projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
lima: delay plbu head command generation to flush stage (v2)
[mesa.git] / src / gallium / drivers / lima / lima_draw.c
1 /*
2 * Copyright (c) 2011-2013 Luc Verhaegen <libv@skynet.be>
3 * Copyright (c) 2017-2019 Lima Project
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sub license,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the
13 * next paragraph) shall be included in all copies or substantial portions
14 * of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 *
24 */
25
26 #include "util/u_math.h"
27 #include "util/format/u_format.h"
28 #include "util/u_debug.h"
29 #include "util/u_half.h"
30 #include "util/u_helpers.h"
31 #include "util/u_inlines.h"
32 #include "util/u_pack_color.h"
33 #include "util/hash_table.h"
34 #include "util/u_split_draw.h"
35 #include "util/u_upload_mgr.h"
36 #include "util/u_prim.h"
37 #include "util/u_vbuf.h"
38
39 #include "lima_context.h"
40 #include "lima_screen.h"
41 #include "lima_resource.h"
42 #include "lima_program.h"
43 #include "lima_bo.h"
44 #include "lima_submit.h"
45 #include "lima_texture.h"
46 #include "lima_util.h"
47 #include "lima_fence.h"
48 #include "lima_format.h"
49
50 #include <drm-uapi/lima_drm.h>
51
52 struct lima_gp_frame_reg {
53 uint32_t vs_cmd_start;
54 uint32_t vs_cmd_end;
55 uint32_t plbu_cmd_start;
56 uint32_t plbu_cmd_end;
57 uint32_t tile_heap_start;
58 uint32_t tile_heap_end;
59 };
60
61 struct lima_pp_frame_reg {
62 uint32_t plbu_array_address;
63 uint32_t render_address;
64 uint32_t unused_0;
65 uint32_t flags;
66 uint32_t clear_value_depth;
67 uint32_t clear_value_stencil;
68 uint32_t clear_value_color;
69 uint32_t clear_value_color_1;
70 uint32_t clear_value_color_2;
71 uint32_t clear_value_color_3;
72 uint32_t width;
73 uint32_t height;
74 uint32_t fragment_stack_address;
75 uint32_t fragment_stack_size;
76 uint32_t unused_1;
77 uint32_t unused_2;
78 uint32_t one;
79 uint32_t supersampled_height;
80 uint32_t dubya;
81 uint32_t onscreen;
82 uint32_t blocking;
83 uint32_t scale;
84 uint32_t foureight;
85 };
86
87 struct lima_pp_wb_reg {
88 uint32_t type;
89 uint32_t address;
90 uint32_t pixel_format;
91 uint32_t downsample_factor;
92 uint32_t pixel_layout;
93 uint32_t pitch;
94 uint32_t mrt_bits;
95 uint32_t mrt_pitch;
96 uint32_t zero;
97 uint32_t unused0;
98 uint32_t unused1;
99 uint32_t unused2;
100 };
101
102 struct lima_render_state {
103 uint32_t blend_color_bg;
104 uint32_t blend_color_ra;
105 uint32_t alpha_blend;
106 uint32_t depth_test;
107 uint32_t depth_range;
108 uint32_t stencil_front;
109 uint32_t stencil_back;
110 uint32_t stencil_test;
111 uint32_t multi_sample;
112 uint32_t shader_address;
113 uint32_t varying_types;
114 uint32_t uniforms_address;
115 uint32_t textures_address;
116 uint32_t aux0;
117 uint32_t aux1;
118 uint32_t varyings_address;
119 };
120
121
122 /* plbu commands */
123 #define PLBU_CMD_BEGIN(array, max) { \
124 int i = 0, max_n = max; \
125 struct util_dynarray *plbu_cmd_array = array; \
126 uint32_t *plbu_cmd = util_dynarray_ensure_cap(plbu_cmd_array, plbu_cmd_array->size + max_n * 4);
127
128 #define PLBU_CMD_END() \
129 assert(i <= max_n); \
130 plbu_cmd_array->size += i * 4; \
131 }
132
133 #define PLBU_CMD(v1, v2) \
134 do { \
135 plbu_cmd[i++] = v1; \
136 plbu_cmd[i++] = v2; \
137 } while (0)
138
139 #define PLBU_CMD_BLOCK_STEP(shift_min, shift_h, shift_w) \
140 PLBU_CMD(((shift_min) << 28) | ((shift_h) << 16) | (shift_w), 0x1000010C)
141 #define PLBU_CMD_TILED_DIMENSIONS(tiled_w, tiled_h) \
142 PLBU_CMD((((tiled_w) - 1) << 24) | (((tiled_h) - 1) << 8), 0x10000109)
143 #define PLBU_CMD_BLOCK_STRIDE(block_w) PLBU_CMD((block_w) & 0xff, 0x30000000)
144 #define PLBU_CMD_ARRAY_ADDRESS(gp_stream, block_num) \
145 PLBU_CMD(gp_stream, 0x28000000 | ((block_num) - 1) | 1)
146 #define PLBU_CMD_VIEWPORT_LEFT(v) PLBU_CMD(v, 0x10000107)
147 #define PLBU_CMD_VIEWPORT_RIGHT(v) PLBU_CMD(v, 0x10000108)
148 #define PLBU_CMD_VIEWPORT_BOTTOM(v) PLBU_CMD(v, 0x10000105)
149 #define PLBU_CMD_VIEWPORT_TOP(v) PLBU_CMD(v, 0x10000106)
150 #define PLBU_CMD_ARRAYS_SEMAPHORE_BEGIN() PLBU_CMD(0x00010002, 0x60000000)
151 #define PLBU_CMD_ARRAYS_SEMAPHORE_END() PLBU_CMD(0x00010001, 0x60000000)
152 #define PLBU_CMD_PRIMITIVE_SETUP(force_point_size, cull, index_size) \
153 PLBU_CMD(0x2200 | ((force_point_size) ? 0x1000 : 0) | \
154 (cull) | ((index_size) << 9), 0x1000010B)
155 #define PLBU_CMD_RSW_VERTEX_ARRAY(rsw, gl_pos) \
156 PLBU_CMD(rsw, 0x80000000 | ((gl_pos) >> 4))
157 #define PLBU_CMD_SCISSORS(minx, maxx, miny, maxy) \
158 PLBU_CMD(((minx) << 30) | ((maxy) - 1) << 15 | (miny), \
159 0x70000000 | ((maxx) - 1) << 13 | ((minx) >> 2))
160 #define PLBU_CMD_UNKNOWN1() PLBU_CMD(0x00000000, 0x1000010A)
161 #define PLBU_CMD_UNKNOWN2() PLBU_CMD(0x00000200, 0x1000010B)
162 #define PLBU_CMD_LOW_PRIM_SIZE(v) PLBU_CMD(v, 0x1000010D)
163 #define PLBU_CMD_DEPTH_RANGE_NEAR(v) PLBU_CMD(v, 0x1000010E)
164 #define PLBU_CMD_DEPTH_RANGE_FAR(v) PLBU_CMD(v, 0x1000010F)
165 #define PLBU_CMD_INDEXED_DEST(gl_pos) PLBU_CMD(gl_pos, 0x10000100)
166 #define PLBU_CMD_INDEXED_PT_SIZE(pt_size) PLBU_CMD(pt_size, 0x10000102)
167 #define PLBU_CMD_INDICES(va) PLBU_CMD(va, 0x10000101)
168 #define PLBU_CMD_DRAW_ARRAYS(mode, start, count) \
169 PLBU_CMD(((count) << 24) | (start), (((mode) & 0x1F) << 16) | ((count) >> 8))
170 #define PLBU_CMD_DRAW_ELEMENTS(mode, start, count) \
171 PLBU_CMD(((count) << 24) | (start), \
172 0x00200000 | (((mode) & 0x1F) << 16) | ((count) >> 8))
173
174 /* vs commands */
175 #define VS_CMD_BEGIN(array, max) { \
176 int i = 0, max_n = max; \
177 struct util_dynarray *vs_cmd_array = array; \
178 uint32_t *vs_cmd = util_dynarray_ensure_cap(vs_cmd_array, vs_cmd_array->size + max_n * 4);
179
180 #define VS_CMD_END() \
181 assert(i <= max_n); \
182 vs_cmd_array->size += i * 4; \
183 }
184
185 #define VS_CMD(v1, v2) \
186 do { \
187 vs_cmd[i++] = v1; \
188 vs_cmd[i++] = v2; \
189 } while (0)
190
191 #define VS_CMD_ARRAYS_SEMAPHORE_BEGIN_1() VS_CMD(0x00028000, 0x50000000)
192 #define VS_CMD_ARRAYS_SEMAPHORE_BEGIN_2() VS_CMD(0x00000001, 0x50000000)
193 #define VS_CMD_ARRAYS_SEMAPHORE_END(index_draw) \
194 VS_CMD((index_draw) ? 0x00018000 : 0x00000000, 0x50000000)
195 #define VS_CMD_UNIFORMS_ADDRESS(addr, size) \
196 VS_CMD(addr, 0x30000000 | ((size) << 12))
197 #define VS_CMD_SHADER_ADDRESS(addr, size) \
198 VS_CMD(addr, 0x40000000 | ((size) << 12))
199 #define VS_CMD_SHADER_INFO(prefetch, size) \
200 VS_CMD(((prefetch) << 20) | ((((size) >> 4) - 1) << 10), 0x10000040)
201 #define VS_CMD_VARYING_ATTRIBUTE_COUNT(nv, na) \
202 VS_CMD((((nv) - 1) << 8) | (((na) - 1) << 24), 0x10000042)
203 #define VS_CMD_UNKNOWN1() VS_CMD(0x00000003, 0x10000041)
204 #define VS_CMD_UNKNOWN2() VS_CMD(0x00000000, 0x60000000)
205 #define VS_CMD_ATTRIBUTES_ADDRESS(addr, na) \
206 VS_CMD(addr, 0x20000000 | ((na) << 17))
207 #define VS_CMD_VARYINGS_ADDRESS(addr, nv) \
208 VS_CMD(addr, 0x20000008 | ((nv) << 17))
209 #define VS_CMD_DRAW(num, index_draw) \
210 VS_CMD(((num) << 24) | ((index_draw) ? 1 : 0), ((num) >> 8))
211
212 static inline bool
213 lima_ctx_dirty(struct lima_context *ctx)
214 {
215 return !!ctx->resolve;
216 }
217
218 static inline struct lima_damage_region *
219 lima_ctx_get_damage(struct lima_context *ctx)
220 {
221 if (!ctx->framebuffer.base.nr_cbufs)
222 return NULL;
223
224 struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
225 struct lima_resource *res = lima_resource(surf->base.texture);
226 return &res->damage;
227 }
228
229 static bool
230 lima_fb_need_reload(struct lima_context *ctx)
231 {
232 /* Depth buffer is always discarded */
233 if (!ctx->framebuffer.base.nr_cbufs)
234 return false;
235
236 struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
237 struct lima_resource *res = lima_resource(surf->base.texture);
238 if (res->damage.region) {
239 /* for EGL_KHR_partial_update, when EGL_EXT_buffer_age is enabled,
240 * we need to reload damage region, otherwise just want to reload
241 * the region not aligned to tile boundary */
242 //if (!res->damage.aligned)
243 // return true;
244 return true;
245 }
246 else if (surf->reload)
247 return true;
248
249 return false;
250 }
251
252 static void
253 lima_pack_reload_plbu_cmd(struct lima_context *ctx)
254 {
255 #define lima_reload_render_state_offset 0x0000
256 #define lima_reload_gl_pos_offset 0x0040
257 #define lima_reload_varying_offset 0x0080
258 #define lima_reload_tex_desc_offset 0x00c0
259 #define lima_reload_tex_array_offset 0x0100
260 #define lima_reload_buffer_size 0x0140
261
262 void *cpu;
263 unsigned offset;
264 struct pipe_resource *pres = NULL;
265 u_upload_alloc(ctx->uploader, 0, lima_reload_buffer_size,
266 0x40, &offset, &pres, &cpu);
267
268 struct lima_resource *res = lima_resource(pres);
269 uint32_t va = res->bo->va + offset;
270
271 struct lima_screen *screen = lima_screen(ctx->base.screen);
272
273 uint32_t reload_shader_first_instr_size =
274 ((uint32_t *)(screen->pp_buffer->map + pp_reload_program_offset))[0] & 0x1f;
275 uint32_t reload_shader_va = screen->pp_buffer->va + pp_reload_program_offset;
276
277 struct lima_render_state reload_render_state = {
278 .alpha_blend = 0xf03b1ad2,
279 .depth_test = 0x0000000e,
280 .depth_range = 0xffff0000,
281 .stencil_front = 0x00000007,
282 .stencil_back = 0x00000007,
283 .multi_sample = 0x0000f007,
284 .shader_address = reload_shader_va | reload_shader_first_instr_size,
285 .varying_types = 0x00000001,
286 .textures_address = va + lima_reload_tex_array_offset,
287 .aux0 = 0x00004021,
288 .varyings_address = va + lima_reload_varying_offset,
289 };
290 memcpy(cpu + lima_reload_render_state_offset, &reload_render_state,
291 sizeof(reload_render_state));
292
293 struct lima_context_framebuffer *fb = &ctx->framebuffer;
294 lima_tex_desc *td = cpu + lima_reload_tex_desc_offset;
295 memset(td, 0, lima_min_tex_desc_size);
296 lima_texture_desc_set_res(ctx, td, fb->base.cbufs[0]->texture, 0, 0);
297 td->unnorm_coords = 1;
298 td->texture_type = LIMA_TEXTURE_TYPE_2D;
299 td->min_img_filter_nearest = 1;
300 td->mag_img_filter_nearest = 1;
301 td->wrap_s_clamp_to_edge = 1;
302 td->wrap_t_clamp_to_edge = 1;
303 td->unknown_2_2 = 0x1;
304
305 uint32_t *ta = cpu + lima_reload_tex_array_offset;
306 ta[0] = va + lima_reload_tex_desc_offset;
307
308 float reload_gl_pos[] = {
309 fb->base.width, 0, 0, 1,
310 0, 0, 0, 1,
311 0, fb->base.height, 0, 1,
312 };
313 memcpy(cpu + lima_reload_gl_pos_offset, reload_gl_pos,
314 sizeof(reload_gl_pos));
315
316 float reload_varying[] = {
317 fb->base.width, 0, 0, 0,
318 0, fb->base.height, 0, 0,
319 };
320 memcpy(cpu + lima_reload_varying_offset, reload_varying,
321 sizeof(reload_varying));
322
323 lima_submit_add_bo(ctx->pp_submit, res->bo, LIMA_SUBMIT_BO_READ);
324 pipe_resource_reference(&pres, NULL);
325
326 PLBU_CMD_BEGIN(&ctx->plbu_cmd_head, 20);
327
328 PLBU_CMD_VIEWPORT_LEFT(0);
329 PLBU_CMD_VIEWPORT_RIGHT(fui(fb->base.width));
330 PLBU_CMD_VIEWPORT_BOTTOM(0);
331 PLBU_CMD_VIEWPORT_TOP(fui(fb->base.height));
332
333 PLBU_CMD_RSW_VERTEX_ARRAY(
334 va + lima_reload_render_state_offset,
335 va + lima_reload_gl_pos_offset);
336
337 PLBU_CMD_UNKNOWN2();
338 PLBU_CMD_UNKNOWN1();
339
340 PLBU_CMD_INDICES(screen->pp_buffer->va + pp_shared_index_offset);
341 PLBU_CMD_INDEXED_DEST(va + lima_reload_gl_pos_offset);
342 PLBU_CMD_DRAW_ELEMENTS(0xf, 0, 3);
343
344 PLBU_CMD_END();
345 }
346
347 static void
348 lima_pack_head_plbu_cmd(struct lima_context *ctx)
349 {
350 struct lima_context_framebuffer *fb = &ctx->framebuffer;
351
352 PLBU_CMD_BEGIN(&ctx->plbu_cmd_head, 10);
353
354 PLBU_CMD_UNKNOWN2();
355 PLBU_CMD_BLOCK_STEP(fb->shift_min, fb->shift_h, fb->shift_w);
356 PLBU_CMD_TILED_DIMENSIONS(fb->tiled_w, fb->tiled_h);
357 PLBU_CMD_BLOCK_STRIDE(fb->block_w);
358
359 PLBU_CMD_ARRAY_ADDRESS(
360 ctx->plb_gp_stream->va + ctx->plb_index * ctx->plb_gp_size,
361 fb->block_w * fb->block_h);
362
363 PLBU_CMD_END();
364
365 if (lima_fb_need_reload(ctx))
366 lima_pack_reload_plbu_cmd(ctx);
367 }
368
369 static bool
370 lima_is_scissor_zero(struct lima_context *ctx)
371 {
372 if (!ctx->rasterizer || !ctx->rasterizer->base.scissor)
373 return false;
374
375 struct pipe_scissor_state *scissor = &ctx->scissor;
376 return
377 scissor->minx == scissor->maxx
378 && scissor->miny == scissor->maxy;
379 }
380
381 static void
382 hilbert_rotate(int n, int *x, int *y, int rx, int ry)
383 {
384 if (ry == 0) {
385 if (rx == 1) {
386 *x = n-1 - *x;
387 *y = n-1 - *y;
388 }
389
390 /* Swap x and y */
391 int t = *x;
392 *x = *y;
393 *y = t;
394 }
395 }
396
397 static void
398 hilbert_coords(int n, int d, int *x, int *y)
399 {
400 int rx, ry, i, t=d;
401
402 *x = *y = 0;
403
404 for (i = 0; (1 << i) < n; i++) {
405
406 rx = 1 & (t / 2);
407 ry = 1 & (t ^ rx);
408
409 hilbert_rotate(1 << i, x, y, rx, ry);
410
411 *x += rx << i;
412 *y += ry << i;
413
414 t /= 4;
415 }
416 }
417
418 static int
419 lima_get_pp_stream_size(int num_pp, int tiled_w, int tiled_h, uint32_t *off)
420 {
421 /* carefully calculate each stream start address:
422 * 1. overflow: each stream size may be different due to
423 * fb->tiled_w * fb->tiled_h can't be divided by num_pp,
424 * extra size should be added to the preceeding stream
425 * 2. alignment: each stream address should be 0x20 aligned
426 */
427 int delta = tiled_w * tiled_h / num_pp * 16 + 16;
428 int remain = tiled_w * tiled_h % num_pp;
429 int offset = 0;
430
431 for (int i = 0; i < num_pp; i++) {
432 off[i] = offset;
433
434 offset += delta;
435 if (remain) {
436 offset += 16;
437 remain--;
438 }
439 offset = align(offset, 0x20);
440 }
441
442 return offset;
443 }
444
445 static bool
446 inside_damage_region(int x, int y, struct lima_damage_region *ds)
447 {
448 if (!ds || !ds->region)
449 return true;
450
451 for (int i = 0; i < ds->num_region; i++) {
452 struct pipe_scissor_state *ss = ds->region + i;
453 if (x >= ss->minx && x < ss->maxx &&
454 y >= ss->miny && y < ss->maxy)
455 return true;
456 }
457
458 return false;
459 }
460
461 static void
462 lima_generate_pp_stream(struct lima_context *ctx, int off_x, int off_y,
463 int tiled_w, int tiled_h)
464 {
465 struct lima_pp_stream_state *ps = &ctx->pp_stream;
466 struct lima_context_framebuffer *fb = &ctx->framebuffer;
467 struct lima_damage_region *damage = lima_ctx_get_damage(ctx);
468 struct lima_screen *screen = lima_screen(ctx->base.screen);
469 int i, num_pp = screen->num_pp;
470
471 /* use hilbert_coords to generates 1D to 2D relationship.
472 * 1D for pp stream index and 2D for plb block x/y on framebuffer.
473 * if multi pp, interleave the 1D index to make each pp's render target
474 * close enough which should result close workload
475 */
476 int max = MAX2(tiled_w, tiled_h);
477 int index = 0;
478 uint32_t *stream[4];
479 int si[4] = {0};
480 int dim = 0;
481 int count = 0;
482
483 /* Don't update count if we get zero rect. We'll just generate
484 * PP stream with just terminators in it.
485 */
486 if ((tiled_w * tiled_h) != 0) {
487 dim = util_logbase2_ceil(max);
488 count = 1 << (dim + dim);
489 }
490
491 for (i = 0; i < num_pp; i++)
492 stream[i] = ps->bo->map + ps->bo_offset + ps->offset[i];
493
494 for (i = 0; i < count; i++) {
495 int x, y;
496 hilbert_coords(max, i, &x, &y);
497 if (x < tiled_w && y < tiled_h) {
498 x += off_x;
499 y += off_y;
500
501 if (!inside_damage_region(x, y, damage))
502 continue;
503
504 int pp = index % num_pp;
505 int offset = ((y >> fb->shift_h) * fb->block_w +
506 (x >> fb->shift_w)) * LIMA_CTX_PLB_BLK_SIZE;
507 int plb_va = ctx->plb[ctx->plb_index]->va + offset;
508
509 stream[pp][si[pp]++] = 0;
510 stream[pp][si[pp]++] = 0xB8000000 | x | (y << 8);
511 stream[pp][si[pp]++] = 0xE0000002 | ((plb_va >> 3) & ~0xE0000003);
512 stream[pp][si[pp]++] = 0xB0000000;
513
514 index++;
515 }
516 }
517
518 for (i = 0; i < num_pp; i++) {
519 stream[i][si[i]++] = 0;
520 stream[i][si[i]++] = 0xBC000000;
521 stream[i][si[i]++] = 0;
522 stream[i][si[i]++] = 0;
523
524 lima_dump_command_stream_print(
525 stream[i], si[i] * 4, false, "pp plb stream %d at va %x\n",
526 i, ps->bo->va + ps->bo_offset + ps->offset[i]);
527 }
528 }
529
530 static void
531 lima_update_damage_pp_stream(struct lima_context *ctx)
532 {
533 struct lima_damage_region *ds = lima_ctx_get_damage(ctx);
534 struct lima_context_framebuffer *fb = &ctx->framebuffer;
535 struct pipe_scissor_state bound;
536
537 if (ds && ds->region) {
538 struct pipe_scissor_state *dbound = &ds->bound;
539 bound.minx = MAX2(dbound->minx, ctx->damage_rect.minx >> 4);
540 bound.miny = MAX2(dbound->miny, ctx->damage_rect.miny >> 4);
541 bound.maxx = MIN2(dbound->maxx, (ctx->damage_rect.maxx + 0xf) >> 4);
542 bound.maxy = MIN2(dbound->maxy, (ctx->damage_rect.maxy + 0xf) >> 4);
543 } else {
544 bound.minx = ctx->damage_rect.minx >> 4;
545 bound.miny = ctx->damage_rect.miny >> 4;
546 bound.maxx = (ctx->damage_rect.maxx + 0xf) >> 4;
547 bound.maxy = (ctx->damage_rect.maxy + 0xf) >> 4;
548 }
549
550 /* Clamp to FB size */
551 bound.minx = MIN2(bound.minx, fb->tiled_w);
552 bound.miny = MIN2(bound.miny, fb->tiled_h);
553 bound.maxx = MIN2(bound.maxx, fb->tiled_w);
554 bound.maxy = MIN2(bound.maxy, fb->tiled_h);
555
556 int tiled_w = bound.maxx - bound.minx;
557 int tiled_h = bound.maxy - bound.miny;
558
559 struct lima_screen *screen = lima_screen(ctx->base.screen);
560 int size = lima_get_pp_stream_size(
561 screen->num_pp, tiled_w, tiled_h, ctx->pp_stream.offset);
562
563 void *cpu;
564 unsigned offset;
565 struct pipe_resource *pres = NULL;
566 u_upload_alloc(ctx->uploader, 0, size, 0x40, &offset, &pres, &cpu);
567
568 struct lima_resource *res = lima_resource(pres);
569 ctx->pp_stream.bo = res->bo;
570 ctx->pp_stream.bo_offset = offset;
571
572 lima_generate_pp_stream(ctx, bound.minx, bound.miny, tiled_w, tiled_h);
573
574 lima_submit_add_bo(ctx->pp_submit, res->bo, LIMA_SUBMIT_BO_READ);
575 pipe_resource_reference(&pres, NULL);
576 }
577
578 static void
579 lima_update_full_pp_stream(struct lima_context *ctx)
580 {
581 struct lima_context_framebuffer *fb = &ctx->framebuffer;
582 struct lima_ctx_plb_pp_stream_key key = {
583 .plb_index = ctx->plb_index,
584 .tiled_w = fb->tiled_w,
585 .tiled_h = fb->tiled_h,
586 };
587
588 struct hash_entry *entry =
589 _mesa_hash_table_search(ctx->plb_pp_stream, &key);
590 struct lima_ctx_plb_pp_stream *s = entry->data;
591
592 if (s->bo) {
593 ctx->pp_stream.bo = s->bo;
594 ctx->pp_stream.bo_offset = 0;
595 memcpy(ctx->pp_stream.offset, s->offset, sizeof(s->offset));
596 }
597 else {
598 struct lima_screen *screen = lima_screen(ctx->base.screen);
599 int size = lima_get_pp_stream_size(
600 screen->num_pp, fb->tiled_w, fb->tiled_h, s->offset);
601 s->bo = lima_bo_create(screen, size, 0);
602 lima_bo_map(s->bo);
603
604 ctx->pp_stream.bo = s->bo;
605 ctx->pp_stream.bo_offset = 0;
606 memcpy(ctx->pp_stream.offset, s->offset, sizeof(s->offset));
607
608 lima_generate_pp_stream(ctx, 0, 0, fb->tiled_w, fb->tiled_h);
609 }
610
611 lima_submit_add_bo(ctx->pp_submit, s->bo, LIMA_SUBMIT_BO_READ);
612 }
613
614 static bool
615 lima_damage_fullscreen(struct lima_context *ctx)
616 {
617 return ctx->damage_rect.minx == 0 &&
618 ctx->damage_rect.miny == 0 &&
619 ctx->damage_rect.maxx == ctx->framebuffer.base.width &&
620 ctx->damage_rect.maxy == ctx->framebuffer.base.height;
621 }
622
623 static void
624 lima_update_pp_stream(struct lima_context *ctx)
625 {
626 struct lima_damage_region *damage = lima_ctx_get_damage(ctx);
627 if ((damage && damage->region) || !lima_damage_fullscreen(ctx))
628 lima_update_damage_pp_stream(ctx);
629 else if (ctx->plb_pp_stream)
630 lima_update_full_pp_stream(ctx);
631 else
632 ctx->pp_stream.bo = NULL;
633 }
634
635 static void
636 lima_update_submit_wb(struct lima_context *ctx)
637 {
638 if (lima_ctx_dirty(ctx))
639 return;
640
641 if (ctx->framebuffer.base.nr_cbufs) {
642 struct lima_resource *res = lima_resource(ctx->framebuffer.base.cbufs[0]->texture);
643 lima_submit_add_bo(ctx->pp_submit, res->bo, LIMA_SUBMIT_BO_WRITE);
644 }
645
646 if (ctx->framebuffer.base.zsbuf) {
647 struct lima_resource *res = lima_resource(ctx->framebuffer.base.zsbuf->texture);
648 lima_submit_add_bo(ctx->pp_submit, res->bo, LIMA_SUBMIT_BO_WRITE);
649 }
650 }
651
652 static void
653 lima_update_submit_bo(struct lima_context *ctx)
654 {
655 lima_submit_add_bo(ctx->gp_submit, ctx->plb_gp_stream, LIMA_SUBMIT_BO_READ);
656 lima_submit_add_bo(ctx->gp_submit, ctx->plb[ctx->plb_index], LIMA_SUBMIT_BO_WRITE);
657 lima_submit_add_bo(ctx->gp_submit, ctx->gp_tile_heap[ctx->plb_index], LIMA_SUBMIT_BO_WRITE);
658
659 lima_dump_command_stream_print(
660 ctx->plb_gp_stream->map + ctx->plb_index * ctx->plb_gp_size,
661 ctx->plb_gp_size, false, "gp plb stream at va %x\n",
662 ctx->plb_gp_stream->va + ctx->plb_index * ctx->plb_gp_size);
663
664 lima_submit_add_bo(ctx->pp_submit, ctx->plb[ctx->plb_index], LIMA_SUBMIT_BO_READ);
665 lima_submit_add_bo(ctx->pp_submit, ctx->gp_tile_heap[ctx->plb_index], LIMA_SUBMIT_BO_READ);
666
667 struct lima_screen *screen = lima_screen(ctx->base.screen);
668 lima_submit_add_bo(ctx->pp_submit, screen->pp_buffer, LIMA_SUBMIT_BO_READ);
669 }
670
671 static void
672 lima_damage_rect_union(struct lima_context *ctx, unsigned minx, unsigned maxx, unsigned miny, unsigned maxy)
673 {
674 ctx->damage_rect.minx = MIN2(ctx->damage_rect.minx, minx);
675 ctx->damage_rect.miny = MIN2(ctx->damage_rect.miny, miny);
676 ctx->damage_rect.maxx = MAX2(ctx->damage_rect.maxx, maxx);
677 ctx->damage_rect.maxy = MAX2(ctx->damage_rect.maxy, maxy);
678 }
679
680 static void
681 lima_clear(struct pipe_context *pctx, unsigned buffers,
682 const union pipe_color_union *color, double depth, unsigned stencil)
683 {
684 struct lima_context *ctx = lima_context(pctx);
685
686 lima_flush(ctx);
687
688 lima_update_submit_wb(ctx);
689
690 ctx->resolve |= buffers;
691
692 /* no need to reload if cleared */
693 if (ctx->framebuffer.base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0)) {
694 struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
695 surf->reload = false;
696 }
697
698 struct lima_context_clear *clear = &ctx->clear;
699 clear->buffers = buffers;
700
701 if (buffers & PIPE_CLEAR_COLOR0) {
702 clear->color_8pc =
703 ((uint32_t)float_to_ubyte(color->f[3]) << 24) |
704 ((uint32_t)float_to_ubyte(color->f[2]) << 16) |
705 ((uint32_t)float_to_ubyte(color->f[1]) << 8) |
706 float_to_ubyte(color->f[0]);
707
708 clear->color_16pc =
709 ((uint64_t)float_to_ushort(color->f[3]) << 48) |
710 ((uint64_t)float_to_ushort(color->f[2]) << 32) |
711 ((uint64_t)float_to_ushort(color->f[1]) << 16) |
712 float_to_ushort(color->f[0]);
713 }
714
715 if (buffers & PIPE_CLEAR_DEPTH)
716 clear->depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
717
718 if (buffers & PIPE_CLEAR_STENCIL)
719 clear->stencil = stencil;
720
721 ctx->dirty |= LIMA_CONTEXT_DIRTY_CLEAR;
722
723 lima_damage_rect_union(ctx, 0, ctx->framebuffer.base.width,
724 0, ctx->framebuffer.base.height);
725 }
726
727 enum lima_attrib_type {
728 LIMA_ATTRIB_FLOAT = 0x000,
729 /* todo: find out what lives here. */
730 LIMA_ATTRIB_I16 = 0x004,
731 LIMA_ATTRIB_U16 = 0x005,
732 LIMA_ATTRIB_I8 = 0x006,
733 LIMA_ATTRIB_U8 = 0x007,
734 LIMA_ATTRIB_I8N = 0x008,
735 LIMA_ATTRIB_U8N = 0x009,
736 LIMA_ATTRIB_I16N = 0x00A,
737 LIMA_ATTRIB_U16N = 0x00B,
738 /* todo: where is the 32 int */
739 /* todo: find out what lives here. */
740 LIMA_ATTRIB_FIXED = 0x101
741 };
742
743 static enum lima_attrib_type
744 lima_pipe_format_to_attrib_type(enum pipe_format format)
745 {
746 const struct util_format_description *desc = util_format_description(format);
747 int i = util_format_get_first_non_void_channel(format);
748 const struct util_format_channel_description *c = desc->channel + i;
749
750 switch (c->type) {
751 case UTIL_FORMAT_TYPE_FLOAT:
752 return LIMA_ATTRIB_FLOAT;
753 case UTIL_FORMAT_TYPE_FIXED:
754 return LIMA_ATTRIB_FIXED;
755 case UTIL_FORMAT_TYPE_SIGNED:
756 if (c->size == 8) {
757 if (c->normalized)
758 return LIMA_ATTRIB_I8N;
759 else
760 return LIMA_ATTRIB_I8;
761 }
762 else if (c->size == 16) {
763 if (c->normalized)
764 return LIMA_ATTRIB_I16N;
765 else
766 return LIMA_ATTRIB_I16;
767 }
768 break;
769 case UTIL_FORMAT_TYPE_UNSIGNED:
770 if (c->size == 8) {
771 if (c->normalized)
772 return LIMA_ATTRIB_U8N;
773 else
774 return LIMA_ATTRIB_U8;
775 }
776 else if (c->size == 16) {
777 if (c->normalized)
778 return LIMA_ATTRIB_U16N;
779 else
780 return LIMA_ATTRIB_U16;
781 }
782 break;
783 }
784
785 return LIMA_ATTRIB_FLOAT;
786 }
787
788 static void
789 lima_pack_vs_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
790 {
791 VS_CMD_BEGIN(&ctx->vs_cmd_array, 24);
792
793 if (!info->index_size) {
794 VS_CMD_ARRAYS_SEMAPHORE_BEGIN_1();
795 VS_CMD_ARRAYS_SEMAPHORE_BEGIN_2();
796 }
797
798 int uniform_size = ctx->vs->uniform_pending_offset + ctx->vs->constant_size + 32;
799 VS_CMD_UNIFORMS_ADDRESS(
800 lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform),
801 align(uniform_size, 16));
802
803 VS_CMD_SHADER_ADDRESS(ctx->vs->bo->va, ctx->vs->shader_size);
804 VS_CMD_SHADER_INFO(ctx->vs->prefetch, ctx->vs->shader_size);
805
806 int num_outputs = ctx->vs->num_outputs;
807 int num_attributes = ctx->vertex_elements->num_elements;
808 VS_CMD_VARYING_ATTRIBUTE_COUNT(num_outputs, MAX2(1, num_attributes));
809
810 VS_CMD_UNKNOWN1();
811
812 VS_CMD_ATTRIBUTES_ADDRESS(
813 lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info),
814 MAX2(1, num_attributes));
815
816 VS_CMD_VARYINGS_ADDRESS(
817 lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info),
818 num_outputs);
819
820 unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
821 VS_CMD_DRAW(num, info->index_size);
822
823 VS_CMD_UNKNOWN2();
824
825 VS_CMD_ARRAYS_SEMAPHORE_END(info->index_size);
826
827 VS_CMD_END();
828 }
829
830 static void
831 lima_pack_plbu_cmd(struct lima_context *ctx, const struct pipe_draw_info *info)
832 {
833 struct lima_context_framebuffer *fb = &ctx->framebuffer;
834 struct lima_vs_shader_state *vs = ctx->vs;
835 unsigned minx, maxx, miny, maxy;
836
837 /* If it's zero scissor, we skip adding all other commands */
838 if (lima_is_scissor_zero(ctx))
839 return;
840
841 PLBU_CMD_BEGIN(&ctx->plbu_cmd_array, 32);
842
843 PLBU_CMD_VIEWPORT_LEFT(fui(ctx->viewport.left));
844 PLBU_CMD_VIEWPORT_RIGHT(fui(ctx->viewport.right));
845 PLBU_CMD_VIEWPORT_BOTTOM(fui(ctx->viewport.bottom));
846 PLBU_CMD_VIEWPORT_TOP(fui(ctx->viewport.top));
847
848 if (!info->index_size)
849 PLBU_CMD_ARRAYS_SEMAPHORE_BEGIN();
850
851 int cf = ctx->rasterizer->base.cull_face;
852 int ccw = ctx->rasterizer->base.front_ccw;
853 uint32_t cull = 0;
854 bool force_point_size = false;
855
856 if (cf != PIPE_FACE_NONE) {
857 if (cf & PIPE_FACE_FRONT)
858 cull |= ccw ? 0x00040000 : 0x00020000;
859 if (cf & PIPE_FACE_BACK)
860 cull |= ccw ? 0x00020000 : 0x00040000;
861 }
862
863 /* Specify point size with PLBU command if shader doesn't write */
864 if (info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1)
865 force_point_size = true;
866
867 /* Specify line width with PLBU command for lines */
868 if (info->mode > PIPE_PRIM_POINTS && info->mode < PIPE_PRIM_TRIANGLES)
869 force_point_size = true;
870
871 PLBU_CMD_PRIMITIVE_SETUP(force_point_size, cull, info->index_size);
872
873 PLBU_CMD_RSW_VERTEX_ARRAY(
874 lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw),
875 ctx->gp_output->va);
876
877 /* TODO
878 * - we should set it only for the first draw that enabled the scissor and for
879 * latter draw only if scissor is dirty
880 */
881 if (ctx->rasterizer->base.scissor) {
882 struct pipe_scissor_state *scissor = &ctx->scissor;
883 minx = scissor->minx;
884 maxx = scissor->maxx;
885 miny = scissor->miny;
886 maxy = scissor->maxy;
887 } else {
888 minx = 0;
889 maxx = fb->base.width;
890 miny = 0;
891 maxy = fb->base.height;
892 }
893
894 minx = MAX2(minx, ctx->viewport.left);
895 maxx = MIN2(maxx, ctx->viewport.right);
896 miny = MAX2(miny, ctx->viewport.bottom);
897 maxy = MIN2(maxy, ctx->viewport.top);
898
899 PLBU_CMD_SCISSORS(minx, maxx, miny, maxy);
900 lima_damage_rect_union(ctx, minx, maxx, miny, maxy);
901
902 PLBU_CMD_UNKNOWN1();
903
904 PLBU_CMD_DEPTH_RANGE_NEAR(fui(ctx->viewport.near));
905 PLBU_CMD_DEPTH_RANGE_FAR(fui(ctx->viewport.far));
906
907 if ((info->mode == PIPE_PRIM_POINTS && ctx->vs->point_size_idx == -1) ||
908 ((info->mode >= PIPE_PRIM_LINES) && (info->mode < PIPE_PRIM_TRIANGLES)))
909 {
910 uint32_t v = info->mode == PIPE_PRIM_POINTS ?
911 fui(ctx->rasterizer->base.point_size) : fui(ctx->rasterizer->base.line_width);
912 PLBU_CMD_LOW_PRIM_SIZE(v);
913 }
914
915 if (info->index_size) {
916 PLBU_CMD_INDEXED_DEST(ctx->gp_output->va);
917 if (vs->point_size_idx != -1)
918 PLBU_CMD_INDEXED_PT_SIZE(ctx->gp_output->va + ctx->gp_output_point_size_offt);
919
920 PLBU_CMD_INDICES(ctx->index_res->bo->va + info->start * info->index_size + ctx->index_offset);
921 }
922 else {
923 /* can this make the attribute info static? */
924 PLBU_CMD_DRAW_ARRAYS(info->mode, info->start, info->count);
925 }
926
927 PLBU_CMD_ARRAYS_SEMAPHORE_END();
928
929 if (info->index_size)
930 PLBU_CMD_DRAW_ELEMENTS(info->mode, ctx->min_index, info->count);
931
932 PLBU_CMD_END();
933 }
934
935 static int
936 lima_blend_func(enum pipe_blend_func pipe)
937 {
938 switch (pipe) {
939 case PIPE_BLEND_ADD:
940 return 2;
941 case PIPE_BLEND_SUBTRACT:
942 return 0;
943 case PIPE_BLEND_REVERSE_SUBTRACT:
944 return 1;
945 case PIPE_BLEND_MIN:
946 return 4;
947 case PIPE_BLEND_MAX:
948 return 5;
949 }
950 return -1;
951 }
952
953 static int
954 lima_blend_factor_has_alpha(enum pipe_blendfactor pipe)
955 {
956 /* Bit 4 is set if the blendfactor uses alpha */
957 switch (pipe) {
958 case PIPE_BLENDFACTOR_SRC_ALPHA:
959 case PIPE_BLENDFACTOR_DST_ALPHA:
960 case PIPE_BLENDFACTOR_CONST_ALPHA:
961 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
962 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
963 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
964 return 1;
965
966 case PIPE_BLENDFACTOR_SRC_COLOR:
967 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
968 case PIPE_BLENDFACTOR_DST_COLOR:
969 case PIPE_BLENDFACTOR_INV_DST_COLOR:
970 case PIPE_BLENDFACTOR_CONST_COLOR:
971 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
972 case PIPE_BLENDFACTOR_ZERO:
973 case PIPE_BLENDFACTOR_ONE:
974 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
975 return 0;
976
977 case PIPE_BLENDFACTOR_SRC1_COLOR:
978 case PIPE_BLENDFACTOR_SRC1_ALPHA:
979 case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
980 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
981 return -1; /* not supported */
982 }
983 return -1;
984 }
985
986 static int
987 lima_blend_factor_is_inv(enum pipe_blendfactor pipe)
988 {
989 /* Bit 3 is set if the blendfactor type is inverted */
990 switch (pipe) {
991 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
992 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
993 case PIPE_BLENDFACTOR_INV_DST_COLOR:
994 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
995 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
996 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
997 case PIPE_BLENDFACTOR_ONE:
998 return 1;
999
1000 case PIPE_BLENDFACTOR_SRC_COLOR:
1001 case PIPE_BLENDFACTOR_SRC_ALPHA:
1002 case PIPE_BLENDFACTOR_DST_COLOR:
1003 case PIPE_BLENDFACTOR_DST_ALPHA:
1004 case PIPE_BLENDFACTOR_CONST_COLOR:
1005 case PIPE_BLENDFACTOR_CONST_ALPHA:
1006 case PIPE_BLENDFACTOR_ZERO:
1007 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
1008 return 0;
1009
1010 case PIPE_BLENDFACTOR_SRC1_COLOR:
1011 case PIPE_BLENDFACTOR_SRC1_ALPHA:
1012 case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
1013 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
1014 return -1; /* not supported */
1015 }
1016 return -1;
1017 }
1018
1019 static int
1020 lima_blend_factor(enum pipe_blendfactor pipe)
1021 {
1022 /* Bits 0-2 indicate the blendfactor type */
1023 switch (pipe) {
1024 case PIPE_BLENDFACTOR_SRC_COLOR:
1025 case PIPE_BLENDFACTOR_SRC_ALPHA:
1026 case PIPE_BLENDFACTOR_INV_SRC_COLOR:
1027 case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
1028 return 0;
1029
1030 case PIPE_BLENDFACTOR_DST_COLOR:
1031 case PIPE_BLENDFACTOR_DST_ALPHA:
1032 case PIPE_BLENDFACTOR_INV_DST_COLOR:
1033 case PIPE_BLENDFACTOR_INV_DST_ALPHA:
1034 return 1;
1035
1036 case PIPE_BLENDFACTOR_CONST_COLOR:
1037 case PIPE_BLENDFACTOR_CONST_ALPHA:
1038 case PIPE_BLENDFACTOR_INV_CONST_COLOR:
1039 case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
1040 return 2;
1041
1042 case PIPE_BLENDFACTOR_ZERO:
1043 case PIPE_BLENDFACTOR_ONE:
1044 return 3;
1045
1046 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
1047 return 4;
1048
1049 case PIPE_BLENDFACTOR_SRC1_COLOR:
1050 case PIPE_BLENDFACTOR_SRC1_ALPHA:
1051 case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
1052 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
1053 return -1; /* not supported */
1054 }
1055 return -1;
1056 }
1057
1058 static int
1059 lima_calculate_alpha_blend(enum pipe_blend_func rgb_func, enum pipe_blend_func alpha_func,
1060 enum pipe_blendfactor rgb_src_factor, enum pipe_blendfactor rgb_dst_factor,
1061 enum pipe_blendfactor alpha_src_factor, enum pipe_blendfactor alpha_dst_factor)
1062 {
1063 /* PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE has to be changed to PIPE_BLENDFACTOR_ONE
1064 * if it is set for alpha_src.
1065 */
1066 if (alpha_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE)
1067 alpha_src_factor = PIPE_BLENDFACTOR_ONE;
1068
1069 return lima_blend_func(rgb_func) |
1070 (lima_blend_func(alpha_func) << 3) |
1071
1072 (lima_blend_factor(rgb_src_factor) << 6) |
1073 (lima_blend_factor_is_inv(rgb_src_factor) << 9) |
1074 (lima_blend_factor_has_alpha(rgb_src_factor) << 10) |
1075
1076 (lima_blend_factor(rgb_dst_factor) << 11) |
1077 (lima_blend_factor_is_inv(rgb_dst_factor) << 14) |
1078 (lima_blend_factor_has_alpha(rgb_dst_factor) << 15) |
1079
1080 (lima_blend_factor(alpha_src_factor) << 16) |
1081 (lima_blend_factor_is_inv(alpha_src_factor) << 19) |
1082
1083 (lima_blend_factor(alpha_dst_factor) << 20) |
1084 (lima_blend_factor_is_inv(alpha_dst_factor) << 23) |
1085 0x0C000000; /* need to check if this is GLESv1 glAlphaFunc */
1086 }
1087
1088 static int
1089 lima_stencil_op(enum pipe_stencil_op pipe)
1090 {
1091 switch (pipe) {
1092 case PIPE_STENCIL_OP_KEEP:
1093 return 0;
1094 case PIPE_STENCIL_OP_ZERO:
1095 return 2;
1096 case PIPE_STENCIL_OP_REPLACE:
1097 return 1;
1098 case PIPE_STENCIL_OP_INCR:
1099 return 6;
1100 case PIPE_STENCIL_OP_DECR:
1101 return 7;
1102 case PIPE_STENCIL_OP_INCR_WRAP:
1103 return 4;
1104 case PIPE_STENCIL_OP_DECR_WRAP:
1105 return 5;
1106 case PIPE_STENCIL_OP_INVERT:
1107 return 3;
1108 }
1109 return -1;
1110 }
1111
1112 static unsigned
1113 lima_calculate_depth_test(struct pipe_depth_state *depth, struct pipe_rasterizer_state *rst)
1114 {
1115 int offset_scale = 0, offset_units = 0;
1116 enum pipe_compare_func func = (depth->enabled ? depth->func : PIPE_FUNC_ALWAYS);
1117
1118 offset_scale = CLAMP(rst->offset_scale * 4, -128, 127);
1119 if (offset_scale < 0)
1120 offset_scale += 0x100;
1121
1122 offset_units = CLAMP(rst->offset_units * 2, -128, 127);
1123 if (offset_units < 0)
1124 offset_units += 0x100;
1125
1126 return (depth->enabled && depth->writemask) |
1127 ((int)func << 1) |
1128 (offset_scale << 16) |
1129 (offset_units << 24) |
1130 0x30; /* find out what is this */
1131 }
1132
1133 static void
1134 lima_pack_render_state(struct lima_context *ctx, const struct pipe_draw_info *info)
1135 {
1136 struct lima_fs_shader_state *fs = ctx->fs;
1137 struct lima_render_state *render =
1138 lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_plb_rsw,
1139 sizeof(*render));
1140 bool early_z = true;
1141 bool pixel_kill = true;
1142
1143 /* do hw support RGBA independ blend?
1144 * PIPE_CAP_INDEP_BLEND_ENABLE
1145 *
1146 * how to handle the no cbuf only zbuf case?
1147 */
1148 struct pipe_rt_blend_state *rt = ctx->blend->base.rt;
1149 render->blend_color_bg = float_to_ubyte(ctx->blend_color.color[2]) |
1150 (float_to_ubyte(ctx->blend_color.color[1]) << 16);
1151 render->blend_color_ra = float_to_ubyte(ctx->blend_color.color[0]) |
1152 (float_to_ubyte(ctx->blend_color.color[3]) << 16);
1153
1154 if (rt->blend_enable) {
1155 render->alpha_blend = lima_calculate_alpha_blend(rt->rgb_func, rt->alpha_func,
1156 rt->rgb_src_factor, rt->rgb_dst_factor,
1157 rt->alpha_src_factor, rt->alpha_dst_factor);
1158 }
1159 else {
1160 /*
1161 * Special handling for blending disabled.
1162 * Binary driver is generating the same alpha_value,
1163 * as when we would just enable blending, without changing/setting any blend equation/params.
1164 * Normaly in this case mesa would set all rt fields (func/factor) to zero.
1165 */
1166 render->alpha_blend = lima_calculate_alpha_blend(PIPE_BLEND_ADD, PIPE_BLEND_ADD,
1167 PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO,
1168 PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO);
1169 }
1170
1171 render->alpha_blend |= (rt->colormask & PIPE_MASK_RGBA) << 28;
1172
1173 struct pipe_rasterizer_state *rst = &ctx->rasterizer->base;
1174 struct pipe_depth_state *depth = &ctx->zsa->base.depth;
1175 render->depth_test = lima_calculate_depth_test(depth, rst);
1176
1177 ushort far, near;
1178
1179 near = float_to_ushort(ctx->viewport.near);
1180 far = float_to_ushort(ctx->viewport.far);
1181
1182 /* Subtract epsilon from 'near' if far == near. Make sure we don't get overflow */
1183 if ((far == near) && (near != 0))
1184 near--;
1185
1186 /* overlap with plbu? any place can remove one? */
1187 render->depth_range = near | (far << 16);
1188
1189 struct pipe_stencil_state *stencil = ctx->zsa->base.stencil;
1190 struct pipe_stencil_ref *ref = &ctx->stencil_ref;
1191
1192 if (stencil[0].enabled) { /* stencil is enabled */
1193 render->stencil_front = stencil[0].func |
1194 (lima_stencil_op(stencil[0].fail_op) << 3) |
1195 (lima_stencil_op(stencil[0].zfail_op) << 6) |
1196 (lima_stencil_op(stencil[0].zpass_op) << 9) |
1197 (ref->ref_value[0] << 16) |
1198 (stencil[0].valuemask << 24);
1199 render->stencil_back = render->stencil_front;
1200 render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[0].writemask & 0xff) << 8;
1201 if (stencil[1].enabled) { /* two-side is enabled */
1202 render->stencil_back = stencil[1].func |
1203 (lima_stencil_op(stencil[1].fail_op) << 3) |
1204 (lima_stencil_op(stencil[1].zfail_op) << 6) |
1205 (lima_stencil_op(stencil[1].zpass_op) << 9) |
1206 (ref->ref_value[1] << 16) |
1207 (stencil[1].valuemask << 24);
1208 render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[1].writemask & 0xff) << 8;
1209 }
1210 /* TODO: Find out, what (render->stecil_test & 0xffff0000) is.
1211 * 0x00ff0000 is probably (float_to_ubyte(alpha->ref_value) << 16)
1212 * (render->multi_sample & 0x00000007 is probably the compare function
1213 * of glAlphaFunc then.
1214 */
1215 }
1216 else {
1217 /* Default values, when stencil is disabled:
1218 * stencil[0|1].valuemask = 0xff
1219 * stencil[0|1].func = PIPE_FUNC_ALWAYS
1220 * stencil[0|1].writemask = 0xff
1221 */
1222 render->stencil_front = 0xff000007;
1223 render->stencil_back = 0xff000007;
1224 render->stencil_test = 0x0000ffff;
1225 }
1226
1227 /* need more investigation */
1228 if (info->mode == PIPE_PRIM_POINTS)
1229 render->multi_sample = 0x0000F007;
1230 else if (info->mode < PIPE_PRIM_TRIANGLES)
1231 render->multi_sample = 0x0000F407;
1232 else
1233 render->multi_sample = 0x0000F807;
1234 if (ctx->framebuffer.base.samples)
1235 render->multi_sample |= 0x68;
1236
1237 render->shader_address =
1238 ctx->fs->bo->va | (((uint32_t *)ctx->fs->bo->map)[0] & 0x1F);
1239
1240 /* seems not needed */
1241 render->uniforms_address = 0x00000000;
1242
1243 render->textures_address = 0x00000000;
1244
1245 render->aux0 = (ctx->vs->varying_stride >> 3);
1246 render->aux1 = 0x00001000;
1247 if (ctx->blend->base.dither)
1248 render->aux1 |= 0x00002000;
1249
1250 if (fs->uses_discard) {
1251 early_z = false;
1252 pixel_kill = false;
1253 }
1254
1255 if (rt->blend_enable)
1256 pixel_kill = false;
1257
1258 if ((rt->colormask & PIPE_MASK_RGBA) != PIPE_MASK_RGBA)
1259 pixel_kill = false;
1260
1261 if (early_z)
1262 render->aux0 |= 0x300;
1263
1264 if (pixel_kill)
1265 render->aux0 |= 0x1000;
1266
1267 if (ctx->tex_stateobj.num_samplers) {
1268 render->textures_address =
1269 lima_ctx_buff_va(ctx, lima_ctx_buff_pp_tex_desc);
1270 render->aux0 |= ctx->tex_stateobj.num_samplers << 14;
1271 render->aux0 |= 0x20;
1272 }
1273
1274 if (ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer) {
1275 render->uniforms_address =
1276 lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array);
1277 uint32_t size = ctx->buffer_state[lima_ctx_buff_pp_uniform].size;
1278 uint32_t bits = 0;
1279 if (size >= 8) {
1280 bits = util_last_bit(size >> 3) - 1;
1281 bits += size & u_bit_consecutive(0, bits + 3) ? 1 : 0;
1282 }
1283 render->uniforms_address |= bits > 0xf ? 0xf : bits;
1284
1285 render->aux0 |= 0x80;
1286 render->aux1 |= 0x10000;
1287 }
1288
1289 if (ctx->vs->num_varyings) {
1290 render->varying_types = 0x00000000;
1291 render->varyings_address = ctx->gp_output->va +
1292 ctx->gp_output_varyings_offt;
1293 for (int i = 0, index = 0; i < ctx->vs->num_outputs; i++) {
1294 int val;
1295
1296 if (i == ctx->vs->gl_pos_idx ||
1297 i == ctx->vs->point_size_idx)
1298 continue;
1299
1300 struct lima_varying_info *v = ctx->vs->varying + i;
1301 if (v->component_size == 4)
1302 val = v->components > 2 ? 0 : 1;
1303 else
1304 val = v->components > 2 ? 2 : 3;
1305
1306 if (index < 10)
1307 render->varying_types |= val << (3 * index);
1308 else if (index == 10) {
1309 render->varying_types |= val << 30;
1310 render->varyings_address |= val >> 2;
1311 }
1312 else if (index == 11)
1313 render->varyings_address |= val << 1;
1314
1315 index++;
1316 }
1317 }
1318 else {
1319 render->varying_types = 0x00000000;
1320 render->varyings_address = 0x00000000;
1321 }
1322
1323 lima_dump_command_stream_print(
1324 render, sizeof(*render), false, "add render state at va %x\n",
1325 lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
1326
1327 lima_dump_rsw_command_stream_print(
1328 render, sizeof(*render), lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
1329
1330 }
1331
1332 static void
1333 lima_update_gp_attribute_info(struct lima_context *ctx, const struct pipe_draw_info *info)
1334 {
1335 struct lima_vertex_element_state *ve = ctx->vertex_elements;
1336 struct lima_context_vertex_buffer *vb = &ctx->vertex_buffers;
1337
1338 uint32_t *attribute =
1339 lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_attribute_info,
1340 MAX2(1, ve->num_elements) * 8);
1341
1342 int n = 0;
1343 for (int i = 0; i < ve->num_elements; i++) {
1344 struct pipe_vertex_element *pve = ve->pipe + i;
1345
1346 assert(pve->vertex_buffer_index < vb->count);
1347 assert(vb->enabled_mask & (1 << pve->vertex_buffer_index));
1348
1349 struct pipe_vertex_buffer *pvb = vb->vb + pve->vertex_buffer_index;
1350 struct lima_resource *res = lima_resource(pvb->buffer.resource);
1351
1352 lima_submit_add_bo(ctx->gp_submit, res->bo, LIMA_SUBMIT_BO_READ);
1353
1354 unsigned start = info->index_size ? (ctx->min_index + info->index_bias) : info->start;
1355 attribute[n++] = res->bo->va + pvb->buffer_offset + pve->src_offset
1356 + start * pvb->stride;
1357 attribute[n++] = (pvb->stride << 11) |
1358 (lima_pipe_format_to_attrib_type(pve->src_format) << 2) |
1359 (util_format_get_nr_components(pve->src_format) - 1);
1360 }
1361
1362 lima_dump_command_stream_print(
1363 attribute, n * 4, false, "update attribute info at va %x\n",
1364 lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info));
1365 }
1366
1367 static void
1368 lima_update_gp_uniform(struct lima_context *ctx)
1369 {
1370 struct lima_context_constant_buffer *ccb =
1371 ctx->const_buffer + PIPE_SHADER_VERTEX;
1372 struct lima_vs_shader_state *vs = ctx->vs;
1373
1374 int size = vs->uniform_pending_offset + vs->constant_size + 32;
1375 void *vs_const_buff =
1376 lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_uniform, size);
1377
1378 if (ccb->buffer)
1379 memcpy(vs_const_buff, ccb->buffer, ccb->size);
1380
1381 memcpy(vs_const_buff + vs->uniform_pending_offset,
1382 ctx->viewport.transform.scale,
1383 sizeof(ctx->viewport.transform.scale));
1384 memcpy(vs_const_buff + vs->uniform_pending_offset + 16,
1385 ctx->viewport.transform.translate,
1386 sizeof(ctx->viewport.transform.translate));
1387
1388 if (vs->constant)
1389 memcpy(vs_const_buff + vs->uniform_pending_offset + 32,
1390 vs->constant, vs->constant_size);
1391
1392 lima_dump_command_stream_print(
1393 vs_const_buff, size, true,
1394 "update gp uniform at va %x\n",
1395 lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform));
1396 }
1397
1398 static void
1399 lima_update_pp_uniform(struct lima_context *ctx)
1400 {
1401 const float *const_buff = ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer;
1402 size_t const_buff_size = ctx->const_buffer[PIPE_SHADER_FRAGMENT].size / sizeof(float);
1403
1404 if (!const_buff)
1405 return;
1406
1407 uint16_t *fp16_const_buff =
1408 lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform,
1409 const_buff_size * sizeof(uint16_t));
1410
1411 uint32_t *array =
1412 lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform_array, 4);
1413
1414 for (int i = 0; i < const_buff_size; i++)
1415 fp16_const_buff[i] = util_float_to_half(const_buff[i]);
1416
1417 *array = lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform);
1418
1419 lima_dump_command_stream_print(
1420 fp16_const_buff, const_buff_size * 2, false, "add pp uniform data at va %x\n",
1421 lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform));
1422 lima_dump_command_stream_print(
1423 array, 4, false, "add pp uniform info at va %x\n",
1424 lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array));
1425 }
1426
1427 static void
1428 lima_update_varying(struct lima_context *ctx, const struct pipe_draw_info *info)
1429 {
1430 struct lima_screen *screen = lima_screen(ctx->base.screen);
1431 struct lima_vs_shader_state *vs = ctx->vs;
1432 uint32_t gp_output_size;
1433 unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : info->count;
1434
1435 uint32_t *varying =
1436 lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_varying_info,
1437 vs->num_outputs * 8);
1438 int n = 0;
1439
1440 int offset = 0;
1441
1442 for (int i = 0; i < vs->num_outputs; i++) {
1443 struct lima_varying_info *v = vs->varying + i;
1444
1445 if (i == vs->gl_pos_idx ||
1446 i == vs->point_size_idx)
1447 continue;
1448
1449 int size = v->component_size * 4;
1450
1451 /* does component_size == 2 need to be 16 aligned? */
1452 if (v->component_size == 4)
1453 offset = align(offset, 16);
1454
1455 v->offset = offset;
1456 offset += size;
1457 }
1458
1459 vs->varying_stride = align(offset, 16);
1460
1461 /* gl_Position is always present, allocate space for it */
1462 gp_output_size = align(4 * 4 * num, 0x40);
1463
1464 /* Allocate space for varyings if there're any */
1465 if (vs->num_varyings) {
1466 ctx->gp_output_varyings_offt = gp_output_size;
1467 gp_output_size += align(vs->varying_stride * num, 0x40);
1468 }
1469
1470 /* Allocate space for gl_PointSize if it's there */
1471 if (vs->point_size_idx != -1) {
1472 ctx->gp_output_point_size_offt = gp_output_size;
1473 gp_output_size += 4 * num;
1474 }
1475
1476 /* gp_output can be too large for the suballocator, so create a
1477 * separate bo for it. The bo cache should prevent performance hit.
1478 */
1479 ctx->gp_output = lima_bo_create(screen, gp_output_size, 0);
1480 assert(ctx->gp_output);
1481 lima_submit_add_bo(ctx->gp_submit, ctx->gp_output, LIMA_SUBMIT_BO_WRITE);
1482 lima_submit_add_bo(ctx->pp_submit, ctx->gp_output, LIMA_SUBMIT_BO_READ);
1483
1484 for (int i = 0; i < vs->num_outputs; i++) {
1485 struct lima_varying_info *v = vs->varying + i;
1486
1487 if (i == vs->gl_pos_idx) {
1488 /* gl_Position */
1489 varying[n++] = ctx->gp_output->va;
1490 varying[n++] = 0x8020;
1491 } else if (i == vs->point_size_idx) {
1492 /* gl_PointSize */
1493 varying[n++] = ctx->gp_output->va + ctx->gp_output_point_size_offt;
1494 varying[n++] = 0x2021;
1495 } else {
1496 /* Varying */
1497 varying[n++] = ctx->gp_output->va + ctx->gp_output_varyings_offt +
1498 v->offset;
1499 varying[n++] = (vs->varying_stride << 11) | (v->components - 1) |
1500 (v->component_size == 2 ? 0x0C : 0);
1501 }
1502 }
1503
1504 lima_dump_command_stream_print(
1505 varying, n * 4, false, "update varying info at va %x\n",
1506 lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info));
1507 }
1508
1509 static void
1510 lima_draw_vbo_update(struct pipe_context *pctx,
1511 const struct pipe_draw_info *info)
1512 {
1513 struct lima_context *ctx = lima_context(pctx);
1514
1515 lima_update_submit_wb(ctx);
1516
1517 lima_update_gp_attribute_info(ctx, info);
1518
1519 if ((ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
1520 ctx->const_buffer[PIPE_SHADER_VERTEX].dirty) ||
1521 ctx->dirty & LIMA_CONTEXT_DIRTY_VIEWPORT ||
1522 ctx->dirty & LIMA_CONTEXT_DIRTY_SHADER_VERT) {
1523 lima_update_gp_uniform(ctx);
1524 ctx->const_buffer[PIPE_SHADER_VERTEX].dirty = false;
1525 }
1526
1527 lima_update_varying(ctx, info);
1528
1529 /* If it's zero scissor, don't build vs cmd list */
1530 if (!lima_is_scissor_zero(ctx))
1531 lima_pack_vs_cmd(ctx, info);
1532
1533 if (ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
1534 ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty) {
1535 lima_update_pp_uniform(ctx);
1536 ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty = false;
1537 }
1538
1539 lima_update_textures(ctx);
1540
1541 lima_pack_render_state(ctx, info);
1542 lima_pack_plbu_cmd(ctx, info);
1543
1544 if (ctx->gp_output) {
1545 lima_bo_unreference(ctx->gp_output); /* held by submit */
1546 ctx->gp_output = NULL;
1547 }
1548
1549 if (ctx->framebuffer.base.zsbuf) {
1550 if (ctx->zsa->base.depth.enabled)
1551 ctx->resolve |= PIPE_CLEAR_DEPTH;
1552 if (ctx->zsa->base.stencil[0].enabled ||
1553 ctx->zsa->base.stencil[1].enabled)
1554 ctx->resolve |= PIPE_CLEAR_STENCIL;
1555 }
1556
1557 if (ctx->framebuffer.base.nr_cbufs)
1558 ctx->resolve |= PIPE_CLEAR_COLOR0;
1559
1560 ctx->dirty = 0;
1561 }
1562
1563 static void
1564 lima_draw_vbo_indexed(struct pipe_context *pctx,
1565 const struct pipe_draw_info *info)
1566 {
1567 struct lima_context *ctx = lima_context(pctx);
1568 struct pipe_resource *indexbuf = NULL;
1569
1570 /* Mali Utgard GPU always need min/max index info for index draw,
1571 * compute it if upper layer does not do for us */
1572 if (info->max_index == ~0u)
1573 u_vbuf_get_minmax_index(pctx, info, &ctx->min_index, &ctx->max_index);
1574 else {
1575 ctx->min_index = info->min_index;
1576 ctx->max_index = info->max_index;
1577 }
1578
1579 if (info->has_user_indices) {
1580 util_upload_index_buffer(&ctx->base, info, &indexbuf, &ctx->index_offset, 0x40);
1581 ctx->index_res = lima_resource(indexbuf);
1582 }
1583 else {
1584 ctx->index_res = lima_resource(info->index.resource);
1585 ctx->index_offset = 0;
1586 }
1587
1588 lima_submit_add_bo(ctx->gp_submit, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1589 lima_submit_add_bo(ctx->pp_submit, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1590 lima_draw_vbo_update(pctx, info);
1591
1592 if (indexbuf)
1593 pipe_resource_reference(&indexbuf, NULL);
1594 }
1595
1596 static void
1597 lima_draw_vbo_count(struct pipe_context *pctx,
1598 const struct pipe_draw_info *info)
1599 {
1600 static const uint32_t max_verts = 65535;
1601
1602 struct pipe_draw_info local_info = *info;
1603 unsigned start = info->start;
1604 unsigned count = info->count;
1605
1606 while (count) {
1607 unsigned this_count = count;
1608 unsigned step;
1609
1610 u_split_draw(info, max_verts, &this_count, &step);
1611
1612 local_info.start = start;
1613 local_info.count = this_count;
1614
1615 lima_draw_vbo_update(pctx, &local_info);
1616
1617 count -= step;
1618 start += step;
1619 }
1620 }
1621
1622 static void
1623 lima_draw_vbo(struct pipe_context *pctx,
1624 const struct pipe_draw_info *info)
1625 {
1626 /* check if draw mode and vertex/index count match,
1627 * otherwise gp will hang */
1628 if (!u_trim_pipe_prim(info->mode, (unsigned*)&info->count)) {
1629 debug_printf("draw mode and vertex/index count mismatch\n");
1630 return;
1631 }
1632
1633 struct lima_context *ctx = lima_context(pctx);
1634
1635 if (!ctx->vs || !ctx->fs) {
1636 debug_warn_once("no shader, skip draw\n");
1637 return;
1638 }
1639
1640 if (!lima_update_vs_state(ctx) || !lima_update_fs_state(ctx))
1641 return;
1642
1643 lima_dump_command_stream_print(
1644 ctx->vs->bo->map, ctx->vs->shader_size, false,
1645 "add vs at va %x\n", ctx->vs->bo->va);
1646
1647 lima_dump_command_stream_print(
1648 ctx->fs->bo->map, ctx->fs->shader_size, false,
1649 "add fs at va %x\n", ctx->fs->bo->va);
1650
1651 lima_submit_add_bo(ctx->gp_submit, ctx->vs->bo, LIMA_SUBMIT_BO_READ);
1652 lima_submit_add_bo(ctx->pp_submit, ctx->fs->bo, LIMA_SUBMIT_BO_READ);
1653
1654 if (info->index_size)
1655 lima_draw_vbo_indexed(pctx, info);
1656 else
1657 lima_draw_vbo_count(pctx, info);
1658 }
1659
1660 static void
1661 lima_finish_plbu_cmd(struct lima_context *ctx)
1662 {
1663 int i = 0;
1664 uint32_t *plbu_cmd = util_dynarray_ensure_cap(&ctx->plbu_cmd_array, ctx->plbu_cmd_array.size + 2 * 4);
1665
1666 plbu_cmd[i++] = 0x00000000;
1667 plbu_cmd[i++] = 0x50000000; /* END */
1668
1669 ctx->plbu_cmd_array.size += i * 4;
1670 }
1671
1672 static void
1673 lima_pack_wb_zsbuf_reg(struct lima_context *ctx, uint32_t *wb_reg, int wb_idx)
1674 {
1675 struct lima_context_framebuffer *fb = &ctx->framebuffer;
1676 struct lima_resource *res = lima_resource(fb->base.zsbuf->texture);
1677 int level = fb->base.zsbuf->u.tex.level;
1678 uint32_t format = lima_format_get_pixel(fb->base.zsbuf->format);
1679
1680 struct lima_pp_wb_reg *wb = (void *)wb_reg;
1681 wb[wb_idx].type = 0x01; /* 1 for depth, stencil */
1682 wb[wb_idx].address = res->bo->va + res->levels[level].offset;
1683 wb[wb_idx].pixel_format = format;
1684 if (res->tiled) {
1685 wb[wb_idx].pixel_layout = 0x2;
1686 wb[wb_idx].pitch = fb->tiled_w;
1687 } else {
1688 wb[wb_idx].pixel_layout = 0x0;
1689 wb[wb_idx].pitch = res->levels[level].stride / 8;
1690 }
1691 wb[wb_idx].mrt_bits = 0;
1692 }
1693
1694 static void
1695 lima_pack_wb_cbuf_reg(struct lima_context *ctx, uint32_t *wb_reg, int wb_idx)
1696 {
1697 struct lima_context_framebuffer *fb = &ctx->framebuffer;
1698 struct lima_resource *res = lima_resource(fb->base.cbufs[0]->texture);
1699 int level = fb->base.cbufs[0]->u.tex.level;
1700 unsigned layer = fb->base.cbufs[0]->u.tex.first_layer;
1701 uint32_t format = lima_format_get_pixel(fb->base.cbufs[0]->format);
1702 bool swap_channels = lima_format_get_swap_rb(fb->base.cbufs[0]->format);
1703
1704 struct lima_pp_wb_reg *wb = (void *)wb_reg;
1705 wb[wb_idx].type = 0x02; /* 2 for color buffer */
1706 wb[wb_idx].address = res->bo->va + res->levels[level].offset + layer * res->levels[level].layer_stride;
1707 wb[wb_idx].pixel_format = format;
1708 if (res->tiled) {
1709 wb[wb_idx].pixel_layout = 0x2;
1710 wb[wb_idx].pitch = fb->tiled_w;
1711 } else {
1712 wb[wb_idx].pixel_layout = 0x0;
1713 wb[wb_idx].pitch = res->levels[level].stride / 8;
1714 }
1715 wb[wb_idx].mrt_bits = swap_channels ? 0x4 : 0x0;
1716 }
1717
1718
1719 static void
1720 lima_pack_pp_frame_reg(struct lima_context *ctx, uint32_t *frame_reg,
1721 uint32_t *wb_reg)
1722 {
1723 struct lima_context_framebuffer *fb = &ctx->framebuffer;
1724 struct lima_pp_frame_reg *frame = (void *)frame_reg;
1725 struct lima_screen *screen = lima_screen(ctx->base.screen);
1726 int wb_idx = 0;
1727
1728 frame->render_address = screen->pp_buffer->va + pp_frame_rsw_offset;
1729 frame->flags = 0x02;
1730 frame->clear_value_depth = ctx->clear.depth;
1731 frame->clear_value_stencil = ctx->clear.stencil;
1732 frame->clear_value_color = ctx->clear.color_8pc;
1733 frame->clear_value_color_1 = ctx->clear.color_8pc;
1734 frame->clear_value_color_2 = ctx->clear.color_8pc;
1735 frame->clear_value_color_3 = ctx->clear.color_8pc;
1736 frame->one = 1;
1737
1738 frame->width = fb->base.width - 1;
1739 frame->height = fb->base.height - 1;
1740
1741 /* frame->fragment_stack_address is overwritten per-pp in the kernel
1742 * by the values of pp_frame.fragment_stack_address[i] */
1743
1744 /* These are "stack size" and "stack offset" shifted,
1745 * here they are assumed to be always the same. */
1746 frame->fragment_stack_size = ctx->pp_max_stack_size << 16 | ctx->pp_max_stack_size;
1747
1748 /* related with MSAA and different value when r4p0/r7p0 */
1749 frame->supersampled_height = fb->base.height * 2 - 1;
1750 frame->scale = 0xE0C;
1751
1752 frame->dubya = 0x77;
1753 frame->onscreen = 1;
1754 frame->blocking = (fb->shift_min << 28) | (fb->shift_h << 16) | fb->shift_w;
1755 frame->foureight = 0x8888;
1756
1757 if (fb->base.nr_cbufs && (ctx->resolve & PIPE_CLEAR_COLOR0))
1758 lima_pack_wb_cbuf_reg(ctx, wb_reg, wb_idx++);
1759
1760 if (fb->base.zsbuf &&
1761 (ctx->resolve & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)))
1762 lima_pack_wb_zsbuf_reg(ctx, wb_reg, wb_idx++);
1763 }
1764
1765 static void
1766 _lima_flush(struct lima_context *ctx, bool end_of_frame)
1767 {
1768 #define pp_stack_pp_size 0x400
1769
1770 lima_pack_head_plbu_cmd(ctx);
1771 lima_finish_plbu_cmd(ctx);
1772
1773 lima_update_submit_bo(ctx);
1774
1775 int vs_cmd_size = ctx->vs_cmd_array.size;
1776 uint32_t vs_cmd_va = 0;
1777
1778 if (vs_cmd_size) {
1779 void *vs_cmd =
1780 lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_vs_cmd, vs_cmd_size);
1781 memcpy(vs_cmd, util_dynarray_begin(&ctx->vs_cmd_array), vs_cmd_size);
1782 util_dynarray_clear(&ctx->vs_cmd_array);
1783 vs_cmd_va = lima_ctx_buff_va(ctx, lima_ctx_buff_gp_vs_cmd);
1784
1785 lima_dump_command_stream_print(
1786 vs_cmd, vs_cmd_size, false, "flush vs cmd at va %x\n", vs_cmd_va);
1787 lima_dump_vs_command_stream_print(vs_cmd, vs_cmd_size, vs_cmd_va);
1788 }
1789
1790 int plbu_cmd_size = ctx->plbu_cmd_array.size + ctx->plbu_cmd_head.size;
1791 void *plbu_cmd =
1792 lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_plbu_cmd, plbu_cmd_size);
1793 memcpy(plbu_cmd,
1794 util_dynarray_begin(&ctx->plbu_cmd_head),
1795 ctx->plbu_cmd_head.size);
1796 memcpy(plbu_cmd + ctx->plbu_cmd_head.size,
1797 util_dynarray_begin(&ctx->plbu_cmd_array),
1798 ctx->plbu_cmd_array.size);
1799 util_dynarray_clear(&ctx->plbu_cmd_array);
1800 util_dynarray_clear(&ctx->plbu_cmd_head);
1801 uint32_t plbu_cmd_va = lima_ctx_buff_va(ctx, lima_ctx_buff_gp_plbu_cmd);
1802
1803 lima_dump_command_stream_print(
1804 plbu_cmd, plbu_cmd_size, false, "flush plbu cmd at va %x\n", plbu_cmd_va);
1805 lima_dump_plbu_command_stream_print(plbu_cmd, plbu_cmd_size, plbu_cmd_va);
1806
1807 struct lima_screen *screen = lima_screen(ctx->base.screen);
1808 struct drm_lima_gp_frame gp_frame;
1809 struct lima_gp_frame_reg *gp_frame_reg = (void *)gp_frame.frame;
1810 gp_frame_reg->vs_cmd_start = vs_cmd_va;
1811 gp_frame_reg->vs_cmd_end = vs_cmd_va + vs_cmd_size;
1812 gp_frame_reg->plbu_cmd_start = plbu_cmd_va;
1813 gp_frame_reg->plbu_cmd_end = plbu_cmd_va + plbu_cmd_size;
1814 gp_frame_reg->tile_heap_start = ctx->gp_tile_heap[ctx->plb_index]->va;
1815 gp_frame_reg->tile_heap_end = ctx->gp_tile_heap[ctx->plb_index]->va + ctx->gp_tile_heap_size;
1816
1817 lima_dump_command_stream_print(
1818 &gp_frame, sizeof(gp_frame), false, "add gp frame\n");
1819
1820 if (!lima_submit_start(ctx->gp_submit, &gp_frame, sizeof(gp_frame)))
1821 fprintf(stderr, "gp submit error\n");
1822
1823 if (lima_dump_command_stream) {
1824 if (lima_submit_wait(ctx->gp_submit, PIPE_TIMEOUT_INFINITE)) {
1825 if (ctx->gp_output) {
1826 float *pos = lima_bo_map(ctx->gp_output);
1827 lima_dump_command_stream_print(
1828 pos, 4 * 4 * 16, true, "gl_pos dump at va %x\n",
1829 ctx->gp_output->va);
1830 }
1831
1832 uint32_t *plb = lima_bo_map(ctx->plb[ctx->plb_index]);
1833 lima_dump_command_stream_print(
1834 plb, LIMA_CTX_PLB_BLK_SIZE, false, "plb dump at va %x\n",
1835 ctx->plb[ctx->plb_index]->va);
1836 }
1837 else {
1838 fprintf(stderr, "gp submit wait error\n");
1839 exit(1);
1840 }
1841 }
1842
1843 uint32_t pp_stack_va = 0;
1844 if (ctx->pp_max_stack_size) {
1845 lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_stack, screen->num_pp *
1846 ctx->pp_max_stack_size * pp_stack_pp_size);
1847 pp_stack_va = lima_ctx_buff_va(ctx, lima_ctx_buff_pp_stack);
1848 }
1849
1850 lima_update_pp_stream(ctx);
1851
1852 struct lima_pp_stream_state *ps = &ctx->pp_stream;
1853 if (screen->gpu_type == DRM_LIMA_PARAM_GPU_ID_MALI400) {
1854 struct drm_lima_m400_pp_frame pp_frame = {0};
1855 lima_pack_pp_frame_reg(ctx, pp_frame.frame, pp_frame.wb);
1856 pp_frame.num_pp = screen->num_pp;
1857
1858 for (int i = 0; i < screen->num_pp; i++) {
1859 pp_frame.plbu_array_address[i] = ps->bo->va + ps->bo_offset + ps->offset[i];
1860 if (ctx->pp_max_stack_size)
1861 pp_frame.fragment_stack_address[i] = pp_stack_va +
1862 ctx->pp_max_stack_size * pp_stack_pp_size * i;
1863 }
1864
1865 lima_dump_command_stream_print(
1866 &pp_frame, sizeof(pp_frame), false, "add pp frame\n");
1867
1868 if (!lima_submit_start(ctx->pp_submit, &pp_frame, sizeof(pp_frame)))
1869 fprintf(stderr, "pp submit error\n");
1870 }
1871 else {
1872 struct drm_lima_m450_pp_frame pp_frame = {0};
1873 lima_pack_pp_frame_reg(ctx, pp_frame.frame, pp_frame.wb);
1874 pp_frame.num_pp = screen->num_pp;
1875
1876 if (ctx->pp_max_stack_size)
1877 for (int i = 0; i < screen->num_pp; i++)
1878 pp_frame.fragment_stack_address[i] = pp_stack_va +
1879 ctx->pp_max_stack_size * pp_stack_pp_size * i;
1880
1881 if (ps->bo) {
1882 for (int i = 0; i < screen->num_pp; i++)
1883 pp_frame.plbu_array_address[i] = ps->bo->va + ps->bo_offset + ps->offset[i];
1884 }
1885 else {
1886 pp_frame.use_dlbu = true;
1887
1888 struct lima_context_framebuffer *fb = &ctx->framebuffer;
1889 pp_frame.dlbu_regs[0] = ctx->plb[ctx->plb_index]->va;
1890 pp_frame.dlbu_regs[1] = ((fb->tiled_h - 1) << 16) | (fb->tiled_w - 1);
1891 unsigned s = util_logbase2(LIMA_CTX_PLB_BLK_SIZE) - 7;
1892 pp_frame.dlbu_regs[2] = (s << 28) | (fb->shift_h << 16) | fb->shift_w;
1893 pp_frame.dlbu_regs[3] = ((fb->tiled_h - 1) << 24) | ((fb->tiled_w - 1) << 16);
1894 }
1895
1896 lima_dump_command_stream_print(
1897 &pp_frame, sizeof(pp_frame), false, "add pp frame\n");
1898
1899 if (!lima_submit_start(ctx->pp_submit, &pp_frame, sizeof(pp_frame)))
1900 fprintf(stderr, "pp submit error\n");
1901 }
1902
1903 if (lima_dump_command_stream) {
1904 if (!lima_submit_wait(ctx->pp_submit, PIPE_TIMEOUT_INFINITE)) {
1905 fprintf(stderr, "pp wait error\n");
1906 exit(1);
1907 }
1908 }
1909
1910 ctx->plb_index = (ctx->plb_index + 1) % lima_ctx_num_plb;
1911
1912 if (ctx->framebuffer.base.nr_cbufs) {
1913 /* Set reload flag for next draw. It'll be unset if buffer is cleared */
1914 struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
1915 surf->reload = true;
1916 }
1917
1918 ctx->pp_max_stack_size = 0;
1919
1920 ctx->damage_rect.minx = ctx->damage_rect.miny = 0xffff;
1921 ctx->damage_rect.maxx = ctx->damage_rect.maxy = 0;
1922
1923 ctx->resolve = 0;
1924
1925 lima_dump_file_next();
1926 }
1927
1928 void
1929 lima_flush(struct lima_context *ctx)
1930 {
1931 if (!lima_ctx_dirty(ctx))
1932 return;
1933
1934 _lima_flush(ctx, false);
1935 }
1936
1937 static void
1938 lima_pipe_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
1939 unsigned flags)
1940 {
1941 struct lima_context *ctx = lima_context(pctx);
1942 if (lima_ctx_dirty(ctx))
1943 _lima_flush(ctx, flags & PIPE_FLUSH_END_OF_FRAME);
1944
1945 if (fence) {
1946 int fd;
1947 if (lima_submit_get_out_sync(ctx->pp_submit, &fd))
1948 *fence = lima_fence_create(fd);
1949 }
1950 }
1951
1952 void
1953 lima_draw_init(struct lima_context *ctx)
1954 {
1955 ctx->base.clear = lima_clear;
1956 ctx->base.draw_vbo = lima_draw_vbo;
1957 ctx->base.flush = lima_pipe_flush;
1958 }