gpu_nested_parallelization_pass.h

Go to the documentation of this file.

#pragma once
 
#include <vector>
#include "sdfg/passes/pass.h"
#include "sdfg/structured_control_flow/map.h"
 
namespace sdfg {
namespace passes {
 
enum class GPUTarget {
    CUDA,
    ROCM,
};

class GPUNestedParallelizationPass : public Pass {
private:
    const std::vector<structured_control_flow::StructuredLoop*>& loops_;
    GPUTarget target_;
    size_t block_size_;
 
public:
    GPUNestedParallelizationPass(
        const std::vector<structured_control_flow::StructuredLoop*>& loops, GPUTarget target, size_t block_size
    );
    ~GPUNestedParallelizationPass() override = default;
 
    bool run_pass(builder::StructuredSDFGBuilder& builder, analysis::AnalysisManager& analysis_manager) override;
 
    std::string name() override {
        return "GPUNestedParallelizationPass";
    }
};
 
} // namespace passes
} // namespace sdfg