1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
//! Metal shader hot-reload: `MtlContext::reload_shaders` rebuilds every live
//! built-in pipeline from the checkout's shader sources. The `reload-shaders`
//! debug command sets the shared flag, and the main thread polls it at the top
//! of `draw_frame`.
//!
//! Entirely a dev-loop concern: the flag exists only when `MtlContext::new` is
//! called with `hot_reload = true`. Production `cn run` never sets it.
#![deny(unsafe_op_in_unsafe_fn)]
use concinnity_core::gfx::mesh_payload;
use concinnity_core::render::error::RenderResult;
use objc2::rc::Retained;
use objc2_metal::{MTLVertexDescriptor, MTLVertexFormat, MTLVertexStepFunction};
use std::sync::atomic::Ordering;
use super::auto_exposure::build_auto_exposure_pipelines;
use super::context::MtlContext;
use super::cull::{build_cull_pipeline, build_shadow_cull_pipeline};
use super::decal::build_decal_pipeline;
use super::descriptors::{VertexAttr, VertexLayout, vertex_descriptor};
use super::fog::build_fog_pipeline;
use super::hiz::build_hiz_pipelines;
use super::init::pipelines::{
BucketBuild, BucketPipelines, build_bindless_sampler_args, build_shadow_bindless_pipeline,
replacement,
};
use super::pipeline::{build_post_pipeline, build_text_pipeline};
use super::post::build_rt_reflection_pipeline;
use super::post::post_device::MtlPostDevice;
// Rebuild a built-in pipeline only when it is currently live. Expands to
// `if $cond { Some($build?) } else { None }`: the rebuild-then-swap pattern
// `reload_shaders` repeats for every optional pipeline: a `None` field stays
// `None`, and any compile error (the `?`) aborts the whole reload before the
// swap, leaving the live pipelines untouched.
macro_rules! rebuild_if_live {
($cond:expr_2021, $build:expr_2021 $(,)?) => {
if $cond { Some($build?) } else { None }
};
}
// Static-vertex-layout descriptor used by the velocity / SSAO / SSR pre-pass
// rebuilds during hot-reload. Matches the layout `MtlContext::new` builds at
// init; kept in sync by construction since both touch the 56-byte static
// `Vertex` struct.
fn static_vertex_descriptor() -> Retained<MTLVertexDescriptor> {
vertex_descriptor(
&[
VertexAttr {
index: 0,
format: MTLVertexFormat::Float3,
offset: 0,
buffer_index: 1,
},
VertexAttr {
index: 1,
format: MTLVertexFormat::Float3,
offset: 12,
buffer_index: 1,
},
VertexAttr {
index: 2,
format: MTLVertexFormat::Float3,
offset: 24,
buffer_index: 1,
},
VertexAttr {
index: 3,
format: MTLVertexFormat::Float3,
offset: 36,
buffer_index: 1,
},
VertexAttr {
index: 4,
format: MTLVertexFormat::Float2,
offset: 48,
buffer_index: 1,
},
],
&[VertexLayout {
buffer_index: 1,
stride: std::mem::size_of::<mesh_payload::Vertex>(),
step: MTLVertexStepFunction::PerVertex,
}],
)
}
impl MtlContext {
// True when the shared shader-reload flag is set. Cheap atomic load; called
// at the top of `draw_frame`. Returns false when hot-reload is off so
// the production path never enters the reload branch.
pub(super) fn shader_reload_requested(&self) -> bool {
self.hot_reload
.reload_pending
.as_ref()
.map(|f| f.load(Ordering::SeqCst))
.unwrap_or(false)
}
// Clear the pending-reload flag. Called after `reload_shaders` regardless
// of outcome so a failed rebuild does not loop forever.
pub(super) fn clear_shader_reload_flag(&self) {
if let Some(flag) = &self.hot_reload.reload_pending {
flag.store(false, Ordering::SeqCst);
}
}
// Rebuild every built-in Metal renderer pipeline from disk-resident source.
// Each pipeline is constructed into a temporary first; only when every
// rebuild succeeds does the context atomically swap them in. Any compile
// or link error logs the underlying message and leaves the live pipelines
// untouched: a typo in a shader edit won't crash the running session.
//
// Covers the main pass and its GPU cull (one builder, since the cull's
// argument encoder comes from the pipeline it feeds), every shader bucket's
// main and pre-pass pair, and the shadow views' pipelines. A world Shader's
// own pair still wins in the main build, so a save to an engine template
// never swaps a world's program for the engine's.
pub(super) fn reload_shaders(&mut self) -> RenderResult<()> {
if !self.hot_reload.enabled {
return Ok(());
}
self.hot_reload.generation += 1;
let device = &self.hw.device;
let hr = true;
// Build every replacement into a temporary first. A `?` early-return
// here means we never overwrite a live pipeline with a failed build:
// any compile error leaves the running session rendering with the
// previous shader source.
let post = build_post_pipeline(device, self.hw.swap_pixel_format, hr)?;
let text = rebuild_if_live!(
self.text.pipeline_state.is_some(),
build_text_pipeline(device, self.hw.swap_pixel_format, hr)
);
// Pipelines only: nothing here encodes, so the device needs no probe set.
let post_device = MtlPostDevice {
device,
sampler: &self.composite.sampler,
cube_sampler: &self.scene.cube_sampler,
probes: None,
timing: None,
hot_reload: hr,
};
let taa = rebuild_if_live!(
self.taa.pass.is_some(),
concinnity_core::render::post::taa::build_pipeline(&post_device)
);
let bloom = rebuild_if_live!(
self.bloom.is_some(),
concinnity_core::render::post::bloom::build_pipelines(&post_device)
);
// The main pass rebuilds together with the GPU cull and the bindless
// argument encoders, whose layouts come from the same sources. A world
// Shader's pair wins here exactly as it does at init, so a save to an
// engine template does not swap a world's program for the engine's.
// Every bucket's main and pre-pass pipelines rebuild as a pair, so
// shading and the G-buffer never compile from different templates.
let build = BucketBuild {
hot_reload: hr,
..self.bucket_build()
};
let main = rebuild_if_live!(
self.cull.main_pipeline.is_some(),
replacement(
device,
0,
self.world_shader.as_ref(),
build,
self.cull.main_pipeline.as_ref(),
None,
)
);
let world_buckets = self.rebuild_world_buckets(build)?;
let main_cull = rebuild_if_live!(main.is_some(), build_cull_pipeline(device, hr));
// The engine sampler block rides the fresh fragment's encoder.
let main_sampler_args = rebuild_if_live!(
main.is_some(),
build_bindless_sampler_args(
device,
hr,
&self.scene.sampler,
&self.shadow.sampler,
&self.scene.cube_sampler,
)
);
// Hi-Z build kernels are engine built-ins (independent of the world
// shader); rebuild them whenever a Hi-Z resource exists so a saved
// edit to `hiz_build.hlsl` is picked up. The texture + mip views are
// kept: only the pipelines swap.
let hiz_samples = self.targets.hdr.sample_count;
let hiz = rebuild_if_live!(
self.cull.hiz.is_some(),
build_hiz_pipelines(device, hr, hiz_samples)
);
let auto_ev = rebuild_if_live!(
self.auto_exposure.pipelines.is_some(),
build_auto_exposure_pipelines(device, hr)
);
let decal = rebuild_if_live!(
self.decal.pipeline.is_some(),
build_decal_pipeline(device, hr)
);
let fog = rebuild_if_live!(self.fog.pipeline.is_some(), build_fog_pipeline(device, hr));
let sky = super::sky::build_sky_pipeline(device, self.targets.hdr.sample_count, hr)?;
let sky_velocity = super::sky::build_sky_velocity_pipeline(device, hr)?;
// The shadow pipeline needs the static vertex layout.
let static_vdesc = static_vertex_descriptor();
let ssao = rebuild_if_live!(
self.ssao.pass.is_some(),
concinnity_core::render::post::ssao::build_pipelines(&post_device)
);
let ssr_resolve = rebuild_if_live!(
self.ssr.resolve.is_some(),
concinnity_core::render::post::ssr::build_pipeline(&post_device)
);
let reflection_composite = rebuild_if_live!(
self.ssr.composite.is_some(),
concinnity_core::render::post::reflection_composite::build_pipelines(&post_device)
);
let ssgi = rebuild_if_live!(
self.ssgi.pass.is_some(),
concinnity_core::render::post::ssgi::build_pipelines(&post_device)
);
let rt_reflections = rebuild_if_live!(
self.rt.pipelines.resolve.is_some(),
build_rt_reflection_pipeline(
device,
&crate::metal::builtin_shaders::RT_REFLECTIONS_FRAG,
hr
)
);
let rt_reflections_textured = rebuild_if_live!(
self.rt.pipelines.resolve_textured.is_some(),
build_rt_reflection_pipeline(
device,
&crate::metal::builtin_shaders::RT_REFLECTIONS_FRAG_TEXTURED,
hr
)
);
// GPU-driven shadow pipelines: the frustum-only shadow
// decision kernel (from cull.hlsl) + the depth-only bindless shadow
// render pipeline. Both engine-internal, so they rebuild here. Gated on
// the live shadow-bindless path.
let shadow_cull = rebuild_if_live!(
self.cull.shadow_pipeline.is_some(),
build_shadow_cull_pipeline(device, hr)
);
let shadow_bindless = rebuild_if_live!(
self.cull.shadow_bindless_pipeline.is_some(),
build_shadow_bindless_pipeline(device, &static_vdesc, hr)
);
// All builds succeeded: swap into the live context. After this
// point the next frame's draw calls bind the freshly compiled
// pipelines.
self.composite.pipeline = post;
if let (Some(p), Some(pass)) = (bloom, self.bloom.as_mut()) {
pass.swap_pipelines(p);
}
if let Some(p) = text {
self.text.pipeline_state = Some(p);
}
if let (Some(p), Some(taa)) = (taa, self.taa.pass.as_mut()) {
taa.swap_pipeline(p);
}
if let (Some(p), Some(cull), Some(sampler_args)) = (main, main_cull, main_sampler_args) {
self.cull.main_pipeline = Some(p);
self.arg_buffers.bindless_sampler_args = Some(sampler_args);
self.cull.pipeline = Some(cull.decide);
self.cull.pipeline_phase2 = Some(cull.decide_phase2);
self.cull.encode_pipeline = Some(cull.encode);
self.cull.icb_arg_encoder = Some(cull.icb_arg_encoder);
// Force the ICB rebuilds on the next frame so every argument buffer
// is re-encoded with the arg encoder the new encode kernel produced.
// The status buffers and the phase-2 ICB are rebuilt by the same
// `ensure_*_capacity` passes that rebuild the ICBs.
self.cull.icbs = Vec::new();
self.cull.icb_arg_buffer = None;
self.cull.icb_capacity = 0;
self.cull.icbs_2 = Vec::new();
self.cull.icb_2_arg_buffer = None;
self.cull.status_buffer = None;
self.cull.shadow_views = Default::default();
self.cull.spot_views = Default::default();
}
if let Some((init_pipeline, downsample_pipeline)) = hiz
&& let Some(h) = self.cull.hiz.as_mut()
{
h.swap_pipelines(init_pipeline, downsample_pipeline);
}
if let Some(p) = auto_ev {
self.auto_exposure.pipelines = Some(p);
}
if let Some(p) = decal {
self.decal.pipeline = Some(p);
}
if let Some(p) = fog {
self.fog.pipeline = Some(p);
}
self.sky.pipeline = sky;
self.sky.velocity_pipeline = sky_velocity;
if let (Some(p), Some(pass)) = (ssao, self.ssao.pass.as_mut()) {
pass.swap_pipelines(p);
}
if let (Some(p), Some(resolve)) = (ssr_resolve, self.ssr.resolve.as_mut()) {
resolve.swap_pipeline(p);
}
if let (Some(p), Some(pass)) = (reflection_composite, self.ssr.composite.as_mut()) {
pass.swap_pipelines(p);
}
if let (Some(p), Some(pass)) = (ssgi, self.ssgi.pass.as_mut()) {
pass.swap_pipelines(p);
}
if let Some(p) = rt_reflections {
self.rt.pipelines.resolve = Some(p);
}
if let Some(p) = rt_reflections_textured {
self.rt.pipelines.resolve_textured = Some(p);
}
if let Some(p) = shadow_cull {
self.cull.shadow_pipeline = Some(p);
}
if let Some(p) = shadow_bindless {
self.cull.shadow_bindless_pipeline = Some(p);
}
for (bucket, pipelines) in world_buckets {
if let Some(live) = self.cull.world_pipelines.get_mut(bucket) {
*live = pipelines;
}
}
Ok(())
}
// Rebuild bucket 0's pipelines from the world default Shader's freshly
// compiled payload, for [`Self::update_world_shader`] on bucket 0, or swap
// in `prepared` when a worker already built them. The replacements are
// built before the swap, so a typo in a shader edit leaves the live
// pipelines untouched and the session keeps rendering.
//
// Every draw a world Shader reaches goes through the GPU-driven pass, so
// these are the pipelines it owns: the main pass and the G-buffer
// pre-pass, which runs its vertex hook too. The cull ICBs inherit the
// render encoder's pipeline state, so they need no re-encode. The shadow
// and cull pipelines compile from engine-internal source and are covered
// by [`Self::reload_shaders`].
pub(super) fn update_default_world_shader(
&mut self,
programs: &concinnity_core::components::ShaderPrograms,
prepared: Option<BucketPipelines>,
) -> RenderResult<()> {
// A scene-less world never built a main pipeline; there is nothing for
// the fresh world-shader programs to replace.
if self.cull.main_pipeline.is_some() {
let pipelines = replacement(
&self.hw.device,
0,
Some(programs),
self.bucket_build(),
self.cull.main_pipeline.as_ref(),
prepared,
)?;
self.cull.main_pipeline = Some(pipelines);
}
self.world_shader = Some(programs.clone());
Ok(())
}
}