projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
r600g: atomize stencil ref state
[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 static void r600_init_block(struct r600_context *ctx,
118 struct r600_block *block,
119 const struct r600_reg *reg, int index, int nreg,
120 unsigned opcode, unsigned offset_base)
121 {
122 int i = index;
123 int j, n = nreg;
124
125 /* initialize block */
126 block->flags = 0;
127 block->status |= R600_BLOCK_STATUS_DIRTY; /* dirty all blocks at start */
128 block->start_offset = reg[i].offset;
129 block->pm4[block->pm4_ndwords++] = PKT3(opcode, n, 0);
130 block->pm4[block->pm4_ndwords++] = (block->start_offset - offset_base) >> 2;
131 block->reg = &block->pm4[block->pm4_ndwords];
132 block->pm4_ndwords += n;
133 block->nreg = n;
134 block->nreg_dirty = n;
135 LIST_INITHEAD(&block->list);
136 LIST_INITHEAD(&block->enable_list);
137
138 for (j = 0; j < n; j++) {
139 if (reg[i+j].flags & REG_FLAG_DIRTY_ALWAYS) {
140 block->flags |= REG_FLAG_DIRTY_ALWAYS;
141 }
142 if (reg[i+j].flags & REG_FLAG_ENABLE_ALWAYS) {
143 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
144 block->status |= R600_BLOCK_STATUS_ENABLED;
145 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
146 LIST_ADDTAIL(&block->list,&ctx->dirty);
147 }
148 }
149 if (reg[i+j].flags & REG_FLAG_FLUSH_CHANGE) {
150 block->flags |= REG_FLAG_FLUSH_CHANGE;
151 }
152
153 if (reg[i+j].flags & REG_FLAG_NEED_BO) {
154 block->nbo++;
155 assert(block->nbo < R600_BLOCK_MAX_BO);
156 block->pm4_bo_index[j] = block->nbo;
157 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0, 0);
158 block->pm4[block->pm4_ndwords++] = 0x00000000;
159 block->reloc[block->nbo].bo_pm4_index = block->pm4_ndwords - 1;
160 }
161 if ((ctx->family > CHIP_R600) &&
162 (ctx->family < CHIP_RV770) && reg[i+j].flags & REG_FLAG_RV6XX_SBU) {
163 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
164 block->pm4[block->pm4_ndwords++] = reg[i+j].sbu_flags;
165 }
166 }
167 /* check that we stay in limit */
168 assert(block->pm4_ndwords < R600_BLOCK_MAX_REG);
169 }
170
171 int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg,
172 unsigned opcode, unsigned offset_base)
173 {
174 struct r600_block *block;
175 struct r600_range *range;
176 int offset;
177
178 for (unsigned i = 0, n = 0; i < nreg; i += n) {
179 /* ignore new block balise */
180 if (reg[i].offset == GROUP_FORCE_NEW_BLOCK) {
181 n = 1;
182 continue;
183 }
184
185 /* ignore regs not on R600 on R600 */
186 if ((reg[i].flags & REG_FLAG_NOT_R600) && ctx->family == CHIP_R600) {
187 n = 1;
188 continue;
189 }
190
191 /* register that need relocation are in their own group */
192 /* find number of consecutive registers */
193 n = 0;
194 offset = reg[i].offset;
195 while (reg[i + n].offset == offset) {
196 n++;
197 offset += 4;
198 if ((n + i) >= nreg)
199 break;
200 if (n >= (R600_BLOCK_MAX_REG - 2))
201 break;
202 }
203
204 /* allocate new block */
205 block = calloc(1, sizeof(struct r600_block));
206 if (block == NULL) {
207 return -ENOMEM;
208 }
209 ctx->nblocks++;
210 for (int j = 0; j < n; j++) {
211 range = &ctx->range[CTX_RANGE_ID(reg[i + j].offset)];
212 /* create block table if it doesn't exist */
213 if (!range->blocks)
214 range->blocks = calloc(1 << HASH_SHIFT, sizeof(void *));
215 if (!range->blocks)
216 return -1;
217
218 range->blocks[CTX_BLOCK_ID(reg[i + j].offset)] = block;
219 }
220
221 r600_init_block(ctx, block, reg, i, n, opcode, offset_base);
222
223 }
224 return 0;
225 }
226
227 /* R600/R700 configuration */
228 static const struct r600_reg r600_config_reg_list[] = {
229 {R_008958_VGT_PRIMITIVE_TYPE, 0, 0},
230 {R_008B40_PA_SC_AA_SAMPLE_LOCS_2S, 0, 0},
231 {R_008B44_PA_SC_AA_SAMPLE_LOCS_4S, 0, 0},
232 {R_008B48_PA_SC_AA_SAMPLE_LOCS_8S_WD0, 0, 0},
233 {R_008B4C_PA_SC_AA_SAMPLE_LOCS_8S_WD1, 0, 0},
234 {R_008C04_SQ_GPR_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0},
235 };
236
237 static const struct r600_reg r600_ctl_const_list[] = {
238 {R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0},
239 };
240
241 static const struct r600_reg r600_context_reg_list[] = {
242 {R_028A4C_PA_SC_MODE_CNTL, 0, 0},
243 {GROUP_FORCE_NEW_BLOCK, 0, 0},
244 {R_028040_CB_COLOR0_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(0)},
245 {GROUP_FORCE_NEW_BLOCK, 0, 0},
246 {R_0280A0_CB_COLOR0_INFO, REG_FLAG_NEED_BO, 0},
247 {R_028060_CB_COLOR0_SIZE, 0, 0},
248 {R_028080_CB_COLOR0_VIEW, 0, 0},
249 {GROUP_FORCE_NEW_BLOCK, 0, 0},
250 {R_0280E0_CB_COLOR0_FRAG, REG_FLAG_NEED_BO, 0},
251 {GROUP_FORCE_NEW_BLOCK, 0, 0},
252 {R_0280C0_CB_COLOR0_TILE, REG_FLAG_NEED_BO, 0},
253 {GROUP_FORCE_NEW_BLOCK, 0, 0},
254 {R_028100_CB_COLOR0_MASK, 0, 0},
255 {R_028044_CB_COLOR1_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(1)},
256 {GROUP_FORCE_NEW_BLOCK, 0, 0},
257 {R_0280A4_CB_COLOR1_INFO, REG_FLAG_NEED_BO, 0},
258 {R_028064_CB_COLOR1_SIZE, 0, 0},
259 {R_028084_CB_COLOR1_VIEW, 0, 0},
260 {GROUP_FORCE_NEW_BLOCK, 0, 0},
261 {R_0280E4_CB_COLOR1_FRAG, REG_FLAG_NEED_BO, 0},
262 {GROUP_FORCE_NEW_BLOCK, 0, 0},
263 {R_0280C4_CB_COLOR1_TILE, REG_FLAG_NEED_BO, 0},
264 {R_028104_CB_COLOR1_MASK, 0, 0},
265 {GROUP_FORCE_NEW_BLOCK, 0, 0},
266 {R_028048_CB_COLOR2_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(2)},
267 {GROUP_FORCE_NEW_BLOCK, 0, 0},
268 {R_0280A8_CB_COLOR2_INFO, REG_FLAG_NEED_BO, 0},
269 {R_028068_CB_COLOR2_SIZE, 0, 0},
270 {R_028088_CB_COLOR2_VIEW, 0, 0},
271 {GROUP_FORCE_NEW_BLOCK, 0, 0},
272 {R_0280E8_CB_COLOR2_FRAG, REG_FLAG_NEED_BO, 0},
273 {GROUP_FORCE_NEW_BLOCK, 0, 0},
274 {R_0280C8_CB_COLOR2_TILE, REG_FLAG_NEED_BO, 0},
275 {R_028108_CB_COLOR2_MASK, 0, 0},
276 {GROUP_FORCE_NEW_BLOCK, 0, 0},
277 {R_02804C_CB_COLOR3_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(3)},
278 {GROUP_FORCE_NEW_BLOCK, 0, 0},
279 {R_0280AC_CB_COLOR3_INFO, REG_FLAG_NEED_BO, 0},
280 {R_02806C_CB_COLOR3_SIZE, 0, 0},
281 {R_02808C_CB_COLOR3_VIEW, 0, 0},
282 {GROUP_FORCE_NEW_BLOCK, 0, 0},
283 {R_0280EC_CB_COLOR3_FRAG, REG_FLAG_NEED_BO, 0},
284 {GROUP_FORCE_NEW_BLOCK, 0, 0},
285 {R_0280CC_CB_COLOR3_TILE, REG_FLAG_NEED_BO, 0},
286 {R_02810C_CB_COLOR3_MASK, 0, 0},
287 {GROUP_FORCE_NEW_BLOCK, 0, 0},
288 {R_028050_CB_COLOR4_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(4)},
289 {GROUP_FORCE_NEW_BLOCK, 0, 0},
290 {R_0280B0_CB_COLOR4_INFO, REG_FLAG_NEED_BO, 0},
291 {R_028070_CB_COLOR4_SIZE, 0, 0},
292 {R_028090_CB_COLOR4_VIEW, 0, 0},
293 {GROUP_FORCE_NEW_BLOCK, 0, 0},
294 {R_0280F0_CB_COLOR4_FRAG, REG_FLAG_NEED_BO, 0},
295 {GROUP_FORCE_NEW_BLOCK, 0, 0},
296 {R_0280D0_CB_COLOR4_TILE, REG_FLAG_NEED_BO, 0},
297 {R_028110_CB_COLOR4_MASK, 0, 0},
298 {GROUP_FORCE_NEW_BLOCK, 0, 0},
299 {R_028054_CB_COLOR5_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(5)},
300 {GROUP_FORCE_NEW_BLOCK, 0, 0},
301 {R_0280B4_CB_COLOR5_INFO, REG_FLAG_NEED_BO, 0},
302 {R_028074_CB_COLOR5_SIZE, 0, 0},
303 {R_028094_CB_COLOR5_VIEW, 0, 0},
304 {GROUP_FORCE_NEW_BLOCK, 0, 0},
305 {R_0280F4_CB_COLOR5_FRAG, REG_FLAG_NEED_BO, 0},
306 {GROUP_FORCE_NEW_BLOCK, 0, 0},
307 {R_0280D4_CB_COLOR5_TILE, REG_FLAG_NEED_BO, 0},
308 {R_028114_CB_COLOR5_MASK, 0, 0},
309 {R_028058_CB_COLOR6_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(6)},
310 {R_0280B8_CB_COLOR6_INFO, REG_FLAG_NEED_BO, 0},
311 {R_028078_CB_COLOR6_SIZE, 0, 0},
312 {R_028098_CB_COLOR6_VIEW, 0, 0},
313 {GROUP_FORCE_NEW_BLOCK, 0, 0},
314 {R_0280F8_CB_COLOR6_FRAG, REG_FLAG_NEED_BO, 0},
315 {GROUP_FORCE_NEW_BLOCK, 0, 0},
316 {R_0280D8_CB_COLOR6_TILE, REG_FLAG_NEED_BO, 0},
317 {R_028118_CB_COLOR6_MASK, 0, 0},
318 {GROUP_FORCE_NEW_BLOCK, 0, 0},
319 {R_02805C_CB_COLOR7_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(7)},
320 {GROUP_FORCE_NEW_BLOCK, 0, 0},
321 {R_0280BC_CB_COLOR7_INFO, REG_FLAG_NEED_BO, 0},
322 {R_02807C_CB_COLOR7_SIZE, 0, 0},
323 {R_02809C_CB_COLOR7_VIEW, 0, 0},
324 {R_0280FC_CB_COLOR7_FRAG, REG_FLAG_NEED_BO, 0},
325 {R_0280DC_CB_COLOR7_TILE, REG_FLAG_NEED_BO, 0},
326 {R_02811C_CB_COLOR7_MASK, 0, 0},
327 {R_028120_CB_CLEAR_RED, 0, 0},
328 {R_028124_CB_CLEAR_GREEN, 0, 0},
329 {R_028128_CB_CLEAR_BLUE, 0, 0},
330 {R_02812C_CB_CLEAR_ALPHA, 0, 0},
331 {R_028414_CB_BLEND_RED, 0, 0},
332 {R_028418_CB_BLEND_GREEN, 0, 0},
333 {R_02841C_CB_BLEND_BLUE, 0, 0},
334 {R_028420_CB_BLEND_ALPHA, 0, 0},
335 {R_028424_CB_FOG_RED, 0, 0},
336 {R_028428_CB_FOG_GREEN, 0, 0},
337 {R_02842C_CB_FOG_BLUE, 0, 0},
338 {R_028780_CB_BLEND0_CONTROL, REG_FLAG_NOT_R600, 0},
339 {R_028784_CB_BLEND1_CONTROL, REG_FLAG_NOT_R600, 0},
340 {R_028788_CB_BLEND2_CONTROL, REG_FLAG_NOT_R600, 0},
341 {R_02878C_CB_BLEND3_CONTROL, REG_FLAG_NOT_R600, 0},
342 {R_028790_CB_BLEND4_CONTROL, REG_FLAG_NOT_R600, 0},
343 {R_028794_CB_BLEND5_CONTROL, REG_FLAG_NOT_R600, 0},
344 {R_028798_CB_BLEND6_CONTROL, REG_FLAG_NOT_R600, 0},
345 {R_02879C_CB_BLEND7_CONTROL, REG_FLAG_NOT_R600, 0},
346 {R_0287A0_CB_SHADER_CONTROL, 0, 0},
347 {R_028800_DB_DEPTH_CONTROL, 0, 0},
348 {R_028804_CB_BLEND_CONTROL, 0, 0},
349 {R_02880C_DB_SHADER_CONTROL, 0, 0},
350 {R_02800C_DB_DEPTH_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_DEPTH},
351 {R_028000_DB_DEPTH_SIZE, 0, 0},
352 {R_028004_DB_DEPTH_VIEW, 0, 0},
353 {GROUP_FORCE_NEW_BLOCK, 0, 0},
354 {R_028010_DB_DEPTH_INFO, REG_FLAG_NEED_BO, 0},
355 {R_028A6C_VGT_GS_OUT_PRIM_TYPE, 0, 0},
356 {R_028D24_DB_HTILE_SURFACE, 0, 0},
357 {R_028D34_DB_PREFETCH_LIMIT, 0, 0},
358 {R_028D44_DB_ALPHA_TO_MASK, 0, 0},
359 {R_028204_PA_SC_WINDOW_SCISSOR_TL, 0, 0},
360 {R_028208_PA_SC_WINDOW_SCISSOR_BR, 0, 0},
361 {R_028250_PA_SC_VPORT_SCISSOR_0_TL, 0, 0},
362 {R_028254_PA_SC_VPORT_SCISSOR_0_BR, 0, 0},
363 {R_02843C_PA_CL_VPORT_XSCALE_0, 0, 0},
364 {R_028440_PA_CL_VPORT_XOFFSET_0, 0, 0},
365 {R_028444_PA_CL_VPORT_YSCALE_0, 0, 0},
366 {R_028448_PA_CL_VPORT_YOFFSET_0, 0, 0},
367 {R_02844C_PA_CL_VPORT_ZSCALE_0, 0, 0},
368 {R_028450_PA_CL_VPORT_ZOFFSET_0, 0, 0},
369 {R_0286D4_SPI_INTERP_CONTROL_0, 0, 0},
370 {R_028810_PA_CL_CLIP_CNTL, 0, 0},
371 {R_028814_PA_SU_SC_MODE_CNTL, 0, 0},
372 {R_02881C_PA_CL_VS_OUT_CNTL, 0, 0},
373 {R_028A00_PA_SU_POINT_SIZE, 0, 0},
374 {R_028A04_PA_SU_POINT_MINMAX, 0, 0},
375 {R_028A08_PA_SU_LINE_CNTL, 0, 0},
376 {R_028A0C_PA_SC_LINE_STIPPLE, 0, 0},
377 {R_028C00_PA_SC_LINE_CNTL, 0, 0},
378 {R_028C04_PA_SC_AA_CONFIG, 0, 0},
379 {R_028C08_PA_SU_VTX_CNTL, 0, 0},
380 {R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL, 0, 0},
381 {R_028DFC_PA_SU_POLY_OFFSET_CLAMP, 0, 0},
382 {R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 0, 0},
383 {R_028E04_PA_SU_POLY_OFFSET_FRONT_OFFSET, 0, 0},
384 {R_028E08_PA_SU_POLY_OFFSET_BACK_SCALE, 0, 0},
385 {R_028E0C_PA_SU_POLY_OFFSET_BACK_OFFSET, 0, 0},
386 {R_028E20_PA_CL_UCP0_X, 0, 0},
387 {R_028E24_PA_CL_UCP0_Y, 0, 0},
388 {R_028E28_PA_CL_UCP0_Z, 0, 0},
389 {R_028E2C_PA_CL_UCP0_W, 0, 0},
390 {R_028E30_PA_CL_UCP1_X, 0, 0},
391 {R_028E34_PA_CL_UCP1_Y, 0, 0},
392 {R_028E38_PA_CL_UCP1_Z, 0, 0},
393 {R_028E3C_PA_CL_UCP1_W, 0, 0},
394 {R_028E40_PA_CL_UCP2_X, 0, 0},
395 {R_028E44_PA_CL_UCP2_Y, 0, 0},
396 {R_028E48_PA_CL_UCP2_Z, 0, 0},
397 {R_028E4C_PA_CL_UCP2_W, 0, 0},
398 {R_028E50_PA_CL_UCP3_X, 0, 0},
399 {R_028E54_PA_CL_UCP3_Y, 0, 0},
400 {R_028E58_PA_CL_UCP3_Z, 0, 0},
401 {R_028E5C_PA_CL_UCP3_W, 0, 0},
402 {R_028E60_PA_CL_UCP4_X, 0, 0},
403 {R_028E64_PA_CL_UCP4_Y, 0, 0},
404 {R_028E68_PA_CL_UCP4_Z, 0, 0},
405 {R_028E6C_PA_CL_UCP4_W, 0, 0},
406 {R_028E70_PA_CL_UCP5_X, 0, 0},
407 {R_028E74_PA_CL_UCP5_Y, 0, 0},
408 {R_028E78_PA_CL_UCP5_Z, 0, 0},
409 {R_028E7C_PA_CL_UCP5_W, 0, 0},
410 {R_028350_SX_MISC, 0, 0},
411 {R_028380_SQ_VTX_SEMANTIC_0, 0, 0},
412 {R_028384_SQ_VTX_SEMANTIC_1, 0, 0},
413 {R_028388_SQ_VTX_SEMANTIC_2, 0, 0},
414 {R_02838C_SQ_VTX_SEMANTIC_3, 0, 0},
415 {R_028390_SQ_VTX_SEMANTIC_4, 0, 0},
416 {R_028394_SQ_VTX_SEMANTIC_5, 0, 0},
417 {R_028398_SQ_VTX_SEMANTIC_6, 0, 0},
418 {R_02839C_SQ_VTX_SEMANTIC_7, 0, 0},
419 {R_0283A0_SQ_VTX_SEMANTIC_8, 0, 0},
420 {R_0283A4_SQ_VTX_SEMANTIC_9, 0, 0},
421 {R_0283A8_SQ_VTX_SEMANTIC_10, 0, 0},
422 {R_0283AC_SQ_VTX_SEMANTIC_11, 0, 0},
423 {R_0283B0_SQ_VTX_SEMANTIC_12, 0, 0},
424 {R_0283B4_SQ_VTX_SEMANTIC_13, 0, 0},
425 {R_0283B8_SQ_VTX_SEMANTIC_14, 0, 0},
426 {R_0283BC_SQ_VTX_SEMANTIC_15, 0, 0},
427 {R_0283C0_SQ_VTX_SEMANTIC_16, 0, 0},
428 {R_0283C4_SQ_VTX_SEMANTIC_17, 0, 0},
429 {R_0283C8_SQ_VTX_SEMANTIC_18, 0, 0},
430 {R_0283CC_SQ_VTX_SEMANTIC_19, 0, 0},
431 {R_0283D0_SQ_VTX_SEMANTIC_20, 0, 0},
432 {R_0283D4_SQ_VTX_SEMANTIC_21, 0, 0},
433 {R_0283D8_SQ_VTX_SEMANTIC_22, 0, 0},
434 {R_0283DC_SQ_VTX_SEMANTIC_23, 0, 0},
435 {R_0283E0_SQ_VTX_SEMANTIC_24, 0, 0},
436 {R_0283E4_SQ_VTX_SEMANTIC_25, 0, 0},
437 {R_0283E8_SQ_VTX_SEMANTIC_26, 0, 0},
438 {R_0283EC_SQ_VTX_SEMANTIC_27, 0, 0},
439 {R_0283F0_SQ_VTX_SEMANTIC_28, 0, 0},
440 {R_0283F4_SQ_VTX_SEMANTIC_29, 0, 0},
441 {R_0283F8_SQ_VTX_SEMANTIC_30, 0, 0},
442 {R_0283FC_SQ_VTX_SEMANTIC_31, 0, 0},
443 {R_028614_SPI_VS_OUT_ID_0, 0, 0},
444 {R_028618_SPI_VS_OUT_ID_1, 0, 0},
445 {R_02861C_SPI_VS_OUT_ID_2, 0, 0},
446 {R_028620_SPI_VS_OUT_ID_3, 0, 0},
447 {R_028624_SPI_VS_OUT_ID_4, 0, 0},
448 {R_028628_SPI_VS_OUT_ID_5, 0, 0},
449 {R_02862C_SPI_VS_OUT_ID_6, 0, 0},
450 {R_028630_SPI_VS_OUT_ID_7, 0, 0},
451 {R_028634_SPI_VS_OUT_ID_8, 0, 0},
452 {R_028638_SPI_VS_OUT_ID_9, 0, 0},
453 {R_0286C4_SPI_VS_OUT_CONFIG, 0, 0},
454 {GROUP_FORCE_NEW_BLOCK, 0, 0},
455 {R_028858_SQ_PGM_START_VS, REG_FLAG_NEED_BO, 0},
456 {GROUP_FORCE_NEW_BLOCK, 0, 0},
457 {R_028868_SQ_PGM_RESOURCES_VS, 0, 0},
458 {GROUP_FORCE_NEW_BLOCK, 0, 0},
459 {R_028894_SQ_PGM_START_FS, REG_FLAG_NEED_BO, 0},
460 {GROUP_FORCE_NEW_BLOCK, 0, 0},
461 {R_0288A4_SQ_PGM_RESOURCES_FS, 0, 0},
462 {R_0288DC_SQ_PGM_CF_OFFSET_FS, 0, 0},
463 {R_028644_SPI_PS_INPUT_CNTL_0, 0, 0},
464 {R_028648_SPI_PS_INPUT_CNTL_1, 0, 0},
465 {R_02864C_SPI_PS_INPUT_CNTL_2, 0, 0},
466 {R_028650_SPI_PS_INPUT_CNTL_3, 0, 0},
467 {R_028654_SPI_PS_INPUT_CNTL_4, 0, 0},
468 {R_028658_SPI_PS_INPUT_CNTL_5, 0, 0},
469 {R_02865C_SPI_PS_INPUT_CNTL_6, 0, 0},
470 {R_028660_SPI_PS_INPUT_CNTL_7, 0, 0},
471 {R_028664_SPI_PS_INPUT_CNTL_8, 0, 0},
472 {R_028668_SPI_PS_INPUT_CNTL_9, 0, 0},
473 {R_02866C_SPI_PS_INPUT_CNTL_10, 0, 0},
474 {R_028670_SPI_PS_INPUT_CNTL_11, 0, 0},
475 {R_028674_SPI_PS_INPUT_CNTL_12, 0, 0},
476 {R_028678_SPI_PS_INPUT_CNTL_13, 0, 0},
477 {R_02867C_SPI_PS_INPUT_CNTL_14, 0, 0},
478 {R_028680_SPI_PS_INPUT_CNTL_15, 0, 0},
479 {R_028684_SPI_PS_INPUT_CNTL_16, 0, 0},
480 {R_028688_SPI_PS_INPUT_CNTL_17, 0, 0},
481 {R_02868C_SPI_PS_INPUT_CNTL_18, 0, 0},
482 {R_028690_SPI_PS_INPUT_CNTL_19, 0, 0},
483 {R_028694_SPI_PS_INPUT_CNTL_20, 0, 0},
484 {R_028698_SPI_PS_INPUT_CNTL_21, 0, 0},
485 {R_02869C_SPI_PS_INPUT_CNTL_22, 0, 0},
486 {R_0286A0_SPI_PS_INPUT_CNTL_23, 0, 0},
487 {R_0286A4_SPI_PS_INPUT_CNTL_24, 0, 0},
488 {R_0286A8_SPI_PS_INPUT_CNTL_25, 0, 0},
489 {R_0286AC_SPI_PS_INPUT_CNTL_26, 0, 0},
490 {R_0286B0_SPI_PS_INPUT_CNTL_27, 0, 0},
491 {R_0286B4_SPI_PS_INPUT_CNTL_28, 0, 0},
492 {R_0286B8_SPI_PS_INPUT_CNTL_29, 0, 0},
493 {R_0286BC_SPI_PS_INPUT_CNTL_30, 0, 0},
494 {R_0286C0_SPI_PS_INPUT_CNTL_31, 0, 0},
495 {R_0286CC_SPI_PS_IN_CONTROL_0, 0, 0},
496 {R_0286D0_SPI_PS_IN_CONTROL_1, 0, 0},
497 {R_0286D8_SPI_INPUT_Z, 0, 0},
498 {GROUP_FORCE_NEW_BLOCK, 0, 0},
499 {R_028840_SQ_PGM_START_PS, REG_FLAG_NEED_BO, 0},
500 {GROUP_FORCE_NEW_BLOCK, 0, 0},
501 {R_028850_SQ_PGM_RESOURCES_PS, 0, 0},
502 {R_028854_SQ_PGM_EXPORTS_PS, 0, 0},
503 {R_028408_VGT_INDX_OFFSET, 0, 0},
504 {R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, 0, 0},
505 {R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, 0, 0},
506 {R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 0, 0},
507 {R_028C20_PA_SC_AA_SAMPLE_LOCS_8S_WD1_MCTX, 0, 0},
508 };
509
510 static int r600_loop_const_init(struct r600_context *ctx, uint32_t offset)
511 {
512 unsigned nreg = 32;
513 struct r600_reg r600_loop_consts[32];
514 int i;
515
516 for (i = 0; i < nreg; i++) {
517 r600_loop_consts[i].offset = R600_LOOP_CONST_OFFSET + ((offset + i) * 4);
518 r600_loop_consts[i].flags = REG_FLAG_DIRTY_ALWAYS;
519 r600_loop_consts[i].sbu_flags = 0;
520 }
521 return r600_context_add_block(ctx, r600_loop_consts, nreg, PKT3_SET_LOOP_CONST, R600_LOOP_CONST_OFFSET);
522 }
523
524 /* initialize */
525 void r600_context_fini(struct r600_context *ctx)
526 {
527 struct r600_block *block;
528 struct r600_range *range;
529
530 if (ctx->range) {
531 for (int i = 0; i < NUM_RANGES; i++) {
532 if (!ctx->range[i].blocks)
533 continue;
534 for (int j = 0; j < (1 << HASH_SHIFT); j++) {
535 block = ctx->range[i].blocks[j];
536 if (block) {
537 for (int k = 0, offset = block->start_offset; k < block->nreg; k++, offset += 4) {
538 range = &ctx->range[CTX_RANGE_ID(offset)];
539 range->blocks[CTX_BLOCK_ID(offset)] = NULL;
540 }
541 for (int k = 1; k <= block->nbo; k++) {
542 pipe_resource_reference((struct pipe_resource**)&block->reloc[k].bo, NULL);
543 }
544 free(block);
545 }
546 }
547 free(ctx->range[i].blocks);
548 }
549 }
550 free(ctx->blocks);
551 }
552
553 int r600_setup_block_table(struct r600_context *ctx)
554 {
555 /* setup block table */
556 int c = 0;
557 ctx->blocks = calloc(ctx->nblocks, sizeof(void*));
558 if (!ctx->blocks)
559 return -ENOMEM;
560 for (int i = 0; i < NUM_RANGES; i++) {
561 if (!ctx->range[i].blocks)
562 continue;
563 for (int j = 0, add; j < (1 << HASH_SHIFT); j++) {
564 if (!ctx->range[i].blocks[j])
565 continue;
566
567 add = 1;
568 for (int k = 0; k < c; k++) {
569 if (ctx->blocks[k] == ctx->range[i].blocks[j]) {
570 add = 0;
571 break;
572 }
573 }
574 if (add) {
575 assert(c < ctx->nblocks);
576 ctx->blocks[c++] = ctx->range[i].blocks[j];
577 j += (ctx->range[i].blocks[j]->nreg) - 1;
578 }
579 }
580 }
581 return 0;
582 }
583
584 int r600_context_init(struct r600_context *ctx)
585 {
586 int r;
587
588 /* add blocks */
589 r = r600_context_add_block(ctx, r600_config_reg_list,
590 Elements(r600_config_reg_list), PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
591 if (r)
592 goto out_err;
593 r = r600_context_add_block(ctx, r600_context_reg_list,
594 Elements(r600_context_reg_list), PKT3_SET_CONTEXT_REG, R600_CONTEXT_REG_OFFSET);
595 if (r)
596 goto out_err;
597 r = r600_context_add_block(ctx, r600_ctl_const_list,
598 Elements(r600_ctl_const_list), PKT3_SET_CTL_CONST, R600_CTL_CONST_OFFSET);
599 if (r)
600 goto out_err;
601
602 /* PS loop const */
603 r600_loop_const_init(ctx, 0);
604 /* VS loop const */
605 r600_loop_const_init(ctx, 32);
606
607 r = r600_setup_block_table(ctx);
608 if (r)
609 goto out_err;
610
611 ctx->max_db = 4;
612 return 0;
613 out_err:
614 r600_context_fini(ctx);
615 return r;
616 }
617
618 void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
619 boolean count_draw_in)
620 {
621 /* The number of dwords we already used in the CS so far. */
622 num_dw += ctx->cs->cdw;
623
624 if (count_draw_in) {
625 unsigned i;
626
627 /* The number of dwords all the dirty states would take. */
628 for (i = 0; i < R600_NUM_ATOMS; i++) {
629 if (ctx->atoms[i] && ctx->atoms[i]->dirty) {
630 num_dw += ctx->atoms[i]->num_dw;
631 }
632 }
633
634 num_dw += ctx->pm4_dirty_cdwords;
635
636 /* The upper-bound of how much a draw command would take. */
637 num_dw += R600_MAX_DRAW_CS_DWORDS;
638 }
639
640 /* Count in queries_suspend. */
641 num_dw += ctx->num_cs_dw_nontimer_queries_suspend;
642 num_dw += ctx->num_cs_dw_timer_queries_suspend;
643
644 /* Count in streamout_end at the end of CS. */
645 num_dw += ctx->num_cs_dw_streamout_end;
646
647 /* Count in render_condition(NULL) at the end of CS. */
648 if (ctx->predicate_drawing) {
649 num_dw += 3;
650 }
651
652 /* Count in framebuffer cache flushes at the end of CS. */
653 num_dw += 44; /* one SURFACE_SYNC and CACHE_FLUSH_AND_INV (r6xx-only) */
654
655 /* Save 16 dwords for the fence mechanism. */
656 num_dw += 16;
657
658 /* Flush if there's not enough space. */
659 if (num_dw > RADEON_MAX_CMDBUF_DWORDS) {
660 r600_flush(&ctx->context, NULL, RADEON_FLUSH_ASYNC);
661 }
662 }
663
664 void r600_context_dirty_block(struct r600_context *ctx,
665 struct r600_block *block,
666 int dirty, int index)
667 {
668 if ((index + 1) > block->nreg_dirty)
669 block->nreg_dirty = index + 1;
670
671 if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
672 block->status |= R600_BLOCK_STATUS_DIRTY;
673 ctx->pm4_dirty_cdwords += block->pm4_ndwords;
674 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
675 block->status |= R600_BLOCK_STATUS_ENABLED;
676 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
677 }
678 LIST_ADDTAIL(&block->list,&ctx->dirty);
679
680 if (block->flags & REG_FLAG_FLUSH_CHANGE) {
681 ctx->flags |= R600_CONTEXT_PS_PARTIAL_FLUSH;
682 }
683 }
684 }
685
686 /**
687 * If reg needs a reloc, this function will add it to its block's reloc list.
688 * @return true if reg needs a reloc, false otherwise
689 */
690 static bool r600_reg_set_block_reloc(struct r600_pipe_reg *reg)
691 {
692 unsigned reloc_id;
693
694 if (!reg->block->pm4_bo_index[reg->id]) {
695 return false;
696 }
697 /* find relocation */
698 reloc_id = reg->block->pm4_bo_index[reg->id];
699 pipe_resource_reference(
700 (struct pipe_resource**)&reg->block->reloc[reloc_id].bo,
701 &reg->bo->b.b);
702 reg->block->reloc[reloc_id].bo_usage = reg->bo_usage;
703 return true;
704 }
705
706 /**
707 * This function will emit all the registers in state directly to the command
708 * stream allowing you to bypass the r600_context dirty list.
709 *
710 * This is used for dispatching compute shaders to avoid mixing compute and
711 * 3D states in the context's dirty list.
712 *
713 * @param pkt_flags Should be either 0 or RADEON_CP_PACKET3_COMPUTE_MODE. This
714 * value will be passed on to r600_context_block_emit_dirty an or'd against
715 * the PKT3 headers.
716 */
717 void r600_context_pipe_state_emit(struct r600_context *ctx,
718 struct r600_pipe_state *state,
719 unsigned pkt_flags)
720 {
721 unsigned i;
722
723 /* Mark all blocks as dirty:
724 * Since two registers can be in the same block, we need to make sure
725 * we mark all the blocks dirty before we emit any of them. If we were
726 * to mark blocks dirty and emit them in the same loop, like this:
727 *
728 * foreach (reg in state->regs) {
729 * mark_dirty(reg->block)
730 * emit_block(reg->block)
731 * }
732 *
733 * Then if we have two registers in this state that are in the same
734 * block, we would end up emitting that block twice.
735 */
736 for (i = 0; i < state->nregs; i++) {
737 struct r600_pipe_reg *reg = &state->regs[i];
738 /* Mark all the registers in the block as dirty */
739 reg->block->nreg_dirty = reg->block->nreg;
740 reg->block->status |= R600_BLOCK_STATUS_DIRTY;
741 /* Update the reloc for this register if necessary. */
742 r600_reg_set_block_reloc(reg);
743 }
744
745 /* Emit the registers writes */
746 for (i = 0; i < state->nregs; i++) {
747 struct r600_pipe_reg *reg = &state->regs[i];
748 if (reg->block->status & R600_BLOCK_STATUS_DIRTY) {
749 r600_context_block_emit_dirty(ctx, reg->block, pkt_flags);
750 }
751 }
752 }
753
754 void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state)
755 {
756 struct r600_block *block;
757 int dirty;
758 for (int i = 0; i < state->nregs; i++) {
759 unsigned id;
760 struct r600_pipe_reg *reg = &state->regs[i];
761
762 block = reg->block;
763 id = reg->id;
764
765 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
766
767 if (reg->value != block->reg[id]) {
768 block->reg[id] = reg->value;
769 dirty |= R600_BLOCK_STATUS_DIRTY;
770 }
771 if (block->flags & REG_FLAG_DIRTY_ALWAYS)
772 dirty |= R600_BLOCK_STATUS_DIRTY;
773 if (r600_reg_set_block_reloc(reg)) {
774 /* always force dirty for relocs for now */
775 dirty |= R600_BLOCK_STATUS_DIRTY;
776 }
777
778 if (dirty)
779 r600_context_dirty_block(ctx, block, dirty, id);
780 }
781 }
782
783 /**
784 * @param pkt_flags should be set to RADEON_CP_PACKET3_COMPUTE_MODE if this
785 * block will be used for compute shaders.
786 */
787 void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block,
788 unsigned pkt_flags)
789 {
790 struct radeon_winsys_cs *cs = ctx->cs;
791 int optional = block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS);
792 int cp_dwords = block->pm4_ndwords, start_dword = 0;
793 int new_dwords = 0;
794 int nbo = block->nbo;
795
796 if (block->nreg_dirty == 0 && optional) {
797 goto out;
798 }
799
800 if (nbo) {
801 for (int j = 0; j < block->nreg; j++) {
802 if (block->pm4_bo_index[j]) {
803 /* find relocation */
804 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]];
805 if (reloc->bo) {
806 block->pm4[reloc->bo_pm4_index] =
807 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage);
808 } else {
809 block->pm4[reloc->bo_pm4_index] = 0;
810 }
811 nbo--;
812 if (nbo == 0)
813 break;
814
815 }
816 }
817 }
818
819 optional &= (block->nreg_dirty != block->nreg);
820 if (optional) {
821 new_dwords = block->nreg_dirty;
822 start_dword = cs->cdw;
823 cp_dwords = new_dwords + 2;
824 }
825 memcpy(&cs->buf[cs->cdw], block->pm4, cp_dwords * 4);
826
827 /* We are applying the pkt_flags after copying the register block to
828 * the the command stream, because it is possible this block will be
829 * emitted with a different pkt_flags, and we don't want to store the
830 * pkt_flags in the block.
831 */
832 cs->buf[cs->cdw] |= pkt_flags;
833 cs->cdw += cp_dwords;
834
835 if (optional) {
836 uint32_t newword;
837
838 newword = cs->buf[start_dword];
839 newword &= PKT_COUNT_C;
840 newword |= PKT_COUNT_S(new_dwords);
841 cs->buf[start_dword] = newword;
842 }
843 out:
844 block->status ^= R600_BLOCK_STATUS_DIRTY;
845 block->nreg_dirty = 0;
846 LIST_DELINIT(&block->list);
847 }
848
849 void r600_flush_emit(struct r600_context *rctx)
850 {
851 struct radeon_winsys_cs *cs = rctx->cs;
852
853 if (!rctx->flags) {
854 return;
855 }
856
857 if (rctx->flags & R600_CONTEXT_PS_PARTIAL_FLUSH) {
858 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
859 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
860 }
861
862 if (rctx->flags & R600_CONTEXT_FLUSH_AND_INV) {
863 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
864 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
865
866 /* DB flushes are special due to errata with hyperz, we need to
867 * insert a no-op, so that the cache has time to really flush.
868 */
869 if (rctx->chip_class <= R700 &&
870 rctx->flags & R600_CONTEXT_HTILE_ERRATA) {
871 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 31, 0);
872 cs->buf[cs->cdw++] = 0xdeadcafe;
873 cs->buf[cs->cdw++] = 0xdeadcafe;
874 cs->buf[cs->cdw++] = 0xdeadcafe;
875 cs->buf[cs->cdw++] = 0xdeadcafe;
876 cs->buf[cs->cdw++] = 0xdeadcafe;
877 cs->buf[cs->cdw++] = 0xdeadcafe;
878 cs->buf[cs->cdw++] = 0xdeadcafe;
879 cs->buf[cs->cdw++] = 0xdeadcafe;
880 cs->buf[cs->cdw++] = 0xdeadcafe;
881 cs->buf[cs->cdw++] = 0xdeadcafe;
882 cs->buf[cs->cdw++] = 0xdeadcafe;
883 cs->buf[cs->cdw++] = 0xdeadcafe;
884 cs->buf[cs->cdw++] = 0xdeadcafe;
885 cs->buf[cs->cdw++] = 0xdeadcafe;
886 cs->buf[cs->cdw++] = 0xdeadcafe;
887 cs->buf[cs->cdw++] = 0xdeadcafe;
888 cs->buf[cs->cdw++] = 0xdeadcafe;
889 cs->buf[cs->cdw++] = 0xdeadcafe;
890 cs->buf[cs->cdw++] = 0xdeadcafe;
891 cs->buf[cs->cdw++] = 0xdeadcafe;
892 cs->buf[cs->cdw++] = 0xdeadcafe;
893 cs->buf[cs->cdw++] = 0xdeadcafe;
894 cs->buf[cs->cdw++] = 0xdeadcafe;
895 cs->buf[cs->cdw++] = 0xdeadcafe;
896 cs->buf[cs->cdw++] = 0xdeadcafe;
897 cs->buf[cs->cdw++] = 0xdeadcafe;
898 cs->buf[cs->cdw++] = 0xdeadcafe;
899 cs->buf[cs->cdw++] = 0xdeadcafe;
900 cs->buf[cs->cdw++] = 0xdeadcafe;
901 cs->buf[cs->cdw++] = 0xdeadcafe;
902 cs->buf[cs->cdw++] = 0xdeadcafe;
903 cs->buf[cs->cdw++] = 0xdeadcafe;
904 }
905 }
906
907 if (rctx->flags & (R600_CONTEXT_CB_FLUSH |
908 R600_CONTEXT_DB_FLUSH |
909 R600_CONTEXT_SHADERCONST_FLUSH |
910 R600_CONTEXT_TEX_FLUSH |
911 R600_CONTEXT_VTX_FLUSH |
912 R600_CONTEXT_STREAMOUT_FLUSH)) {
913 /* anything left (cb, vtx, shader, streamout) can be flushed
914 * using the surface sync packet
915 */
916 unsigned flags = 0;
917
918 if (rctx->flags & R600_CONTEXT_CB_FLUSH) {
919 flags |= S_0085F0_CB_ACTION_ENA(1) |
920 S_0085F0_CB0_DEST_BASE_ENA(1) |
921 S_0085F0_CB1_DEST_BASE_ENA(1) |
922 S_0085F0_CB2_DEST_BASE_ENA(1) |
923 S_0085F0_CB3_DEST_BASE_ENA(1) |
924 S_0085F0_CB4_DEST_BASE_ENA(1) |
925 S_0085F0_CB5_DEST_BASE_ENA(1) |
926 S_0085F0_CB6_DEST_BASE_ENA(1) |
927 S_0085F0_CB7_DEST_BASE_ENA(1);
928
929 if (rctx->chip_class >= EVERGREEN) {
930 flags |= S_0085F0_CB8_DEST_BASE_ENA(1) |
931 S_0085F0_CB9_DEST_BASE_ENA(1) |
932 S_0085F0_CB10_DEST_BASE_ENA(1) |
933 S_0085F0_CB11_DEST_BASE_ENA(1);
934 }
935
936 /* RV670 errata
937 * (CB1_DEST_BASE_ENA is also required, which is
938 * included unconditionally above). */
939 if (rctx->family == CHIP_RV670 ||
940 rctx->family == CHIP_RS780 ||
941 rctx->family == CHIP_RS880) {
942 flags |= S_0085F0_DEST_BASE_0_ENA(1);
943 }
944 }
945
946 if (rctx->flags & R600_CONTEXT_STREAMOUT_FLUSH) {
947 flags |= S_0085F0_SO0_DEST_BASE_ENA(1) |
948 S_0085F0_SO1_DEST_BASE_ENA(1) |
949 S_0085F0_SO2_DEST_BASE_ENA(1) |
950 S_0085F0_SO3_DEST_BASE_ENA(1) |
951 S_0085F0_SMX_ACTION_ENA(1);
952
953 /* RV670 errata */
954 if (rctx->family == CHIP_RV670 ||
955 rctx->family == CHIP_RS780 ||
956 rctx->family == CHIP_RS880) {
957 flags |= S_0085F0_DEST_BASE_0_ENA(1);
958 }
959 }
960
961 flags |= (rctx->flags & R600_CONTEXT_DB_FLUSH) ? S_0085F0_DB_ACTION_ENA(1) |
962 S_0085F0_DB_DEST_BASE_ENA(1): 0;
963 flags |= (rctx->flags & R600_CONTEXT_SHADERCONST_FLUSH) ? S_0085F0_SH_ACTION_ENA(1) : 0;
964 flags |= (rctx->flags & R600_CONTEXT_TEX_FLUSH) ? S_0085F0_TC_ACTION_ENA(1) : 0;
965 flags |= (rctx->flags & R600_CONTEXT_VTX_FLUSH) ? S_0085F0_VC_ACTION_ENA(1) : 0;
966
967 cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_SYNC, 3, 0);
968 cs->buf[cs->cdw++] = flags; /* CP_COHER_CNTL */
969 cs->buf[cs->cdw++] = 0xffffffff; /* CP_COHER_SIZE */
970 cs->buf[cs->cdw++] = 0; /* CP_COHER_BASE */
971 cs->buf[cs->cdw++] = 0x0000000A; /* POLL_INTERVAL */
972 }
973
974 if (rctx->flags & R600_CONTEXT_WAIT_IDLE) {
975 /* wait for things to settle */
976 r600_write_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
977 }
978
979 /* everything is properly flushed */
980 rctx->flags = 0;
981 }
982
983 void r600_context_flush(struct r600_context *ctx, unsigned flags)
984 {
985 struct radeon_winsys_cs *cs = ctx->cs;
986
987 if (cs->cdw == ctx->start_cs_cmd.atom.num_dw)
988 return;
989
990 ctx->timer_queries_suspended = false;
991 ctx->nontimer_queries_suspended = false;
992 ctx->streamout_suspended = false;
993
994 /* suspend queries */
995 if (ctx->num_cs_dw_timer_queries_suspend) {
996 r600_suspend_timer_queries(ctx);
997 ctx->timer_queries_suspended = true;
998 }
999 if (ctx->num_cs_dw_nontimer_queries_suspend) {
1000 r600_suspend_nontimer_queries(ctx);
1001 ctx->nontimer_queries_suspended = true;
1002 }
1003
1004 if (ctx->num_cs_dw_streamout_end) {
1005 r600_context_streamout_end(ctx);
1006 ctx->streamout_suspended = true;
1007 }
1008
1009 /* partial flush is needed to avoid lockups on some chips with user fences */
1010 ctx->flags |= R600_CONTEXT_PS_PARTIAL_FLUSH;
1011
1012 /* flush the framebuffer */
1013 ctx->flags |= R600_CONTEXT_CB_FLUSH | R600_CONTEXT_DB_FLUSH;
1014
1015 /* R6xx errata */
1016 if (ctx->chip_class == R600) {
1017 ctx->flags |= R600_CONTEXT_FLUSH_AND_INV;
1018 }
1019
1020 r600_flush_emit(ctx);
1021
1022 /* old kernels and userspace don't set SX_MISC, so we must reset it to 0 here */
1023 if (ctx->chip_class <= R700) {
1024 r600_write_context_reg(cs, R_028350_SX_MISC, 0);
1025 }
1026
1027 /* force to keep tiling flags */
1028 flags |= RADEON_FLUSH_KEEP_TILING_FLAGS;
1029
1030 /* Flush the CS. */
1031 ctx->ws->cs_flush(ctx->cs, flags);
1032
1033 r600_begin_new_cs(ctx);
1034 }
1035
1036 void r600_begin_new_cs(struct r600_context *ctx)
1037 {
1038 struct r600_block *enable_block = NULL;
1039 unsigned shader;
1040
1041 ctx->pm4_dirty_cdwords = 0;
1042 ctx->flags = 0;
1043
1044 /* Begin a new CS. */
1045 r600_emit_atom(ctx, &ctx->start_cs_cmd.atom);
1046
1047 /* Re-emit states. */
1048 r600_atom_dirty(ctx, &ctx->alphatest_state.atom);
1049 r600_atom_dirty(ctx, &ctx->cb_misc_state.atom);
1050 r600_atom_dirty(ctx, &ctx->db_misc_state.atom);
1051 r600_atom_dirty(ctx, &ctx->sample_mask.atom);
1052 r600_atom_dirty(ctx, &ctx->stencil_ref.atom);
1053
1054 if (ctx->chip_class <= R700) {
1055 r600_atom_dirty(ctx, &ctx->seamless_cube_map.atom);
1056 }
1057
1058 ctx->vertex_buffer_state.dirty_mask = ctx->vertex_buffer_state.enabled_mask;
1059 r600_vertex_buffers_dirty(ctx);
1060
1061 /* Re-emit shader resources. */
1062 for (shader = 0; shader < PIPE_SHADER_TYPES; shader++) {
1063 struct r600_constbuf_state *constbuf = &ctx->constbuf_state[shader];
1064 struct r600_textures_info *samplers = &ctx->samplers[shader];
1065
1066 constbuf->dirty_mask = constbuf->enabled_mask;
1067 samplers->views.dirty_mask = samplers->views.enabled_mask;
1068 samplers->states.dirty_mask = samplers->states.enabled_mask;
1069
1070 r600_constant_buffers_dirty(ctx, constbuf);
1071 r600_sampler_views_dirty(ctx, &samplers->views);
1072 r600_sampler_states_dirty(ctx, &samplers->states);
1073 }
1074
1075 if (ctx->streamout_suspended) {
1076 ctx->streamout_start = TRUE;
1077 ctx->streamout_append_bitmask = ~0;
1078 }
1079
1080 /* resume queries */
1081 if (ctx->timer_queries_suspended) {
1082 r600_resume_timer_queries(ctx);
1083 }
1084 if (ctx->nontimer_queries_suspended) {
1085 r600_resume_nontimer_queries(ctx);
1086 }
1087
1088 /* set all valid group as dirty so they get reemited on
1089 * next draw command
1090 */
1091 LIST_FOR_EACH_ENTRY(enable_block, &ctx->enable_list, enable_list) {
1092 if(!(enable_block->status & R600_BLOCK_STATUS_DIRTY)) {
1093 LIST_ADDTAIL(&enable_block->list,&ctx->dirty);
1094 enable_block->status |= R600_BLOCK_STATUS_DIRTY;
1095 }
1096 ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords;
1097 enable_block->nreg_dirty = enable_block->nreg;
1098 }
1099 }
1100
1101 void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value)
1102 {
1103 struct radeon_winsys_cs *cs = ctx->cs;
1104 uint64_t va;
1105
1106 r600_need_cs_space(ctx, 10, FALSE);
1107
1108 va = r600_resource_va(&ctx->screen->screen, (void*)fence_bo);
1109 va = va + (offset << 2);
1110
1111 ctx->flags &= ~R600_CONTEXT_PS_PARTIAL_FLUSH;
1112 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
1113 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
1114
1115 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1116 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1117 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* ADDRESS_LO */
1118 /* DATA_SEL | INT_EN | ADDRESS_HI */
1119 cs->buf[cs->cdw++] = (1 << 29) | (0 << 24) | ((va >> 32UL) & 0xFF);
1120 cs->buf[cs->cdw++] = value; /* DATA_LO */
1121 cs->buf[cs->cdw++] = 0; /* DATA_HI */
1122 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1123 cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, fence_bo, RADEON_USAGE_WRITE);
1124 }
1125
1126 static void r600_flush_vgt_streamout(struct r600_context *ctx)
1127 {
1128 struct radeon_winsys_cs *cs = ctx->cs;
1129
1130 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0);
1131 cs->buf[cs->cdw++] = (R_008490_CP_STRMOUT_CNTL - R600_CONFIG_REG_OFFSET) >> 2;
1132 cs->buf[cs->cdw++] = 0;
1133
1134 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
1135 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0);
1136
1137 cs->buf[cs->cdw++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0);
1138 cs->buf[cs->cdw++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */
1139 cs->buf[cs->cdw++] = R_008490_CP_STRMOUT_CNTL >> 2; /* register */
1140 cs->buf[cs->cdw++] = 0;
1141 cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* reference value */
1142 cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* mask */
1143 cs->buf[cs->cdw++] = 4; /* poll interval */
1144 }
1145
1146 static void r600_set_streamout_enable(struct r600_context *ctx, unsigned buffer_enable_bit)
1147 {
1148 struct radeon_winsys_cs *cs = ctx->cs;
1149
1150 if (buffer_enable_bit) {
1151 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1152 cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1153 cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(1);
1154
1155 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1156 cs->buf[cs->cdw++] = (R_028B20_VGT_STRMOUT_BUFFER_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1157 cs->buf[cs->cdw++] = buffer_enable_bit;
1158 } else {
1159 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1160 cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1161 cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(0);
1162 }
1163 }
1164
1165 void r600_context_streamout_begin(struct r600_context *ctx)
1166 {
1167 struct radeon_winsys_cs *cs = ctx->cs;
1168 struct r600_so_target **t = ctx->so_targets;
1169 unsigned *stride_in_dw = ctx->vs_shader->so.stride;
1170 unsigned buffer_en, i, update_flags = 0;
1171 uint64_t va;
1172
1173 buffer_en = (ctx->num_so_targets >= 1 && t[0] ? 1 : 0) |
1174 (ctx->num_so_targets >= 2 && t[1] ? 2 : 0) |
1175 (ctx->num_so_targets >= 3 && t[2] ? 4 : 0) |
1176 (ctx->num_so_targets >= 4 && t[3] ? 8 : 0);
1177
1178 ctx->num_cs_dw_streamout_end =
1179 12 + /* flush_vgt_streamout */
1180 util_bitcount(buffer_en) * 8 + /* STRMOUT_BUFFER_UPDATE */
1181 3 /* set_streamout_enable(0) */;
1182
1183 r600_need_cs_space(ctx,
1184 12 + /* flush_vgt_streamout */
1185 6 + /* set_streamout_enable */
1186 util_bitcount(buffer_en) * 7 + /* SET_CONTEXT_REG */
1187 (ctx->chip_class == R700 ? util_bitcount(buffer_en) * 5 : 0) + /* STRMOUT_BASE_UPDATE */
1188 util_bitcount(buffer_en & ctx->streamout_append_bitmask) * 8 + /* STRMOUT_BUFFER_UPDATE */
1189 util_bitcount(buffer_en & ~ctx->streamout_append_bitmask) * 6 + /* STRMOUT_BUFFER_UPDATE */
1190 (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770 ? 2 : 0) + /* SURFACE_BASE_UPDATE */
1191 ctx->num_cs_dw_streamout_end, TRUE);
1192
1193 if (ctx->chip_class >= EVERGREEN) {
1194 evergreen_flush_vgt_streamout(ctx);
1195 evergreen_set_streamout_enable(ctx, buffer_en);
1196 } else {
1197 r600_flush_vgt_streamout(ctx);
1198 r600_set_streamout_enable(ctx, buffer_en);
1199 }
1200
1201 for (i = 0; i < ctx->num_so_targets; i++) {
1202 if (t[i]) {
1203 t[i]->stride_in_dw = stride_in_dw[i];
1204 t[i]->so_index = i;
1205 va = r600_resource_va(&ctx->screen->screen,
1206 (void*)t[i]->b.buffer);
1207
1208 update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i);
1209
1210 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 3, 0);
1211 cs->buf[cs->cdw++] = (R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 +
1212 16*i - R600_CONTEXT_REG_OFFSET) >> 2;
1213 cs->buf[cs->cdw++] = (t[i]->b.buffer_offset +
1214 t[i]->b.buffer_size) >> 2; /* BUFFER_SIZE (in DW) */
1215 cs->buf[cs->cdw++] = stride_in_dw[i]; /* VTX_STRIDE (in DW) */
1216 cs->buf[cs->cdw++] = va >> 8; /* BUFFER_BASE */
1217
1218 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1219 cs->buf[cs->cdw++] =
1220 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1221 RADEON_USAGE_WRITE);
1222
1223 /* R7xx requires this packet after updating BUFFER_BASE.
1224 * Without this, R7xx locks up. */
1225 if (ctx->chip_class == R700) {
1226 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BASE_UPDATE, 1, 0);
1227 cs->buf[cs->cdw++] = i;
1228 cs->buf[cs->cdw++] = va >> 8;
1229
1230 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1231 cs->buf[cs->cdw++] =
1232 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1233 RADEON_USAGE_WRITE);
1234 }
1235
1236 if (ctx->streamout_append_bitmask & (1 << i)) {
1237 va = r600_resource_va(&ctx->screen->screen,
1238 (void*)t[i]->filled_size);
1239 /* Append. */
1240 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1241 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1242 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_MEM); /* control */
1243 cs->buf[cs->cdw++] = 0; /* unused */
1244 cs->buf[cs->cdw++] = 0; /* unused */
1245 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* src address lo */
1246 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* src address hi */
1247
1248 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1249 cs->buf[cs->cdw++] =
1250 r600_context_bo_reloc(ctx, t[i]->filled_size,
1251 RADEON_USAGE_READ);
1252 } else {
1253 /* Start from the beginning. */
1254 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1255 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1256 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_PACKET); /* control */
1257 cs->buf[cs->cdw++] = 0; /* unused */
1258 cs->buf[cs->cdw++] = 0; /* unused */
1259 cs->buf[cs->cdw++] = t[i]->b.buffer_offset >> 2; /* buffer offset in DW */
1260 cs->buf[cs->cdw++] = 0; /* unused */
1261 }
1262 }
1263 }
1264
1265 if (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770) {
1266 cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
1267 cs->buf[cs->cdw++] = update_flags;
1268 }
1269 }
1270
1271 void r600_context_streamout_end(struct r600_context *ctx)
1272 {
1273 struct radeon_winsys_cs *cs = ctx->cs;
1274 struct r600_so_target **t = ctx->so_targets;
1275 unsigned i;
1276 uint64_t va;
1277
1278 if (ctx->chip_class >= EVERGREEN) {
1279 evergreen_flush_vgt_streamout(ctx);
1280 } else {
1281 r600_flush_vgt_streamout(ctx);
1282 }
1283
1284 for (i = 0; i < ctx->num_so_targets; i++) {
1285 if (t[i]) {
1286 va = r600_resource_va(&ctx->screen->screen,
1287 (void*)t[i]->filled_size);
1288 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1289 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1290 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_NONE) |
1291 STRMOUT_STORE_BUFFER_FILLED_SIZE; /* control */
1292 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* dst address lo */
1293 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* dst address hi */
1294 cs->buf[cs->cdw++] = 0; /* unused */
1295 cs->buf[cs->cdw++] = 0; /* unused */
1296
1297 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1298 cs->buf[cs->cdw++] =
1299 r600_context_bo_reloc(ctx, t[i]->filled_size,
1300 RADEON_USAGE_WRITE);
1301
1302 }
1303 }
1304
1305 if (ctx->chip_class >= EVERGREEN) {
1306 evergreen_set_streamout_enable(ctx, 0);
1307 } else {
1308 r600_set_streamout_enable(ctx, 0);
1309 }
1310 ctx->flags |= R600_CONTEXT_STREAMOUT_FLUSH;
1311
1312 /* R6xx errata */
1313 if (ctx->chip_class == R600) {
1314 ctx->flags |= R600_CONTEXT_FLUSH_AND_INV;
1315 }
1316 ctx->num_cs_dw_streamout_end = 0;
1317 }