summaryrefslogtreecommitdiff
path: root/test/routines
diff options
context:
space:
mode:
Diffstat (limited to 'test/routines')
-rw-r--r--test/routines/level1/xamax.hpp27
-rw-r--r--test/routines/levelx/xcol2im.hpp3
-rw-r--r--test/routines/levelx/xconvgemm.hpp3
-rw-r--r--test/routines/levelx/xhad.hpp5
-rw-r--r--test/routines/levelx/xim2col.hpp3
-rw-r--r--test/routines/levelx/xinvert.hpp3
-rw-r--r--test/routines/levelx/xomatcopy.hpp3
7 files changed, 30 insertions, 17 deletions
diff --git a/test/routines/level1/xamax.hpp b/test/routines/level1/xamax.hpp
index 71c1a0ec..aa0f8e10 100644
--- a/test/routines/level1/xamax.hpp
+++ b/test/routines/level1/xamax.hpp
@@ -35,15 +35,15 @@ class TestXamax {
kArgXInc,
kArgXOffset, kArgImaxOffset};
}
- static std::vector<std::string> BuffersIn() { return {kBufVecX, kBufScalar}; }
- static std::vector<std::string> BuffersOut() { return {kBufScalar}; }
+ static std::vector<std::string> BuffersIn() { return {kBufVecX, kBufScalarUint}; }
+ static std::vector<std::string> BuffersOut() { return {kBufScalarUint}; }
// 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 GetSizeImax(const Arguments<T> &args) {
- return (1 + args.imax_offset) * 2; // always a 4-byte integer, this is a hack for FP16
+ return args.imax_offset + 1;
}
// Describes how to set the sizes of all the buffers
@@ -73,13 +73,13 @@ class TestXamax {
auto queue_plain = queue();
auto event = cl_event{};
auto status = Amax<T>(args.n,
- buffers.scalar(), args.imax_offset,
+ buffers.scalar_uint(), args.imax_offset,
buffers.x_vec(), args.x_offset, args.x_inc,
&queue_plain, &event);
if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
#elif CUDA_API
auto status = Amax<T>(args.n,
- buffers.scalar(), args.imax_offset,
+ buffers.scalar_uint(), args.imax_offset,
buffers.x_vec(), args.x_offset, args.x_inc,
queue.GetContext()(), queue.GetDevice()());
cuStreamSynchronize(queue());
@@ -93,7 +93,7 @@ class TestXamax {
auto queue_plain = queue();
auto event = cl_event{};
auto status = clblasXamax<T>(args.n,
- buffers.scalar, args.imax_offset,
+ buffers.scalar_uint, args.imax_offset,
buffers.x_vec, args.x_offset, args.x_inc,
1, &queue_plain, 0, nullptr, &event);
clWaitForEvents(1, &event);
@@ -105,7 +105,7 @@ class TestXamax {
#ifdef CLBLAST_REF_CBLAS
static StatusCode RunReference2(const Arguments<T> &args, BuffersHost<T> &buffers_host, Queue &) {
cblasXamax(args.n,
- buffers_host.scalar, args.imax_offset,
+ buffers_host.scalar_uint, args.imax_offset,
buffers_host.x_vec, args.x_offset, args.x_inc);
return StatusCode::kSuccess;
}
@@ -115,7 +115,7 @@ class TestXamax {
#ifdef CLBLAST_REF_CUBLAS
static StatusCode RunReference3(const Arguments<T> &args, BuffersCUDA<T> &buffers, Queue &) {
auto status = cublasXamax(reinterpret_cast<cublasHandle_t>(args.cublas_handle), args.n,
- buffers.scalar, args.imax_offset,
+ buffers.scalar_uint, args.imax_offset,
buffers.x_vec, args.x_offset, args.x_inc);
if (status == CUBLAS_STATUS_SUCCESS) { return StatusCode::kSuccess; } else { return StatusCode::kUnknownError; }
}
@@ -123,8 +123,15 @@ class TestXamax {
// 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.scalar_size, static_cast<T>(0));
- buffers.scalar.Read(queue, args.scalar_size, result);
+ std::vector<unsigned int> result_uint(args.scalar_size, 0);
+ buffers.scalar_uint.Read(queue, args.scalar_size, result_uint);
+ // The result is an integer. However, since the test infrastructure assumes results of
+ // type 'T' (float/double/float2/double2/half), we store the results into T instead.
+ // The values might then become meaningless, but a comparison for testing should still
+ // be valid to verify correctness.
+ auto result_as_T = static_cast<T>(result_uint[0]);
+ std::vector<T> result(args.scalar_size);
+ result[0] = result_as_T;
return result;
}
diff --git a/test/routines/levelx/xcol2im.hpp b/test/routines/levelx/xcol2im.hpp
index c740e4c7..cdac3493 100644
--- a/test/routines/levelx/xcol2im.hpp
+++ b/test/routines/levelx/xcol2im.hpp
@@ -201,7 +201,8 @@ StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &bu
auto a_buffer2 = HalfToFloatBuffer(buffers_host.a_mat);
auto b_buffer2 = HalfToFloatBuffer(buffers_host.b_mat);
auto dummy = std::vector<float>(0);
- auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy};
+ auto dummy_uint = std::vector<unsigned int>(0);
+ auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy, dummy_uint};
auto args2 = Arguments<float>();
args2.a_size = args.a_size; args2.b_size = args.b_size;
args2.kernel_mode = args.kernel_mode;
diff --git a/test/routines/levelx/xconvgemm.hpp b/test/routines/levelx/xconvgemm.hpp
index 786bb733..9085d927 100644
--- a/test/routines/levelx/xconvgemm.hpp
+++ b/test/routines/levelx/xconvgemm.hpp
@@ -230,7 +230,8 @@ StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &bu
auto b_buffer2 = HalfToFloatBuffer(buffers_host.b_mat);
auto c_buffer2 = HalfToFloatBuffer(buffers_host.c_mat);
auto dummy = std::vector<float>(0);
- auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, c_buffer2, dummy, dummy};
+ auto dummy_uint = std::vector<unsigned int>(0);
+ auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, c_buffer2, dummy, dummy, dummy_uint};
auto args2 = Arguments<float>();
args2.a_size = args.a_size; args2.b_size = args.b_size; args2.c_size = args.c_size;
args2.kernel_mode = args.kernel_mode;
diff --git a/test/routines/levelx/xhad.hpp b/test/routines/levelx/xhad.hpp
index 3e40de87..49629452 100644
--- a/test/routines/levelx/xhad.hpp
+++ b/test/routines/levelx/xhad.hpp
@@ -39,7 +39,8 @@ StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &bu
auto y_buffer2 = HalfToFloatBuffer(buffers_host.y_vec);
auto c_buffer2 = HalfToFloatBuffer(buffers_host.c_mat);
auto dummy = std::vector<float>(0);
- auto buffers2 = BuffersHost<float>{x_buffer2, y_buffer2, dummy, dummy, c_buffer2, dummy, dummy};
+ auto dummy_uint = std::vector<unsigned int>(0);
+ auto buffers2 = BuffersHost<float>{x_buffer2, y_buffer2, dummy, dummy, c_buffer2, dummy, dummy, dummy_uint};
auto args2 = Arguments<float>();
args2.x_size = args.x_size; args2.y_size = args.y_size; args2.c_size = args.c_size;
args2.x_inc = args.x_inc; args2.y_inc = args.y_inc; args2.n = args.n;
@@ -152,7 +153,7 @@ public:
// 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) {
+ static size_t GetResultIndex(const Arguments<T> &, const size_t id1, const size_t) {
return id1; // * args.z_inc + args.z_offset;
}
diff --git a/test/routines/levelx/xim2col.hpp b/test/routines/levelx/xim2col.hpp
index 2a5ebf8e..ab3b6a0f 100644
--- a/test/routines/levelx/xim2col.hpp
+++ b/test/routines/levelx/xim2col.hpp
@@ -200,7 +200,8 @@ StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &bu
auto a_buffer2 = HalfToFloatBuffer(buffers_host.a_mat);
auto b_buffer2 = HalfToFloatBuffer(buffers_host.b_mat);
auto dummy = std::vector<float>(0);
- auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy};
+ auto dummy_uint = std::vector<unsigned int>(0);
+ auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy, dummy_uint};
auto args2 = Arguments<float>();
args2.a_size = args.a_size; args2.b_size = args.b_size;
args2.kernel_mode = args.kernel_mode;
diff --git a/test/routines/levelx/xinvert.hpp b/test/routines/levelx/xinvert.hpp
index 126856ac..54905de1 100644
--- a/test/routines/levelx/xinvert.hpp
+++ b/test/routines/levelx/xinvert.hpp
@@ -108,7 +108,8 @@ StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &bu
auto a_buffer2 = HalfToFloatBuffer(buffers_host.a_mat);
auto b_buffer2 = HalfToFloatBuffer(buffers_host.b_mat);
auto dummy = std::vector<float>(0);
- auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy};
+ auto dummy_uint = std::vector<unsigned int>(0);
+ auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy, dummy_uint};
auto args2 = Arguments<float>();
args2.a_size = args.a_size; args2.b_size = args.b_size;
args2.a_ld = args.a_ld; args2.m = args.m; args2.n = args.n;
diff --git a/test/routines/levelx/xomatcopy.hpp b/test/routines/levelx/xomatcopy.hpp
index 4a93b29d..a4d98e79 100644
--- a/test/routines/levelx/xomatcopy.hpp
+++ b/test/routines/levelx/xomatcopy.hpp
@@ -59,7 +59,8 @@ StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &bu
auto a_buffer2 = HalfToFloatBuffer(buffers_host.a_mat);
auto b_buffer2 = HalfToFloatBuffer(buffers_host.b_mat);
auto dummy = std::vector<float>(0);
- auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy};
+ auto dummy_uint = std::vector<unsigned int>(0);
+ auto buffers2 = BuffersHost<float>{dummy, dummy, a_buffer2, b_buffer2, dummy, dummy, dummy, dummy_uint};
auto args2 = Arguments<float>();
args2.a_size = args.a_size; args2.b_size = args.b_size;
args2.a_ld = args.a_ld; args2.b_ld = args.b_ld; args2.m = args.m; args2.n = args.n;
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-02 07:54:09 +0000