From fa84ac36f23c1aebb5facf946b21d8c0f1a4a46d Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Sat, 28 Jul 2018 16:01:03 +0200 Subject: The tuners now also check for valid local thread configurations and skip invalid ones completely, saving compilation time --- src/tuning/configurations.hpp | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-) (limited to 'src/tuning/configurations.hpp') diff --git a/src/tuning/configurations.hpp b/src/tuning/configurations.hpp index faa5498f..4b9ba93f 100644 --- a/src/tuning/configurations.hpp +++ b/src/tuning/configurations.hpp @@ -50,6 +50,9 @@ struct LocalMemSizeInfo { // function to find all configurations. It also applies the user-defined constraints within. std::vector SetConfigurations(const Device& device, const std::vector parameters, + const std::vector& local_size_base, + const TransformVector& mul_local_config, + const TransformVector& div_local_config, const Constraints& constraints, const LocalMemSizeInfo& local_mem_size_info); @@ -58,11 +61,16 @@ std::vector SetConfigurations(const Device& device, // At the end of each chain (when all parameters are considered), the function stores the result // into the configuration list. void PopulateConfigurations(const std::vector ¶meters, + const std::vector local_size_base, + const TransformVector& mul_local_config, + const TransformVector& div_local_config, const size_t index, const Configuration &config, std::vector &configuration, const size_t local_mem_max, const Constraints& constraints, - const LocalMemSizeInfo& local_mem_size_info); + const LocalMemSizeInfo& local_mem_size_info, + const std::vector& max_work_item_sizes, + const size_t max_work_group_size); // Loops over all user-defined constraints to check whether or not the configuration is valid. // Assumes initially all configurations are valid, then returns false if one of the constraints has @@ -71,7 +79,12 @@ void PopulateConfigurations(const std::vector ¶meters, bool ValidConfiguration(const Configuration &config, const size_t local_mem_max, const Constraints& constraints, - const LocalMemSizeInfo& local_mem_size_info); + const LocalMemSizeInfo& local_mem_size_info, + const std::vector local_size_base, + const TransformVector& mul_local_config, + const TransformVector& div_local_config, + const std::vector& max_work_item_sizes, + const size_t max_work_group_size); // Processes multipliers and dividers to obtain the final thread configuration std::vector SetThreadConfiguration(const Configuration& config, -- cgit v1.2.3