mlonmcu 0.4.dev0__tar.gz → 0.4.1__tar.gz

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mlonmcu
-Version: 0.4.dev0
+Version: 0.4.1
 Summary: This project contains research code related to the deployment of inferenceor learning applications on tiny micro-controllers.
 Home-page: https://github.com/tum-ei-eda/mlonmcu
 Author: TUM Department of Electrical and Computer Engineering - Chair of Electronic Design Automation
@@ -21,6 +21,7 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Requires-Python: >=3.7
 Description-Content-Type: text/markdown
+Provides-Extra: all-prod
 Provides-Extra: espidf
 Provides-Extra: etiss
 Provides-Extra: gvsoc_pulp
@@ -35,7 +36,6 @@ Provides-Extra: tvm-autotuning
 Provides-Extra: visualize
 Provides-Extra: visualize_tflite
 Provides-Extra: zephyr
-Provides-Extra: all-prod
 License-File: LICENSE
 License-File: NOTICE
 
@@ -43,14 +43,15 @@ License-File: NOTICE
 
 [![pypi package](https://badge.fury.io/py/mlonmcu.svg)](https://pypi.org/project/mlonmcu)
 [![readthedocs](https://readthedocs.org/projects/mlonmcu/badge/?version=latest)](https://mlonmcu.readthedocs.io/en/latest/?version=latest)
-![covergae](https://byob.yarr.is/tum-ei-eda/mlonmcu/coverage)
+![coverage](https://byob.yarr.is/tum-ei-eda/mlonmcu/coverage)
 [![GitHub license](https://img.shields.io/github/license/tum-ei-eda/mlonmcu.svg)](https://github.com/tum-ei-eda/mlonmcu/blob/main/LICENSE)
 
 [![cicd workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/cicd.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/cicd.yml)
+[![lint workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/style.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/style.yml)
 [![demo workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/demo.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/demo.yml)
-[![container workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/container.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/container.yml)
 [![bench workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/bench.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/bench.yml)
 
+
 This project contains research code related to the deployment of inference or learning applications on tiny micro-controllers.
 
 
@@ -87,7 +88,7 @@ sudo apt install python3-pip python3-venv
 sudo apt install libboost-all-dev graphviz doxygen libtinfo-dev zlib1g-dev texinfo unzip device-tree-compiler tree g++
 
 # Optional (depending on configuration)
-sudo apt install ninja-build flex
+sudo apt install ninja-build flex lsb-release libelf-dev
 ```
 
 Also make sure that your default Python is at least v3.7. If the `python` command is not available in your shell or points Python v2.7 check out `python-is-python3`.
@@ -174,6 +175,14 @@ with mlonmcu.context.MlonMcuContext() as context:
     pass
 ```
 
+## List of interesting MLonMCU forks
+
+- MINRES TGC support: https://github.com/Minres/mlonmcu/tree/develop
+
+## List of existing MLonMCU extensions/plugins
+
+- ABC Example Plugin: coming soon!
+- MINRES TGC Support: coming soon!
 
 ## Development
 
@@ -193,10 +202,6 @@ Documentation is mainly generated automatically from doctrings (triggered via `m
 
 Regarding coding style, it is recommended to run `black` before every commit. The default line length should be given in the `setup.cfg` file.
 
-## Credits
-
-This is a research project proposed by the Chair of Design Automation of the Technical University of Munich.
-
 ### Developers
 
 - Rafael Stahl (TUM) [@rafzi]
@@ -207,6 +212,33 @@ This is a research project proposed by the Chair of Design Automation of the Tec
 
   - Came up with MLonMCU Python package
 
+## Publications
+
+- **MLonMCU: TinyML Benchmarking with Fast Retargeting** ([https://dl.acm.org/doi/10.1145/3637543.3652878](https://dl.acm.org/doi/abs/10.1145/3615338.3618128))
+
+  *CODAI '23: Proceedings of the 2023 Workshop on Compilers, Deployment, and Tooling for Edge AI*
+
+  BibTeX
+
+  ```bibtex
+  @inproceedings{10.1145/3615338.3618128,
+    author = {van Kempen, Philipp and Stahl, Rafael and Mueller-Gritschneder, Daniel and Schlichtmann, Ulf},
+    title = {MLonMCU: TinyML Benchmarking with Fast Retargeting},
+    year = {2024},
+    isbn = {9798400703379},
+    publisher = {Association for Computing Machinery},
+    address = {New York, NY, USA},
+    url = {https://doi.org/10.1145/3615338.3618128},
+    doi = {10.1145/3615338.3618128},
+    abstract = {While there exist many ways to deploy machine learning models on microcontrollers, it is non-trivial to choose the optimal combination of frameworks and targets for a given application. Thus, automating the end-to-end benchmarking flow is of high relevance nowadays. A tool called MLonMCU is proposed in this paper and demonstrated by benchmarking the state-of-the-art TinyML frameworks TFLite for Microcontrollers and TVM effortlessly with a large number of configurations in a low amount of time.},
+    booktitle = {Proceedings of the 2023 Workshop on Compilers, Deployment, and Tooling for Edge AI},
+    pages = {32–36},
+    numpages = {5},
+    keywords = {TinyML, neural networks, microcontrollers},
+    location = {Hamburg, Germany},
+    series = {CODAI '23}
+  }
+  ```
 
 ### Other
 This package was created with Cookiecutter_ and the `audreyr/cookiecutter-pypackage`_ project template. However most of the templates was manually changed to be in Markdown instead of reStructuredText.
@@ -215,6 +247,14 @@ This package was created with Cookiecutter_ and the `audreyr/cookiecutter-pypack
 - **`audreyr/cookiecutter-pypackage`:** https://github.com/audreyr/cookiecutter-pypackage
 
 
+## Acknowledgment
+
+<img src="./BMBF_gefoerdert_2017_en.jpg" alt="drawing" height="75" align="left" >
+
+This research is partially funded by the German Federal Ministry of Education and Research (BMBF) within
+the projects [Scale4Edge](https://www.edacentrum.de/scale4edge/) (grant number 16ME0127) and [MANNHEIM-FlexKI](https://www.edacentrum.de/projekte/MANNHEIM-FlexKI) (grant number 01IS22086L).
+
+
 # History
 
 ## 0.1.0 (2021-11-12)

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/README.md

@@ -2,14 +2,15 @@
 
 [![pypi package](https://badge.fury.io/py/mlonmcu.svg)](https://pypi.org/project/mlonmcu)
 [![readthedocs](https://readthedocs.org/projects/mlonmcu/badge/?version=latest)](https://mlonmcu.readthedocs.io/en/latest/?version=latest)
-![covergae](https://byob.yarr.is/tum-ei-eda/mlonmcu/coverage)
+![coverage](https://byob.yarr.is/tum-ei-eda/mlonmcu/coverage)
 [![GitHub license](https://img.shields.io/github/license/tum-ei-eda/mlonmcu.svg)](https://github.com/tum-ei-eda/mlonmcu/blob/main/LICENSE)
 
 [![cicd workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/cicd.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/cicd.yml)
+[![lint workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/style.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/style.yml)
 [![demo workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/demo.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/demo.yml)
-[![container workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/container.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/container.yml)
 [![bench workflow](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/bench.yml/badge.svg)](https://github.com/tum-ei-eda/mlonmcu/actions/workflows/bench.yml)
 
+
 This project contains research code related to the deployment of inference or learning applications on tiny micro-controllers.
 
 
@@ -46,7 +47,7 @@ sudo apt install python3-pip python3-venv
 sudo apt install libboost-all-dev graphviz doxygen libtinfo-dev zlib1g-dev texinfo unzip device-tree-compiler tree g++
 
 # Optional (depending on configuration)
-sudo apt install ninja-build flex
+sudo apt install ninja-build flex lsb-release libelf-dev
 ```
 
 Also make sure that your default Python is at least v3.7. If the `python` command is not available in your shell or points Python v2.7 check out `python-is-python3`.
@@ -133,6 +134,14 @@ with mlonmcu.context.MlonMcuContext() as context:
     pass
 ```
 
+## List of interesting MLonMCU forks
+
+- MINRES TGC support: https://github.com/Minres/mlonmcu/tree/develop
+
+## List of existing MLonMCU extensions/plugins
+
+- ABC Example Plugin: coming soon!
+- MINRES TGC Support: coming soon!
 
 ## Development
 
@@ -152,10 +161,6 @@ Documentation is mainly generated automatically from doctrings (triggered via `m
 
 Regarding coding style, it is recommended to run `black` before every commit. The default line length should be given in the `setup.cfg` file.
 
-## Credits
-
-This is a research project proposed by the Chair of Design Automation of the Technical University of Munich.
-
 ### Developers
 
 - Rafael Stahl (TUM) [@rafzi]
@@ -166,9 +171,44 @@ This is a research project proposed by the Chair of Design Automation of the Tec
 
   - Came up with MLonMCU Python package
 
+## Publications
+
+- **MLonMCU: TinyML Benchmarking with Fast Retargeting** ([https://dl.acm.org/doi/10.1145/3637543.3652878](https://dl.acm.org/doi/abs/10.1145/3615338.3618128))
+
+  *CODAI '23: Proceedings of the 2023 Workshop on Compilers, Deployment, and Tooling for Edge AI*
+
+  BibTeX
+
+  ```bibtex
+  @inproceedings{10.1145/3615338.3618128,
+    author = {van Kempen, Philipp and Stahl, Rafael and Mueller-Gritschneder, Daniel and Schlichtmann, Ulf},
+    title = {MLonMCU: TinyML Benchmarking with Fast Retargeting},
+    year = {2024},
+    isbn = {9798400703379},
+    publisher = {Association for Computing Machinery},
+    address = {New York, NY, USA},
+    url = {https://doi.org/10.1145/3615338.3618128},
+    doi = {10.1145/3615338.3618128},
+    abstract = {While there exist many ways to deploy machine learning models on microcontrollers, it is non-trivial to choose the optimal combination of frameworks and targets for a given application. Thus, automating the end-to-end benchmarking flow is of high relevance nowadays. A tool called MLonMCU is proposed in this paper and demonstrated by benchmarking the state-of-the-art TinyML frameworks TFLite for Microcontrollers and TVM effortlessly with a large number of configurations in a low amount of time.},
+    booktitle = {Proceedings of the 2023 Workshop on Compilers, Deployment, and Tooling for Edge AI},
+    pages = {32–36},
+    numpages = {5},
+    keywords = {TinyML, neural networks, microcontrollers},
+    location = {Hamburg, Germany},
+    series = {CODAI '23}
+  }
+  ```
 
 ### Other
 This package was created with Cookiecutter_ and the `audreyr/cookiecutter-pypackage`_ project template. However most of the templates was manually changed to be in Markdown instead of reStructuredText.
 
 - **Cookiecutter:** https://github.com/audreyr/cookiecutter
 - **`audreyr/cookiecutter-pypackage`:** https://github.com/audreyr/cookiecutter-pypackage
+
+
+## Acknowledgment
+
+<img src="./BMBF_gefoerdert_2017_en.jpg" alt="drawing" height="75" align="left" >
+
+This research is partially funded by the German Federal Ministry of Education and Research (BMBF) within
+the projects [Scale4Edge](https://www.edacentrum.de/scale4edge/) (grant number 16ME0127) and [MANNHEIM-FlexKI](https://www.edacentrum.de/projekte/MANNHEIM-FlexKI) (grant number 01IS22086L).

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/mlonmcu/artifact.py

@@ -40,7 +40,7 @@ class ArtifactFormat(Enum):  # TODO: ArtifactType, ArtifactKind?
     DATA = 6
     NUMPY = 7
     PARAMS = 8
-    JSON = 9  # ?
+    JSON = 9  # TODO: how about YAML or more general: DICT?
     PATH = 10  # NOT A DIRECTORY?
     RAW = 11
     BIN = 11
@@ -50,6 +50,8 @@ class ArtifactFormat(Enum):  # TODO: ArtifactType, ArtifactKind?
 
 def lookup_artifacts(artifacts, name=None, fmt=None, flags=None, first_only=False):
     """Utility to get a matching artifact for a given set of properties."""
+    if isinstance(name, Path):
+        name = str(name)
     matches = []
     # Warning: if neither name, fmt nor flags is provided, the first artifact (first_only=True)
     # or all (first_only=False) are returned

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/mlonmcu/cli/common.py

@@ -119,9 +119,11 @@ def add_flow_options(parser):
                 metavar="FEATURES",
                 nargs="+",
                 action="append",
-                help=f"Generator statement for features. (Also available: --feature-gen2 ... --feature-gen{number})"
-                if i == 0
-                else argparse.SUPPRESS,
+                help=(
+                    f"Generator statement for features. (Also available: --feature-gen2 ... --feature-gen{number})"
+                    if i == 0
+                    else argparse.SUPPRESS
+                ),
             )
             flow_parser.add_argument(
                 "--config-gen" + suffix,
@@ -129,9 +131,11 @@ def add_flow_options(parser):
                 metavar="KEY=VALUE",
                 nargs="+",
                 action="append",
-                help=f"Generator statement for configs. (Also available: --config-gen2 ... --config-gen{number})"
-                if i == 0
-                else argparse.SUPPRESS,
+                help=(
+                    f"Generator statement for configs. (Also available: --config-gen2 ... --config-gen{number})"
+                    if i == 0
+                    else argparse.SUPPRESS
+                ),
             )
 
     add_gen_args(flow_parser, NUM_GEN_ARGS)

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/mlonmcu/cli/main.py

@@ -81,6 +81,7 @@ def handle_docker(args):
 # def main(args):
 def main(args=None):
     """Console script for mlonmcu."""
+
     parser = argparse.ArgumentParser(
         description="ML on MCU Flow",
         formatter_class=argparse.ArgumentDefaultsHelpFormatter,

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/mlonmcu/context/context.py

@@ -226,7 +226,7 @@ def resolve_environment_file(name: str = None, path: str = None) -> Path:
         env_file = lookup_environment()
         if not env_file:
             raise RuntimeError("Lookup for mlonmcu environment was not successful.")
-    return env_file
+    return Path(env_file)
 
 
 def setup_logging(environment):
@@ -273,6 +273,14 @@ class MlonMcuContext:
         env_file = resolve_environment_file(name=name, path=path)
         assert env_file is not None, "Unable to find a MLonMCU environment"
         self.environment = UserEnvironment.from_file(env_file)  # TODO: move to __enter__
+        actual_home = env_file.parent.resolve()
+        specified_home = Path(self.environment.home).resolve()
+        if specified_home != actual_home:
+            logger.warning(
+                "The HOME directory %s specified in environment.yml does not match workspace directory %s!",
+                specified_home,
+                actual_home,
+            )
         setup_logging(self.environment)
         assert deps_lock in ["read", "write"]
         if deps_lock == "read":

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/mlonmcu/environment/init.py

@@ -17,7 +17,6 @@
 # limitations under the License.
 #
 import sys
-import git
 from pathlib import Path
 import venv
 import os
@@ -32,6 +31,7 @@ from .config import (
 )
 from .templates import write_environment_yaml_from_template
 from mlonmcu.utils import in_virtualenv, ask_user
+from mlonmcu.setup import utils
 from mlonmcu.logging import get_logger
 
 logger = get_logger()
@@ -46,10 +46,10 @@ def create_environment_directories(path, directories):
         (path / directory).mkdir(parents=True, exist_ok=True)
 
 
-def clone_models_repo(
-    dest, url="https://github.com/tum-ei-eda/mlonmcu-models.git"
-):  # TODO: how to get submodule url/ref
-    git.Repo.clone_from(url, dest)
+def clone_models_repo(dest, url=None, ref=None, refresh=False, recursive=False):
+    if url is None:
+        url = "https://github.com/tum-ei-eda/mlonmcu-models.git"
+    utils.clone(url, dest, ref, refresh=refresh, recursive=False)
 
 
 def create_venv_directory(base, hidden=True):
@@ -157,7 +157,11 @@ def initialize_environment(
                 interactive=interactive,
             )
         ):
-            clone_models_repo(models_subdir)
+            custom_url = None
+            custom_ref = None
+            custom_url = config.get("models.clone_url", None)
+            custom_ref = config.get("models.clone_ref", None)
+            clone_models_repo(models_subdir, url=custom_url, ref=custom_ref)
         else:
             subdirs.append("models")
 

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/mlonmcu/feature/features.py

@@ -92,7 +92,8 @@ def get_matching_features(features, feature_type):
     return [feature for feature in features if feature_type in feature.types()]
 
 
-@register_feature("debug_arena", depends=["debug"])
+# @register_feature("debug_arena", depends=["debug"])
+@register_feature("debug_arena")
 class DebugArena(BackendFeature):
     """Enable verbose printing of arena usage for debugging."""
 
@@ -432,8 +433,9 @@ class Vext(SetupFeature, TargetFeature, PlatformFeature):
 
     DEFAULTS = {
         **FeatureBase.DEFAULTS,
-        "vlen": 128,  # 64 does not work with every toolchain
-        "elen": 64,  # some toolchains may generate auto-vectorized programs with elen 64
+        # None -> use target setting
+        "vlen": None,  # 64 does not work with every toolchain
+        "elen": None,  # some toolchains may generate auto-vectorized programs with elen 64
         # use target-side settings by default
         "spec": None,
         "embedded": None,
@@ -444,11 +446,13 @@ class Vext(SetupFeature, TargetFeature, PlatformFeature):
 
     @property
     def vlen(self):
-        return int(self.config["vlen"])
+        value = self.config["vlen"]
+        return None if value is None else int(value)
 
     @property
     def elen(self):
-        return int(self.config["elen"])
+        value = self.config["elen"]
+        return None if value is None else int(value)
 
     @property
     def spec(self):
@@ -491,8 +495,8 @@ class Vext(SetupFeature, TargetFeature, PlatformFeature):
         return {
             "muriscvnn.lib": ["vext"],
             "tflmc.exe": ["vext"],
-            "riscv_gcc.install_dir": ["vext"],
-            "riscv_gcc.name": ["vext"],
+            # "riscv_gcc.install_dir": ["vext"],
+            # "riscv_gcc.name": ["vext"],
         }
 
 
@@ -535,6 +539,73 @@ class Pext(SetupFeature, TargetFeature, PlatformFeature):
         }
 
 
+@register_feature("bext")
+class Bext(SetupFeature, TargetFeature, PlatformFeature):
+    """Enable bitmanipulation extension for supported RISC-V targets"""
+
+    DEFAULTS = {
+        **FeatureBase.DEFAULTS,
+        # use target-side settings by default
+        "spec": None,
+        "zba": True,
+        "zbb": True,
+        "zbc": True,
+        "zbs": True,
+    }
+
+    def __init__(self, features=None, config=None):
+        super().__init__("bext", features=features, config=config)
+
+    @property
+    def spec(self):
+        return self.config["spec"]
+
+    @property
+    def zba(self):
+        value = self.config["zba"]
+        return str2bool(value) if not isinstance(value, (bool, int)) else value
+
+    @property
+    def zbb(self):
+        value = self.config["zbb"]
+        return str2bool(value) if not isinstance(value, (bool, int)) else value
+
+    @property
+    def zbc(self):
+        value = self.config["zbc"]
+        return str2bool(value) if not isinstance(value, (bool, int)) else value
+
+    @property
+    def zbs(self):
+        value = self.config["zbs"]
+        return str2bool(value) if not isinstance(value, (bool, int)) else value
+
+    def get_target_config(self, target):
+        return filter_none(
+            {
+                f"{target}.enable_bext": True,  # Handle via arch characters in the future
+                f"{target}.bext_spec": self.spec,
+                f"{target}.bext_zba": self.zba,
+                f"{target}.bext_zbb": self.zbb,
+                f"{target}.bext_zbc": self.zbc,
+                f"{target}.bext_zbs": self.zbs,
+            }
+        )
+
+    def get_platform_defs(self, platform):
+        assert platform in ["mlif"], f"Unsupported feature '{self.name}' for platform '{platform}'"
+        return {"RISCV_PEXT": self.enabled}
+
+    def get_required_cache_flags(self):
+        # These will be merged automatically with existing ones
+        return {
+            # "riscv_gcc.install_dir": ["bext"],
+            # "riscv_gcc.name": ["bext"],
+            "riscv_gcc.install_dir": [],
+            "riscv_gcc.name": [],
+        }
+
+
 @register_feature("debug")
 class Debug(SetupFeature, PlatformFeature):
     """Enable debugging ability of target software."""
@@ -715,6 +786,40 @@ class Usmp(BackendFeature):
     # -> enable this via backend
 
 
+@register_feature("fuse_ops")
+class FuseOps(BackendFeature):
+    """TODO"""
+
+    DEFAULTS = {
+        **FeatureBase.DEFAULTS,
+        "max_depth": 100,
+    }
+
+    def __init__(self, features=None, config=None):
+        super().__init__("fuse_ops", features=features, config=config)
+
+    @property
+    def max_depth(self):
+        return int(self.config["max_depth"])
+
+    def add_backend_config(self, backend, config):
+        # assert backend in ["tvmaot"], f"Unsupported feature '{self.name}' for backend '{backend}'"
+        # TODO: tvm only
+        if f"{backend}.extra_pass_config" in config:
+            tmp = config[f"{backend}.extra_pass_config"]
+        elif "extra_pass_config" in config:
+            tmp = config["extra_pass_config"]
+        else:
+            tmp = {}
+        if isinstance(tmp, str):
+            import ast
+
+            tmp = ast.literal_eval(tmp)
+        assert isinstance(tmp, dict)
+        tmp["relay.FuseOps.max_depth"] = self.max_depth
+        config.update({f"{backend}.extra_pass_config": tmp})
+
+
 @register_feature("moiopt")
 class MOIOPT(BackendFeature):
     """Memory-Optimizing, Inter-Operator Tiling - currently only supported with custom TVM"""
@@ -1077,6 +1182,44 @@ class DisableLegalize(BackendFeature, SetupFeature):
         return ret
 
 
+@register_feature("uma_backends")
+class UMABackends(BackendFeature):
+    """Add directories that contain UMA backends."""
+
+    REQUIRED = set()
+
+    DEFAULTS = {
+        **FeatureBase.DEFAULTS,
+        "uma_dir": "",
+        "uma_target": "",
+    }
+
+    def __init__(self, features=None, config=None):
+        super().__init__("uma_backends", features=features, config=config)
+
+    @property
+    def uma_dir(self):
+        return self.config["uma_dir"]
+
+    @property
+    def uma_target(self):
+        return self.config["uma_target"]
+
+    def add_backend_config(self, backend, config):
+        assert backend in SUPPORTED_TVM_BACKENDS, f"Unsupported feature '{self.name}' for backend '{backend}'"
+        tvmcArgs = ["--experimental-tvmc-extension", self.uma_dir]
+        if f"{backend}.tvmc_extra_args" in config:
+            config[f"{backend}.tvmc_extra_args"].extend(tvmcArgs)
+        else:
+            config[f"{backend}.tvmc_extra_args"] = tvmcArgs
+        extras = config.get(f"{backend}.extra_target", [])
+        if self.uma_target not in extras:
+            if isinstance(extras, str):
+                extras = [extras]
+            extras.append(self.uma_target)
+        config[f"{backend}.extra_target"] = extras
+
+
 @register_feature("demo")
 class Demo(PlatformFeature):
     """Run demo application instead of benchmarking code."""
@@ -1297,8 +1440,8 @@ class LogInstructions(TargetFeature):
                                     instrs.append(line)
                                 else:
                                     new_lines.append(line)
-                            content = ("\n".join(instrs),)
-                            return "\n".join(new_lines)
+                            content = "\n".join(instrs)
+                            stdout = "\n".join(new_lines)
                         else:
                             assert target in ["spike", "ovpsim", "corev_ovpsim"]
                             log_file = Path(directory) / "instrs.txt"
@@ -1916,6 +2059,34 @@ class SplitLayers(FrontendFeature):
         )
 
 
+@register_feature("tflite_analyze")
+class TfLiteAnalyze(FrontendFeature):
+    """Get the estimated ROM, RAM and MACs from a TFLite model."""
+
+    REQUIRED = {"tflite_analyze.exe"}
+
+    def __init__(self, features=None, config=None):
+        super().__init__("tflite_analyze", features=features, config=config)
+
+    @property
+    def tflite_analyze_exe(self):
+        return self.config["tflite_analyze.exe"]
+
+    def get_frontend_config(self, frontend):
+        assert frontend in ["tflite"], f"Unsupported feature '{self.name}' for frontend '{frontend}'"
+        return filter_none(
+            {
+                f"{frontend}.analyze_enable": self.enabled,
+                f"{frontend}.analyze_script": self.tflite_analyze_exe,
+            }
+        )
+
+    def update_formats(self, frontend, input_formats, output_formats):
+        assert frontend in ["tflite"], f"Unsupported feature '{self.name}' for frontend '{frontend}'"
+        if self.enabled:
+            output_formats.append(ArtifactFormat.TEXT)
+
+
 # @register_feature("hpmcounter")
 class HpmCounter(TargetFeature, PlatformFeature):  # TODO: SetupFeature?
     """Use RISC-V Performance Counters"""
@@ -2009,3 +2180,40 @@ class CV32HpmCounter(HpmCounter):  # TODO: SetupFeature?
 
     def __init__(self, features=None, config=None):
         super().__init__("cv32_hpmcounter", features=features, config=config)
+
+
+@register_feature("vanilla_accelerator")
+class VanillaAccelerator(TargetFeature):
+    """TODO"""
+
+    DEFAULTS = {
+        **FeatureBase.DEFAULTS,
+        "plugin_name": "VanillaAccelerator",
+        "base_addr": None,
+    }
+
+    def __init__(self, features=None, config=None):
+        super().__init__("vanilla_accelerator", features=features, config=config)
+
+    @property
+    def plugin_name(self):
+        value = self.config["plugin_name"]
+        return value
+
+    @property
+    def base_addr(self):
+        value = self.config["base_addr"]
+        return value
+
+    def add_target_config(self, target, config):
+        assert target in ["etiss"]
+        if not self.enabled:
+            return
+        plugins_new = config.get(f"{target}.plugins", [])
+        plugins_new.append(self.plugin_name)
+        config.update({f"{target}.plugins": plugins_new})
+        if self.base_addr is not None:
+            extra_plugin_config = config.get(f"{target}.extra_plugin_config", {})
+            assert self.name not in extra_plugin_config
+            extra_plugin_config[self.name]["baseaddr"] = self.base_addr
+            config.update({f"{target}.extra_plugin_config": extra_plugin_config})

{mlonmcu-0.4.dev0 → mlonmcu-0.4.1}/mlonmcu/flow/tflm/backend/tflmc.py

@@ -106,7 +106,7 @@ size_t model_outputs();
             args.append(str(self.model))
             args.append(str(Path(tmpdirname) / f"{self.prefix}.cc"))
             args.append(f"{self.prefix}_")
-            out = utils.exec_getout(tflmc_exe, *args, live=self.print_outputs, print_output=False)
+            out = utils.exec_getout(tflmc_exe, *args, live=self.print_outputs)
             files = [f for f in os.listdir(tmpdirname) if os.path.isfile(os.path.join(tmpdirname, f))]
             # TODO: ensure that main file is processed first
             for filename in files: