Debian Codegraph @

src/gallium/drivers/nouveau/nv50/nv50_state.c

raw
1 /*
2 * Copyright 2010 Christoph Bumiller
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 */
22
23 #include "pipe/p_defines.h"
24 #include "util/u_framebuffer.h"
25 #include "util/u_helpers.h"
26 #include "util/u_inlines.h"
27 #include "util/u_transfer.h"
28 #include "util/format_srgb.h"
29
30 #include "compiler/nir/nir.h"
31 #include "nir/tgsi_to_nir.h"
32
33 #include "nv50/nv50_stateobj.h"
34 #include "nv50/nv50_context.h"
35 #include "nv50/nv50_query_hw.h"
36
37 #include "nv50/nv50_3d.xml.h"
38 #include "nv50/g80_texture.xml.h"
39
40 #include "nouveau_gldefs.h"
41
42 /* Caveats:
43 * ! pipe_sampler_state.unnormalized_coords is ignored - rectangle textures
44 * will use non-normalized coordinates, everything else won't
45 * (The relevant bit is in the TIC entry and not the TSC entry.)
46 *
47 * ! pipe_sampler_state.seamless_cube_map is ignored - seamless filtering is
48 * always activated on NVA0 +
49 * (Give me the global bit, otherwise it's not worth the CPU work.)
50 *
51 * ! pipe_sampler_state.border_color is not swizzled according to the texture
52 * swizzle in pipe_sampler_view
53 * (This will be ugly with indirect independent texture/sampler access,
54 * we'd have to emulate the logic in the shader. GL doesn't have that,
55 * D3D doesn't have swizzle, if we knew what we were implementing we'd be
56 * good.)
57 *
58 * ! pipe_rasterizer_state.line_last_pixel is ignored - it is never drawn
59 *
60 * ! pipe_rasterizer_state.flatshade_first also applies to QUADS
61 * (There's a GL query for that, forcing an exception is just ridiculous.)
62 *
63 * ! pipe_rasterizer_state.sprite_coord_enable is masked with 0xff on NVC0
64 * (The hardware only has 8 slots meant for TexCoord and we have to assign
65 * in advance to maintain elegant separate shader objects.)
66 */
67
68 static inline uint32_t
69 nv50_colormask(unsigned mask)
70 {
71 uint32_t ret = 0;
72
73 if (mask & PIPE_MASK_R)
74 ret |= 0x0001;
75 if (mask & PIPE_MASK_G)
76 ret |= 0x0010;
77 if (mask & PIPE_MASK_B)
78 ret |= 0x0100;
79 if (mask & PIPE_MASK_A)
80 ret |= 0x1000;
81
82 return ret;
83 }
84
85 #define NV50_BLEND_FACTOR_CASE(a, b) \
86 case PIPE_BLENDFACTOR_##a: return NV50_BLEND_FACTOR_##b
87
88 static inline uint32_t
89 nv50_blend_fac(unsigned factor)
90 {
91 switch (factor) {
92 NV50_BLEND_FACTOR_CASE(ONE, ONE);
93 NV50_BLEND_FACTOR_CASE(SRC_COLOR, SRC_COLOR);
94 NV50_BLEND_FACTOR_CASE(SRC_ALPHA, SRC_ALPHA);
95 NV50_BLEND_FACTOR_CASE(DST_ALPHA, DST_ALPHA);
96 NV50_BLEND_FACTOR_CASE(DST_COLOR, DST_COLOR);
97 NV50_BLEND_FACTOR_CASE(SRC_ALPHA_SATURATE, SRC_ALPHA_SATURATE);
98 NV50_BLEND_FACTOR_CASE(CONST_COLOR, CONSTANT_COLOR);
99 NV50_BLEND_FACTOR_CASE(CONST_ALPHA, CONSTANT_ALPHA);
100 NV50_BLEND_FACTOR_CASE(SRC1_COLOR, SRC1_COLOR);
101 NV50_BLEND_FACTOR_CASE(SRC1_ALPHA, SRC1_ALPHA);
102 NV50_BLEND_FACTOR_CASE(ZERO, ZERO);
103 NV50_BLEND_FACTOR_CASE(INV_SRC_COLOR, ONE_MINUS_SRC_COLOR);
104 NV50_BLEND_FACTOR_CASE(INV_SRC_ALPHA, ONE_MINUS_SRC_ALPHA);
105 NV50_BLEND_FACTOR_CASE(INV_DST_ALPHA, ONE_MINUS_DST_ALPHA);
106 NV50_BLEND_FACTOR_CASE(INV_DST_COLOR, ONE_MINUS_DST_COLOR);
107 NV50_BLEND_FACTOR_CASE(INV_CONST_COLOR, ONE_MINUS_CONSTANT_COLOR);
108 NV50_BLEND_FACTOR_CASE(INV_CONST_ALPHA, ONE_MINUS_CONSTANT_ALPHA);
109 NV50_BLEND_FACTOR_CASE(INV_SRC1_COLOR, ONE_MINUS_SRC1_COLOR);
110 NV50_BLEND_FACTOR_CASE(INV_SRC1_ALPHA, ONE_MINUS_SRC1_ALPHA);
111 default:
112 return NV50_BLEND_FACTOR_ZERO;
113 }
114 }
115
116 static void *
117 nv50_blend_state_create(struct pipe_context *pipe,
118 const struct pipe_blend_state *cso)
119 {
120 struct nv50_blend_stateobj *so = CALLOC_STRUCT(nv50_blend_stateobj);
121 int i;
122 bool emit_common_func = cso->rt[0].blend_enable;
123 uint32_t ms;
124
125 if (nv50_context(pipe)->screen->tesla->oclass >= NVA3_3D_CLASS) {
126 SB_BEGIN_3D(so, BLEND_INDEPENDENT, 1);
127 SB_DATA (so, cso->independent_blend_enable);
128 }
129
130 so->pipe = *cso;
131
132 SB_BEGIN_3D(so, COLOR_MASK_COMMON, 1);
133 SB_DATA (so, !cso->independent_blend_enable);
134
135 SB_BEGIN_3D(so, BLEND_ENABLE_COMMON, 1);
136 SB_DATA (so, !cso->independent_blend_enable);
137
138 if (cso->independent_blend_enable) {
139 SB_BEGIN_3D(so, BLEND_ENABLE(0), 8);
140 for (i = 0; i < 8; ++i) {
141 SB_DATA(so, cso->rt[i].blend_enable);
142 if (cso->rt[i].blend_enable)
143 emit_common_func = true;
144 }
145
146 if (nv50_context(pipe)->screen->tesla->oclass >= NVA3_3D_CLASS) {
147 emit_common_func = false;
148
149 for (i = 0; i < 8; ++i) {
150 if (!cso->rt[i].blend_enable)
151 continue;
152 SB_BEGIN_3D_(so, NVA3_3D_IBLEND_EQUATION_RGB(i), 6);
153 SB_DATA (so, nvgl_blend_eqn(cso->rt[i].rgb_func));
154 SB_DATA (so, nv50_blend_fac(cso->rt[i].rgb_src_factor));
155 SB_DATA (so, nv50_blend_fac(cso->rt[i].rgb_dst_factor));
156 SB_DATA (so, nvgl_blend_eqn(cso->rt[i].alpha_func));
157 SB_DATA (so, nv50_blend_fac(cso->rt[i].alpha_src_factor));
158 SB_DATA (so, nv50_blend_fac(cso->rt[i].alpha_dst_factor));
159 }
160 }
161 } else {
162 SB_BEGIN_3D(so, BLEND_ENABLE(0), 1);
163 SB_DATA (so, cso->rt[0].blend_enable);
164 }
165
166 if (emit_common_func) {
167 SB_BEGIN_3D(so, BLEND_EQUATION_RGB, 5);
168 SB_DATA (so, nvgl_blend_eqn(cso->rt[0].rgb_func));
169 SB_DATA (so, nv50_blend_fac(cso->rt[0].rgb_src_factor));
170 SB_DATA (so, nv50_blend_fac(cso->rt[0].rgb_dst_factor));
171 SB_DATA (so, nvgl_blend_eqn(cso->rt[0].alpha_func));
172 SB_DATA (so, nv50_blend_fac(cso->rt[0].alpha_src_factor));
173 SB_BEGIN_3D(so, BLEND_FUNC_DST_ALPHA, 1);
174 SB_DATA (so, nv50_blend_fac(cso->rt[0].alpha_dst_factor));
175 }
176
177 if (cso->logicop_enable) {
178 SB_BEGIN_3D(so, LOGIC_OP_ENABLE, 2);
179 SB_DATA (so, 1);
180 SB_DATA (so, nvgl_logicop_func(cso->logicop_func));
181 } else {
182 SB_BEGIN_3D(so, LOGIC_OP_ENABLE, 1);
183 SB_DATA (so, 0);
184 }
185
186 if (cso->independent_blend_enable) {
187 SB_BEGIN_3D(so, COLOR_MASK(0), 8);
188 for (i = 0; i < 8; ++i)
189 SB_DATA(so, nv50_colormask(cso->rt[i].colormask));
190 } else {
191 SB_BEGIN_3D(so, COLOR_MASK(0), 1);
192 SB_DATA (so, nv50_colormask(cso->rt[0].colormask));
193 }
194
195 ms = 0;
196 if (cso->alpha_to_coverage)
197 ms |= NV50_3D_MULTISAMPLE_CTRL_ALPHA_TO_COVERAGE;
198 if (cso->alpha_to_one)
199 ms |= NV50_3D_MULTISAMPLE_CTRL_ALPHA_TO_ONE;
200
201 SB_BEGIN_3D(so, MULTISAMPLE_CTRL, 1);
202 SB_DATA (so, ms);
203
204 assert(so->size <= ARRAY_SIZE(so->state));
205 return so;
206 }
207
208 static void
209 nv50_blend_state_bind(struct pipe_context *pipe, void *hwcso)
210 {
211 struct nv50_context *nv50 = nv50_context(pipe);
212
213 nv50->blend = hwcso;
214 nv50->dirty_3d |= NV50_NEW_3D_BLEND;
215 }
216
217 static void
218 nv50_blend_state_delete(struct pipe_context *pipe, void *hwcso)
219 {
220 FREE(hwcso);
221 }
222
223 /* NOTE: ignoring line_last_pixel */
224 static void *
225 nv50_rasterizer_state_create(struct pipe_context *pipe,
226 const struct pipe_rasterizer_state *cso)
227 {
228 struct nv50_rasterizer_stateobj *so;
229 uint32_t reg;
230
231 so = CALLOC_STRUCT(nv50_rasterizer_stateobj);
232 if (!so)
233 return NULL;
234 so->pipe = *cso;
235
236 #ifndef NV50_SCISSORS_CLIPPING
237 for (int i = 0; i < NV50_MAX_VIEWPORTS; i++) {
238 SB_BEGIN_3D(so, SCISSOR_ENABLE(i), 1);
239 SB_DATA (so, cso->scissor);
240 }
241 #endif
242
243 SB_BEGIN_3D(so, SHADE_MODEL, 1);
244 SB_DATA (so, cso->flatshade ? NV50_3D_SHADE_MODEL_FLAT :
245 NV50_3D_SHADE_MODEL_SMOOTH);
246 SB_BEGIN_3D(so, PROVOKING_VERTEX_LAST, 1);
247 SB_DATA (so, !cso->flatshade_first);
248 SB_BEGIN_3D(so, VERTEX_TWO_SIDE_ENABLE, 1);
249 SB_DATA (so, cso->light_twoside);
250
251 SB_BEGIN_3D(so, FRAG_COLOR_CLAMP_EN, 1);
252 SB_DATA (so, cso->clamp_fragment_color ? 0x11111111 : 0x00000000);
253
254 SB_BEGIN_3D(so, MULTISAMPLE_ENABLE, 1);
255 SB_DATA (so, cso->multisample);
256
257 SB_BEGIN_3D(so, LINE_WIDTH, 1);
258 SB_DATA (so, fui(cso->line_width));
259 SB_BEGIN_3D(so, LINE_SMOOTH_ENABLE, 1);
260 SB_DATA (so, cso->line_smooth);
261
262 SB_BEGIN_3D(so, LINE_STIPPLE_ENABLE, 1);
263 if (cso->line_stipple_enable) {
264 SB_DATA (so, 1);
265 SB_BEGIN_3D(so, LINE_STIPPLE, 1);
266 SB_DATA (so, (cso->line_stipple_pattern << 8) |
267 cso->line_stipple_factor);
268 } else {
269 SB_DATA (so, 0);
270 }
271
272 if (!cso->point_size_per_vertex) {
273 SB_BEGIN_3D(so, POINT_SIZE, 1);
274 SB_DATA (so, fui(cso->point_size));
275 }
276 SB_BEGIN_3D(so, POINT_SPRITE_ENABLE, 1);
277 SB_DATA (so, cso->point_quad_rasterization);
278 SB_BEGIN_3D(so, POINT_SMOOTH_ENABLE, 1);
279 SB_DATA (so, cso->point_smooth);
280
281 SB_BEGIN_3D(so, POLYGON_MODE_FRONT, 3);
282 SB_DATA (so, nvgl_polygon_mode(cso->fill_front));
283 SB_DATA (so, nvgl_polygon_mode(cso->fill_back));
284 SB_DATA (so, cso->poly_smooth);
285
286 SB_BEGIN_3D(so, CULL_FACE_ENABLE, 3);
287 SB_DATA (so, cso->cull_face != PIPE_FACE_NONE);
288 SB_DATA (so, cso->front_ccw ? NV50_3D_FRONT_FACE_CCW :
289 NV50_3D_FRONT_FACE_CW);
290 switch (cso->cull_face) {
291 case PIPE_FACE_FRONT_AND_BACK:
292 SB_DATA(so, NV50_3D_CULL_FACE_FRONT_AND_BACK);
293 break;
294 case PIPE_FACE_FRONT:
295 SB_DATA(so, NV50_3D_CULL_FACE_FRONT);
296 break;
297 case PIPE_FACE_BACK:
298 default:
299 SB_DATA(so, NV50_3D_CULL_FACE_BACK);
300 break;
301 }
302
303 SB_BEGIN_3D(so, POLYGON_STIPPLE_ENABLE, 1);
304 SB_DATA (so, cso->poly_stipple_enable);
305 SB_BEGIN_3D(so, POLYGON_OFFSET_POINT_ENABLE, 3);
306 SB_DATA (so, cso->offset_point);
307 SB_DATA (so, cso->offset_line);
308 SB_DATA (so, cso->offset_tri);
309
310 if (cso->offset_point || cso->offset_line || cso->offset_tri) {
311 SB_BEGIN_3D(so, POLYGON_OFFSET_FACTOR, 1);
312 SB_DATA (so, fui(cso->offset_scale));
313 SB_BEGIN_3D(so, POLYGON_OFFSET_UNITS, 1);
314 SB_DATA (so, fui(cso->offset_units * 2.0f));
315 SB_BEGIN_3D(so, POLYGON_OFFSET_CLAMP, 1);
316 SB_DATA (so, fui(cso->offset_clamp));
317 }
318
319 if (cso->depth_clip_near) {
320 reg = 0;
321 } else {
322 reg =
323 NV50_3D_VIEW_VOLUME_CLIP_CTRL_DEPTH_CLAMP_NEAR |
324 NV50_3D_VIEW_VOLUME_CLIP_CTRL_DEPTH_CLAMP_FAR |
325 NV50_3D_VIEW_VOLUME_CLIP_CTRL_UNK12_UNK1;
326 }
327 #ifndef NV50_SCISSORS_CLIPPING
328 reg |=
329 NV50_3D_VIEW_VOLUME_CLIP_CTRL_UNK7 |
330 NV50_3D_VIEW_VOLUME_CLIP_CTRL_UNK12_UNK1;
331 #endif
332 SB_BEGIN_3D(so, VIEW_VOLUME_CLIP_CTRL, 1);
333 SB_DATA (so, reg);
334
335 SB_BEGIN_3D(so, DEPTH_CLIP_NEGATIVE_Z, 1);
336 SB_DATA (so, cso->clip_halfz);
337
338 SB_BEGIN_3D(so, PIXEL_CENTER_INTEGER, 1);
339 SB_DATA (so, !cso->half_pixel_center);
340
341 assert(so->size <= ARRAY_SIZE(so->state));
342 return (void *)so;
343 }
344
345 static void
346 nv50_rasterizer_state_bind(struct pipe_context *pipe, void *hwcso)
347 {
348 struct nv50_context *nv50 = nv50_context(pipe);
349
350 nv50->rast = hwcso;
351 nv50->dirty_3d |= NV50_NEW_3D_RASTERIZER;
352 }
353
354 static void
355 nv50_rasterizer_state_delete(struct pipe_context *pipe, void *hwcso)
356 {
357 FREE(hwcso);
358 }
359
360 static void *
361 nv50_zsa_state_create(struct pipe_context *pipe,
362 const struct pipe_depth_stencil_alpha_state *cso)
363 {
364 struct nv50_zsa_stateobj *so = CALLOC_STRUCT(nv50_zsa_stateobj);
365
366 so->pipe = *cso;
367
368 SB_BEGIN_3D(so, DEPTH_WRITE_ENABLE, 1);
369 SB_DATA (so, cso->depth_writemask);
370 SB_BEGIN_3D(so, DEPTH_TEST_ENABLE, 1);
371 if (cso->depth_enabled) {
372 SB_DATA (so, 1);
373 SB_BEGIN_3D(so, DEPTH_TEST_FUNC, 1);
374 SB_DATA (so, nvgl_comparison_op(cso->depth_func));
375 } else {
376 SB_DATA (so, 0);
377 }
378
379 SB_BEGIN_3D(so, DEPTH_BOUNDS_EN, 1);
380 if (cso->depth_bounds_test) {
381 SB_DATA (so, 1);
382 SB_BEGIN_3D(so, DEPTH_BOUNDS(0), 2);
383 SB_DATA (so, fui(cso->depth_bounds_min));
384 SB_DATA (so, fui(cso->depth_bounds_max));
385 } else {
386 SB_DATA (so, 0);
387 }
388
389 if (cso->stencil[0].enabled) {
390 SB_BEGIN_3D(so, STENCIL_ENABLE, 5);
391 SB_DATA (so, 1);
392 SB_DATA (so, nvgl_stencil_op(cso->stencil[0].fail_op));
393 SB_DATA (so, nvgl_stencil_op(cso->stencil[0].zfail_op));
394 SB_DATA (so, nvgl_stencil_op(cso->stencil[0].zpass_op));
395 SB_DATA (so, nvgl_comparison_op(cso->stencil[0].func));
396 SB_BEGIN_3D(so, STENCIL_FRONT_MASK, 2);
397 SB_DATA (so, cso->stencil[0].writemask);
398 SB_DATA (so, cso->stencil[0].valuemask);
399 } else {
400 SB_BEGIN_3D(so, STENCIL_ENABLE, 1);
401 SB_DATA (so, 0);
402 }
403
404 if (cso->stencil[1].enabled) {
405 assert(cso->stencil[0].enabled);
406 SB_BEGIN_3D(so, STENCIL_TWO_SIDE_ENABLE, 5);
407 SB_DATA (so, 1);
408 SB_DATA (so, nvgl_stencil_op(cso->stencil[1].fail_op));
409 SB_DATA (so, nvgl_stencil_op(cso->stencil[1].zfail_op));
410 SB_DATA (so, nvgl_stencil_op(cso->stencil[1].zpass_op));
411 SB_DATA (so, nvgl_comparison_op(cso->stencil[1].func));
412 SB_BEGIN_3D(so, STENCIL_BACK_MASK, 2);
413 SB_DATA (so, cso->stencil[1].writemask);
414 SB_DATA (so, cso->stencil[1].valuemask);
415 } else {
416 SB_BEGIN_3D(so, STENCIL_TWO_SIDE_ENABLE, 1);
417 SB_DATA (so, 0);
418 }
419
420 SB_BEGIN_3D(so, ALPHA_TEST_ENABLE, 1);
421 if (cso->alpha_enabled) {
422 SB_DATA (so, 1);
423 SB_BEGIN_3D(so, ALPHA_TEST_REF, 2);
424 SB_DATA (so, fui(cso->alpha_ref_value));
425 SB_DATA (so, nvgl_comparison_op(cso->alpha_func));
426 } else {
427 SB_DATA (so, 0);
428 }
429
430 SB_BEGIN_3D(so, CB_ADDR, 1);
431 SB_DATA (so, NV50_CB_AUX_ALPHATEST_OFFSET << (8 - 2) | NV50_CB_AUX);
432 SB_BEGIN_3D(so, CB_DATA(0), 1);
433 SB_DATA (so, fui(cso->alpha_ref_value));
434
435 assert(so->size <= ARRAY_SIZE(so->state));
436 return (void *)so;
437 }
438
439 static void
440 nv50_zsa_state_bind(struct pipe_context *pipe, void *hwcso)
441 {
442 struct nv50_context *nv50 = nv50_context(pipe);
443
444 nv50->zsa = hwcso;
445 nv50->dirty_3d |= NV50_NEW_3D_ZSA;
446 }
447
448 static void
449 nv50_zsa_state_delete(struct pipe_context *pipe, void *hwcso)
450 {
451 FREE(hwcso);
452 }
453
454 /* ====================== SAMPLERS AND TEXTURES ================================
455 */
456
457 static inline unsigned
458 nv50_tsc_wrap_mode(unsigned wrap)
459 {
460 switch (wrap) {
461 case PIPE_TEX_WRAP_REPEAT:
462 return G80_TSC_WRAP_WRAP;
463 case PIPE_TEX_WRAP_MIRROR_REPEAT:
464 return G80_TSC_WRAP_MIRROR;
465 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
466 return G80_TSC_WRAP_CLAMP_TO_EDGE;
467 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
468 return G80_TSC_WRAP_BORDER;
469 case PIPE_TEX_WRAP_CLAMP:
470 return G80_TSC_WRAP_CLAMP_OGL;
471 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
472 return G80_TSC_WRAP_MIRROR_ONCE_CLAMP_TO_EDGE;
473 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
474 return G80_TSC_WRAP_MIRROR_ONCE_BORDER;
475 case PIPE_TEX_WRAP_MIRROR_CLAMP:
476 return G80_TSC_WRAP_MIRROR_ONCE_CLAMP_OGL;
477 default:
478 NOUVEAU_ERR("unknown wrap mode: %d\n", wrap);
479 return G80_TSC_WRAP_WRAP;
480 }
481 }
482
483 void *
484 nv50_sampler_state_create(struct pipe_context *pipe,
485 const struct pipe_sampler_state *cso)
486 {
487 struct nv50_tsc_entry *so = MALLOC_STRUCT(nv50_tsc_entry);
488 float f[2];
489
490 so->id = -1;
491
492 so->tsc[0] = (0x00026000 |
493 (nv50_tsc_wrap_mode(cso->wrap_s) << 0) |
494 (nv50_tsc_wrap_mode(cso->wrap_t) << 3) |
495 (nv50_tsc_wrap_mode(cso->wrap_r) << 6));
496
497 switch (cso->mag_img_filter) {
498 case PIPE_TEX_FILTER_LINEAR:
499 so->tsc[1] = G80_TSC_1_MAG_FILTER_LINEAR;
500 break;
501 case PIPE_TEX_FILTER_NEAREST:
502 default:
503 so->tsc[1] = G80_TSC_1_MAG_FILTER_NEAREST;
504 break;
505 }
506
507 switch (cso->min_img_filter) {
508 case PIPE_TEX_FILTER_LINEAR:
509 so->tsc[1] |= G80_TSC_1_MIN_FILTER_LINEAR;
510 break;
511 case PIPE_TEX_FILTER_NEAREST:
512 default:
513 so->tsc[1] |= G80_TSC_1_MIN_FILTER_NEAREST;
514 break;
515 }
516
517 switch (cso->min_mip_filter) {
518 case PIPE_TEX_MIPFILTER_LINEAR:
519 so->tsc[1] |= G80_TSC_1_MIP_FILTER_LINEAR;
520 break;
521 case PIPE_TEX_MIPFILTER_NEAREST:
522 so->tsc[1] |= G80_TSC_1_MIP_FILTER_NEAREST;
523 break;
524 case PIPE_TEX_MIPFILTER_NONE:
525 default:
526 so->tsc[1] |= G80_TSC_1_MIP_FILTER_NONE;
527 break;
528 }
529
530 if (nouveau_screen(pipe->screen)->class_3d >= NVE4_3D_CLASS) {
531 if (cso->seamless_cube_map)
532 so->tsc[1] |= GK104_TSC_1_CUBEMAP_INTERFACE_FILTERING;
533 if (cso->unnormalized_coords)
534 so->tsc[1] |= GK104_TSC_1_FLOAT_COORD_NORMALIZATION_FORCE_UNNORMALIZED_COORDS;
535 } else {
536 so->seamless_cube_map = cso->seamless_cube_map;
537 }
538
539 if (nouveau_screen(pipe->screen)->class_3d >= GM200_3D_CLASS) {
540 if (cso->reduction_mode == PIPE_TEX_REDUCTION_MIN)
541 so->tsc[1] |= GM204_TSC_1_REDUCTION_MODE_MIN;
542 if (cso->reduction_mode == PIPE_TEX_REDUCTION_MAX)
543 so->tsc[1] |= GM204_TSC_1_REDUCTION_MODE_MAX;
544 }
545
546 if (cso->max_anisotropy >= 16)
547 so->tsc[0] |= (7 << 20);
548 else
549 if (cso->max_anisotropy >= 12)
550 so->tsc[0] |= (6 << 20);
551 else {
552 so->tsc[0] |= (cso->max_anisotropy >> 1) << 20;
553
554 if (cso->max_anisotropy >= 4)
555 so->tsc[1] |= 6 << G80_TSC_1_TRILIN_OPT__SHIFT;
556 else
557 if (cso->max_anisotropy >= 2)
558 so->tsc[1] |= 4 << G80_TSC_1_TRILIN_OPT__SHIFT;
559 }
560
561 if (cso->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
562 /* NOTE: must be deactivated for non-shadow textures */
563 so->tsc[0] |= (1 << 9);
564 so->tsc[0] |= (nvgl_comparison_op(cso->compare_func) & 0x7) << 10;
565 }
566
567 f[0] = CLAMP(cso->lod_bias, -16.0f, 15.0f);
568 so->tsc[1] |= ((int)(f[0] * 256.0f) & 0x1fff) << 12;
569
570 f[0] = CLAMP(cso->min_lod, 0.0f, 15.0f);
571 f[1] = CLAMP(cso->max_lod, 0.0f, 15.0f);
572 so->tsc[2] =
573 (((int)(f[1] * 256.0f) & 0xfff) << 12) | ((int)(f[0] * 256.0f) & 0xfff);
574
575 so->tsc[2] |=
576 util_format_linear_float_to_srgb_8unorm(cso->border_color.f[0]) << 24;
577 so->tsc[3] =
578 util_format_linear_float_to_srgb_8unorm(cso->border_color.f[1]) << 12;
579 so->tsc[3] |=
580 util_format_linear_float_to_srgb_8unorm(cso->border_color.f[2]) << 20;
581
582 so->tsc[4] = fui(cso->border_color.f[0]);
583 so->tsc[5] = fui(cso->border_color.f[1]);
584 so->tsc[6] = fui(cso->border_color.f[2]);
585 so->tsc[7] = fui(cso->border_color.f[3]);
586
587 return (void *)so;
588 }
589
590 static void
591 nv50_sampler_state_delete(struct pipe_context *pipe, void *hwcso)
592 {
593 unsigned s, i;
594
595 for (s = 0; s < NV50_MAX_SHADER_STAGES; ++s) {
596 assert(nv50_context(pipe)->num_samplers[s] <= PIPE_MAX_SAMPLERS);
597 for (i = 0; i < nv50_context(pipe)->num_samplers[s]; ++i)
598 if (nv50_context(pipe)->samplers[s][i] == hwcso)
599 nv50_context(pipe)->samplers[s][i] = NULL;
600 }
601
602 nv50_screen_tsc_free(nv50_context(pipe)->screen, nv50_tsc_entry(hwcso));
603
604 FREE(hwcso);
605 }
606
607 static inline void
608 nv50_stage_sampler_states_bind(struct nv50_context *nv50, int s,
609 unsigned nr, void **hwcsos)
610 {
611 unsigned highest_found = 0;
612 unsigned i;
613
614 assert(nr <= PIPE_MAX_SAMPLERS);
615 for (i = 0; i < nr; ++i) {
616 struct nv50_tsc_entry *hwcso = hwcsos ? nv50_tsc_entry(hwcsos[i]) : NULL;
617 struct nv50_tsc_entry *old = nv50->samplers[s][i];
618
619 if (hwcso)
620 highest_found = i;
621
622 nv50->samplers[s][i] = hwcso;
623 if (old)
624 nv50_screen_tsc_unlock(nv50->screen, old);
625 }
626 assert(nv50->num_samplers[s] <= PIPE_MAX_SAMPLERS);
627 if (nr >= nv50->num_samplers[s])
628 nv50->num_samplers[s] = highest_found + 1;
629 }
630
631 static void
632 nv50_bind_sampler_states(struct pipe_context *pipe,
633 mesa_shader_stage shader, unsigned start,
634 unsigned num_samplers, void **samplers)
635 {
636 unsigned s = nv50_context_shader_stage(shader);
637
638 assert(start == 0);
639 nv50_stage_sampler_states_bind(nv50_context(pipe), s, num_samplers,
640 samplers);
641
642 if (unlikely(s == NV50_SHADER_STAGE_COMPUTE))
643 nv50_context(pipe)->dirty_cp |= NV50_NEW_CP_SAMPLERS;
644 else
645 nv50_context(pipe)->dirty_3d |= NV50_NEW_3D_SAMPLERS;
646 }
647
648
649
650 /* NOTE: only called when not referenced anywhere, won't be bound */
651 static void
652 nv50_sampler_view_destroy(struct pipe_context *pipe,
653 struct pipe_sampler_view *view)
654 {
655 pipe_resource_reference(&view->texture, NULL);
656
657 nv50_screen_tic_free(nv50_context(pipe)->screen, nv50_tic_entry(view));
658
659 FREE(nv50_tic_entry(view));
660 }
661
662 static inline void
663 nv50_stage_set_sampler_views(struct nv50_context *nv50, int s,
664 unsigned nr,
665 struct pipe_sampler_view **views)
666 {
667 unsigned i;
668
669 assert(nr <= PIPE_MAX_SAMPLERS);
670 for (i = 0; i < nr; ++i) {
671 struct pipe_sampler_view *view = views ? views[i] : NULL;
672 struct nv50_tic_entry *old = nv50_tic_entry(nv50->textures[s][i]);
673 if (old)
674 nv50_screen_tic_unlock(nv50->screen, old);
675
676 if (view && view->texture) {
677 struct pipe_resource *res = view->texture;
678 if (res->target == PIPE_BUFFER &&
679 (res->flags & PIPE_RESOURCE_FLAG_MAP_COHERENT))
680 nv50->textures_coherent[s] |= 1 << i;
681 else
682 nv50->textures_coherent[s] &= ~(1 << i);
683 } else {
684 nv50->textures_coherent[s] &= ~(1 << i);
685 }
686
687 pipe_sampler_view_reference(&nv50->textures[s][i], view);
688 }
689
690 assert(nv50->num_textures[s] <= PIPE_MAX_SAMPLERS);
691 for (i = nr; i < nv50->num_textures[s]; ++i) {
692 struct nv50_tic_entry *old = nv50_tic_entry(nv50->textures[s][i]);
693 if (!old)
694 continue;
695 nv50_screen_tic_unlock(nv50->screen, old);
696
697 pipe_sampler_view_reference(&nv50->textures[s][i], NULL);
698 }
699
700 nv50->num_textures[s] = nr;
701 }
702
703 static void
704 nv50_set_sampler_views(struct pipe_context *pipe, mesa_shader_stage shader,
705 unsigned start, unsigned nr,
706 unsigned unbind_num_trailing_slots,
707 struct pipe_sampler_view **views)
708 {
709 struct nv50_context *nv50 = nv50_context(pipe);
710 unsigned s = nv50_context_shader_stage(shader);
711
712 assert(start == 0);
713 nv50_stage_set_sampler_views(nv50, s, nr, views);
714
715 if (unlikely(s == NV50_SHADER_STAGE_COMPUTE)) {
716 nouveau_bufctx_reset(nv50->bufctx_cp, NV50_BIND_CP_TEXTURES);
717
718 nv50->dirty_cp |= NV50_NEW_CP_TEXTURES;
719 } else {
720 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_TEXTURES);
721
722 nv50->dirty_3d |= NV50_NEW_3D_TEXTURES;
723 }
724 }
725
726
727
728 /* ============================= SHADERS =======================================
729 */
730
731 static void *
732 nv50_sp_state_create(struct pipe_context *pipe,
733 const struct pipe_shader_state *cso,
734 mesa_shader_stage type)
735 {
736 struct nv50_program *prog;
737
738 prog = CALLOC_STRUCT(nv50_program);
739 if (!prog)
740 return NULL;
741
742 prog->type = type;
743
744 switch (cso->type) {
745 case PIPE_SHADER_IR_TGSI:
746 prog->nir = tgsi_to_nir(cso->tokens, pipe->screen, false);
747 break;
748 case PIPE_SHADER_IR_NIR:
749 prog->nir = cso->ir.nir;
750 break;
751 default:
752 assert(!"unsupported IR!");
753 free(prog);
754 return NULL;
755 }
756
757 if (cso->stream_output.num_outputs)
758 prog->stream_output = cso->stream_output;
759
760 prog->translated = nv50_program_translate(
761 prog, nv50_context(pipe)->screen->base.device->chipset,
762 &nouveau_context(pipe)->debug);
763
764 return (void *)prog;
765 }
766
767 static void
768 nv50_sp_state_delete(struct pipe_context *pipe, void *hwcso)
769 {
770 struct nv50_context *nv50 = nv50_context(pipe);
771 struct nv50_program *prog = (struct nv50_program *)hwcso;
772
773 simple_mtx_lock(&nv50->screen->state_lock);
774 nv50_program_destroy(nv50, prog);
775 simple_mtx_unlock(&nv50->screen->state_lock);
776
777 ralloc_free(prog->nir);
778 FREE(prog);
779 }
780
781 static void *
782 nv50_vp_state_create(struct pipe_context *pipe,
783 const struct pipe_shader_state *cso)
784 {
785 return nv50_sp_state_create(pipe, cso, MESA_SHADER_VERTEX);
786 }
787
788 static void
789 nv50_vp_state_bind(struct pipe_context *pipe, void *hwcso)
790 {
791 struct nv50_context *nv50 = nv50_context(pipe);
792
793 nv50->vertprog = hwcso;
794 nv50->dirty_3d |= NV50_NEW_3D_VERTPROG;
795 }
796
797 static void *
798 nv50_fp_state_create(struct pipe_context *pipe,
799 const struct pipe_shader_state *cso)
800 {
801 return nv50_sp_state_create(pipe, cso, MESA_SHADER_FRAGMENT);
802 }
803
804 static void
805 nv50_fp_state_bind(struct pipe_context *pipe, void *hwcso)
806 {
807 struct nv50_context *nv50 = nv50_context(pipe);
808
809 nv50->fragprog = hwcso;
810 nv50->dirty_3d |= NV50_NEW_3D_FRAGPROG;
811 }
812
813 static void *
814 nv50_gp_state_create(struct pipe_context *pipe,
815 const struct pipe_shader_state *cso)
816 {
817 return nv50_sp_state_create(pipe, cso, MESA_SHADER_GEOMETRY);
818 }
819
820 static void
821 nv50_gp_state_bind(struct pipe_context *pipe, void *hwcso)
822 {
823 struct nv50_context *nv50 = nv50_context(pipe);
824
825 nv50->gmtyprog = hwcso;
826 nv50->dirty_3d |= NV50_NEW_3D_GMTYPROG;
827 }
828
829 static void *
830 nv50_cp_state_create(struct pipe_context *pipe,
831 const struct pipe_compute_state *cso)
832 {
833 struct nv50_program *prog;
834
835 prog = CALLOC_STRUCT(nv50_program);
836 if (!prog)
837 return NULL;
838 prog->type = MESA_SHADER_COMPUTE;
839
840 switch(cso->ir_type) {
841 case PIPE_SHADER_IR_TGSI: {
842 const struct tgsi_token *tokens = cso->prog;
843 prog->nir = tgsi_to_nir(tokens, pipe->screen, false);
844 break;
845 }
846 case PIPE_SHADER_IR_NIR:
847 prog->nir = (nir_shader *)cso->prog;
848 break;
849 default:
850 assert(!"unsupported IR!");
851 free(prog);
852 return NULL;
853 }
854
855 prog->cp.smem_size = cso->static_shared_mem;
856
857 return (void *)prog;
858 }
859
860 static void
861 nv50_cp_state_bind(struct pipe_context *pipe, void *hwcso)
862 {
863 struct nv50_context *nv50 = nv50_context(pipe);
864
865 nv50->compprog = hwcso;
866 nv50->dirty_cp |= NV50_NEW_CP_PROGRAM;
867 }
868
869 static void
870 nv50_get_compute_state_info(struct pipe_context *pipe, void *hwcso,
871 struct pipe_compute_state_object_info *info)
872 {
873 struct nv50_context *nv50 = nv50_context(pipe);
874 struct nv50_program *prog = (struct nv50_program *)hwcso;
875 uint16_t obj_class = nv50->screen->compute->oclass;
876 uint32_t smregs = obj_class >= NVA3_COMPUTE_CLASS ? 16384 : 8192;
877 uint32_t threads = smregs / align(prog->max_gpr, 4);
878
879 info->max_threads = MIN2(ROUND_DOWN_TO(threads, 32), 512);
880 info->private_memory = prog->tls_space;
881 info->preferred_simd_size = 32;
882 info->simd_sizes = 32;
883 }
884
885 static void
886 nv50_set_constant_buffer(struct pipe_context *pipe,
887 mesa_shader_stage shader, uint index,
888 const struct pipe_constant_buffer *cb)
889 {
890 struct nv50_context *nv50 = nv50_context(pipe);
891 struct pipe_resource *res = cb ? cb->buffer : NULL;
892 const unsigned s = nv50_context_shader_stage(shader);
893 const unsigned i = index;
894
895 if (unlikely(shader == MESA_SHADER_COMPUTE)) {
896 if (nv50->constbuf[s][i].user)
897 nv50->constbuf[s][i].u.buf = NULL;
898 else
899 if (nv50->constbuf[s][i].u.buf)
900 nouveau_bufctx_reset(nv50->bufctx_cp, NV50_BIND_CP_CB(i));
901
902 nv50->dirty_cp |= NV50_NEW_CP_CONSTBUF;
903 } else {
904 if (nv50->constbuf[s][i].user)
905 nv50->constbuf[s][i].u.buf = NULL;
906 else
907 if (nv50->constbuf[s][i].u.buf)
908 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_CB(s, i));
909
910 nv50->dirty_3d |= NV50_NEW_3D_CONSTBUF;
911 }
912 nv50->constbuf_dirty[s] |= 1 << i;
913
914 if (nv50->constbuf[s][i].u.buf)
915 nv04_resource(nv50->constbuf[s][i].u.buf)->cb_bindings[s] &= ~(1 << i);
916
917 pipe_resource_reference(&nv50->constbuf[s][i].u.buf, res);
918
919 nv50->constbuf[s][i].user = (cb && cb->user_buffer) ? true : false;
920 if (nv50->constbuf[s][i].user) {
921 nv50->constbuf[s][i].u.data = cb->user_buffer;
922 nv50->constbuf[s][i].size = MIN2(cb->buffer_size, 0x10000);
923 nv50->constbuf_valid[s] |= 1 << i;
924 nv50->constbuf_coherent[s] &= ~(1 << i);
925 } else
926 if (cb) {
927 nv50->constbuf[s][i].offset = cb->buffer_offset;
928 nv50->constbuf[s][i].size = MIN2(align(cb->buffer_size, 0x100), 0x10000);
929 nv50->constbuf_valid[s] |= 1 << i;
930 if (res && res->flags & PIPE_RESOURCE_FLAG_MAP_COHERENT)
931 nv50->constbuf_coherent[s] |= 1 << i;
932 else
933 nv50->constbuf_coherent[s] &= ~(1 << i);
934 }
935 else {
936 nv50->constbuf_valid[s] &= ~(1 << i);
937 nv50->constbuf_coherent[s] &= ~(1 << i);
938 }
939 }
940
941 /* =============================================================================
942 */
943
944 static void
945 nv50_set_blend_color(struct pipe_context *pipe,
946 const struct pipe_blend_color *bcol)
947 {
948 struct nv50_context *nv50 = nv50_context(pipe);
949
950 nv50->blend_colour = *bcol;
951 nv50->dirty_3d |= NV50_NEW_3D_BLEND_COLOUR;
952 }
953
954 static void
955 nv50_set_stencil_ref(struct pipe_context *pipe,
956 const struct pipe_stencil_ref sr)
957 {
958 struct nv50_context *nv50 = nv50_context(pipe);
959
960 nv50->stencil_ref = sr;
961 nv50->dirty_3d |= NV50_NEW_3D_STENCIL_REF;
962 }
963
964 static void
965 nv50_set_clip_state(struct pipe_context *pipe,
966 const struct pipe_clip_state *clip)
967 {
968 struct nv50_context *nv50 = nv50_context(pipe);
969
970 memcpy(nv50->clip.ucp, clip->ucp, sizeof(clip->ucp));
971
972 nv50->dirty_3d |= NV50_NEW_3D_CLIP;
973 }
974
975 static void
976 nv50_set_sample_mask(struct pipe_context *pipe, unsigned sample_mask)
977 {
978 struct nv50_context *nv50 = nv50_context(pipe);
979
980 nv50->sample_mask = sample_mask;
981 nv50->dirty_3d |= NV50_NEW_3D_SAMPLE_MASK;
982 }
983
984 static void
985 nv50_set_min_samples(struct pipe_context *pipe, unsigned min_samples)
986 {
987 struct nv50_context *nv50 = nv50_context(pipe);
988
989 if (nv50->min_samples != min_samples) {
990 nv50->min_samples = min_samples;
991 nv50->dirty_3d |= NV50_NEW_3D_MIN_SAMPLES;
992 }
993 }
994
995 static void
996 nv50_set_framebuffer_state(struct pipe_context *pipe,
997 const struct pipe_framebuffer_state *fb)
998 {
999 struct nv50_context *nv50 = nv50_context(pipe);
1000
1001 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_FB);
1002
1003 nv50_framebuffer_init(pipe, fb, nv50->fb_cbufs, &nv50->fb_zsbuf);
1004 util_copy_framebuffer_state(&nv50->framebuffer, fb);
1005
1006 nv50->dirty_3d |= NV50_NEW_3D_FRAMEBUFFER | NV50_NEW_3D_TEXTURES;
1007 }
1008
1009 static void
1010 nv50_set_polygon_stipple(struct pipe_context *pipe,
1011 const struct pipe_poly_stipple *stipple)
1012 {
1013 struct nv50_context *nv50 = nv50_context(pipe);
1014
1015 nv50->stipple = *stipple;
1016 nv50->dirty_3d |= NV50_NEW_3D_STIPPLE;
1017 }
1018
1019 static void
1020 nv50_set_scissor_states(struct pipe_context *pipe,
1021 unsigned start_slot,
1022 unsigned num_scissors,
1023 const struct pipe_scissor_state *scissor)
1024 {
1025 struct nv50_context *nv50 = nv50_context(pipe);
1026 int i;
1027
1028 assert(start_slot + num_scissors <= NV50_MAX_VIEWPORTS);
1029 for (i = 0; i < num_scissors; i++) {
1030 if (!memcmp(&nv50->scissors[start_slot + i], &scissor[i], sizeof(*scissor)))
1031 continue;
1032 nv50->scissors[start_slot + i] = scissor[i];
1033 nv50->scissors_dirty |= 1 << (start_slot + i);
1034 nv50->dirty_3d |= NV50_NEW_3D_SCISSOR;
1035 }
1036 }
1037
1038 static void
1039 nv50_set_viewport_states(struct pipe_context *pipe,
1040 unsigned start_slot,
1041 unsigned num_viewports,
1042 const struct pipe_viewport_state *vpt)
1043 {
1044 struct nv50_context *nv50 = nv50_context(pipe);
1045 int i;
1046
1047 assert(start_slot + num_viewports <= NV50_MAX_VIEWPORTS);
1048 for (i = 0; i < num_viewports; i++) {
1049 if (!memcmp(&nv50->viewports[start_slot + i], &vpt[i], sizeof(*vpt)))
1050 continue;
1051 nv50->viewports[start_slot + i] = vpt[i];
1052 nv50->viewports_dirty |= 1 << (start_slot + i);
1053 nv50->dirty_3d |= NV50_NEW_3D_VIEWPORT;
1054 }
1055 }
1056
1057 static void
1058 nv50_set_window_rectangles(struct pipe_context *pipe,
1059 bool include,
1060 unsigned num_rectangles,
1061 const struct pipe_scissor_state *rectangles)
1062 {
1063 struct nv50_context *nv50 = nv50_context(pipe);
1064
1065 nv50->window_rect.inclusive = include;
1066 nv50->window_rect.rects = MIN2(num_rectangles, NV50_MAX_WINDOW_RECTANGLES);
1067 memcpy(nv50->window_rect.rect, rectangles,
1068 sizeof(struct pipe_scissor_state) * nv50->window_rect.rects);
1069
1070 nv50->dirty_3d |= NV50_NEW_3D_WINDOW_RECTS;
1071 }
1072
1073 static void
1074 nv50_set_vertex_buffers(struct pipe_context *pipe,
1075 unsigned count,
1076 const struct pipe_vertex_buffer *vb)
1077 {
1078 struct nv50_context *nv50 = nv50_context(pipe);
1079 unsigned i;
1080
1081 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_VERTEX);
1082 nv50->dirty_3d |= NV50_NEW_3D_ARRAYS;
1083
1084 unsigned last_count = nv50->num_vtxbufs;
1085 util_set_vertex_buffers_count(nv50->vtxbuf, &nv50->num_vtxbufs, vb,
1086 count);
1087
1088 unsigned clear_mask =
1089 last_count > count ? BITFIELD_RANGE(count, last_count - count) : 0;
1090 nv50->vbo_user &= clear_mask;
1091 nv50->vbo_constant &= clear_mask;
1092 nv50->vtxbufs_coherent &= clear_mask;
1093
1094 if (!vb) {
1095 clear_mask = ~u_bit_consecutive(0, count);
1096 nv50->vbo_user &= clear_mask;
1097 nv50->vbo_constant &= clear_mask;
1098 nv50->vtxbufs_coherent &= clear_mask;
1099 return;
1100 }
1101
1102 for (i = 0; i < count; ++i) {
1103 unsigned dst_index = i;
1104
1105 if (vb[i].is_user_buffer) {
1106 nv50->vbo_user |= 1 << dst_index;
1107 nv50->vtxbufs_coherent &= ~(1 << dst_index);
1108 } else {
1109 nv50->vbo_user &= ~(1 << dst_index);
1110
1111 if (vb[i].buffer.resource &&
1112 vb[i].buffer.resource->flags & PIPE_RESOURCE_FLAG_MAP_COHERENT)
1113 nv50->vtxbufs_coherent |= (1 << dst_index);
1114 else
1115 nv50->vtxbufs_coherent &= ~(1 << dst_index);
1116 }
1117 }
1118 }
1119
1120 static void
1121 nv50_vertex_state_bind(struct pipe_context *pipe, void *hwcso)
1122 {
1123 struct nv50_context *nv50 = nv50_context(pipe);
1124
1125 nv50->vertex = hwcso;
1126 nv50->dirty_3d |= NV50_NEW_3D_VERTEX;
1127 }
1128
1129 static struct pipe_stream_output_target *
1130 nv50_so_target_create(struct pipe_context *pipe,
1131 struct pipe_resource *res,
1132 unsigned offset, unsigned size)
1133 {
1134 struct nv04_resource *buf = (struct nv04_resource *)res;
1135 struct nv50_so_target *targ = MALLOC_STRUCT(nv50_so_target);
1136 if (!targ)
1137 return NULL;
1138
1139 if (nouveau_context(pipe)->screen->class_3d >= NVA0_3D_CLASS) {
1140 targ->pq = pipe->create_query(pipe,
1141 NVA0_HW_QUERY_STREAM_OUTPUT_BUFFER_OFFSET, 0);
1142 if (!targ->pq) {
1143 FREE(targ);
1144 return NULL;
1145 }
1146 } else {
1147 targ->pq = NULL;
1148 }
1149 targ->clean = true;
1150
1151 targ->pipe.buffer_size = size;
1152 targ->pipe.buffer_offset = offset;
1153 targ->pipe.context = pipe;
1154 targ->pipe.buffer = NULL;
1155 pipe_resource_reference(&targ->pipe.buffer, res);
1156 pipe_reference_init(&targ->pipe.reference, 1);
1157
1158 assert(buf->base.target == PIPE_BUFFER);
1159 util_range_add(&buf->base, &buf->valid_buffer_range, offset, offset + size);
1160
1161 return &targ->pipe;
1162 }
1163
1164 static void
1165 nva0_so_target_save_offset(struct pipe_context *pipe,
1166 struct pipe_stream_output_target *ptarg,
1167 unsigned index, bool serialize)
1168 {
1169 struct nv50_so_target *targ = nv50_so_target(ptarg);
1170
1171 if (serialize) {
1172 struct nouveau_pushbuf *push = nv50_context(pipe)->base.pushbuf;
1173 PUSH_SPACE(push, 2);
1174 BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
1175 PUSH_DATA (push, 0);
1176 }
1177
1178 nv50_query(targ->pq)->index = index;
1179 pipe->end_query(pipe, targ->pq);
1180 }
1181
1182 static void
1183 nv50_so_target_destroy(struct pipe_context *pipe,
1184 struct pipe_stream_output_target *ptarg)
1185 {
1186 struct nv50_so_target *targ = nv50_so_target(ptarg);
1187 if (targ->pq)
1188 pipe->destroy_query(pipe, targ->pq);
1189 pipe_resource_reference(&targ->pipe.buffer, NULL);
1190 FREE(targ);
1191 }
1192
1193 static void
1194 nv50_set_stream_output_targets(struct pipe_context *pipe,
1195 unsigned num_targets,
1196 struct pipe_stream_output_target **targets,
1197 const unsigned *offsets,
1198 enum mesa_prim output_prim)
1199 {
1200 struct nv50_context *nv50 = nv50_context(pipe);
1201 unsigned i;
1202 bool serialize = true;
1203 const bool can_resume = nv50->screen->base.class_3d >= NVA0_3D_CLASS;
1204
1205 assert(num_targets <= 4);
1206
1207 for (i = 0; i < num_targets; ++i) {
1208 const bool changed = nv50->so_target[i] != targets[i];
1209 const bool append = (offsets[i] == (unsigned)-1);
1210 if (!changed && append)
1211 continue;
1212 nv50->so_targets_dirty |= 1 << i;
1213
1214 if (can_resume && changed && nv50->so_target[i]) {
1215 nva0_so_target_save_offset(pipe, nv50->so_target[i], i, serialize);
1216 serialize = false;
1217 }
1218
1219 if (targets[i] && !append) {
1220 nv50_so_target(targets[i])->clean = true;
1221 nv50->so_used[i] = 0;
1222 }
1223
1224 pipe_so_target_reference(&nv50->so_target[i], targets[i]);
1225 }
1226 for (; i < nv50->num_so_targets; ++i) {
1227 if (can_resume && nv50->so_target[i]) {
1228 nva0_so_target_save_offset(pipe, nv50->so_target[i], i, serialize);
1229 serialize = false;
1230 }
1231 pipe_so_target_reference(&nv50->so_target[i], NULL);
1232 nv50->so_targets_dirty |= 1 << i;
1233 }
1234 nv50->num_so_targets = num_targets;
1235
1236 if (nv50->so_targets_dirty) {
1237 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_SO);
1238 nv50->dirty_3d |= NV50_NEW_3D_STRMOUT;
1239 }
1240 }
1241
1242 static bool
1243 nv50_bind_images_range(struct nv50_context *nv50,
1244 unsigned start, unsigned nr,
1245 const struct pipe_image_view *pimages)
1246 {
1247 const unsigned end = start + nr;
1248 unsigned mask = 0;
1249 unsigned i;
1250
1251 if (pimages) {
1252 for (i = start; i < end; ++i) {
1253 struct pipe_image_view *img = &nv50->images[i];
1254 const unsigned p = i - start;
1255
1256 if (img->resource == pimages[p].resource &&
1257 img->format == pimages[p].format &&
1258 img->access == pimages[p].access) {
1259 if (img->resource == NULL)
1260 continue;
1261 if (img->resource->target == PIPE_BUFFER &&
1262 img->u.buf.offset == pimages[p].u.buf.offset &&
1263 img->u.buf.size == pimages[p].u.buf.size)
1264 continue;
1265 if (img->resource->target != PIPE_BUFFER &&
1266 img->u.tex.first_layer == pimages[p].u.tex.first_layer &&
1267 img->u.tex.last_layer == pimages[p].u.tex.last_layer &&
1268 img->u.tex.level == pimages[p].u.tex.level)
1269 continue;
1270 }
1271
1272 mask |= (1 << i);
1273 if (pimages[p].resource)
1274 nv50->images_valid |= (1 << i);
1275 else
1276 nv50->images_valid &= ~(1 << i);
1277
1278 img->format = pimages[p].format;
1279 img->access = pimages[p].access;
1280 if (pimages[p].resource && pimages[p].resource->target == PIPE_BUFFER)
1281 img->u.buf = pimages[p].u.buf;
1282 else
1283 img->u.tex = pimages[p].u.tex;
1284
1285 pipe_resource_reference(
1286 &img->resource, pimages[p].resource);
1287 }
1288 if (!mask)
1289 return false;
1290 } else {
1291 mask = ((1 << nr) - 1) << start;
1292 if (!(nv50->images_valid & mask))
1293 return false;
1294 for (i = start; i < end; ++i) {
1295 pipe_resource_reference(&nv50->images[i].resource, NULL);
1296 }
1297 nv50->images_valid &= ~mask;
1298 }
1299 nv50->images_dirty |= mask;
1300
1301 nouveau_bufctx_reset(nv50->bufctx_cp, NV50_BIND_CP_SUF);
1302
1303 return true;
1304 }
1305
1306 static void
1307 nv50_set_shader_images(struct pipe_context *pipe,
1308 mesa_shader_stage shader,
1309 unsigned start, unsigned nr,
1310 unsigned unbind_num_trailing_slots,
1311 const struct pipe_image_view *images)
1312 {
1313 const unsigned s = nv50_context_shader_stage(shader);
1314
1315 if (s != NV50_SHADER_STAGE_COMPUTE)
1316 return;
1317
1318 nv50_bind_images_range(nv50_context(pipe), start + nr,
1319 unbind_num_trailing_slots, NULL);
1320
1321 if (!nv50_bind_images_range(nv50_context(pipe), start, nr, images))
1322 return;
1323
1324 nv50_context(pipe)->dirty_cp |= NV50_NEW_CP_SURFACES;
1325 }
1326
1327 static bool
1328 nv50_bind_buffers_range(struct nv50_context *nv50,
1329 unsigned start, unsigned nr,
1330 const struct pipe_shader_buffer *pbuffers)
1331 {
1332 const unsigned end = start + nr;
1333 unsigned mask = 0;
1334 unsigned i;
1335
1336 if (pbuffers) {
1337 for (i = start; i < end; ++i) {
1338 struct pipe_shader_buffer *buf = &nv50->buffers[i];
1339 const unsigned p = i - start;
1340 if (buf->buffer == pbuffers[p].buffer &&
1341 buf->buffer_offset == pbuffers[p].buffer_offset &&
1342 buf->buffer_size == pbuffers[p].buffer_size)
1343 continue;
1344
1345 mask |= (1 << i);
1346 if (pbuffers[p].buffer)
1347 nv50->buffers_valid |= (1 << i);
1348 else
1349 nv50->buffers_valid &= ~(1 << i);
1350 buf->buffer_offset = pbuffers[p].buffer_offset;
1351 buf->buffer_size = pbuffers[p].buffer_size;
1352 pipe_resource_reference(&buf->buffer, pbuffers[p].buffer);
1353 }
1354 if (!mask)
1355 return false;
1356 } else {
1357 mask = ((1 << nr) - 1) << start;
1358 if (!(nv50->buffers_valid & mask))
1359 return false;
1360 for (i = start; i < end; ++i)
1361 pipe_resource_reference(&nv50->buffers[i].buffer, NULL);
1362 nv50->buffers_valid &= ~mask;
1363 }
1364 nv50->buffers_dirty |= mask;
1365
1366 nouveau_bufctx_reset(nv50->bufctx_cp, NV50_BIND_CP_BUF);
1367
1368 return true;
1369 }
1370
1371 static void
1372 nv50_set_shader_buffers(struct pipe_context *pipe,
1373 mesa_shader_stage shader,
1374 unsigned start, unsigned nr,
1375 const struct pipe_shader_buffer *buffers,
1376 unsigned writable_bitmask)
1377 {
1378 const unsigned s = nv50_context_shader_stage(shader);
1379
1380 if (s != NV50_SHADER_STAGE_COMPUTE)
1381 return;
1382
1383 if (!nv50_bind_buffers_range(nv50_context(pipe), start, nr, buffers))
1384 return;
1385
1386 nv50_context(pipe)->dirty_cp |= NV50_NEW_CP_BUFFERS;
1387 }
1388
1389 static inline void
1390 nv50_set_global_handle(uint32_t *phandle, struct pipe_resource *res)
1391 {
1392 struct nv04_resource *buf = nv04_resource(res);
1393 if (buf) {
1394 uint64_t limit = (buf->address + buf->base.width0) - 1;
1395 if (limit < (1ULL << 32)) {
1396 *phandle = (uint32_t)buf->address;
1397 } else {
1398 NOUVEAU_ERR("Cannot map into TGSI_RESOURCE_GLOBAL: "
1399 "resource not contained within 32-bit address space !\n");
1400 *phandle = 0;
1401 }
1402 } else {
1403 *phandle = 0;
1404 }
1405 }
1406
1407 static void
1408 nv50_set_global_bindings(struct pipe_context *pipe,
1409 unsigned start, unsigned nr,
1410 struct pipe_resource **resources,
1411 uint32_t **handles)
1412 {
1413 struct nv50_context *nv50 = nv50_context(pipe);
1414 struct pipe_resource **ptr;
1415 unsigned i;
1416 const unsigned end = start + nr;
1417
1418 if (nv50->global_residents.size < (end * sizeof(struct pipe_resource *))) {
1419 const unsigned old_size = nv50->global_residents.size;
1420 if (util_dynarray_resize(&nv50->global_residents, struct pipe_resource *, end)) {
1421 memset((uint8_t *)nv50->global_residents.data + old_size, 0,
1422 nv50->global_residents.size - old_size);
1423 } else {
1424 NOUVEAU_ERR("Could not resize global residents array\n");
1425 return;
1426 }
1427 }
1428
1429 if (resources) {
1430 ptr = util_dynarray_element(
1431 &nv50->global_residents, struct pipe_resource *, start);
1432 for (i = 0; i < nr; ++i) {
1433 pipe_resource_reference(&ptr[i], resources[i]);
1434 nv50_set_global_handle(handles[i], resources[i]);
1435 }
1436 } else {
1437 ptr = util_dynarray_element(
1438 &nv50->global_residents, struct pipe_resource *, start);
1439 for (i = 0; i < nr; ++i)
1440 pipe_resource_reference(&ptr[i], NULL);
1441 }
1442
1443 nouveau_bufctx_reset(nv50->bufctx_cp, NV50_BIND_CP_GLOBAL);
1444
1445 nv50->dirty_cp |= NV50_NEW_CP_GLOBALS;
1446 }
1447
1448 void
1449 nv50_init_state_functions(struct nv50_context *nv50)
1450 {
1451 struct pipe_context *pipe = &nv50->base.pipe;
1452
1453 pipe->create_blend_state = nv50_blend_state_create;
1454 pipe->bind_blend_state = nv50_blend_state_bind;
1455 pipe->delete_blend_state = nv50_blend_state_delete;
1456
1457 pipe->create_rasterizer_state = nv50_rasterizer_state_create;
1458 pipe->bind_rasterizer_state = nv50_rasterizer_state_bind;
1459 pipe->delete_rasterizer_state = nv50_rasterizer_state_delete;
1460
1461 pipe->create_depth_stencil_alpha_state = nv50_zsa_state_create;
1462 pipe->bind_depth_stencil_alpha_state = nv50_zsa_state_bind;
1463 pipe->delete_depth_stencil_alpha_state = nv50_zsa_state_delete;
1464
1465 pipe->create_sampler_state = nv50_sampler_state_create;
1466 pipe->delete_sampler_state = nv50_sampler_state_delete;
1467 pipe->bind_sampler_states = nv50_bind_sampler_states;
1468
1469 pipe->create_sampler_view = nv50_create_sampler_view;
1470 pipe->sampler_view_destroy = nv50_sampler_view_destroy;
1471 pipe->sampler_view_release = u_default_sampler_view_release;
1472 pipe->resource_release = u_default_resource_release;
1473 pipe->set_sampler_views = nv50_set_sampler_views;
1474
1475 pipe->create_vs_state = nv50_vp_state_create;
1476 pipe->create_fs_state = nv50_fp_state_create;
1477 pipe->create_gs_state = nv50_gp_state_create;
1478 pipe->create_compute_state = nv50_cp_state_create;
1479 pipe->bind_vs_state = nv50_vp_state_bind;
1480 pipe->bind_fs_state = nv50_fp_state_bind;
1481 pipe->bind_gs_state = nv50_gp_state_bind;
1482 pipe->bind_compute_state = nv50_cp_state_bind;
1483 pipe->delete_vs_state = nv50_sp_state_delete;
1484 pipe->delete_fs_state = nv50_sp_state_delete;
1485 pipe->delete_gs_state = nv50_sp_state_delete;
1486 pipe->delete_compute_state = nv50_sp_state_delete;
1487
1488 pipe->get_compute_state_info = nv50_get_compute_state_info;
1489
1490 pipe->set_blend_color = nv50_set_blend_color;
1491 pipe->set_stencil_ref = nv50_set_stencil_ref;
1492 pipe->set_clip_state = nv50_set_clip_state;
1493 pipe->set_sample_mask = nv50_set_sample_mask;
1494 pipe->set_min_samples = nv50_set_min_samples;
1495 pipe->set_constant_buffer = nv50_set_constant_buffer;
1496 pipe->set_framebuffer_state = nv50_set_framebuffer_state;
1497 pipe->set_polygon_stipple = nv50_set_polygon_stipple;
1498 pipe->set_scissor_states = nv50_set_scissor_states;
1499 pipe->set_viewport_states = nv50_set_viewport_states;
1500 pipe->set_window_rectangles = nv50_set_window_rectangles;
1501
1502 pipe->create_vertex_elements_state = nv50_vertex_state_create;
1503 pipe->delete_vertex_elements_state = nv50_vertex_state_delete;
1504 pipe->bind_vertex_elements_state = nv50_vertex_state_bind;
1505
1506 pipe->set_vertex_buffers = nv50_set_vertex_buffers;
1507
1508 pipe->create_stream_output_target = nv50_so_target_create;
1509 pipe->stream_output_target_destroy = nv50_so_target_destroy;
1510 pipe->set_stream_output_targets = nv50_set_stream_output_targets;
1511
1512 pipe->set_global_binding = nv50_set_global_bindings;
1513 pipe->set_shader_images = nv50_set_shader_images;
1514 pipe->set_shader_buffers = nv50_set_shader_buffers;
1515
1516 nv50->sample_mask = ~0;
1517 nv50->min_samples = 1;
1518 }