summaryrefslogtreecommitdiff
path: root/src/utilities
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2018-09-16 20:01:18 +0200
committerCedric Nugteren <web@cedricnugteren.nl>2018-09-16 20:01:18 +0200
commit83ba3d4b7ba3a9cb5fbd2c1ad2bb14b2addd39fb (patch)
tree58900a63158d08e76342b46372fcc59015b4d3ca /src/utilities
parentb7d833901213d03fe5e7f10c15741f55c6c1eb54 (diff)
parentc163868e1822a97750b4380f0d9cdd38369f9f0b (diff)
Merge branch 'master' into convgemm_multi_kernel
Diffstat (limited to 'src/utilities')
-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";
}