summaryrefslogtreecommitdiff
path: root/test/wrapper_clblas.h
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2016-03-25 10:00:40 +0100
committerCedric Nugteren <web@cedricnugteren.nl>2016-03-25 10:00:40 +0100
commit3876096c30ad4eed5769dbc88dbfe75b7571718a (patch)
treef902f2ee43df982152ada5d8d8d0573653cef58a /test/wrapper_clblas.h
parent49822c8ead3313e88a08f31162870e88f8ad2bb5 (diff)
Added prototypes for SNRM2/DNRM2 routines
Diffstat (limited to 'test/wrapper_clblas.h')
-rw-r--r--test/wrapper_clblas.h38
1 files changed, 38 insertions, 0 deletions
diff --git a/test/wrapper_clblas.h b/test/wrapper_clblas.h
index 23a02a45..501f0bc5 100644
--- a/test/wrapper_clblas.h
+++ b/test/wrapper_clblas.h
@@ -350,6 +350,44 @@ clblasStatus clblasXdotc<double2>(const size_t n,
num_queues, queues, num_wait_events, wait_events, events);
}
+// Forwards the clBLAS calls for SNRM2/DNRM2
+template <typename T>
+clblasStatus clblasXnrm2(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);
+template <>
+clblasStatus clblasXnrm2<float>(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<float>(context, n*x_inc + x_offset);
+ return clblasSnrm2(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<double>(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<double>(context, n*x_inc + x_offset);
+ return clblasDnrm2(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
// =================================================================================================