Cycles: fix crash rendering textured objects in OpenCL
[blender-staging.git] / intern / cycles / kernel / svm / svm_image.h
1 /*
2  * Copyright 2011, Blender Foundation.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License
6  * as published by the Free Software Foundation; either version 2
7  * of the License, or (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software Foundation,
16  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
17  */
18
19 CCL_NAMESPACE_BEGIN
20
21 #ifdef __KERNEL_OPENCL__
22
23 /* For OpenCL all images are packed in a single array, and we do manual lookup
24  * and interpolation. */
25
26 __device_inline float4 svm_image_texture_read(KernelGlobals *kg, int offset)
27 {
28         uchar4 r = kernel_tex_fetch(__tex_image_packed, offset);
29         float f = 1.0f/255.0f;
30         return make_float4(r.x*f, r.y*f, r.z*f, r.w*f);
31 }
32
33 __device_inline int svm_image_texture_wrap_periodic(int x, int width)
34 {
35         x %= width;
36         if(x < 0)
37                 x += width;
38         return x;
39 }
40
41 __device_inline int svm_image_texture_wrap_clamp(int x, int width)
42 {
43         return clamp(x, 0, width-1);
44 }
45
46 __device_inline float svm_image_texture_frac(float x, int *ix)
47 {
48         int i = (int)x - ((x < 0.0f)? 1: 0);
49         *ix = i;
50         return x - (float)i;
51 }
52
53 __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint srgb)
54 {
55         /* first slots are used by float textures, which are not supported here */
56         if(id < TEX_NUM_FLOAT_IMAGES)
57                 return make_float4(1.0f, 0.0f, 1.0f, 1.0f);
58
59         id -= TEX_NUM_FLOAT_IMAGES;
60
61         uint4 info = kernel_tex_fetch(__tex_image_packed_info, id);
62         uint width = info.x;
63         uint height = info.y;
64         uint offset = info.z;
65         uint periodic = info.w;
66
67         int ix, iy, nix, niy;
68         float tx = svm_image_texture_frac(x*width, &ix);
69         float ty = svm_image_texture_frac(y*height, &iy);
70
71         if(periodic) {
72                 ix = svm_image_texture_wrap_periodic(ix, width);
73                 iy = svm_image_texture_wrap_periodic(iy, height);
74
75                 nix = svm_image_texture_wrap_periodic(ix+1, width);
76                 niy = svm_image_texture_wrap_periodic(iy+1, height);
77         }
78         else {
79                 ix = svm_image_texture_wrap_clamp(ix, width);
80                 iy = svm_image_texture_wrap_clamp(iy, height);
81
82                 nix = svm_image_texture_wrap_clamp(ix+1, width);
83                 niy = svm_image_texture_wrap_clamp(iy+1, height);
84         }
85
86         float4 r = (1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + iy*width);
87         r += (1.0f - ty)*tx*svm_image_texture_read(kg, offset + nix + iy*width);
88         r += ty*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + niy*width);
89         r += ty*tx*svm_image_texture_read(kg, offset + nix + niy*width);
90
91         if(srgb) {
92                 r.x = color_srgb_to_scene_linear(r.x);
93                 r.y = color_srgb_to_scene_linear(r.y);
94                 r.z = color_srgb_to_scene_linear(r.z);
95         }
96
97         return r;
98 }
99
100 #else
101
102 __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint srgb)
103 {
104         float4 r;
105
106 #ifdef __KERNEL_CPU__
107         r = kernel_tex_image_interp(id, x, y);
108 #else
109         /* not particularly proud of this massive switch, what are the
110          * alternatives?
111          * - use a single big 1D texture, and do our own lookup/filtering
112          * - group by size and use a 3d texture, performance impact
113          * - group into larger texture with some padding for correct lerp
114          *
115          * also note that cuda has 128 textures limit, we use 100 now, since
116          * we still need some for other storage */
117
118         switch(id) {
119                 case 0: r = kernel_tex_image_interp(__tex_image_float_000, x, y); break;
120                 case 1: r = kernel_tex_image_interp(__tex_image_float_001, x, y); break;
121                 case 2: r = kernel_tex_image_interp(__tex_image_float_002, x, y); break;
122                 case 3: r = kernel_tex_image_interp(__tex_image_float_003, x, y); break;
123                 case 4: r = kernel_tex_image_interp(__tex_image_float_004, x, y); break;
124                 case 5: r = kernel_tex_image_interp(__tex_image_005, x, y); break;
125                 case 6: r = kernel_tex_image_interp(__tex_image_006, x, y); break;
126                 case 7: r = kernel_tex_image_interp(__tex_image_007, x, y); break;
127                 case 8: r = kernel_tex_image_interp(__tex_image_008, x, y); break;
128                 case 9: r = kernel_tex_image_interp(__tex_image_009, x, y); break;
129                 case 10: r = kernel_tex_image_interp(__tex_image_010, x, y); break;
130                 case 11: r = kernel_tex_image_interp(__tex_image_011, x, y); break;
131                 case 12: r = kernel_tex_image_interp(__tex_image_012, x, y); break;
132                 case 13: r = kernel_tex_image_interp(__tex_image_013, x, y); break;
133                 case 14: r = kernel_tex_image_interp(__tex_image_014, x, y); break;
134                 case 15: r = kernel_tex_image_interp(__tex_image_015, x, y); break;
135                 case 16: r = kernel_tex_image_interp(__tex_image_016, x, y); break;
136                 case 17: r = kernel_tex_image_interp(__tex_image_017, x, y); break;
137                 case 18: r = kernel_tex_image_interp(__tex_image_018, x, y); break;
138                 case 19: r = kernel_tex_image_interp(__tex_image_019, x, y); break;
139                 case 20: r = kernel_tex_image_interp(__tex_image_020, x, y); break;
140                 case 21: r = kernel_tex_image_interp(__tex_image_021, x, y); break;
141                 case 22: r = kernel_tex_image_interp(__tex_image_022, x, y); break;
142                 case 23: r = kernel_tex_image_interp(__tex_image_023, x, y); break;
143                 case 24: r = kernel_tex_image_interp(__tex_image_024, x, y); break;
144                 case 25: r = kernel_tex_image_interp(__tex_image_025, x, y); break;
145                 case 26: r = kernel_tex_image_interp(__tex_image_026, x, y); break;
146                 case 27: r = kernel_tex_image_interp(__tex_image_027, x, y); break;
147                 case 28: r = kernel_tex_image_interp(__tex_image_028, x, y); break;
148                 case 29: r = kernel_tex_image_interp(__tex_image_029, x, y); break;
149                 case 30: r = kernel_tex_image_interp(__tex_image_030, x, y); break;
150                 case 31: r = kernel_tex_image_interp(__tex_image_031, x, y); break;
151                 case 32: r = kernel_tex_image_interp(__tex_image_032, x, y); break;
152                 case 33: r = kernel_tex_image_interp(__tex_image_033, x, y); break;
153                 case 34: r = kernel_tex_image_interp(__tex_image_034, x, y); break;
154                 case 35: r = kernel_tex_image_interp(__tex_image_035, x, y); break;
155                 case 36: r = kernel_tex_image_interp(__tex_image_036, x, y); break;
156                 case 37: r = kernel_tex_image_interp(__tex_image_037, x, y); break;
157                 case 38: r = kernel_tex_image_interp(__tex_image_038, x, y); break;
158                 case 39: r = kernel_tex_image_interp(__tex_image_039, x, y); break;
159                 case 40: r = kernel_tex_image_interp(__tex_image_040, x, y); break;
160                 case 41: r = kernel_tex_image_interp(__tex_image_041, x, y); break;
161                 case 42: r = kernel_tex_image_interp(__tex_image_042, x, y); break;
162                 case 43: r = kernel_tex_image_interp(__tex_image_043, x, y); break;
163                 case 44: r = kernel_tex_image_interp(__tex_image_044, x, y); break;
164                 case 45: r = kernel_tex_image_interp(__tex_image_045, x, y); break;
165                 case 46: r = kernel_tex_image_interp(__tex_image_046, x, y); break;
166                 case 47: r = kernel_tex_image_interp(__tex_image_047, x, y); break;
167                 case 48: r = kernel_tex_image_interp(__tex_image_048, x, y); break;
168                 case 49: r = kernel_tex_image_interp(__tex_image_049, x, y); break;
169                 case 50: r = kernel_tex_image_interp(__tex_image_050, x, y); break;
170                 case 51: r = kernel_tex_image_interp(__tex_image_051, x, y); break;
171                 case 52: r = kernel_tex_image_interp(__tex_image_052, x, y); break;
172                 case 53: r = kernel_tex_image_interp(__tex_image_053, x, y); break;
173                 case 54: r = kernel_tex_image_interp(__tex_image_054, x, y); break;
174                 case 55: r = kernel_tex_image_interp(__tex_image_055, x, y); break;
175                 case 56: r = kernel_tex_image_interp(__tex_image_056, x, y); break;
176                 case 57: r = kernel_tex_image_interp(__tex_image_057, x, y); break;
177                 case 58: r = kernel_tex_image_interp(__tex_image_058, x, y); break;
178                 case 59: r = kernel_tex_image_interp(__tex_image_059, x, y); break;
179                 case 60: r = kernel_tex_image_interp(__tex_image_060, x, y); break;
180                 case 61: r = kernel_tex_image_interp(__tex_image_061, x, y); break;
181                 case 62: r = kernel_tex_image_interp(__tex_image_062, x, y); break;
182                 case 63: r = kernel_tex_image_interp(__tex_image_063, x, y); break;
183                 case 64: r = kernel_tex_image_interp(__tex_image_064, x, y); break;
184                 case 65: r = kernel_tex_image_interp(__tex_image_065, x, y); break;
185                 case 66: r = kernel_tex_image_interp(__tex_image_066, x, y); break;
186                 case 67: r = kernel_tex_image_interp(__tex_image_067, x, y); break;
187                 case 68: r = kernel_tex_image_interp(__tex_image_068, x, y); break;
188                 case 69: r = kernel_tex_image_interp(__tex_image_069, x, y); break;
189                 case 70: r = kernel_tex_image_interp(__tex_image_070, x, y); break;
190                 case 71: r = kernel_tex_image_interp(__tex_image_071, x, y); break;
191                 case 72: r = kernel_tex_image_interp(__tex_image_072, x, y); break;
192                 case 73: r = kernel_tex_image_interp(__tex_image_073, x, y); break;
193                 case 74: r = kernel_tex_image_interp(__tex_image_074, x, y); break;
194                 case 75: r = kernel_tex_image_interp(__tex_image_075, x, y); break;
195                 case 76: r = kernel_tex_image_interp(__tex_image_076, x, y); break;
196                 case 77: r = kernel_tex_image_interp(__tex_image_077, x, y); break;
197                 case 78: r = kernel_tex_image_interp(__tex_image_078, x, y); break;
198                 case 79: r = kernel_tex_image_interp(__tex_image_079, x, y); break;
199                 case 80: r = kernel_tex_image_interp(__tex_image_080, x, y); break;
200                 case 81: r = kernel_tex_image_interp(__tex_image_081, x, y); break;
201                 case 82: r = kernel_tex_image_interp(__tex_image_082, x, y); break;
202                 case 83: r = kernel_tex_image_interp(__tex_image_083, x, y); break;
203                 case 84: r = kernel_tex_image_interp(__tex_image_084, x, y); break;
204                 case 85: r = kernel_tex_image_interp(__tex_image_085, x, y); break;
205                 case 86: r = kernel_tex_image_interp(__tex_image_086, x, y); break;
206                 case 87: r = kernel_tex_image_interp(__tex_image_087, x, y); break;
207                 case 88: r = kernel_tex_image_interp(__tex_image_088, x, y); break;
208                 case 89: r = kernel_tex_image_interp(__tex_image_089, x, y); break;
209                 case 90: r = kernel_tex_image_interp(__tex_image_090, x, y); break;
210                 case 91: r = kernel_tex_image_interp(__tex_image_091, x, y); break;
211                 case 92: r = kernel_tex_image_interp(__tex_image_092, x, y); break;
212                 case 93: r = kernel_tex_image_interp(__tex_image_093, x, y); break;
213                 case 94: r = kernel_tex_image_interp(__tex_image_094, x, y); break;
214                 case 95: r = kernel_tex_image_interp(__tex_image_095, x, y); break;
215                 case 96: r = kernel_tex_image_interp(__tex_image_096, x, y); break;
216                 case 97: r = kernel_tex_image_interp(__tex_image_097, x, y); break;
217                 case 98: r = kernel_tex_image_interp(__tex_image_098, x, y); break;
218                 case 99: r = kernel_tex_image_interp(__tex_image_099, x, y); break;
219                 default: 
220                         kernel_assert(0);
221                         return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
222         }
223 #endif
224
225         if(srgb) {
226                 r.x = color_srgb_to_scene_linear(r.x);
227                 r.y = color_srgb_to_scene_linear(r.y);
228                 r.z = color_srgb_to_scene_linear(r.z);
229         }
230
231         return r;
232 }
233
234 #endif
235
236 __device void svm_node_tex_image(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
237 {
238         uint id = node.y;
239         uint co_offset, out_offset, alpha_offset, srgb;
240
241         decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &srgb);
242
243         float3 co = stack_load_float3(stack, co_offset);
244         float4 f = svm_image_texture(kg, id, co.x, co.y, srgb);
245
246         if(stack_valid(out_offset))
247                 stack_store_float3(stack, out_offset, make_float3(f.x, f.y, f.z));
248         if(stack_valid(alpha_offset))
249                 stack_store_float(stack, alpha_offset, f.w);
250 }
251
252 __device void svm_node_tex_image_box(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
253 {
254         /* get object space normal */
255         float3 N = sd->N;
256
257         N = sd->N;
258         if(sd->object != ~0)
259                 object_inverse_normal_transform(kg, sd, &N);
260
261         /* project from direction vector to barycentric coordinates in triangles */
262         N.x = fabsf(N.x);
263         N.y = fabsf(N.y);
264         N.z = fabsf(N.z);
265
266         N /= (N.x + N.y + N.z);
267
268         /* basic idea is to think of this as a triangle, each corner representing
269          * one of the 3 faces of the cube. in the corners we have single textures,
270          * in between we blend between two textures, and in the middle we a blend
271          * between three textures.
272          *
273          * the Nxyz values are the barycentric coordinates in an equilateral
274          * triangle, which in case of blending, in the middle has a smaller
275          * equilateral triangle where 3 textures blend. this divides things into
276          * 7 zones, with an if() test for each zone */
277
278         float3 weight = make_float3(0.0f, 0.0f, 0.0f);
279         float blend = __int_as_float(node.w);
280         float limit = 0.5f*(1.0f + blend);
281
282         /* first test for corners with single texture */
283         if(N.x > limit*(N.x + N.y) && N.x > limit*(N.x + N.z)) {
284                 weight.x = 1.0f;
285         }
286         else if(N.y > limit*(N.x + N.y) && N.y > limit*(N.y + N.z)) {
287                 weight.y = 1.0f;
288         }
289         else if(N.z > limit*(N.x + N.z) && N.z > limit*(N.y + N.z)) {
290                 weight.z = 1.0f;
291         }
292         else if(blend > 0.0f) {
293                 /* in case of blending, test for mixes between two textures */
294                 if(N.z < (1.0f - limit)*(N.y + N.x)) {
295                         weight.x = N.x/(N.x + N.y);
296                         weight.x = clamp((weight.x - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
297                         weight.y = 1.0f - weight.x;
298                 }
299                 else if(N.x < (1.0f - limit)*(N.y + N.z)) {
300                         weight.y = N.y/(N.y + N.z);
301                         weight.y = clamp((weight.y - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
302                         weight.z = 1.0f - weight.y;
303                 }
304                 else if(N.y < (1.0f - limit)*(N.x + N.z)) {
305                         weight.x = N.x/(N.x + N.z);
306                         weight.x = clamp((weight.x - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
307                         weight.z = 1.0f - weight.x;
308                 }
309                 else {
310                         /* last case, we have a mix between three */
311                         weight.x = ((2.0f - limit)*N.x + (limit - 1.0f))/(2.0f*limit - 1.0f);
312                         weight.y = ((2.0f - limit)*N.y + (limit - 1.0f))/(2.0f*limit - 1.0f);
313                         weight.z = ((2.0f - limit)*N.z + (limit - 1.0f))/(2.0f*limit - 1.0f);
314                 }
315         }
316
317         /* now fetch textures */
318         uint co_offset, out_offset, alpha_offset, srgb;
319         decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &srgb);
320
321         float3 co = stack_load_float3(stack, co_offset);
322         uint id = node.y;
323
324         float4 f = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
325
326         if(weight.x > 0.0f)
327                 f += weight.x*svm_image_texture(kg, id, co.y, co.z, srgb);
328         if(weight.y > 0.0f)
329                 f += weight.y*svm_image_texture(kg, id, co.x, co.z, srgb);
330         if(weight.z > 0.0f)
331                 f += weight.z*svm_image_texture(kg, id, co.y, co.x, srgb);
332
333         if(stack_valid(out_offset))
334                 stack_store_float3(stack, out_offset, make_float3(f.x, f.y, f.z));
335         if(stack_valid(alpha_offset))
336                 stack_store_float(stack, alpha_offset, f.w);
337 }
338
339
340 __device void svm_node_tex_environment(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
341 {
342         uint id = node.y;
343         uint co_offset, out_offset, alpha_offset, srgb;
344         uint projection = node.w;
345
346         decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &srgb);
347
348         float3 co = stack_load_float3(stack, co_offset);
349         float2 uv;
350
351         co = normalize(co);
352         
353         if(projection == 0)
354                 uv = direction_to_equirectangular(co);
355         else
356                 uv = direction_to_mirrorball(co);
357
358         float4 f = svm_image_texture(kg, id, uv.x, uv.y, srgb);
359
360         if(stack_valid(out_offset))
361                 stack_store_float3(stack, out_offset, make_float3(f.x, f.y, f.z));
362         if(stack_valid(alpha_offset))
363                 stack_store_float(stack, alpha_offset, f.w);
364 }
365
366 CCL_NAMESPACE_END
367