1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "util/u_inlines.h"
27 #include "pipe/p_defines.h"
28 #include "util/u_math.h"
29 #include "util/u_memory.h"
30 #include "util/u_bitmask.h"
31 #include "tgsi/tgsi_ureg.h"
32
33 #include "svga_context.h"
34 #include "svga_state.h"
35 #include "svga_cmd.h"
36 #include "svga_tgsi.h"
37
38 #include "svga_hw_reg.h"
39
40
41
compare_fs_keys(const struct svga_fs_compile_key * a,const struct svga_fs_compile_key * b)42 static INLINE int compare_fs_keys( const struct svga_fs_compile_key *a,
43 const struct svga_fs_compile_key *b )
44 {
45 unsigned keysize_a = svga_fs_key_size( a );
46 unsigned keysize_b = svga_fs_key_size( b );
47
48 if (keysize_a != keysize_b) {
49 return (int)(keysize_a - keysize_b);
50 }
51 return memcmp( a, b, keysize_a );
52 }
53
54
search_fs_key(struct svga_fragment_shader * fs,const struct svga_fs_compile_key * key)55 static struct svga_shader_result *search_fs_key( struct svga_fragment_shader *fs,
56 const struct svga_fs_compile_key *key )
57 {
58 struct svga_shader_result *result = fs->base.results;
59
60 assert(key);
61
62 for ( ; result; result = result->next) {
63 if (compare_fs_keys( key, &result->key.fkey ) == 0)
64 return result;
65 }
66
67 return NULL;
68 }
69
70
71 /**
72 * If we fail to compile a fragment shader (because it uses too many
73 * registers, for example) we'll use a dummy/fallback shader that
74 * simply emits a constant color.
75 */
76 static const struct tgsi_token *
get_dummy_fragment_shader(void)77 get_dummy_fragment_shader(void)
78 {
79 static const float red[4] = { 1.0, 0.0, 0.0, 0.0 };
80 struct ureg_program *ureg;
81 const struct tgsi_token *tokens;
82 struct ureg_src src;
83 struct ureg_dst dst;
84 unsigned num_tokens;
85
86 ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
87 if (!ureg)
88 return NULL;
89
90 dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
91 src = ureg_DECL_immediate(ureg, red, 4);
92 ureg_MOV(ureg, dst, src);
93 ureg_END(ureg);
94
95 tokens = ureg_get_tokens(ureg, &num_tokens);
96
97 ureg_destroy(ureg);
98
99 return tokens;
100 }
101
102
compile_fs(struct svga_context * svga,struct svga_fragment_shader * fs,const struct svga_fs_compile_key * key,struct svga_shader_result ** out_result)103 static enum pipe_error compile_fs( struct svga_context *svga,
104 struct svga_fragment_shader *fs,
105 const struct svga_fs_compile_key *key,
106 struct svga_shader_result **out_result )
107 {
108 struct svga_shader_result *result;
109 enum pipe_error ret = PIPE_ERROR;
110
111 result = svga_translate_fragment_program( fs, key );
112 if (result == NULL) {
113 /* some problem during translation, try the dummy shader */
114 const struct tgsi_token *dummy = get_dummy_fragment_shader();
115 if (!dummy) {
116 ret = PIPE_ERROR_OUT_OF_MEMORY;
117 goto fail;
118 }
119 debug_printf("Failed to compile fragment shader, using dummy shader instead.\n");
120 FREE((void *) fs->base.tokens);
121 fs->base.tokens = dummy;
122 result = svga_translate_fragment_program(fs, key);
123 if (result == NULL) {
124 ret = PIPE_ERROR;
125 goto fail;
126 }
127 }
128
129 result->id = util_bitmask_add(svga->fs_bm);
130 if(result->id == UTIL_BITMASK_INVALID_INDEX) {
131 ret = PIPE_ERROR_OUT_OF_MEMORY;
132 goto fail;
133 }
134
135 ret = SVGA3D_DefineShader(svga->swc,
136 result->id,
137 SVGA3D_SHADERTYPE_PS,
138 result->tokens,
139 result->nr_tokens * sizeof result->tokens[0]);
140 if (ret != PIPE_OK)
141 goto fail;
142
143 *out_result = result;
144 result->next = fs->base.results;
145 fs->base.results = result;
146 return PIPE_OK;
147
148 fail:
149 if (result) {
150 if (result->id != UTIL_BITMASK_INVALID_INDEX)
151 util_bitmask_clear( svga->fs_bm, result->id );
152 svga_destroy_shader_result( result );
153 }
154 return ret;
155 }
156
157
158 /* SVGA_NEW_TEXTURE_BINDING
159 * SVGA_NEW_RAST
160 * SVGA_NEW_NEED_SWTNL
161 * SVGA_NEW_SAMPLER
162 */
163 static enum pipe_error
make_fs_key(const struct svga_context * svga,struct svga_fragment_shader * fs,struct svga_fs_compile_key * key)164 make_fs_key(const struct svga_context *svga,
165 struct svga_fragment_shader *fs,
166 struct svga_fs_compile_key *key)
167 {
168 int i;
169 int idx = 0;
170
171 memset(key, 0, sizeof *key);
172
173 /* Only need fragment shader fixup for twoside lighting if doing
174 * hwtnl. Otherwise the draw module does the whole job for us.
175 *
176 * SVGA_NEW_SWTNL
177 */
178 if (!svga->state.sw.need_swtnl) {
179 /* SVGA_NEW_RAST
180 */
181 key->light_twoside = svga->curr.rast->templ.light_twoside;
182 key->front_ccw = svga->curr.rast->templ.front_ccw;
183 }
184
185 /* The blend workaround for simulating logicop xor behaviour
186 * requires that the incoming fragment color be white. This change
187 * achieves that by creating a variant of the current fragment
188 * shader that overrides all output colors with 1,1,1,1
189 *
190 * This will work for most shaders, including those containing
191 * TEXKIL and/or depth-write. However, it will break on the
192 * combination of xor-logicop plus alphatest.
193 *
194 * Ultimately, we could implement alphatest in the shader using
195 * texkil prior to overriding the outgoing fragment color.
196 *
197 * SVGA_NEW_BLEND
198 */
199 if (svga->curr.blend->need_white_fragments) {
200 key->white_fragments = 1;
201 }
202
203 /* XXX: want to limit this to the textures that the shader actually
204 * refers to.
205 *
206 * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
207 */
208 for (i = 0; i < svga->curr.num_sampler_views; i++) {
209 if (svga->curr.sampler_views[i]) {
210 assert(svga->curr.sampler[i]);
211 assert(svga->curr.sampler_views[i]->texture);
212 key->tex[i].texture_target = svga->curr.sampler_views[i]->texture->target;
213 if (!svga->curr.sampler[i]->normalized_coords) {
214 key->tex[i].width_height_idx = idx++;
215 key->tex[i].unnormalized = TRUE;
216 ++key->num_unnormalized_coords;
217 }
218
219 key->tex[i].swizzle_r = svga->curr.sampler_views[i]->swizzle_r;
220 key->tex[i].swizzle_g = svga->curr.sampler_views[i]->swizzle_g;
221 key->tex[i].swizzle_b = svga->curr.sampler_views[i]->swizzle_b;
222 key->tex[i].swizzle_a = svga->curr.sampler_views[i]->swizzle_a;
223 }
224 }
225 key->num_textures = svga->curr.num_sampler_views;
226
227 idx = 0;
228 for (i = 0; i < svga->curr.num_samplers; ++i) {
229 if (svga->curr.sampler[i]) {
230 key->tex[i].compare_mode = svga->curr.sampler[i]->compare_mode;
231 key->tex[i].compare_func = svga->curr.sampler[i]->compare_func;
232 }
233 }
234
235 /* sprite coord gen state */
236 for (i = 0; i < svga->curr.num_samplers; ++i) {
237 key->tex[i].sprite_texgen =
238 svga->curr.rast->templ.sprite_coord_enable & (1 << i);
239 }
240
241 key->sprite_origin_lower_left = (svga->curr.rast->templ.sprite_coord_mode
242 == PIPE_SPRITE_COORD_LOWER_LEFT);
243
244 return PIPE_OK;
245 }
246
247
248
249 static enum pipe_error
emit_hw_fs(struct svga_context * svga,unsigned dirty)250 emit_hw_fs(struct svga_context *svga, unsigned dirty)
251 {
252 struct svga_shader_result *result = NULL;
253 unsigned id = SVGA3D_INVALID_ID;
254 enum pipe_error ret = PIPE_OK;
255
256 struct svga_fragment_shader *fs = svga->curr.fs;
257 struct svga_fs_compile_key key;
258
259 /* SVGA_NEW_BLEND
260 * SVGA_NEW_TEXTURE_BINDING
261 * SVGA_NEW_RAST
262 * SVGA_NEW_NEED_SWTNL
263 * SVGA_NEW_SAMPLER
264 */
265 ret = make_fs_key( svga, fs, &key );
266 if (ret != PIPE_OK)
267 return ret;
268
269 result = search_fs_key( fs, &key );
270 if (!result) {
271 ret = compile_fs( svga, fs, &key, &result );
272 if (ret != PIPE_OK)
273 return ret;
274 }
275
276 assert (result);
277 id = result->id;
278
279 assert(id != SVGA3D_INVALID_ID);
280
281 if (result != svga->state.hw_draw.fs) {
282 ret = SVGA3D_SetShader(svga->swc,
283 SVGA3D_SHADERTYPE_PS,
284 id );
285 if (ret != PIPE_OK)
286 return ret;
287
288 svga->dirty |= SVGA_NEW_FS_RESULT;
289 svga->state.hw_draw.fs = result;
290 }
291
292 return PIPE_OK;
293 }
294
295 struct svga_tracked_state svga_hw_fs =
296 {
297 "fragment shader (hwtnl)",
298 (SVGA_NEW_FS |
299 SVGA_NEW_TEXTURE_BINDING |
300 SVGA_NEW_NEED_SWTNL |
301 SVGA_NEW_RAST |
302 SVGA_NEW_SAMPLER |
303 SVGA_NEW_BLEND),
304 emit_hw_fs
305 };
306
307
308
309