Dusk/WeirdUtils
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

V34: use full-float JFA seed render targets

Browse Source
This commit is contained in:
Dusk-92
2026-08-29 10:18:28 +02:00
parent 35b3d00519
commit f81851bb84
1 changed files with 13 additions and 80 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/outline/d3d9_hook.zig
+13 -80
View File
@@ -531,8 +531,8 @@ fn ensureResources(device: *anyopaque) void {
}
debug_resource_stage = 3; // silhouette surface ready
// JFA A RT (G16R16F)
if (deviceCreateTexture(device, vp.Width, vp.Height, 1, types.D3DUSAGE_RENDERTARGET, types.D3DFMT_G16R16F, types.D3DPOOL_DEFAULT, &rt_jfa_a_tex) < 0) {
// V34: JFA A RT (G32R32F) - full-float seed UVs avoid screen-position quantization.
if (deviceCreateTexture(device, vp.Width, vp.Height, 1, types.D3DUSAGE_RENDERTARGET, types.D3DFMT_G32R32F, types.D3DPOOL_DEFAULT, &rt_jfa_a_tex) < 0) {
releaseResources();
return;
}
@@ -544,8 +544,8 @@ fn ensureResources(device: *anyopaque) void {
}
debug_resource_stage = 5; // JFA A surface ready
// JFA B RT (G16R16F)
if (deviceCreateTexture(device, vp.Width, vp.Height, 1, types.D3DUSAGE_RENDERTARGET, types.D3DFMT_G16R16F, types.D3DPOOL_DEFAULT, &rt_jfa_b_tex) < 0) {
// V34: JFA B RT (G32R32F) - full-float seed UVs avoid screen-position quantization.
if (deviceCreateTexture(device, vp.Width, vp.Height, 1, types.D3DUSAGE_RENDERTARGET, types.D3DFMT_G32R32F, types.D3DPOOL_DEFAULT, &rt_jfa_b_tex) < 0) {
releaseResources();
return;
}
@@ -628,81 +628,17 @@ const material_mask_src =
"texkill r1\n" ++
"mov oC0, c0\n";
/// V33 POLISH: JFA init with conservative 3x3 pinhole/crack cleanup.
/// c0.xy = one texel in UV space.
/// A transparent centre pixel is promoted to a seed only when at least 5 of
/// its 8 neighbours are occupied. This fills tiny interior holes/cracks while
/// avoiding the 1px outward dilation caused by a simple max/any-neighbour filter.
/// JFA init: sample silhouette, output own UV as seed or sentinel (-1,-1).
/// Sentinel must be outside [0,1] UV space so it never wins distance comparisons.
const jfa_init_src =
"ps_3_0\n" ++
"def c1, 0.0, 1.0, -0.002, -4.5\n" ++
"def c2, -1.0, -1.0, 0.0, 0.0\n" ++
"def c3, -1.0, 0.0, 0.0, 0.0\n" ++
"def c4, -1.0, 1.0, 0.0, 0.0\n" ++
"def c5, 0.0, -1.0, 0.0, 0.0\n" ++
"def c6, 0.0, 1.0, 0.0, 0.0\n" ++
"def c7, 1.0, -1.0, 0.0, 0.0\n" ++
"def c8, 1.0, 0.0, 0.0, 0.0\n" ++
"def c9, 1.0, 1.0, 0.0, 0.0\n" ++
"def c10, -1.0, -1.0, -0.5, 0.0\n" ++
"def c0, -1.0, -1.0, -0.002, 0.0\n" ++
"dcl_2d s0\n" ++
"dcl_texcoord0 v0\n" ++
"mov r7.xy, c0.xy\n" ++
// Centre occupancy.
"texld r0, v0, s0\n" ++
"add r0.x, r0.a, c1.z\n" ++
"cmp r0.x, r0.x, c1.y, c1.x\n" ++
"mov r6.x, c1.x\n" ++
// Count occupied neighbours.
"mad r4.xy, c2.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
"mad r4.xy, c3.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
"mad r4.xy, c4.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
"mad r4.xy, c5.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
"mad r4.xy, c6.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
"mad r4.xy, c7.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
"mad r4.xy, c8.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
"mad r4.xy, c9.xy, r7.xy, v0.xy\n" ++
"texld r5, r4, s0\n" ++
"add r5.x, r5.a, c1.z\n" ++
"cmp r5.x, r5.x, c1.y, c1.x\n" ++
"add r6.x, r6.x, r5.x\n" ++
// Fill only strongly surrounded empty pixels (5/8 or more).
"add r6.x, r6.x, c1.w\n" ++
"cmp r6.y, r6.x, c1.y, c1.x\n" ++
"max r6.y, r6.y, r0.x\n" ++
// Convert 0/1 to a signed test because cmp treats zero as true.
"add r6.z, r6.y, c10.z\n" ++
// Valid centre/filled pixel becomes its own seed; otherwise sentinel.
"cmp oC0.xy, r6.z, v0.xy, c10.xy\n" ++
"mov oC0.zw, c1.xx\n";
"add r0.x, r0.a, c0.z\n" ++ // alpha - 0.002
"cmp oC0.xy, r0.x, v0.xy, c0.xy\n" ++ // >= 0 → own UV (seed), < 0 → sentinel
"mov oC0.zw, c0.ww\n";
/// JFA propagation: 9-tap (self + 8 neighbors), keeps nearest seed.
/// c0.xy = step_size_uv (set per-pass by CPU).
@@ -1439,18 +1375,15 @@ fn runJfaPipeline(device: *anyopaque) void {
const fw = @as(f32, @floatFromInt(@max(vp.Width, 1)));
const fh = @as(f32, @floatFromInt(@max(vp.Height, 1)));
var c0: [4]f32 = undefined;
// Pass 1: V33 conservative 3x3 cleanup + JFA Init (silhouette → JFA_A).
// Pass 1: JFA Init (silhouette → JFA_A).
deviceSetRenderTarget(device, 0, rt_jfa_a_surf.?);
deviceSetTexture(device, 0, rt_silhouette_tex);
c0 = .{ 1.0 / fw, 1.0 / fh, 0.0, 0.0 };
deviceSetPSConstF(device, 0, &c0);
deviceSetPtr(device, types.VT.SetPixelShader, jfa_init_ps.?);
deviceDrawPrimitiveUP(device, types.D3DPT_TRIANGLESTRIP, 2, @ptrCast(&quad), qstride);
// JFA Propagation: steps [8, 4, 2, 1] ping-ponging between A and B.
deviceSetPtr(device, types.VT.SetPixelShader, jfa_prop_ps.?);
var c0: [4]f32 = undefined;
// step=8 (JFA_A → JFA_B)
deviceSetRenderTarget(device, 0, rt_jfa_b_surf.?);
@@ -1484,7 +1417,7 @@ fn runJfaPipeline(device: *anyopaque) void {
if (saved_rt0) |rt| deviceSetRenderTarget(device, 0, rt);
deviceSetTexture(device, 0, rt_jfa_a_tex);
deviceSetTexture(device, 1, rt_silhouette_tex);
// V32 POLISH: hard 3px edge, no feather.
// V34: hard 3px edge unchanged; JFA seed precision is now full-float.
c0 = [4]f32{ fw, fh, 3.0, 0.0 };
deviceSetPSConstF(device, 0, &c0);
deviceSetPtr(device, types.VT.SetPixelShader, jfa_decode_ps.?);