projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
v3d: Set the SO offsets correctly if we have to re-emit.
[mesa.git] / src / gallium / drivers / v3d / v3dx_emit.c
1 /*
2 * Copyright © 2014-2017 Broadcom
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 (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "util/u_format.h"
25 #include "util/u_half.h"
26 #include "v3d_context.h"
27 #include "broadcom/common/v3d_macros.h"
28 #include "broadcom/cle/v3dx_pack.h"
29 #include "broadcom/compiler/v3d_compiler.h"
30
31 static uint8_t
32 v3d_factor(enum pipe_blendfactor factor, bool dst_alpha_one)
33 {
34 /* We may get a bad blendfactor when blending is disabled. */
35 if (factor == 0)
36 return V3D_BLEND_FACTOR_ZERO;
37
38 switch (factor) {
39 case PIPE_BLENDFACTOR_ZERO:
40 return V3D_BLEND_FACTOR_ZERO;
41 case PIPE_BLENDFACTOR_ONE:
42 return V3D_BLEND_FACTOR_ONE;
43 case PIPE_BLENDFACTOR_SRC_COLOR:
44 return V3D_BLEND_FACTOR_SRC_COLOR;
45 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
46 return V3D_BLEND_FACTOR_INV_SRC_COLOR;
47 case PIPE_BLENDFACTOR_DST_COLOR:
48 return V3D_BLEND_FACTOR_DST_COLOR;
49 case PIPE_BLENDFACTOR_INV_DST_COLOR:
50 return V3D_BLEND_FACTOR_INV_DST_COLOR;
51 case PIPE_BLENDFACTOR_SRC_ALPHA:
52 return V3D_BLEND_FACTOR_SRC_ALPHA;
53 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
54 return V3D_BLEND_FACTOR_INV_SRC_ALPHA;
55 case PIPE_BLENDFACTOR_DST_ALPHA:
56 return (dst_alpha_one ?
57 V3D_BLEND_FACTOR_ONE :
58 V3D_BLEND_FACTOR_DST_ALPHA);
59 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
60 return (dst_alpha_one ?
61 V3D_BLEND_FACTOR_ZERO :
62 V3D_BLEND_FACTOR_INV_DST_ALPHA);
63 case PIPE_BLENDFACTOR_CONST_COLOR:
64 return V3D_BLEND_FACTOR_CONST_COLOR;
65 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
66 return V3D_BLEND_FACTOR_INV_CONST_COLOR;
67 case PIPE_BLENDFACTOR_CONST_ALPHA:
68 return V3D_BLEND_FACTOR_CONST_ALPHA;
69 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
70 return V3D_BLEND_FACTOR_INV_CONST_ALPHA;
71 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
72 return V3D_BLEND_FACTOR_SRC_ALPHA_SATURATE;
73 default:
74 unreachable("Bad blend factor");
75 }
76 }
77
78 static inline uint16_t
79 swizzled_border_color(const struct v3d_device_info *devinfo,
80 struct pipe_sampler_state *sampler,
81 struct v3d_sampler_view *sview,
82 int chan)
83 {
84 const struct util_format_description *desc =
85 util_format_description(sview->base.format);
86 uint8_t swiz = chan;
87
88 /* If we're doing swizzling in the sampler, then only rearrange the
89 * border color for the mismatch between the VC5 texture format and
90 * the PIPE_FORMAT, since GL_ARB_texture_swizzle will be handled by
91 * the sampler's swizzle.
92 *
93 * For swizzling in the shader, we don't do any pre-swizzling of the
94 * border color.
95 */
96 if (v3d_get_tex_return_size(devinfo, sview->base.format,
97 sampler->compare_mode) != 32)
98 swiz = desc->swizzle[swiz];
99
100 switch (swiz) {
101 case PIPE_SWIZZLE_0:
102 return util_float_to_half(0.0);
103 case PIPE_SWIZZLE_1:
104 return util_float_to_half(1.0);
105 default:
106 return util_float_to_half(sampler->border_color.f[swiz]);
107 }
108 }
109
110 #if V3D_VERSION < 40
111 static uint32_t
112 translate_swizzle(unsigned char pipe_swizzle)
113 {
114 switch (pipe_swizzle) {
115 case PIPE_SWIZZLE_0:
116 return 0;
117 case PIPE_SWIZZLE_1:
118 return 1;
119 case PIPE_SWIZZLE_X:
120 case PIPE_SWIZZLE_Y:
121 case PIPE_SWIZZLE_Z:
122 case PIPE_SWIZZLE_W:
123 return 2 + pipe_swizzle;
124 default:
125 unreachable("unknown swizzle");
126 }
127 }
128
129 static void
130 emit_one_texture(struct v3d_context *v3d, struct v3d_texture_stateobj *stage_tex,
131 int i)
132 {
133 struct v3d_job *job = v3d->job;
134 struct pipe_sampler_state *psampler = stage_tex->samplers[i];
135 struct v3d_sampler_state *sampler = v3d_sampler_state(psampler);
136 struct pipe_sampler_view *psview = stage_tex->textures[i];
137 struct v3d_sampler_view *sview = v3d_sampler_view(psview);
138 struct pipe_resource *prsc = psview->texture;
139 struct v3d_resource *rsc = v3d_resource(prsc);
140 const struct v3d_device_info *devinfo = &v3d->screen->devinfo;
141
142 stage_tex->texture_state[i].offset =
143 v3d_cl_ensure_space(&job->indirect,
144 cl_packet_length(TEXTURE_SHADER_STATE),
145 32);
146 v3d_bo_set_reference(&stage_tex->texture_state[i].bo,
147 job->indirect.bo);
148
149 uint32_t return_size = v3d_get_tex_return_size(devinfo, psview->format,
150 psampler->compare_mode);
151
152 struct V3D33_TEXTURE_SHADER_STATE unpacked = {
153 /* XXX */
154 .border_color_red = swizzled_border_color(devinfo, psampler,
155 sview, 0),
156 .border_color_green = swizzled_border_color(devinfo, psampler,
157 sview, 1),
158 .border_color_blue = swizzled_border_color(devinfo, psampler,
159 sview, 2),
160 .border_color_alpha = swizzled_border_color(devinfo, psampler,
161 sview, 3),
162
163 /* In the normal texturing path, the LOD gets clamped between
164 * min/max, and the base_level field (set in the sampler view
165 * from first_level) only decides where the min/mag switch
166 * happens, so we need to use the LOD clamps to keep us
167 * between min and max.
168 *
169 * For txf, the LOD clamp is still used, despite GL not
170 * wanting that. We will need to have a separate
171 * TEXTURE_SHADER_STATE that ignores psview->min/max_lod to
172 * support txf properly.
173 */
174 .min_level_of_detail = MIN2(psview->u.tex.first_level +
175 MAX2(psampler->min_lod, 0),
176 psview->u.tex.last_level),
177 .max_level_of_detail = MIN2(psview->u.tex.first_level +
178 psampler->max_lod,
179 psview->u.tex.last_level),
180
181 .texture_base_pointer = cl_address(rsc->bo,
182 rsc->slices[0].offset),
183
184 .output_32_bit = return_size == 32,
185 };
186
187 /* Set up the sampler swizzle if we're doing 16-bit sampling. For
188 * 32-bit, we leave swizzling up to the shader compiler.
189 *
190 * Note: Contrary to the docs, the swizzle still applies even if the
191 * return size is 32. It's just that you probably want to swizzle in
192 * the shader, because you need the Y/Z/W channels to be defined.
193 */
194 if (return_size == 32) {
195 unpacked.swizzle_r = translate_swizzle(PIPE_SWIZZLE_X);
196 unpacked.swizzle_g = translate_swizzle(PIPE_SWIZZLE_Y);
197 unpacked.swizzle_b = translate_swizzle(PIPE_SWIZZLE_Z);
198 unpacked.swizzle_a = translate_swizzle(PIPE_SWIZZLE_W);
199 } else {
200 unpacked.swizzle_r = translate_swizzle(sview->swizzle[0]);
201 unpacked.swizzle_g = translate_swizzle(sview->swizzle[1]);
202 unpacked.swizzle_b = translate_swizzle(sview->swizzle[2]);
203 unpacked.swizzle_a = translate_swizzle(sview->swizzle[3]);
204 }
205
206 int min_img_filter = psampler->min_img_filter;
207 int min_mip_filter = psampler->min_mip_filter;
208 int mag_img_filter = psampler->mag_img_filter;
209
210 if (return_size == 32) {
211 min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
212 min_img_filter = PIPE_TEX_FILTER_NEAREST;
213 mag_img_filter = PIPE_TEX_FILTER_NEAREST;
214 }
215
216 bool min_nearest = min_img_filter == PIPE_TEX_FILTER_NEAREST;
217 switch (min_mip_filter) {
218 case PIPE_TEX_MIPFILTER_NONE:
219 unpacked.filter += min_nearest ? 2 : 0;
220 break;
221 case PIPE_TEX_MIPFILTER_NEAREST:
222 unpacked.filter += min_nearest ? 4 : 8;
223 break;
224 case PIPE_TEX_MIPFILTER_LINEAR:
225 unpacked.filter += min_nearest ? 4 : 8;
226 unpacked.filter += 2;
227 break;
228 }
229
230 if (mag_img_filter == PIPE_TEX_FILTER_NEAREST)
231 unpacked.filter++;
232
233 if (psampler->max_anisotropy > 8)
234 unpacked.filter = V3D_TMU_FILTER_ANISOTROPIC_16_1;
235 else if (psampler->max_anisotropy > 4)
236 unpacked.filter = V3D_TMU_FILTER_ANISOTROPIC_8_1;
237 else if (psampler->max_anisotropy > 2)
238 unpacked.filter = V3D_TMU_FILTER_ANISOTROPIC_4_1;
239 else if (psampler->max_anisotropy)
240 unpacked.filter = V3D_TMU_FILTER_ANISOTROPIC_2_1;
241
242 uint8_t packed[cl_packet_length(TEXTURE_SHADER_STATE)];
243 cl_packet_pack(TEXTURE_SHADER_STATE)(&job->indirect, packed, &unpacked);
244
245 for (int i = 0; i < ARRAY_SIZE(packed); i++)
246 packed[i] |= sview->texture_shader_state[i] | sampler->texture_shader_state[i];
247
248 /* TMU indirect structs need to be 32b aligned. */
249 v3d_cl_ensure_space(&job->indirect, ARRAY_SIZE(packed), 32);
250 cl_emit_prepacked(&job->indirect, &packed);
251 }
252
253 static void
254 emit_textures(struct v3d_context *v3d, struct v3d_texture_stateobj *stage_tex)
255 {
256 for (int i = 0; i < stage_tex->num_textures; i++) {
257 if (stage_tex->textures[i])
258 emit_one_texture(v3d, stage_tex, i);
259 }
260 }
261 #endif /* V3D_VERSION < 40 */
262
263 static uint32_t
264 translate_colormask(struct v3d_context *v3d, uint32_t colormask, int rt)
265 {
266 if (v3d->swap_color_rb & (1 << rt)) {
267 colormask = ((colormask & (2 | 8)) |
268 ((colormask & 1) << 2) |
269 ((colormask & 4) >> 2));
270 }
271
272 return (~colormask) & 0xf;
273 }
274
275 static void
276 emit_rt_blend(struct v3d_context *v3d, struct v3d_job *job,
277 struct pipe_blend_state *blend, int rt)
278 {
279 cl_emit(&job->bcl, BLEND_CONFIG, config) {
280 struct pipe_rt_blend_state *rtblend = &blend->rt[rt];
281
282 #if V3D_VERSION >= 40
283 config.render_target_mask = 1 << rt;
284 #else
285 assert(rt == 0);
286 #endif
287
288 config.colour_blend_mode = rtblend->rgb_func;
289 config.colour_blend_dst_factor =
290 v3d_factor(rtblend->rgb_dst_factor,
291 v3d->blend_dst_alpha_one);
292 config.colour_blend_src_factor =
293 v3d_factor(rtblend->rgb_src_factor,
294 v3d->blend_dst_alpha_one);
295
296 config.alpha_blend_mode = rtblend->alpha_func;
297 config.alpha_blend_dst_factor =
298 v3d_factor(rtblend->alpha_dst_factor,
299 v3d->blend_dst_alpha_one);
300 config.alpha_blend_src_factor =
301 v3d_factor(rtblend->alpha_src_factor,
302 v3d->blend_dst_alpha_one);
303 }
304 }
305
306 void
307 v3dX(emit_state)(struct pipe_context *pctx)
308 {
309 struct v3d_context *v3d = v3d_context(pctx);
310 struct v3d_job *job = v3d->job;
311 bool rasterizer_discard = v3d->rasterizer->base.rasterizer_discard;
312
313 if (v3d->dirty & (VC5_DIRTY_SCISSOR | VC5_DIRTY_VIEWPORT |
314 VC5_DIRTY_RASTERIZER)) {
315 float *vpscale = v3d->viewport.scale;
316 float *vptranslate = v3d->viewport.translate;
317 float vp_minx = -fabsf(vpscale[0]) + vptranslate[0];
318 float vp_maxx = fabsf(vpscale[0]) + vptranslate[0];
319 float vp_miny = -fabsf(vpscale[1]) + vptranslate[1];
320 float vp_maxy = fabsf(vpscale[1]) + vptranslate[1];
321
322 /* Clip to the scissor if it's enabled, but still clip to the
323 * drawable regardless since that controls where the binner
324 * tries to put things.
325 *
326 * Additionally, always clip the rendering to the viewport,
327 * since the hardware does guardband clipping, meaning
328 * primitives would rasterize outside of the view volume.
329 */
330 uint32_t minx, miny, maxx, maxy;
331 if (!v3d->rasterizer->base.scissor) {
332 minx = MAX2(vp_minx, 0);
333 miny = MAX2(vp_miny, 0);
334 maxx = MIN2(vp_maxx, job->draw_width);
335 maxy = MIN2(vp_maxy, job->draw_height);
336 } else {
337 minx = MAX2(vp_minx, v3d->scissor.minx);
338 miny = MAX2(vp_miny, v3d->scissor.miny);
339 maxx = MIN2(vp_maxx, v3d->scissor.maxx);
340 maxy = MIN2(vp_maxy, v3d->scissor.maxy);
341 }
342
343 cl_emit(&job->bcl, CLIP_WINDOW, clip) {
344 clip.clip_window_left_pixel_coordinate = minx;
345 clip.clip_window_bottom_pixel_coordinate = miny;
346 if (maxx > minx && maxy > miny) {
347 clip.clip_window_width_in_pixels = maxx - minx;
348 clip.clip_window_height_in_pixels = maxy - miny;
349 } else if (V3D_VERSION < 41) {
350 /* The HW won't entirely clip out when scissor
351 * w/h is 0. Just treat it the same as
352 * rasterizer discard.
353 */
354 rasterizer_discard = true;
355 clip.clip_window_width_in_pixels = 1;
356 clip.clip_window_height_in_pixels = 1;
357 }
358 }
359
360 job->draw_min_x = MIN2(job->draw_min_x, minx);
361 job->draw_min_y = MIN2(job->draw_min_y, miny);
362 job->draw_max_x = MAX2(job->draw_max_x, maxx);
363 job->draw_max_y = MAX2(job->draw_max_y, maxy);
364 }
365
366 if (v3d->dirty & (VC5_DIRTY_RASTERIZER |
367 VC5_DIRTY_ZSA |
368 VC5_DIRTY_BLEND |
369 VC5_DIRTY_COMPILED_FS)) {
370 cl_emit(&job->bcl, CONFIGURATION_BITS, config) {
371 config.enable_forward_facing_primitive =
372 !rasterizer_discard &&
373 !(v3d->rasterizer->base.cull_face &
374 PIPE_FACE_FRONT);
375 config.enable_reverse_facing_primitive =
376 !rasterizer_discard &&
377 !(v3d->rasterizer->base.cull_face &
378 PIPE_FACE_BACK);
379 /* This seems backwards, but it's what gets the
380 * clipflat test to pass.
381 */
382 config.clockwise_primitives =
383 v3d->rasterizer->base.front_ccw;
384
385 config.enable_depth_offset =
386 v3d->rasterizer->base.offset_tri;
387
388 /* V3D follows GL behavior where the sample mask only
389 * applies when MSAA is enabled. Gallium has sample
390 * mask apply anyway, and the MSAA blit shaders will
391 * set sample mask without explicitly setting
392 * rasterizer oversample. Just force it on here,
393 * since the blit shaders are the only way to have
394 * !multisample && samplemask != 0xf.
395 */
396 config.rasterizer_oversample_mode =
397 v3d->rasterizer->base.multisample ||
398 v3d->sample_mask != 0xf;
399
400 config.direct3d_provoking_vertex =
401 v3d->rasterizer->base.flatshade_first;
402
403 config.blend_enable = v3d->blend->rt[0].blend_enable;
404
405 /* Note: EZ state may update based on the compiled FS,
406 * along with ZSA
407 */
408 config.early_z_updates_enable =
409 (job->ez_state != VC5_EZ_DISABLED);
410 if (v3d->zsa->base.depth.enabled) {
411 config.z_updates_enable =
412 v3d->zsa->base.depth.writemask;
413 config.early_z_enable =
414 config.early_z_updates_enable;
415 config.depth_test_function =
416 v3d->zsa->base.depth.func;
417 } else {
418 config.depth_test_function = PIPE_FUNC_ALWAYS;
419 }
420
421 config.stencil_enable =
422 v3d->zsa->base.stencil[0].enabled;
423 }
424
425 }
426
427 if (v3d->dirty & VC5_DIRTY_RASTERIZER &&
428 v3d->rasterizer->base.offset_tri) {
429 cl_emit(&job->bcl, DEPTH_OFFSET, depth) {
430 depth.depth_offset_factor =
431 v3d->rasterizer->offset_factor;
432 if (job->zsbuf &&
433 job->zsbuf->format == PIPE_FORMAT_Z16_UNORM) {
434 depth.depth_offset_units =
435 v3d->rasterizer->z16_offset_units;
436 } else {
437 depth.depth_offset_units =
438 v3d->rasterizer->offset_units;
439 }
440 }
441 }
442
443 if (v3d->dirty & VC5_DIRTY_RASTERIZER) {
444 cl_emit(&job->bcl, POINT_SIZE, point_size) {
445 point_size.point_size = v3d->rasterizer->point_size;
446 }
447
448 cl_emit(&job->bcl, LINE_WIDTH, line_width) {
449 line_width.line_width = v3d->rasterizer->base.line_width;
450 }
451 }
452
453 if (v3d->dirty & VC5_DIRTY_VIEWPORT) {
454 cl_emit(&job->bcl, CLIPPER_XY_SCALING, clip) {
455 clip.viewport_half_width_in_1_256th_of_pixel =
456 v3d->viewport.scale[0] * 256.0f;
457 clip.viewport_half_height_in_1_256th_of_pixel =
458 v3d->viewport.scale[1] * 256.0f;
459 }
460
461 cl_emit(&job->bcl, CLIPPER_Z_SCALE_AND_OFFSET, clip) {
462 clip.viewport_z_offset_zc_to_zs =
463 v3d->viewport.translate[2];
464 clip.viewport_z_scale_zc_to_zs =
465 v3d->viewport.scale[2];
466 }
467 cl_emit(&job->bcl, CLIPPER_Z_MIN_MAX_CLIPPING_PLANES, clip) {
468 clip.minimum_zw = (v3d->viewport.translate[2] -
469 v3d->viewport.scale[2]);
470 clip.maximum_zw = (v3d->viewport.translate[2] +
471 v3d->viewport.scale[2]);
472 }
473
474 cl_emit(&job->bcl, VIEWPORT_OFFSET, vp) {
475 vp.viewport_centre_x_coordinate =
476 v3d->viewport.translate[0];
477 vp.viewport_centre_y_coordinate =
478 v3d->viewport.translate[1];
479 }
480 }
481
482 if (v3d->dirty & VC5_DIRTY_BLEND && v3d->blend->rt[0].blend_enable) {
483 struct pipe_blend_state *blend = v3d->blend;
484
485 if (blend->independent_blend_enable) {
486 for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++)
487 emit_rt_blend(v3d, job, blend, i);
488 } else {
489 emit_rt_blend(v3d, job, blend, 0);
490 }
491 }
492
493 if (v3d->dirty & VC5_DIRTY_BLEND) {
494 struct pipe_blend_state *blend = v3d->blend;
495
496 cl_emit(&job->bcl, COLOUR_WRITE_MASKS, mask) {
497 if (blend->independent_blend_enable) {
498 mask.render_target_0_per_colour_component_write_masks =
499 translate_colormask(v3d, blend->rt[0].colormask, 0);
500 mask.render_target_1_per_colour_component_write_masks =
501 translate_colormask(v3d, blend->rt[1].colormask, 1);
502 mask.render_target_2_per_colour_component_write_masks =
503 translate_colormask(v3d, blend->rt[2].colormask, 2);
504 mask.render_target_3_per_colour_component_write_masks =
505 translate_colormask(v3d, blend->rt[3].colormask, 3);
506 } else {
507 mask.render_target_0_per_colour_component_write_masks =
508 translate_colormask(v3d, blend->rt[0].colormask, 0);
509 mask.render_target_1_per_colour_component_write_masks =
510 translate_colormask(v3d, blend->rt[0].colormask, 1);
511 mask.render_target_2_per_colour_component_write_masks =
512 translate_colormask(v3d, blend->rt[0].colormask, 2);
513 mask.render_target_3_per_colour_component_write_masks =
514 translate_colormask(v3d, blend->rt[0].colormask, 3);
515 }
516 }
517 }
518
519 /* GFXH-1431: On V3D 3.x, writing BLEND_CONFIG resets the constant
520 * color.
521 */
522 if (v3d->dirty & VC5_DIRTY_BLEND_COLOR ||
523 (V3D_VERSION < 41 && (v3d->dirty & VC5_DIRTY_BLEND))) {
524 cl_emit(&job->bcl, BLEND_CONSTANT_COLOUR, colour) {
525 colour.red_f16 = (v3d->swap_color_rb ?
526 v3d->blend_color.hf[2] :
527 v3d->blend_color.hf[0]);
528 colour.green_f16 = v3d->blend_color.hf[1];
529 colour.blue_f16 = (v3d->swap_color_rb ?
530 v3d->blend_color.hf[0] :
531 v3d->blend_color.hf[2]);
532 colour.alpha_f16 = v3d->blend_color.hf[3];
533 }
534 }
535
536 if (v3d->dirty & (VC5_DIRTY_ZSA | VC5_DIRTY_STENCIL_REF)) {
537 struct pipe_stencil_state *front = &v3d->zsa->base.stencil[0];
538 struct pipe_stencil_state *back = &v3d->zsa->base.stencil[1];
539
540 if (front->enabled) {
541 cl_emit_with_prepacked(&job->bcl, STENCIL_CONFIG,
542 v3d->zsa->stencil_front, config) {
543 config.stencil_ref_value =
544 v3d->stencil_ref.ref_value[0];
545 }
546 }
547
548 if (back->enabled) {
549 cl_emit_with_prepacked(&job->bcl, STENCIL_CONFIG,
550 v3d->zsa->stencil_back, config) {
551 config.stencil_ref_value =
552 v3d->stencil_ref.ref_value[1];
553 }
554 }
555 }
556
557 #if V3D_VERSION < 40
558 /* Pre-4.x, we have texture state that depends on both the sampler and
559 * the view, so we merge them together at draw time.
560 */
561 if (v3d->dirty & VC5_DIRTY_FRAGTEX)
562 emit_textures(v3d, &v3d->fragtex);
563
564 if (v3d->dirty & VC5_DIRTY_VERTTEX)
565 emit_textures(v3d, &v3d->verttex);
566 #endif
567
568 if (v3d->dirty & VC5_DIRTY_FLAT_SHADE_FLAGS) {
569 bool emitted_any = false;
570
571 for (int i = 0; i < ARRAY_SIZE(v3d->prog.fs->prog_data.fs->flat_shade_flags); i++) {
572 if (!v3d->prog.fs->prog_data.fs->flat_shade_flags[i])
573 continue;
574
575 cl_emit(&job->bcl, FLAT_SHADE_FLAGS, flags) {
576 flags.varying_offset_v0 = i;
577
578 if (emitted_any) {
579 flags.action_for_flat_shade_flags_of_lower_numbered_varyings =
580 V3D_VARYING_FLAGS_ACTION_UNCHANGED;
581 flags.action_for_flat_shade_flags_of_higher_numbered_varyings =
582 V3D_VARYING_FLAGS_ACTION_UNCHANGED;
583 } else {
584 flags.action_for_flat_shade_flags_of_lower_numbered_varyings =
585 ((i == 0) ?
586 V3D_VARYING_FLAGS_ACTION_UNCHANGED :
587 V3D_VARYING_FLAGS_ACTION_ZEROED);
588
589 flags.action_for_flat_shade_flags_of_higher_numbered_varyings =
590 V3D_VARYING_FLAGS_ACTION_ZEROED;
591 }
592
593 flags.flat_shade_flags_for_varyings_v024 =
594 v3d->prog.fs->prog_data.fs->flat_shade_flags[i];
595 }
596
597 emitted_any = true;
598 }
599
600 if (!emitted_any) {
601 cl_emit(&job->bcl, ZERO_ALL_FLAT_SHADE_FLAGS, flags);
602 }
603 }
604
605 #if V3D_VERSION >= 40
606 if (v3d->dirty & VC5_DIRTY_CENTROID_FLAGS) {
607 bool emitted_any = false;
608
609 for (int i = 0; i < ARRAY_SIZE(v3d->prog.fs->prog_data.fs->centroid_flags); i++) {
610 if (!v3d->prog.fs->prog_data.fs->centroid_flags[i])
611 continue;
612
613 cl_emit(&job->bcl, CENTROID_FLAGS, flags) {
614 flags.varying_offset_v0 = i;
615
616 if (emitted_any) {
617 flags.action_for_centroid_flags_of_lower_numbered_varyings =
618 V3D_VARYING_FLAGS_ACTION_UNCHANGED;
619 flags.action_for_centroid_flags_of_higher_numbered_varyings =
620 V3D_VARYING_FLAGS_ACTION_UNCHANGED;
621 } else {
622 flags.action_for_centroid_flags_of_lower_numbered_varyings =
623 ((i == 0) ?
624 V3D_VARYING_FLAGS_ACTION_UNCHANGED :
625 V3D_VARYING_FLAGS_ACTION_ZEROED);
626
627 flags.action_for_centroid_flags_of_higher_numbered_varyings =
628 V3D_VARYING_FLAGS_ACTION_ZEROED;
629 }
630
631 flags.centroid_flags_for_varyings_v024 =
632 v3d->prog.fs->prog_data.fs->centroid_flags[i];
633 }
634
635 emitted_any = true;
636 }
637
638 if (!emitted_any) {
639 cl_emit(&job->bcl, ZERO_ALL_CENTROID_FLAGS, flags);
640 }
641 }
642 #endif
643
644 /* Set up the transform feedback data specs (which VPM entries to
645 * output to which buffers).
646 */
647 if (v3d->dirty & (VC5_DIRTY_STREAMOUT |
648 VC5_DIRTY_RASTERIZER |
649 VC5_DIRTY_PRIM_MODE)) {
650 struct v3d_streamout_stateobj *so = &v3d->streamout;
651
652 if (so->num_targets) {
653 bool psiz_per_vertex = (v3d->prim_mode == PIPE_PRIM_POINTS &&
654 v3d->rasterizer->base.point_size_per_vertex);
655 uint16_t *tf_specs = (psiz_per_vertex ?
656 v3d->prog.bind_vs->tf_specs_psiz :
657 v3d->prog.bind_vs->tf_specs);
658
659 #if V3D_VERSION >= 40
660 job->tf_enabled = (v3d->prog.bind_vs->num_tf_specs != 0 &&
661 v3d->active_queries);
662
663 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_SPECS, tfe) {
664 tfe.number_of_16_bit_output_data_specs_following =
665 v3d->prog.bind_vs->num_tf_specs;
666 tfe.enable = job->tf_enabled;
667 };
668 #else /* V3D_VERSION < 40 */
669 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_ENABLE, tfe) {
670 tfe.number_of_32_bit_output_buffer_address_following =
671 so->num_targets;
672 tfe.number_of_16_bit_output_data_specs_following =
673 v3d->prog.bind_vs->num_tf_specs;
674 };
675 #endif /* V3D_VERSION < 40 */
676 for (int i = 0; i < v3d->prog.bind_vs->num_tf_specs; i++) {
677 cl_emit_prepacked(&job->bcl, &tf_specs[i]);
678 }
679 } else if (job->tf_enabled) {
680 #if V3D_VERSION >= 40
681 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_SPECS, tfe) {
682 tfe.enable = false;
683 };
684 job->tf_enabled = false;
685 #endif /* V3D_VERSION >= 40 */
686 }
687 }
688
689 /* Set up the trasnform feedback buffers. */
690 if (v3d->dirty & VC5_DIRTY_STREAMOUT) {
691 struct v3d_streamout_stateobj *so = &v3d->streamout;
692 for (int i = 0; i < so->num_targets; i++) {
693 const struct pipe_stream_output_target *target =
694 so->targets[i];
695 struct v3d_resource *rsc = target ?
696 v3d_resource(target->buffer) : NULL;
697 struct pipe_shader_state *vs = &v3d->prog.bind_vs->base;
698 struct pipe_stream_output_info *info = &vs->stream_output;
699 uint32_t offset = (v3d->streamout.offsets[i] *
700 info->stride[i] * 4);
701
702 #if V3D_VERSION >= 40
703 if (!target)
704 continue;
705
706 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_BUFFER, output) {
707 output.buffer_address =
708 cl_address(rsc->bo,
709 target->buffer_offset +
710 offset);
711 output.buffer_size_in_32_bit_words =
712 (target->buffer_size - offset) >> 2;
713 output.buffer_number = i;
714 }
715 #else /* V3D_VERSION < 40 */
716 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_OUTPUT_ADDRESS, output) {
717 if (target) {
718 output.address =
719 cl_address(rsc->bo,
720 target->buffer_offset +
721 offset);
722 }
723 };
724 #endif /* V3D_VERSION < 40 */
725 if (target) {
726 v3d_job_add_write_resource(v3d->job,
727 target->buffer);
728 }
729 /* XXX: buffer_size? */
730 }
731 }
732
733 if (v3d->dirty & VC5_DIRTY_OQ) {
734 cl_emit(&job->bcl, OCCLUSION_QUERY_COUNTER, counter) {
735 job->oq_enabled = v3d->active_queries && v3d->current_oq;
736 if (job->oq_enabled) {
737 counter.address = cl_address(v3d->current_oq, 0);
738 }
739 }
740 }
741
742 #if V3D_VERSION >= 40
743 if (v3d->dirty & VC5_DIRTY_SAMPLE_STATE) {
744 cl_emit(&job->bcl, SAMPLE_STATE, state) {
745 /* Note: SampleCoverage was handled at the
746 * state_tracker level by converting to sample_mask.
747 */
748 state.coverage = fui(1.0) >> 16;
749 state.mask = job->msaa ? v3d->sample_mask : 0xf;
750 }
751 }
752 #endif
753 }