@woosh/meep-engine 3.55.0 → 3.57.0

package/src/shade/renderer/fluid/pressure/shader_fluid_mg_residual_restrict.d.ts

@@ -1,17 +1,19 @@
 /**
- * The pressure V-cycle's residual and restriction (GPU_FLUID_CASCADES_PROPOSAL_2026_10_16.md §8): one
- * workgroup an 8×8×4 tile of a level, one invocation a cell.
+ * The pressure V-cycle's residual, restriction, and coarse weight reduction
+ * (GPU_FLUID_CASCADES_PROPOSAL_2026_10_16.md §8): one workgroup an 8×8×4 tile of a level, one invocation
+ * a cell.
  *
  * The workgroup loads its tile and halo from `input` as the smoother does, without a prolongation (the
  * finest level's ghosts its parent's pressure, from the step's `ghost_pressure`);
  * each invocation takes its cell's residual (`chunk_fluid_mg_relax`), 0 outside the level or with no
- * degree of freedom, into workgroup memory. After a barrier, the first 32 invocations each write one of
- * the tile's 4×4×2 coarse cells inside the coarser level: `coarse_rhs`, the 8-child mean
- * (`chunk_fluid_mg_restrict`), and `coarse_error`, 0, which the coarser level's down leg starts from.
+ * degree of freedom, and its cell's weight word into workgroup memory. After a barrier, the first 32
+ * invocations each write one of the tile's 4×4×2 coarse cells inside the coarser level: `coarse_rhs`,
+ * the 8-child mean (`chunk_fluid_mg_restrict`), `coarse_error`, 0, which the coarser level's down leg
+ * starts from, and `coarse_words`, the 8-child face-open reduction (`chunk_fluid_mg_coarse_word`).
  *
  * Non-finite input and RHS are read as 0 and counted into `nonfinite[0]`, as the smoother does.
  *
- * Dispatched as `ceil(size / (8, 8, 4))` workgroups. Workgroup memory: 3,424 bytes.
+ * Dispatched as `ceil(size / (8, 8, 4))` workgroups. Workgroup memory: 4,448 bytes.
  */
 export const shader_fluid_mg_residual_restrict: ComputeShader;
 import { ComputeShader } from "../../shader/ComputeShader.js";

package/src/shade/renderer/fluid/pressure/shader_fluid_mg_residual_restrict.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"shader_fluid_mg_residual_restrict.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/fluid/pressure/shader_fluid_mg_residual_restrict.js"],"names":[],"mappings":"AA2BA;;;;;;;;;;;;;;GAcG;AACH,8DA0GG;8BAnJ2B,+BAA+B"}
+{"version":3,"file":"shader_fluid_mg_residual_restrict.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/fluid/pressure/shader_fluid_mg_residual_restrict.js"],"names":[],"mappings":"AA6BA;;;;;;;;;;;;;;;;GAgBG;AACH,8DAuHG;8BApK2B,+BAA+B"}

package/src/shade/renderer/fluid/pressure/shader_fluid_mg_residual_restrict.js

@@ -1,149 +1,166 @@
-import { CodeChunk } from "../../shader/compiler/CodeChunk.js";
-import { ComputeShader } from "../../shader/ComputeShader.js";
-import { chunk_grid_tile_halo_f32 } from "../../shader/chunk/grid/chunk_grid_tile_halo_f32.js";
-import { GRID_TILE_WINDOW_STRUCT } from "../../shader/chunk/grid/GRID_TILE_WINDOW_STRUCT.js";
-import { chunk_index_to_grid3d } from "../../shader/chunk/grid/chunk_index_to_grid3d.js";
-import { chunk_is_finite } from "../../shader/chunk/math/chunk_is_finite.js";
-import { chunk_subgroup_atomic_add } from "../../shader/chunk/subgroup/chunk_subgroup_atomic_add.js";
-import { ShaderResourceSetDescriptor } from "../../shader/resource/ShaderResourceSetDescriptor.js";
-import { atomic_u32, f32, u32 } from "../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
-import { WebGPUArray } from "../../shader/type/WebGPUArray.js";
-import { chunk_fluid_mg_relax } from "./chunk_fluid_mg_relax.js";
-import { chunk_fluid_mg_restrict } from "./chunk_fluid_mg_restrict.js";
-import { chunk_fluid_pressure_ghost } from "./chunk_fluid_pressure_ghost.js";
-import { FLUID_MG_PASS_STRUCT } from "./FLUID_MG_PASS_STRUCT.js";
-
-const resources = new ShaderResourceSetDescriptor();
-
-resources.createGroup()
-    .addUniform('params', FLUID_MG_PASS_STRUCT)
-    .addStorageBuffer('input', WebGPUArray.from(f32), false)
-    .addStorageBuffer('rhs', WebGPUArray.from(f32), false)
-    .addStorageBuffer('words', WebGPUArray.from(u32), false)
-    .addStorageBuffer('coarse_rhs', WebGPUArray.from(f32), true)
-    .addStorageBuffer('coarse_error', WebGPUArray.from(f32), true)
-    .addStorageBuffer('nonfinite', WebGPUArray.from(atomic_u32), true)
-    .addStorageBuffer('ghost_pressure', WebGPUArray.from(f32), false);
-
-/**
- * The pressure V-cycle's residual and restriction (GPU_FLUID_CASCADES_PROPOSAL_2026_10_16.md §8): one
- * workgroup an 8×8×4 tile of a level, one invocation a cell.
- *
- * The workgroup loads its tile and halo from `input` as the smoother does, without a prolongation (the
- * finest level's ghosts its parent's pressure, from the step's `ghost_pressure`);
- * each invocation takes its cell's residual (`chunk_fluid_mg_relax`), 0 outside the level or with no
- * degree of freedom, into workgroup memory. After a barrier, the first 32 invocations each write one of
- * the tile's 4×4×2 coarse cells inside the coarser level: `coarse_rhs`, the 8-child mean
- * (`chunk_fluid_mg_restrict`), and `coarse_error`, 0, which the coarser level's down leg starts from.
- *
- * Non-finite input and RHS are read as 0 and counted into `nonfinite[0]`, as the smoother does.
- *
- * Dispatched as `ceil(size / (8, 8, 4))` workgroups. Workgroup memory: 3,424 bytes.
- */
-export const shader_fluid_mg_residual_restrict = ComputeShader.from({
-    label: 'Fluid / pressure / residual + restrict',
-    resources,
-    body: CodeChunk.from(
-        //language=WGSL
-        `
-const FLUID_MG_SOURCE_PRESSURE = 0u;
-const FLUID_MG_SOURCE_ERROR = 1u;
-
-var<workgroup> tile_p: array<f32, GRID_TILE_HALO_CELLS>;
-var<workgroup> tile_r: array<f32, GRID_TILE_THREADS>;
-
-fn fluid_mg_input_index(storage_cell: vec3<u32>) -> u32 {
-    return storage_cell.x + params.input_period.x * (storage_cell.y + params.input_period.y * storage_cell.z);
-}
-
-fn grid_tile_f32_fetch_inside(source: u32, storage_cell: vec3<u32>) -> f32 {
-    let value = input[fluid_mg_input_index(storage_cell)];
-
-    return select(0.0, value, is_finite(value));
-}
-
-fn grid_tile_f32_fetch_outside(source: u32, world_cell: vec3<i32>) -> f32 {
-    if (source == FLUID_MG_SOURCE_PRESSURE) {
-        return fluid_pressure_ghost(world_cell);
-    }
-
-    return 0.0;
-}
-
-
-// where the finest level's window is, for its pressure ghosts
-fn fluid_pressure_ghost_corner() -> vec3<i32> {
-    return params.corner;
-}
-
-fn fluid_pressure_ghost_size() -> vec3<u32> {
-    return params.size;
-}
-fn tile_read_p(local: vec3<i32>) -> f32 {
-    return grid_tile_f32_read(&tile_p, local);
-}
-
-// a tile-local interior cell's residual
-fn tile_read_r(local: vec3<i32>) -> f32 {
-    let c = vec3<u32>(local);
-
-    return tile_r[c.x + 8u * (c.y + 8u * c.z)];
-}
-
-@compute @workgroup_size(8, 8, 4)
-fn main(
-    @builtin(workgroup_id) group: vec3<u32>,
-    @builtin(local_invocation_id) local_id: vec3<u32>,
-    @builtin(local_invocation_index) lane: u32,
-) {
-    let window = ${GRID_TILE_WINDOW_STRUCT.wgsl_ref}(params.corner, params.input_corner_storage, params.size, params.input_period);
-    let source = select(FLUID_MG_SOURCE_ERROR, FLUID_MG_SOURCE_PRESSURE, params.level == 0u);
-    let tile_origin = vec3<i32>(group) * GRID_TILE_SIZE;
-
-    grid_tile_f32_load(&tile_p, window, source, tile_origin, lane);
-
-    let local = vec3<i32>(local_id);
-    let cell = vec3<u32>(tile_origin + local);
-
-    var residual = 0.0;
-    var zeroed = 0u;
-
-    if (all(cell < params.size)) {
-        let linear = cell.x + params.size.x * (cell.y + params.size.y * cell.z);
-        let rhs_value = rhs[linear];
-        let held = input[fluid_mg_input_index(grid_tile_window_storage(vec3<i32>(cell), params.input_corner_storage, params.input_period))];
-        let f = select(0.0, rhs_value, is_finite(rhs_value));
-        let stencil = fluid_mg_stencil(local, cell, params.size, words[linear], params.edge_weight);
-
-        residual = fluid_mg_residual(tile_read_p(local), stencil, f, params.h2);
-        zeroed = select(0u, 1u, !is_finite(held)) + select(0u, 1u, !is_finite(rhs_value));
-    }
-
-    tile_r[lane] = residual;
-
-    subgroup_atomic_add_u32(&nonfinite[0], zeroed);
-
-    workgroupBarrier();
-
-    if (lane < 32u) {
-        let coarse_local = vec3<i32>(index_to_grid3d(lane, vec2<u32>(4u, 4u)));
-        let coarse_cell = vec3<u32>(tile_origin / 2 + coarse_local);
-
-        if (all(coarse_cell < params.coarse_size)) {
-            let coarse_linear = coarse_cell.x + params.coarse_size.x * (coarse_cell.y + params.coarse_size.y * coarse_cell.z);
-
-            coarse_rhs[coarse_linear] = fluid_mg_restrict(coarse_local);
-            coarse_error[coarse_linear] = 0.0;
-        }
-    }
-}
-`, [
-            chunk_grid_tile_halo_f32,
-            chunk_fluid_mg_relax,
-            chunk_fluid_mg_restrict,
-            chunk_fluid_pressure_ghost,
-            chunk_index_to_grid3d,
-            chunk_is_finite,
-            chunk_subgroup_atomic_add,
-        ]),
-});
+import { CodeChunk } from "../../shader/compiler/CodeChunk.js";
+import { ComputeShader } from "../../shader/ComputeShader.js";
+import { chunk_grid_tile_halo_f32 } from "../../shader/chunk/grid/chunk_grid_tile_halo_f32.js";
+import { GRID_TILE_WINDOW_STRUCT } from "../../shader/chunk/grid/GRID_TILE_WINDOW_STRUCT.js";
+import { chunk_index_to_grid3d } from "../../shader/chunk/grid/chunk_index_to_grid3d.js";
+import { chunk_is_finite } from "../../shader/chunk/math/chunk_is_finite.js";
+import { chunk_subgroup_atomic_add } from "../../shader/chunk/subgroup/chunk_subgroup_atomic_add.js";
+import { ShaderResourceSetDescriptor } from "../../shader/resource/ShaderResourceSetDescriptor.js";
+import { atomic_u32, f32, u32 } from "../../shader/type/WEBGPU_WGSL_PRIMITIVE_TYPES.js";
+import { WebGPUArray } from "../../shader/type/WebGPUArray.js";
+import { chunk_fluid_mg_coarse_word } from "./chunk_fluid_mg_coarse_word.js";
+import { chunk_fluid_mg_relax } from "./chunk_fluid_mg_relax.js";
+import { chunk_fluid_mg_restrict } from "./chunk_fluid_mg_restrict.js";
+import { chunk_fluid_pressure_ghost } from "./chunk_fluid_pressure_ghost.js";
+import { FLUID_MG_PASS_STRUCT } from "./FLUID_MG_PASS_STRUCT.js";
+
+const resources = new ShaderResourceSetDescriptor();
+
+resources.createGroup()
+    .addUniform('params', FLUID_MG_PASS_STRUCT)
+    .addStorageBuffer('input', WebGPUArray.from(f32), false)
+    .addStorageBuffer('rhs', WebGPUArray.from(f32), false)
+    .addStorageBuffer('words', WebGPUArray.from(u32), false)
+    .addStorageBuffer('coarse_rhs', WebGPUArray.from(f32), true)
+    .addStorageBuffer('coarse_error', WebGPUArray.from(f32), true)
+    .addStorageBuffer('coarse_words', WebGPUArray.from(u32), true)
+    .addStorageBuffer('nonfinite', WebGPUArray.from(atomic_u32), true)
+    .addStorageBuffer('ghost_pressure', WebGPUArray.from(f32), false);
+
+/**
+ * The pressure V-cycle's residual, restriction, and coarse weight reduction
+ * (GPU_FLUID_CASCADES_PROPOSAL_2026_10_16.md §8): one workgroup an 8×8×4 tile of a level, one invocation
+ * a cell.
+ *
+ * The workgroup loads its tile and halo from `input` as the smoother does, without a prolongation (the
+ * finest level's ghosts its parent's pressure, from the step's `ghost_pressure`);
+ * each invocation takes its cell's residual (`chunk_fluid_mg_relax`), 0 outside the level or with no
+ * degree of freedom, and its cell's weight word into workgroup memory. After a barrier, the first 32
+ * invocations each write one of the tile's 4×4×2 coarse cells inside the coarser level: `coarse_rhs`,
+ * the 8-child mean (`chunk_fluid_mg_restrict`), `coarse_error`, 0, which the coarser level's down leg
+ * starts from, and `coarse_words`, the 8-child face-open reduction (`chunk_fluid_mg_coarse_word`).
+ *
+ * Non-finite input and RHS are read as 0 and counted into `nonfinite[0]`, as the smoother does.
+ *
+ * Dispatched as `ceil(size / (8, 8, 4))` workgroups. Workgroup memory: 4,448 bytes.
+ */
+export const shader_fluid_mg_residual_restrict = ComputeShader.from({
+    label: 'Fluid / pressure / residual + restrict',
+    resources,
+    body: CodeChunk.from(
+        //language=WGSL
+        `
+const FLUID_MG_SOURCE_PRESSURE = 0u;
+const FLUID_MG_SOURCE_ERROR = 1u;
+
+var<workgroup> tile_p: array<f32, GRID_TILE_HALO_CELLS>;
+var<workgroup> tile_r: array<f32, GRID_TILE_THREADS>;
+var<workgroup> tile_w: array<u32, GRID_TILE_THREADS>;
+
+fn fluid_mg_input_index(storage_cell: vec3<u32>) -> u32 {
+    return storage_cell.x + params.input_period.x * (storage_cell.y + params.input_period.y * storage_cell.z);
+}
+
+fn grid_tile_f32_fetch_inside(source: u32, storage_cell: vec3<u32>) -> f32 {
+    let value = input[fluid_mg_input_index(storage_cell)];
+
+    return select(0.0, value, is_finite(value));
+}
+
+fn grid_tile_f32_fetch_outside(source: u32, world_cell: vec3<i32>) -> f32 {
+    if (source == FLUID_MG_SOURCE_PRESSURE) {
+        return fluid_pressure_ghost(world_cell);
+    }
+
+    return 0.0;
+}
+
+
+// where the finest level's window is, for its pressure ghosts
+fn fluid_pressure_ghost_corner() -> vec3<i32> {
+    return params.corner;
+}
+
+fn fluid_pressure_ghost_size() -> vec3<u32> {
+    return params.size;
+}
+fn tile_read_p(local: vec3<i32>) -> f32 {
+    return grid_tile_f32_read(&tile_p, local);
+}
+
+// a tile-local interior cell's residual
+fn tile_read_r(local: vec3<i32>) -> f32 {
+    let c = vec3<u32>(local);
+
+    return tile_r[c.x + 8u * (c.y + 8u * c.z)];
+}
+
+// a tile-local interior cell's weight word
+fn fluid_mg_read_child_word(local: vec3<i32>) -> u32 {
+    let c = vec3<u32>(local);
+
+    return tile_w[c.x + 8u * (c.y + 8u * c.z)];
+}
+
+@compute @workgroup_size(8, 8, 4)
+fn main(
+    @builtin(workgroup_id) group: vec3<u32>,
+    @builtin(local_invocation_id) local_id: vec3<u32>,
+    @builtin(local_invocation_index) lane: u32,
+) {
+    let window = ${GRID_TILE_WINDOW_STRUCT.wgsl_ref}(params.corner, params.input_corner_storage, params.size, params.input_period);
+    let source = select(FLUID_MG_SOURCE_ERROR, FLUID_MG_SOURCE_PRESSURE, params.level == 0u);
+    let tile_origin = vec3<i32>(group) * GRID_TILE_SIZE;
+
+    grid_tile_f32_load(&tile_p, window, source, tile_origin, lane);
+
+    let local = vec3<i32>(local_id);
+    let cell = vec3<u32>(tile_origin + local);
+
+    var residual = 0.0;
+    var word = 0u;
+    var zeroed = 0u;
+
+    if (all(cell < params.size)) {
+        let linear = cell.x + params.size.x * (cell.y + params.size.y * cell.z);
+        let rhs_value = rhs[linear];
+        let held = input[fluid_mg_input_index(grid_tile_window_storage(vec3<i32>(cell), params.input_corner_storage, params.input_period))];
+        let f = select(0.0, rhs_value, is_finite(rhs_value));
+        word = words[linear];
+        let stencil = fluid_mg_stencil(local, cell, params.size, word, params.edge_weight);
+
+        residual = fluid_mg_residual(tile_read_p(local), stencil, f, params.h2);
+        zeroed = select(0u, 1u, !is_finite(held)) + select(0u, 1u, !is_finite(rhs_value));
+    }
+
+    tile_r[lane] = residual;
+    tile_w[lane] = word;
+
+    subgroup_atomic_add_u32(&nonfinite[0], zeroed);
+
+    workgroupBarrier();
+
+    if (lane < 32u) {
+        let coarse_local = vec3<i32>(index_to_grid3d(lane, vec2<u32>(4u, 4u)));
+        let coarse_cell = vec3<u32>(tile_origin / 2 + coarse_local);
+
+        if (all(coarse_cell < params.coarse_size)) {
+            let coarse_linear = coarse_cell.x + params.coarse_size.x * (coarse_cell.y + params.coarse_size.y * coarse_cell.z);
+
+            coarse_rhs[coarse_linear] = fluid_mg_restrict(coarse_local);
+            coarse_error[coarse_linear] = 0.0;
+            coarse_words[coarse_linear] = fluid_mg_coarse_word(coarse_local);
+        }
+    }
+}
+`, [
+            chunk_grid_tile_halo_f32,
+            chunk_fluid_mg_relax,
+            chunk_fluid_mg_restrict,
+            chunk_fluid_mg_coarse_word,
+            chunk_fluid_pressure_ghost,
+            chunk_index_to_grid3d,
+            chunk_is_finite,
+            chunk_subgroup_atomic_add,
+        ]),
+});

package/src/shade/renderer/fluid/pressure/shader_fluid_mg_smooth.d.ts

@@ -3,11 +3,13 @@
  * workgroup an 8×8×4 tile of a level, one invocation a cell.
  *
  * 1. The workgroup loads its tile and a one-cell halo from `input` into workgroup memory
- *    (`chunk_grid_tile_halo_f32`). A value that is not finite is loaded as 0. With `prolong`, every
- *    value inside the level is loaded with the coarser level's error added trilinearly
- *    (`chunk_fluid_mg_prolong`), which is the separate prolongation pass folded in. A cell beyond the
- *    level is its ghost: the cascade's Dirichlet pressure on the finest level, its parent's at the
- *    ghost cell's centre (`chunk_fluid_pressure_ghost`, from the step's `ghost_pressure`), 0 below it.
+ *    (`chunk_grid_tile_halo_f32`). A value that is not finite is loaded as 0. With `prolong`, the 6×6×4
+ *    coarse error neighbourhood covering the tile and its halo is first staged into workgroup memory
+ *    (`tile_coarse`, 144 cells, out-of-bounds coarse cells zeroed), and every value inside the level is
+ *    loaded with the coarser level's error added trilinearly (`fluid_mg_prolong_cached`), which is the
+ *    separate prolongation pass folded in. A cell beyond the level is its ghost: the cascade's Dirichlet
+ *    pressure on the finest level, its parent's at the ghost cell's centre (`chunk_fluid_pressure_ghost`,
+ *    from the step's `ghost_pressure`), 0 below it.
  * 2. `sweeps` times, for red then black, each cell of a colour in `colours` with a degree of freedom
  *    relaxes toward its equation (`chunk_fluid_mg_relax`) against its neighbours in the tile, the
  *    halo frozen. A colour reads only the other colour, so its cells update together, and a barrier
@@ -23,7 +25,7 @@
  * Each invocation counts its own cell's input and RHS when not finite, and the workgroup adds the counts
  * to `nonfinite[0]` with one atomic per subgroup.
  *
- * Dispatched as `ceil(size / (8, 8, 4))` workgroups. Workgroup memory: 2,400 bytes.
+ * Dispatched as `ceil(size / (8, 8, 4))` workgroups. Workgroup memory: 2,976 bytes.
  */
 export const shader_fluid_mg_smooth: ComputeShader;
 import { ComputeShader } from "../../shader/ComputeShader.js";

package/src/shade/renderer/fluid/pressure/shader_fluid_mg_smooth.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"shader_fluid_mg_smooth.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/fluid/pressure/shader_fluid_mg_smooth.js"],"names":[],"mappings":"AA2BA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,mDAgIG;8BArL2B,+BAA+B"}
+{"version":3,"file":"shader_fluid_mg_smooth.d.ts","sourceRoot":"","sources":["../../../../../../src/shade/renderer/fluid/pressure/shader_fluid_mg_smooth.js"],"names":[],"mappings":"AA2BA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,mDAkKG;8BAzN2B,+BAA+B"}