[GR-74217][GR-73570] Update numpy and pandas versions in benchmarks.

PullRequest: graalpython/4587

Author: timfel

ci.jsonnet

@@ -5,7 +5,7 @@
 (import "ci/python-gate.libsonnet") +
 (import "ci/python-bench.libsonnet") +
 {
-    overlay: "c004ac2843125f797eb9f9a28ae76b51952ad9d7",
+    overlay: "40bd8048e1a6ba45494605955ffe748ae4db20be",
     specVersion: "8",
     // Until buildbot issues around CI tiers are resolved, we cannot use them
     // tierConfig: self.tierConfig,
@@ -21,6 +21,7 @@
         GRAAL_ENTERPRISE_GIT: "",
         CI_OVERLAYS_GIT: "",
         BENCHMARK_CONFIG_GIT: "",
+        NUMPY_REPO_GIT: "",
         PANDAS_REPO_GIT: "",
         PIP_EXTRA_INDEX_URL: "",
         WATCHDOG_GIT: "",

ci/python-bench.libsonnet

@@ -92,6 +92,7 @@
     local environment(os, arch) = self.environment(os, arch) + {
         BENCH_RESULTS_FILE_PATH: "bench-results.json",
         PANDAS_REPO_URL: $.overlay_imports.PANDAS_REPO_GIT,
+        NUMPY_REPO_URL: $.overlay_imports.NUMPY_REPO_GIT,
     } + if $.overlay_imports.PIP_EXTRA_INDEX_URL != "" then {
         PIP_EXTRA_INDEX_URL: $.overlay_imports.PIP_EXTRA_INDEX_URL,
     } else {},

ci/python-gate.libsonnet

@@ -81,7 +81,6 @@
         linux: {
             common: {
                 LIBGMP: {name:"libgmp", version:"6.1.0", platformspecific:true},
-                NUMPY_BENCHMARKS_DIR: {name: "numpy", version: "1.26.4", platformspecific: false},
                 PYPY_HOME: {name: "pypy3", version: "3.10-v7.3.12", platformspecific: true},
                 PYPY_BENCHMARKS_DIR: {name: "pypybenchmarks", version: "84f401a8f55a", platformspecific: false},
             },

mx.graalpython/mx_graalpython_python_benchmarks.py

@@ -70,17 +70,45 @@
 ]
 
 SKIPPED_NUMPY_BENCHMARKS = [
+    "bench_core.CountNonzero.time_count_nonzero(1, 1000000, <class 'str'>)",  # Times out
     "bench_core.CountNonzero.time_count_nonzero(2, 1000000, <class 'str'>)",  # Times out
     "bench_core.CountNonzero.time_count_nonzero(3, 1000000, <class 'str'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero(1, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero(2, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero(3, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_axis(1, 1000000, <class 'str'>)",  # Times out
     "bench_core.CountNonzero.time_count_nonzero_axis(2, 1000000, <class 'str'>)",  # Times out
     "bench_core.CountNonzero.time_count_nonzero_axis(3, 1000000, <class 'str'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_axis(1, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_axis(2, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_axis(3, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_multi_axis(1, 1000000, <class 'str'>)",  # Times out
     "bench_core.CountNonzero.time_count_nonzero_multi_axis(2, 1000000, <class 'str'>)",  # Times out
     "bench_core.CountNonzero.time_count_nonzero_multi_axis(3, 1000000, <class 'str'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_multi_axis(1, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_multi_axis(2, 1000000, <class 'object'>)",  # Times out
+    "bench_core.CountNonzero.time_count_nonzero_multi_axis(3, 1000000, <class 'object'>)",  # Times out
     "bench_linalg.LinalgSmallArrays.time_det_small_array",  # TODO fails with numpy.linalg.LinAlgError
     "bench_indexing.IndexingSeparate.time_mmap_fancy_indexing",  # Hangs in periodic job GR-73912
     "bench_indexing.IndexingStructured0D.time_array_slice",  # Hangs in periodic job GR-73912
 ]
 
+SKIPPED_NUMPY_BENCHMARK_PATTERNS = [
+    # These patterns match ASV's expanded benchmark names. They keep the small
+    # cases while dropping large Cartesian-product outliers that dominate the
+    # periodic job runtime.
+    r"bench_core\.CorrConv\.time_(?:correlate|convolve)\((?:100000,|[^)]*, (?:1000|10000), )",
+    r"bench_core\.CountNonzero\.time_count_nonzero(?:_axis|_multi_axis)?\([1-3], (?:10000|1000000), ",
+    r"bench_core\.Nonzero\.time_nonzero(?:_sparse|_dense)?\([^)]*, \((?:1000000,|1000, 1000)\)\)",
+    r"bench_core\.StatsMethods\.time_(?:max|mean|min|prod|std|sum|var)\('[^']+', 10000\)",
+    r"bench_indexing\.Indexing\.time_op\('(?:complex64|complex128|object|O,i)'",
+    r"bench_indexing\.Indexing\.time_op\('[^']+', 'indexes_rand_'",
+    r"bench_indexing\.Indexing\.time_op\('[^']+', '[^']+', 'np\.ix_\(I, I\)'",
+    r"bench_indexing\.IndexingWith1DArr\.time_(?:getitem|setitem)_ordered\(\((?:1000, 2|2, 1000, 1|1000, 3)\), ",
+    r"bench_linalg\.LinAlgTransposeVdot\.time_(?:transpose|vdot)\(\(64, 64\), ",
+    r"bench_linalg\.MatmulStrided\.time_matmul\('matmul_m(?:150|400)_",
+]
+
 DEFAULT_PANDAS_BENCHMARKS = [
     "reshape",
     "replace"
@@ -199,11 +227,12 @@
 ]
 
 
-def create_asv_benchmark_selection(benchmarks, skipped=()):
+def create_asv_benchmark_selection(benchmarks, skipped=(), skipped_patterns=()):
     regex = '|'.join(benchmarks)
-    if not skipped:
-        return regex
     negative_lookaheads = [re.escape(skip) + (r'\b' if not skip.endswith(')') else '') for skip in skipped]
+    negative_lookaheads.extend(skipped_patterns)
+    if not negative_lookaheads:
+        return regex
     return '^(?!' + '|'.join(negative_lookaheads) + ')(' + regex + ')'
 
 
@@ -435,7 +464,7 @@ class PySuite(mx_benchmark.TemporaryWorkdirMixin, mx_benchmark.VmBenchmarkSuite)
 
 
 class PyPerformanceSuite(PySuite):
-    VERSION = "1.0.6"
+    VERSION = "1.14.0"
 
     def name(self):
         return "pyperformance-suite"
@@ -600,10 +629,11 @@ def _vmRun(self, vm, workdir, command, benchmarks, bmSuiteArgs):
 
 
 class NumPySuite(PySuite):
-    VERSION = "1.26.4"
+    VERSION = "2.4.4"
 
     BENCHMARK_REQ = [
         "asv==0.5.1",
+        "asv-runner==0.2.1",
         f"setuptools=={SETUPTOOLS_PIN}",
         "distlib==0.3.6",
         "filelock==3.8.0",
@@ -648,13 +678,14 @@ def _vmRun(self, vm, workdir, command, benchmarks, bmSuiteArgs):
                 shutil.copytree(artifact, npdir)
             else:
                 mx.warn("NUMPY_BENCHMARKS_DIR is not set, cloning numpy repository")
+                repo_url = os.environ.get("NUMPY_REPO_URL", "https://github.com/numpy/numpy.git")
                 mx.run(
                     [
                         "git",
                         "clone",
                         "--depth",
                         "1",
-                        "https://github.com/numpy/numpy.git",
+                        repo_url,
                         "--branch",
                         f"v{self.VERSION}",
                         "--single-branch",
@@ -691,7 +722,9 @@ def _vmRun(self, vm, workdir, command, benchmarks, bmSuiteArgs):
                 "--python=same",
                 "--set-commit-hash",
                 f"v{self.VERSION}",
-                "-b", create_asv_benchmark_selection(benchmarks, skipped=SKIPPED_NUMPY_BENCHMARKS),
+                "-b", create_asv_benchmark_selection(
+                    benchmarks, skipped=SKIPPED_NUMPY_BENCHMARKS, skipped_patterns=SKIPPED_NUMPY_BENCHMARK_PATTERNS
+                ),
             ],
             cwd=benchdir,
             nonZeroIsFatal=False,
@@ -708,7 +741,7 @@ def _vmRun(self, vm, workdir, command, benchmarks, bmSuiteArgs):
 
 
 class PandasSuite(PySuite):
-    VERSION = "1.5.2"
+    VERSION = "3.0.2"
     VERSION_TAG = "v" + VERSION
 
     BENCHMARK_REQ = [
@@ -721,6 +754,7 @@ class PandasSuite(PySuite):
         "virtualenv==20.16.3",
         "packaging==24.0",
         "jinja2",
+        "tzdata==2026.2",
         f"numpy=={NumPySuite.VERSION}",
         f"pandas=={VERSION}",
     ]