23906044628

Committed 02 Apr 2026 02:40PM UTC coverage: 64.553% (+0.08%) from 64.474%

Build # 23906044628

Build Type

Pull #632

github

Committed by

web-flow

Commit Message

Merge b2698daab into 3125b927b

Pull Request Pull Request #632: Separate can_be_applied and apply for GPUTilling during Loop Scheduling

Run Details

282 of 348 new or added lines in 16 files covered. (81.03%)

29 existing lines in 10 files now uncovered.

28998 of 44921 relevant lines covered (64.55%)

453.01 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

79.49

/opt/src/passes/offloading/gpu_nested_parallelization_pass.cpp

#include "sdfg/passes/offloading/gpu_nested_parallelization_pass.h"

#include "sdfg/analysis/loop_analysis.h"
#include "sdfg/transformations/offloading/cuda_parallelize_nested_map.h"
#include "sdfg/transformations/offloading/rocm_parallelize_nested_map.h"

namespace sdfg {
namespace passes {

GPUNestedParallelizationPass::GPUNestedParallelizationPass(
    const std::vector<structured_control_flow::Map*>& maps, GPUTarget target, size_t block_size
)
    : maps_(maps), target_(target), block_size_(block_size) {}

bool GPUNestedParallelizationPass::
    run_pass(builder::StructuredSDFGBuilder& builder, analysis::AnalysisManager& analysis_manager) {
    if (maps_.empty()) {
        return false;
    }

    auto& loop_analysis = analysis_manager.get<analysis::LoopAnalysis>();

    // Phase 1: Collect all applicable nested maps
    std::vector<structured_control_flow::Map*> candidates;

    for (auto* map : maps_) {
        auto descendants = loop_analysis.descendants(map);
        for (auto* descendant : descendants) {
            if (auto* nested_map = dynamic_cast<structured_control_flow::Map*>(descendant)) {
                bool applicable = false;
                if (target_ == GPUTarget::CUDA) {
                    transformations::CUDAParallelizeNestedMap transform(*nested_map, block_size_);
                    applicable = transform.can_be_applied(builder, analysis_manager);
                } else {
                    transformations::ROCMParallelizeNestedMap transform(*nested_map, block_size_);
                    applicable = transform.can_be_applied(builder, analysis_manager);
                }
                if (applicable) {
                    candidates.push_back(nested_map);
                }
            }
        }
    }

    if (candidates.empty()) {
        return false;
    }

    // Phase 2: Apply all parallelizations
    for (auto* nested_map : candidates) {
        if (target_ == GPUTarget::CUDA) {
            transformations::CUDAParallelizeNestedMap transform(*nested_map, block_size_);
            transform.apply(builder, analysis_manager);
        } else {
            transformations::ROCMParallelizeNestedMap transform(*nested_map, block_size_);
            transform.apply(builder, analysis_manager);
        }
    }

    return true;
}

} // namespace passes
} // namespace sdfg

1	#include "sdfg/passes/offloading/gpu_nested_parallelization_pass.h"
2
3	#include "sdfg/analysis/loop_analysis.h"
4	#include "sdfg/transformations/offloading/cuda_parallelize_nested_map.h"
5	#include "sdfg/transformations/offloading/rocm_parallelize_nested_map.h"
6
7	namespace sdfg {
8	namespace passes {
9
10	GPUNestedParallelizationPass::GPUNestedParallelizationPass(
11	const std::vector<structured_control_flow::Map*>& maps, GPUTarget target, size_t block_size
12	)
13	: maps_(maps), target_(target), block_size_(block_size) {}	3✔
14
15	bool GPUNestedParallelizationPass::
16	run_pass(builder::StructuredSDFGBuilder& builder, analysis::AnalysisManager& analysis_manager) {	3✔
17	if (maps_.empty()) {	3✔
NEW 18	return false;	×
NEW 19	}	×
20
21	auto& loop_analysis = analysis_manager.get<analysis::LoopAnalysis>();	3✔
22
23	// Phase 1: Collect all applicable nested maps
24	std::vector<structured_control_flow::Map*> candidates;	3✔
25
26	for (auto* map : maps_) {	4✔
27	auto descendants = loop_analysis.descendants(map);	4✔
28	for (auto* descendant : descendants) {	4✔
29	if (auto* nested_map = dynamic_cast<structured_control_flow::Map*>(descendant)) {	2✔
30	bool applicable = false;	2✔
31	if (target_ == GPUTarget::CUDA) {	2✔
32	transformations::CUDAParallelizeNestedMap transform(*nested_map, block_size_);	2✔
33	applicable = transform.can_be_applied(builder, analysis_manager);	2✔
34	} else {	2✔
NEW 35	transformations::ROCMParallelizeNestedMap transform(*nested_map, block_size_);	×
NEW 36	applicable = transform.can_be_applied(builder, analysis_manager);	×
NEW 37	}	×
38	if (applicable) {	2✔
39	candidates.push_back(nested_map);	2✔
40	}	2✔
41	}	2✔
42	}	2✔
43	}	4✔
44
45	if (candidates.empty()) {	3✔
46	return false;	1✔
47	}	1✔
48
49	// Phase 2: Apply all parallelizations
50	for (auto* nested_map : candidates) {	2✔
51	if (target_ == GPUTarget::CUDA) {	2✔
52	transformations::CUDAParallelizeNestedMap transform(*nested_map, block_size_);	2✔
53	transform.apply(builder, analysis_manager);	2✔
54	} else {	2✔
NEW 55	transformations::ROCMParallelizeNestedMap transform(*nested_map, block_size_);	×
NEW 56	transform.apply(builder, analysis_manager);	×
NEW 57	}	×
58	}	2✔
59
60	return true;	2✔
61	}	3✔
62
63	} // namespace passes
64	} // namespace sdfg

daisytuner / docc / 23906044628

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous