projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
r600g: improve flushed depth texture handling v2
[mesa.git] / src / gallium / drivers / r600 / r600_hw_context.c
1 /*
2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Jerome Glisse
25 */
26 #include "r600_hw_context_priv.h"
27 #include "r600d.h"
28 #include "util/u_memory.h"
29 #include <errno.h>
30
31 /* Get backends mask */
32 void r600_get_backend_mask(struct r600_context *ctx)
33 {
34 struct radeon_winsys_cs *cs = ctx->cs;
35 struct r600_resource *buffer;
36 uint32_t *results;
37 unsigned num_backends = ctx->screen->info.r600_num_backends;
38 unsigned i, mask = 0;
39 uint64_t va;
40
41 /* if backend_map query is supported by the kernel */
42 if (ctx->screen->info.r600_backend_map_valid) {
43 unsigned num_tile_pipes = ctx->screen->info.r600_num_tile_pipes;
44 unsigned backend_map = ctx->screen->info.r600_backend_map;
45 unsigned item_width, item_mask;
46
47 if (ctx->chip_class >= EVERGREEN) {
48 item_width = 4;
49 item_mask = 0x7;
50 } else {
51 item_width = 2;
52 item_mask = 0x3;
53 }
54
55 while(num_tile_pipes--) {
56 i = backend_map & item_mask;
57 mask |= (1<<i);
58 backend_map >>= item_width;
59 }
60 if (mask != 0) {
61 ctx->backend_mask = mask;
62 return;
63 }
64 }
65
66 /* otherwise backup path for older kernels */
67
68 /* create buffer for event data */
69 buffer = (struct r600_resource*)
70 pipe_buffer_create(&ctx->screen->screen, PIPE_BIND_CUSTOM,
71 PIPE_USAGE_STAGING, ctx->max_db*16);
72 if (!buffer)
73 goto err;
74
75 va = r600_resource_va(&ctx->screen->screen, (void*)buffer);
76
77 /* initialize buffer with zeroes */
78 results = ctx->ws->buffer_map(buffer->cs_buf, ctx->cs, PIPE_TRANSFER_WRITE);
79 if (results) {
80 memset(results, 0, ctx->max_db * 4 * 4);
81 ctx->ws->buffer_unmap(buffer->cs_buf);
82
83 /* emit EVENT_WRITE for ZPASS_DONE */
84 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
85 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
86 cs->buf[cs->cdw++] = va;
87 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFF;
88
89 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
90 cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, buffer, RADEON_USAGE_WRITE);
91
92 /* analyze results */
93 results = ctx->ws->buffer_map(buffer->cs_buf, ctx->cs, PIPE_TRANSFER_READ);
94 if (results) {
95 for(i = 0; i < ctx->max_db; i++) {
96 /* at least highest bit will be set if backend is used */
97 if (results[i*4 + 1])
98 mask |= (1<<i);
99 }
100 ctx->ws->buffer_unmap(buffer->cs_buf);
101 }
102 }
103
104 pipe_resource_reference((struct pipe_resource**)&buffer, NULL);
105
106 if (mask != 0) {
107 ctx->backend_mask = mask;
108 return;
109 }
110
111 err:
112 /* fallback to old method - set num_backends lower bits to 1 */
113 ctx->backend_mask = (~((uint32_t)0))>>(32-num_backends);
114 return;
115 }
116
117 void r600_context_ps_partial_flush(struct r600_context *ctx)
118 {
119 struct radeon_winsys_cs *cs = ctx->cs;
120
121 if (!(ctx->flags & R600_CONTEXT_DRAW_PENDING))
122 return;
123
124 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
125 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
126
127 ctx->flags &= ~R600_CONTEXT_DRAW_PENDING;
128 }
129
130 static void r600_init_block(struct r600_context *ctx,
131 struct r600_block *block,
132 const struct r600_reg *reg, int index, int nreg,
133 unsigned opcode, unsigned offset_base)
134 {
135 int i = index;
136 int j, n = nreg;
137
138 /* initialize block */
139 if (opcode == PKT3_SET_RESOURCE) {
140 block->flags = BLOCK_FLAG_RESOURCE;
141 block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY; /* dirty all blocks at start */
142 } else {
143 block->flags = 0;
144 block->status |= R600_BLOCK_STATUS_DIRTY; /* dirty all blocks at start */
145 }
146 block->start_offset = reg[i].offset;
147 block->pm4[block->pm4_ndwords++] = PKT3(opcode, n, 0);
148 block->pm4[block->pm4_ndwords++] = (block->start_offset - offset_base) >> 2;
149 block->reg = &block->pm4[block->pm4_ndwords];
150 block->pm4_ndwords += n;
151 block->nreg = n;
152 block->nreg_dirty = n;
153 LIST_INITHEAD(&block->list);
154 LIST_INITHEAD(&block->enable_list);
155
156 for (j = 0; j < n; j++) {
157 if (reg[i+j].flags & REG_FLAG_DIRTY_ALWAYS) {
158 block->flags |= REG_FLAG_DIRTY_ALWAYS;
159 }
160 if (reg[i+j].flags & REG_FLAG_ENABLE_ALWAYS) {
161 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
162 block->status |= R600_BLOCK_STATUS_ENABLED;
163 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
164 LIST_ADDTAIL(&block->list,&ctx->dirty);
165 }
166 }
167 if (reg[i+j].flags & REG_FLAG_FLUSH_CHANGE) {
168 block->flags |= REG_FLAG_FLUSH_CHANGE;
169 }
170
171 if (reg[i+j].flags & REG_FLAG_NEED_BO) {
172 block->nbo++;
173 assert(block->nbo < R600_BLOCK_MAX_BO);
174 block->pm4_bo_index[j] = block->nbo;
175 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0, 0);
176 block->pm4[block->pm4_ndwords++] = 0x00000000;
177 block->reloc[block->nbo].bo_pm4_index = block->pm4_ndwords - 1;
178 }
179 if ((ctx->family > CHIP_R600) &&
180 (ctx->family < CHIP_RV770) && reg[i+j].flags & REG_FLAG_RV6XX_SBU) {
181 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
182 block->pm4[block->pm4_ndwords++] = reg[i+j].sbu_flags;
183 }
184 }
185 /* check that we stay in limit */
186 assert(block->pm4_ndwords < R600_BLOCK_MAX_REG);
187 }
188
189 int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg,
190 unsigned opcode, unsigned offset_base)
191 {
192 struct r600_block *block;
193 struct r600_range *range;
194 int offset;
195
196 for (unsigned i = 0, n = 0; i < nreg; i += n) {
197 /* ignore new block balise */
198 if (reg[i].offset == GROUP_FORCE_NEW_BLOCK) {
199 n = 1;
200 continue;
201 }
202
203 /* ignore regs not on R600 on R600 */
204 if ((reg[i].flags & REG_FLAG_NOT_R600) && ctx->family == CHIP_R600) {
205 n = 1;
206 continue;
207 }
208
209 /* register that need relocation are in their own group */
210 /* find number of consecutive registers */
211 n = 0;
212 offset = reg[i].offset;
213 while (reg[i + n].offset == offset) {
214 n++;
215 offset += 4;
216 if ((n + i) >= nreg)
217 break;
218 if (n >= (R600_BLOCK_MAX_REG - 2))
219 break;
220 }
221
222 /* allocate new block */
223 block = calloc(1, sizeof(struct r600_block));
224 if (block == NULL) {
225 return -ENOMEM;
226 }
227 ctx->nblocks++;
228 for (int j = 0; j < n; j++) {
229 range = &ctx->range[CTX_RANGE_ID(reg[i + j].offset)];
230 /* create block table if it doesn't exist */
231 if (!range->blocks)
232 range->blocks = calloc(1 << HASH_SHIFT, sizeof(void *));
233 if (!range->blocks)
234 return -1;
235
236 range->blocks[CTX_BLOCK_ID(reg[i + j].offset)] = block;
237 }
238
239 r600_init_block(ctx, block, reg, i, n, opcode, offset_base);
240
241 }
242 return 0;
243 }
244
245 /* R600/R700 configuration */
246 static const struct r600_reg r600_config_reg_list[] = {
247 {R_008958_VGT_PRIMITIVE_TYPE, 0, 0},
248 {R_008C04_SQ_GPR_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0},
249 {R_009508_TA_CNTL_AUX, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0},
250 };
251
252 static const struct r600_reg r600_ctl_const_list[] = {
253 {R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0},
254 };
255
256 static const struct r600_reg r600_context_reg_list[] = {
257 {R_028A4C_PA_SC_MODE_CNTL, 0, 0},
258 {GROUP_FORCE_NEW_BLOCK, 0, 0},
259 {R_028040_CB_COLOR0_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(0)},
260 {GROUP_FORCE_NEW_BLOCK, 0, 0},
261 {R_0280A0_CB_COLOR0_INFO, REG_FLAG_NEED_BO, 0},
262 {R_028060_CB_COLOR0_SIZE, 0, 0},
263 {R_028080_CB_COLOR0_VIEW, 0, 0},
264 {GROUP_FORCE_NEW_BLOCK, 0, 0},
265 {R_0280E0_CB_COLOR0_FRAG, REG_FLAG_NEED_BO, 0},
266 {GROUP_FORCE_NEW_BLOCK, 0, 0},
267 {R_0280C0_CB_COLOR0_TILE, REG_FLAG_NEED_BO, 0},
268 {GROUP_FORCE_NEW_BLOCK, 0, 0},
269 {R_028044_CB_COLOR1_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(1)},
270 {GROUP_FORCE_NEW_BLOCK, 0, 0},
271 {R_0280A4_CB_COLOR1_INFO, REG_FLAG_NEED_BO, 0},
272 {R_028064_CB_COLOR1_SIZE, 0, 0},
273 {R_028084_CB_COLOR1_VIEW, 0, 0},
274 {GROUP_FORCE_NEW_BLOCK, 0, 0},
275 {R_0280E4_CB_COLOR1_FRAG, REG_FLAG_NEED_BO, 0},
276 {GROUP_FORCE_NEW_BLOCK, 0, 0},
277 {R_0280C4_CB_COLOR1_TILE, REG_FLAG_NEED_BO, 0},
278 {GROUP_FORCE_NEW_BLOCK, 0, 0},
279 {R_028048_CB_COLOR2_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(2)},
280 {GROUP_FORCE_NEW_BLOCK, 0, 0},
281 {R_0280A8_CB_COLOR2_INFO, REG_FLAG_NEED_BO, 0},
282 {R_028068_CB_COLOR2_SIZE, 0, 0},
283 {R_028088_CB_COLOR2_VIEW, 0, 0},
284 {GROUP_FORCE_NEW_BLOCK, 0, 0},
285 {R_0280E8_CB_COLOR2_FRAG, REG_FLAG_NEED_BO, 0},
286 {GROUP_FORCE_NEW_BLOCK, 0, 0},
287 {R_0280C8_CB_COLOR2_TILE, REG_FLAG_NEED_BO, 0},
288 {GROUP_FORCE_NEW_BLOCK, 0, 0},
289 {R_02804C_CB_COLOR3_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(3)},
290 {GROUP_FORCE_NEW_BLOCK, 0, 0},
291 {R_0280AC_CB_COLOR3_INFO, REG_FLAG_NEED_BO, 0},
292 {R_02806C_CB_COLOR3_SIZE, 0, 0},
293 {R_02808C_CB_COLOR3_VIEW, 0, 0},
294 {GROUP_FORCE_NEW_BLOCK, 0, 0},
295 {R_0280EC_CB_COLOR3_FRAG, REG_FLAG_NEED_BO, 0},
296 {GROUP_FORCE_NEW_BLOCK, 0, 0},
297 {R_0280CC_CB_COLOR3_TILE, REG_FLAG_NEED_BO, 0},
298 {GROUP_FORCE_NEW_BLOCK, 0, 0},
299 {R_028050_CB_COLOR4_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(4)},
300 {GROUP_FORCE_NEW_BLOCK, 0, 0},
301 {R_0280B0_CB_COLOR4_INFO, REG_FLAG_NEED_BO, 0},
302 {R_028070_CB_COLOR4_SIZE, 0, 0},
303 {R_028090_CB_COLOR4_VIEW, 0, 0},
304 {GROUP_FORCE_NEW_BLOCK, 0, 0},
305 {R_0280F0_CB_COLOR4_FRAG, REG_FLAG_NEED_BO, 0},
306 {GROUP_FORCE_NEW_BLOCK, 0, 0},
307 {R_0280D0_CB_COLOR4_TILE, REG_FLAG_NEED_BO, 0},
308 {GROUP_FORCE_NEW_BLOCK, 0, 0},
309 {R_028054_CB_COLOR5_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(5)},
310 {GROUP_FORCE_NEW_BLOCK, 0, 0},
311 {R_0280B4_CB_COLOR5_INFO, REG_FLAG_NEED_BO, 0},
312 {R_028074_CB_COLOR5_SIZE, 0, 0},
313 {R_028094_CB_COLOR5_VIEW, 0, 0},
314 {GROUP_FORCE_NEW_BLOCK, 0, 0},
315 {R_0280F4_CB_COLOR5_FRAG, REG_FLAG_NEED_BO, 0},
316 {GROUP_FORCE_NEW_BLOCK, 0, 0},
317 {R_0280D4_CB_COLOR5_TILE, REG_FLAG_NEED_BO, 0},
318 {R_028058_CB_COLOR6_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(6)},
319 {R_0280B8_CB_COLOR6_INFO, REG_FLAG_NEED_BO, 0},
320 {R_028078_CB_COLOR6_SIZE, 0, 0},
321 {R_028098_CB_COLOR6_VIEW, 0, 0},
322 {GROUP_FORCE_NEW_BLOCK, 0, 0},
323 {R_0280F8_CB_COLOR6_FRAG, REG_FLAG_NEED_BO, 0},
324 {GROUP_FORCE_NEW_BLOCK, 0, 0},
325 {R_0280D8_CB_COLOR6_TILE, REG_FLAG_NEED_BO, 0},
326 {GROUP_FORCE_NEW_BLOCK, 0, 0},
327 {R_02805C_CB_COLOR7_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(7)},
328 {GROUP_FORCE_NEW_BLOCK, 0, 0},
329 {R_0280BC_CB_COLOR7_INFO, REG_FLAG_NEED_BO, 0},
330 {R_02807C_CB_COLOR7_SIZE, 0, 0},
331 {R_02809C_CB_COLOR7_VIEW, 0, 0},
332 {R_0280FC_CB_COLOR7_FRAG, REG_FLAG_NEED_BO, 0},
333 {R_0280DC_CB_COLOR7_TILE, REG_FLAG_NEED_BO, 0},
334 {R_028120_CB_CLEAR_RED, 0, 0},
335 {R_028124_CB_CLEAR_GREEN, 0, 0},
336 {R_028128_CB_CLEAR_BLUE, 0, 0},
337 {R_02812C_CB_CLEAR_ALPHA, 0, 0},
338 {R_02823C_CB_SHADER_MASK, 0, 0},
339 {R_028238_CB_TARGET_MASK, 0, 0},
340 {R_028410_SX_ALPHA_TEST_CONTROL, 0, 0},
341 {R_028414_CB_BLEND_RED, 0, 0},
342 {R_028418_CB_BLEND_GREEN, 0, 0},
343 {R_02841C_CB_BLEND_BLUE, 0, 0},
344 {R_028420_CB_BLEND_ALPHA, 0, 0},
345 {R_028424_CB_FOG_RED, 0, 0},
346 {R_028428_CB_FOG_GREEN, 0, 0},
347 {R_02842C_CB_FOG_BLUE, 0, 0},
348 {R_028430_DB_STENCILREFMASK, 0, 0},
349 {R_028434_DB_STENCILREFMASK_BF, 0, 0},
350 {R_028438_SX_ALPHA_REF, 0, 0},
351 {R_028780_CB_BLEND0_CONTROL, REG_FLAG_NOT_R600, 0},
352 {R_028784_CB_BLEND1_CONTROL, REG_FLAG_NOT_R600, 0},
353 {R_028788_CB_BLEND2_CONTROL, REG_FLAG_NOT_R600, 0},
354 {R_02878C_CB_BLEND3_CONTROL, REG_FLAG_NOT_R600, 0},
355 {R_028790_CB_BLEND4_CONTROL, REG_FLAG_NOT_R600, 0},
356 {R_028794_CB_BLEND5_CONTROL, REG_FLAG_NOT_R600, 0},
357 {R_028798_CB_BLEND6_CONTROL, REG_FLAG_NOT_R600, 0},
358 {R_02879C_CB_BLEND7_CONTROL, REG_FLAG_NOT_R600, 0},
359 {R_0287A0_CB_SHADER_CONTROL, 0, 0},
360 {R_028800_DB_DEPTH_CONTROL, 0, 0},
361 {R_028804_CB_BLEND_CONTROL, 0, 0},
362 {R_028808_CB_COLOR_CONTROL, 0, 0},
363 {R_02880C_DB_SHADER_CONTROL, 0, 0},
364 {R_02800C_DB_DEPTH_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_DEPTH},
365 {R_028000_DB_DEPTH_SIZE, 0, 0},
366 {R_028004_DB_DEPTH_VIEW, 0, 0},
367 {GROUP_FORCE_NEW_BLOCK, 0, 0},
368 {R_028010_DB_DEPTH_INFO, REG_FLAG_NEED_BO, 0},
369 {R_028A6C_VGT_GS_OUT_PRIM_TYPE, 0, 0},
370 {R_028D24_DB_HTILE_SURFACE, 0, 0},
371 {R_028D34_DB_PREFETCH_LIMIT, 0, 0},
372 {R_028204_PA_SC_WINDOW_SCISSOR_TL, 0, 0},
373 {R_028208_PA_SC_WINDOW_SCISSOR_BR, 0, 0},
374 {R_028250_PA_SC_VPORT_SCISSOR_0_TL, 0, 0},
375 {R_028254_PA_SC_VPORT_SCISSOR_0_BR, 0, 0},
376 {R_02843C_PA_CL_VPORT_XSCALE_0, 0, 0},
377 {R_028440_PA_CL_VPORT_XOFFSET_0, 0, 0},
378 {R_028444_PA_CL_VPORT_YSCALE_0, 0, 0},
379 {R_028448_PA_CL_VPORT_YOFFSET_0, 0, 0},
380 {R_02844C_PA_CL_VPORT_ZSCALE_0, 0, 0},
381 {R_028450_PA_CL_VPORT_ZOFFSET_0, 0, 0},
382 {R_0286D4_SPI_INTERP_CONTROL_0, 0, 0},
383 {R_028810_PA_CL_CLIP_CNTL, 0, 0},
384 {R_028814_PA_SU_SC_MODE_CNTL, 0, 0},
385 {R_02881C_PA_CL_VS_OUT_CNTL, 0, 0},
386 {R_028A00_PA_SU_POINT_SIZE, 0, 0},
387 {R_028A04_PA_SU_POINT_MINMAX, 0, 0},
388 {R_028A08_PA_SU_LINE_CNTL, 0, 0},
389 {R_028A0C_PA_SC_LINE_STIPPLE, 0, 0},
390 {R_028C08_PA_SU_VTX_CNTL, 0, 0},
391 {R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL, 0, 0},
392 {R_028DFC_PA_SU_POLY_OFFSET_CLAMP, 0, 0},
393 {R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 0, 0},
394 {R_028E04_PA_SU_POLY_OFFSET_FRONT_OFFSET, 0, 0},
395 {R_028E08_PA_SU_POLY_OFFSET_BACK_SCALE, 0, 0},
396 {R_028E0C_PA_SU_POLY_OFFSET_BACK_OFFSET, 0, 0},
397 {R_028E20_PA_CL_UCP0_X, 0, 0},
398 {R_028E24_PA_CL_UCP0_Y, 0, 0},
399 {R_028E28_PA_CL_UCP0_Z, 0, 0},
400 {R_028E2C_PA_CL_UCP0_W, 0, 0},
401 {R_028E30_PA_CL_UCP1_X, 0, 0},
402 {R_028E34_PA_CL_UCP1_Y, 0, 0},
403 {R_028E38_PA_CL_UCP1_Z, 0, 0},
404 {R_028E3C_PA_CL_UCP1_W, 0, 0},
405 {R_028E40_PA_CL_UCP2_X, 0, 0},
406 {R_028E44_PA_CL_UCP2_Y, 0, 0},
407 {R_028E48_PA_CL_UCP2_Z, 0, 0},
408 {R_028E4C_PA_CL_UCP2_W, 0, 0},
409 {R_028E50_PA_CL_UCP3_X, 0, 0},
410 {R_028E54_PA_CL_UCP3_Y, 0, 0},
411 {R_028E58_PA_CL_UCP3_Z, 0, 0},
412 {R_028E5C_PA_CL_UCP3_W, 0, 0},
413 {R_028E60_PA_CL_UCP4_X, 0, 0},
414 {R_028E64_PA_CL_UCP4_Y, 0, 0},
415 {R_028E68_PA_CL_UCP4_Z, 0, 0},
416 {R_028E6C_PA_CL_UCP4_W, 0, 0},
417 {R_028E70_PA_CL_UCP5_X, 0, 0},
418 {R_028E74_PA_CL_UCP5_Y, 0, 0},
419 {R_028E78_PA_CL_UCP5_Z, 0, 0},
420 {R_028E7C_PA_CL_UCP5_W, 0, 0},
421 {R_028350_SX_MISC, 0, 0},
422 {R_028380_SQ_VTX_SEMANTIC_0, 0, 0},
423 {R_028384_SQ_VTX_SEMANTIC_1, 0, 0},
424 {R_028388_SQ_VTX_SEMANTIC_2, 0, 0},
425 {R_02838C_SQ_VTX_SEMANTIC_3, 0, 0},
426 {R_028390_SQ_VTX_SEMANTIC_4, 0, 0},
427 {R_028394_SQ_VTX_SEMANTIC_5, 0, 0},
428 {R_028398_SQ_VTX_SEMANTIC_6, 0, 0},
429 {R_02839C_SQ_VTX_SEMANTIC_7, 0, 0},
430 {R_0283A0_SQ_VTX_SEMANTIC_8, 0, 0},
431 {R_0283A4_SQ_VTX_SEMANTIC_9, 0, 0},
432 {R_0283A8_SQ_VTX_SEMANTIC_10, 0, 0},
433 {R_0283AC_SQ_VTX_SEMANTIC_11, 0, 0},
434 {R_0283B0_SQ_VTX_SEMANTIC_12, 0, 0},
435 {R_0283B4_SQ_VTX_SEMANTIC_13, 0, 0},
436 {R_0283B8_SQ_VTX_SEMANTIC_14, 0, 0},
437 {R_0283BC_SQ_VTX_SEMANTIC_15, 0, 0},
438 {R_0283C0_SQ_VTX_SEMANTIC_16, 0, 0},
439 {R_0283C4_SQ_VTX_SEMANTIC_17, 0, 0},
440 {R_0283C8_SQ_VTX_SEMANTIC_18, 0, 0},
441 {R_0283CC_SQ_VTX_SEMANTIC_19, 0, 0},
442 {R_0283D0_SQ_VTX_SEMANTIC_20, 0, 0},
443 {R_0283D4_SQ_VTX_SEMANTIC_21, 0, 0},
444 {R_0283D8_SQ_VTX_SEMANTIC_22, 0, 0},
445 {R_0283DC_SQ_VTX_SEMANTIC_23, 0, 0},
446 {R_0283E0_SQ_VTX_SEMANTIC_24, 0, 0},
447 {R_0283E4_SQ_VTX_SEMANTIC_25, 0, 0},
448 {R_0283E8_SQ_VTX_SEMANTIC_26, 0, 0},
449 {R_0283EC_SQ_VTX_SEMANTIC_27, 0, 0},
450 {R_0283F0_SQ_VTX_SEMANTIC_28, 0, 0},
451 {R_0283F4_SQ_VTX_SEMANTIC_29, 0, 0},
452 {R_0283F8_SQ_VTX_SEMANTIC_30, 0, 0},
453 {R_0283FC_SQ_VTX_SEMANTIC_31, 0, 0},
454 {R_028614_SPI_VS_OUT_ID_0, 0, 0},
455 {R_028618_SPI_VS_OUT_ID_1, 0, 0},
456 {R_02861C_SPI_VS_OUT_ID_2, 0, 0},
457 {R_028620_SPI_VS_OUT_ID_3, 0, 0},
458 {R_028624_SPI_VS_OUT_ID_4, 0, 0},
459 {R_028628_SPI_VS_OUT_ID_5, 0, 0},
460 {R_02862C_SPI_VS_OUT_ID_6, 0, 0},
461 {R_028630_SPI_VS_OUT_ID_7, 0, 0},
462 {R_028634_SPI_VS_OUT_ID_8, 0, 0},
463 {R_028638_SPI_VS_OUT_ID_9, 0, 0},
464 {R_0286C4_SPI_VS_OUT_CONFIG, 0, 0},
465 {GROUP_FORCE_NEW_BLOCK, 0, 0},
466 {R_028858_SQ_PGM_START_VS, REG_FLAG_NEED_BO, 0},
467 {GROUP_FORCE_NEW_BLOCK, 0, 0},
468 {R_028868_SQ_PGM_RESOURCES_VS, 0, 0},
469 {GROUP_FORCE_NEW_BLOCK, 0, 0},
470 {R_028894_SQ_PGM_START_FS, REG_FLAG_NEED_BO, 0},
471 {GROUP_FORCE_NEW_BLOCK, 0, 0},
472 {R_0288A4_SQ_PGM_RESOURCES_FS, 0, 0},
473 {R_0288DC_SQ_PGM_CF_OFFSET_FS, 0, 0},
474 {R_028644_SPI_PS_INPUT_CNTL_0, 0, 0},
475 {R_028648_SPI_PS_INPUT_CNTL_1, 0, 0},
476 {R_02864C_SPI_PS_INPUT_CNTL_2, 0, 0},
477 {R_028650_SPI_PS_INPUT_CNTL_3, 0, 0},
478 {R_028654_SPI_PS_INPUT_CNTL_4, 0, 0},
479 {R_028658_SPI_PS_INPUT_CNTL_5, 0, 0},
480 {R_02865C_SPI_PS_INPUT_CNTL_6, 0, 0},
481 {R_028660_SPI_PS_INPUT_CNTL_7, 0, 0},
482 {R_028664_SPI_PS_INPUT_CNTL_8, 0, 0},
483 {R_028668_SPI_PS_INPUT_CNTL_9, 0, 0},
484 {R_02866C_SPI_PS_INPUT_CNTL_10, 0, 0},
485 {R_028670_SPI_PS_INPUT_CNTL_11, 0, 0},
486 {R_028674_SPI_PS_INPUT_CNTL_12, 0, 0},
487 {R_028678_SPI_PS_INPUT_CNTL_13, 0, 0},
488 {R_02867C_SPI_PS_INPUT_CNTL_14, 0, 0},
489 {R_028680_SPI_PS_INPUT_CNTL_15, 0, 0},
490 {R_028684_SPI_PS_INPUT_CNTL_16, 0, 0},
491 {R_028688_SPI_PS_INPUT_CNTL_17, 0, 0},
492 {R_02868C_SPI_PS_INPUT_CNTL_18, 0, 0},
493 {R_028690_SPI_PS_INPUT_CNTL_19, 0, 0},
494 {R_028694_SPI_PS_INPUT_CNTL_20, 0, 0},
495 {R_028698_SPI_PS_INPUT_CNTL_21, 0, 0},
496 {R_02869C_SPI_PS_INPUT_CNTL_22, 0, 0},
497 {R_0286A0_SPI_PS_INPUT_CNTL_23, 0, 0},
498 {R_0286A4_SPI_PS_INPUT_CNTL_24, 0, 0},
499 {R_0286A8_SPI_PS_INPUT_CNTL_25, 0, 0},
500 {R_0286AC_SPI_PS_INPUT_CNTL_26, 0, 0},
501 {R_0286B0_SPI_PS_INPUT_CNTL_27, 0, 0},
502 {R_0286B4_SPI_PS_INPUT_CNTL_28, 0, 0},
503 {R_0286B8_SPI_PS_INPUT_CNTL_29, 0, 0},
504 {R_0286BC_SPI_PS_INPUT_CNTL_30, 0, 0},
505 {R_0286C0_SPI_PS_INPUT_CNTL_31, 0, 0},
506 {R_0286CC_SPI_PS_IN_CONTROL_0, 0, 0},
507 {R_0286D0_SPI_PS_IN_CONTROL_1, 0, 0},
508 {R_0286D8_SPI_INPUT_Z, 0, 0},
509 {GROUP_FORCE_NEW_BLOCK, 0, 0},
510 {R_028840_SQ_PGM_START_PS, REG_FLAG_NEED_BO, 0},
511 {GROUP_FORCE_NEW_BLOCK, 0, 0},
512 {R_028850_SQ_PGM_RESOURCES_PS, 0, 0},
513 {R_028854_SQ_PGM_EXPORTS_PS, 0, 0},
514 {R_028408_VGT_INDX_OFFSET, 0, 0},
515 {R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, 0, 0},
516 {R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, 0, 0},
517 };
518
519 /* SHADER RESOURCE R600/R700 */
520 int r600_resource_init(struct r600_context *ctx, struct r600_range *range, unsigned offset, unsigned nblocks, unsigned stride, struct r600_reg *reg, int nreg, unsigned offset_base)
521 {
522 int i;
523 struct r600_block *block;
524 range->blocks = calloc(nblocks, sizeof(struct r600_block *));
525 if (range->blocks == NULL)
526 return -ENOMEM;
527
528 reg[0].offset += offset;
529 for (i = 0; i < nblocks; i++) {
530 block = calloc(1, sizeof(struct r600_block));
531 if (block == NULL) {
532 return -ENOMEM;
533 }
534 ctx->nblocks++;
535 range->blocks[i] = block;
536 r600_init_block(ctx, block, reg, 0, nreg, PKT3_SET_RESOURCE, offset_base);
537
538 reg[0].offset += stride;
539 }
540 return 0;
541 }
542
543
544 static int r600_resource_range_init(struct r600_context *ctx, struct r600_range *range, unsigned offset, unsigned nblocks, unsigned stride)
545 {
546 struct r600_reg r600_shader_resource[] = {
547 {R_038000_RESOURCE0_WORD0, REG_FLAG_NEED_BO, 0},
548 {R_038004_RESOURCE0_WORD1, REG_FLAG_NEED_BO, 0},
549 {R_038008_RESOURCE0_WORD2, 0, 0},
550 {R_03800C_RESOURCE0_WORD3, 0, 0},
551 {R_038010_RESOURCE0_WORD4, 0, 0},
552 {R_038014_RESOURCE0_WORD5, 0, 0},
553 {R_038018_RESOURCE0_WORD6, 0, 0},
554 };
555 unsigned nreg = Elements(r600_shader_resource);
556
557 return r600_resource_init(ctx, range, offset, nblocks, stride, r600_shader_resource, nreg, R600_RESOURCE_OFFSET);
558 }
559
560 /* SHADER SAMPLER R600/R700/EG/CM */
561 int r600_state_sampler_init(struct r600_context *ctx, uint32_t offset)
562 {
563 struct r600_reg r600_shader_sampler[] = {
564 {R_03C000_SQ_TEX_SAMPLER_WORD0_0, 0, 0},
565 {R_03C004_SQ_TEX_SAMPLER_WORD1_0, 0, 0},
566 {R_03C008_SQ_TEX_SAMPLER_WORD2_0, 0, 0},
567 };
568 unsigned nreg = Elements(r600_shader_sampler);
569
570 for (int i = 0; i < nreg; i++) {
571 r600_shader_sampler[i].offset += offset;
572 }
573 return r600_context_add_block(ctx, r600_shader_sampler, nreg, PKT3_SET_SAMPLER, R600_SAMPLER_OFFSET);
574 }
575
576 /* SHADER SAMPLER BORDER R600/R700 */
577 static int r600_state_sampler_border_init(struct r600_context *ctx, uint32_t offset)
578 {
579 struct r600_reg r600_shader_sampler_border[] = {
580 {R_00A400_TD_PS_SAMPLER0_BORDER_RED, 0, 0},
581 {R_00A404_TD_PS_SAMPLER0_BORDER_GREEN, 0, 0},
582 {R_00A408_TD_PS_SAMPLER0_BORDER_BLUE, 0, 0},
583 {R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA, 0, 0},
584 };
585 unsigned nreg = Elements(r600_shader_sampler_border);
586
587 for (int i = 0; i < nreg; i++) {
588 r600_shader_sampler_border[i].offset += offset;
589 }
590 return r600_context_add_block(ctx, r600_shader_sampler_border, nreg, PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
591 }
592
593 static int r600_loop_const_init(struct r600_context *ctx, uint32_t offset)
594 {
595 unsigned nreg = 32;
596 struct r600_reg r600_loop_consts[32];
597 int i;
598
599 for (i = 0; i < nreg; i++) {
600 r600_loop_consts[i].offset = R600_LOOP_CONST_OFFSET + ((offset + i) * 4);
601 r600_loop_consts[i].flags = REG_FLAG_DIRTY_ALWAYS;
602 r600_loop_consts[i].sbu_flags = 0;
603 }
604 return r600_context_add_block(ctx, r600_loop_consts, nreg, PKT3_SET_LOOP_CONST, R600_LOOP_CONST_OFFSET);
605 }
606
607 static void r600_free_resource_range(struct r600_context *ctx, struct r600_range *range, int nblocks)
608 {
609 struct r600_block *block;
610 int i;
611
612 if (!range->blocks) {
613 return; /* nothing to do */
614 }
615
616 for (i = 0; i < nblocks; i++) {
617 block = range->blocks[i];
618 if (block) {
619 for (int k = 1; k <= block->nbo; k++)
620 pipe_resource_reference((struct pipe_resource**)&block->reloc[k].bo, NULL);
621 free(block);
622 }
623 }
624 free(range->blocks);
625 }
626
627 /* initialize */
628 void r600_context_fini(struct r600_context *ctx)
629 {
630 struct r600_block *block;
631 struct r600_range *range;
632
633 if (ctx->range) {
634 for (int i = 0; i < NUM_RANGES; i++) {
635 if (!ctx->range[i].blocks)
636 continue;
637 for (int j = 0; j < (1 << HASH_SHIFT); j++) {
638 block = ctx->range[i].blocks[j];
639 if (block) {
640 for (int k = 0, offset = block->start_offset; k < block->nreg; k++, offset += 4) {
641 range = &ctx->range[CTX_RANGE_ID(offset)];
642 range->blocks[CTX_BLOCK_ID(offset)] = NULL;
643 }
644 for (int k = 1; k <= block->nbo; k++) {
645 pipe_resource_reference((struct pipe_resource**)&block->reloc[k].bo, NULL);
646 }
647 free(block);
648 }
649 }
650 free(ctx->range[i].blocks);
651 }
652 }
653 r600_free_resource_range(ctx, &ctx->ps_resources, ctx->num_ps_resources);
654 r600_free_resource_range(ctx, &ctx->vs_resources, ctx->num_vs_resources);
655 free(ctx->blocks);
656 }
657
658 static void r600_add_resource_block(struct r600_context *ctx, struct r600_range *range, int num_blocks, int *index)
659 {
660 int c = *index;
661 for (int j = 0; j < num_blocks; j++) {
662 if (!range->blocks[j])
663 continue;
664
665 ctx->blocks[c++] = range->blocks[j];
666 }
667 *index = c;
668 }
669
670 int r600_setup_block_table(struct r600_context *ctx)
671 {
672 /* setup block table */
673 int c = 0;
674 ctx->blocks = calloc(ctx->nblocks, sizeof(void*));
675 if (!ctx->blocks)
676 return -ENOMEM;
677 for (int i = 0; i < NUM_RANGES; i++) {
678 if (!ctx->range[i].blocks)
679 continue;
680 for (int j = 0, add; j < (1 << HASH_SHIFT); j++) {
681 if (!ctx->range[i].blocks[j])
682 continue;
683
684 add = 1;
685 for (int k = 0; k < c; k++) {
686 if (ctx->blocks[k] == ctx->range[i].blocks[j]) {
687 add = 0;
688 break;
689 }
690 }
691 if (add) {
692 assert(c < ctx->nblocks);
693 ctx->blocks[c++] = ctx->range[i].blocks[j];
694 j += (ctx->range[i].blocks[j]->nreg) - 1;
695 }
696 }
697 }
698
699 r600_add_resource_block(ctx, &ctx->ps_resources, ctx->num_ps_resources, &c);
700 r600_add_resource_block(ctx, &ctx->vs_resources, ctx->num_vs_resources, &c);
701 return 0;
702 }
703
704 int r600_context_init(struct r600_context *ctx)
705 {
706 int r;
707
708 /* add blocks */
709 r = r600_context_add_block(ctx, r600_config_reg_list,
710 Elements(r600_config_reg_list), PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
711 if (r)
712 goto out_err;
713 r = r600_context_add_block(ctx, r600_context_reg_list,
714 Elements(r600_context_reg_list), PKT3_SET_CONTEXT_REG, R600_CONTEXT_REG_OFFSET);
715 if (r)
716 goto out_err;
717 r = r600_context_add_block(ctx, r600_ctl_const_list,
718 Elements(r600_ctl_const_list), PKT3_SET_CTL_CONST, R600_CTL_CONST_OFFSET);
719 if (r)
720 goto out_err;
721
722 /* PS SAMPLER BORDER */
723 for (int j = 0, offset = 0; j < 18; j++, offset += 0x10) {
724 r = r600_state_sampler_border_init(ctx, offset);
725 if (r)
726 goto out_err;
727 }
728
729 /* VS SAMPLER BORDER */
730 for (int j = 0, offset = 0x200; j < 18; j++, offset += 0x10) {
731 r = r600_state_sampler_border_init(ctx, offset);
732 if (r)
733 goto out_err;
734 }
735 /* PS SAMPLER */
736 for (int j = 0, offset = 0; j < 18; j++, offset += 0xC) {
737 r = r600_state_sampler_init(ctx, offset);
738 if (r)
739 goto out_err;
740 }
741 /* VS SAMPLER */
742 for (int j = 0, offset = 0xD8; j < 18; j++, offset += 0xC) {
743 r = r600_state_sampler_init(ctx, offset);
744 if (r)
745 goto out_err;
746 }
747
748 ctx->num_ps_resources = 160;
749 ctx->num_vs_resources = 160;
750 r = r600_resource_range_init(ctx, &ctx->ps_resources, 0, 160, 0x1c);
751 if (r)
752 goto out_err;
753 r = r600_resource_range_init(ctx, &ctx->vs_resources, 0x1180, 160, 0x1c);
754 if (r)
755 goto out_err;
756
757 /* PS loop const */
758 r600_loop_const_init(ctx, 0);
759 /* VS loop const */
760 r600_loop_const_init(ctx, 32);
761
762 r = r600_setup_block_table(ctx);
763 if (r)
764 goto out_err;
765
766 ctx->max_db = 4;
767 return 0;
768 out_err:
769 r600_context_fini(ctx);
770 return r;
771 }
772
773 void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
774 boolean count_draw_in)
775 {
776 struct r600_atom *state;
777
778 /* The number of dwords we already used in the CS so far. */
779 num_dw += ctx->cs->cdw;
780
781 if (count_draw_in) {
782 /* The number of dwords all the dirty states would take. */
783 LIST_FOR_EACH_ENTRY(state, &ctx->dirty_states, head) {
784 num_dw += state->num_dw;
785 }
786
787 num_dw += ctx->pm4_dirty_cdwords;
788
789 /* The upper-bound of how much a draw command would take. */
790 num_dw += R600_MAX_DRAW_CS_DWORDS;
791 }
792
793 /* Count in queries_suspend. */
794 num_dw += ctx->num_cs_dw_nontimer_queries_suspend;
795 num_dw += ctx->num_cs_dw_timer_queries_suspend;
796
797 /* Count in streamout_end at the end of CS. */
798 num_dw += ctx->num_cs_dw_streamout_end;
799
800 /* Count in render_condition(NULL) at the end of CS. */
801 if (ctx->predicate_drawing) {
802 num_dw += 3;
803 }
804
805 /* Count in framebuffer cache flushes at the end of CS. */
806 num_dw += 7; /* one SURFACE_SYNC and CACHE_FLUSH_AND_INV (r6xx-only) */
807
808 /* Save 16 dwords for the fence mechanism. */
809 num_dw += 16;
810
811 /* Flush if there's not enough space. */
812 if (num_dw > RADEON_MAX_CMDBUF_DWORDS) {
813 r600_flush(&ctx->context, NULL, RADEON_FLUSH_ASYNC);
814 }
815 }
816
817 void r600_context_dirty_block(struct r600_context *ctx,
818 struct r600_block *block,
819 int dirty, int index)
820 {
821 if ((index + 1) > block->nreg_dirty)
822 block->nreg_dirty = index + 1;
823
824 if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
825 block->status |= R600_BLOCK_STATUS_DIRTY;
826 ctx->pm4_dirty_cdwords += block->pm4_ndwords;
827 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
828 block->status |= R600_BLOCK_STATUS_ENABLED;
829 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
830 }
831 LIST_ADDTAIL(&block->list,&ctx->dirty);
832
833 if (block->flags & REG_FLAG_FLUSH_CHANGE) {
834 r600_context_ps_partial_flush(ctx);
835 }
836 }
837 }
838
839 /**
840 * If reg needs a reloc, this function will add it to its block's reloc list.
841 * @return true if reg needs a reloc, false otherwise
842 */
843 static bool r600_reg_set_block_reloc(struct r600_pipe_reg *reg)
844 {
845 unsigned reloc_id;
846
847 if (!reg->block->pm4_bo_index[reg->id]) {
848 return false;
849 }
850 /* find relocation */
851 reloc_id = reg->block->pm4_bo_index[reg->id];
852 pipe_resource_reference(
853 (struct pipe_resource**)&reg->block->reloc[reloc_id].bo,
854 &reg->bo->b.b);
855 reg->block->reloc[reloc_id].bo_usage = reg->bo_usage;
856 return true;
857 }
858
859 /**
860 * This function will emit all the registers in state directly to the command
861 * stream allowing you to bypass the r600_context dirty list.
862 *
863 * This is used for dispatching compute shaders to avoid mixing compute and
864 * 3D states in the context's dirty list.
865 *
866 * @param pkt_flags Should be either 0 or RADEON_CP_PACKET3_COMPUTE_MODE. This
867 * value will be passed on to r600_context_block_emit_dirty an or'd against
868 * the PKT3 headers.
869 */
870 void r600_context_pipe_state_emit(struct r600_context *ctx,
871 struct r600_pipe_state *state,
872 unsigned pkt_flags)
873 {
874 unsigned i;
875
876 /* Mark all blocks as dirty:
877 * Since two registers can be in the same block, we need to make sure
878 * we mark all the blocks dirty before we emit any of them. If we were
879 * to mark blocks dirty and emit them in the same loop, like this:
880 *
881 * foreach (reg in state->regs) {
882 * mark_dirty(reg->block)
883 * emit_block(reg->block)
884 * }
885 *
886 * Then if we have two registers in this state that are in the same
887 * block, we would end up emitting that block twice.
888 */
889 for (i = 0; i < state->nregs; i++) {
890 struct r600_pipe_reg *reg = &state->regs[i];
891 /* Mark all the registers in the block as dirty */
892 reg->block->nreg_dirty = reg->block->nreg;
893 reg->block->status |= R600_BLOCK_STATUS_DIRTY;
894 /* Update the reloc for this register if necessary. */
895 r600_reg_set_block_reloc(reg);
896 }
897
898 /* Emit the registers writes */
899 for (i = 0; i < state->nregs; i++) {
900 struct r600_pipe_reg *reg = &state->regs[i];
901 if (reg->block->status & R600_BLOCK_STATUS_DIRTY) {
902 r600_context_block_emit_dirty(ctx, reg->block, pkt_flags);
903 }
904 }
905 }
906
907 void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state)
908 {
909 struct r600_block *block;
910 int dirty;
911 for (int i = 0; i < state->nregs; i++) {
912 unsigned id;
913 struct r600_pipe_reg *reg = &state->regs[i];
914
915 block = reg->block;
916 id = reg->id;
917
918 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
919
920 if (reg->value != block->reg[id]) {
921 block->reg[id] = reg->value;
922 dirty |= R600_BLOCK_STATUS_DIRTY;
923 }
924 if (block->flags & REG_FLAG_DIRTY_ALWAYS)
925 dirty |= R600_BLOCK_STATUS_DIRTY;
926 if (r600_reg_set_block_reloc(reg)) {
927 /* always force dirty for relocs for now */
928 dirty |= R600_BLOCK_STATUS_DIRTY;
929 }
930
931 if (dirty)
932 r600_context_dirty_block(ctx, block, dirty, id);
933 }
934 }
935
936 static void r600_context_dirty_resource_block(struct r600_context *ctx,
937 struct r600_block *block,
938 int dirty, int index)
939 {
940 block->nreg_dirty = index + 1;
941
942 if ((dirty != (block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
943 block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
944 ctx->pm4_dirty_cdwords += block->pm4_ndwords;
945 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
946 block->status |= R600_BLOCK_STATUS_ENABLED;
947 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
948 }
949 LIST_ADDTAIL(&block->list,&ctx->resource_dirty);
950 }
951 }
952
953 void r600_context_pipe_state_set_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, struct r600_block *block)
954 {
955 int dirty;
956 int num_regs = ctx->chip_class >= EVERGREEN ? 8 : 7;
957
958 if (state == NULL) {
959 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_RESOURCE_DIRTY);
960 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, NULL);
961 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, NULL);
962 LIST_DELINIT(&block->list);
963 LIST_DELINIT(&block->enable_list);
964 return;
965 }
966
967 dirty = block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY;
968
969 if (memcmp(block->reg, state->val, num_regs*4)) {
970 memcpy(block->reg, state->val, num_regs * 4);
971 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
972 }
973
974 /* if no BOs on block, force dirty */
975 if (!block->reloc[1].bo || !block->reloc[2].bo)
976 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
977
978 if (!dirty) {
979 if ((block->reloc[1].bo->buf != state->bo[0]->buf) ||
980 (block->reloc[2].bo->buf != state->bo[1]->buf))
981 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
982 }
983
984 if (dirty) {
985 /* TEXTURE RESOURCE */
986 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, &state->bo[0]->b.b);
987 block->reloc[1].bo_usage = state->bo_usage[0];
988 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, &state->bo[1]->b.b);
989 block->reloc[2].bo_usage = state->bo_usage[1];
990
991 r600_context_dirty_resource_block(ctx, block, dirty, num_regs - 1);
992 }
993 }
994
995 void r600_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid)
996 {
997 struct r600_block *block = ctx->ps_resources.blocks[rid];
998
999 r600_context_pipe_state_set_resource(ctx, state, block);
1000 }
1001
1002 void r600_context_pipe_state_set_vs_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid)
1003 {
1004 struct r600_block *block = ctx->vs_resources.blocks[rid];
1005
1006 r600_context_pipe_state_set_resource(ctx, state, block);
1007 }
1008
1009 void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
1010 {
1011 struct r600_range *range;
1012 struct r600_block *block;
1013 int i;
1014 int dirty;
1015
1016 range = &ctx->range[CTX_RANGE_ID(offset)];
1017 block = range->blocks[CTX_BLOCK_ID(offset)];
1018 if (state == NULL) {
1019 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
1020 LIST_DELINIT(&block->list);
1021 LIST_DELINIT(&block->enable_list);
1022 return;
1023 }
1024 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
1025
1026 for (i = 0; i < 3; i++) {
1027 if (block->reg[i] != state->regs[i].value) {
1028 block->reg[i] = state->regs[i].value;
1029 dirty |= R600_BLOCK_STATUS_DIRTY;
1030 }
1031 }
1032
1033 if (dirty)
1034 r600_context_dirty_block(ctx, block, dirty, 2);
1035 }
1036
1037 static inline void r600_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
1038 {
1039 struct r600_range *range;
1040 struct r600_block *block;
1041 int i;
1042 int dirty;
1043
1044 range = &ctx->range[CTX_RANGE_ID(offset)];
1045 block = range->blocks[CTX_BLOCK_ID(offset)];
1046 if (state == NULL) {
1047 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
1048 LIST_DELINIT(&block->list);
1049 LIST_DELINIT(&block->enable_list);
1050 return;
1051 }
1052 if (state->nregs <= 3) {
1053 return;
1054 }
1055 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
1056 for (i = 0; i < 4; i++) {
1057 if (block->reg[i] != state->regs[i + 3].value) {
1058 block->reg[i] = state->regs[i + 3].value;
1059 dirty |= R600_BLOCK_STATUS_DIRTY;
1060 }
1061 }
1062
1063 /* We have to flush the shaders before we change the border color
1064 * registers, or previous draw commands that haven't completed yet
1065 * will end up using the new border color. */
1066 if (dirty & R600_BLOCK_STATUS_DIRTY)
1067 r600_context_ps_partial_flush(ctx);
1068 if (dirty)
1069 r600_context_dirty_block(ctx, block, dirty, 3);
1070 }
1071
1072 void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
1073 {
1074 unsigned offset;
1075
1076 offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*id;
1077 r600_context_pipe_state_set_sampler(ctx, state, offset);
1078 offset = R_00A400_TD_PS_SAMPLER0_BORDER_RED + 16*id;
1079 r600_context_pipe_state_set_sampler_border(ctx, state, offset);
1080 }
1081
1082 void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
1083 {
1084 unsigned offset;
1085
1086 offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*(id + 18);
1087 r600_context_pipe_state_set_sampler(ctx, state, offset);
1088 offset = R_00A600_TD_VS_SAMPLER0_BORDER_RED + 16*id;
1089 r600_context_pipe_state_set_sampler_border(ctx, state, offset);
1090 }
1091
1092 /**
1093 * @param pkt_flags should be set to RADEON_CP_PACKET3_COMPUTE_MODE if this
1094 * block will be used for compute shaders.
1095 */
1096 void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block,
1097 unsigned pkt_flags)
1098 {
1099 struct radeon_winsys_cs *cs = ctx->cs;
1100 int optional = block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS);
1101 int cp_dwords = block->pm4_ndwords, start_dword = 0;
1102 int new_dwords = 0;
1103 int nbo = block->nbo;
1104
1105 if (block->nreg_dirty == 0 && optional) {
1106 goto out;
1107 }
1108
1109 if (nbo) {
1110 for (int j = 0; j < block->nreg; j++) {
1111 if (block->pm4_bo_index[j]) {
1112 /* find relocation */
1113 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]];
1114 if (reloc->bo) {
1115 block->pm4[reloc->bo_pm4_index] =
1116 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage);
1117 } else {
1118 block->pm4[reloc->bo_pm4_index] = 0;
1119 }
1120 nbo--;
1121 if (nbo == 0)
1122 break;
1123
1124 }
1125 }
1126 }
1127
1128 optional &= (block->nreg_dirty != block->nreg);
1129 if (optional) {
1130 new_dwords = block->nreg_dirty;
1131 start_dword = cs->cdw;
1132 cp_dwords = new_dwords + 2;
1133 }
1134 memcpy(&cs->buf[cs->cdw], block->pm4, cp_dwords * 4);
1135
1136 /* We are applying the pkt_flags after copying the register block to
1137 * the the command stream, because it is possible this block will be
1138 * emitted with a different pkt_flags, and we don't want to store the
1139 * pkt_flags in the block.
1140 */
1141 cs->buf[cs->cdw] |= pkt_flags;
1142 cs->cdw += cp_dwords;
1143
1144 if (optional) {
1145 uint32_t newword;
1146
1147 newword = cs->buf[start_dword];
1148 newword &= PKT_COUNT_C;
1149 newword |= PKT_COUNT_S(new_dwords);
1150 cs->buf[start_dword] = newword;
1151 }
1152 out:
1153 block->status ^= R600_BLOCK_STATUS_DIRTY;
1154 block->nreg_dirty = 0;
1155 LIST_DELINIT(&block->list);
1156 }
1157
1158 void r600_context_block_resource_emit_dirty(struct r600_context *ctx, struct r600_block *block)
1159 {
1160 struct radeon_winsys_cs *cs = ctx->cs;
1161 int cp_dwords = block->pm4_ndwords;
1162 int nbo = block->nbo;
1163
1164 for (int j = 0; j < nbo; j++) {
1165 if (block->pm4_bo_index[j]) {
1166 /* find relocation */
1167 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]];
1168 block->pm4[reloc->bo_pm4_index] =
1169 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage);
1170 }
1171 }
1172
1173 memcpy(&cs->buf[cs->cdw], block->pm4, cp_dwords * 4);
1174 cs->cdw += cp_dwords;
1175
1176 block->status ^= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1177 block->nreg_dirty = 0;
1178 LIST_DELINIT(&block->list);
1179 }
1180
1181 void r600_inval_shader_cache(struct r600_context *ctx)
1182 {
1183 ctx->surface_sync_cmd.flush_flags |= S_0085F0_SH_ACTION_ENA(1);
1184 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1185 }
1186
1187 void r600_inval_texture_cache(struct r600_context *ctx)
1188 {
1189 ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
1190 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1191 }
1192
1193 void r600_inval_vertex_cache(struct r600_context *ctx)
1194 {
1195 if (ctx->has_vertex_cache) {
1196 ctx->surface_sync_cmd.flush_flags |= S_0085F0_VC_ACTION_ENA(1);
1197 } else {
1198 /* Some GPUs don't have the vertex cache and must use the texture cache instead. */
1199 ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
1200 }
1201 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1202 }
1203
1204 void r600_flush_framebuffer(struct r600_context *ctx, bool flush_now)
1205 {
1206 if (!(ctx->flags & R600_CONTEXT_DST_CACHES_DIRTY))
1207 return;
1208
1209 ctx->surface_sync_cmd.flush_flags |=
1210 r600_get_cb_flush_flags(ctx) |
1211 (ctx->framebuffer.zsbuf ? S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1) : 0);
1212
1213 if (flush_now) {
1214 r600_emit_atom(ctx, &ctx->surface_sync_cmd.atom);
1215 } else {
1216 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1217 }
1218
1219 /* Also add a complete cache flush to work around broken flushing on R6xx. */
1220 if (ctx->chip_class == R600) {
1221 if (flush_now) {
1222 r600_emit_atom(ctx, &ctx->r6xx_flush_and_inv_cmd);
1223 } else {
1224 r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
1225 }
1226 }
1227
1228 ctx->flags &= ~R600_CONTEXT_DST_CACHES_DIRTY;
1229 }
1230
1231 void r600_context_flush(struct r600_context *ctx, unsigned flags)
1232 {
1233 struct radeon_winsys_cs *cs = ctx->cs;
1234 struct r600_block *enable_block = NULL;
1235 bool timer_queries_suspended = false;
1236 bool nontimer_queries_suspended = false;
1237 bool streamout_suspended = false;
1238
1239 if (cs->cdw == ctx->start_cs_cmd.atom.num_dw)
1240 return;
1241
1242 /* suspend queries */
1243 if (ctx->num_cs_dw_timer_queries_suspend) {
1244 r600_suspend_timer_queries(ctx);
1245 timer_queries_suspended = true;
1246 }
1247 if (ctx->num_cs_dw_nontimer_queries_suspend) {
1248 r600_suspend_nontimer_queries(ctx);
1249 nontimer_queries_suspended = true;
1250 }
1251
1252 if (ctx->num_cs_dw_streamout_end) {
1253 r600_context_streamout_end(ctx);
1254 streamout_suspended = true;
1255 }
1256
1257 r600_flush_framebuffer(ctx, true);
1258
1259 /* partial flush is needed to avoid lockups on some chips with user fences */
1260 r600_context_ps_partial_flush(ctx);
1261
1262 /* old kernels and userspace don't set SX_MISC, so we must reset it to 0 here */
1263 if (ctx->chip_class <= R700) {
1264 r600_write_context_reg(cs, R_028350_SX_MISC, 0);
1265 }
1266
1267 /* force to keep tiling flags */
1268 flags |= RADEON_FLUSH_KEEP_TILING_FLAGS;
1269
1270 /* Flush the CS. */
1271 ctx->ws->cs_flush(ctx->cs, flags);
1272
1273 ctx->pm4_dirty_cdwords = 0;
1274 ctx->flags = 0;
1275
1276 /* Begin a new CS. */
1277 r600_emit_atom(ctx, &ctx->start_cs_cmd.atom);
1278
1279 /* Invalidate caches. */
1280 r600_inval_vertex_cache(ctx);
1281 r600_inval_texture_cache(ctx);
1282 r600_inval_shader_cache(ctx);
1283 r600_flush_framebuffer(ctx, false);
1284
1285 /* Re-emit states. */
1286 r600_atom_dirty(ctx, &ctx->db_misc_state.atom);
1287 r600_atom_dirty(ctx, &ctx->vertex_buffer_state);
1288
1289 ctx->vs_constbuf_state.dirty_mask = ctx->vs_constbuf_state.enabled_mask;
1290 ctx->ps_constbuf_state.dirty_mask = ctx->ps_constbuf_state.enabled_mask;
1291 r600_constant_buffers_dirty(ctx, &ctx->vs_constbuf_state);
1292 r600_constant_buffers_dirty(ctx, &ctx->ps_constbuf_state);
1293
1294 if (streamout_suspended) {
1295 ctx->streamout_start = TRUE;
1296 ctx->streamout_append_bitmask = ~0;
1297 }
1298
1299 /* resume queries */
1300 if (timer_queries_suspended) {
1301 r600_resume_timer_queries(ctx);
1302 }
1303 if (nontimer_queries_suspended) {
1304 r600_resume_nontimer_queries(ctx);
1305 }
1306
1307 /* set all valid group as dirty so they get reemited on
1308 * next draw command
1309 */
1310 LIST_FOR_EACH_ENTRY(enable_block, &ctx->enable_list, enable_list) {
1311 if (!(enable_block->flags & BLOCK_FLAG_RESOURCE)) {
1312 if(!(enable_block->status & R600_BLOCK_STATUS_DIRTY)) {
1313 LIST_ADDTAIL(&enable_block->list,&ctx->dirty);
1314 enable_block->status |= R600_BLOCK_STATUS_DIRTY;
1315 }
1316 } else {
1317 if(!(enable_block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) {
1318 LIST_ADDTAIL(&enable_block->list,&ctx->resource_dirty);
1319 enable_block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1320 }
1321 }
1322 ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords;
1323 enable_block->nreg_dirty = enable_block->nreg;
1324 }
1325 }
1326
1327 void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value)
1328 {
1329 struct radeon_winsys_cs *cs = ctx->cs;
1330 uint64_t va;
1331
1332 r600_need_cs_space(ctx, 10, FALSE);
1333
1334 va = r600_resource_va(&ctx->screen->screen, (void*)fence_bo);
1335 va = va + (offset << 2);
1336
1337 r600_context_ps_partial_flush(ctx);
1338 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1339 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1340 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* ADDRESS_LO */
1341 /* DATA_SEL | INT_EN | ADDRESS_HI */
1342 cs->buf[cs->cdw++] = (1 << 29) | (0 << 24) | ((va >> 32UL) & 0xFF);
1343 cs->buf[cs->cdw++] = value; /* DATA_LO */
1344 cs->buf[cs->cdw++] = 0; /* DATA_HI */
1345 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1346 cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, fence_bo, RADEON_USAGE_WRITE);
1347 }
1348
1349 static void r600_flush_vgt_streamout(struct r600_context *ctx)
1350 {
1351 struct radeon_winsys_cs *cs = ctx->cs;
1352
1353 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0);
1354 cs->buf[cs->cdw++] = (R_008490_CP_STRMOUT_CNTL - R600_CONFIG_REG_OFFSET) >> 2;
1355 cs->buf[cs->cdw++] = 0;
1356
1357 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
1358 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0);
1359
1360 cs->buf[cs->cdw++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0);
1361 cs->buf[cs->cdw++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */
1362 cs->buf[cs->cdw++] = R_008490_CP_STRMOUT_CNTL >> 2; /* register */
1363 cs->buf[cs->cdw++] = 0;
1364 cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* reference value */
1365 cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* mask */
1366 cs->buf[cs->cdw++] = 4; /* poll interval */
1367 }
1368
1369 static void r600_set_streamout_enable(struct r600_context *ctx, unsigned buffer_enable_bit)
1370 {
1371 struct radeon_winsys_cs *cs = ctx->cs;
1372
1373 if (buffer_enable_bit) {
1374 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1375 cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1376 cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(1);
1377
1378 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1379 cs->buf[cs->cdw++] = (R_028B20_VGT_STRMOUT_BUFFER_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1380 cs->buf[cs->cdw++] = buffer_enable_bit;
1381 } else {
1382 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1383 cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1384 cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(0);
1385 }
1386 }
1387
1388 void r600_context_streamout_begin(struct r600_context *ctx)
1389 {
1390 struct radeon_winsys_cs *cs = ctx->cs;
1391 struct r600_so_target **t = ctx->so_targets;
1392 unsigned *stride_in_dw = ctx->vs_shader->so.stride;
1393 unsigned buffer_en, i, update_flags = 0;
1394 uint64_t va;
1395
1396 buffer_en = (ctx->num_so_targets >= 1 && t[0] ? 1 : 0) |
1397 (ctx->num_so_targets >= 2 && t[1] ? 2 : 0) |
1398 (ctx->num_so_targets >= 3 && t[2] ? 4 : 0) |
1399 (ctx->num_so_targets >= 4 && t[3] ? 8 : 0);
1400
1401 ctx->num_cs_dw_streamout_end =
1402 12 + /* flush_vgt_streamout */
1403 util_bitcount(buffer_en) * 8 + /* STRMOUT_BUFFER_UPDATE */
1404 3 /* set_streamout_enable(0) */;
1405
1406 r600_need_cs_space(ctx,
1407 12 + /* flush_vgt_streamout */
1408 6 + /* set_streamout_enable */
1409 util_bitcount(buffer_en) * 7 + /* SET_CONTEXT_REG */
1410 (ctx->chip_class == R700 ? util_bitcount(buffer_en) * 5 : 0) + /* STRMOUT_BASE_UPDATE */
1411 util_bitcount(buffer_en & ctx->streamout_append_bitmask) * 8 + /* STRMOUT_BUFFER_UPDATE */
1412 util_bitcount(buffer_en & ~ctx->streamout_append_bitmask) * 6 + /* STRMOUT_BUFFER_UPDATE */
1413 (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770 ? 2 : 0) + /* SURFACE_BASE_UPDATE */
1414 ctx->num_cs_dw_streamout_end, TRUE);
1415
1416 if (ctx->chip_class >= EVERGREEN) {
1417 evergreen_flush_vgt_streamout(ctx);
1418 evergreen_set_streamout_enable(ctx, buffer_en);
1419 } else {
1420 r600_flush_vgt_streamout(ctx);
1421 r600_set_streamout_enable(ctx, buffer_en);
1422 }
1423
1424 for (i = 0; i < ctx->num_so_targets; i++) {
1425 if (t[i]) {
1426 t[i]->stride_in_dw = stride_in_dw[i];
1427 t[i]->so_index = i;
1428 va = r600_resource_va(&ctx->screen->screen,
1429 (void*)t[i]->b.buffer);
1430
1431 update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i);
1432
1433 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 3, 0);
1434 cs->buf[cs->cdw++] = (R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 +
1435 16*i - R600_CONTEXT_REG_OFFSET) >> 2;
1436 cs->buf[cs->cdw++] = (t[i]->b.buffer_offset +
1437 t[i]->b.buffer_size) >> 2; /* BUFFER_SIZE (in DW) */
1438 cs->buf[cs->cdw++] = stride_in_dw[i]; /* VTX_STRIDE (in DW) */
1439 cs->buf[cs->cdw++] = va >> 8; /* BUFFER_BASE */
1440
1441 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1442 cs->buf[cs->cdw++] =
1443 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1444 RADEON_USAGE_WRITE);
1445
1446 /* R7xx requires this packet after updating BUFFER_BASE.
1447 * Without this, R7xx locks up. */
1448 if (ctx->chip_class == R700) {
1449 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BASE_UPDATE, 1, 0);
1450 cs->buf[cs->cdw++] = i;
1451 cs->buf[cs->cdw++] = va >> 8;
1452
1453 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1454 cs->buf[cs->cdw++] =
1455 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1456 RADEON_USAGE_WRITE);
1457 }
1458
1459 if (ctx->streamout_append_bitmask & (1 << i)) {
1460 va = r600_resource_va(&ctx->screen->screen,
1461 (void*)t[i]->filled_size);
1462 /* Append. */
1463 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1464 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1465 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_MEM); /* control */
1466 cs->buf[cs->cdw++] = 0; /* unused */
1467 cs->buf[cs->cdw++] = 0; /* unused */
1468 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* src address lo */
1469 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* src address hi */
1470
1471 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1472 cs->buf[cs->cdw++] =
1473 r600_context_bo_reloc(ctx, t[i]->filled_size,
1474 RADEON_USAGE_READ);
1475 } else {
1476 /* Start from the beginning. */
1477 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1478 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1479 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_PACKET); /* control */
1480 cs->buf[cs->cdw++] = 0; /* unused */
1481 cs->buf[cs->cdw++] = 0; /* unused */
1482 cs->buf[cs->cdw++] = t[i]->b.buffer_offset >> 2; /* buffer offset in DW */
1483 cs->buf[cs->cdw++] = 0; /* unused */
1484 }
1485 }
1486 }
1487
1488 if (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770) {
1489 cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
1490 cs->buf[cs->cdw++] = update_flags;
1491 }
1492 }
1493
1494 void r600_context_streamout_end(struct r600_context *ctx)
1495 {
1496 struct radeon_winsys_cs *cs = ctx->cs;
1497 struct r600_so_target **t = ctx->so_targets;
1498 unsigned i, flush_flags = 0;
1499 uint64_t va;
1500
1501 if (ctx->chip_class >= EVERGREEN) {
1502 evergreen_flush_vgt_streamout(ctx);
1503 } else {
1504 r600_flush_vgt_streamout(ctx);
1505 }
1506
1507 for (i = 0; i < ctx->num_so_targets; i++) {
1508 if (t[i]) {
1509 va = r600_resource_va(&ctx->screen->screen,
1510 (void*)t[i]->filled_size);
1511 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1512 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1513 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_NONE) |
1514 STRMOUT_STORE_BUFFER_FILLED_SIZE; /* control */
1515 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* dst address lo */
1516 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* dst address hi */
1517 cs->buf[cs->cdw++] = 0; /* unused */
1518 cs->buf[cs->cdw++] = 0; /* unused */
1519
1520 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1521 cs->buf[cs->cdw++] =
1522 r600_context_bo_reloc(ctx, t[i]->filled_size,
1523 RADEON_USAGE_WRITE);
1524
1525 flush_flags |= S_0085F0_SO0_DEST_BASE_ENA(1) << i;
1526 }
1527 }
1528
1529 if (ctx->chip_class >= EVERGREEN) {
1530 evergreen_set_streamout_enable(ctx, 0);
1531 } else {
1532 r600_set_streamout_enable(ctx, 0);
1533 }
1534
1535 /* This is needed to fix cache flushes on r600. */
1536 if (ctx->chip_class == R600) {
1537 if (ctx->family == CHIP_RV670 ||
1538 ctx->family == CHIP_RS780 ||
1539 ctx->family == CHIP_RS880) {
1540 flush_flags |= S_0085F0_DEST_BASE_0_ENA(1);
1541 }
1542
1543 r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
1544 }
1545
1546 /* Flush streamout caches. */
1547 ctx->surface_sync_cmd.flush_flags |=
1548 S_0085F0_SMX_ACTION_ENA(1) | flush_flags;
1549 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1550
1551 ctx->num_cs_dw_streamout_end = 0;
1552
1553 #if 0
1554 for (i = 0; i < ctx->num_so_targets; i++) {
1555 if (!t[i])
1556 continue;
1557
1558 uint32_t *ptr = ctx->ws->buffer_map(t[i]->filled_size->buf, ctx->cs, RADEON_USAGE_READ);
1559 printf("FILLED_SIZE%i: %u\n", i, *ptr);
1560 ctx->ws->buffer_unmap(t[i]->filled_size->buf);
1561 }
1562 #endif
1563 }
1564
1565 void r600_context_draw_opaque_count(struct r600_context *ctx, struct r600_so_target *t)
1566 {
1567 struct radeon_winsys_cs *cs = ctx->cs;
1568 uint64_t va = r600_resource_va(&ctx->screen->screen,
1569 (void*)t->filled_size);
1570
1571 r600_need_cs_space(ctx, 14 + 21, TRUE);
1572
1573 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1574 cs->buf[cs->cdw++] = (R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET - R600_CONTEXT_REG_OFFSET) >> 2;
1575 cs->buf[cs->cdw++] = 0;
1576
1577 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1578 cs->buf[cs->cdw++] = (R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE - R600_CONTEXT_REG_OFFSET) >> 2;
1579 cs->buf[cs->cdw++] = t->stride_in_dw;
1580
1581 cs->buf[cs->cdw++] = PKT3(PKT3_COPY_DW, 4, 0);
1582 cs->buf[cs->cdw++] = COPY_DW_SRC_IS_MEM | COPY_DW_DST_IS_REG;
1583 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* src address lo */
1584 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* src address hi */
1585 cs->buf[cs->cdw++] = R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2; /* dst register */
1586 cs->buf[cs->cdw++] = 0; /* unused */
1587
1588 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1589 cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, t->filled_size, RADEON_USAGE_READ);
1590 }