makepad/libs/audio_synth
2026-10-01 15:49:24 +02:00
..
src audio_synth: a recipe voice blocks DC -- a one-pole 10 Hz high-pass on each voice's output: a pulse narrower than half its period, a driven or filter-swept layer sat off centre (pulse patches measured 0.3-0.5 of their RMS as offset); a 40 Hz kick loses 0.3 dB. Checked: cargo test --release -p makepad-audio-synth --features instruments; makepad-score-player shader_library (the native pluck's tuning test reads -9.6 cents at note 91, its threshold now 5). 2026-10-01 15:49:24 +02:00
tests render, model, platform, audio_synth, ai: shadow atlases clipped per cascade and face, portable math for identical model builds everywhere, foliage and city rendering, Score replaces the synth sequencer, and sandbox audio generation fixes 2026-09-30 08:11:15 +02:00
Cargo.toml render, model, scene, audio_synth, platform: the game renderer's lights, shadows, GI, VFX and HDR, a character kit, a shared synth, and the platform hooks the Sandbox needed -- libs/render is split into renderer/ and shaders/ modules (pure moves first) and gets stable cascaded shadow maps with a local shadow atlas, cascaded voxel-traced probe GI with SSAO (the CPU BVH placement and analytic emitters are gone), an opt-in HDR scene pass (Renderer::set_hdr_output: AgX, bloom, auto-exposure, FXAA), GPU particles with presets and decals (libs/scene/vfx, with a CPU fallback where float16 blending is missing), world streaming, skinned animation layers and a composed HUD (the HUD document no longer caps its element count; CameraState gains a far plane). Fab and Stage keep the 8-bit lane and in-scene VFX; only hosts that opt in change. libs/model gains the character kit (body, head, gear, clips, presets, FPS arms), construction shapes and decals, bevels, model programs with weathering, and import; mesh_edit decimation collapses every independent candidate per adjacency pass instead of one (a 4.6k-face mesh took 10 s). libs/csg drops the document/rig evaluator the Sandbox no longer uses. New libs/audio_synth (makepad-audio-synth): band-limited oscillators, filters, physically inspired engines, a sound-recipe format, FDN reverb, a spatializer and a look-ahead limiter, with the piano, drum kits, SoundFonts and Ironfish behind instruments; Stage's audio_dsp re-exports it. Platform: GpuInfo::float16_blend_targets, with Metal building an RGBA16Float PSO pair per Bgra8 shader on first use and Vulkan reporting the device's RGBA16F blend support; the WGSL backend forwards only the instance fields the fragment stage reads (39 vec4 varyings overflowed NVIDIA's limit and lost the device); an unbound Vulkan texture_depth slot gets the depth dummy instead of a colour view; an over-budget remote raw grab is box-halved until it fits instead of failing; Windows implements game_input_infos and XInput haptics, DirectInput FFB starts its constant force once and takes exclusive mode only in the foreground, iOS/tvOS get Core Haptics, and game_input gains an additive force API. The AI hub's Claude Code provider can serve one MCP server's tools natively (--mcp-config/--allowedTools, no think-tag litter for redacted thinking), the MCP server answers tool results with image content, and the fleet Qwen provider keeps its vision observations on the row they annotate. widgets gain the cyberpunk desktop style, and a Splash isolate inherits the stylesheet of the VM that allocates it. Checked: cargo check --workspace on macOS (clean apart from apps/source-library's own warnings), makepad-platform/widgets/render/scene/model/audio-synth/mesh-edit/csg for x86_64-unknown-linux-gnu and x86_64-pc-windows-msvc (clean); tests: platform 318, render 488, widgets-core 1307, model 158, scene 35, audio-synth 40, mesh-edit 9, all passing. 2026-09-29 01:55:13 +02:00