1 /*
2 * Copyright 2008 Corbin Simpson <[email protected]>
3 * Copyright 2009 Marek Olšák <[email protected]>
4 * SPDX-License-Identifier: MIT
5 */
6
7 /* r300_emit: Functions for emitting state. */
8
9 #include "util/format/u_format.h"
10 #include "util/u_math.h"
11
12 #include "r300_context.h"
13 #include "r300_cb.h"
14 #include "r300_cs.h"
15 #include "r300_emit.h"
16 #include "r300_fs.h"
17 #include "r300_screen.h"
18 #include "r300_screen_buffer.h"
19 #include "r300_vs.h"
20
r300_emit_blend_state(struct r300_context * r300,unsigned size,void * state)21 void r300_emit_blend_state(struct r300_context* r300,
22 unsigned size, void* state)
23 {
24 struct r300_blend_state* blend = (struct r300_blend_state*)state;
25 struct pipe_framebuffer_state* fb =
26 (struct pipe_framebuffer_state*)r300->fb_state.state;
27 struct pipe_surface *cb;
28 CS_LOCALS(r300);
29
30 cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL;
31
32 if (cb) {
33 if (cb->format == PIPE_FORMAT_R16G16B16A16_FLOAT) {
34 WRITE_CS_TABLE(blend->cb_noclamp, size);
35 } else if (cb->format == PIPE_FORMAT_R16G16B16X16_FLOAT) {
36 WRITE_CS_TABLE(blend->cb_noclamp_noalpha, size);
37 } else {
38 unsigned swz = r300_surface(cb)->colormask_swizzle;
39 WRITE_CS_TABLE(blend->cb_clamp[swz], size);
40 }
41 } else {
42 WRITE_CS_TABLE(blend->cb_no_readwrite, size);
43 }
44 }
45
r300_emit_blend_color_state(struct r300_context * r300,unsigned size,void * state)46 void r300_emit_blend_color_state(struct r300_context* r300,
47 unsigned size, void* state)
48 {
49 struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state;
50 CS_LOCALS(r300);
51
52 WRITE_CS_TABLE(bc->cb, size);
53 }
54
r300_emit_clip_state(struct r300_context * r300,unsigned size,void * state)55 void r300_emit_clip_state(struct r300_context* r300,
56 unsigned size, void* state)
57 {
58 struct r300_clip_state* clip = (struct r300_clip_state*)state;
59 CS_LOCALS(r300);
60
61 WRITE_CS_TABLE(clip->cb, size);
62 }
63
r300_emit_dsa_state(struct r300_context * r300,unsigned size,void * state)64 void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state)
65 {
66 struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
67 struct pipe_framebuffer_state* fb =
68 (struct pipe_framebuffer_state*)r300->fb_state.state;
69 bool is_r500 = r300->screen->caps.is_r500;
70 CS_LOCALS(r300);
71 uint32_t alpha_func = dsa->alpha_function;
72
73 /* Choose the alpha ref value between 8-bit (FG_ALPHA_FUNC.AM_VAL) and
74 * 16-bit (FG_ALPHA_VALUE). */
75 if (is_r500 && (alpha_func & R300_FG_ALPHA_FUNC_ENABLE)) {
76 struct pipe_surface *cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL;
77
78 if (cb &&
79 (cb->format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
80 cb->format == PIPE_FORMAT_R16G16B16X16_FLOAT)) {
81 alpha_func |= R500_FG_ALPHA_FUNC_FP16_ENABLE;
82 } else {
83 alpha_func |= R500_FG_ALPHA_FUNC_8BIT;
84 }
85 }
86
87 /* Setup alpha-to-coverage. */
88 if (r300->alpha_to_coverage && r300->msaa_enable) {
89 /* Always set 3/6, it improves precision even for 2x and 4x MSAA. */
90 alpha_func |= R300_FG_ALPHA_FUNC_MASK_ENABLE |
91 R300_FG_ALPHA_FUNC_CFG_3_OF_6;
92 }
93
94 BEGIN_CS(size);
95 OUT_CS_REG(R300_FG_ALPHA_FUNC, alpha_func);
96 OUT_CS_TABLE(fb->zsbuf ? &dsa->cb_begin : dsa->cb_zb_no_readwrite, size-2);
97 END_CS;
98 }
99
get_rc_constant_state(float vec[4],struct r300_context * r300,struct rc_constant * constant)100 static void get_rc_constant_state(
101 float vec[4],
102 struct r300_context * r300,
103 struct rc_constant * constant)
104 {
105 struct r300_textures_state* texstate = r300->textures_state.state;
106 struct r300_resource *tex;
107
108 assert(constant->Type == RC_CONSTANT_STATE);
109
110 /* vec should either be (0, 0, 0, 1), which should be a relatively safe
111 * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
112 * state factors. */
113
114 switch (constant->u.State[0]) {
115 /* Factor for converting rectangle coords to
116 * normalized coords. Should only show up on non-r500. */
117 case RC_STATE_R300_TEXRECT_FACTOR:
118 tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture);
119 vec[0] = 1.0 / tex->tex.width0;
120 vec[1] = 1.0 / tex->tex.height0;
121 vec[2] = 0;
122 vec[3] = 1;
123 break;
124
125 case RC_STATE_R300_TEXSCALE_FACTOR:
126 tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture);
127 /* Add a small number to the texture size to work around rounding errors in hw. */
128 vec[0] = tex->b.width0 / (tex->tex.width0 + 0.001f);
129 vec[1] = tex->b.height0 / (tex->tex.height0 + 0.001f);
130 vec[2] = tex->b.depth0 / (tex->tex.depth0 + 0.001f);
131 vec[3] = 1;
132 break;
133
134 case RC_STATE_R300_VIEWPORT_SCALE:
135 vec[0] = r300->viewport.scale[0];
136 vec[1] = r300->viewport.scale[1];
137 vec[2] = r300->viewport.scale[2];
138 vec[3] = 1;
139 break;
140
141 case RC_STATE_R300_VIEWPORT_OFFSET:
142 vec[0] = r300->viewport.translate[0];
143 vec[1] = r300->viewport.translate[1];
144 vec[2] = r300->viewport.translate[2];
145 vec[3] = 1;
146 break;
147
148 default:
149 fprintf(stderr, "r300: Implementation error: "
150 "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
151 vec[0] = 0;
152 vec[1] = 0;
153 vec[2] = 0;
154 vec[3] = 1;
155 }
156 }
157
158 /* Convert a normal single-precision float into the 7.16 format
159 * used by the R300 fragment shader.
160 */
pack_float24(float f)161 uint32_t pack_float24(float f)
162 {
163 union {
164 float fl;
165 uint32_t u;
166 } u;
167 float mantissa;
168 int exponent;
169 uint32_t float24 = 0;
170
171 if (f == 0.0)
172 return 0;
173
174 u.fl = f;
175
176 mantissa = frexpf(f, &exponent);
177
178 /* Handle -ve */
179 if (mantissa < 0) {
180 float24 |= (1 << 23);
181 mantissa = mantissa * -1.0;
182 }
183 /* Handle exponent, bias of 63 */
184 exponent += 62;
185 float24 |= (exponent << 16);
186 /* Kill 7 LSB of mantissa */
187 float24 |= (u.u & 0x7FFFFF) >> 7;
188
189 return float24;
190 }
191
r300_emit_fs(struct r300_context * r300,unsigned size,void * state)192 void r300_emit_fs(struct r300_context* r300, unsigned size, void *state)
193 {
194 struct r300_fragment_shader *fs = r300_fs(r300);
195 CS_LOCALS(r300);
196
197 WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
198 }
199
r300_emit_fs_constants(struct r300_context * r300,unsigned size,void * state)200 void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
201 {
202 struct r300_fragment_shader *fs = r300_fs(r300);
203 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
204 unsigned count = fs->shader->externals_count;
205 unsigned i, j;
206 CS_LOCALS(r300);
207
208 if (count == 0)
209 return;
210
211 BEGIN_CS(size);
212 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count * 4);
213 if (buf->remap_table){
214 for (i = 0; i < count; i++) {
215 for (j = 0; j < 4; j++) {
216 unsigned swz = buf->remap_table[i].swizzle[j];
217 unsigned index = buf->remap_table[i].index[j];
218 if (index == -1)
219 OUT_CS(pack_float24(0.0f));
220 else {
221 OUT_CS(pack_float24(*(float*)&buf->ptr[index * 4 + swz]));
222 }
223 }
224 }
225 } else {
226 for (i = 0; i < count; i++)
227 for (j = 0; j < 4; j++)
228 OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j]));
229 }
230
231 END_CS;
232 }
233
r300_emit_fs_rc_constant_state(struct r300_context * r300,unsigned size,void * state)234 void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state)
235 {
236 struct r300_fragment_shader *fs = r300_fs(r300);
237 struct rc_constant_list *constants = &fs->shader->code.constants;
238 unsigned i;
239 unsigned count = fs->shader->rc_state_count;
240 unsigned first = fs->shader->externals_count;
241 unsigned end = constants->Count;
242 unsigned j;
243 CS_LOCALS(r300);
244
245 if (count == 0)
246 return;
247
248 BEGIN_CS(size);
249 for(i = first; i < end; ++i) {
250 if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
251 float data[4];
252
253 get_rc_constant_state(data, r300, &constants->Constants[i]);
254
255 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
256 for (j = 0; j < 4; j++)
257 OUT_CS(pack_float24(data[j]));
258 }
259 }
260 END_CS;
261 }
262
r500_emit_fs(struct r300_context * r300,unsigned size,void * state)263 void r500_emit_fs(struct r300_context* r300, unsigned size, void *state)
264 {
265 struct r300_fragment_shader *fs = r300_fs(r300);
266 CS_LOCALS(r300);
267
268 WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
269 }
270
r500_emit_fs_constants(struct r300_context * r300,unsigned size,void * state)271 void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
272 {
273 struct r300_fragment_shader *fs = r300_fs(r300);
274 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
275 unsigned count = fs->shader->externals_count;
276 CS_LOCALS(r300);
277
278 if (count == 0)
279 return;
280
281 BEGIN_CS(size);
282 OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
283 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count * 4);
284 if (buf->remap_table){
285 for (unsigned i = 0; i < count; i++) {
286 uint32_t data[4] = {};
287 for (unsigned chan = 0; chan < 4; chan++){
288 if (buf->remap_table[i].swizzle[chan] != RC_SWIZZLE_UNUSED)
289 data[chan] = buf->ptr[buf->remap_table[i].index[chan] * 4 + buf->remap_table[i].swizzle[chan]];
290 }
291 OUT_CS_TABLE(data, 4);
292 }
293 } else {
294 OUT_CS_TABLE(buf->ptr, count * 4);
295 }
296 END_CS;
297 }
298
r500_emit_fs_rc_constant_state(struct r300_context * r300,unsigned size,void * state)299 void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state)
300 {
301 struct r300_fragment_shader *fs = r300_fs(r300);
302 struct rc_constant_list *constants = &fs->shader->code.constants;
303 unsigned i;
304 unsigned count = fs->shader->rc_state_count;
305 unsigned first = fs->shader->externals_count;
306 unsigned end = constants->Count;
307 CS_LOCALS(r300);
308
309 if (count == 0)
310 return;
311
312 BEGIN_CS(size);
313 for(i = first; i < end; ++i) {
314 if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
315 float data[4];
316
317 get_rc_constant_state(data, r300, &constants->Constants[i]);
318
319 OUT_CS_REG(R500_GA_US_VECTOR_INDEX,
320 R500_GA_US_VECTOR_INDEX_TYPE_CONST |
321 (i & R500_GA_US_VECTOR_INDEX_MASK));
322 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
323 OUT_CS_TABLE(data, 4);
324 }
325 }
326 END_CS;
327 }
328
r300_emit_gpu_flush(struct r300_context * r300,unsigned size,void * state)329 void r300_emit_gpu_flush(struct r300_context *r300, unsigned size, void *state)
330 {
331 struct r300_gpu_flush *gpuflush = (struct r300_gpu_flush*)state;
332 struct pipe_framebuffer_state* fb =
333 (struct pipe_framebuffer_state*)r300->fb_state.state;
334 uint32_t height = fb->height;
335 uint32_t width = fb->width;
336 CS_LOCALS(r300);
337
338 if (r300->cbzb_clear) {
339 struct r300_surface *surf = r300_surface(fb->cbufs[0]);
340
341 height = surf->cbzb_height;
342 width = surf->cbzb_width;
343 }
344
345 DBG(r300, DBG_SCISSOR,
346 "r300: Scissor width: %i, height: %i, CBZB clear: %s\n",
347 width, height, r300->cbzb_clear ? "YES" : "NO");
348
349 BEGIN_CS(size);
350
351 /* Set up scissors.
352 * By writing to the SC registers, SC & US assert idle. */
353 OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
354 if (r300->screen->caps.is_r500) {
355 OUT_CS(0);
356 OUT_CS(((width - 1) << R300_SCISSORS_X_SHIFT) |
357 ((height - 1) << R300_SCISSORS_Y_SHIFT));
358 } else {
359 OUT_CS((1440 << R300_SCISSORS_X_SHIFT) |
360 (1440 << R300_SCISSORS_Y_SHIFT));
361 OUT_CS(((width + 1440-1) << R300_SCISSORS_X_SHIFT) |
362 ((height + 1440-1) << R300_SCISSORS_Y_SHIFT));
363 }
364
365 /* Flush CB & ZB caches and wait until the 3D engine is idle and clean. */
366 OUT_CS_TABLE(gpuflush->cb_flush_clean, 6);
367 END_CS;
368 }
369
r300_emit_aa_state(struct r300_context * r300,unsigned size,void * state)370 void r300_emit_aa_state(struct r300_context *r300, unsigned size, void *state)
371 {
372 struct r300_aa_state *aa = (struct r300_aa_state*)state;
373 CS_LOCALS(r300);
374
375 BEGIN_CS(size);
376 OUT_CS_REG(R300_GB_AA_CONFIG, aa->aa_config);
377
378 if (aa->dest) {
379 OUT_CS_REG_SEQ(R300_RB3D_AARESOLVE_OFFSET, 3);
380 OUT_CS(aa->dest->offset);
381 OUT_CS(aa->dest->pitch & R300_RB3D_AARESOLVE_PITCH_MASK);
382 OUT_CS(R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
383 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE);
384 OUT_CS_RELOC(aa->dest);
385 } else {
386 OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0);
387 }
388
389 END_CS;
390 }
391
r300_emit_fb_state(struct r300_context * r300,unsigned size,void * state)392 void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
393 {
394 struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
395 struct r300_surface* surf;
396 unsigned i;
397 uint32_t rb3d_cctl = 0;
398
399 CS_LOCALS(r300);
400
401 BEGIN_CS(size);
402
403 if (r300->screen->caps.is_r500) {
404 rb3d_cctl = R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE;
405 }
406 /* NUM_MULTIWRITES replicates COLOR[0] to all colorbuffers. */
407 if (fb->nr_cbufs && r300->fb_multiwrite) {
408 rb3d_cctl |= R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs);
409 }
410 if (r300->cmask_in_use) {
411 rb3d_cctl |= R300_RB3D_CCTL_AA_COMPRESSION_ENABLE |
412 R300_RB3D_CCTL_CMASK_ENABLE;
413 }
414
415 OUT_CS_REG(R300_RB3D_CCTL, rb3d_cctl);
416
417 /* Set up colorbuffers. */
418 for (i = 0; i < fb->nr_cbufs; i++) {
419 surf = r300_surface(r300_get_nonnull_cb(fb, i));
420
421 OUT_CS_REG(R300_RB3D_COLOROFFSET0 + (4 * i), surf->offset);
422 OUT_CS_RELOC(surf);
423
424 OUT_CS_REG(R300_RB3D_COLORPITCH0 + (4 * i), surf->pitch);
425 OUT_CS_RELOC(surf);
426
427 if (r300->cmask_in_use && i == 0) {
428 OUT_CS_REG(R300_RB3D_CMASK_OFFSET0, 0);
429 OUT_CS_REG(R300_RB3D_CMASK_PITCH0, surf->pitch_cmask);
430 OUT_CS_REG(R300_RB3D_COLOR_CLEAR_VALUE, r300->color_clear_value);
431 if (r300->screen->caps.is_r500) {
432 OUT_CS_REG_SEQ(R500_RB3D_COLOR_CLEAR_VALUE_AR, 2);
433 OUT_CS(r300->color_clear_value_ar);
434 OUT_CS(r300->color_clear_value_gb);
435 }
436 }
437 }
438
439 /* Set up the ZB part of the CBZB clear. */
440 if (r300->cbzb_clear) {
441 surf = r300_surface(fb->cbufs[0]);
442
443 OUT_CS_REG(R300_ZB_FORMAT, surf->cbzb_format);
444
445 OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->cbzb_midpoint_offset);
446 OUT_CS_RELOC(surf);
447
448 OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->cbzb_pitch);
449 OUT_CS_RELOC(surf);
450
451 DBG(r300, DBG_CBZB,
452 "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format,
453 surf->cbzb_pitch);
454 }
455 /* Set up a zbuffer. */
456 else if (fb->zsbuf) {
457 surf = r300_surface(fb->zsbuf);
458
459 OUT_CS_REG(R300_ZB_FORMAT, surf->format);
460
461 OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->offset);
462 OUT_CS_RELOC(surf);
463
464 OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->pitch);
465 OUT_CS_RELOC(surf);
466
467 if (r300->hyperz_enabled) {
468 /* HiZ RAM. */
469 OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
470 OUT_CS_REG(R300_ZB_HIZ_PITCH, surf->pitch_hiz);
471 /* Z Mask RAM. (compressed zbuffer) */
472 OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
473 OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf->pitch_zmask);
474 }
475 }
476
477 END_CS;
478 }
479
r300_emit_hyperz_state(struct r300_context * r300,unsigned size,void * state)480 void r300_emit_hyperz_state(struct r300_context *r300,
481 unsigned size, void *state)
482 {
483 struct r300_hyperz_state *z = state;
484 CS_LOCALS(r300);
485
486 if (z->flush)
487 WRITE_CS_TABLE(&z->cb_flush_begin, size);
488 else
489 WRITE_CS_TABLE(&z->cb_begin, size - 2);
490 }
491
r300_emit_hyperz_end(struct r300_context * r300)492 void r300_emit_hyperz_end(struct r300_context *r300)
493 {
494 struct r300_hyperz_state z =
495 *(struct r300_hyperz_state*)r300->hyperz_state.state;
496
497 z.flush = 1;
498 z.zb_bw_cntl = 0;
499 z.zb_depthclearvalue = 0;
500 z.sc_hyperz = R300_SC_HYPERZ_ADJ_2;
501 z.gb_z_peq_config = 0;
502
503 r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z);
504 }
505
506 #define R300_NIBBLES(x0, y0, x1, y1, x2, y2, d0y, d0x) \
507 (((x0) & 0xf) | (((y0) & 0xf) << 4) | \
508 (((x1) & 0xf) << 8) | (((y1) & 0xf) << 12) | \
509 (((x2) & 0xf) << 16) | (((y2) & 0xf) << 20) | \
510 (((d0y) & 0xf) << 24) | (((d0x) & 0xf) << 28))
511
r300_get_mspos(int index,unsigned * p)512 static unsigned r300_get_mspos(int index, unsigned *p)
513 {
514 unsigned reg, i, distx, disty, dist;
515
516 if (index == 0) {
517 /* MSPOS0 contains positions for samples 0,1,2 as (X,Y) pairs of nibbles,
518 * followed by a (Y,X) pair containing the minimum distance from the pixel
519 * edge:
520 * X0, Y0, X1, Y1, X2, Y2, D0_Y, D0_X
521 *
522 * There is a quirk when setting D0_X. The value represents the distance
523 * from the left edge of the pixel quad to the first sample in subpixels.
524 * All values less than eight should use the actual value, but „7‟ should
525 * be used for the distance „8‟. The hardware will convert 7 into 8 internally.
526 */
527 distx = 11;
528 for (i = 0; i < 12; i += 2) {
529 if (p[i] < distx)
530 distx = p[i];
531 }
532
533 disty = 11;
534 for (i = 1; i < 12; i += 2) {
535 if (p[i] < disty)
536 disty = p[i];
537 }
538
539 if (distx == 8)
540 distx = 7;
541
542 reg = R300_NIBBLES(p[0], p[1], p[2], p[3], p[4], p[5], disty, distx);
543 } else {
544 /* MSPOS1 contains positions for samples 3,4,5 as (X,Y) pairs of nibbles,
545 * followed by the minimum distance from the pixel edge (not sure if X or Y):
546 * X3, Y3, X4, Y4, X5, Y5, D1
547 */
548 dist = 11;
549 for (i = 0; i < 12; i++) {
550 if (p[i] < dist)
551 dist = p[i];
552 }
553
554 reg = R300_NIBBLES(p[6], p[7], p[8], p[9], p[10], p[11], dist, 0);
555 }
556 return reg;
557 }
558
r300_emit_fb_state_pipelined(struct r300_context * r300,unsigned size,void * state)559 void r300_emit_fb_state_pipelined(struct r300_context *r300,
560 unsigned size, void *state)
561 {
562 /* The sample coordinates are in the range [0,11], because
563 * GB_TILE_CONFIG.SUBPIXEL is set to the 1/12 subpixel precision.
564 *
565 * Some sample coordinates reach to neighboring pixels and should not be used.
566 * (e.g. Y=11)
567 *
568 * The unused samples must be set to the positions of other valid samples. */
569 static unsigned sample_locs_1x[12] = {
570 6,6, 6,6, 6,6, 6,6, 6,6, 6,6
571 };
572 static unsigned sample_locs_2x[12] = {
573 3,9, 9,3, 9,3, 9,3, 9,3, 9,3
574 };
575 static unsigned sample_locs_4x[12] = {
576 4,4, 8,8, 2,10, 10,2, 10,2, 10,2
577 };
578 static unsigned sample_locs_6x[12] = {
579 3,1, 7,3, 11,5, 1,7, 5,9, 9,10
580 };
581
582 struct pipe_framebuffer_state* fb =
583 (struct pipe_framebuffer_state*)r300->fb_state.state;
584 unsigned i, num_cbufs = fb->nr_cbufs;
585 unsigned mspos0, mspos1;
586 CS_LOCALS(r300);
587
588 /* If we use the multiwrite feature, the colorbuffers 2,3,4 must be
589 * marked as UNUSED in the US block. */
590 if (r300->fb_multiwrite) {
591 num_cbufs = MIN2(num_cbufs, 1);
592 }
593
594 BEGIN_CS(size);
595
596 /* Colorbuffer format in the US block.
597 * (must be written after unpipelined regs) */
598 OUT_CS_REG_SEQ(R300_US_OUT_FMT_0, 4);
599 for (i = 0; i < num_cbufs; i++) {
600 OUT_CS(r300_surface(r300_get_nonnull_cb(fb, i))->format);
601 }
602 for (; i < 1; i++) {
603 OUT_CS(R300_US_OUT_FMT_C4_8 |
604 R300_C0_SEL_B | R300_C1_SEL_G |
605 R300_C2_SEL_R | R300_C3_SEL_A);
606 }
607 for (; i < 4; i++) {
608 OUT_CS(R300_US_OUT_FMT_UNUSED);
609 }
610
611 /* Set sample positions. It depends on the framebuffer sample count.
612 * These are pipelined regs and as such cannot be moved to the AA state.
613 */
614 switch (r300->num_samples) {
615 default:
616 mspos0 = r300_get_mspos(0, sample_locs_1x);
617 mspos1 = r300_get_mspos(1, sample_locs_1x);
618 break;
619 case 2:
620 mspos0 = r300_get_mspos(0, sample_locs_2x);
621 mspos1 = r300_get_mspos(1, sample_locs_2x);
622 break;
623 case 4:
624 mspos0 = r300_get_mspos(0, sample_locs_4x);
625 mspos1 = r300_get_mspos(1, sample_locs_4x);
626 break;
627 case 6:
628 mspos0 = r300_get_mspos(0, sample_locs_6x);
629 mspos1 = r300_get_mspos(1, sample_locs_6x);
630 break;
631 }
632
633 OUT_CS_REG_SEQ(R300_GB_MSPOS0, 2);
634 OUT_CS(mspos0);
635 OUT_CS(mspos1);
636 END_CS;
637 }
638
r300_emit_query_start(struct r300_context * r300,unsigned size,void * state)639 void r300_emit_query_start(struct r300_context *r300, unsigned size, void*state)
640 {
641 struct r300_query *query = r300->query_current;
642 CS_LOCALS(r300);
643
644 if (!query)
645 return;
646
647 BEGIN_CS(size);
648 if (r300->screen->caps.family == CHIP_RV530) {
649 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
650 } else {
651 OUT_CS_REG(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL);
652 }
653 OUT_CS_REG(R300_ZB_ZPASS_DATA, 0);
654 END_CS;
655 query->begin_emitted = true;
656 }
657
r300_emit_query_end_frag_pipes(struct r300_context * r300,struct r300_query * query)658 static void r300_emit_query_end_frag_pipes(struct r300_context *r300,
659 struct r300_query *query)
660 {
661 struct r300_capabilities* caps = &r300->screen->caps;
662 uint32_t gb_pipes = r300->screen->info.r300_num_gb_pipes;
663 CS_LOCALS(r300);
664
665 assert(gb_pipes);
666
667 BEGIN_CS(6 * gb_pipes + 2);
668 /* I'm not so sure I like this switch, but it's hard to be elegant
669 * when there's so many special cases...
670 *
671 * So here's the basic idea. For each pipe, enable writes to it only,
672 * then put out the relocation for ZPASS_ADDR, taking into account a
673 * 4-byte offset for each pipe. RV380 and older are special; they have
674 * only two pipes, and the second pipe's enable is on bit 3, not bit 1,
675 * so there's a chipset cap for that. */
676 switch (gb_pipes) {
677 case 4:
678 /* pipe 3 only */
679 OUT_CS_REG(R300_SU_REG_DEST, 1 << 3);
680 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 3) * 4);
681 OUT_CS_RELOC(r300->query_current);
682 FALLTHROUGH;
683 case 3:
684 /* pipe 2 only */
685 OUT_CS_REG(R300_SU_REG_DEST, 1 << 2);
686 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 2) * 4);
687 OUT_CS_RELOC(r300->query_current);
688 FALLTHROUGH;
689 case 2:
690 /* pipe 1 only */
691 /* As mentioned above, accommodate RV380 and older. */
692 OUT_CS_REG(R300_SU_REG_DEST,
693 1 << (caps->high_second_pipe ? 3 : 1));
694 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4);
695 OUT_CS_RELOC(r300->query_current);
696 FALLTHROUGH;
697 case 1:
698 /* pipe 0 only */
699 OUT_CS_REG(R300_SU_REG_DEST, 1 << 0);
700 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4);
701 OUT_CS_RELOC(r300->query_current);
702 break;
703 default:
704 fprintf(stderr, "r300: Implementation error: Chipset reports %d"
705 " pixel pipes!\n", gb_pipes);
706 abort();
707 }
708
709 /* And, finally, reset it to normal... */
710 OUT_CS_REG(R300_SU_REG_DEST, 0xF);
711 END_CS;
712 }
713
rv530_emit_query_end_single_z(struct r300_context * r300,struct r300_query * query)714 static void rv530_emit_query_end_single_z(struct r300_context *r300,
715 struct r300_query *query)
716 {
717 CS_LOCALS(r300);
718
719 BEGIN_CS(8);
720 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
721 OUT_CS_REG(R300_ZB_ZPASS_ADDR, query->num_results * 4);
722 OUT_CS_RELOC(r300->query_current);
723 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
724 END_CS;
725 }
726
rv530_emit_query_end_double_z(struct r300_context * r300,struct r300_query * query)727 static void rv530_emit_query_end_double_z(struct r300_context *r300,
728 struct r300_query *query)
729 {
730 CS_LOCALS(r300);
731
732 BEGIN_CS(14);
733 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
734 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4);
735 OUT_CS_RELOC(r300->query_current);
736 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1);
737 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4);
738 OUT_CS_RELOC(r300->query_current);
739 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
740 END_CS;
741 }
742
r300_emit_query_end(struct r300_context * r300)743 void r300_emit_query_end(struct r300_context* r300)
744 {
745 struct r300_capabilities *caps = &r300->screen->caps;
746 struct r300_query *query = r300->query_current;
747
748 if (!query)
749 return;
750
751 if (query->begin_emitted == false)
752 return;
753
754 if (caps->family == CHIP_RV530) {
755 if (r300->screen->info.r300_num_z_pipes == 2)
756 rv530_emit_query_end_double_z(r300, query);
757 else
758 rv530_emit_query_end_single_z(r300, query);
759 } else
760 r300_emit_query_end_frag_pipes(r300, query);
761
762 query->begin_emitted = false;
763 query->num_results += query->num_pipes;
764
765 /* XXX grab all the results and reset the counter. */
766 if (query->num_results >= query->buf->size / 4 - 4) {
767 query->num_results = (query->buf->size / 4) / 2;
768 fprintf(stderr, "r300: Rewinding OQBO...\n");
769 }
770 }
771
r300_emit_invariant_state(struct r300_context * r300,unsigned size,void * state)772 void r300_emit_invariant_state(struct r300_context *r300,
773 unsigned size, void *state)
774 {
775 CS_LOCALS(r300);
776 WRITE_CS_TABLE(state, size);
777 }
778
r300_emit_rs_state(struct r300_context * r300,unsigned size,void * state)779 void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state)
780 {
781 struct r300_rs_state* rs = state;
782 CS_LOCALS(r300);
783
784 BEGIN_CS(size);
785 OUT_CS_TABLE(rs->cb_main, RS_STATE_MAIN_SIZE);
786 if (rs->polygon_offset_enable) {
787 if (r300->zbuffer_bpp == 16) {
788 OUT_CS_TABLE(rs->cb_poly_offset_zb16, 5);
789 } else {
790 OUT_CS_TABLE(rs->cb_poly_offset_zb24, 5);
791 }
792 }
793 END_CS;
794 }
795
r300_emit_rs_block_state(struct r300_context * r300,unsigned size,void * state)796 void r300_emit_rs_block_state(struct r300_context* r300,
797 unsigned size, void* state)
798 {
799 struct r300_rs_block* rs = (struct r300_rs_block*)state;
800 unsigned i;
801 /* It's the same for both INST and IP tables */
802 unsigned count = (rs->inst_count & R300_RS_INST_COUNT_MASK) + 1;
803 CS_LOCALS(r300);
804
805 if (DBG_ON(r300, DBG_RS_BLOCK)) {
806 r500_dump_rs_block(rs);
807
808 fprintf(stderr, "r300: RS emit:\n");
809
810 for (i = 0; i < count; i++)
811 fprintf(stderr, " : ip %d: 0x%08x\n", i, rs->ip[i]);
812
813 for (i = 0; i < count; i++)
814 fprintf(stderr, " : inst %d: 0x%08x\n", i, rs->inst[i]);
815
816 fprintf(stderr, " : count: 0x%08x inst_count: 0x%08x\n",
817 rs->count, rs->inst_count);
818 }
819
820 BEGIN_CS(size);
821 OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
822 OUT_CS(rs->vap_vtx_state_cntl);
823 OUT_CS(rs->vap_vsm_vtx_assm);
824 OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
825 OUT_CS(rs->vap_out_vtx_fmt[0]);
826 OUT_CS(rs->vap_out_vtx_fmt[1]);
827 OUT_CS_REG_SEQ(R300_GB_ENABLE, 1);
828 OUT_CS(rs->gb_enable);
829
830 if (r300->screen->caps.is_r500) {
831 OUT_CS_REG_SEQ(R500_RS_IP_0, count);
832 } else {
833 OUT_CS_REG_SEQ(R300_RS_IP_0, count);
834 }
835 OUT_CS_TABLE(rs->ip, count);
836
837 OUT_CS_REG_SEQ(R300_RS_COUNT, 2);
838 OUT_CS(rs->count);
839 OUT_CS(rs->inst_count);
840
841 if (r300->screen->caps.is_r500) {
842 OUT_CS_REG_SEQ(R500_RS_INST_0, count);
843 } else {
844 OUT_CS_REG_SEQ(R300_RS_INST_0, count);
845 }
846 OUT_CS_TABLE(rs->inst, count);
847 END_CS;
848 }
849
r300_emit_sample_mask(struct r300_context * r300,unsigned size,void * state)850 void r300_emit_sample_mask(struct r300_context *r300,
851 unsigned size, void *state)
852 {
853 unsigned mask = (*(unsigned*)state) & ((1 << 6)-1);
854 CS_LOCALS(r300);
855
856 BEGIN_CS(size);
857 OUT_CS_REG(R300_SC_SCREENDOOR,
858 mask | (mask << 6) | (mask << 12) | (mask << 18));
859 END_CS;
860 }
861
r300_emit_scissor_state(struct r300_context * r300,unsigned size,void * state)862 void r300_emit_scissor_state(struct r300_context* r300,
863 unsigned size, void* state)
864 {
865 struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state;
866 CS_LOCALS(r300);
867
868 BEGIN_CS(size);
869 OUT_CS_REG_SEQ(R300_SC_CLIPRECT_TL_0, 2);
870 if (r300->screen->caps.is_r500) {
871 OUT_CS((scissor->minx << R300_CLIPRECT_X_SHIFT) |
872 (scissor->miny << R300_CLIPRECT_Y_SHIFT));
873 OUT_CS(((scissor->maxx - 1) << R300_CLIPRECT_X_SHIFT) |
874 ((scissor->maxy - 1) << R300_CLIPRECT_Y_SHIFT));
875 } else {
876 OUT_CS(((scissor->minx + 1440) << R300_CLIPRECT_X_SHIFT) |
877 ((scissor->miny + 1440) << R300_CLIPRECT_Y_SHIFT));
878 OUT_CS(((scissor->maxx + 1440-1) << R300_CLIPRECT_X_SHIFT) |
879 ((scissor->maxy + 1440-1) << R300_CLIPRECT_Y_SHIFT));
880 }
881 END_CS;
882 }
883
r300_emit_textures_state(struct r300_context * r300,unsigned size,void * state)884 void r300_emit_textures_state(struct r300_context *r300,
885 unsigned size, void *state)
886 {
887 struct r300_textures_state *allstate = (struct r300_textures_state*)state;
888 struct r300_texture_sampler_state *texstate;
889 struct r300_resource *tex;
890 unsigned i;
891 bool has_us_format = r300->screen->caps.has_us_format;
892 CS_LOCALS(r300);
893
894 BEGIN_CS(size);
895 OUT_CS_REG(R300_TX_ENABLE, allstate->tx_enable);
896
897 for (i = 0; i < allstate->count; i++) {
898 if ((1 << i) & allstate->tx_enable) {
899 texstate = &allstate->regs[i];
900 tex = r300_resource(allstate->sampler_views[i]->base.texture);
901
902 OUT_CS_REG(R300_TX_FILTER0_0 + (i * 4), texstate->filter0);
903 OUT_CS_REG(R300_TX_FILTER1_0 + (i * 4), texstate->filter1);
904 OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (i * 4),
905 texstate->border_color);
906
907 OUT_CS_REG(R300_TX_FORMAT0_0 + (i * 4), texstate->format.format0);
908 OUT_CS_REG(R300_TX_FORMAT1_0 + (i * 4), texstate->format.format1);
909 OUT_CS_REG(R300_TX_FORMAT2_0 + (i * 4), texstate->format.format2);
910
911 OUT_CS_REG(R300_TX_OFFSET_0 + (i * 4), texstate->format.tile_config);
912 OUT_CS_RELOC(tex);
913
914 if (has_us_format) {
915 OUT_CS_REG(R500_US_FORMAT0_0 + (i * 4),
916 texstate->format.us_format0);
917 }
918 }
919 }
920 END_CS;
921 }
922
r300_emit_vertex_arrays(struct r300_context * r300,int offset,bool indexed,int instance_id)923 void r300_emit_vertex_arrays(struct r300_context* r300, int offset,
924 bool indexed, int instance_id)
925 {
926 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
927 struct pipe_vertex_element *velem = r300->velems->velem;
928 struct r300_resource *buf;
929 int i;
930 unsigned vertex_array_count = r300->velems->count;
931 unsigned packet_size = (vertex_array_count * 3 + 1) / 2;
932 struct pipe_vertex_buffer *vb1, *vb2;
933 unsigned *hw_format_size = r300->velems->format_size;
934 unsigned size1, size2, offset1, offset2, stride1, stride2;
935 CS_LOCALS(r300);
936
937 BEGIN_CS(2 + packet_size + vertex_array_count * 2);
938 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
939 OUT_CS(vertex_array_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
940
941 if (instance_id == -1) {
942 /* Non-instanced arrays. This ignores instance_divisor and instance_id. */
943 for (i = 0; i < vertex_array_count - 1; i += 2) {
944 vb1 = &vbuf[velem[i].vertex_buffer_index];
945 vb2 = &vbuf[velem[i+1].vertex_buffer_index];
946 size1 = hw_format_size[i];
947 size2 = hw_format_size[i+1];
948
949 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(velem[i].src_stride) |
950 R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(velem[i+1].src_stride));
951 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * velem[i].src_stride);
952 OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * velem[i+1].src_stride);
953 }
954
955 if (vertex_array_count & 1) {
956 vb1 = &vbuf[velem[i].vertex_buffer_index];
957 size1 = hw_format_size[i];
958
959 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(velem[i].src_stride));
960 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * velem[i].src_stride);
961 }
962
963 for (i = 0; i < vertex_array_count; i++) {
964 buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer.resource);
965 OUT_CS_RELOC(buf);
966 }
967 } else {
968 /* Instanced arrays. */
969 for (i = 0; i < vertex_array_count - 1; i += 2) {
970 vb1 = &vbuf[velem[i].vertex_buffer_index];
971 vb2 = &vbuf[velem[i+1].vertex_buffer_index];
972 size1 = hw_format_size[i];
973 size2 = hw_format_size[i+1];
974
975 if (velem[i].instance_divisor) {
976 stride1 = 0;
977 offset1 = vb1->buffer_offset + velem[i].src_offset +
978 (instance_id / velem[i].instance_divisor) * velem[i].src_stride;
979 } else {
980 stride1 = velem[i].src_stride;
981 offset1 = vb1->buffer_offset + velem[i].src_offset + offset * velem[i].src_stride;
982 }
983 if (velem[i+1].instance_divisor) {
984 stride2 = 0;
985 offset2 = vb2->buffer_offset + velem[i+1].src_offset +
986 (instance_id / velem[i+1].instance_divisor) * velem[i+1].src_stride;
987 } else {
988 stride2 = velem[i+1].src_stride;
989 offset2 = vb2->buffer_offset + velem[i+1].src_offset + offset * velem[i+1].src_stride;
990 }
991
992 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1) |
993 R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(stride2));
994 OUT_CS(offset1);
995 OUT_CS(offset2);
996 }
997
998 if (vertex_array_count & 1) {
999 vb1 = &vbuf[velem[i].vertex_buffer_index];
1000 size1 = hw_format_size[i];
1001
1002 if (velem[i].instance_divisor) {
1003 stride1 = 0;
1004 offset1 = vb1->buffer_offset + velem[i].src_offset +
1005 (instance_id / velem[i].instance_divisor) * velem[i].src_stride;
1006 } else {
1007 stride1 = velem[i].src_stride;
1008 offset1 = vb1->buffer_offset + velem[i].src_offset + offset * velem[i].src_stride;
1009 }
1010
1011 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1));
1012 OUT_CS(offset1);
1013 }
1014
1015 for (i = 0; i < vertex_array_count; i++) {
1016 buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer.resource);
1017 OUT_CS_RELOC(buf);
1018 }
1019 }
1020 END_CS;
1021 }
1022
r300_emit_vertex_arrays_swtcl(struct r300_context * r300,bool indexed)1023 void r300_emit_vertex_arrays_swtcl(struct r300_context *r300, bool indexed)
1024 {
1025 CS_LOCALS(r300);
1026
1027 DBG(r300, DBG_SWTCL, "r300: Preparing vertex buffer %p for render, "
1028 "vertex size %d\n", r300->vbo,
1029 r300->vertex_info.size);
1030 /* Set the pointer to our vertex buffer. The emitted values are this:
1031 * PACKET3 [3D_LOAD_VBPNTR]
1032 * COUNT [1]
1033 * FORMAT [size | stride << 8]
1034 * OFFSET [offset into BO]
1035 * VBPNTR [relocated BO]
1036 */
1037 BEGIN_CS(7);
1038 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3);
1039 OUT_CS(1 | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
1040 OUT_CS(r300->vertex_info.size |
1041 (r300->vertex_info.size << 8));
1042 OUT_CS(r300->draw_vbo_offset);
1043 OUT_CS(0);
1044
1045 assert(r300->vbo);
1046 OUT_CS(0xc0001000); /* PKT3_NOP */
1047 OUT_CS(r300->rws->cs_lookup_buffer(&r300->cs, r300->vbo) * 4);
1048 END_CS;
1049 }
1050
r300_emit_vertex_stream_state(struct r300_context * r300,unsigned size,void * state)1051 void r300_emit_vertex_stream_state(struct r300_context* r300,
1052 unsigned size, void* state)
1053 {
1054 struct r300_vertex_stream_state *streams =
1055 (struct r300_vertex_stream_state*)state;
1056 unsigned i;
1057 CS_LOCALS(r300);
1058
1059 if (DBG_ON(r300, DBG_PSC)) {
1060 fprintf(stderr, "r300: PSC emit:\n");
1061
1062 for (i = 0; i < streams->count; i++) {
1063 fprintf(stderr, " : prog_stream_cntl%d: 0x%08x\n", i,
1064 streams->vap_prog_stream_cntl[i]);
1065 }
1066
1067 for (i = 0; i < streams->count; i++) {
1068 fprintf(stderr, " : prog_stream_cntl_ext%d: 0x%08x\n", i,
1069 streams->vap_prog_stream_cntl_ext[i]);
1070 }
1071 }
1072
1073 BEGIN_CS(size);
1074 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count);
1075 OUT_CS_TABLE(streams->vap_prog_stream_cntl, streams->count);
1076 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count);
1077 OUT_CS_TABLE(streams->vap_prog_stream_cntl_ext, streams->count);
1078 END_CS;
1079 }
1080
r300_emit_pvs_flush(struct r300_context * r300,unsigned size,void * state)1081 void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state)
1082 {
1083 CS_LOCALS(r300);
1084
1085 BEGIN_CS(size);
1086 OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
1087 END_CS;
1088 }
1089
r300_emit_vap_invariant_state(struct r300_context * r300,unsigned size,void * state)1090 void r300_emit_vap_invariant_state(struct r300_context *r300,
1091 unsigned size, void *state)
1092 {
1093 CS_LOCALS(r300);
1094 WRITE_CS_TABLE(state, size);
1095 }
1096
r300_emit_vs_state(struct r300_context * r300,unsigned size,void * state)1097 void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
1098 {
1099 struct r300_vertex_shader_code* vs = ((struct r300_vertex_shader*)state)->shader;
1100 struct r300_vertex_program_code* code = &vs->code;
1101 struct r300_screen* r300screen = r300->screen;
1102 unsigned instruction_count = code->length / 4;
1103
1104 unsigned vtx_mem_size = r300screen->caps.is_r500 ? 128 : 72;
1105 unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1);
1106 unsigned output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
1107 unsigned temp_count = MAX2(code->num_temporaries, 1);
1108
1109 unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count,
1110 vtx_mem_size / output_count, 10);
1111 unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5);
1112
1113 CS_LOCALS(r300);
1114
1115 BEGIN_CS(size);
1116
1117 /* R300_VAP_PVS_CODE_CNTL_0
1118 * R300_VAP_PVS_CONST_CNTL
1119 * R300_VAP_PVS_CODE_CNTL_1
1120 * See the r5xx docs for instructions on how to use these. */
1121 OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_0, R300_PVS_FIRST_INST(0) |
1122 R300_PVS_XYZW_VALID_INST(code->last_pos_write) |
1123 R300_PVS_LAST_INST(instruction_count - 1));
1124 OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_1, code->last_input_read);
1125
1126 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
1127 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length);
1128 OUT_CS_TABLE(code->body.d, code->length);
1129
1130 OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
1131 R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
1132 R300_PVS_NUM_FPUS(r300screen->caps.num_vert_fpus) |
1133 R300_PVS_VF_MAX_VTX_NUM(12) |
1134 (r300->clip_halfz ? R300_DX_CLIP_SPACE_DEF : 0) |
1135 (r300screen->caps.is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
1136
1137 /* Emit flow control instructions. Even if there are no fc instructions,
1138 * we still need to write the registers to make sure they are cleared. */
1139 OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
1140 if (r300screen->caps.is_r500) {
1141 OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, R300_VS_MAX_FC_OPS * 2);
1142 OUT_CS_TABLE(code->fc_op_addrs.r500, R300_VS_MAX_FC_OPS * 2);
1143 } else {
1144 OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, R300_VS_MAX_FC_OPS);
1145 OUT_CS_TABLE(code->fc_op_addrs.r300, R300_VS_MAX_FC_OPS);
1146 }
1147 OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, R300_VS_MAX_FC_OPS);
1148 OUT_CS_TABLE(code->fc_loop_index, R300_VS_MAX_FC_OPS);
1149
1150 END_CS;
1151 }
1152
r300_emit_vs_constants(struct r300_context * r300,unsigned size,void * state)1153 void r300_emit_vs_constants(struct r300_context* r300,
1154 unsigned size, void *state)
1155 {
1156 unsigned count = r300_vs(r300)->shader->externals_count;
1157 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
1158 struct r300_vertex_shader_code *vs = r300_vs(r300)->shader;
1159 unsigned i;
1160 int imm_first = vs->externals_count;
1161 int imm_end = vs->code.constants.Count;
1162 int imm_count = vs->immediates_count;
1163 CS_LOCALS(r300);
1164
1165 BEGIN_CS(size);
1166 OUT_CS_REG(R300_VAP_PVS_CONST_CNTL,
1167 R300_PVS_CONST_BASE_OFFSET(buf->buffer_base) |
1168 R300_PVS_MAX_CONST_ADDR(MAX2(imm_end - 1, 0)));
1169 if (vs->externals_count) {
1170 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
1171 (r300->screen->caps.is_r500 ?
1172 R500_PVS_CONST_START : R300_PVS_CONST_START) + buf->buffer_base);
1173 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
1174 if (buf->remap_table){
1175 uint32_t *data = buf->ptr;
1176 for (i = 0; i < count; i++) {
1177 uint32_t constant[4];
1178 for (unsigned chan = 0; chan < 4; chan++) {
1179 constant[chan] = data[buf->remap_table[i].index[chan] * 4 +
1180 buf->remap_table[i].swizzle[chan]];
1181 }
1182 OUT_CS_TABLE(constant, 4);
1183 }
1184 } else {
1185 OUT_CS_TABLE(buf->ptr, count * 4);
1186 }
1187 }
1188
1189 /* Emit immediates. */
1190 if (imm_count) {
1191 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
1192 (r300->screen->caps.is_r500 ?
1193 R500_PVS_CONST_START : R300_PVS_CONST_START) +
1194 buf->buffer_base + imm_first);
1195 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, imm_count * 4);
1196 for (i = imm_first; i < imm_end; i++) {
1197 const float *data = vs->code.constants.Constants[i].u.Immediate;
1198 OUT_CS_TABLE(data, 4);
1199 }
1200 }
1201 END_CS;
1202 }
1203
r300_emit_viewport_state(struct r300_context * r300,unsigned size,void * state)1204 void r300_emit_viewport_state(struct r300_context* r300,
1205 unsigned size, void* state)
1206 {
1207 struct r300_viewport_state* viewport = (struct r300_viewport_state*)state;
1208 CS_LOCALS(r300);
1209
1210 BEGIN_CS(size);
1211 OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
1212 OUT_CS_TABLE(&viewport->xscale, 6);
1213 OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
1214 END_CS;
1215 }
1216
r300_emit_hiz_clear(struct r300_context * r300,unsigned size,void * state)1217 void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
1218 {
1219 struct pipe_framebuffer_state *fb =
1220 (struct pipe_framebuffer_state*)r300->fb_state.state;
1221 struct r300_resource* tex;
1222 CS_LOCALS(r300);
1223
1224 tex = r300_resource(fb->zsbuf->texture);
1225
1226 BEGIN_CS(size);
1227 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
1228 OUT_CS(0);
1229 OUT_CS(tex->tex.hiz_dwords[fb->zsbuf->u.tex.level]);
1230 OUT_CS(r300->hiz_clear_value);
1231 END_CS;
1232
1233 /* Mark the current zbuffer's hiz ram as in use. */
1234 r300->hiz_in_use = true;
1235 r300->hiz_func = HIZ_FUNC_NONE;
1236 r300_mark_atom_dirty(r300, &r300->hyperz_state);
1237 }
1238
r300_emit_zmask_clear(struct r300_context * r300,unsigned size,void * state)1239 void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
1240 {
1241 struct pipe_framebuffer_state *fb =
1242 (struct pipe_framebuffer_state*)r300->fb_state.state;
1243 struct r300_resource *tex;
1244 CS_LOCALS(r300);
1245
1246 tex = r300_resource(fb->zsbuf->texture);
1247
1248 BEGIN_CS(size);
1249 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2);
1250 OUT_CS(0);
1251 OUT_CS(tex->tex.zmask_dwords[fb->zsbuf->u.tex.level]);
1252 OUT_CS(0);
1253 END_CS;
1254
1255 /* Mark the current zbuffer's zmask as in use. */
1256 r300->zmask_in_use = true;
1257 r300_mark_atom_dirty(r300, &r300->hyperz_state);
1258 }
1259
r300_emit_cmask_clear(struct r300_context * r300,unsigned size,void * state)1260 void r300_emit_cmask_clear(struct r300_context *r300, unsigned size, void *state)
1261 {
1262 struct pipe_framebuffer_state *fb =
1263 (struct pipe_framebuffer_state*)r300->fb_state.state;
1264 struct r300_resource *tex;
1265 CS_LOCALS(r300);
1266
1267 tex = r300_resource(fb->cbufs[0]->texture);
1268
1269 BEGIN_CS(size);
1270 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_CMASK, 2);
1271 OUT_CS(0);
1272 OUT_CS(tex->tex.cmask_dwords);
1273 OUT_CS(0);
1274 END_CS;
1275
1276 /* Mark the current zbuffer's zmask as in use. */
1277 r300->cmask_in_use = true;
1278 r300_mark_fb_state_dirty(r300, R300_CHANGED_CMASK_ENABLE);
1279 }
1280
r300_emit_ztop_state(struct r300_context * r300,unsigned size,void * state)1281 void r300_emit_ztop_state(struct r300_context* r300,
1282 unsigned size, void* state)
1283 {
1284 struct r300_ztop_state* ztop = (struct r300_ztop_state*)state;
1285 CS_LOCALS(r300);
1286
1287 BEGIN_CS(size);
1288 OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top);
1289 END_CS;
1290 }
1291
r300_emit_texture_cache_inval(struct r300_context * r300,unsigned size,void * state)1292 void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state)
1293 {
1294 CS_LOCALS(r300);
1295
1296 BEGIN_CS(size);
1297 OUT_CS_REG(R300_TX_INVALTAGS, 0);
1298 END_CS;
1299 }
1300
r300_emit_buffer_validate(struct r300_context * r300,bool do_validate_vertex_buffers,struct pipe_resource * index_buffer)1301 bool r300_emit_buffer_validate(struct r300_context *r300,
1302 bool do_validate_vertex_buffers,
1303 struct pipe_resource *index_buffer)
1304 {
1305 struct pipe_framebuffer_state *fb =
1306 (struct pipe_framebuffer_state*)r300->fb_state.state;
1307 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
1308 struct r300_textures_state *texstate =
1309 (struct r300_textures_state*)r300->textures_state.state;
1310 struct r300_resource *tex;
1311 unsigned i;
1312 bool flushed = false;
1313
1314 validate:
1315 if (r300->fb_state.dirty) {
1316 /* Color buffers... */
1317 for (i = 0; i < fb->nr_cbufs; i++) {
1318 if (!fb->cbufs[i])
1319 continue;
1320 tex = r300_resource(fb->cbufs[i]->texture);
1321 assert(tex && tex->buf && "cbuf is marked, but NULL!");
1322 r300->rws->cs_add_buffer(&r300->cs, tex->buf,
1323 RADEON_USAGE_READWRITE | RADEON_USAGE_SYNCHRONIZED |
1324 (tex->b.nr_samples > 1 ?
1325 RADEON_PRIO_COLOR_BUFFER_MSAA :
1326 RADEON_PRIO_COLOR_BUFFER),
1327 r300_surface(fb->cbufs[i])->domain);
1328 }
1329 /* ...depth buffer... */
1330 if (fb->zsbuf) {
1331 tex = r300_resource(fb->zsbuf->texture);
1332 assert(tex && tex->buf && "zsbuf is marked, but NULL!");
1333 r300->rws->cs_add_buffer(&r300->cs, tex->buf,
1334 RADEON_USAGE_READWRITE | RADEON_USAGE_SYNCHRONIZED |
1335 (tex->b.nr_samples > 1 ?
1336 RADEON_PRIO_DEPTH_BUFFER_MSAA :
1337 RADEON_PRIO_DEPTH_BUFFER),
1338 r300_surface(fb->zsbuf)->domain);
1339 }
1340 }
1341 /* The AA resolve buffer. */
1342 if (r300->aa_state.dirty) {
1343 if (aa->dest) {
1344 r300->rws->cs_add_buffer(&r300->cs, aa->dest->buf,
1345 RADEON_USAGE_WRITE | RADEON_USAGE_SYNCHRONIZED |
1346 RADEON_PRIO_COLOR_BUFFER,
1347 aa->dest->domain);
1348 }
1349 }
1350 if (r300->textures_state.dirty) {
1351 /* ...textures... */
1352 for (i = 0; i < texstate->count; i++) {
1353 if (!(texstate->tx_enable & (1U << i))) {
1354 continue;
1355 }
1356
1357 tex = r300_resource(texstate->sampler_views[i]->base.texture);
1358 r300->rws->cs_add_buffer(&r300->cs, tex->buf,
1359 RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1360 RADEON_PRIO_SAMPLER_TEXTURE,
1361 tex->domain);
1362 }
1363 }
1364 /* ...occlusion query buffer... */
1365 if (r300->query_current)
1366 r300->rws->cs_add_buffer(&r300->cs, r300->query_current->buf,
1367 RADEON_USAGE_WRITE | RADEON_USAGE_SYNCHRONIZED |
1368 RADEON_PRIO_QUERY,
1369 RADEON_DOMAIN_GTT);
1370 /* ...vertex buffer for SWTCL path... */
1371 if (r300->vbo)
1372 r300->rws->cs_add_buffer(&r300->cs, r300->vbo,
1373 RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1374 RADEON_PRIO_VERTEX_BUFFER,
1375 RADEON_DOMAIN_GTT);
1376 /* ...vertex buffers for HWTCL path... */
1377 if (do_validate_vertex_buffers && r300->vertex_arrays_dirty) {
1378 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
1379 struct pipe_vertex_buffer *last = r300->vertex_buffer +
1380 r300->nr_vertex_buffers;
1381 struct pipe_resource *buf;
1382
1383 for (; vbuf != last; vbuf++) {
1384 buf = vbuf->buffer.resource;
1385 if (!buf)
1386 continue;
1387
1388 r300->rws->cs_add_buffer(&r300->cs, r300_resource(buf)->buf,
1389 RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1390 RADEON_PRIO_SAMPLER_BUFFER,
1391 r300_resource(buf)->domain);
1392 }
1393 }
1394 /* ...and index buffer for HWTCL path. */
1395 if (index_buffer)
1396 r300->rws->cs_add_buffer(&r300->cs, r300_resource(index_buffer)->buf,
1397 RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1398 RADEON_PRIO_INDEX_BUFFER,
1399 r300_resource(index_buffer)->domain);
1400
1401 /* Now do the validation (flush is called inside cs_validate on failure). */
1402 if (!r300->rws->cs_validate(&r300->cs)) {
1403 /* Ooops, an infinite loop, give up. */
1404 if (flushed)
1405 return false;
1406
1407 flushed = true;
1408 goto validate;
1409 }
1410
1411 return true;
1412 }
1413
r300_get_num_dirty_dwords(struct r300_context * r300)1414 unsigned r300_get_num_dirty_dwords(struct r300_context *r300)
1415 {
1416 struct r300_atom* atom;
1417 unsigned dwords = 0;
1418
1419 foreach_dirty_atom(r300, atom) {
1420 if (atom->dirty) {
1421 dwords += atom->size;
1422 }
1423 }
1424
1425 /* let's reserve some more, just in case */
1426 dwords += 32;
1427
1428 return dwords;
1429 }
1430
r300_get_num_cs_end_dwords(struct r300_context * r300)1431 unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
1432 {
1433 unsigned dwords = 0;
1434
1435 /* Emitted in flush. */
1436 dwords += 26; /* emit_query_end */
1437 dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
1438 if (r300->screen->caps.is_r500)
1439 dwords += 2; /* emit_index_bias */
1440 dwords += 3; /* MSPOS */
1441
1442 return dwords;
1443 }
1444
1445 /* Emit all dirty state. */
r300_emit_dirty_state(struct r300_context * r300)1446 void r300_emit_dirty_state(struct r300_context* r300)
1447 {
1448 struct r300_atom *atom;
1449
1450 foreach_dirty_atom(r300, atom) {
1451 if (atom->dirty) {
1452 atom->emit(r300, atom->size, atom->state);
1453 atom->dirty = false;
1454 }
1455 }
1456
1457 r300->first_dirty = NULL;
1458 r300->last_dirty = NULL;
1459 r300->dirty_hw++;
1460 }
1461