summaryrefslogtreecommitdiff
path: root/test/wrapper_clblas.h
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2016-03-25 10:30:38 +0100
committerCedric Nugteren <web@cedricnugteren.nl>2016-03-25 10:30:38 +0100
commit1d5a702d9d31afa320a15ed9fa79471aec314f4a (patch)
tree52a073c0909b23c8e2c29e467c19c81dbcf34f5a /test/wrapper_clblas.h
parent3876096c30ad4eed5769dbc88dbfe75b7571718a (diff)
Added prototypes for ScNRM2/DzNRM2 routines
Diffstat (limited to 'test/wrapper_clblas.h')
-rw-r--r--test/wrapper_clblas.h32
1 files changed, 31 insertions, 1 deletions
diff --git a/test/wrapper_clblas.h b/test/wrapper_clblas.h
index 501f0bc5..37d9eee5 100644
--- a/test/wrapper_clblas.h
+++ b/test/wrapper_clblas.h
@@ -350,7 +350,7 @@ clblasStatus clblasXdotc<double2>(const size_t n,
num_queues, queues, num_wait_events, wait_events, events);
}
-// Forwards the clBLAS calls for SNRM2/DNRM2
+// Forwards the clBLAS calls for SNRM2/DNRM2/ScNRM2/DzNRM2
template <typename T>
clblasStatus clblasXnrm2(const size_t n,
cl_mem nrm2_buffer, const size_t nrm2_offset,
@@ -387,6 +387,36 @@ clblasStatus clblasXnrm2<double>(const size_t n,
scratch_buffer(),
num_queues, queues, num_wait_events, wait_events, events);
}
+template <>
+clblasStatus clblasXnrm2<float2>(const size_t n,
+ cl_mem nrm2_buffer, const size_t nrm2_offset,
+ const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+ cl_uint num_queues, cl_command_queue *queues,
+ cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+ auto queue = Queue(queues[0]);
+ auto context = queue.GetContext();
+ auto scratch_buffer = Buffer<float2>(context, n*x_inc + x_offset);
+ return clblasScnrm2(n,
+ nrm2_buffer, nrm2_offset,
+ x_buffer, x_offset, static_cast<int>(x_inc),
+ scratch_buffer(),
+ num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXnrm2<double2>(const size_t n,
+ cl_mem nrm2_buffer, const size_t nrm2_offset,
+ const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+ cl_uint num_queues, cl_command_queue *queues,
+ cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+ auto queue = Queue(queues[0]);
+ auto context = queue.GetContext();
+ auto scratch_buffer = Buffer<double2>(context, n*x_inc + x_offset);
+ return clblasDznrm2(n,
+ nrm2_buffer, nrm2_offset,
+ x_buffer, x_offset, static_cast<int>(x_inc),
+ scratch_buffer(),
+ num_queues, queues, num_wait_events, wait_events, events);
+}
// =================================================================================================
// BLAS level-2 (matrix-vector) routines