A game about forced loneliness, made by TACStudios
tacstudios.tngl.sh
1#pragma kernel StpSetup
2
3#pragma multi_compile _ ENABLE_DEBUG_MODE
4#pragma multi_compile _ ENABLE_STENCIL_RESPONSIVE
5#pragma multi_compile _ ENABLE_LARGE_KERNEL
6
7#pragma multi_compile _ UNITY_DEVICE_SUPPORTS_NATIVE_16BIT
8
9// TODO: Re-enable support for wave reductions (usage of UNITY_DEVICE_SUPPORTS_WAVE_ANY keyword)
10//
11// We've run into many platform specific problems when trying to use wave operations for STP's reductions so they're being
12// disabled for now. Enabling support for wave operations also causes us to use DXC on the 32-bit path on some Qualcomm Android
13// devices and this triggers visual artifacts that we have no other way to work around at the moment.
14
15#pragma multi_compile _ DISABLE_TEXTURE2D_X_ARRAY
16
17#pragma only_renderers d3d11 playstation xboxone xboxseries vulkan metal switch
18
19#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Common.hlsl"
20#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Color.hlsl"
21#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/UnityInstancing.hlsl"
22
23#define STP_PAT 1
24
25#include "Packages/com.unity.render-pipelines.core/Runtime/STP/StpCommon.hlsl"
26
27//
28// Input
29//
30
31TEXTURE2D_X(_StpInputColor);
32TEXTURE2D_X(_StpInputDepth);
33TEXTURE2D_X(_StpInputMotion);
34
35#if defined(ENABLE_STENCIL_RESPONSIVE)
36 TYPED_TEXTURE2D_X(uint2, _StpInputStencil);
37#endif
38
39//
40// Intermediate Output
41//
42
43RW_TEXTURE2D_X(float4, _StpIntermediateColor);
44RW_TEXTURE2D_X(float, _StpIntermediateConvergence);
45
46//
47// History Input/Output
48//
49
50TYPED_TEXTURE2D_X(uint, _StpPriorDepthMotion);
51RW_TEXTURE2D_X(uint, _StpDepthMotion);
52
53TEXTURE2D_X(_StpPriorLuma);
54RW_TEXTURE2D_X(float2, _StpLuma);
55
56TEXTURE2D_X(_StpPriorConvergence);
57
58TEXTURE2D_X(_StpPriorFeedback);
59
60#define STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET (SLICE_ARRAY_INDEX * STPSETUPPERVIEWCONSTANTS_COUNT)
61
62#if defined(SHADER_API_PSSL) || defined(SHADER_API_SWITCH) || (defined(SHADER_API_METAL) && !defined(SHADER_API_MOBILE))
63 // Force usage of the 32-bit reduction path even in 16-bit environments
64 #define STP_FORCE_32BIT_REDUCTION
65#endif
66
67#if defined(SHADER_API_PSSL) || defined(SHADER_API_GAMECORE) || defined(SHADER_API_METAL) || (defined(SHADER_API_VULKAN) && defined(SHADER_API_MOBILE))
68 // Force usage of group shared memory instead using wave operations
69 #define STP_FORCE_GROUPSHARED
70#endif
71
72// Enable the use of wave operations when they're supported by the current hardware and usage of groupshared hasn't been forced.
73#if defined(UNITY_HW_SUPPORTS_WAVE) && !defined(STP_FORCE_GROUPSHARED)
74 #define STP_ENABLE_WAVEOPS
75#endif
76
77// STP requires a 4x4 reduction which must be implemented by either wave operations, or group shared memory.
78#if !defined(STP_ENABLE_WAVEOPS)
79#if defined(STP_16BIT) && !defined(STP_FORCE_32BIT_REDUCTION)
80groupshared uint4 gs_StpScratch[STP_GROUP_SIZE];
81#else
82groupshared float4 gs_StpScratch[STP_GROUP_SIZE * 2];
83#endif
84#endif
85
86// In some cases, we have to expose the 32-bit reduction code in the 16-bit path
87#if defined(STP_32BIT) || defined(STP_FORCE_32BIT_REDUCTION)
88void StpPat4x4MaxF8(StpMU1 i, inout StpF4 a, inout StpF4 b)
89{
90#if defined(STP_ENABLE_WAVEOPS)
91 a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 1)));
92 a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 1)));
93 a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 1)));
94 a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 1)));
95 b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 1)));
96 b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 1)));
97 b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 1)));
98 b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 1)));
99
100 a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 2)));
101 a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 2)));
102 a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 2)));
103 a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 2)));
104 b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 2)));
105 b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 2)));
106 b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 2)));
107 b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 2)));
108
109 a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 4)));
110 a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 4)));
111 a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 4)));
112 a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 4)));
113 b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 4)));
114 b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 4)));
115 b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 4)));
116 b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 4)));
117
118 a.x = max(a.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 8)));
119 a.y = max(a.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 8)));
120 a.z = max(a.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 8)));
121 a.w = max(a.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 8)));
122 b.x = max(b.x, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.x), WaveGetLaneIndex() ^ 8)));
123 b.y = max(b.y, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.y), WaveGetLaneIndex() ^ 8)));
124 b.z = max(b.z, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.z), WaveGetLaneIndex() ^ 8)));
125 b.w = max(b.w, StpF1_U1(WaveReadLaneAt(StpU1_F1(b.w), WaveGetLaneIndex() ^ 8)));
126#else
127 gs_StpScratch[i] = a;
128 gs_StpScratch[i + STP_GROUP_SIZE] = b;
129
130 GroupMemoryBarrierWithGroupSync();
131
132 // 2x2 Reduction
133 {
134 StpMU1 offset = (i & ~StpMU1(3));
135
136 StpMU1 a0 = offset + ((i + StpMU1(1)) & StpMU1(3));
137 StpMU1 a1 = offset + ((i + StpMU1(2)) & StpMU1(3));
138 StpMU1 a2 = offset + ((i + StpMU1(3)) & StpMU1(3));
139
140 float4 x0 = gs_StpScratch[a0];
141 float4 x1 = gs_StpScratch[a1];
142 float4 x2 = gs_StpScratch[a2];
143
144 float4 y0 = gs_StpScratch[a0 + STP_GROUP_SIZE];
145 float4 y1 = gs_StpScratch[a1 + STP_GROUP_SIZE];
146 float4 y2 = gs_StpScratch[a2 + STP_GROUP_SIZE];
147
148 GroupMemoryBarrierWithGroupSync();
149
150 a = max(max(max(a, x0), x1), x2);
151 b = max(max(max(b, y0), y1), y2);
152 }
153
154 gs_StpScratch[i] = a;
155 gs_StpScratch[i + STP_GROUP_SIZE] = b;
156
157 GroupMemoryBarrierWithGroupSync();
158
159 // 4x4 Reduction
160 {
161 StpMU1 offset = (i & ~StpMU1(15));
162
163 StpMU1 a0 = offset + ((i + StpMU1(4)) & StpMU1(15));
164 StpMU1 a1 = offset + ((i + StpMU1(8)) & StpMU1(15));
165 StpMU1 a2 = offset + ((i + StpMU1(12)) & StpMU1(15));
166
167 float4 x0 = gs_StpScratch[a0];
168 float4 x1 = gs_StpScratch[a1];
169 float4 x2 = gs_StpScratch[a2];
170
171 float4 y0 = gs_StpScratch[a0 + STP_GROUP_SIZE];
172 float4 y1 = gs_StpScratch[a1 + STP_GROUP_SIZE];
173 float4 y2 = gs_StpScratch[a2 + STP_GROUP_SIZE];
174
175 GroupMemoryBarrierWithGroupSync();
176
177 a = max(max(max(a, x0), x1), x2);
178 b = max(max(max(b, y0), y1), y2);
179 }
180#endif
181}
182void StpPat4x4SumF4(StpMU1 i, inout StpF4 a)
183{
184#if defined(STP_ENABLE_WAVEOPS)
185 a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 1));
186 a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 1));
187 a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 1));
188 a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 1));
189
190 a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 2));
191 a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 2));
192 a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 2));
193 a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 2));
194
195 a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 4));
196 a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 4));
197 a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 4));
198 a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 4));
199
200 a.x += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.x), WaveGetLaneIndex() ^ 8));
201 a.y += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.y), WaveGetLaneIndex() ^ 8));
202 a.z += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.z), WaveGetLaneIndex() ^ 8));
203 a.w += StpF1_U1(WaveReadLaneAt(StpU1_F1(a.w), WaveGetLaneIndex() ^ 8));
204#else
205 gs_StpScratch[i] = a;
206
207 GroupMemoryBarrierWithGroupSync();
208
209 // 2x2 Reduction
210 {
211 StpMU1 offset = (i & ~StpMU1(3));
212
213 StpMU1 a0 = offset + ((i + StpMU1(1)) & StpMU1(3));
214 StpMU1 a1 = offset + ((i + StpMU1(2)) & StpMU1(3));
215 StpMU1 a2 = offset + ((i + StpMU1(3)) & StpMU1(3));
216
217 float4 x0 = gs_StpScratch[a0];
218 float4 x1 = gs_StpScratch[a1];
219 float4 x2 = gs_StpScratch[a2];
220
221 GroupMemoryBarrierWithGroupSync();
222
223 a = a + x0 + x1 + x2;
224 }
225
226 gs_StpScratch[i] = a;
227
228 GroupMemoryBarrierWithGroupSync();
229
230 // 4x4 Reduction
231 {
232 StpMU1 offset = (i & ~StpMU1(15));
233
234 StpMU1 a0 = offset + ((i + StpMU1(4)) & StpMU1(15));
235 StpMU1 a1 = offset + ((i + StpMU1(8)) & StpMU1(15));
236 StpMU1 a2 = offset + ((i + StpMU1(12)) & StpMU1(15));
237
238 float4 x0 = gs_StpScratch[a0];
239 float4 x1 = gs_StpScratch[a1];
240 float4 x2 = gs_StpScratch[a2];
241
242 GroupMemoryBarrierWithGroupSync();
243
244 a = a + x0 + x1 + x2;
245 }
246#endif
247}
248#endif
249
250#if defined(STP_16BIT)
251void StpPat4x4MaxH8(StpW1 i, inout StpH4 a, inout StpH4 b)
252{
253#if defined(STP_FORCE_32BIT_REDUCTION)
254 StpPat4x4MaxF8(i, a, b);
255#else
256#if defined(STP_ENABLE_WAVEOPS)
257 a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 1)));
258 a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 1)));
259 b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 1)));
260 b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 1)));
261
262 a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 2)));
263 a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 2)));
264 b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 2)));
265 b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 2)));
266
267 a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 4)));
268 a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 4)));
269 b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 4)));
270 b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 4)));
271
272 a.xy = max(a.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 8)));
273 a.zw = max(a.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 8)));
274 b.xy = max(b.xy, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.xy), WaveGetLaneIndex() ^ 8)));
275 b.zw = max(b.zw, StpH2_U1(WaveReadLaneAt(StpU1_H2(b.zw), WaveGetLaneIndex() ^ 8)));
276#else
277 gs_StpScratch[i] = StpU4(StpU1_H2(a.xy), StpU1_H2(a.zw), StpU1_H2(b.xy), StpU1_H2(b.zw));
278
279 GroupMemoryBarrierWithGroupSync();
280
281 // 2x2 Reduction
282 {
283 StpW1 offset = (i & ~StpW1(3));
284
285 StpW1 a0 = offset + ((i + StpW1(1)) & StpW1(3));
286 StpW1 a1 = offset + ((i + StpW1(2)) & StpW1(3));
287 StpW1 a2 = offset + ((i + StpW1(3)) & StpW1(3));
288
289 uint4 x0 = gs_StpScratch[a0];
290 uint4 x1 = gs_StpScratch[a1];
291 uint4 x2 = gs_StpScratch[a2];
292
293 GroupMemoryBarrierWithGroupSync();
294
295 a.xy = max(max(max(a.xy, StpH2_U1(x0.x)), StpH2_U1(x1.x)), StpH2_U1(x2.x));
296 a.zw = max(max(max(a.zw, StpH2_U1(x0.y)), StpH2_U1(x1.y)), StpH2_U1(x2.y));
297 b.xy = max(max(max(b.xy, StpH2_U1(x0.z)), StpH2_U1(x1.z)), StpH2_U1(x2.z));
298 b.zw = max(max(max(b.zw, StpH2_U1(x0.w)), StpH2_U1(x1.w)), StpH2_U1(x2.w));
299 }
300
301 gs_StpScratch[i] = StpU4(StpU1_H2(a.xy), StpU1_H2(a.zw), StpU1_H2(b.xy), StpU1_H2(b.zw));
302
303 GroupMemoryBarrierWithGroupSync();
304
305 // 4x4 Reduction
306 {
307 StpW1 offset = (i & ~StpW1(15));
308
309 StpW1 a0 = offset + ((i + StpW1(4)) & StpW1(15));
310 StpW1 a1 = offset + ((i + StpW1(8)) & StpW1(15));
311 StpW1 a2 = offset + ((i + StpW1(12)) & StpW1(15));
312
313 uint4 x0 = gs_StpScratch[a0];
314 uint4 x1 = gs_StpScratch[a1];
315 uint4 x2 = gs_StpScratch[a2];
316
317 GroupMemoryBarrierWithGroupSync();
318
319 a.xy = max(max(max(a.xy, StpH2_U1(x0.x)), StpH2_U1(x1.x)), StpH2_U1(x2.x));
320 a.zw = max(max(max(a.zw, StpH2_U1(x0.y)), StpH2_U1(x1.y)), StpH2_U1(x2.y));
321 b.xy = max(max(max(b.xy, StpH2_U1(x0.z)), StpH2_U1(x1.z)), StpH2_U1(x2.z));
322 b.zw = max(max(max(b.zw, StpH2_U1(x0.w)), StpH2_U1(x1.w)), StpH2_U1(x2.w));
323 }
324#endif
325#endif
326}
327void StpPat4x4SumH4(StpW1 i, inout StpH4 a)
328{
329#if defined(STP_FORCE_32BIT_REDUCTION)
330 StpPat4x4SumF4(i, a);
331#else
332#if defined(STP_ENABLE_WAVEOPS)
333 a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 1));
334 a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 1));
335
336 a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 2));
337 a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 2));
338
339 a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 4));
340 a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 4));
341
342 a.xy += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.xy), WaveGetLaneIndex() ^ 8));
343 a.zw += StpH2_U1(WaveReadLaneAt(StpU1_H2(a.zw), WaveGetLaneIndex() ^ 8));
344#else
345 gs_StpScratch[i].xy = StpU2(StpU1_H2(a.xy), StpU1_H2(a.zw));
346
347 GroupMemoryBarrierWithGroupSync();
348
349 // 2x2 Reduction
350 {
351 StpW1 offset = (i & ~StpW1(3));
352
353 StpW1 a0 = offset + ((i + StpW1(1)) & StpW1(3));
354 StpW1 a1 = offset + ((i + StpW1(2)) & StpW1(3));
355 StpW1 a2 = offset + ((i + StpW1(3)) & StpW1(3));
356
357 uint2 x0 = gs_StpScratch[a0].xy;
358 uint2 x1 = gs_StpScratch[a1].xy;
359 uint2 x2 = gs_StpScratch[a2].xy;
360
361 GroupMemoryBarrierWithGroupSync();
362
363 a.xy = a.xy + StpH2_U1(x0.x) + StpH2_U1(x1.x) + StpH2_U1(x2.x);
364 a.zw = a.zw + StpH2_U1(x0.y) + StpH2_U1(x1.y) + StpH2_U1(x2.y);
365 }
366
367 gs_StpScratch[i].xy = StpU2(StpU1_H2(a.xy), StpU1_H2(a.zw));
368
369 GroupMemoryBarrierWithGroupSync();
370
371 // 4x4 Reduction
372 {
373 StpW1 offset = (i & ~StpW1(15));
374
375 StpW1 a0 = offset + ((i + StpW1(4)) & StpW1(15));
376 StpW1 a1 = offset + ((i + StpW1(8)) & StpW1(15));
377 StpW1 a2 = offset + ((i + StpW1(12)) & StpW1(15));
378
379 uint2 x0 = gs_StpScratch[a0].xy;
380 uint2 x1 = gs_StpScratch[a1].xy;
381 uint2 x2 = gs_StpScratch[a2].xy;
382
383 GroupMemoryBarrierWithGroupSync();
384
385 a.xy = a.xy + StpH2_U1(x0.x) + StpH2_U1(x1.x) + StpH2_U1(x2.x);
386 a.zw = a.zw + StpH2_U1(x0.y) + StpH2_U1(x1.y) + StpH2_U1(x2.y);
387 }
388#endif
389#endif
390}
391StpH1 StpPatPriConH(StpF2 p) { return (StpH1)SAMPLE_TEXTURE2D_X_LOD(_StpPriorConvergence, s_linear_clamp_sampler, p, 0); }
392
393// These are separate to support inline operation (pass merged instead of loads).
394StpF2 StpPatDatMotH(StpW2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputMotion, o, 0).xy; }
395StpH3 StpPatDatColH(StpW2 o) { return (StpH3)LOAD_TEXTURE2D_X_LOD(_StpInputColor, o, 0).rgb; }
396StpF1 StpPatDatZH(StpW2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputDepth, o, 0).x; }
397// This provides a place to convert Z from depth to linear if not inlined and actually loaded.
398StpF1 StpPatFixZH(StpF1 z) { return 1.0 / (STP_ZBUFFER_PARAMS_Z * z + STP_ZBUFFER_PARAMS_W); }
399StpU1 StpPatDatRH(StpW2 o) {
400 #if defined(ENABLE_STENCIL_RESPONSIVE)
401 return GetStencilValue(LOAD_TEXTURE2D_X_LOD(_StpInputStencil, o, 0).xy);
402 #endif // defined(ENABLE_STENCIL_RESPONSIVE)
403 return StpU1_(0); }
404StpH1 StpPatFixRH(StpU1 v) {
405 // Activate the "responsive" feature when we don't have valid history textures.
406 bool hasValidHistory = DecodeHasValidHistory(STP_COMMON_CONSTANT);
407 bool excludeTaa = false;
408 #if defined(ENABLE_STENCIL_RESPONSIVE)
409 excludeTaa = (v & DecodeStencilMask(STP_COMMON_CONSTANT)) != 0;
410 #endif // defined(ENABLE_STENCIL_RESPONSIVE)
411 return (hasValidHistory && !excludeTaa) ? StpH1_(1.0) : StpH1_(0.0); }
412
413StpH1 StpPatDitH(StpW2 o) { return StpDitH1(o); }
414StpH4 StpPatPriFedH(StpF2 p) { return (StpH4)SAMPLE_TEXTURE2D_X_LOD(_StpPriorFeedback, s_linear_clamp_sampler, p, 0); }
415StpH4 StpPatPriFedR4H(StpF2 p) { return (StpH4)GATHER_RED_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
416StpH4 StpPatPriFedG4H(StpF2 p) { return (StpH4)GATHER_GREEN_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
417StpH4 StpPatPriFedB4H(StpF2 p) { return (StpH4)GATHER_BLUE_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
418StpH2 StpPatPriLumH(StpF2 p) { return (StpH2)SAMPLE_TEXTURE2D_X_LOD(_StpPriorLuma, s_linear_clamp_sampler, p, 0); }
419StpU4 StpPatPriMot4H(StpF2 p) { return GATHER_RED_TEXTURE2D_X(_StpPriorDepthMotion, s_point_clamp_sampler, p); }
420
421void StpPatStMotH(StpW2 p, StpU1 v) { _StpDepthMotion[COORD_TEXTURE2D_X(p)] = v; }
422void StpPatStColH(StpW2 p, StpH4 v) { _StpIntermediateColor[COORD_TEXTURE2D_X(p)] = v; }
423void StpPatStLumH(StpW2 p, StpH2 v) { _StpLuma[COORD_TEXTURE2D_X(p)] = v; }
424void StpPatStCnvH(StpW2 p, StpH1 v) { _StpIntermediateConvergence[COORD_TEXTURE2D_X(p >> StpW1(2))] = v; }
425#endif
426
427#if defined(STP_32BIT)
428StpMF1 StpPatPriConF(StpF2 p) { return (StpMF1)SAMPLE_TEXTURE2D_X_LOD(_StpPriorConvergence, s_linear_clamp_sampler, p, 0); }
429
430// These are separate to support inline operation (pass merged instead of loads).
431StpF2 StpPatDatMotF(StpMU2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputMotion, o, 0).xy; }
432StpMF3 StpPatDatColF(StpMU2 o) { return (StpMF3)LOAD_TEXTURE2D_X_LOD(_StpInputColor, o, 0).rgb; }
433StpF1 StpPatDatZF(StpMU2 o) { return LOAD_TEXTURE2D_X_LOD(_StpInputDepth, o, 0).x; }
434// This provides a place to convert Z from depth to linear if not inlined and actually loaded.
435StpF1 StpPatFixZF(StpF1 z) { return 1.0 / (STP_ZBUFFER_PARAMS_Z * z + STP_ZBUFFER_PARAMS_W); }
436StpU1 StpPatDatRF(StpMU2 o) {
437 #if defined(ENABLE_STENCIL_RESPONSIVE)
438 return GetStencilValue(LOAD_TEXTURE2D_X_LOD(_StpInputStencil, o, 0).xy);
439 #endif // defined(ENABLE_STENCIL_RESPONSIVE)
440 return StpU1_(0); }
441StpMF1 StpPatFixRF(StpU1 v) {
442 // Activate the "responsive" feature when we don't have valid history textures.
443 bool hasValidHistory = DecodeHasValidHistory(STP_COMMON_CONSTANT);
444 bool excludeTaa = false;
445 #if defined(ENABLE_STENCIL_RESPONSIVE)
446 excludeTaa = (v & DecodeStencilMask(STP_COMMON_CONSTANT)) != 0;
447 #endif // defined(ENABLE_STENCIL_RESPONSIVE)
448 return (hasValidHistory && !excludeTaa) ? StpMF1_(1.0) : StpMF1_(0.0); }
449
450StpMF1 StpPatDitF(StpMU2 o) { return (StpMF1)StpDitF1(o); }
451StpMF4 StpPatPriFedF(StpF2 p) { return (StpMF4)SAMPLE_TEXTURE2D_X_LOD(_StpPriorFeedback, s_linear_clamp_sampler, p, 0); }
452StpMF4 StpPatPriFedR4F(StpF2 p) { return (StpMF4)GATHER_RED_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
453StpMF4 StpPatPriFedG4F(StpF2 p) { return (StpMF4)GATHER_GREEN_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
454StpMF4 StpPatPriFedB4F(StpF2 p) { return (StpMF4)GATHER_BLUE_TEXTURE2D_X(_StpPriorFeedback, s_point_clamp_sampler, p); }
455StpMF2 StpPatPriLumF(StpF2 p) { return (StpMF2)SAMPLE_TEXTURE2D_X_LOD(_StpPriorLuma, s_linear_clamp_sampler, p, 0); }
456StpU4 StpPatPriMot4F(StpF2 p) { return GATHER_RED_TEXTURE2D_X(_StpPriorDepthMotion, s_point_clamp_sampler, p); }
457
458void StpPatStMotF(StpMU2 p, StpU1 v) { _StpDepthMotion[COORD_TEXTURE2D_X(p)] = v; }
459void StpPatStColF(StpMU2 p, StpMF4 v) { _StpIntermediateColor[COORD_TEXTURE2D_X(p)] = v; }
460void StpPatStLumF(StpMU2 p, StpMF2 v) { _StpLuma[COORD_TEXTURE2D_X(p)] = v; }
461void StpPatStCnvF(StpMU2 p, StpMF1 v) { _StpIntermediateConvergence[COORD_TEXTURE2D_X(p >> StpMU1(2))] = v; }
462#endif
463
464#define THREADING_BLOCK_SIZE STP_GROUP_SIZE
465#include "Packages/com.unity.render-pipelines.core/ShaderLibrary/Threading.hlsl"
466
467[numthreads(STP_GROUP_SIZE, 1, 1)]
468void StpSetup(Threading::Group group)
469{
470 UNITY_XR_ASSIGN_VIEW_INDEX(group.groupID.z);
471
472#if defined(STP_16BIT)
473 StpW1 lane = StpW1_(group.groupIndex);
474 StpW2 groupPos = ComputeGroupPos(StpW2(group.groupID.xy));
475 StpW2 pos = groupPos + StpRemapLaneTo8x16H(lane);
476#else
477 StpMU1 lane = StpMU1_(group.groupIndex);
478 StpMU2 groupPos = ComputeGroupPos(StpMU2(group.groupID.xy));
479 StpMU2 pos = groupPos + StpRemapLaneTo8x16F(lane);
480#endif
481
482#if defined(STP_16BIT)
483 StpPatH(
484 lane,
485 pos,
486#else
487 StpPatF(
488 lane,
489 pos,
490#endif
491
492 asuint(_StpSetupConstants0),
493 asuint(_StpSetupConstants1),
494 asuint(_StpSetupConstants2),
495 asuint(_StpSetupConstants3),
496 asuint(_StpSetupConstants4),
497 asuint(_StpSetupConstants5),
498
499 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 0]),
500 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 1]),
501 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 2]),
502 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 3]),
503 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 4]),
504 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 5]),
505 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 6]),
506 asuint(_StpSetupPerViewConstants[STP_SETUP_PER_VIEW_CONSTANTS_STEREO_OFFSET + 7])
507 );
508}
509