Cycles / CUDA: Increase maximum image textures on GPU.
[blender-staging.git] / intern / cycles / kernel / svm / svm_image.h
1 /*
2  * Copyright 2011-2013 Blender Foundation
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  * http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License
15  */
16
17 CCL_NAMESPACE_BEGIN
18
19 #ifdef __KERNEL_OPENCL__
20
21 /* For OpenCL all images are packed in a single array, and we do manual lookup
22  * and interpolation. */
23
24 ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, int offset)
25 {
26         uchar4 r = kernel_tex_fetch(__tex_image_packed, offset);
27         float f = 1.0f/255.0f;
28         return make_float4(r.x*f, r.y*f, r.z*f, r.w*f);
29 }
30
31 ccl_device_inline int svm_image_texture_wrap_periodic(int x, int width)
32 {
33         x %= width;
34         if(x < 0)
35                 x += width;
36         return x;
37 }
38
39 ccl_device_inline int svm_image_texture_wrap_clamp(int x, int width)
40 {
41         return clamp(x, 0, width-1);
42 }
43
44 ccl_device_inline float svm_image_texture_frac(float x, int *ix)
45 {
46         int i = float_to_int(x) - ((x < 0.0f)? 1: 0);
47         *ix = i;
48         return x - (float)i;
49 }
50
51 ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint srgb, uint use_alpha)
52 {
53         /* first slots are used by float textures, which are not supported here */
54         if(id < TEX_NUM_FLOAT_IMAGES)
55                 return make_float4(1.0f, 0.0f, 1.0f, 1.0f);
56
57         id -= TEX_NUM_FLOAT_IMAGES;
58
59         uint4 info = kernel_tex_fetch(__tex_image_packed_info, id);
60         uint width = info.x;
61         uint height = info.y;
62         uint offset = info.z;
63         uint periodic = (info.w & 0x1);
64         uint interpolation = info.w >> 1;
65
66         float4 r;
67         int ix, iy, nix, niy;
68         if (interpolation == INTERPOLATION_CLOSEST) {
69                 svm_image_texture_frac(x*width, &ix);
70                 svm_image_texture_frac(y*height, &iy);
71
72                 if(periodic) {
73                         ix = svm_image_texture_wrap_periodic(ix, width);
74                         iy = svm_image_texture_wrap_periodic(iy, height);
75                 }
76                 else {
77                         ix = svm_image_texture_wrap_clamp(ix, width);
78                         iy = svm_image_texture_wrap_clamp(iy, height);
79
80                 }
81                 r = svm_image_texture_read(kg, offset + ix + iy*width);
82         }
83         else { /* We default to linear interpolation if it is not closest */
84                 float tx = svm_image_texture_frac(x*width, &ix);
85                 float ty = svm_image_texture_frac(y*height, &iy);
86
87                 if(periodic) {
88                         ix = svm_image_texture_wrap_periodic(ix, width);
89                         iy = svm_image_texture_wrap_periodic(iy, height);
90
91                         nix = svm_image_texture_wrap_periodic(ix+1, width);
92                         niy = svm_image_texture_wrap_periodic(iy+1, height);
93                 }
94                 else {
95                         ix = svm_image_texture_wrap_clamp(ix, width);
96                         iy = svm_image_texture_wrap_clamp(iy, height);
97
98                         nix = svm_image_texture_wrap_clamp(ix+1, width);
99                         niy = svm_image_texture_wrap_clamp(iy+1, height);
100                 }
101
102
103                 r = (1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + iy*width);
104                 r += (1.0f - ty)*tx*svm_image_texture_read(kg, offset + nix + iy*width);
105                 r += ty*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + niy*width);
106                 r += ty*tx*svm_image_texture_read(kg, offset + nix + niy*width);
107         }
108
109         if(use_alpha && r.w != 1.0f && r.w != 0.0f) {
110                 float invw = 1.0f/r.w;
111                 r.x *= invw;
112                 r.y *= invw;
113                 r.z *= invw;
114
115                 if(id >= TEX_NUM_FLOAT_IMAGES) {
116                         r.x = min(r.x, 1.0f);
117                         r.y = min(r.y, 1.0f);
118                         r.z = min(r.z, 1.0f);
119                 }
120         }
121
122         if(srgb) {
123                 r.x = color_srgb_to_scene_linear(r.x);
124                 r.y = color_srgb_to_scene_linear(r.y);
125                 r.z = color_srgb_to_scene_linear(r.z);
126         }
127
128         return r;
129 }
130
131 #else
132
133 ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint srgb, uint use_alpha)
134 {
135 #ifdef __KERNEL_CPU__
136 #ifdef __KERNEL_SSE2__
137         __m128 r_m128;
138         float4 &r = (float4 &)r_m128;
139         r = kernel_tex_image_interp(id, x, y);
140 #else
141         float4 r = kernel_tex_image_interp(id, x, y);
142 #endif
143 #else
144         float4 r;
145
146         /* not particularly proud of this massive switch, what are the
147          * alternatives?
148          * - use a single big 1D texture, and do our own lookup/filtering
149          * - group by size and use a 3d texture, performance impact
150          * - group into larger texture with some padding for correct lerp
151          *
152          * also note that cuda has a textures limit (128 for Fermi, 256 for Kepler),
153          * and we cannot use all since we still need some for other storage */
154
155         switch(id) {
156                 case 0: r = kernel_tex_image_interp(__tex_image_float_000, x, y); break;
157                 case 1: r = kernel_tex_image_interp(__tex_image_float_001, x, y); break;
158                 case 2: r = kernel_tex_image_interp(__tex_image_float_002, x, y); break;
159                 case 3: r = kernel_tex_image_interp(__tex_image_float_003, x, y); break;
160                 case 4: r = kernel_tex_image_interp(__tex_image_float_004, x, y); break;
161                 case 5: r = kernel_tex_image_interp(__tex_image_005, x, y); break;
162                 case 6: r = kernel_tex_image_interp(__tex_image_006, x, y); break;
163                 case 7: r = kernel_tex_image_interp(__tex_image_007, x, y); break;
164                 case 8: r = kernel_tex_image_interp(__tex_image_008, x, y); break;
165                 case 9: r = kernel_tex_image_interp(__tex_image_009, x, y); break;
166                 case 10: r = kernel_tex_image_interp(__tex_image_010, x, y); break;
167                 case 11: r = kernel_tex_image_interp(__tex_image_011, x, y); break;
168                 case 12: r = kernel_tex_image_interp(__tex_image_012, x, y); break;
169                 case 13: r = kernel_tex_image_interp(__tex_image_013, x, y); break;
170                 case 14: r = kernel_tex_image_interp(__tex_image_014, x, y); break;
171                 case 15: r = kernel_tex_image_interp(__tex_image_015, x, y); break;
172                 case 16: r = kernel_tex_image_interp(__tex_image_016, x, y); break;
173                 case 17: r = kernel_tex_image_interp(__tex_image_017, x, y); break;
174                 case 18: r = kernel_tex_image_interp(__tex_image_018, x, y); break;
175                 case 19: r = kernel_tex_image_interp(__tex_image_019, x, y); break;
176                 case 20: r = kernel_tex_image_interp(__tex_image_020, x, y); break;
177                 case 21: r = kernel_tex_image_interp(__tex_image_021, x, y); break;
178                 case 22: r = kernel_tex_image_interp(__tex_image_022, x, y); break;
179                 case 23: r = kernel_tex_image_interp(__tex_image_023, x, y); break;
180                 case 24: r = kernel_tex_image_interp(__tex_image_024, x, y); break;
181                 case 25: r = kernel_tex_image_interp(__tex_image_025, x, y); break;
182                 case 26: r = kernel_tex_image_interp(__tex_image_026, x, y); break;
183                 case 27: r = kernel_tex_image_interp(__tex_image_027, x, y); break;
184                 case 28: r = kernel_tex_image_interp(__tex_image_028, x, y); break;
185                 case 29: r = kernel_tex_image_interp(__tex_image_029, x, y); break;
186                 case 30: r = kernel_tex_image_interp(__tex_image_030, x, y); break;
187                 case 31: r = kernel_tex_image_interp(__tex_image_031, x, y); break;
188                 case 32: r = kernel_tex_image_interp(__tex_image_032, x, y); break;
189                 case 33: r = kernel_tex_image_interp(__tex_image_033, x, y); break;
190                 case 34: r = kernel_tex_image_interp(__tex_image_034, x, y); break;
191                 case 35: r = kernel_tex_image_interp(__tex_image_035, x, y); break;
192                 case 36: r = kernel_tex_image_interp(__tex_image_036, x, y); break;
193                 case 37: r = kernel_tex_image_interp(__tex_image_037, x, y); break;
194                 case 38: r = kernel_tex_image_interp(__tex_image_038, x, y); break;
195                 case 39: r = kernel_tex_image_interp(__tex_image_039, x, y); break;
196                 case 40: r = kernel_tex_image_interp(__tex_image_040, x, y); break;
197                 case 41: r = kernel_tex_image_interp(__tex_image_041, x, y); break;
198                 case 42: r = kernel_tex_image_interp(__tex_image_042, x, y); break;
199                 case 43: r = kernel_tex_image_interp(__tex_image_043, x, y); break;
200                 case 44: r = kernel_tex_image_interp(__tex_image_044, x, y); break;
201                 case 45: r = kernel_tex_image_interp(__tex_image_045, x, y); break;
202                 case 46: r = kernel_tex_image_interp(__tex_image_046, x, y); break;
203                 case 47: r = kernel_tex_image_interp(__tex_image_047, x, y); break;
204                 case 48: r = kernel_tex_image_interp(__tex_image_048, x, y); break;
205                 case 49: r = kernel_tex_image_interp(__tex_image_049, x, y); break;
206                 case 50: r = kernel_tex_image_interp(__tex_image_050, x, y); break;
207                 case 51: r = kernel_tex_image_interp(__tex_image_051, x, y); break;
208                 case 52: r = kernel_tex_image_interp(__tex_image_052, x, y); break;
209                 case 53: r = kernel_tex_image_interp(__tex_image_053, x, y); break;
210                 case 54: r = kernel_tex_image_interp(__tex_image_054, x, y); break;
211                 case 55: r = kernel_tex_image_interp(__tex_image_055, x, y); break;
212                 case 56: r = kernel_tex_image_interp(__tex_image_056, x, y); break;
213                 case 57: r = kernel_tex_image_interp(__tex_image_057, x, y); break;
214                 case 58: r = kernel_tex_image_interp(__tex_image_058, x, y); break;
215                 case 59: r = kernel_tex_image_interp(__tex_image_059, x, y); break;
216                 case 60: r = kernel_tex_image_interp(__tex_image_060, x, y); break;
217                 case 61: r = kernel_tex_image_interp(__tex_image_061, x, y); break;
218                 case 62: r = kernel_tex_image_interp(__tex_image_062, x, y); break;
219                 case 63: r = kernel_tex_image_interp(__tex_image_063, x, y); break;
220                 case 64: r = kernel_tex_image_interp(__tex_image_064, x, y); break;
221                 case 65: r = kernel_tex_image_interp(__tex_image_065, x, y); break;
222                 case 66: r = kernel_tex_image_interp(__tex_image_066, x, y); break;
223                 case 67: r = kernel_tex_image_interp(__tex_image_067, x, y); break;
224                 case 68: r = kernel_tex_image_interp(__tex_image_068, x, y); break;
225                 case 69: r = kernel_tex_image_interp(__tex_image_069, x, y); break;
226                 case 70: r = kernel_tex_image_interp(__tex_image_070, x, y); break;
227                 case 71: r = kernel_tex_image_interp(__tex_image_071, x, y); break;
228                 case 72: r = kernel_tex_image_interp(__tex_image_072, x, y); break;
229                 case 73: r = kernel_tex_image_interp(__tex_image_073, x, y); break;
230                 case 74: r = kernel_tex_image_interp(__tex_image_074, x, y); break;
231                 case 75: r = kernel_tex_image_interp(__tex_image_075, x, y); break;
232                 case 76: r = kernel_tex_image_interp(__tex_image_076, x, y); break;
233                 case 77: r = kernel_tex_image_interp(__tex_image_077, x, y); break;
234                 case 78: r = kernel_tex_image_interp(__tex_image_078, x, y); break;
235                 case 79: r = kernel_tex_image_interp(__tex_image_079, x, y); break;
236                 case 80: r = kernel_tex_image_interp(__tex_image_080, x, y); break;
237                 case 81: r = kernel_tex_image_interp(__tex_image_081, x, y); break;
238                 case 82: r = kernel_tex_image_interp(__tex_image_082, x, y); break;
239                 case 83: r = kernel_tex_image_interp(__tex_image_083, x, y); break;
240                 case 84: r = kernel_tex_image_interp(__tex_image_084, x, y); break;
241                 case 85: r = kernel_tex_image_interp(__tex_image_085, x, y); break;
242                 case 86: r = kernel_tex_image_interp(__tex_image_086, x, y); break;
243                 case 87: r = kernel_tex_image_interp(__tex_image_087, x, y); break;
244                 case 88: r = kernel_tex_image_interp(__tex_image_088, x, y); break;
245                 case 89: r = kernel_tex_image_interp(__tex_image_089, x, y); break;
246                 case 90: r = kernel_tex_image_interp(__tex_image_090, x, y); break;
247                 case 91: r = kernel_tex_image_interp(__tex_image_091, x, y); break;
248                 case 92: r = kernel_tex_image_interp(__tex_image_092, x, y); break;
249                 case 93: r = kernel_tex_image_interp(__tex_image_093, x, y); break;
250                 case 94: r = kernel_tex_image_interp(__tex_image_094, x, y); break;
251                 case 95: r = kernel_tex_image_interp(__tex_image_095, x, y); break;
252                 case 96: r = kernel_tex_image_interp(__tex_image_096, x, y); break;
253                 case 97: r = kernel_tex_image_interp(__tex_image_097, x, y); break;
254                 case 98: r = kernel_tex_image_interp(__tex_image_098, x, y); break;
255                 case 99: r = kernel_tex_image_interp(__tex_image_099, x, y); break;
256
257 #if __CUDA_ARCH__ >= 300
258                 case 100: r = kernel_tex_image_interp(__tex_image_100, x, y); break;
259                 case 101: r = kernel_tex_image_interp(__tex_image_101, x, y); break;
260                 case 102: r = kernel_tex_image_interp(__tex_image_102, x, y); break;
261                 case 103: r = kernel_tex_image_interp(__tex_image_103, x, y); break;
262                 case 104: r = kernel_tex_image_interp(__tex_image_104, x, y); break;
263                 case 105: r = kernel_tex_image_interp(__tex_image_105, x, y); break;
264                 case 106: r = kernel_tex_image_interp(__tex_image_106, x, y); break;
265                 case 107: r = kernel_tex_image_interp(__tex_image_107, x, y); break;
266                 case 108: r = kernel_tex_image_interp(__tex_image_108, x, y); break;
267                 case 109: r = kernel_tex_image_interp(__tex_image_109, x, y); break;
268                 case 110: r = kernel_tex_image_interp(__tex_image_110, x, y); break;
269                 case 111: r = kernel_tex_image_interp(__tex_image_111, x, y); break;
270                 case 112: r = kernel_tex_image_interp(__tex_image_112, x, y); break;
271                 case 113: r = kernel_tex_image_interp(__tex_image_113, x, y); break;
272                 case 114: r = kernel_tex_image_interp(__tex_image_114, x, y); break;
273                 case 115: r = kernel_tex_image_interp(__tex_image_115, x, y); break;
274                 case 116: r = kernel_tex_image_interp(__tex_image_116, x, y); break;
275                 case 117: r = kernel_tex_image_interp(__tex_image_117, x, y); break;
276                 case 118: r = kernel_tex_image_interp(__tex_image_118, x, y); break;
277                 case 119: r = kernel_tex_image_interp(__tex_image_119, x, y); break;
278                 case 120: r = kernel_tex_image_interp(__tex_image_120, x, y); break;
279                 case 121: r = kernel_tex_image_interp(__tex_image_121, x, y); break;
280                 case 122: r = kernel_tex_image_interp(__tex_image_122, x, y); break;
281                 case 123: r = kernel_tex_image_interp(__tex_image_123, x, y); break;
282                 case 124: r = kernel_tex_image_interp(__tex_image_124, x, y); break;
283                 case 125: r = kernel_tex_image_interp(__tex_image_125, x, y); break;
284                 case 126: r = kernel_tex_image_interp(__tex_image_126, x, y); break;
285                 case 127: r = kernel_tex_image_interp(__tex_image_127, x, y); break;
286                 case 128: r = kernel_tex_image_interp(__tex_image_128, x, y); break;
287                 case 129: r = kernel_tex_image_interp(__tex_image_129, x, y); break;
288                 case 130: r = kernel_tex_image_interp(__tex_image_130, x, y); break;
289                 case 131: r = kernel_tex_image_interp(__tex_image_131, x, y); break;
290                 case 132: r = kernel_tex_image_interp(__tex_image_132, x, y); break;
291                 case 133: r = kernel_tex_image_interp(__tex_image_133, x, y); break;
292                 case 134: r = kernel_tex_image_interp(__tex_image_134, x, y); break;
293                 case 135: r = kernel_tex_image_interp(__tex_image_135, x, y); break;
294                 case 136: r = kernel_tex_image_interp(__tex_image_136, x, y); break;
295                 case 137: r = kernel_tex_image_interp(__tex_image_137, x, y); break;
296                 case 138: r = kernel_tex_image_interp(__tex_image_138, x, y); break;
297                 case 139: r = kernel_tex_image_interp(__tex_image_139, x, y); break;
298                 case 140: r = kernel_tex_image_interp(__tex_image_140, x, y); break;
299                 case 141: r = kernel_tex_image_interp(__tex_image_141, x, y); break;
300                 case 142: r = kernel_tex_image_interp(__tex_image_142, x, y); break;
301                 case 143: r = kernel_tex_image_interp(__tex_image_143, x, y); break;
302                 case 144: r = kernel_tex_image_interp(__tex_image_144, x, y); break;
303                 case 145: r = kernel_tex_image_interp(__tex_image_145, x, y); break;
304                 case 146: r = kernel_tex_image_interp(__tex_image_146, x, y); break;
305                 case 147: r = kernel_tex_image_interp(__tex_image_147, x, y); break;
306                 case 148: r = kernel_tex_image_interp(__tex_image_148, x, y); break;
307                 case 149: r = kernel_tex_image_interp(__tex_image_149, x, y); break;
308                 case 150: r = kernel_tex_image_interp(__tex_image_150, x, y); break;
309 #endif
310
311                 default:
312                         kernel_assert(0);
313                         return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
314         }
315 #endif
316
317 #ifdef __KERNEL_SSE2__
318         float alpha = r.w;
319
320         if(use_alpha && alpha != 1.0f && alpha != 0.0f) {
321                 r_m128 = _mm_div_ps(r_m128, _mm_set1_ps(alpha));
322                 if(id >= TEX_NUM_FLOAT_IMAGES)
323                         r_m128 = _mm_min_ps(r_m128, _mm_set1_ps(1.0f));
324                 r.w = alpha;
325         }
326
327         if(srgb) {
328                 r_m128 = color_srgb_to_scene_linear(r_m128);
329                 r.w = alpha;
330         }
331 #else
332         if(use_alpha && r.w != 1.0f && r.w != 0.0f) {
333                 float invw = 1.0f/r.w;
334                 r.x *= invw;
335                 r.y *= invw;
336                 r.z *= invw;
337
338                 if(id >= TEX_NUM_FLOAT_IMAGES) {
339                         r.x = min(r.x, 1.0f);
340                         r.y = min(r.y, 1.0f);
341                         r.z = min(r.z, 1.0f);
342                 }
343         }
344
345         if(srgb) {
346                 r.x = color_srgb_to_scene_linear(r.x);
347                 r.y = color_srgb_to_scene_linear(r.y);
348                 r.z = color_srgb_to_scene_linear(r.z);
349         }
350 #endif
351
352         return r;
353 }
354
355 #endif
356
357 ccl_device void svm_node_tex_image(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
358 {
359         uint id = node.y;
360         uint co_offset, out_offset, alpha_offset, srgb;
361
362         decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &srgb);
363
364         float3 co = stack_load_float3(stack, co_offset);
365         uint use_alpha = stack_valid(alpha_offset);
366         float4 f = svm_image_texture(kg, id, co.x, co.y, srgb, use_alpha);
367
368         if(stack_valid(out_offset))
369                 stack_store_float3(stack, out_offset, make_float3(f.x, f.y, f.z));
370         if(stack_valid(alpha_offset))
371                 stack_store_float(stack, alpha_offset, f.w);
372 }
373
374 ccl_device void svm_node_tex_image_box(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
375 {
376         /* get object space normal */
377         float3 N = sd->N;
378
379         N = sd->N;
380         if(sd->object != OBJECT_NONE)
381                 object_inverse_normal_transform(kg, sd, &N);
382
383         /* project from direction vector to barycentric coordinates in triangles */
384         N.x = fabsf(N.x);
385         N.y = fabsf(N.y);
386         N.z = fabsf(N.z);
387
388         N /= (N.x + N.y + N.z);
389
390         /* basic idea is to think of this as a triangle, each corner representing
391          * one of the 3 faces of the cube. in the corners we have single textures,
392          * in between we blend between two textures, and in the middle we a blend
393          * between three textures.
394          *
395          * the Nxyz values are the barycentric coordinates in an equilateral
396          * triangle, which in case of blending, in the middle has a smaller
397          * equilateral triangle where 3 textures blend. this divides things into
398          * 7 zones, with an if() test for each zone */
399
400         float3 weight = make_float3(0.0f, 0.0f, 0.0f);
401         float blend = __int_as_float(node.w);
402         float limit = 0.5f*(1.0f + blend);
403
404         /* first test for corners with single texture */
405         if(N.x > limit*(N.x + N.y) && N.x > limit*(N.x + N.z)) {
406                 weight.x = 1.0f;
407         }
408         else if(N.y > limit*(N.x + N.y) && N.y > limit*(N.y + N.z)) {
409                 weight.y = 1.0f;
410         }
411         else if(N.z > limit*(N.x + N.z) && N.z > limit*(N.y + N.z)) {
412                 weight.z = 1.0f;
413         }
414         else if(blend > 0.0f) {
415                 /* in case of blending, test for mixes between two textures */
416                 if(N.z < (1.0f - limit)*(N.y + N.x)) {
417                         weight.x = N.x/(N.x + N.y);
418                         weight.x = clamp((weight.x - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
419                         weight.y = 1.0f - weight.x;
420                 }
421                 else if(N.x < (1.0f - limit)*(N.y + N.z)) {
422                         weight.y = N.y/(N.y + N.z);
423                         weight.y = clamp((weight.y - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
424                         weight.z = 1.0f - weight.y;
425                 }
426                 else if(N.y < (1.0f - limit)*(N.x + N.z)) {
427                         weight.x = N.x/(N.x + N.z);
428                         weight.x = clamp((weight.x - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
429                         weight.z = 1.0f - weight.x;
430                 }
431                 else {
432                         /* last case, we have a mix between three */
433                         weight.x = ((2.0f - limit)*N.x + (limit - 1.0f))/(2.0f*limit - 1.0f);
434                         weight.y = ((2.0f - limit)*N.y + (limit - 1.0f))/(2.0f*limit - 1.0f);
435                         weight.z = ((2.0f - limit)*N.z + (limit - 1.0f))/(2.0f*limit - 1.0f);
436                 }
437         }
438
439         /* now fetch textures */
440         uint co_offset, out_offset, alpha_offset, srgb;
441         decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &srgb);
442
443         float3 co = stack_load_float3(stack, co_offset);
444         uint id = node.y;
445
446         float4 f = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
447         uint use_alpha = stack_valid(alpha_offset);
448
449         if(weight.x > 0.0f)
450                 f += weight.x*svm_image_texture(kg, id, co.y, co.z, srgb, use_alpha);
451         if(weight.y > 0.0f)
452                 f += weight.y*svm_image_texture(kg, id, co.x, co.z, srgb, use_alpha);
453         if(weight.z > 0.0f)
454                 f += weight.z*svm_image_texture(kg, id, co.y, co.x, srgb, use_alpha);
455
456         if(stack_valid(out_offset))
457                 stack_store_float3(stack, out_offset, make_float3(f.x, f.y, f.z));
458         if(stack_valid(alpha_offset))
459                 stack_store_float(stack, alpha_offset, f.w);
460 }
461
462
463 ccl_device void svm_node_tex_environment(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
464 {
465         uint id = node.y;
466         uint co_offset, out_offset, alpha_offset, srgb;
467         uint projection = node.w;
468
469         decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &srgb);
470
471         float3 co = stack_load_float3(stack, co_offset);
472         float2 uv;
473
474         co = normalize(co);
475         
476         if(projection == 0)
477                 uv = direction_to_equirectangular(co);
478         else
479                 uv = direction_to_mirrorball(co);
480
481         uint use_alpha = stack_valid(alpha_offset);
482         float4 f = svm_image_texture(kg, id, uv.x, uv.y, srgb, use_alpha);
483
484         if(stack_valid(out_offset))
485                 stack_store_float3(stack, out_offset, make_float3(f.x, f.y, f.z));
486         if(stack_valid(alpha_offset))
487                 stack_store_float(stack, alpha_offset, f.w);
488 }
489
490 CCL_NAMESPACE_END
491