diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2018-03-22 21:01:02 +0100 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2018-03-22 21:01:02 +0100 |
commit | 9fb6550dd02c54fafbb03e20516a394d9cd63f3f (patch) | |
tree | 143bd103186eb0d3fd9672ab8a41ebb076d1107d /src/tuning/kernels/transpose_fast.cpp | |
parent | 7a2371213bb900c8c726c346f85920199f424d82 (diff) |
Added the OpenCL local memory size constraint to the tuners
Diffstat (limited to 'src/tuning/kernels/transpose_fast.cpp')
-rw-r--r-- | src/tuning/kernels/transpose_fast.cpp | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/src/tuning/kernels/transpose_fast.cpp b/src/tuning/kernels/transpose_fast.cpp index 6b37a31d..024f7385 100644 --- a/src/tuning/kernels/transpose_fast.cpp +++ b/src/tuning/kernels/transpose_fast.cpp @@ -22,11 +22,11 @@ using double2 = clblast::double2; int main(int argc, char *argv[]) { const auto command_line_args = clblast::RetrieveCommandLineArguments(argc, argv); switch(clblast::GetPrecision(command_line_args)) { - case clblast::Precision::kHalf: clblast::Tuner<half>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<half>, clblast::TransposeTestValidArguments<half>, clblast::TransposeSetConstraints, clblast::TransposeSetArguments<half>); break; - case clblast::Precision::kSingle: clblast::Tuner<float>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<float>, clblast::TransposeTestValidArguments<float>, clblast::TransposeSetConstraints, clblast::TransposeSetArguments<float>); break; - case clblast::Precision::kDouble: clblast::Tuner<double>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<double>, clblast::TransposeTestValidArguments<double>, clblast::TransposeSetConstraints, clblast::TransposeSetArguments<double>); break; - case clblast::Precision::kComplexSingle: clblast::Tuner<float2>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<float2>, clblast::TransposeTestValidArguments<float2>, clblast::TransposeSetConstraints, clblast::TransposeSetArguments<float2>); break; - case clblast::Precision::kComplexDouble: clblast::Tuner<double2>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<double2>, clblast::TransposeTestValidArguments<double2>, clblast::TransposeSetConstraints, clblast::TransposeSetArguments<double2>); break; + case clblast::Precision::kHalf: clblast::Tuner<half>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<half>, clblast::TransposeTestValidArguments<half>, clblast::TransposeSetConstraints, clblast::TransposeComputeLocalMemSize<half>, clblast::TransposeSetArguments<half>); break; + case clblast::Precision::kSingle: clblast::Tuner<float>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<float>, clblast::TransposeTestValidArguments<float>, clblast::TransposeSetConstraints, clblast::TransposeComputeLocalMemSize<float>, clblast::TransposeSetArguments<float>); break; + case clblast::Precision::kDouble: clblast::Tuner<double>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<double>, clblast::TransposeTestValidArguments<double>, clblast::TransposeSetConstraints, clblast::TransposeComputeLocalMemSize<double>, clblast::TransposeSetArguments<double>); break; + case clblast::Precision::kComplexSingle: clblast::Tuner<float2>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<float2>, clblast::TransposeTestValidArguments<float2>, clblast::TransposeSetConstraints, clblast::TransposeComputeLocalMemSize<float2>, clblast::TransposeSetArguments<float2>); break; + case clblast::Precision::kComplexDouble: clblast::Tuner<double2>(argc, argv, 0, clblast::TransposeGetTunerDefaults, clblast::TransposeGetTunerSettings<double2>, clblast::TransposeTestValidArguments<double2>, clblast::TransposeSetConstraints, clblast::TransposeComputeLocalMemSize<double2>, clblast::TransposeSetArguments<double2>); break; } return 0; } |