Improved CPU knobs

gnommo/cache.py

@@ -20,32 +20,52 @@ _assets_config: Optional[dict] = None
 _perf_config: Optional[dict] = None
 
 
-def get_ffmpeg_thread_count() -> int:
-    """Return FFmpeg thread count based on [performance] cpu_limit in ~/.gnommo.conf.
-
-    cpu_limit is a fraction of logical CPUs (e.g. 0.8 = 80%).
-    Defaults to 1 when not configured, which is safe on memory-constrained machines.
-
-    Example ~/.gnommo.conf:
-        [performance]
-        cpu_limit = 0.8
-    """
+def _load_perf_config() -> dict:
+    """Read and cache the [performance] section of ~/.gnommo.conf."""
     global _perf_config
-    if _perf_config is None:
-        config_path = Path.home() / ".gnommo.conf"
-        _perf_config = {}
-        if config_path.exists():
-            cfg = configparser.ConfigParser()
-            cfg.read(config_path)
-            if cfg.has_option("performance", "cpu_limit"):
+    if _perf_config is not None:
+        return _perf_config
+
+    _perf_config = {}
+    config_path = Path.home() / ".gnommo.conf"
+    if config_path.exists():
+        cfg = configparser.ConfigParser()
+        cfg.read(config_path)
+        for key in ("cpu_limit_preprocess", "cpu_limit_render", "cpu_limit"):
+            if cfg.has_option("performance", key):
                 try:
-                    _perf_config["cpu_limit"] = float(
-                        cfg.get("performance", "cpu_limit")
-                    )
+                    _perf_config[key] = float(cfg.get("performance", key))
                 except ValueError:
                     pass
+        if cfg.has_option("performance", "render_chunk_slides"):
+            try:
+                _perf_config["render_chunk_slides"] = int(
+                    cfg.get("performance", "render_chunk_slides")
+                )
+            except ValueError:
+                pass
+    return _perf_config
 
-    cpu_limit = _perf_config.get("cpu_limit")
+
+def get_ffmpeg_thread_count(stage: str = "preprocess") -> int:
+    """Return the FFmpeg thread count for a pipeline stage from ~/.gnommo.conf.
+
+    Preprocessing and rendering scale differently, so they read separate CPU
+    fractions of the logical core count:
+
+        [performance]
+        cpu_limit_preprocess = 0.8   # throughput-bound; safe at high parallelism
+        cpu_limit_render     = 0.15  # -filter_complex spawns swscaler threads per
+                                     # layer and OOMs at high core counts
+
+    `stage` is "preprocess" or "render". The legacy single `cpu_limit` key is the
+    fallback for either stage when its specific key is absent. Each value is a
+    fraction of logical CPUs (0.8 = 80%); defaults to 1 thread when nothing is
+    configured, which is safe on memory-constrained machines.
+    """
+    cfg = _load_perf_config()
+    key = "cpu_limit_render" if stage == "render" else "cpu_limit_preprocess"
+    cpu_limit = cfg.get(key, cfg.get("cpu_limit"))
     if cpu_limit is None:
         return 1
     cpu_count = os.cpu_count() or 1
@@ -62,10 +82,7 @@ def get_render_chunk_size() -> Optional[int]:
         [performance]
         render_chunk_slides = 15
     """
-    global _perf_config
-    if _perf_config is None:
-        get_ffmpeg_thread_count()  # populates _perf_config
-    val = _perf_config.get("render_chunk_slides")
+    val = _load_perf_config().get("render_chunk_slides")
     if val is None:
         return None
     try:

gnommo/preprocessor.py

@@ -21,10 +21,10 @@ from typing import Union, Optional
 
 
 def _tc() -> str:
-    """Return FFmpeg thread count string from ~/.gnommo.conf [performance] cpu_limit."""
+    """FFmpeg thread count for preprocessing (~/.gnommo.conf cpu_limit_preprocess)."""
     from .cache import get_ffmpeg_thread_count
 
-    return str(get_ffmpeg_thread_count())
+    return str(get_ffmpeg_thread_count("preprocess"))
 
 
 # Number of parallel workers for chunk processing

gnommo/renderer.py

@@ -447,7 +447,7 @@ def build_ffmpeg_command(plan: RenderPlan, output_path: Path) -> list[str]:
     # thread per core no matter what — the real cause of the render-stage memory blowup.
     from .cache import get_ffmpeg_thread_count
 
-    _tc = str(get_ffmpeg_thread_count())
+    _tc = str(get_ffmpeg_thread_count("render"))
     cmd.extend(
         ["-threads", _tc, "-filter_threads", _tc, "-filter_complex_threads", _tc]
     )