1 /**********************************************************
2 * Copyright 2008-2022 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 "translate/translate.h"
32 #include "tgsi/tgsi_ureg.h"
33
34 #include "svga_context.h"
35 #include "svga_state.h"
36 #include "svga_cmd.h"
37 #include "svga_shader.h"
38 #include "svga_tgsi.h"
39
40 #include "svga_hw_reg.h"
41
42
43 /**
44 * If we fail to compile a vertex shader we'll use a dummy/fallback shader
45 * that simply emits a (0,0,0,1) vertex position.
46 */
47 static const struct tgsi_token *
get_dummy_vertex_shader(void)48 get_dummy_vertex_shader(void)
49 {
50 static const float zero[4] = { 0.0, 0.0, 0.0, 1.0 };
51 struct ureg_program *ureg;
52 const struct tgsi_token *tokens;
53 struct ureg_src src;
54 struct ureg_dst dst;
55
56 ureg = ureg_create(PIPE_SHADER_VERTEX);
57 if (!ureg)
58 return NULL;
59
60 dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
61 src = ureg_DECL_immediate(ureg, zero, 4);
62 ureg_MOV(ureg, dst, src);
63 ureg_END(ureg);
64
65 tokens = ureg_get_tokens(ureg, NULL);
66
67 ureg_destroy(ureg);
68
69 return tokens;
70 }
71
72
73 /**
74 * Replace the given shader's instruction with a simple / dummy shader.
75 * We use this when normal shader translation fails.
76 */
77 struct svga_shader_variant *
svga_get_compiled_dummy_vertex_shader(struct svga_context * svga,struct svga_shader * shader,const struct svga_compile_key * key)78 svga_get_compiled_dummy_vertex_shader(struct svga_context *svga,
79 struct svga_shader *shader,
80 const struct svga_compile_key *key)
81 {
82 struct svga_vertex_shader *vs = (struct svga_vertex_shader *)shader;
83 const struct tgsi_token *dummy = get_dummy_vertex_shader();
84 struct svga_shader_variant *variant;
85
86 if (!dummy) {
87 return NULL;
88 }
89
90 FREE((void *) vs->base.tokens);
91 vs->base.tokens = dummy;
92
93 svga_tgsi_scan_shader(&vs->base);
94
95 variant = svga_tgsi_compile_shader(svga, shader, key);
96 return variant;
97 }
98
99
100 /* SVGA_NEW_PRESCALE, SVGA_NEW_RAST, SVGA_NEW_FS
101 */
102 static void
make_vs_key(struct svga_context * svga,struct svga_compile_key * key)103 make_vs_key(struct svga_context *svga, struct svga_compile_key *key)
104 {
105 struct svga_vertex_shader *vs = svga->curr.vs;
106
107 memset(key, 0, sizeof *key);
108
109 if (svga->state.sw.need_swtnl && svga_have_vgpu10(svga)) {
110 /* Set both of these flags, to match compile_passthrough_vs() */
111 key->vs.passthrough = 1;
112 key->vs.undo_viewport = 1;
113 return;
114 }
115
116 if (svga_have_vgpu10(svga)) {
117 key->vs.need_vertex_id_bias = 1;
118 }
119
120 /* SVGA_NEW_PRESCALE */
121 key->vs.need_prescale = svga->state.hw_clear.prescale[0].enabled &&
122 (svga->curr.tes == NULL) &&
123 (svga->curr.gs == NULL);
124
125 /* SVGA_NEW_RAST */
126 key->vs.allow_psiz = svga->curr.rast->templ.point_size_per_vertex;
127
128 /* SVGA_NEW_FS */
129 key->vs.fs_generic_inputs = svga->curr.fs->base.info.generic_inputs_mask;
130
131 svga_remap_generics(key->vs.fs_generic_inputs, key->generic_remap_table);
132
133 /* SVGA_NEW_VELEMENT */
134 key->vs.adjust_attrib_range = svga->curr.velems->adjust_attrib_range;
135 key->vs.adjust_attrib_w_1 = svga->curr.velems->adjust_attrib_w_1;
136 key->vs.attrib_is_pure_int = svga->curr.velems->attrib_is_pure_int;
137 key->vs.adjust_attrib_itof = svga->curr.velems->adjust_attrib_itof;
138 key->vs.adjust_attrib_utof = svga->curr.velems->adjust_attrib_utof;
139 key->vs.attrib_is_bgra = svga->curr.velems->attrib_is_bgra;
140 key->vs.attrib_puint_to_snorm = svga->curr.velems->attrib_puint_to_snorm;
141 key->vs.attrib_puint_to_uscaled = svga->curr.velems->attrib_puint_to_uscaled;
142 key->vs.attrib_puint_to_sscaled = svga->curr.velems->attrib_puint_to_sscaled;
143
144 /* SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER */
145 svga_init_shader_key_common(svga, PIPE_SHADER_VERTEX, &vs->base, key);
146
147 /* SVGA_NEW_RAST */
148 key->clip_plane_enable = svga->curr.rast->templ.clip_plane_enable;
149
150 /* Determine if this shader is the last shader in the vertex
151 * processing stage.
152 */
153 key->last_vertex_stage = !(svga->curr.gs ||
154 svga->curr.tcs || svga->curr.tes);
155 }
156
157
158 /**
159 * svga_reemit_vs_bindings - Reemit the vertex shader bindings
160 */
161 enum pipe_error
svga_reemit_vs_bindings(struct svga_context * svga)162 svga_reemit_vs_bindings(struct svga_context *svga)
163 {
164 enum pipe_error ret;
165 struct svga_winsys_gb_shader *gbshader = NULL;
166 SVGA3dShaderId shaderId = SVGA3D_INVALID_ID;
167
168 assert(svga->rebind.flags.vs);
169 assert(svga_have_gb_objects(svga));
170
171 if (svga->state.hw_draw.vs) {
172 gbshader = svga->state.hw_draw.vs->gb_shader;
173 shaderId = svga->state.hw_draw.vs->id;
174 }
175
176 if (!svga_need_to_rebind_resources(svga)) {
177 ret = svga->swc->resource_rebind(svga->swc, NULL, gbshader,
178 SVGA_RELOC_READ);
179 }
180 else {
181 if (svga_have_vgpu10(svga))
182 ret = SVGA3D_vgpu10_SetShader(svga->swc, SVGA3D_SHADERTYPE_VS,
183 gbshader, shaderId);
184 else
185 ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_VS, gbshader);
186 }
187
188 if (ret != PIPE_OK)
189 return ret;
190
191 svga->rebind.flags.vs = FALSE;
192 return PIPE_OK;
193 }
194
195
196 /**
197 * The current vertex shader is already executed by the 'draw'
198 * module, so we just need to generate a simple vertex shader
199 * to pass through all those VS outputs that will
200 * be consumed by the fragment shader.
201 * Used when we employ the 'draw' module.
202 */
203 static enum pipe_error
compile_passthrough_vs(struct svga_context * svga,struct svga_vertex_shader * vs,struct svga_fragment_shader * fs,struct svga_shader_variant ** out_variant)204 compile_passthrough_vs(struct svga_context *svga,
205 struct svga_vertex_shader *vs,
206 struct svga_fragment_shader *fs,
207 struct svga_shader_variant **out_variant)
208 {
209 struct svga_shader_variant *variant = NULL;
210 unsigned num_inputs;
211 unsigned i;
212 unsigned num_elements;
213 struct svga_vertex_shader new_vs;
214 struct ureg_src src[PIPE_MAX_SHADER_INPUTS];
215 struct ureg_dst dst[PIPE_MAX_SHADER_OUTPUTS];
216 struct ureg_program *ureg;
217 struct svga_compile_key key;
218 enum pipe_error ret;
219
220 assert(svga_have_vgpu10(svga));
221 assert(fs);
222
223 num_inputs = fs->base.tgsi_info.num_inputs;
224
225 ureg = ureg_create(PIPE_SHADER_VERTEX);
226 if (!ureg)
227 return PIPE_ERROR_OUT_OF_MEMORY;
228
229 /* draw will always add position */
230 dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
231 src[0] = ureg_DECL_vs_input(ureg, 0);
232 num_elements = 1;
233
234 /**
235 * swtnl backend redefines the input layout based on the
236 * fragment shader's inputs. So we only need to passthrough
237 * those inputs that will be consumed by the fragment shader.
238 * Note: DX10 requires the number of vertex elements
239 * specified in the input layout to be no less than the
240 * number of inputs to the vertex shader.
241 */
242 for (i = 0; i < num_inputs; i++) {
243 switch (fs->base.tgsi_info.input_semantic_name[i]) {
244 case TGSI_SEMANTIC_COLOR:
245 case TGSI_SEMANTIC_GENERIC:
246 case TGSI_SEMANTIC_FOG:
247 dst[num_elements] = ureg_DECL_output(ureg,
248 fs->base.tgsi_info.input_semantic_name[i],
249 fs->base.tgsi_info.input_semantic_index[i]);
250 src[num_elements] = ureg_DECL_vs_input(ureg, num_elements);
251 num_elements++;
252 break;
253 default:
254 break;
255 }
256 }
257
258 for (i = 0; i < num_elements; i++) {
259 ureg_MOV(ureg, dst[i], src[i]);
260 }
261
262 ureg_END(ureg);
263
264 memset(&new_vs, 0, sizeof(new_vs));
265 new_vs.base.tokens = ureg_get_tokens(ureg, NULL);
266 svga_tgsi_scan_shader(&new_vs.base);
267
268 memset(&key, 0, sizeof(key));
269 key.vs.undo_viewport = 1;
270
271 ret = svga_compile_shader(svga, &new_vs.base, &key, &variant);
272 if (ret != PIPE_OK)
273 return ret;
274
275 ureg_free_tokens(new_vs.base.tokens);
276 ureg_destroy(ureg);
277
278 /* Overwrite the variant key to indicate it's a pass-through VS */
279 memset(&variant->key, 0, sizeof(variant->key));
280 variant->key.vs.passthrough = 1;
281 variant->key.vs.undo_viewport = 1;
282
283 *out_variant = variant;
284
285 return PIPE_OK;
286 }
287
288
289 static enum pipe_error
emit_hw_vs(struct svga_context * svga,uint64_t dirty)290 emit_hw_vs(struct svga_context *svga, uint64_t dirty)
291 {
292 struct svga_shader_variant *variant;
293 struct svga_vertex_shader *vs = svga->curr.vs;
294 struct svga_fragment_shader *fs = svga->curr.fs;
295 enum pipe_error ret = PIPE_OK;
296 struct svga_compile_key key;
297
298 SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_EMITVS);
299
300 /* If there is an active geometry shader, and it has stream output
301 * defined, then we will skip the stream output from the vertex shader
302 */
303 if (!svga_have_gs_streamout(svga)) {
304 /* No GS stream out */
305 if (svga_have_vs_streamout(svga)) {
306 /* Set VS stream out */
307 ret = svga_set_stream_output(svga, vs->base.stream_output);
308 }
309 else {
310 /* turn off stream out */
311 ret = svga_set_stream_output(svga, NULL);
312 }
313 if (ret != PIPE_OK) {
314 goto done;
315 }
316 }
317
318 /* SVGA_NEW_NEED_SWTNL */
319 if (svga->state.sw.need_swtnl && !svga_have_vgpu10(svga)) {
320 /* No vertex shader is needed */
321 variant = NULL;
322 }
323 else {
324 make_vs_key(svga, &key);
325
326 /* See if we already have a VS variant that matches the key */
327 variant = svga_search_shader_key(&vs->base, &key);
328
329 if (!variant) {
330 /* Create VS variant now */
331 if (key.vs.passthrough) {
332 ret = compile_passthrough_vs(svga, vs, fs, &variant);
333 }
334 else {
335 ret = svga_compile_shader(svga, &vs->base, &key, &variant);
336 }
337 if (ret != PIPE_OK)
338 goto done;
339 }
340 }
341
342 if (variant != svga->state.hw_draw.vs) {
343 /* Bind the new variant */
344 if (variant) {
345 ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_VS, variant);
346 if (ret != PIPE_OK)
347 goto done;
348 svga->rebind.flags.vs = FALSE;
349 }
350
351 svga->dirty |= SVGA_NEW_VS_VARIANT;
352 svga->state.hw_draw.vs = variant;
353 }
354
355 done:
356 SVGA_STATS_TIME_POP(svga_sws(svga));
357 return ret;
358 }
359
360 struct svga_tracked_state svga_hw_vs =
361 {
362 "vertex shader (hwtnl)",
363 (SVGA_NEW_VS |
364 SVGA_NEW_FS |
365 SVGA_NEW_TEXTURE_BINDING |
366 SVGA_NEW_SAMPLER |
367 SVGA_NEW_RAST |
368 SVGA_NEW_PRESCALE |
369 SVGA_NEW_VELEMENT |
370 SVGA_NEW_NEED_SWTNL |
371 SVGA_NEW_VS_RAW_BUFFER),
372 emit_hw_vs
373 };
374