summaryrefslogtreecommitdiff
path: root/src/utilities/compile.cpp
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2018-09-15 21:47:04 +0200
committerGitHub <noreply@github.com>2018-09-15 21:47:04 +0200
commitc163868e1822a97750b4380f0d9cdd38369f9f0b (patch)
treefbdbc5b5697be8cf4237cc11f87d0b1649f4190d /src/utilities/compile.cpp
parent91dbd580ab2f5d2363d51ba4e3fc9735f1c7a937 (diff)
parent0f6dd01e513db036191786aed3d03a77e2e8c5dc (diff)
Merge pull request #318 from CNugteren/CLBlast-315-preprocessor-gemmk1-issue
Fixed pre-processor issues with the new GEMMK=1 kernel
Diffstat (limited to 'src/utilities/compile.cpp')
-rw-r--r--src/utilities/compile.cpp3
1 files changed, 2 insertions, 1 deletions
diff --git a/src/utilities/compile.cpp b/src/utilities/compile.cpp
index 835f54b4..00cb90cb 100644
--- a/src/utilities/compile.cpp
+++ b/src/utilities/compile.cpp
@@ -59,7 +59,8 @@ std::shared_ptr<Program> CompileFromSource(
}
// For Intel GPUs with subgroup support, use subgroup shuffling.
- if (device.IsGPU() && device.HasExtension(kKhronosIntelSubgroups)) {
+ if (device.IsGPU() && device.HasExtension(kKhronosIntelSubgroups) &&
+ (precision == Precision::kSingle || precision == Precision::kHalf)) {
header_string += "#define USE_SUBGROUP_SHUFFLING 1\n";
header_string += "#define SUBGROUP_SHUFFLING_INTEL 1\n";
}