17637380013

Committed 11 Sep 2025 07:29AM UTC coverage: 59.755% (+0.6%) from 59.145%

Build # 17637380013

Build Type

push

github

Committed by

web-flow

Commit Message

New debug info (#210)

* initial draft

* update data structure and construction logic

* finalize DebugInfo draft

* fix tests

* Update serializer and fix tests

* fix append bug

* update data structure

* sdfg builder update

* const ref vectors

* update implementation and partial tests

* compiling state

* update serializer interface

* update dot test

* reset interface to debug_info in json to maintain compatibility with tools

* first review batch

* second batch of changes

* merge fixes

Run Details

777 of 1111 new or added lines in 46 files covered. (69.94%)

11 existing lines in 11 files now uncovered.

9755 of 16325 relevant lines covered (59.75%)

115.06 hits per line

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

0.0

/src/data_flow/library_nodes/math/ml/batch_normalization.cpp

#include "sdfg/data_flow/library_nodes/math/ml/batch_normalization.h"

#include "sdfg/analysis/analysis.h"
#include "sdfg/analysis/scope_analysis.h"
#include "sdfg/builder/structured_sdfg_builder.h"

namespace sdfg {
namespace math {
namespace ml {

BatchNormalizationNode::BatchNormalizationNode(
    size_t element_id,
    const DebugInfoRegion &debug_info,
    const graph::Vertex vertex,
    data_flow::DataFlowGraph &parent,
    int axis,
    const std::string &epsilon
)
    : MathNode(
          element_id,
          debug_info,
          vertex,
          parent,
          LibraryNodeType_BatchNormalization,
          {"Y"},
          {"X", "Scale", "B", "input_mean", "input_var"},
          data_flow::ImplementationType_NONE
      ),
      axis_(axis), epsilon_(epsilon) {}

void BatchNormalizationNode::validate(const Function &) const { /* TODO */ }

bool BatchNormalizationNode::expand(builder::StructuredSDFGBuilder &builder, analysis::AnalysisManager &analysis_manager) {
    auto &dataflow = this->get_parent();
    auto &block = static_cast<structured_control_flow::Block &>(*dataflow.get_parent());

    auto &scope_analysis = analysis_manager.get<analysis::ScopeAnalysis>();
    auto &parent = static_cast<structured_control_flow::Sequence &>(*scope_analysis.parent_scope(&block));
    int index = parent.index(block);
    auto &transition = parent.at(index).second;

    // Locate edges
    const data_flow::Memlet *iedge_input = nullptr;
    const data_flow::Memlet *iedge_scale = nullptr;
    const data_flow::Memlet *iedge_bias = nullptr;
    const data_flow::Memlet *iedge_mean = nullptr;
    const data_flow::Memlet *iedge_var = nullptr;
    const data_flow::Memlet *oedge_output = nullptr;
    for (const auto &edge : dataflow.in_edges(*this)) {
        if (edge.dst_conn() == "X") {
            iedge_input = &edge;
        } else if (edge.dst_conn() == "Scale") {
            iedge_scale = &edge;
        } else if (edge.dst_conn() == "B") {
            iedge_bias = &edge;
        } else if (edge.dst_conn() == "input_mean") {
            iedge_mean = &edge;
        } else if (edge.dst_conn() == "input_var") {
            iedge_var = &edge;
        }
    }
    for (const auto &edge : dataflow.out_edges(*this)) {
        if (edge.src_conn() == "Y") {
            oedge_output = &edge;
        }
    }
    if (!iedge_input || !iedge_scale || !iedge_bias || !iedge_mean || !iedge_var || !oedge_output) return false;

    std::string input_name = static_cast<const data_flow::AccessNode &>(iedge_input->src()).data();
    std::string scale_name = static_cast<const data_flow::AccessNode &>(iedge_scale->src()).data();
    std::string bias_name = static_cast<const data_flow::AccessNode &>(iedge_bias->src()).data();
    std::string mean_name = static_cast<const data_flow::AccessNode &>(iedge_mean->src()).data();
    std::string var_name = static_cast<const data_flow::AccessNode &>(iedge_var->src()).data();
    std::string output_name = static_cast<const data_flow::AccessNode &>(oedge_output->dst()).data();

    // Create new sequence before
    auto &new_sequence = builder.add_sequence_before(
        parent, block, transition.assignments(), builder.debug_info().get_region(block.debug_info().indices())
    );
    structured_control_flow::Sequence *last_scope = &new_sequence;

    // Create maps over output subset dims (parallel dims)
    data_flow::Subset domain_begin = oedge_output->begin_subset();
    data_flow::Subset domain_end = oedge_output->end_subset();

    std::vector<symbolic::Expression> loop_syms;
    structured_control_flow::Map *last_map = nullptr;
    for (size_t d = 0; d < domain_begin.size(); ++d) {
        std::string indvar_str = builder.find_new_name("_i");
        builder.add_container(indvar_str, types::Scalar(types::PrimitiveType::UInt64));
        auto indvar = symbolic::symbol(indvar_str);
        auto init = domain_begin[d];
        auto update = symbolic::add(indvar, symbolic::one());
        auto cond = symbolic::Lt(indvar, symbolic::add(domain_end[d], symbolic::one()));
        last_map = &builder.add_map(
            *last_scope,
            indvar,
            cond,
            init,
            update,
            structured_control_flow::ScheduleType_Sequential::create(),
            {},
            builder.subject().debug_info().get_region(block.debug_info().indices())
        );
        last_scope = &last_map->root();
        loop_syms.push_back(indvar);
    }

    // Create normalization block
    auto &norm_block = builder.add_block(*last_scope);

    // Create access nodes for normalization
    auto &input_access_norm = builder.add_access(norm_block, input_name);
    auto &scale_access_norm = builder.add_access(norm_block, scale_name);
    auto &bias_access_norm = builder.add_access(norm_block, bias_name);
    auto &mean_access_norm = builder.add_access(norm_block, mean_name);
    auto &var_access_norm = builder.add_access(norm_block, var_name);
    auto &output_access_norm = builder.add_access(norm_block, output_name);

    // Add epsilon to variance and compute standard deviation
    auto &add_epsilon_tasklet =
        builder.add_tasklet(norm_block, data_flow::TaskletCode::add, "_out", {"_in1", epsilon_});
    auto &var_eps_access = builder.add_access(norm_block, builder.find_new_name("_var_eps"));
    builder.add_computational_memlet(
        norm_block, var_access_norm, add_epsilon_tasklet, "_in1", loop_syms, iedge_var->base_type()
    );
    builder
        .add_computational_memlet(norm_block, add_epsilon_tasklet, "_out", var_eps_access, {}, iedge_var->base_type());

    auto &sqrt_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::sqrt, "_out", {"_in"});
    auto &std_dev_access = builder.add_access(norm_block, builder.find_new_name("_std_dev"));
    builder.add_computational_memlet(norm_block, var_eps_access, sqrt_tasklet, "_in", {}, iedge_var->base_type());
    builder.add_computational_memlet(norm_block, sqrt_tasklet, "_out", std_dev_access, {}, iedge_var->base_type());

    // Normalize: (x - mean) / std_dev
    auto &sub_norm_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::sub, "_out", {"_in1", "_in2"});
    auto &centered_access = builder.add_access(norm_block, builder.find_new_name("_centered"));
    builder.add_computational_memlet(
        norm_block, input_access_norm, sub_norm_tasklet, "_in1", loop_syms, iedge_input->base_type()
    );
    builder.add_computational_memlet(
        norm_block, mean_access_norm, sub_norm_tasklet, "_in2", loop_syms, iedge_mean->base_type()
    );
    builder
        .add_computational_memlet(norm_block, sub_norm_tasklet, "_out", centered_access, {}, iedge_input->base_type());

    auto &div_norm_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::div, "_out", {"_in1", "_in2"});
    auto &normalized_access = builder.add_access(norm_block, builder.find_new_name("_normalized"));
    builder
        .add_computational_memlet(norm_block, centered_access, div_norm_tasklet, "_in1", {}, iedge_input->base_type());
    builder
        .add_computational_memlet(norm_block, std_dev_access, div_norm_tasklet, "_in2", loop_syms, iedge_var->base_type());
    builder
        .add_computational_memlet(norm_block, div_norm_tasklet, "_out", normalized_access, {}, iedge_input->base_type());

    // Apply scale and bias: scale * normalized + bias
    auto &mul_scale_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::mul, "_out", {"_in1", "_in2"});
    auto &scaled_access = builder.add_access(norm_block, builder.find_new_name("_scaled"));
    builder
        .add_computational_memlet(norm_block, normalized_access, mul_scale_tasklet, "_in1", {}, iedge_input->base_type());
    builder.add_computational_memlet(
        norm_block, scale_access_norm, mul_scale_tasklet, "_in2", loop_syms, iedge_scale->base_type()
    );
    builder.add_computational_memlet(norm_block, mul_scale_tasklet, "_out", scaled_access, {}, iedge_input->base_type());

    auto &add_bias_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::add, "_out", {"_in1", "_in2"});
    builder.add_computational_memlet(norm_block, scaled_access, add_bias_tasklet, "_in1", {}, iedge_input->base_type());
    builder.add_computational_memlet(
        norm_block, bias_access_norm, add_bias_tasklet, "_in2", loop_syms, iedge_bias->base_type()
    );
    builder.add_computational_memlet(
        norm_block, add_bias_tasklet, "_out", output_access_norm, loop_syms, oedge_output->base_type()
    );

    // Cleanup old block
    builder.remove_memlet(block, *iedge_input);
    builder.remove_memlet(block, *iedge_scale);
    if (iedge_bias) {
        builder.remove_memlet(block, *iedge_bias);
    }
    builder.remove_memlet(block, *iedge_mean);
    builder.remove_memlet(block, *iedge_var);
    builder.remove_memlet(block, *oedge_output);
    builder.remove_node(block, *this);
    builder.remove_child(parent, index + 1);

    return true;
}

std::unique_ptr<data_flow::DataFlowNode> BatchNormalizationNode::
    clone(size_t element_id, const graph::Vertex vertex, data_flow::DataFlowGraph &parent) const {
    return std::unique_ptr<data_flow::DataFlowNode>(
        new BatchNormalizationNode(element_id, this->debug_info(), vertex, parent, axis_, epsilon_)
    );
}

nlohmann::json BatchNormalizationNodeSerializer::serialize(const data_flow::LibraryNode &library_node) {
    const BatchNormalizationNode &node = static_cast<const BatchNormalizationNode &>(library_node);
    nlohmann::json j;

    j["code"] = node.code().value();
    j["axis"] = node.axis();
    j["epsilon"] = node.epsilon();

    return j;
}

data_flow::LibraryNode &BatchNormalizationNodeSerializer::deserialize(
    const nlohmann::json &j, builder::StructuredSDFGBuilder &builder, structured_control_flow::Block &parent
) {
    auto code = j["code"].get<std::string>();
    if (code != LibraryNodeType_BatchNormalization.value()) {
        throw std::runtime_error("Invalid library node code");
    }

    sdfg::serializer::JSONSerializer serializer;
    DebugInfoRegion debug_info = serializer.json_to_debug_info_region(j["debug_info"], builder.debug_info());

    auto axis = j["axis"].get<int>();
    auto epsilon = j["epsilon"].get<std::string>();

    return builder.add_library_node<BatchNormalizationNode>(parent, debug_info, axis, epsilon);
}

} // namespace ml
} // namespace math
} // namespace sdfg

1	#include "sdfg/data_flow/library_nodes/math/ml/batch_normalization.h"
2
3	#include "sdfg/analysis/analysis.h"
4	#include "sdfg/analysis/scope_analysis.h"
5	#include "sdfg/builder/structured_sdfg_builder.h"
6
7	namespace sdfg {
8	namespace math {
9	namespace ml {
10
11	BatchNormalizationNode::BatchNormalizationNode(	×
12	size_t element_id,
13	const DebugInfoRegion &debug_info,
14	const graph::Vertex vertex,
15	data_flow::DataFlowGraph &parent,
16	int axis,
17	const std::string &epsilon
18	)
19	: MathNode(	×
20	element_id,	×
21	debug_info,	×
22	vertex,	×
23	parent,	×
24	LibraryNodeType_BatchNormalization,
25	{"Y"},	×
26	{"X", "Scale", "B", "input_mean", "input_var"},	×
27	data_flow::ImplementationType_NONE
28	),
29	axis_(axis), epsilon_(epsilon) {}	×
30
31	void BatchNormalizationNode::validate(const Function &) const { /* TODO */ }	×
32
33	bool BatchNormalizationNode::expand(builder::StructuredSDFGBuilder &builder, analysis::AnalysisManager &analysis_manager) {	×
34	auto &dataflow = this->get_parent();	×
35	auto &block = static_cast<structured_control_flow::Block &>(*dataflow.get_parent());	×
36
37	auto &scope_analysis = analysis_manager.get<analysis::ScopeAnalysis>();	×
38	auto &parent = static_cast<structured_control_flow::Sequence &>(*scope_analysis.parent_scope(&block));	×
39	int index = parent.index(block);	×
40	auto &transition = parent.at(index).second;	×
41
42	// Locate edges
43	const data_flow::Memlet *iedge_input = nullptr;	×
44	const data_flow::Memlet *iedge_scale = nullptr;	×
45	const data_flow::Memlet *iedge_bias = nullptr;	×
46	const data_flow::Memlet *iedge_mean = nullptr;	×
47	const data_flow::Memlet *iedge_var = nullptr;	×
48	const data_flow::Memlet *oedge_output = nullptr;	×
49	for (const auto &edge : dataflow.in_edges(*this)) {	×
50	if (edge.dst_conn() == "X") {	×
51	iedge_input = &edge;	×
52	} else if (edge.dst_conn() == "Scale") {	×
53	iedge_scale = &edge;	×
54	} else if (edge.dst_conn() == "B") {	×
55	iedge_bias = &edge;	×
56	} else if (edge.dst_conn() == "input_mean") {	×
57	iedge_mean = &edge;	×
58	} else if (edge.dst_conn() == "input_var") {	×
59	iedge_var = &edge;	×
60	}	×
61	}
62	for (const auto &edge : dataflow.out_edges(*this)) {	×
63	if (edge.src_conn() == "Y") {	×
64	oedge_output = &edge;	×
65	}	×
66	}
67	if (!iedge_input \|\| !iedge_scale \|\| !iedge_bias \|\| !iedge_mean \|\| !iedge_var \|\| !oedge_output) return false;	×
68
69	std::string input_name = static_cast<const data_flow::AccessNode &>(iedge_input->src()).data();	×
70	std::string scale_name = static_cast<const data_flow::AccessNode &>(iedge_scale->src()).data();	×
71	std::string bias_name = static_cast<const data_flow::AccessNode &>(iedge_bias->src()).data();	×
72	std::string mean_name = static_cast<const data_flow::AccessNode &>(iedge_mean->src()).data();	×
73	std::string var_name = static_cast<const data_flow::AccessNode &>(iedge_var->src()).data();	×
74	std::string output_name = static_cast<const data_flow::AccessNode &>(oedge_output->dst()).data();	×
75
76	// Create new sequence before
NEW 77	auto &new_sequence = builder.add_sequence_before(	×
NEW 78	parent, block, transition.assignments(), builder.debug_info().get_region(block.debug_info().indices())	×
79	);
UNCOV 80	structured_control_flow::Sequence *last_scope = &new_sequence;	×
81
82	// Create maps over output subset dims (parallel dims)
83	data_flow::Subset domain_begin = oedge_output->begin_subset();	×
84	data_flow::Subset domain_end = oedge_output->end_subset();	×
85
86	std::vector<symbolic::Expression> loop_syms;	×
87	structured_control_flow::Map *last_map = nullptr;	×
88	for (size_t d = 0; d < domain_begin.size(); ++d) {	×
89	std::string indvar_str = builder.find_new_name("_i");	×
90	builder.add_container(indvar_str, types::Scalar(types::PrimitiveType::UInt64));	×
91	auto indvar = symbolic::symbol(indvar_str);	×
92	auto init = domain_begin[d];	×
93	auto update = symbolic::add(indvar, symbolic::one());	×
94	auto cond = symbolic::Lt(indvar, symbolic::add(domain_end[d], symbolic::one()));	×
95	last_map = &builder.add_map(	×
96	*last_scope,	×
97	indvar,
98	cond,
99	init,
100	update,
101	structured_control_flow::ScheduleType_Sequential::create(),	×
102	{},	×
NEW 103	builder.subject().debug_info().get_region(block.debug_info().indices())	×
104	);
105	last_scope = &last_map->root();	×
106	loop_syms.push_back(indvar);	×
107	}	×
108
109	// Create normalization block
110	auto &norm_block = builder.add_block(*last_scope);	×
111
112	// Create access nodes for normalization
113	auto &input_access_norm = builder.add_access(norm_block, input_name);	×
114	auto &scale_access_norm = builder.add_access(norm_block, scale_name);	×
115	auto &bias_access_norm = builder.add_access(norm_block, bias_name);	×
116	auto &mean_access_norm = builder.add_access(norm_block, mean_name);	×
117	auto &var_access_norm = builder.add_access(norm_block, var_name);	×
118	auto &output_access_norm = builder.add_access(norm_block, output_name);	×
119
120	// Add epsilon to variance and compute standard deviation
121	auto &add_epsilon_tasklet =	×
122	builder.add_tasklet(norm_block, data_flow::TaskletCode::add, "_out", {"_in1", epsilon_});	×
123	auto &var_eps_access = builder.add_access(norm_block, builder.find_new_name("_var_eps"));	×
124	builder.add_computational_memlet(	×
125	norm_block, var_access_norm, add_epsilon_tasklet, "_in1", loop_syms, iedge_var->base_type()	×
126	);
127	builder	×
128	.add_computational_memlet(norm_block, add_epsilon_tasklet, "_out", var_eps_access, {}, iedge_var->base_type());	×
129
130	auto &sqrt_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::sqrt, "_out", {"_in"});	×
131	auto &std_dev_access = builder.add_access(norm_block, builder.find_new_name("_std_dev"));	×
132	builder.add_computational_memlet(norm_block, var_eps_access, sqrt_tasklet, "_in", {}, iedge_var->base_type());	×
133	builder.add_computational_memlet(norm_block, sqrt_tasklet, "_out", std_dev_access, {}, iedge_var->base_type());	×
134
135	// Normalize: (x - mean) / std_dev
136	auto &sub_norm_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::sub, "_out", {"_in1", "_in2"});	×
137	auto &centered_access = builder.add_access(norm_block, builder.find_new_name("_centered"));	×
138	builder.add_computational_memlet(	×
139	norm_block, input_access_norm, sub_norm_tasklet, "_in1", loop_syms, iedge_input->base_type()	×
140	);
141	builder.add_computational_memlet(	×
142	norm_block, mean_access_norm, sub_norm_tasklet, "_in2", loop_syms, iedge_mean->base_type()	×
143	);
144	builder	×
145	.add_computational_memlet(norm_block, sub_norm_tasklet, "_out", centered_access, {}, iedge_input->base_type());	×
146
147	auto &div_norm_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::div, "_out", {"_in1", "_in2"});	×
148	auto &normalized_access = builder.add_access(norm_block, builder.find_new_name("_normalized"));	×
149	builder	×
150	.add_computational_memlet(norm_block, centered_access, div_norm_tasklet, "_in1", {}, iedge_input->base_type());	×
151	builder	×
152	.add_computational_memlet(norm_block, std_dev_access, div_norm_tasklet, "_in2", loop_syms, iedge_var->base_type());	×
153	builder	×
154	.add_computational_memlet(norm_block, div_norm_tasklet, "_out", normalized_access, {}, iedge_input->base_type());	×
155
156	// Apply scale and bias: scale * normalized + bias
157	auto &mul_scale_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::mul, "_out", {"_in1", "_in2"});	×
158	auto &scaled_access = builder.add_access(norm_block, builder.find_new_name("_scaled"));	×
159	builder	×
160	.add_computational_memlet(norm_block, normalized_access, mul_scale_tasklet, "_in1", {}, iedge_input->base_type());	×
161	builder.add_computational_memlet(	×
162	norm_block, scale_access_norm, mul_scale_tasklet, "_in2", loop_syms, iedge_scale->base_type()	×
163	);
164	builder.add_computational_memlet(norm_block, mul_scale_tasklet, "_out", scaled_access, {}, iedge_input->base_type());	×
165
166	auto &add_bias_tasklet = builder.add_tasklet(norm_block, data_flow::TaskletCode::add, "_out", {"_in1", "_in2"});	×
167	builder.add_computational_memlet(norm_block, scaled_access, add_bias_tasklet, "_in1", {}, iedge_input->base_type());	×
168	builder.add_computational_memlet(	×
169	norm_block, bias_access_norm, add_bias_tasklet, "_in2", loop_syms, iedge_bias->base_type()	×
170	);
171	builder.add_computational_memlet(	×
172	norm_block, add_bias_tasklet, "_out", output_access_norm, loop_syms, oedge_output->base_type()	×
173	);
174
175	// Cleanup old block
176	builder.remove_memlet(block, *iedge_input);	×
177	builder.remove_memlet(block, *iedge_scale);	×
178	if (iedge_bias) {	×
179	builder.remove_memlet(block, *iedge_bias);	×
180	}	×
181	builder.remove_memlet(block, *iedge_mean);	×
182	builder.remove_memlet(block, *iedge_var);	×
183	builder.remove_memlet(block, *oedge_output);	×
184	builder.remove_node(block, *this);	×
185	builder.remove_child(parent, index + 1);	×
186
187	return true;	×
188	}	×
189
190	std::unique_ptr<data_flow::DataFlowNode> BatchNormalizationNode::
191	clone(size_t element_id, const graph::Vertex vertex, data_flow::DataFlowGraph &parent) const {	×
192	return std::unique_ptr<data_flow::DataFlowNode>(	×
193	new BatchNormalizationNode(element_id, this->debug_info(), vertex, parent, axis_, epsilon_)	×
194	);
195	}	×
196
197	nlohmann::json BatchNormalizationNodeSerializer::serialize(const data_flow::LibraryNode &library_node) {	×
198	const BatchNormalizationNode &node = static_cast<const BatchNormalizationNode &>(library_node);	×
199	nlohmann::json j;	×
200
201	j["code"] = node.code().value();	×
202	j["axis"] = node.axis();	×
203	j["epsilon"] = node.epsilon();	×
204
205	return j;	×
206	}	×
207
208	data_flow::LibraryNode &BatchNormalizationNodeSerializer::deserialize(	×
209	const nlohmann::json &j, builder::StructuredSDFGBuilder &builder, structured_control_flow::Block &parent
210	) {
211	auto code = j["code"].get<std::string>();	×
212	if (code != LibraryNodeType_BatchNormalization.value()) {	×
213	throw std::runtime_error("Invalid library node code");	×
214	}
215
216	sdfg::serializer::JSONSerializer serializer;	×
NEW 217	DebugInfoRegion debug_info = serializer.json_to_debug_info_region(j["debug_info"], builder.debug_info());	×
218
219	auto axis = j["axis"].get<int>();	×
220	auto epsilon = j["epsilon"].get<std::string>();	×
221
222	return builder.add_library_node<BatchNormalizationNode>(parent, debug_info, axis, epsilon);	×
223	}	×
224
225	} // namespace ml
226	} // namespace math
227	} // namespace sdfg

daisytuner / sdfglib / 17637380013

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