n; }
+  static size_t DefaultLDB(const Arguments<T> &) { return 1; } // N/A for this routine
+  static size_t DefaultLDC(const Arguments<T> &) { return 1; } // N/A for this routine
+
+  // Describes which transpose options are relevant for this routine
+  using Transposes = std::vector<Transpose>;
+  static Transposes GetATransposes(const Transposes &all) { return all; }
+  static Transposes GetBTransposes(const Transposes &) { return {}; } // N/A for this routine
+
+  // Describes how to run the CLBlast routine
+  static StatusCode RunRoutine(const Arguments<T> &args, const Buffers<T> &buffers, Queue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto status = Tpmv<T>(args.layout, args.triangle, args.a_transpose, args.diagonal,
+                          args.n,
+                          buffers.ap_mat(), args.ap_offset,
+                          buffers.x_vec(), args.x_offset, args.x_inc,
+                          &queue_plain, &event);
+    clWaitForEvents(1, &event);
+    return status;
+  }
+
+  // Describes how to run the clBLAS routine (for correctness/performance comparison)
+  static StatusCode RunReference(const Arguments<T> &args, const Buffers<T> &buffers, Queue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto status = clblasXtpmv<T>(static_cast<clblasOrder>(args.layout),
+                                 static_cast<clblasUplo>(args.triangle),
+                                 static_cast<clblasTranspose>(args.a_transpose),
+                                 static_cast<clblasDiag>(args.diagonal),
+                                 args.n,
+                                 buffers.ap_mat(), args.ap_offset,
+                                 buffers.x_vec(), args.x_offset, args.x_inc,
+                                 1, &queue_plain, 0, nullptr, &event);
+    clWaitForEvents(1, &event);
+    return static_cast<StatusCode>(status);
+  }
+
+  // Describes how to download the results of the computation (more importantly: which buffer)
+  static std::vector<T> DownloadResult(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
+    std::vector<T> result(args.x_size, static_cast<T>(0));
+    buffers.x_vec.Read(queue, args.x_size, result);
+    return result;
+  }
+
+  // Describes how to compute the indices of the result buffer
+  static size_t ResultID1(const Arguments<T> &args) {
+    return args.n;
+  }
+  static size_t ResultID2(const Arguments<T> &) { return 1; } // N/A for this routine
+  static size_t GetResultIndex(const Arguments<T> &args, const size_t id1, const size_t) {
+    return id1*args.x_inc + args.x_offset;
+  }
+
+  // Describes how to compute performance metrics
+  static size_t GetFlops(const Arguments<T> &args) {
+    return 2 * args.n * args.n;
+  }
+  static size_t GetBytes(const Arguments<T> &args) {
+    return (((args.n*(args.n+1)) / 2) + 2*args.n + args.n) * sizeof(T);
+  }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_TEST_ROUTINES_XTPMV_H_
+#endif
diff --git a/test/routines/level2/xtrmv.h b/test/routines/level2/xtrmv.h
new file mode 100644
index 00000000..1c0c6fd8
--- /dev/null
+++ b/test/routines/level2/xtrmv.h
@@ -0,0 +1,125 @@
+
+// =================================================================================================
+// 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 a class with static methods to describe the Xtrmv routine. Examples of
+// such 'descriptions' are how to calculate the size a of buffer or how to run the routine. These
+// static methods are used by the correctness tester and the performance tester.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_TEST_ROUTINES_XTRMV_H_
+#define CLBLAST_TEST_ROUTINES_XTRMV_H_
+
+#include <vector>
+#include <string>
+
+#include "wrapper_clblas.h"
+
+namespace clblast {
+// =================================================================================================
+
+// See comment at top of file for a description of the class
+template <typename T>
+class TestXtrmv {
+ public:
+
+  // The BLAS level: 1, 2, or 3
+  static size_t BLASLevel() { return 2; }
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() {
+    return {kArgN,
+            kArgLayout, kArgTriangle, kArgATransp, kArgDiagonal,
+            kArgALeadDim, kArgXInc,
+            kArgAOffset, kArgXOffset};
+  }
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeX(const Arguments<T> &args) {
+    return args.n * args.x_inc + args.x_offset;
+  }
+  static size_t GetSizeA(const Arguments<T> &args) {
+    return args.n * args.a_ld + args.a_offset;
+  }
+
+  // Describes how to set the sizes of all the buffers
+  static void SetSizes(Arguments<T> &args) {
+    args.a_size = GetSizeA(args);
+    args.x_size = GetSizeX(args);
+  }
+
+  // Describes what the default values of the leading dimensions of the matrices are
+  static size_t DefaultLDA(const Arguments<T> &args) { return args.n; }
+  static size_t DefaultLDB(const Arguments<T> &) { return 1; } // N/A for this routine
+  static size_t DefaultLDC(const Arguments<T> &) { return 1; } // N/A for this routine
+
+  // Describes which transpose options are relevant for this routine
+  using Transposes = std::vector<Transpose>;
+  static Transposes GetATransposes(const Transposes &all) { return all; }
+  static Transposes GetBTransposes(const Transposes &) { return {}; } // N/A for this routine
+
+  // Describes how to run the CLBlast routine
+  static StatusCode RunRoutine(const Arguments<T> &args, const Buffers<T> &buffers, Queue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto status = Trmv<T>(args.layout, args.triangle, args.a_transpose, args.diagonal,
+                          args.n,
+                          buffers.a_mat(), args.a_offset, args.a_ld,
+                          buffers.x_vec(), args.x_offset, args.x_inc,
+                          &queue_plain, &event);
+    clWaitForEvents(1, &event);
+    return status;
+  }
+
+  // Describes how to run the clBLAS routine (for correctness/performance comparison)
+  static StatusCode RunReference(const Arguments<T> &args, const Buffers<T> &buffers, Queue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto status = clblasXtrmv<T>(static_cast<clblasOrder>(args.layout),
+                                 static_cast<clblasUplo>(args.triangle),
+                                 static_cast<clblasTranspose>(args.a_transpose),
+                                 static_cast<clblasDiag>(args.diagonal),
+                                 args.n,
+                                 buffers.a_mat(), args.a_offset, args.a_ld,
+                                 buffers.x_vec(), args.x_offset, args.x_inc,
+                                 1, &queue_plain, 0, nullptr, &event);
+    clWaitForEvents(1, &event);
+    return static_cast<StatusCode>(status);
+  }
+
+  // Describes how to download the results of the computation (more importantly: which buffer)
+  static std::vector<T> DownloadResult(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
+    std::vector<T> result(args.x_size, static_cast<T>(0));
+    buffers.x_vec.Read(queue, args.x_size, result);
+    return result;
+  }
+
+  // Describes how to compute the indices of the result buffer
+  static size_t ResultID1(const Arguments<T> &args) {
+    return args.n;
+  }
+  static size_t ResultID2(const Arguments<T> &) { return 1; } // N/A for this routine
+  static size_t GetResultIndex(const Arguments<T> &args, const size_t id1, const size_t) {
+    return id1*args.x_inc + args.x_offset;
+  }
+
+  // Describes how to compute performance metrics
+  static size_t GetFlops(const Arguments<T> &args) {
+    return 2 * args.n * args.n;
+  }
+  static size_t GetBytes(const Arguments<T> &args) {
+    return (args.n*args.n + 2*args.n + args.n) * sizeof(T);
+  }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_TEST_ROUTINES_XTRMV_H_
+#endif