P9 (WIP): compiled kernels (rust/simurg-core, PyO3) with bit-identical pure-Python fallback

Rust ports of the hot paths found by profiling: fused per-locus GenCall calling
and scoring, BAF/LRR, libm maps with CPython math semantics, both nearest-neighbour
searches of the normalization fit, and the BPM LocusEntry parser. Address
coverage is memoised per IDAT layout. SIMURG_PURE_PYTHON=1 forces the pure paths.
On the Pi: genotype_sample 3.4 s -> 1.1 s, 36-sample GSA run 131 s -> 46 s, BPM
read 4x faster; 36 GSA + 6 Omni2.5 GTCs and QC tables byte-identical to the pure
path; tests/unit/test_core_kernels.py checks every kernel on edge cases.
bench/profile_genotype.py, bench/scaling.py, cloud task p9 (thread scaling).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01814GKVzZVL3FuF3DeiZXhv

src/simurg/genotype/baf_lrr.py

@@ -11,6 +11,7 @@ import math
 import numpy as np
 import numpy.typing as npt
 
+from simurg._core import core
 from simurg.genotype.clusters import ClusterArrays
 from simurg.genotype.gencall import _libm
 
@@ -22,6 +23,17 @@ NEG_NAN = np.frombuffer(np.uint32(0xFFC00000).tobytes(), dtype=np.float32)[0]  #
 def baf_lrr(
     theta: F32Arr, r: F32Arr, raw_zero: npt.NDArray[np.bool_], c: ClusterArrays, intensity_only: npt.NDArray[np.bool_]
 ) -> tuple[F32Arr, F32Arr]:
+    if core is not None:
+        return core.baf_lrr(  # type: ignore[no-any-return]
+            np.ascontiguousarray(theta, np.float32),
+            np.ascontiguousarray(r, np.float32),
+            np.ascontiguousarray(raw_zero, np.bool_),
+            np.ascontiguousarray(c.present, np.bool_),
+            np.ascontiguousarray(c.theta_mean, np.float32).ravel(),
+            np.ascontiguousarray(c.r_mean, np.float32).ravel(),
+            np.ascontiguousarray(c.mean_r, np.float32),
+            np.ascontiguousarray(intensity_only, np.bool_),
+        )
     t = np.asarray(theta, np.float32)
     rr = np.asarray(r, np.float32)
     at, bt, ct = c.theta_mean[:, 0], c.theta_mean[:, 1], c.theta_mean[:, 2]

src/simurg/genotype/gencall.py

@@ -15,6 +15,7 @@ from dataclasses import dataclass
 import numpy as np
 import numpy.typing as npt
 
+from simurg._core import core
 from simurg.genotype.clusters import ClusterArrays
 
 F32 = np.float32
@@ -42,8 +43,17 @@ def _c_call(fn, v: float) -> float:  # type: ignore[no-untyped-def]
         return math.inf
 
 
+_CORE_FN = {math.exp: "exp", math.log10: "log10", math.log2: "log2"}
+
+
 def _libm(fn, x: F64Arr) -> F64Arr:  # type: ignore[no-untyped-def]
-    """Element-wise libm call (bit-identical to C)."""
+    """Element-wise libm call (bit-identical to C); compiled when ``simurg_core`` is installed."""
+    if core is not None:
+        xa = np.ascontiguousarray(x, dtype=np.float64).ravel()
+        if (e := getattr(fn, "exponent", None)) is not None:
+            return core.pow(xa, e).reshape(np.shape(x))  # type: ignore[no-any-return]
+        if (name := _CORE_FN.get(fn)) is not None:
+            return getattr(core, name)(xa).reshape(np.shape(x))  # type: ignore[no-any-return]
     return np.fromiter((_c_call(fn, v) for v in x.tolist()), dtype=np.float64, count=len(x))
 
 
@@ -51,6 +61,7 @@ def _pow(e: float):  # type: ignore[no-untyped-def]
     def f(v: float) -> float:
         return math.pow(v, e)
 
+    f.exponent = e  # type: ignore[attr-defined]
     return f
 
 
@@ -137,6 +148,23 @@ def call_genotypes(
     cutoff: float = DEFAULT_CUTOFF,
 ) -> Calls:
     """Bayesian calls and GenCall scores for one sample (plan §7.5)."""
+    if core is not None:
+        f32 = [np.ascontiguousarray(a, np.float32).ravel() for a in (c.theta_mean, c.theta_dev, c.r_mean, c.r_dev)]
+        geno, score = core.call_genotypes(
+            np.ascontiguousarray(theta, np.float32),
+            np.ascontiguousarray(r, np.float32),
+            np.ascontiguousarray(raw_zero, np.bool_),
+            np.ascontiguousarray(c.present, np.bool_),
+            np.ascontiguousarray(c.n, np.int32).ravel(),
+            *f32,
+            np.ascontiguousarray(c.intensity_threshold, np.float32),
+            np.ascontiguousarray(c.gentrain, np.float32),
+            np.ascontiguousarray(intensity_only, np.bool_),
+            F32(cutoff),
+        )
+        return Calls(
+            genotypes=geno, scores=score, num_calls=int((geno > 0).sum()), num_intensity_only=int(intensity_only.sum())
+        )
     n = len(theta)
     t = np.asarray(theta, np.float32)
     rr = np.asarray(r, np.float32)