Location

include/mop/profile.h   — Public API, MopFrameStats struct
src/util/profile.c       — High-resolution timer, stats retrieval

Overview

The profile module exposes per-frame timing data and counters collected during mop_viewport_render. After each render call, the viewport stores a MopFrameStats struct that can be retrieved to display or log performance information. Timing uses platform-native high-resolution clocks: mach_absolute_time on macOS and clock_gettime(CLOCK_MONOTONIC) on Linux.

Types

MopFrameStats

typedef struct MopFrameStats {
    double   frame_time_ms;
    double   clear_ms;
    double   transform_ms;
    double   rasterize_ms;
    uint32_t triangle_count;
    uint32_t pixel_count;
} MopFrameStats;

All time values are in milliseconds as double for sub-millisecond precision.

Functions

mop_viewport_get_stats

MopFrameStats mop_viewport_get_stats(const MopViewport *viewport);

Retrieves the statistics from the most recent mop_viewport_render call. The struct is returned by value. If viewport is NULL, a zeroed MopFrameStats is returned.

This function reads the stats stored during the last render pass. It does not trigger a new render. Call it after mop_viewport_render to get meaningful data.

Timing Fields

The three timing fields partition the render pipeline into distinct phases:

frame_time_ms
|-- clear_ms        Framebuffer clear (color, depth, object-ID)
|-- transform_ms    Vertex shader: MVP multiply, viewport transform, clipping
|-- rasterize_ms    Fragment processing: edge function, shading, depth test, blending

Note that clear_ms + transform_ms + rasterize_ms may not exactly equal frame_time_ms due to overhead from setup, subsystem updates, and profiling instrumentation itself.

What Each Field Measures

• frame_time_ms -- End-to-end time from the start of mop_viewport_render to its return. This is the primary metric for overall frame budget. At 60 FPS the budget is approximately 16.7 ms.

• clear_ms -- Time to zero out the color buffer, reset the depth buffer to maximum depth, and clear the object-ID buffer. On the CPU backend this is proportional to pixel_count.

• transform_ms -- Time to multiply each vertex by the model-view-projection matrix, perform perspective division, and apply the viewport transform. Scales with total vertex count across all visible meshes.

• rasterize_ms -- Time to iterate over triangles, evaluate edge functions, interpolate vertex attributes, perform depth testing, and write fragments to the framebuffer. This is typically the dominant cost on the CPU backend and scales with both triangle count and covered pixel area.

• triangle_count -- The sum of index_count / 3 for every active mesh. Useful for correlating render time with scene complexity.

• pixel_count -- Total framebuffer resolution (width * height). Useful for computing per-pixel costs (e.g. clear_ms / pixel_count gives the cost per pixel of the clear pass).

Usage

mop_viewport_render(viewport);

MopFrameStats stats = mop_viewport_get_stats(viewport);
printf("Frame: %.2f ms (clear=%.2f, xform=%.2f, raster=%.2f)\n",
       stats.frame_time_ms, stats.clear_ms,
       stats.transform_ms, stats.rasterize_ms);
printf("Triangles: %u, Pixels: %u\n",
       stats.triangle_count, stats.pixel_count);