summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--include/clblast.h32
-rw-r--r--samples/tuning_api.cpp4
-rw-r--r--src/tuning/tuning_api.cpp126
3 files changed, 146 insertions, 16 deletions
diff --git a/include/clblast.h b/include/clblast.h
index d6118e19..c5d2ddc2 100644
--- a/include/clblast.h
+++ b/include/clblast.h
@@ -706,8 +706,36 @@ StatusCode PUBLIC_API OverrideParameters(const cl_device_id device, const std::s
// =================================================================================================
template <typename T>
-StatusCode PUBLIC_API TuneCopyMatrixFast(cl_command_queue* queue, const size_t m, const size_t n,
- const double fraction, std::unordered_map<std::string,size_t> &parameters);
+StatusCode PUBLIC_API TuneXaxpy(cl_command_queue* queue, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
+
+template <typename T>
+StatusCode PUBLIC_API TuneXdot(cl_command_queue* queue, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
+
+template <typename T>
+StatusCode PUBLIC_API TuneXgemv(cl_command_queue* queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
+
+template <typename T>
+StatusCode PUBLIC_API TuneXger(cl_command_queue* queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
+
+template <typename T>
+StatusCode PUBLIC_API TuneCopy(cl_command_queue* queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
+
+template <typename T>
+StatusCode PUBLIC_API TunePad(cl_command_queue* queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
+
+template <typename T>
+StatusCode PUBLIC_API TuneTranspose(cl_command_queue* queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
+
+template <typename T>
+StatusCode PUBLIC_API TunePadtranspose(cl_command_queue* queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters);
// =================================================================================================
diff --git a/samples/tuning_api.cpp b/samples/tuning_api.cpp
index 3c9ba876..f92b6909 100644
--- a/samples/tuning_api.cpp
+++ b/samples/tuning_api.cpp
@@ -60,10 +60,10 @@ int main() {
printf("Starting the tuning...\n");
std::unordered_map<std::string,size_t> parameters;
auto queue_plain = queue();
- auto status = clblast::TuneCopyMatrixFast<float>(&queue_plain, m, n, fraction, parameters);
+ auto status = clblast::TuneCopy<float>(&queue_plain, m, n, fraction, parameters);
// Tuning completed. See "clblast.h" for status codes (0 -> success).
- printf("Completed TuneCopyMatrixFast with status %d (0 == OK), found parameters:\n", static_cast<int>(status));
+ printf("Completed TuneCopy with status %d (0 == OK), found parameters:\n", static_cast<int>(status));
for (const auto parameter: parameters) {
printf("> %s = %zu\n", parameter.first.c_str(), parameter.second);
}
diff --git a/src/tuning/tuning_api.cpp b/src/tuning/tuning_api.cpp
index 94a9a367..61cb0389 100644
--- a/src/tuning/tuning_api.cpp
+++ b/src/tuning/tuning_api.cpp
@@ -26,23 +26,125 @@ namespace clblast {
// =================================================================================================
template <typename T>
-StatusCode TuneCopyMatrixFast(RawCommandQueue * queue, const size_t m, const size_t n,
- const double fraction, std::unordered_map<std::string,size_t> &parameters) {
- auto args = Arguments<T>();
- args.m = m;
- args.n = n;
- args.fraction = fraction;
+StatusCode TuneXaxpy(RawCommandQueue * queue, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.n = n;
auto queue_cpp = Queue(*queue);
return TunerAPI<T>(queue_cpp, args, 0, GetTunerDefaults, GetTunerSettings<T>,
TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
}
+template StatusCode TuneXaxpy<half>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXaxpy<float>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXaxpy<double>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXaxpy<float2>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXaxpy<double2>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
-// Compiles the above
-template StatusCode TuneCopyMatrixFast<half>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
-template StatusCode TuneCopyMatrixFast<float>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
-template StatusCode TuneCopyMatrixFast<double>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
-template StatusCode TuneCopyMatrixFast<float2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
-template StatusCode TuneCopyMatrixFast<double2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template <typename T>
+StatusCode TuneXdot(RawCommandQueue * queue, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.n = n;
+ auto queue_cpp = Queue(*queue);
+ auto status = TunerAPI<T>(queue_cpp, args, 1, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+ if (status != StatusCode::kSuccess) { return status; }
+ return TunerAPI<T>(queue_cpp, args, 2, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+}
+template StatusCode TuneXdot<half>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXdot<float>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXdot<double>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXdot<float2>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXdot<double2>(RawCommandQueue*, const size_t, const double, std::unordered_map<std::string,size_t>&);
+
+template <typename T>
+StatusCode TuneXgemv(RawCommandQueue * queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.m = m; args.n = n;
+ auto queue_cpp = Queue(*queue);
+ auto status = TunerAPI<T>(queue_cpp, args, 1, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+ if (status != StatusCode::kSuccess) { return status; }
+ status = TunerAPI<T>(queue_cpp, args, 2, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+ if (status != StatusCode::kSuccess) { return status; }
+ return TunerAPI<T>(queue_cpp, args, 3, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+}
+template StatusCode TuneXgemv<half>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXgemv<float>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXgemv<double>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXgemv<float2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXgemv<double2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+
+template <typename T>
+StatusCode TuneXger(RawCommandQueue * queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.m = m; args.n = n;
+ auto queue_cpp = Queue(*queue);
+ return TunerAPI<T>(queue_cpp, args, 0, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+}
+template StatusCode TuneXger<half>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXger<float>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXger<double>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXger<float2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneXger<double2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+
+template <typename T>
+StatusCode TuneCopy(RawCommandQueue * queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.m = m; args.n = n;
+ auto queue_cpp = Queue(*queue);
+ return TunerAPI<T>(queue_cpp, args, 0, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+}
+template StatusCode TuneCopy<half>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneCopy<float>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneCopy<double>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneCopy<float2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneCopy<double2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+
+template <typename T>
+StatusCode TunePad(RawCommandQueue * queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.m = m; args.n = n;
+ auto queue_cpp = Queue(*queue);
+ return TunerAPI<T>(queue_cpp, args, 0, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+}
+template StatusCode TunePad<half>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePad<float>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePad<double>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePad<float2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePad<double2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+
+template <typename T>
+StatusCode TuneTranspose(RawCommandQueue * queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.m = m; args.n = n;
+ auto queue_cpp = Queue(*queue);
+ return TunerAPI<T>(queue_cpp, args, 0, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+}
+template StatusCode TuneTranspose<half>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneTranspose<float>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneTranspose<double>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneTranspose<float2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TuneTranspose<double2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+
+template <typename T>
+StatusCode TunePadtranspose(RawCommandQueue * queue, const size_t m, const size_t n,
+ const double fraction, std::unordered_map<std::string,size_t> &parameters) {
+ auto args = Arguments<T>(); args.fraction = fraction; args.m = m; args.n = n;
+ auto queue_cpp = Queue(*queue);
+ return TunerAPI<T>(queue_cpp, args, 0, GetTunerDefaults, GetTunerSettings<T>,
+ TestValidArguments<T>, SetConstraints, SetArguments<T>, parameters);
+}
+template StatusCode TunePadtranspose<half>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePadtranspose<float>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePadtranspose<double>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePadtranspose<float2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
+template StatusCode TunePadtranspose<double2>(RawCommandQueue*, const size_t, const size_t, const double, std::unordered_map<std::string,size_t>&);
// =================================================================================================