edgevoid/armorpaint
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

metal: fix gpu hang during envmap filtering

Browse Source
This commit is contained in:
luboslenco
2026-01-23 13:58:39 +01:00
parent 8760df13f7
commit ec26d76efa
2 changed files with 28 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
paint/shaders/prefilter_envmap.kong
+3 -3
View File
@@ -1,7 +1,7 @@
#[set(everything)]
const constants: {
params: float4; // level, constants.params.y
params: float4; // level, samples, alpha, pass
};
#[set(everything)]
@@ -65,13 +65,13 @@ fun cos_weighted_hemisphere_direction(n: float3, co: float2, seed: float): float
}
fun prefilter_envmap_frag(input: vert_out): float4 {
var color: float4 = float4(0.0, 0.0, 0.0, 1.0);
var color: float4 = float4(0.0, 0.0, 0.0, constants.params.z);
var n: float3 = reverse_equirect(input.tex);
//for (var i: int = 0; i < int(constants.params.y); i += 1) {
var i: int = 0;
while (i < int(constants.params.y)) {
var dir: float3 = normalize(lerp3(n, cos_weighted_hemisphere_direction(n, input.tex, float(i)), constants.params.x));
var dir: float3 = normalize(lerp3(n, cos_weighted_hemisphere_direction(n, input.tex, float(i + constants.params.w * constants.params.y)), constants.params.x));
var sampled: float3 = sample(radiance, sampler_linear, equirect(dir)).rgb;
sampled = min3(sampled, float3(10.0, 10.0, 10.0));
color.rgb = color.rgb + sampled;
paint/sources/import_envmap.ts
+25 -16
View File
@@ -11,9 +11,11 @@ let import_envmap_mips: gpu_texture_t[] = null;
function import_envmap_run(path: string, image: gpu_texture_t) {
// Init
if (import_envmap_pipeline == null) {
import_envmap_pipeline = gpu_create_pipeline();
import_envmap_pipeline.vertex_shader = sys_get_shader("prefilter_envmap.vert");
import_envmap_pipeline.fragment_shader = sys_get_shader("prefilter_envmap.frag");
import_envmap_pipeline = gpu_create_pipeline();
import_envmap_pipeline.vertex_shader = sys_get_shader("prefilter_envmap.vert");
import_envmap_pipeline.fragment_shader = sys_get_shader("prefilter_envmap.frag");
import_envmap_pipeline.blend_source = gpu_blend_t.SOURCE_ALPHA;
import_envmap_pipeline.blend_destination = gpu_blend_t.ONE;
let vs: gpu_vertex_structure_t = {};
gpu_vertex_struct_add(vs, "pos", gpu_vertex_data_t.F32_2X);
import_envmap_pipeline.input_layout = vs;
@@ -68,22 +70,29 @@ function import_envmap_run(path: string, image: gpu_texture_t) {
}
function import_envmap_get_radiance_mip(mip: gpu_texture_t, level: i32, radiance: gpu_texture_t) {
_gpu_begin(mip);
gpu_set_vertex_buffer(const_data_screen_aligned_vb);
gpu_set_index_buffer(const_data_screen_aligned_ib);
gpu_set_pipeline(import_envmap_pipeline);
import_envmap_params.x = (level + 1) / 10;
/// if (arm_macos || arm_ios)
import_envmap_params.y = 1024 * 2; // Prevent gpu hang
/// if arm_metal
let pass_count: i32 = 8; // 32;
import_envmap_params.y = 512;
/// else
let pass_count: i32 = 1;
import_envmap_params.y = 1024 * 16;
/// end
gpu_set_float4(import_envmap_params_loc, import_envmap_params.x, import_envmap_params.y, import_envmap_params.z, import_envmap_params.w);
gpu_set_texture(import_envmap_radiance_loc, radiance);
let noise: gpu_texture_t = data_get_image("bnoise256.k");
gpu_set_texture(import_envmap_noise_loc, noise);
gpu_draw();
gpu_end();
import_envmap_params.z = 1.0 / pass_count;
import_envmap_params.x = (level + 1) / 10;
for (let i: i32 = 0; i < pass_count; ++i) {
_gpu_begin(mip, null, null, i == 0 ? gpu_clear_t.COLOR : gpu_clear_t.NONE, 0x00000000);
gpu_set_vertex_buffer(const_data_screen_aligned_vb);
gpu_set_index_buffer(const_data_screen_aligned_ib);
gpu_set_pipeline(import_envmap_pipeline);
import_envmap_params.w = i;
gpu_set_float4(import_envmap_params_loc, import_envmap_params.x, import_envmap_params.y, import_envmap_params.z, import_envmap_params.w);
gpu_set_texture(import_envmap_radiance_loc, radiance);
let noise: gpu_texture_t = data_get_image("bnoise256.k");
gpu_set_texture(import_envmap_noise_loc, noise);
gpu_draw();
gpu_end();
}
}
function import_envmap_reverse_equirect(x: f32, y: f32): vec4_t {