Added the HER2K routine, tester, and client

4 files changed, 306 insertions, 0 deletions

diff --git a/test/correctness/routines/xher2k.cc b/test/correctness/routines/xher2k.cc
new file mode 100644
index 00000000..7c0e5a92
--- /dev/null
+++ b/test/correctness/routines/xher2k.cc
@@ -0,0 +1,100 @@
+
+// =================================================================================================
+// 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 tests for the Xher2k routine.
+//
+// =================================================================================================
+
+#include "correctness/testblas.h"
+#include "routines/xher2k.h"
+
+namespace clblast {
+// =================================================================================================
+
+// The correctness tester
+template <typename T, typename U>
+void RunTest(int argc, char *argv[], const bool silent, const std::string &name) {
+
+  // Creates a tester
+  TestBlas<T,U> tester{argc, argv, silent, name, TestXher2k<T,U>::GetOptions(),
+                       TestXher2k<T,U>::RunRoutine, TestXher2k<T,U>::RunReference,
+                       TestXher2k<T,U>::DownloadResult, TestXher2k<T,U>::GetResultIndex,
+                       TestXher2k<T,U>::ResultID1, TestXher2k<T,U>::ResultID2};
+
+  // This variable holds the arguments relevant for this routine
+  auto args = Arguments<U>{};
+
+  // Loops over the test-cases from a data-layout point of view
+  for (auto &layout: tester.kLayouts) { args.layout = layout;
+    for (auto &triangle: tester.kTriangles) { args.triangle = triangle;
+      for (auto &ab_transpose: {Transpose::kNo, Transpose::kConjugate}) { // Regular transpose not a
+        args.a_transpose = ab_transpose;                                  // valid BLAS option
+        args.b_transpose = ab_transpose;
+
+        // Creates the arguments vector for the regular tests
+        auto regular_test_vector = std::vector<Arguments<U>>{};
+        for (auto &n: tester.kMatrixDims) { args.n = n;
+          for (auto &k: tester.kMatrixDims) { args.k = k;
+            for (auto &a_ld: tester.kMatrixDims) { args.a_ld = a_ld;
+              for (auto &a_offset: tester.kOffsets) { args.a_offset = a_offset;
+                for (auto &b_ld: tester.kMatrixDims) { args.b_ld = b_ld;
+                  for (auto &b_offset: tester.kOffsets) { args.b_offset = b_offset;
+                    for (auto &c_ld: tester.kMatrixDims) { args.c_ld = c_ld;
+                      for (auto &c_offset: tester.kOffsets) { args.c_offset = c_offset;
+                        for (auto &alpha: tester.kAlphaValues) { args.alpha = alpha;
+                          for (auto &beta: tester.kBetaValues) { args.beta = beta;
+                            args.a_size = TestXher2k<T,U>::GetSizeA(args);
+                            args.b_size = TestXher2k<T,U>::GetSizeB(args);
+                            args.c_size = TestXher2k<T,U>::GetSizeC(args);
+                            if (args.a_size<1 || args.b_size<1 || args.c_size<1) { continue; }
+                            regular_test_vector.push_back(args);
+                          }
+                        }
+                      }
+                    }
+                  }
+                }
+              }
+            }
+          }
+        }
+
+        // Creates the arguments vector for the invalid-buffer tests
+        auto invalid_test_vector = std::vector<Arguments<U>>{};
+        args.n = args.k = tester.kBufferSize;
+        args.a_ld = args.b_ld = args.c_ld = tester.kBufferSize;
+        args.a_offset = args.b_offset = args.c_offset = 0;
+        for (auto &a_size: tester.kMatSizes) { args.a_size = a_size;
+          for (auto &b_size: tester.kMatSizes) { args.b_size = b_size;
+            for (auto &c_size: tester.kMatSizes) { args.c_size = c_size;
+              invalid_test_vector.push_back(args);
+            }
+          }
+        }
+
+        // Runs the tests
+        const auto case_name = ToString(layout)+" "+ToString(triangle)+" "+ToString(ab_transpose);
+        tester.TestRegular(regular_test_vector, case_name);
+        tester.TestInvalid(invalid_test_vector, case_name);
+      }
+    }
+  }
+}
+
+// =================================================================================================
+} // namespace clblast
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  clblast::RunTest<clblast::float2,float>(argc, argv, false, "CHER2K");
+  clblast::RunTest<clblast::double2,double>(argc, argv, true, "ZHER2K");
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/performance/routines/xher2k.cc b/test/performance/routines/xher2k.cc
new file mode 100644
index 00000000..1b505737
--- /dev/null
+++ b/test/performance/routines/xher2k.cc
@@ -0,0 +1,40 @@
+
+// =================================================================================================
+// 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 command-line interface performance tester.
+//
+// =================================================================================================
+
+#include "performance/client.h"
+#include "routines/xher2k.h"
+
+// =================================================================================================
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv)) {
+    case clblast::Precision::kHalf:
+      throw std::runtime_error("Unsupported precision mode");
+    case clblast::Precision::kSingle:
+      throw std::runtime_error("Unsupported precision mode");
+    case clblast::Precision::kDouble:
+      throw std::runtime_error("Unsupported precision mode");
+    case clblast::Precision::kComplexSingle:
+      clblast::RunClient<clblast::TestXher2k<float2,float>, float2, float>(argc, argv); break;
+    case clblast::Precision::kComplexDouble:
+      clblast::RunClient<clblast::TestXher2k<double2,double>, double2, double>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/routines/xher2k.h b/test/routines/xher2k.h
new file mode 100644
index 00000000..f13e8a62
--- /dev/null
+++ b/test/routines/xher2k.h
@@ -0,0 +1,132 @@
+
+// =================================================================================================
+// 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 Xher2k 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_XHER2K_H_
+#define CLBLAST_TEST_ROUTINES_XHER2K_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, typename U>
+class TestXher2k {
+ public:
+
+  // The list of arguments relevant for this routine
+  static std::vector<std::string> GetOptions() {
+    return {kArgN, kArgK,
+            kArgLayout, kArgTriangle, kArgATransp,
+            kArgALeadDim, kArgBLeadDim, kArgCLeadDim,
+            kArgAOffset, kArgBOffset, kArgCOffset,
+            kArgAlpha, kArgBeta};
+  }
+
+  // Describes how to obtain the sizes of the buffers
+  static size_t GetSizeA(const Arguments<U> &args) {
+    auto a_rotated = (args.layout == Layout::kColMajor && args.a_transpose != Transpose::kNo) ||
+                     (args.layout == Layout::kRowMajor && args.a_transpose == Transpose::kNo);
+    auto a_two = (a_rotated) ? args.n : args.k;
+    return a_two * args.a_ld + args.a_offset;
+  }
+  static size_t GetSizeB(const Arguments<U> &args) {
+    auto b_rotated = (args.layout == Layout::kColMajor && args.b_transpose != Transpose::kNo) ||
+                     (args.layout == Layout::kRowMajor && args.b_transpose == Transpose::kNo);
+    auto b_two = (b_rotated) ? args.n : args.k;
+    return b_two * args.b_ld + args.b_offset;
+  }
+  static size_t GetSizeC(const Arguments<U> &args) {
+    return args.n * args.c_ld + args.c_offset;
+  }
+
+  // Describes how to set the sizes of all the buffers
+  static void SetSizes(Arguments<U> &args) {
+    args.a_size = GetSizeA(args);
+    args.b_size = GetSizeB(args);
+    args.c_size = GetSizeC(args);
+  }
+
+  // Describes what the default values of the leading dimensions of the matrices are
+  static size_t DefaultLDA(const Arguments<U> &args) { return args.k; }
+  static size_t DefaultLDB(const Arguments<U> &args) { return args.k; }
+  static size_t DefaultLDC(const Arguments<U> &args) { return args.n; }
+
+  // Describes how to run the CLBlast routine
+  static StatusCode RunRoutine(const Arguments<U> &args, const Buffers &buffers,
+                               CommandQueue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto alpha2 = T{args.alpha, args.alpha};
+    auto status = Her2k(args.layout, args.triangle, args.a_transpose,
+                        args.n, args.k, alpha2,
+                        buffers.a_mat(), args.a_offset, args.a_ld,
+                        buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                        buffers.c_mat(), args.c_offset, args.c_ld,
+                        &queue_plain, &event);
+    clWaitForEvents(1, &event);
+    return status;
+  }
+
+  // Describes how to run the clBLAS routine (for correctness/performance comparison)
+  static StatusCode RunReference(const Arguments<U> &args, const Buffers &buffers,
+                                 CommandQueue &queue) {
+    auto queue_plain = queue();
+    auto event = cl_event{};
+    auto alpha2 = T{args.alpha, args.alpha};
+    auto status = clblasXher2k(static_cast<clblasOrder>(args.layout),
+                               static_cast<clblasUplo>(args.triangle),
+                               static_cast<clblasTranspose>(args.a_transpose),
+                               args.n, args.k, alpha2,
+                               buffers.a_mat(), args.a_offset, args.a_ld,
+                               buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
+                               buffers.c_mat(), args.c_offset, args.c_ld,
+                               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<U> &args, Buffers &buffers,
+                                       CommandQueue &queue) {
+    std::vector<T> result(args.c_size, static_cast<T>(0));
+    buffers.c_mat.ReadBuffer(queue, args.c_size*sizeof(T), result);
+    return result;
+  }
+
+  // Describes how to compute the indices of the result buffer
+  static size_t ResultID1(const Arguments<U> &args) { return args.n; }
+  static size_t ResultID2(const Arguments<U> &args) { return args.n; }
+  static size_t GetResultIndex(const Arguments<U> &args, const size_t id1, const size_t id2) {
+    return id1*args.c_ld + id2 + args.c_offset;
+  }
+
+  // Describes how to compute performance metrics
+  static size_t GetFlops(const Arguments<U> &args) {
+    return 2 * args.n * args.n * args.k;
+  }
+  static size_t GetBytes(const Arguments<U> &args) {
+    return (args.n*args.k + args.n*args.n) * sizeof(T);
+  }
+};
+
+// =================================================================================================
+} // namespace clblast
+
+// CLBLAST_TEST_ROUTINES_XHER2K_H_
+#endif
diff --git a/test/wrapper_clblas.h b/test/wrapper_clblas.h
index 360adfc8..03f31a58 100644
--- a/test/wrapper_clblas.h
+++ b/test/wrapper_clblas.h
@@ -419,6 +419,40 @@ clblasStatus clblasXsyr2k(
                         num_queues, queues, num_wait_events, wait_events, events);
 }
 
+// This calls {clblasCher2k, clblasZher2k} with the arguments forwarded.
+clblasStatus clblasXher2k(
+  clblasOrder layout, clblasUplo triangle, clblasTranspose ab_transpose,
+  size_t n, size_t k, float2 alpha,
+  const cl_mem a_mat, size_t a_offset, size_t a_ld,
+  const cl_mem b_mat, size_t b_offset, size_t b_ld, float beta,
+  const cl_mem c_mat, size_t c_offset, size_t c_ld,
+  cl_uint num_queues, cl_command_queue *queues,
+  cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+    auto cl_alpha = cl_float2{{alpha.real(), alpha.imag()}};
+    return clblasCher2k(layout, triangle, ab_transpose,
+                        n, k, cl_alpha,
+                        a_mat, a_offset, a_ld,
+                        b_mat, b_offset, b_ld, beta,
+                        c_mat, c_offset, c_ld,
+                        num_queues, queues, num_wait_events, wait_events, events);
+}
+clblasStatus clblasXher2k(
+  clblasOrder layout, clblasUplo triangle, clblasTranspose ab_transpose,
+  size_t n, size_t k, double2 alpha,
+  const cl_mem a_mat, size_t a_offset, size_t a_ld,
+  const cl_mem b_mat, size_t b_offset, size_t b_ld, double beta,
+  const cl_mem c_mat, size_t c_offset, size_t c_ld,
+  cl_uint num_queues, cl_command_queue *queues,
+  cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+    auto cl_alpha = cl_double2{{alpha.real(), alpha.imag()}};
+    return clblasZher2k(layout, triangle, ab_transpose,
+                        n, k, cl_alpha,
+                        a_mat, a_offset, a_ld,
+                        b_mat, b_offset, b_ld, beta,
+                        c_mat, c_offset, c_ld,
+                        num_queues, queues, num_wait_events, wait_events, events);
+}
+
 // This calls {clblasStrmm, clblasDtrmm, clblasCtrmm, clblasZtrmm} with the arguments forwarded.
 clblasStatus clblasXtrmm(
   clblasOrder layout, clblasSide side, clblasUplo triangle,