diff options
| author | Cedric Nugteren <web@cedricnugteren.nl> | 2019-05-08 22:01:56 +0200 |
|---|---|---|
| committer | Cedric Nugteren <web@cedricnugteren.nl> | 2019-05-08 22:01:56 +0200 |
| commit | 9cbffc9b7cbcfa6d6c8a048c45c1ad52d22effb3 (patch) | |
| tree | 26b0bbb1f887e49d2eb02658322bf0624e9e68a8 | |
| parent | c5a82f6978b3c459d70a04f0bf3904b424c867b5 (diff) | |
Changed back to cl_intel_subgroups as suggested
| -rw-r--r-- | src/kernels/level3/xgemm_part1.opencl | 2 | ||||
| -rw-r--r-- | src/utilities/compile.cpp | 1 | ||||
| -rw-r--r-- | src/utilities/utilities.hpp | 1 |
3 files changed, 1 insertions, 3 deletions
diff --git a/src/kernels/level3/xgemm_part1.opencl b/src/kernels/level3/xgemm_part1.opencl index 306280bc..9d46ab7e 100644 --- a/src/kernels/level3/xgemm_part1.opencl +++ b/src/kernels/level3/xgemm_part1.opencl @@ -127,7 +127,7 @@ R"( // Intel subgroups (https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_subgroups.html) #if USE_SUBGROUP_SHUFFLING == 1 && SUBGROUP_SHUFFLING_INTEL == 1 - #pragma OPENCL EXTENSION cl_intel_device_side_avc_motion_estimation: enable + #pragma OPENCL EXTENSION cl_intel_subgroups: enable #define SUBGROUP_SIZE 8 // Assumes subgroup size is always 8 on Intel GPUs #endif diff --git a/src/utilities/compile.cpp b/src/utilities/compile.cpp index 46cef1d7..00cb90cb 100644 --- a/src/utilities/compile.cpp +++ b/src/utilities/compile.cpp @@ -60,7 +60,6 @@ std::shared_ptr<Program> CompileFromSource( // For Intel GPUs with subgroup support, use subgroup shuffling. if (device.IsGPU() && device.HasExtension(kKhronosIntelSubgroups) && - device.HasExtension(kKhronosIntelAvcMotionEstimation) && (precision == Precision::kSingle || precision == Precision::kHalf)) { header_string += "#define USE_SUBGROUP_SHUFFLING 1\n"; header_string += "#define SUBGROUP_SHUFFLING_INTEL 1\n"; diff --git a/src/utilities/utilities.hpp b/src/utilities/utilities.hpp index 77221277..23486d35 100644 --- a/src/utilities/utilities.hpp +++ b/src/utilities/utilities.hpp @@ -48,7 +48,6 @@ using double2 = std::complex<double>; const std::string kKhronosAttributesAMD = "cl_amd_device_attribute_query"; const std::string kKhronosAttributesNVIDIA = "cl_nv_device_attribute_query"; const std::string kKhronosIntelSubgroups = "cl_intel_subgroups"; -const std::string kKhronosIntelAvcMotionEstimation = "cl_intel_device_side_avc_motion_estimation"; // Catched an unknown error constexpr auto kUnknownError = -999; |