summaryrefslogtreecommitdiff
path: root/src/kernels/level1/xdot.opencl
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2016-09-13 21:14:51 +0200
committerGitHub <noreply@github.com>2016-09-13 21:14:51 +0200
commitf07ac22f5b57d22756d779d2e53620f988d786ee (patch)
treee8bcbc331683ca6fd807f5a5b83bb05c6e6fed69 /src/kernels/level1/xdot.opencl
parent7c13bacf129291e3e295ecb6e833788477085fa0 (diff)
parent4b94afda941a86f363064ff02f97e21eb9618794 (diff)
Merge pull request #99 from CNugteren/development
Update to version 0.9.0
Diffstat (limited to 'src/kernels/level1/xdot.opencl')
-rw-r--r--src/kernels/level1/xdot.opencl16
1 files changed, 8 insertions, 8 deletions
diff --git a/src/kernels/level1/xdot.opencl b/src/kernels/level1/xdot.opencl
index e13eb3c1..02f04ea7 100644
--- a/src/kernels/level1/xdot.opencl
+++ b/src/kernels/level1/xdot.opencl
@@ -30,11 +30,11 @@ R"(
// =================================================================================================
// The main reduction kernel, performing the multiplication and the majority of the sum operation
-__attribute__((reqd_work_group_size(WGS1, 1, 1)))
-__kernel void Xdot(const int n,
- const __global real* restrict xgm, const int x_offset, const int x_inc,
- const __global real* restrict ygm, const int y_offset, const int y_inc,
- __global real* output, const int do_conjugate) {
+__kernel __attribute__((reqd_work_group_size(WGS1, 1, 1)))
+void Xdot(const int n,
+ const __global real* restrict xgm, const int x_offset, const int x_inc,
+ const __global real* restrict ygm, const int y_offset, const int y_inc,
+ __global real* output, const int do_conjugate) {
__local real lm[WGS1];
const int lid = get_local_id(0);
const int wgid = get_group_id(0);
@@ -73,9 +73,9 @@ __kernel void Xdot(const int n,
// The epilogue reduction kernel, performing the final bit of the sum operation. This kernel has to
// be launched with a single workgroup only.
-__attribute__((reqd_work_group_size(WGS2, 1, 1)))
-__kernel void XdotEpilogue(const __global real* restrict input,
- __global real* dot, const int dot_offset) {
+__kernel __attribute__((reqd_work_group_size(WGS2, 1, 1)))
+void XdotEpilogue(const __global real* restrict input,
+ __global real* dot, const int dot_offset) {
__local real lm[WGS2];
const int lid = get_local_id(0);