[datasets] Update AnghaBench to v1.

compiler_gym/envs/llvm/datasets/__init__.py

@@ -212,6 +212,25 @@ def get_llvm_datasets(site_data_base: Optional[Path] = None) -> Iterable[Dataset
     site_data_base = site_data_base or site_data_path("llvm-v0")
 
     yield AnghaBenchDataset(site_data_base=site_data_base, sort_order=0)
+    # Add legacy version of Anghabench using an old manifest.
+    anghabench_v0_manifest_url, anghabench_v0_manifest_sha256 = {
+        "darwin": (
+            "https://dl.fbaipublicfiles.com/compiler_gym/llvm_bitcodes-10.0.0-anghabench-v0-macos-manifest.bz2",
+            "39464256405aacefdb7550a7f990c9c578264c132804eec3daac091fa3c21bd1",
+        ),
+        "linux": (
+            "https://dl.fbaipublicfiles.com/compiler_gym/llvm_bitcodes-10.0.0-anghabench-v0-linux-manifest.bz2",
+            "a038d25d39ee9472662a9704dfff19c9e3512ff6a70f1067af85c5cb3784b477",
+        ),
+    }[sys.platform]
+    yield AnghaBenchDataset(
+        name="benchmark://anghabench-v0",
+        site_data_base=site_data_base,
+        sort_order=0,
+        manifest_url=anghabench_v0_manifest_url,
+        manifest_sha256=anghabench_v0_manifest_sha256,
+        deprecated="Please use anghabench-v1",
+    )
     yield BlasDataset(site_data_base=site_data_base, sort_order=0)
     yield CLgenDataset(site_data_base=site_data_base, sort_order=0)
     yield CBenchDataset(site_data_base=site_data_base, sort_order=-1)

compiler_gym/envs/llvm/datasets/anghabench.py

@@ -6,6 +6,7 @@
 import sys
 from concurrent.futures import as_completed
 from pathlib import Path
+from typing import Optional
 
 from compiler_gym.datasets import Benchmark, TarDatasetWithManifest
 from compiler_gym.datasets.benchmark import BenchmarkWithSource
@@ -38,28 +39,36 @@ class AnghaBenchDataset(TarDatasetWithManifest):
     overhead of compiling it from C to bitcode. This is a one-off cost.
     """
 
-    def __init__(self, site_data_base: Path, sort_order: int = 0):
-        manifest_url, manifest_sha256 = {
+    def __init__(
+        self,
+        site_data_base: Path,
+        sort_order: int = 0,
+        manifest_url: Optional[str] = None,
+        manifest_sha256: Optional[str] = None,
+        deprecated: Optional[str] = None,
+        name: Optional[str] = None,
+    ):
+        manifest_url_, manifest_sha256_ = {
             "darwin": (
-                "https://dl.fbaipublicfiles.com/compiler_gym/llvm_bitcodes-10.0.0-anghabench-v0-macos-manifest.bz2",
-                "39464256405aacefdb7550a7f990c9c578264c132804eec3daac091fa3c21bd1",
+                "https://dl.fbaipublicfiles.com/compiler_gym/llvm_bitcodes-10.0.0-anghabench-v1-macos-manifest.bz2",
+                "96ead63da5f8efa07fd0370f0c6e452b59bed840828b8b19402102b1ce3ee109",
             ),
             "linux": (
-                "https://dl.fbaipublicfiles.com/compiler_gym/llvm_bitcodes-10.0.0-anghabench-v0-linux-manifest.bz2",
-                "a038d25d39ee9472662a9704dfff19c9e3512ff6a70f1067af85c5cb3784b477",
+                "https://dl.fbaipublicfiles.com/compiler_gym/llvm_bitcodes-10.0.0-anghabench-v1-linux-manifest.bz2",
+                "14df85f650199498cf769715e9f0d7841d09f9fa62a95b8ecc242bdaf227f33a",
             ),
         }[sys.platform]
         super().__init__(
-            name="benchmark://anghabench-v0",
+            name=name or "benchmark://anghabench-v1",
             description="Compile-only C/C++ functions extracted from GitHub",
             references={
                 "Paper": "https://homepages.dcc.ufmg.br/~fernando/publications/papers/FaustinoCGO21.pdf",
                 "Homepage": "http://cuda.dcc.ufmg.br/angha/",
             },
             license="Unknown. See: https://github.com/brenocfg/AnghaBench/issues/1",
             site_data_base=site_data_base,
-            manifest_urls=[manifest_url],
-            manifest_sha256=manifest_sha256,
+            manifest_urls=[manifest_url or manifest_url_],
+            manifest_sha256=manifest_sha256 or manifest_sha256_,
             tar_urls=[
                 "https://github.com/brenocfg/AnghaBench/archive/d8034ac8562b8c978376008f4b33df01b8887b19.tar.gz"
             ],
@@ -68,6 +77,7 @@ def __init__(self, site_data_base: Path, sort_order: int = 0):
             tar_compression="gz",
             benchmark_file_suffix=".bc",
             sort_order=sort_order,
+            deprecated=deprecated,
         )
 
     def benchmark(self, uri: str) -> Benchmark:

docs/source/llvm/index.rst

@@ -21,7 +21,7 @@ We provide several datasets of open-source LLVM-IR benchmarks for use:
 +----------------------------+--------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------------------+
 | Dataset                    | Num. Benchmarks [#f1]_   | Description                                                                                                                                                                                                        | Validatable [#f2]_   |
 +============================+==========================+====================================================================================================================================================================================================================+======================+
-| benchmark://anghabench-v0  | 1,042,976                | Compile-only C/C++ functions extracted from GitHub [`Homepage <http://cuda.dcc.ufmg.br/angha/>`__, `Paper <https://homepages.dcc.ufmg.br/~fernando/publications/papers/FaustinoCGO21.pdf>`__]                      | No                   |
+| benchmark://anghabench-v1  | 1,041,333                | Compile-only C/C++ functions extracted from GitHub [`Homepage <http://cuda.dcc.ufmg.br/angha/>`__, `Paper <https://homepages.dcc.ufmg.br/~fernando/publications/papers/FaustinoCGO21.pdf>`__]                      | No                   |
 +----------------------------+--------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------------------+
 | benchmark://blas-v0        | 300                      | Basic linear algebra kernels [`Homepage <http://www.netlib.org/blas/>`__, `Paper <https://strum355.netsoc.co/books/PDF/Basic%20Linear%20Algebra%20Subprograms%20for%20Fortran%20Usage%20-%20BLAS%20(1979).pdf>`__] | No                   |
 +----------------------------+--------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------------------+

tests/llvm/datasets/anghabench_test.py

@@ -23,33 +23,33 @@
 def anghabench_dataset() -> AnghaBenchDataset:
     env = gym.make("llvm-v0")
     try:
-        ds = env.datasets["anghabench-v0"]
+        ds = env.datasets["anghabench-v1"]
     finally:
         env.close()
     yield ds
 
 
 def test_anghabench_size(anghabench_dataset: AnghaBenchDataset):
     if sys.platform == "darwin":
-        assert anghabench_dataset.size == 1042908
+        assert anghabench_dataset.size == 1041265
     else:
-        assert anghabench_dataset.size == 1042976
+        assert anghabench_dataset.size == 1041333
 
 
 def test_missing_benchmark_name(anghabench_dataset: AnghaBenchDataset, mocker):
     # Mock install() so that on CI it doesn't download and unpack the tarfile.
     mocker.patch.object(anghabench_dataset, "install")
 
     with pytest.raises(
-        LookupError, match=r"^No benchmark specified: benchmark://anghabench-v0$"
+        LookupError, match=r"^No benchmark specified: benchmark://anghabench-v1$"
     ):
-        anghabench_dataset.benchmark("benchmark://anghabench-v0")
+        anghabench_dataset.benchmark("benchmark://anghabench-v1")
     anghabench_dataset.install.assert_called_once()
 
     with pytest.raises(
-        LookupError, match=r"^No benchmark specified: benchmark://anghabench-v0/$"
+        LookupError, match=r"^No benchmark specified: benchmark://anghabench-v1/$"
     ):
-        anghabench_dataset.benchmark("benchmark://anghabench-v0/")
+        anghabench_dataset.benchmark("benchmark://anghabench-v1/")
     assert anghabench_dataset.install.call_count == 2
 
 

tests/llvm/datasets/llvm_datasets_test.py

@@ -14,7 +14,7 @@ def test_default_dataset_list():
     try:
         assert list(d.name for d in env.datasets) == [
             "benchmark://cbench-v1",
-            "benchmark://anghabench-v0",
+            "benchmark://anghabench-v1",
             "benchmark://blas-v0",
             "benchmark://clgen-v0",
             "benchmark://github-v0",