Added a proper data-preparation function for the TRSM tests

1 files changed, 17 insertions, 35 deletions

diff --git a/test/routines/level3/xtrsm.hpp b/test/routines/level3/xtrsm.hpp
index 1ffaef35..0da4189d 100644
--- a/test/routines/level3/xtrsm.hpp
+++ b/test/routines/level3/xtrsm.hpp
@@ -19,6 +19,8 @@
 #include <vector>
 #include <string>
 
+#include "test/routines/level3/xtrsm_data.hpp"
+
 #ifdef CLBLAST_REF_CLBLAS
   #include "test/wrapper_clblas.hpp"
 #endif
@@ -29,38 +31,6 @@
 namespace clblast {
 // =================================================================================================
 
-// Prepares the data
-template <typename T>
-void PrepareData(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-  const auto k = (args.side == Side::kLeft) ? args.m : args.n;
-  if (args.a_ld < k) { return; }
-  if (args.a_size <= 0 || args.b_size <= 0) { return; }
-
-  // Copies input buffers to the host
-  std::vector<T> a_mat_cpu(args.a_size, static_cast<T>(0));
-  std::vector<T> b_mat_cpu(args.b_size, static_cast<T>(0));
-  buffers.a_mat.Read(queue, args.a_size, a_mat_cpu);
-  buffers.b_mat.Read(queue, args.b_size, b_mat_cpu);
-
-  // Generates 'proper' input for the TRSM routine
-  // TODO: Improve this
-  for (auto i = size_t{0}; i < k; ++i) {
-    for (auto j = size_t{0}; j < k; ++j) {
-      auto value = a_mat_cpu[j*args.a_ld + i + args.a_offset];
-      value *= ConstantTwo<T>();
-      if (IsCloseToZero(value)) { value += ConstantOne<T>(); }
-      a_mat_cpu[j*args.a_ld + i + args.a_offset] = value;
-    }
-  }
-
-  // Copies input buffers back to the OpenCL device
-  buffers.a_mat.Write(queue, args.a_size, a_mat_cpu);
-  buffers.b_mat.Write(queue, args.b_size, b_mat_cpu);
-  return;
-}
-
-// =================================================================================================
-
 // See comment at top of file for a description of the class
 template <typename T>
 class TestXtrsm {
@@ -105,9 +75,23 @@ class TestXtrsm {
   static Transposes GetATransposes(const Transposes &all) { return all; }
   static Transposes GetBTransposes(const Transposes &) { return {}; } // N/A for this routine
 
+  // Describes how to prepare the input data
+  static void PrepareData(const Arguments<T> &args, Queue &queue, const int seed,
+                          std::vector<T>&, std::vector<T>&,
+                          std::vector<T>& a_source_, std::vector<T>& b_source_, std::vector<T>&,
+                          std::vector<T>&, std::vector<T>&) {
+    const auto k = (args.side == Side::kLeft) ? args.m : args.n;
+    const auto b_one = (args.layout == Layout::kRowMajor) ? args.n : args.m;
+    if (args.a_ld < k) { return; }
+    if (args.b_ld < b_one) { return; }
+    if (args.a_size <= 0 || args.b_size <= 0) { return; }
+
+    // TODO: This is a copy of the clBLAS random matrix generation, make it work properly
+    GenerateProperTrsmMatrices(args, seed, &a_source_[args.a_offset], &b_source_[args.b_offset]);
+  }
+
   // Describes how to run the CLBlast routine
   static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-    PrepareData(args, buffers, queue);
     auto queue_plain = queue();
     auto event = cl_event{};
     auto status = Trsm(args.layout, args.side, args.triangle, args.a_transpose, args.diagonal,
@@ -122,7 +106,6 @@ class TestXtrsm {
   // Describes how to run the clBLAS routine (for correctness/performance comparison)
   #ifdef CLBLAST_REF_CLBLAS
     static StatusCode RunReference1(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-      PrepareData(args, buffers, queue);
       auto queue_plain = queue();
       auto event = cl_event{};
       auto status = clblasXtrsm(convertToCLBLAS(args.layout),
@@ -142,7 +125,6 @@ class TestXtrsm {
   // Describes how to run the CPU BLAS routine (for correctness/performance comparison)
   #ifdef CLBLAST_REF_CBLAS
     static StatusCode RunReference2(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
-      PrepareData(args, buffers, queue);
       std::vector<T> a_mat_cpu(args.a_size, static_cast<T>(0));
       std::vector<T> b_mat_cpu(args.b_size, static_cast<T>(0));
       buffers.a_mat.Read(queue, args.a_size, a_mat_cpu);