25436733018

Committed 06 May 2026 12:59PM UTC coverage: 65.227% (+0.004%) from 65.223%

Build # 25436733018

Build Type

Pull #681

github

Committed by

web-flow

Commit Message

Merge e31167aa2 into 84dc11a09

Pull Request Pull Request #681: Target specific expand

Coverage Stats

55 of 59 new or added lines in 3 files covered. (93.22%)

27 existing lines in 2 files now uncovered.

31624 of 48483 relevant lines covered (65.23%)

2351.88 hits per line

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

73.64

/python/docc/compiler/docc_program.py

from abc import ABC, abstractmethod
import sys
from typing import Any, Dict, Optional
import os
import re

from docc.sdfg import StructuredSDFG
from docc.sdfg._sdfg import (
    _enable_statistics,
    _statistics_enabled_by_env,
    _statistics_summary,
)
from docc.compiler.compiled_sdfg import CompiledSDFG
from docc.compiler.target_registry import (
    get_target_schedule_fn,
    get_target_compile_fn,
    get_target_expand_fn,
)


def _cuda_expand_fn(sdfg, category: str, kwargs: Dict[str, Any]) -> None:
    sdfg.expand_cuda()
    sdfg.expand()


def _parse_docc_debug() -> dict[str, str]:
    debug_env = os.environ.get("DOCC_DEBUG", "")
    debug_dict = {}
    if debug_env:
        for entry in re.split(r"[;:]", debug_env):
            if not entry:
                continue
            parts = entry.split("=", 1)
            key = parts[0].strip()
            value = parts[1].strip() if len(parts) > 1 else ""
            debug_dict[key] = value
    return debug_dict


def _is_debug_dump(flags: dict[str, str]) -> bool:
    return "dump" in flags


def _is_debug_compile(flags: dict[str, str]) -> bool:
    return "build" in flags


def _get_build_thread_count(flags: dict[str, str]) -> int:
    return int(flags.get("build_threads", "0"))


class DoccProgram(ABC):

    def __init__(
        self,
        name: str,
        target: str = "none",
        category: str = "server",
        instrumentation_mode: Optional[str] = None,
        capture_args: Optional[bool] = None,
        remote_tuning: bool = False,
    ):
        self.name = name
        self.target = target
        self.category = category
        self.remote_tuning = remote_tuning
        self.last_sdfg: Optional[StructuredSDFG] = None
        self.cache: dict = {}
        debug_flags = _parse_docc_debug()
        self.debug_dump: bool = _is_debug_dump(debug_flags)
        self.debug_build: bool = _is_debug_compile(debug_flags)
        self.build_thread_count: int = _get_build_thread_count(debug_flags)

        # Check environment variable DOCC_CI
        docc_ci = os.environ.get("DOCC_CI", "")
        if docc_ci:
            if docc_ci == "regions":
                if instrumentation_mode is None:
                    instrumentation_mode = "ols"
            elif docc_ci == "arg-capture":
                if capture_args is None:
                    capture_args = True
            else:
                # Full mode (or unknown value treated as full)
                if instrumentation_mode is None:
                    instrumentation_mode = "ols"
                if capture_args is None:
                    capture_args = True

        self.instrumentation_mode = instrumentation_mode
        self.capture_args = capture_args

        if target == "cuda":
            from docc.python import register_target_overrides

            register_target_overrides(
                "cuda",
                schedule_fn=None,
                compile_fn=None,
                expand_fn=_cuda_expand_fn,
            )

    @abstractmethod
    def __call__(self, *args: Any) -> Any:
        pass

    @abstractmethod
    def compile(self, *args: Any, output_folder: Optional[str] = None) -> CompiledSDFG:
        pass

    def sdfg_pipe(
        self,
        sdfg: StructuredSDFG,
        output_folder: Optional[str],
        instrumentation_mode: str,
        capture_args: bool,
    ) -> str:

        if self.debug_dump:
            sdfg.dump(output_folder, "py0.parsed", dump_dot=True)

        # Enable statistics if envvar is set
        if _statistics_enabled_by_env():
            _enable_statistics()

        sdfg.validate()

        # Tensor targets keep tensor nodes
        custom_expand_fn = get_target_expand_fn(self.target)
        if custom_expand_fn is not None:
            custom_expand_fn(sdfg, self.category, {})
        else:
            sdfg.expand()
        if self.debug_dump:
            sdfg.dump(output_folder, "py1.expanded", dump_dot=True)

        # Simplify pipelines
        sdfg.simplify()
        if self.debug_dump:
            sdfg.dump(output_folder, "py2.opt", dump_dot=True)

        # Normalization for scheduling
        if self.target != "none":
            sdfg.normalize()

        if self.debug_dump or instrumentation_mode or capture_args:
            sdfg.dump(
                output_folder,
                "py3.norm",
                dump_dot=self.debug_dump,
                dump_json=True,
                record_for_instrumentation=True,
            )

        # Schedule if target is specified

        if self.target != "none":
            # Check for custom registered target first
            custom_schedule_fn = get_target_schedule_fn(self.target)
            if custom_schedule_fn is not None:
                custom_schedule_fn(
                    sdfg, self.category, {"remote_tuning": self.remote_tuning}
                )
            else:
                sdfg.schedule(self.target, self.category, self.remote_tuning)

            if self.debug_dump:
                sdfg.dump(output_folder, "py4.post_sched", dump_dot=True)

        self.last_sdfg = sdfg

        custom_compile_fn = get_target_compile_fn(self.target)
        if custom_compile_fn is not None:
            lib_path = custom_compile_fn(
                sdfg,
                output_folder,
                instrumentation_mode,
                capture_args,
                {"debug_build": self.debug_build, "threads": self.build_thread_count},
            )
        else:
            lib_path = sdfg._compile(
                output_folder=output_folder,
                target=self.target,
                instrumentation_mode=instrumentation_mode,
                capture_args=capture_args,
                debug_build=self.debug_build,
                threads=self.build_thread_count,
            )

        # Dump statistics after compile
        if _statistics_enabled_by_env():
            print(_statistics_summary(), file=sys.stderr)

        return lib_path

    @abstractmethod
    def to_sdfg(self, *args: Any) -> StructuredSDFG:
        pass

    @abstractmethod
    def _convert_inputs(self, args: tuple) -> tuple:
        pass

    @abstractmethod
    def _convert_outputs(self, result: Any, original_args: tuple) -> Any:
        pass

    def _get_cache_key(self, *args: Any) -> str:
        return ""

1	from abc import ABC, abstractmethod	4✔
2	import sys	4✔
3	from typing import Any, Dict, Optional	4✔
4	import os	4✔
5	import re	4✔
6
7	from docc.sdfg import StructuredSDFG	4✔
8	from docc.sdfg._sdfg import (	4✔
9	_enable_statistics,
10	_statistics_enabled_by_env,
11	_statistics_summary,
12	)
13	from docc.compiler.compiled_sdfg import CompiledSDFG	4✔
14	from docc.compiler.target_registry import (	4✔
15	get_target_schedule_fn,
16	get_target_compile_fn,
17	get_target_expand_fn,
18	)
19
20
21	def _cuda_expand_fn(sdfg, category: str, kwargs: Dict[str, Any]) -> None:	4✔
22	sdfg.expand_cuda()	4✔
23	sdfg.expand()	4✔
24
25
26	def _parse_docc_debug() -> dict[str, str]:	4✔
27	debug_env = os.environ.get("DOCC_DEBUG", "")	4✔
28	debug_dict = {}	4✔
29	if debug_env:	4✔
30	for entry in re.split(r"[;:]", debug_env):	×
UNCOV 31	if not entry:	×
UNCOV 32	continue	×
UNCOV 33	parts = entry.split("=", 1)	×
UNCOV 34	key = parts[0].strip()	×
35	value = parts[1].strip() if len(parts) > 1 else ""	×
36	debug_dict[key] = value	×
37	return debug_dict	4✔
38
39
40	def _is_debug_dump(flags: dict[str, str]) -> bool:	4✔
41	return "dump" in flags	4✔
42
43
44	def _is_debug_compile(flags: dict[str, str]) -> bool:	4✔
45	return "build" in flags	4✔
46
47
48	def _get_build_thread_count(flags: dict[str, str]) -> int:	4✔
49	return int(flags.get("build_threads", "0"))	4✔
50
51
52	class DoccProgram(ABC):	4✔
53
54	def __init__(	4✔
55	self,
56	name: str,
57	target: str = "none",
58	category: str = "server",
59	instrumentation_mode: Optional[str] = None,
60	capture_args: Optional[bool] = None,
61	remote_tuning: bool = False,
62	):
63	self.name = name	4✔
64	self.target = target	4✔
65	self.category = category	4✔
66	self.remote_tuning = remote_tuning	4✔
67	self.last_sdfg: Optional[StructuredSDFG] = None	4✔
68	self.cache: dict = {}	4✔
69	debug_flags = _parse_docc_debug()	4✔
70	self.debug_dump: bool = _is_debug_dump(debug_flags)	4✔
71	self.debug_build: bool = _is_debug_compile(debug_flags)	4✔
72	self.build_thread_count: int = _get_build_thread_count(debug_flags)	4✔
73
74	# Check environment variable DOCC_CI
75	docc_ci = os.environ.get("DOCC_CI", "")	4✔
76	if docc_ci:	4✔
UNCOV 77	if docc_ci == "regions":	×
UNCOV 78	if instrumentation_mode is None:	×
UNCOV 79	instrumentation_mode = "ols"	×
UNCOV 80	elif docc_ci == "arg-capture":	×
UNCOV 81	if capture_args is None:	×
82	capture_args = True	×
83	else:
84	# Full mode (or unknown value treated as full)
85	if instrumentation_mode is None:	×
86	instrumentation_mode = "ols"	×
87	if capture_args is None:	×
UNCOV 88	capture_args = True	×
89
90	self.instrumentation_mode = instrumentation_mode	4✔
91	self.capture_args = capture_args	4✔
92
93	if target == "cuda":	4✔
94	from docc.python import register_target_overrides	4✔
95
96	register_target_overrides(	4✔
97	"cuda",
98	schedule_fn=None,
99	compile_fn=None,
100	expand_fn=_cuda_expand_fn,
101	)
102
103	@abstractmethod	4✔
104	def __call__(self, *args: Any) -> Any:	4✔
105	pass	×
106
107	@abstractmethod	4✔
108	def compile(self, *args: Any, output_folder: Optional[str] = None) -> CompiledSDFG:	4✔
109	pass	×
110
111	def sdfg_pipe(	4✔
112	self,
113	sdfg: StructuredSDFG,
114	output_folder: Optional[str],
115	instrumentation_mode: str,
116	capture_args: bool,
117	) -> str:
118
119	if self.debug_dump:	4✔
UNCOV 120	sdfg.dump(output_folder, "py0.parsed", dump_dot=True)	×
121
122	# Enable statistics if envvar is set
123	if _statistics_enabled_by_env():	4✔
UNCOV 124	_enable_statistics()	×
125
126	sdfg.validate()	4✔
127
128	# Tensor targets keep tensor nodes
129	custom_expand_fn = get_target_expand_fn(self.target)	4✔
130	if custom_expand_fn is not None:	4✔
131	custom_expand_fn(sdfg, self.category, {})	4✔
132	else:
133	sdfg.expand()	4✔
134	if self.debug_dump:	4✔
UNCOV 135	sdfg.dump(output_folder, "py1.expanded", dump_dot=True)	×
136
137	# Simplify pipelines
138	sdfg.simplify()	4✔
139	if self.debug_dump:	4✔
UNCOV 140	sdfg.dump(output_folder, "py2.opt", dump_dot=True)	×
141
142	# Normalization for scheduling
143	if self.target != "none":	4✔
144	sdfg.normalize()	4✔
145
146	if self.debug_dump or instrumentation_mode or capture_args:	4✔
147	sdfg.dump(	4✔
148	output_folder,
149	"py3.norm",
150	dump_dot=self.debug_dump,
151	dump_json=True,
152	record_for_instrumentation=True,
153	)
154
155	# Schedule if target is specified
156
157	if self.target != "none":	4✔
158	# Check for custom registered target first
159	custom_schedule_fn = get_target_schedule_fn(self.target)	4✔
160	if custom_schedule_fn is not None:	4✔
161	custom_schedule_fn(	4✔
162	sdfg, self.category, {"remote_tuning": self.remote_tuning}
163	)
164	else:
165	sdfg.schedule(self.target, self.category, self.remote_tuning)	4✔
166
167	if self.debug_dump:	4✔
UNCOV 168	sdfg.dump(output_folder, "py4.post_sched", dump_dot=True)	×
169
170	self.last_sdfg = sdfg	4✔
171
172	custom_compile_fn = get_target_compile_fn(self.target)	4✔
173	if custom_compile_fn is not None:	4✔
174	lib_path = custom_compile_fn(	4✔
175	sdfg,
176	output_folder,
177	instrumentation_mode,
178	capture_args,
179	{"debug_build": self.debug_build, "threads": self.build_thread_count},
180	)
181	else:
182	lib_path = sdfg._compile(	4✔
183	output_folder=output_folder,
184	target=self.target,
185	instrumentation_mode=instrumentation_mode,
186	capture_args=capture_args,
187	debug_build=self.debug_build,
188	threads=self.build_thread_count,
189	)
190
191	# Dump statistics after compile
192	if _statistics_enabled_by_env():	4✔
UNCOV 193	print(_statistics_summary(), file=sys.stderr)	×
194
195	return lib_path	4✔
196
197	@abstractmethod	4✔
198	def to_sdfg(self, *args: Any) -> StructuredSDFG:	4✔
UNCOV 199	pass	×
200
201	@abstractmethod	4✔
202	def _convert_inputs(self, args: tuple) -> tuple:	4✔
UNCOV 203	pass	×
204
205	@abstractmethod	4✔
206	def _convert_outputs(self, result: Any, original_args: tuple) -> Any:	4✔
207	pass	×
208
209	def _get_cache_key(self, *args: Any) -> str:	4✔
UNCOV 210	return ""	×

daisytuner / docc / 25436733018

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