1 /**
2 * Copyright (C) 2010 Jorge Jimenez (jorge@iryoku.com)
3 * Copyright (C) 2010 Belen Masia (bmasia@unizar.es)
4 * Copyright (C) 2010 Jose I. Echevarria (joseignacioechevarria@gmail.com)
5 * Copyright (C) 2010 Fernando Navarro (fernandn@microsoft.com)
6 * Copyright (C) 2010 Diego Gutierrez (diegog@unizar.es)
7 * Copyright (C) 2011 Lauri Kasanen (cand@gmx.com)
8 * All rights reserved.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions are met:
12 *
13 * 1. Redistributions of source code must retain the above copyright notice,
14 * this list of conditions and the following disclaimer.
15 *
16 * 2. Redistributions in binary form must reproduce the following statement:
17 *
18 * "Uses Jimenez's MLAA. Copyright (C) 2010 by Jorge Jimenez, Belen Masia,
19 * Jose I. Echevarria, Fernando Navarro and Diego Gutierrez."
20 *
21 * Only for use in the Mesa project, this point 2 is filled by naming the
22 * technique Jimenez's MLAA in the Mesa config options.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS
25 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL COPYRIGHT HOLDERS OR CONTRIBUTORS
28 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
31 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
32 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
33 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
34 * POSSIBILITY OF SUCH DAMAGE.
35 *
36 * The views and conclusions contained in the software and documentation are
37 * those of the authors and should not be interpreted as representing official
38 * policies, either expressed or implied, of the copyright holders.
39 */
40
41 #include "pipe/p_compiler.h"
42
43 #include "postprocess/postprocess.h"
44 #include "postprocess/pp_mlaa.h"
45 #include "postprocess/pp_filters.h"
46 #include "postprocess/pp_private.h"
47
48 #include "util/u_box.h"
49 #include "util/u_sampler.h"
50 #include "util/u_inlines.h"
51 #include "util/u_memory.h"
52 #include "util/u_string.h"
53 #include "pipe/p_screen.h"
54
55 #define IMM_SPACE 80
56
57 static float constants[] = { 1, 1, 0, 0 };
58 static unsigned int dimensions[2] = { 0, 0 };
59
60 /** Run function of the MLAA filter. */
61 static void
pp_jimenezmlaa_run(struct pp_queue_t * ppq,struct pipe_resource * in,struct pipe_resource * out,unsigned int n,bool iscolor)62 pp_jimenezmlaa_run(struct pp_queue_t *ppq, struct pipe_resource *in,
63 struct pipe_resource *out, unsigned int n, bool iscolor)
64 {
65
66 struct pp_program *p = ppq->p;
67
68 struct pipe_depth_stencil_alpha_state mstencil;
69 struct pipe_sampler_view v_tmp, *arr[3];
70
71 unsigned int w = 0;
72 unsigned int h = 0;
73
74 const struct pipe_stencil_ref ref = { {1} };
75
76 /* Insufficient initialization checks. */
77 assert(p);
78 assert(ppq);
79 assert(ppq->areamaptex);
80 assert(ppq->inner_tmp);
81 assert(ppq->shaders[n]);
82
83 w = p->framebuffer.width;
84 h = p->framebuffer.height;
85
86 memset(&mstencil, 0, sizeof(mstencil));
87
88 cso_set_stencil_ref(p->cso, &ref);
89
90 /* Init the pixel size constant */
91 if (dimensions[0] != p->framebuffer.width ||
92 dimensions[1] != p->framebuffer.height) {
93 constants[0] = 1.0f / p->framebuffer.width;
94 constants[1] = 1.0f / p->framebuffer.height;
95
96 dimensions[0] = p->framebuffer.width;
97 dimensions[1] = p->framebuffer.height;
98 }
99
100 cso_set_constant_user_buffer(p->cso, PIPE_SHADER_VERTEX,
101 0, constants, sizeof(constants));
102 cso_set_constant_user_buffer(p->cso, PIPE_SHADER_FRAGMENT,
103 0, constants, sizeof(constants));
104
105 mstencil.stencil[0].enabled = 1;
106 mstencil.stencil[0].valuemask = mstencil.stencil[0].writemask = ~0;
107 mstencil.stencil[0].func = PIPE_FUNC_ALWAYS;
108 mstencil.stencil[0].fail_op = PIPE_STENCIL_OP_KEEP;
109 mstencil.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
110 mstencil.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
111
112 p->framebuffer.zsbuf = ppq->stencils;
113
114 /* First pass: depth edge detection */
115 if (iscolor)
116 pp_filter_setup_in(p, in);
117 else
118 pp_filter_setup_in(p, ppq->depth);
119
120 pp_filter_setup_out(p, ppq->inner_tmp[0]);
121
122 pp_filter_set_fb(p);
123 pp_filter_misc_state(p);
124 cso_set_depth_stencil_alpha(p->cso, &mstencil);
125 p->pipe->clear(p->pipe, PIPE_CLEAR_STENCIL | PIPE_CLEAR_COLOR0, NULL,
126 &p->clear_color, 0, 0);
127
128 {
129 const struct pipe_sampler_state *samplers[] = {&p->sampler_point};
130 cso_set_samplers(p->cso, PIPE_SHADER_FRAGMENT, 1, samplers);
131 }
132 cso_set_sampler_views(p->cso, PIPE_SHADER_FRAGMENT, 1, &p->view);
133
134 cso_set_vertex_shader_handle(p->cso, ppq->shaders[n][1]); /* offsetvs */
135 cso_set_fragment_shader_handle(p->cso, ppq->shaders[n][2]);
136
137 pp_filter_draw(p);
138 pp_filter_end_pass(p);
139
140
141 /* Second pass: blend weights */
142 /* Sampler order: areamap, edgesmap, edgesmapL (reversed, thx compiler) */
143 mstencil.stencil[0].func = PIPE_FUNC_EQUAL;
144 mstencil.stencil[0].zpass_op = PIPE_STENCIL_OP_KEEP;
145 cso_set_depth_stencil_alpha(p->cso, &mstencil);
146
147 pp_filter_setup_in(p, ppq->areamaptex);
148 pp_filter_setup_out(p, ppq->inner_tmp[1]);
149
150 u_sampler_view_default_template(&v_tmp, ppq->inner_tmp[0],
151 ppq->inner_tmp[0]->format);
152 arr[1] = arr[2] = p->pipe->create_sampler_view(p->pipe,
153 ppq->inner_tmp[0], &v_tmp);
154
155 pp_filter_set_clear_fb(p);
156
157 {
158 const struct pipe_sampler_state *samplers[] =
159 {&p->sampler_point, &p->sampler_point, &p->sampler};
160 cso_set_samplers(p->cso, PIPE_SHADER_FRAGMENT, 3, samplers);
161 }
162
163 arr[0] = p->view;
164 cso_set_sampler_views(p->cso, PIPE_SHADER_FRAGMENT, 3, arr);
165
166 cso_set_vertex_shader_handle(p->cso, ppq->shaders[n][0]); /* passvs */
167 cso_set_fragment_shader_handle(p->cso, ppq->shaders[n][3]);
168
169 pp_filter_draw(p);
170 pp_filter_end_pass(p);
171 pipe_sampler_view_reference(&arr[1], NULL);
172
173
174 /* Third pass: smoothed edges */
175 /* Sampler order: colormap, blendmap (wtf compiler) */
176 pp_filter_setup_in(p, ppq->inner_tmp[1]);
177 pp_filter_setup_out(p, out);
178
179 pp_filter_set_fb(p);
180
181 /* Blit the input to the output */
182 pp_blit(p->pipe, in, 0, 0,
183 w, h, 0, p->framebuffer.cbufs[0],
184 0, 0, w, h);
185
186 u_sampler_view_default_template(&v_tmp, in, in->format);
187 arr[0] = p->pipe->create_sampler_view(p->pipe, in, &v_tmp);
188
189 {
190 const struct pipe_sampler_state *samplers[] =
191 {&p->sampler_point, &p->sampler_point};
192 cso_set_samplers(p->cso, PIPE_SHADER_FRAGMENT, 2, samplers);
193 }
194
195 arr[1] = p->view;
196 cso_set_sampler_views(p->cso, PIPE_SHADER_FRAGMENT, 2, arr);
197
198 cso_set_vertex_shader_handle(p->cso, ppq->shaders[n][1]); /* offsetvs */
199 cso_set_fragment_shader_handle(p->cso, ppq->shaders[n][4]);
200
201 p->blend.rt[0].blend_enable = 1;
202 cso_set_blend(p->cso, &p->blend);
203
204 pp_filter_draw(p);
205 pp_filter_end_pass(p);
206 pipe_sampler_view_reference(&arr[0], NULL);
207
208 p->blend.rt[0].blend_enable = 0;
209 p->framebuffer.zsbuf = NULL;
210 }
211
212 /** The init function of the MLAA filter. */
213 static bool
pp_jimenezmlaa_init_run(struct pp_queue_t * ppq,unsigned int n,unsigned int val,bool iscolor)214 pp_jimenezmlaa_init_run(struct pp_queue_t *ppq, unsigned int n,
215 unsigned int val, bool iscolor)
216 {
217
218 struct pipe_box box;
219 struct pipe_resource res;
220 char *tmp_text = NULL;
221
222 tmp_text = CALLOC(sizeof(blend2fs_1) + sizeof(blend2fs_2) +
223 IMM_SPACE, sizeof(char));
224
225 if (!tmp_text) {
226 pp_debug("Failed to allocate shader space\n");
227 return FALSE;
228 }
229
230 pp_debug("mlaa: using %u max search steps\n", val);
231
232 sprintf(tmp_text, "%s"
233 "IMM FLT32 { %.8f, 0.0000, 0.0000, 0.0000}\n"
234 "%s\n", blend2fs_1, (float) val, blend2fs_2);
235
236 memset(&res, 0, sizeof(res));
237
238 res.target = PIPE_TEXTURE_2D;
239 res.format = PIPE_FORMAT_R8G8_UNORM;
240 res.width0 = res.height0 = 165;
241 res.bind = PIPE_BIND_SAMPLER_VIEW;
242 res.usage = PIPE_USAGE_DEFAULT;
243 res.depth0 = res.array_size = res.nr_samples = res.nr_storage_samples = 1;
244
245 if (!ppq->p->screen->is_format_supported(ppq->p->screen, res.format,
246 res.target, 1, 1, res.bind))
247 pp_debug("Areamap format not supported\n");
248
249 ppq->areamaptex = ppq->p->screen->resource_create(ppq->p->screen, &res);
250
251 if (ppq->areamaptex == NULL) {
252 pp_debug("Failed to allocate area map texture\n");
253 goto fail;
254 }
255
256 u_box_2d(0, 0, 165, 165, &box);
257
258 ppq->p->pipe->texture_subdata(ppq->p->pipe, ppq->areamaptex, 0,
259 PIPE_MAP_WRITE, &box,
260 areamap, 165 * 2, sizeof(areamap));
261
262 ppq->shaders[n][1] = pp_tgsi_to_state(ppq->p->pipe, offsetvs, true,
263 "offsetvs");
264 if (iscolor)
265 ppq->shaders[n][2] = pp_tgsi_to_state(ppq->p->pipe, color1fs,
266 false, "color1fs");
267 else
268 ppq->shaders[n][2] = pp_tgsi_to_state(ppq->p->pipe, depth1fs,
269 false, "depth1fs");
270 ppq->shaders[n][3] = pp_tgsi_to_state(ppq->p->pipe, tmp_text, false,
271 "blend2fs");
272 ppq->shaders[n][4] = pp_tgsi_to_state(ppq->p->pipe, neigh3fs, false,
273 "neigh3fs");
274
275 FREE(tmp_text);
276
277 return TRUE;
278
279 fail:
280
281 FREE(tmp_text);
282
283 /*
284 * Call the common free function for destruction of partially initialized
285 * resources.
286 */
287 pp_jimenezmlaa_free(ppq, n);
288
289 return FALSE;
290 }
291
292 /** Short wrapper to init the depth version. */
293 bool
pp_jimenezmlaa_init(struct pp_queue_t * ppq,unsigned int n,unsigned int val)294 pp_jimenezmlaa_init(struct pp_queue_t *ppq, unsigned int n, unsigned int val)
295 {
296 return pp_jimenezmlaa_init_run(ppq, n, val, false);
297 }
298
299 /** Short wrapper to init the color version. */
300 bool
pp_jimenezmlaa_init_color(struct pp_queue_t * ppq,unsigned int n,unsigned int val)301 pp_jimenezmlaa_init_color(struct pp_queue_t *ppq, unsigned int n,
302 unsigned int val)
303 {
304 return pp_jimenezmlaa_init_run(ppq, n, val, true);
305 }
306
307 /** Short wrapper to run the depth version. */
308 void
pp_jimenezmlaa(struct pp_queue_t * ppq,struct pipe_resource * in,struct pipe_resource * out,unsigned int n)309 pp_jimenezmlaa(struct pp_queue_t *ppq, struct pipe_resource *in,
310 struct pipe_resource *out, unsigned int n)
311 {
312 if (!ppq->depth) {
313 return;
314 }
315 pp_jimenezmlaa_run(ppq, in, out, n, false);
316 }
317
318 /** Short wrapper to run the color version. */
319 void
pp_jimenezmlaa_color(struct pp_queue_t * ppq,struct pipe_resource * in,struct pipe_resource * out,unsigned int n)320 pp_jimenezmlaa_color(struct pp_queue_t *ppq, struct pipe_resource *in,
321 struct pipe_resource *out, unsigned int n)
322 {
323 pp_jimenezmlaa_run(ppq, in, out, n, true);
324 }
325
326
327 /**
328 * Short wrapper to free the mlaa filter resources. Shaders are freed in
329 * the common code in pp_free.
330 */
331 void
pp_jimenezmlaa_free(struct pp_queue_t * ppq,unsigned int n)332 pp_jimenezmlaa_free(struct pp_queue_t *ppq, unsigned int n)
333 {
334 pipe_resource_reference(&ppq->areamaptex, NULL);
335 }
336
337