projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
r600g: remove debug code for streamout
[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 block->flags = 0;
140 block->status |= R600_BLOCK_STATUS_DIRTY; /* dirty all blocks at start */
141 block->start_offset = reg[i].offset;
142 block->pm4[block->pm4_ndwords++] = PKT3(opcode, n, 0);
143 block->pm4[block->pm4_ndwords++] = (block->start_offset - offset_base) >> 2;
144 block->reg = &block->pm4[block->pm4_ndwords];
145 block->pm4_ndwords += n;
146 block->nreg = n;
147 block->nreg_dirty = n;
148 LIST_INITHEAD(&block->list);
149 LIST_INITHEAD(&block->enable_list);
150
151 for (j = 0; j < n; j++) {
152 if (reg[i+j].flags & REG_FLAG_DIRTY_ALWAYS) {
153 block->flags |= REG_FLAG_DIRTY_ALWAYS;
154 }
155 if (reg[i+j].flags & REG_FLAG_ENABLE_ALWAYS) {
156 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
157 block->status |= R600_BLOCK_STATUS_ENABLED;
158 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
159 LIST_ADDTAIL(&block->list,&ctx->dirty);
160 }
161 }
162 if (reg[i+j].flags & REG_FLAG_FLUSH_CHANGE) {
163 block->flags |= REG_FLAG_FLUSH_CHANGE;
164 }
165
166 if (reg[i+j].flags & REG_FLAG_NEED_BO) {
167 block->nbo++;
168 assert(block->nbo < R600_BLOCK_MAX_BO);
169 block->pm4_bo_index[j] = block->nbo;
170 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0, 0);
171 block->pm4[block->pm4_ndwords++] = 0x00000000;
172 block->reloc[block->nbo].bo_pm4_index = block->pm4_ndwords - 1;
173 }
174 if ((ctx->family > CHIP_R600) &&
175 (ctx->family < CHIP_RV770) && reg[i+j].flags & REG_FLAG_RV6XX_SBU) {
176 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
177 block->pm4[block->pm4_ndwords++] = reg[i+j].sbu_flags;
178 }
179 }
180 /* check that we stay in limit */
181 assert(block->pm4_ndwords < R600_BLOCK_MAX_REG);
182 }
183
184 int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg,
185 unsigned opcode, unsigned offset_base)
186 {
187 struct r600_block *block;
188 struct r600_range *range;
189 int offset;
190
191 for (unsigned i = 0, n = 0; i < nreg; i += n) {
192 /* ignore new block balise */
193 if (reg[i].offset == GROUP_FORCE_NEW_BLOCK) {
194 n = 1;
195 continue;
196 }
197
198 /* ignore regs not on R600 on R600 */
199 if ((reg[i].flags & REG_FLAG_NOT_R600) && ctx->family == CHIP_R600) {
200 n = 1;
201 continue;
202 }
203
204 /* register that need relocation are in their own group */
205 /* find number of consecutive registers */
206 n = 0;
207 offset = reg[i].offset;
208 while (reg[i + n].offset == offset) {
209 n++;
210 offset += 4;
211 if ((n + i) >= nreg)
212 break;
213 if (n >= (R600_BLOCK_MAX_REG - 2))
214 break;
215 }
216
217 /* allocate new block */
218 block = calloc(1, sizeof(struct r600_block));
219 if (block == NULL) {
220 return -ENOMEM;
221 }
222 ctx->nblocks++;
223 for (int j = 0; j < n; j++) {
224 range = &ctx->range[CTX_RANGE_ID(reg[i + j].offset)];
225 /* create block table if it doesn't exist */
226 if (!range->blocks)
227 range->blocks = calloc(1 << HASH_SHIFT, sizeof(void *));
228 if (!range->blocks)
229 return -1;
230
231 range->blocks[CTX_BLOCK_ID(reg[i + j].offset)] = block;
232 }
233
234 r600_init_block(ctx, block, reg, i, n, opcode, offset_base);
235
236 }
237 return 0;
238 }
239
240 /* R600/R700 configuration */
241 static const struct r600_reg r600_config_reg_list[] = {
242 {R_008958_VGT_PRIMITIVE_TYPE, 0, 0},
243 {R_008C04_SQ_GPR_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0},
244 {R_009508_TA_CNTL_AUX, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0},
245 };
246
247 static const struct r600_reg r600_ctl_const_list[] = {
248 {R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0},
249 };
250
251 static const struct r600_reg r600_context_reg_list[] = {
252 {R_028A4C_PA_SC_MODE_CNTL, 0, 0},
253 {GROUP_FORCE_NEW_BLOCK, 0, 0},
254 {R_028040_CB_COLOR0_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(0)},
255 {GROUP_FORCE_NEW_BLOCK, 0, 0},
256 {R_0280A0_CB_COLOR0_INFO, REG_FLAG_NEED_BO, 0},
257 {R_028060_CB_COLOR0_SIZE, 0, 0},
258 {R_028080_CB_COLOR0_VIEW, 0, 0},
259 {GROUP_FORCE_NEW_BLOCK, 0, 0},
260 {R_0280E0_CB_COLOR0_FRAG, REG_FLAG_NEED_BO, 0},
261 {GROUP_FORCE_NEW_BLOCK, 0, 0},
262 {R_0280C0_CB_COLOR0_TILE, REG_FLAG_NEED_BO, 0},
263 {GROUP_FORCE_NEW_BLOCK, 0, 0},
264 {R_028044_CB_COLOR1_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(1)},
265 {GROUP_FORCE_NEW_BLOCK, 0, 0},
266 {R_0280A4_CB_COLOR1_INFO, REG_FLAG_NEED_BO, 0},
267 {R_028064_CB_COLOR1_SIZE, 0, 0},
268 {R_028084_CB_COLOR1_VIEW, 0, 0},
269 {GROUP_FORCE_NEW_BLOCK, 0, 0},
270 {R_0280E4_CB_COLOR1_FRAG, REG_FLAG_NEED_BO, 0},
271 {GROUP_FORCE_NEW_BLOCK, 0, 0},
272 {R_0280C4_CB_COLOR1_TILE, REG_FLAG_NEED_BO, 0},
273 {GROUP_FORCE_NEW_BLOCK, 0, 0},
274 {R_028048_CB_COLOR2_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(2)},
275 {GROUP_FORCE_NEW_BLOCK, 0, 0},
276 {R_0280A8_CB_COLOR2_INFO, REG_FLAG_NEED_BO, 0},
277 {R_028068_CB_COLOR2_SIZE, 0, 0},
278 {R_028088_CB_COLOR2_VIEW, 0, 0},
279 {GROUP_FORCE_NEW_BLOCK, 0, 0},
280 {R_0280E8_CB_COLOR2_FRAG, REG_FLAG_NEED_BO, 0},
281 {GROUP_FORCE_NEW_BLOCK, 0, 0},
282 {R_0280C8_CB_COLOR2_TILE, REG_FLAG_NEED_BO, 0},
283 {GROUP_FORCE_NEW_BLOCK, 0, 0},
284 {R_02804C_CB_COLOR3_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(3)},
285 {GROUP_FORCE_NEW_BLOCK, 0, 0},
286 {R_0280AC_CB_COLOR3_INFO, REG_FLAG_NEED_BO, 0},
287 {R_02806C_CB_COLOR3_SIZE, 0, 0},
288 {R_02808C_CB_COLOR3_VIEW, 0, 0},
289 {GROUP_FORCE_NEW_BLOCK, 0, 0},
290 {R_0280EC_CB_COLOR3_FRAG, REG_FLAG_NEED_BO, 0},
291 {GROUP_FORCE_NEW_BLOCK, 0, 0},
292 {R_0280CC_CB_COLOR3_TILE, REG_FLAG_NEED_BO, 0},
293 {GROUP_FORCE_NEW_BLOCK, 0, 0},
294 {R_028050_CB_COLOR4_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(4)},
295 {GROUP_FORCE_NEW_BLOCK, 0, 0},
296 {R_0280B0_CB_COLOR4_INFO, REG_FLAG_NEED_BO, 0},
297 {R_028070_CB_COLOR4_SIZE, 0, 0},
298 {R_028090_CB_COLOR4_VIEW, 0, 0},
299 {GROUP_FORCE_NEW_BLOCK, 0, 0},
300 {R_0280F0_CB_COLOR4_FRAG, REG_FLAG_NEED_BO, 0},
301 {GROUP_FORCE_NEW_BLOCK, 0, 0},
302 {R_0280D0_CB_COLOR4_TILE, REG_FLAG_NEED_BO, 0},
303 {GROUP_FORCE_NEW_BLOCK, 0, 0},
304 {R_028054_CB_COLOR5_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(5)},
305 {GROUP_FORCE_NEW_BLOCK, 0, 0},
306 {R_0280B4_CB_COLOR5_INFO, REG_FLAG_NEED_BO, 0},
307 {R_028074_CB_COLOR5_SIZE, 0, 0},
308 {R_028094_CB_COLOR5_VIEW, 0, 0},
309 {GROUP_FORCE_NEW_BLOCK, 0, 0},
310 {R_0280F4_CB_COLOR5_FRAG, REG_FLAG_NEED_BO, 0},
311 {GROUP_FORCE_NEW_BLOCK, 0, 0},
312 {R_0280D4_CB_COLOR5_TILE, REG_FLAG_NEED_BO, 0},
313 {R_028058_CB_COLOR6_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(6)},
314 {R_0280B8_CB_COLOR6_INFO, REG_FLAG_NEED_BO, 0},
315 {R_028078_CB_COLOR6_SIZE, 0, 0},
316 {R_028098_CB_COLOR6_VIEW, 0, 0},
317 {GROUP_FORCE_NEW_BLOCK, 0, 0},
318 {R_0280F8_CB_COLOR6_FRAG, REG_FLAG_NEED_BO, 0},
319 {GROUP_FORCE_NEW_BLOCK, 0, 0},
320 {R_0280D8_CB_COLOR6_TILE, REG_FLAG_NEED_BO, 0},
321 {GROUP_FORCE_NEW_BLOCK, 0, 0},
322 {R_02805C_CB_COLOR7_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(7)},
323 {GROUP_FORCE_NEW_BLOCK, 0, 0},
324 {R_0280BC_CB_COLOR7_INFO, REG_FLAG_NEED_BO, 0},
325 {R_02807C_CB_COLOR7_SIZE, 0, 0},
326 {R_02809C_CB_COLOR7_VIEW, 0, 0},
327 {R_0280FC_CB_COLOR7_FRAG, REG_FLAG_NEED_BO, 0},
328 {R_0280DC_CB_COLOR7_TILE, REG_FLAG_NEED_BO, 0},
329 {R_028120_CB_CLEAR_RED, 0, 0},
330 {R_028124_CB_CLEAR_GREEN, 0, 0},
331 {R_028128_CB_CLEAR_BLUE, 0, 0},
332 {R_02812C_CB_CLEAR_ALPHA, 0, 0},
333 {R_028414_CB_BLEND_RED, 0, 0},
334 {R_028418_CB_BLEND_GREEN, 0, 0},
335 {R_02841C_CB_BLEND_BLUE, 0, 0},
336 {R_028420_CB_BLEND_ALPHA, 0, 0},
337 {R_028424_CB_FOG_RED, 0, 0},
338 {R_028428_CB_FOG_GREEN, 0, 0},
339 {R_02842C_CB_FOG_BLUE, 0, 0},
340 {R_028430_DB_STENCILREFMASK, 0, 0},
341 {R_028434_DB_STENCILREFMASK_BF, 0, 0},
342 {R_028780_CB_BLEND0_CONTROL, REG_FLAG_NOT_R600, 0},
343 {R_028784_CB_BLEND1_CONTROL, REG_FLAG_NOT_R600, 0},
344 {R_028788_CB_BLEND2_CONTROL, REG_FLAG_NOT_R600, 0},
345 {R_02878C_CB_BLEND3_CONTROL, REG_FLAG_NOT_R600, 0},
346 {R_028790_CB_BLEND4_CONTROL, REG_FLAG_NOT_R600, 0},
347 {R_028794_CB_BLEND5_CONTROL, REG_FLAG_NOT_R600, 0},
348 {R_028798_CB_BLEND6_CONTROL, REG_FLAG_NOT_R600, 0},
349 {R_02879C_CB_BLEND7_CONTROL, REG_FLAG_NOT_R600, 0},
350 {R_0287A0_CB_SHADER_CONTROL, 0, 0},
351 {R_028800_DB_DEPTH_CONTROL, 0, 0},
352 {R_028804_CB_BLEND_CONTROL, 0, 0},
353 {R_02880C_DB_SHADER_CONTROL, 0, 0},
354 {R_02800C_DB_DEPTH_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_DEPTH},
355 {R_028000_DB_DEPTH_SIZE, 0, 0},
356 {R_028004_DB_DEPTH_VIEW, 0, 0},
357 {GROUP_FORCE_NEW_BLOCK, 0, 0},
358 {R_028010_DB_DEPTH_INFO, REG_FLAG_NEED_BO, 0},
359 {R_028A6C_VGT_GS_OUT_PRIM_TYPE, 0, 0},
360 {R_028D24_DB_HTILE_SURFACE, 0, 0},
361 {R_028D34_DB_PREFETCH_LIMIT, 0, 0},
362 {R_028204_PA_SC_WINDOW_SCISSOR_TL, 0, 0},
363 {R_028208_PA_SC_WINDOW_SCISSOR_BR, 0, 0},
364 {R_028250_PA_SC_VPORT_SCISSOR_0_TL, 0, 0},
365 {R_028254_PA_SC_VPORT_SCISSOR_0_BR, 0, 0},
366 {R_02843C_PA_CL_VPORT_XSCALE_0, 0, 0},
367 {R_028440_PA_CL_VPORT_XOFFSET_0, 0, 0},
368 {R_028444_PA_CL_VPORT_YSCALE_0, 0, 0},
369 {R_028448_PA_CL_VPORT_YOFFSET_0, 0, 0},
370 {R_02844C_PA_CL_VPORT_ZSCALE_0, 0, 0},
371 {R_028450_PA_CL_VPORT_ZOFFSET_0, 0, 0},
372 {R_0286D4_SPI_INTERP_CONTROL_0, 0, 0},
373 {R_028810_PA_CL_CLIP_CNTL, 0, 0},
374 {R_028814_PA_SU_SC_MODE_CNTL, 0, 0},
375 {R_02881C_PA_CL_VS_OUT_CNTL, 0, 0},
376 {R_028A00_PA_SU_POINT_SIZE, 0, 0},
377 {R_028A04_PA_SU_POINT_MINMAX, 0, 0},
378 {R_028A08_PA_SU_LINE_CNTL, 0, 0},
379 {R_028A0C_PA_SC_LINE_STIPPLE, 0, 0},
380 {R_028C08_PA_SU_VTX_CNTL, 0, 0},
381 {R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL, 0, 0},
382 {R_028DFC_PA_SU_POLY_OFFSET_CLAMP, 0, 0},
383 {R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 0, 0},
384 {R_028E04_PA_SU_POLY_OFFSET_FRONT_OFFSET, 0, 0},
385 {R_028E08_PA_SU_POLY_OFFSET_BACK_SCALE, 0, 0},
386 {R_028E0C_PA_SU_POLY_OFFSET_BACK_OFFSET, 0, 0},
387 {R_028E20_PA_CL_UCP0_X, 0, 0},
388 {R_028E24_PA_CL_UCP0_Y, 0, 0},
389 {R_028E28_PA_CL_UCP0_Z, 0, 0},
390 {R_028E2C_PA_CL_UCP0_W, 0, 0},
391 {R_028E30_PA_CL_UCP1_X, 0, 0},
392 {R_028E34_PA_CL_UCP1_Y, 0, 0},
393 {R_028E38_PA_CL_UCP1_Z, 0, 0},
394 {R_028E3C_PA_CL_UCP1_W, 0, 0},
395 {R_028E40_PA_CL_UCP2_X, 0, 0},
396 {R_028E44_PA_CL_UCP2_Y, 0, 0},
397 {R_028E48_PA_CL_UCP2_Z, 0, 0},
398 {R_028E4C_PA_CL_UCP2_W, 0, 0},
399 {R_028E50_PA_CL_UCP3_X, 0, 0},
400 {R_028E54_PA_CL_UCP3_Y, 0, 0},
401 {R_028E58_PA_CL_UCP3_Z, 0, 0},
402 {R_028E5C_PA_CL_UCP3_W, 0, 0},
403 {R_028E60_PA_CL_UCP4_X, 0, 0},
404 {R_028E64_PA_CL_UCP4_Y, 0, 0},
405 {R_028E68_PA_CL_UCP4_Z, 0, 0},
406 {R_028E6C_PA_CL_UCP4_W, 0, 0},
407 {R_028E70_PA_CL_UCP5_X, 0, 0},
408 {R_028E74_PA_CL_UCP5_Y, 0, 0},
409 {R_028E78_PA_CL_UCP5_Z, 0, 0},
410 {R_028E7C_PA_CL_UCP5_W, 0, 0},
411 {R_028350_SX_MISC, 0, 0},
412 {R_028380_SQ_VTX_SEMANTIC_0, 0, 0},
413 {R_028384_SQ_VTX_SEMANTIC_1, 0, 0},
414 {R_028388_SQ_VTX_SEMANTIC_2, 0, 0},
415 {R_02838C_SQ_VTX_SEMANTIC_3, 0, 0},
416 {R_028390_SQ_VTX_SEMANTIC_4, 0, 0},
417 {R_028394_SQ_VTX_SEMANTIC_5, 0, 0},
418 {R_028398_SQ_VTX_SEMANTIC_6, 0, 0},
419 {R_02839C_SQ_VTX_SEMANTIC_7, 0, 0},
420 {R_0283A0_SQ_VTX_SEMANTIC_8, 0, 0},
421 {R_0283A4_SQ_VTX_SEMANTIC_9, 0, 0},
422 {R_0283A8_SQ_VTX_SEMANTIC_10, 0, 0},
423 {R_0283AC_SQ_VTX_SEMANTIC_11, 0, 0},
424 {R_0283B0_SQ_VTX_SEMANTIC_12, 0, 0},
425 {R_0283B4_SQ_VTX_SEMANTIC_13, 0, 0},
426 {R_0283B8_SQ_VTX_SEMANTIC_14, 0, 0},
427 {R_0283BC_SQ_VTX_SEMANTIC_15, 0, 0},
428 {R_0283C0_SQ_VTX_SEMANTIC_16, 0, 0},
429 {R_0283C4_SQ_VTX_SEMANTIC_17, 0, 0},
430 {R_0283C8_SQ_VTX_SEMANTIC_18, 0, 0},
431 {R_0283CC_SQ_VTX_SEMANTIC_19, 0, 0},
432 {R_0283D0_SQ_VTX_SEMANTIC_20, 0, 0},
433 {R_0283D4_SQ_VTX_SEMANTIC_21, 0, 0},
434 {R_0283D8_SQ_VTX_SEMANTIC_22, 0, 0},
435 {R_0283DC_SQ_VTX_SEMANTIC_23, 0, 0},
436 {R_0283E0_SQ_VTX_SEMANTIC_24, 0, 0},
437 {R_0283E4_SQ_VTX_SEMANTIC_25, 0, 0},
438 {R_0283E8_SQ_VTX_SEMANTIC_26, 0, 0},
439 {R_0283EC_SQ_VTX_SEMANTIC_27, 0, 0},
440 {R_0283F0_SQ_VTX_SEMANTIC_28, 0, 0},
441 {R_0283F4_SQ_VTX_SEMANTIC_29, 0, 0},
442 {R_0283F8_SQ_VTX_SEMANTIC_30, 0, 0},
443 {R_0283FC_SQ_VTX_SEMANTIC_31, 0, 0},
444 {R_028614_SPI_VS_OUT_ID_0, 0, 0},
445 {R_028618_SPI_VS_OUT_ID_1, 0, 0},
446 {R_02861C_SPI_VS_OUT_ID_2, 0, 0},
447 {R_028620_SPI_VS_OUT_ID_3, 0, 0},
448 {R_028624_SPI_VS_OUT_ID_4, 0, 0},
449 {R_028628_SPI_VS_OUT_ID_5, 0, 0},
450 {R_02862C_SPI_VS_OUT_ID_6, 0, 0},
451 {R_028630_SPI_VS_OUT_ID_7, 0, 0},
452 {R_028634_SPI_VS_OUT_ID_8, 0, 0},
453 {R_028638_SPI_VS_OUT_ID_9, 0, 0},
454 {R_0286C4_SPI_VS_OUT_CONFIG, 0, 0},
455 {GROUP_FORCE_NEW_BLOCK, 0, 0},
456 {R_028858_SQ_PGM_START_VS, REG_FLAG_NEED_BO, 0},
457 {GROUP_FORCE_NEW_BLOCK, 0, 0},
458 {R_028868_SQ_PGM_RESOURCES_VS, 0, 0},
459 {GROUP_FORCE_NEW_BLOCK, 0, 0},
460 {R_028894_SQ_PGM_START_FS, REG_FLAG_NEED_BO, 0},
461 {GROUP_FORCE_NEW_BLOCK, 0, 0},
462 {R_0288A4_SQ_PGM_RESOURCES_FS, 0, 0},
463 {R_0288DC_SQ_PGM_CF_OFFSET_FS, 0, 0},
464 {R_028644_SPI_PS_INPUT_CNTL_0, 0, 0},
465 {R_028648_SPI_PS_INPUT_CNTL_1, 0, 0},
466 {R_02864C_SPI_PS_INPUT_CNTL_2, 0, 0},
467 {R_028650_SPI_PS_INPUT_CNTL_3, 0, 0},
468 {R_028654_SPI_PS_INPUT_CNTL_4, 0, 0},
469 {R_028658_SPI_PS_INPUT_CNTL_5, 0, 0},
470 {R_02865C_SPI_PS_INPUT_CNTL_6, 0, 0},
471 {R_028660_SPI_PS_INPUT_CNTL_7, 0, 0},
472 {R_028664_SPI_PS_INPUT_CNTL_8, 0, 0},
473 {R_028668_SPI_PS_INPUT_CNTL_9, 0, 0},
474 {R_02866C_SPI_PS_INPUT_CNTL_10, 0, 0},
475 {R_028670_SPI_PS_INPUT_CNTL_11, 0, 0},
476 {R_028674_SPI_PS_INPUT_CNTL_12, 0, 0},
477 {R_028678_SPI_PS_INPUT_CNTL_13, 0, 0},
478 {R_02867C_SPI_PS_INPUT_CNTL_14, 0, 0},
479 {R_028680_SPI_PS_INPUT_CNTL_15, 0, 0},
480 {R_028684_SPI_PS_INPUT_CNTL_16, 0, 0},
481 {R_028688_SPI_PS_INPUT_CNTL_17, 0, 0},
482 {R_02868C_SPI_PS_INPUT_CNTL_18, 0, 0},
483 {R_028690_SPI_PS_INPUT_CNTL_19, 0, 0},
484 {R_028694_SPI_PS_INPUT_CNTL_20, 0, 0},
485 {R_028698_SPI_PS_INPUT_CNTL_21, 0, 0},
486 {R_02869C_SPI_PS_INPUT_CNTL_22, 0, 0},
487 {R_0286A0_SPI_PS_INPUT_CNTL_23, 0, 0},
488 {R_0286A4_SPI_PS_INPUT_CNTL_24, 0, 0},
489 {R_0286A8_SPI_PS_INPUT_CNTL_25, 0, 0},
490 {R_0286AC_SPI_PS_INPUT_CNTL_26, 0, 0},
491 {R_0286B0_SPI_PS_INPUT_CNTL_27, 0, 0},
492 {R_0286B4_SPI_PS_INPUT_CNTL_28, 0, 0},
493 {R_0286B8_SPI_PS_INPUT_CNTL_29, 0, 0},
494 {R_0286BC_SPI_PS_INPUT_CNTL_30, 0, 0},
495 {R_0286C0_SPI_PS_INPUT_CNTL_31, 0, 0},
496 {R_0286CC_SPI_PS_IN_CONTROL_0, 0, 0},
497 {R_0286D0_SPI_PS_IN_CONTROL_1, 0, 0},
498 {R_0286D8_SPI_INPUT_Z, 0, 0},
499 {GROUP_FORCE_NEW_BLOCK, 0, 0},
500 {R_028840_SQ_PGM_START_PS, REG_FLAG_NEED_BO, 0},
501 {GROUP_FORCE_NEW_BLOCK, 0, 0},
502 {R_028850_SQ_PGM_RESOURCES_PS, 0, 0},
503 {R_028854_SQ_PGM_EXPORTS_PS, 0, 0},
504 {R_028408_VGT_INDX_OFFSET, 0, 0},
505 {R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, 0, 0},
506 {R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, 0, 0},
507 };
508
509 /* SHADER SAMPLER R600/R700/EG/CM */
510 int r600_state_sampler_init(struct r600_context *ctx, uint32_t offset)
511 {
512 struct r600_reg r600_shader_sampler[] = {
513 {R_03C000_SQ_TEX_SAMPLER_WORD0_0, 0, 0},
514 {R_03C004_SQ_TEX_SAMPLER_WORD1_0, 0, 0},
515 {R_03C008_SQ_TEX_SAMPLER_WORD2_0, 0, 0},
516 };
517 unsigned nreg = Elements(r600_shader_sampler);
518
519 for (int i = 0; i < nreg; i++) {
520 r600_shader_sampler[i].offset += offset;
521 }
522 return r600_context_add_block(ctx, r600_shader_sampler, nreg, PKT3_SET_SAMPLER, R600_SAMPLER_OFFSET);
523 }
524
525 /* SHADER SAMPLER BORDER R600/R700 */
526 static int r600_state_sampler_border_init(struct r600_context *ctx, uint32_t offset)
527 {
528 struct r600_reg r600_shader_sampler_border[] = {
529 {R_00A400_TD_PS_SAMPLER0_BORDER_RED, 0, 0},
530 {R_00A404_TD_PS_SAMPLER0_BORDER_GREEN, 0, 0},
531 {R_00A408_TD_PS_SAMPLER0_BORDER_BLUE, 0, 0},
532 {R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA, 0, 0},
533 };
534 unsigned nreg = Elements(r600_shader_sampler_border);
535
536 for (int i = 0; i < nreg; i++) {
537 r600_shader_sampler_border[i].offset += offset;
538 }
539 return r600_context_add_block(ctx, r600_shader_sampler_border, nreg, PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
540 }
541
542 static int r600_loop_const_init(struct r600_context *ctx, uint32_t offset)
543 {
544 unsigned nreg = 32;
545 struct r600_reg r600_loop_consts[32];
546 int i;
547
548 for (i = 0; i < nreg; i++) {
549 r600_loop_consts[i].offset = R600_LOOP_CONST_OFFSET + ((offset + i) * 4);
550 r600_loop_consts[i].flags = REG_FLAG_DIRTY_ALWAYS;
551 r600_loop_consts[i].sbu_flags = 0;
552 }
553 return r600_context_add_block(ctx, r600_loop_consts, nreg, PKT3_SET_LOOP_CONST, R600_LOOP_CONST_OFFSET);
554 }
555
556 /* initialize */
557 void r600_context_fini(struct r600_context *ctx)
558 {
559 struct r600_block *block;
560 struct r600_range *range;
561
562 if (ctx->range) {
563 for (int i = 0; i < NUM_RANGES; i++) {
564 if (!ctx->range[i].blocks)
565 continue;
566 for (int j = 0; j < (1 << HASH_SHIFT); j++) {
567 block = ctx->range[i].blocks[j];
568 if (block) {
569 for (int k = 0, offset = block->start_offset; k < block->nreg; k++, offset += 4) {
570 range = &ctx->range[CTX_RANGE_ID(offset)];
571 range->blocks[CTX_BLOCK_ID(offset)] = NULL;
572 }
573 for (int k = 1; k <= block->nbo; k++) {
574 pipe_resource_reference((struct pipe_resource**)&block->reloc[k].bo, NULL);
575 }
576 free(block);
577 }
578 }
579 free(ctx->range[i].blocks);
580 }
581 }
582 free(ctx->blocks);
583 }
584
585 int r600_setup_block_table(struct r600_context *ctx)
586 {
587 /* setup block table */
588 int c = 0;
589 ctx->blocks = calloc(ctx->nblocks, sizeof(void*));
590 if (!ctx->blocks)
591 return -ENOMEM;
592 for (int i = 0; i < NUM_RANGES; i++) {
593 if (!ctx->range[i].blocks)
594 continue;
595 for (int j = 0, add; j < (1 << HASH_SHIFT); j++) {
596 if (!ctx->range[i].blocks[j])
597 continue;
598
599 add = 1;
600 for (int k = 0; k < c; k++) {
601 if (ctx->blocks[k] == ctx->range[i].blocks[j]) {
602 add = 0;
603 break;
604 }
605 }
606 if (add) {
607 assert(c < ctx->nblocks);
608 ctx->blocks[c++] = ctx->range[i].blocks[j];
609 j += (ctx->range[i].blocks[j]->nreg) - 1;
610 }
611 }
612 }
613 return 0;
614 }
615
616 int r600_context_init(struct r600_context *ctx)
617 {
618 int r;
619
620 /* add blocks */
621 r = r600_context_add_block(ctx, r600_config_reg_list,
622 Elements(r600_config_reg_list), PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
623 if (r)
624 goto out_err;
625 r = r600_context_add_block(ctx, r600_context_reg_list,
626 Elements(r600_context_reg_list), PKT3_SET_CONTEXT_REG, R600_CONTEXT_REG_OFFSET);
627 if (r)
628 goto out_err;
629 r = r600_context_add_block(ctx, r600_ctl_const_list,
630 Elements(r600_ctl_const_list), PKT3_SET_CTL_CONST, R600_CTL_CONST_OFFSET);
631 if (r)
632 goto out_err;
633
634 /* PS SAMPLER BORDER */
635 for (int j = 0, offset = 0; j < 18; j++, offset += 0x10) {
636 r = r600_state_sampler_border_init(ctx, offset);
637 if (r)
638 goto out_err;
639 }
640
641 /* VS SAMPLER BORDER */
642 for (int j = 0, offset = 0x200; j < 18; j++, offset += 0x10) {
643 r = r600_state_sampler_border_init(ctx, offset);
644 if (r)
645 goto out_err;
646 }
647 /* PS SAMPLER */
648 for (int j = 0, offset = 0; j < 18; j++, offset += 0xC) {
649 r = r600_state_sampler_init(ctx, offset);
650 if (r)
651 goto out_err;
652 }
653 /* VS SAMPLER */
654 for (int j = 0, offset = 0xD8; j < 18; j++, offset += 0xC) {
655 r = r600_state_sampler_init(ctx, offset);
656 if (r)
657 goto out_err;
658 }
659
660 /* PS loop const */
661 r600_loop_const_init(ctx, 0);
662 /* VS loop const */
663 r600_loop_const_init(ctx, 32);
664
665 r = r600_setup_block_table(ctx);
666 if (r)
667 goto out_err;
668
669 ctx->max_db = 4;
670 return 0;
671 out_err:
672 r600_context_fini(ctx);
673 return r;
674 }
675
676 void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
677 boolean count_draw_in)
678 {
679 struct r600_atom *state;
680
681 /* The number of dwords we already used in the CS so far. */
682 num_dw += ctx->cs->cdw;
683
684 if (count_draw_in) {
685 /* The number of dwords all the dirty states would take. */
686 LIST_FOR_EACH_ENTRY(state, &ctx->dirty_states, head) {
687 num_dw += state->num_dw;
688 }
689
690 num_dw += ctx->pm4_dirty_cdwords;
691
692 /* The upper-bound of how much a draw command would take. */
693 num_dw += R600_MAX_DRAW_CS_DWORDS;
694 }
695
696 /* Count in queries_suspend. */
697 num_dw += ctx->num_cs_dw_nontimer_queries_suspend;
698 num_dw += ctx->num_cs_dw_timer_queries_suspend;
699
700 /* Count in streamout_end at the end of CS. */
701 num_dw += ctx->num_cs_dw_streamout_end;
702
703 /* Count in render_condition(NULL) at the end of CS. */
704 if (ctx->predicate_drawing) {
705 num_dw += 3;
706 }
707
708 /* Count in framebuffer cache flushes at the end of CS. */
709 num_dw += 7; /* one SURFACE_SYNC and CACHE_FLUSH_AND_INV (r6xx-only) */
710
711 /* Save 16 dwords for the fence mechanism. */
712 num_dw += 16;
713
714 /* Flush if there's not enough space. */
715 if (num_dw > RADEON_MAX_CMDBUF_DWORDS) {
716 r600_flush(&ctx->context, NULL, RADEON_FLUSH_ASYNC);
717 }
718 }
719
720 void r600_context_dirty_block(struct r600_context *ctx,
721 struct r600_block *block,
722 int dirty, int index)
723 {
724 if ((index + 1) > block->nreg_dirty)
725 block->nreg_dirty = index + 1;
726
727 if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
728 block->status |= R600_BLOCK_STATUS_DIRTY;
729 ctx->pm4_dirty_cdwords += block->pm4_ndwords;
730 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
731 block->status |= R600_BLOCK_STATUS_ENABLED;
732 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
733 }
734 LIST_ADDTAIL(&block->list,&ctx->dirty);
735
736 if (block->flags & REG_FLAG_FLUSH_CHANGE) {
737 r600_context_ps_partial_flush(ctx);
738 }
739 }
740 }
741
742 /**
743 * If reg needs a reloc, this function will add it to its block's reloc list.
744 * @return true if reg needs a reloc, false otherwise
745 */
746 static bool r600_reg_set_block_reloc(struct r600_pipe_reg *reg)
747 {
748 unsigned reloc_id;
749
750 if (!reg->block->pm4_bo_index[reg->id]) {
751 return false;
752 }
753 /* find relocation */
754 reloc_id = reg->block->pm4_bo_index[reg->id];
755 pipe_resource_reference(
756 (struct pipe_resource**)&reg->block->reloc[reloc_id].bo,
757 &reg->bo->b.b);
758 reg->block->reloc[reloc_id].bo_usage = reg->bo_usage;
759 return true;
760 }
761
762 /**
763 * This function will emit all the registers in state directly to the command
764 * stream allowing you to bypass the r600_context dirty list.
765 *
766 * This is used for dispatching compute shaders to avoid mixing compute and
767 * 3D states in the context's dirty list.
768 *
769 * @param pkt_flags Should be either 0 or RADEON_CP_PACKET3_COMPUTE_MODE. This
770 * value will be passed on to r600_context_block_emit_dirty an or'd against
771 * the PKT3 headers.
772 */
773 void r600_context_pipe_state_emit(struct r600_context *ctx,
774 struct r600_pipe_state *state,
775 unsigned pkt_flags)
776 {
777 unsigned i;
778
779 /* Mark all blocks as dirty:
780 * Since two registers can be in the same block, we need to make sure
781 * we mark all the blocks dirty before we emit any of them. If we were
782 * to mark blocks dirty and emit them in the same loop, like this:
783 *
784 * foreach (reg in state->regs) {
785 * mark_dirty(reg->block)
786 * emit_block(reg->block)
787 * }
788 *
789 * Then if we have two registers in this state that are in the same
790 * block, we would end up emitting that block twice.
791 */
792 for (i = 0; i < state->nregs; i++) {
793 struct r600_pipe_reg *reg = &state->regs[i];
794 /* Mark all the registers in the block as dirty */
795 reg->block->nreg_dirty = reg->block->nreg;
796 reg->block->status |= R600_BLOCK_STATUS_DIRTY;
797 /* Update the reloc for this register if necessary. */
798 r600_reg_set_block_reloc(reg);
799 }
800
801 /* Emit the registers writes */
802 for (i = 0; i < state->nregs; i++) {
803 struct r600_pipe_reg *reg = &state->regs[i];
804 if (reg->block->status & R600_BLOCK_STATUS_DIRTY) {
805 r600_context_block_emit_dirty(ctx, reg->block, pkt_flags);
806 }
807 }
808 }
809
810 void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state)
811 {
812 struct r600_block *block;
813 int dirty;
814 for (int i = 0; i < state->nregs; i++) {
815 unsigned id;
816 struct r600_pipe_reg *reg = &state->regs[i];
817
818 block = reg->block;
819 id = reg->id;
820
821 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
822
823 if (reg->value != block->reg[id]) {
824 block->reg[id] = reg->value;
825 dirty |= R600_BLOCK_STATUS_DIRTY;
826 }
827 if (block->flags & REG_FLAG_DIRTY_ALWAYS)
828 dirty |= R600_BLOCK_STATUS_DIRTY;
829 if (r600_reg_set_block_reloc(reg)) {
830 /* always force dirty for relocs for now */
831 dirty |= R600_BLOCK_STATUS_DIRTY;
832 }
833
834 if (dirty)
835 r600_context_dirty_block(ctx, block, dirty, id);
836 }
837 }
838
839 void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
840 {
841 struct r600_range *range;
842 struct r600_block *block;
843 int i;
844 int dirty;
845
846 range = &ctx->range[CTX_RANGE_ID(offset)];
847 block = range->blocks[CTX_BLOCK_ID(offset)];
848 if (state == NULL) {
849 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
850 LIST_DELINIT(&block->list);
851 LIST_DELINIT(&block->enable_list);
852 return;
853 }
854 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
855
856 for (i = 0; i < 3; i++) {
857 if (block->reg[i] != state->regs[i].value) {
858 block->reg[i] = state->regs[i].value;
859 dirty |= R600_BLOCK_STATUS_DIRTY;
860 }
861 }
862
863 if (dirty)
864 r600_context_dirty_block(ctx, block, dirty, 2);
865 }
866
867 static inline void r600_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
868 {
869 struct r600_range *range;
870 struct r600_block *block;
871 int i;
872 int dirty;
873
874 range = &ctx->range[CTX_RANGE_ID(offset)];
875 block = range->blocks[CTX_BLOCK_ID(offset)];
876 if (state == NULL) {
877 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
878 LIST_DELINIT(&block->list);
879 LIST_DELINIT(&block->enable_list);
880 return;
881 }
882 if (state->nregs <= 3) {
883 return;
884 }
885 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
886 for (i = 0; i < 4; i++) {
887 if (block->reg[i] != state->regs[i + 3].value) {
888 block->reg[i] = state->regs[i + 3].value;
889 dirty |= R600_BLOCK_STATUS_DIRTY;
890 }
891 }
892
893 /* We have to flush the shaders before we change the border color
894 * registers, or previous draw commands that haven't completed yet
895 * will end up using the new border color. */
896 if (dirty & R600_BLOCK_STATUS_DIRTY)
897 r600_context_ps_partial_flush(ctx);
898 if (dirty)
899 r600_context_dirty_block(ctx, block, dirty, 3);
900 }
901
902 void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
903 {
904 unsigned offset;
905
906 offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*id;
907 r600_context_pipe_state_set_sampler(ctx, state, offset);
908 offset = R_00A400_TD_PS_SAMPLER0_BORDER_RED + 16*id;
909 r600_context_pipe_state_set_sampler_border(ctx, state, offset);
910 }
911
912 void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
913 {
914 unsigned offset;
915
916 offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*(id + 18);
917 r600_context_pipe_state_set_sampler(ctx, state, offset);
918 offset = R_00A600_TD_VS_SAMPLER0_BORDER_RED + 16*id;
919 r600_context_pipe_state_set_sampler_border(ctx, state, offset);
920 }
921
922 /**
923 * @param pkt_flags should be set to RADEON_CP_PACKET3_COMPUTE_MODE if this
924 * block will be used for compute shaders.
925 */
926 void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block,
927 unsigned pkt_flags)
928 {
929 struct radeon_winsys_cs *cs = ctx->cs;
930 int optional = block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS);
931 int cp_dwords = block->pm4_ndwords, start_dword = 0;
932 int new_dwords = 0;
933 int nbo = block->nbo;
934
935 if (block->nreg_dirty == 0 && optional) {
936 goto out;
937 }
938
939 if (nbo) {
940 for (int j = 0; j < block->nreg; j++) {
941 if (block->pm4_bo_index[j]) {
942 /* find relocation */
943 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]];
944 if (reloc->bo) {
945 block->pm4[reloc->bo_pm4_index] =
946 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage);
947 } else {
948 block->pm4[reloc->bo_pm4_index] = 0;
949 }
950 nbo--;
951 if (nbo == 0)
952 break;
953
954 }
955 }
956 }
957
958 optional &= (block->nreg_dirty != block->nreg);
959 if (optional) {
960 new_dwords = block->nreg_dirty;
961 start_dword = cs->cdw;
962 cp_dwords = new_dwords + 2;
963 }
964 memcpy(&cs->buf[cs->cdw], block->pm4, cp_dwords * 4);
965
966 /* We are applying the pkt_flags after copying the register block to
967 * the the command stream, because it is possible this block will be
968 * emitted with a different pkt_flags, and we don't want to store the
969 * pkt_flags in the block.
970 */
971 cs->buf[cs->cdw] |= pkt_flags;
972 cs->cdw += cp_dwords;
973
974 if (optional) {
975 uint32_t newword;
976
977 newword = cs->buf[start_dword];
978 newword &= PKT_COUNT_C;
979 newword |= PKT_COUNT_S(new_dwords);
980 cs->buf[start_dword] = newword;
981 }
982 out:
983 block->status ^= R600_BLOCK_STATUS_DIRTY;
984 block->nreg_dirty = 0;
985 LIST_DELINIT(&block->list);
986 }
987
988 void r600_inval_shader_cache(struct r600_context *ctx)
989 {
990 ctx->surface_sync_cmd.flush_flags |= S_0085F0_SH_ACTION_ENA(1);
991 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
992 }
993
994 void r600_inval_texture_cache(struct r600_context *ctx)
995 {
996 ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
997 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
998 }
999
1000 void r600_inval_vertex_cache(struct r600_context *ctx)
1001 {
1002 if (ctx->has_vertex_cache) {
1003 ctx->surface_sync_cmd.flush_flags |= S_0085F0_VC_ACTION_ENA(1);
1004 } else {
1005 /* Some GPUs don't have the vertex cache and must use the texture cache instead. */
1006 ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
1007 }
1008 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1009 }
1010
1011 void r600_flush_framebuffer(struct r600_context *ctx, bool flush_now)
1012 {
1013 if (!(ctx->flags & R600_CONTEXT_DST_CACHES_DIRTY))
1014 return;
1015
1016 ctx->surface_sync_cmd.flush_flags |=
1017 r600_get_cb_flush_flags(ctx) |
1018 (ctx->framebuffer.zsbuf ? S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1) : 0);
1019
1020 if (flush_now) {
1021 r600_emit_atom(ctx, &ctx->surface_sync_cmd.atom);
1022 } else {
1023 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1024 }
1025
1026 /* Also add a complete cache flush to work around broken flushing on R6xx. */
1027 if (ctx->chip_class == R600) {
1028 if (flush_now) {
1029 r600_emit_atom(ctx, &ctx->r6xx_flush_and_inv_cmd);
1030 } else {
1031 r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
1032 }
1033 }
1034
1035 ctx->flags &= ~R600_CONTEXT_DST_CACHES_DIRTY;
1036 }
1037
1038 void r600_context_flush(struct r600_context *ctx, unsigned flags)
1039 {
1040 struct radeon_winsys_cs *cs = ctx->cs;
1041 struct r600_block *enable_block = NULL;
1042 bool timer_queries_suspended = false;
1043 bool nontimer_queries_suspended = false;
1044 bool streamout_suspended = false;
1045
1046 if (cs->cdw == ctx->start_cs_cmd.atom.num_dw)
1047 return;
1048
1049 /* suspend queries */
1050 if (ctx->num_cs_dw_timer_queries_suspend) {
1051 r600_suspend_timer_queries(ctx);
1052 timer_queries_suspended = true;
1053 }
1054 if (ctx->num_cs_dw_nontimer_queries_suspend) {
1055 r600_suspend_nontimer_queries(ctx);
1056 nontimer_queries_suspended = true;
1057 }
1058
1059 if (ctx->num_cs_dw_streamout_end) {
1060 r600_context_streamout_end(ctx);
1061 streamout_suspended = true;
1062 }
1063
1064 r600_flush_framebuffer(ctx, true);
1065
1066 /* partial flush is needed to avoid lockups on some chips with user fences */
1067 r600_context_ps_partial_flush(ctx);
1068
1069 /* old kernels and userspace don't set SX_MISC, so we must reset it to 0 here */
1070 if (ctx->chip_class <= R700) {
1071 r600_write_context_reg(cs, R_028350_SX_MISC, 0);
1072 }
1073
1074 /* force to keep tiling flags */
1075 flags |= RADEON_FLUSH_KEEP_TILING_FLAGS;
1076
1077 /* Flush the CS. */
1078 ctx->ws->cs_flush(ctx->cs, flags);
1079
1080 ctx->pm4_dirty_cdwords = 0;
1081 ctx->flags = 0;
1082
1083 /* Begin a new CS. */
1084 r600_emit_atom(ctx, &ctx->start_cs_cmd.atom);
1085
1086 /* Invalidate caches. */
1087 r600_inval_texture_cache(ctx);
1088 r600_flush_framebuffer(ctx, false);
1089
1090 /* Re-emit states. */
1091 r600_atom_dirty(ctx, &ctx->alphatest_state.atom);
1092 r600_atom_dirty(ctx, &ctx->cb_misc_state.atom);
1093 r600_atom_dirty(ctx, &ctx->db_misc_state.atom);
1094
1095 ctx->vertex_buffer_state.dirty_mask = ctx->vertex_buffer_state.enabled_mask;
1096 r600_vertex_buffers_dirty(ctx);
1097
1098 ctx->vs_constbuf_state.dirty_mask = ctx->vs_constbuf_state.enabled_mask;
1099 ctx->ps_constbuf_state.dirty_mask = ctx->ps_constbuf_state.enabled_mask;
1100 r600_constant_buffers_dirty(ctx, &ctx->vs_constbuf_state);
1101 r600_constant_buffers_dirty(ctx, &ctx->ps_constbuf_state);
1102
1103 ctx->vs_samplers.views.dirty_mask = ctx->vs_samplers.views.enabled_mask;
1104 ctx->ps_samplers.views.dirty_mask = ctx->ps_samplers.views.enabled_mask;
1105 r600_sampler_views_dirty(ctx, &ctx->vs_samplers.views);
1106 r600_sampler_views_dirty(ctx, &ctx->ps_samplers.views);
1107
1108 if (streamout_suspended) {
1109 ctx->streamout_start = TRUE;
1110 ctx->streamout_append_bitmask = ~0;
1111 }
1112
1113 /* resume queries */
1114 if (timer_queries_suspended) {
1115 r600_resume_timer_queries(ctx);
1116 }
1117 if (nontimer_queries_suspended) {
1118 r600_resume_nontimer_queries(ctx);
1119 }
1120
1121 /* set all valid group as dirty so they get reemited on
1122 * next draw command
1123 */
1124 LIST_FOR_EACH_ENTRY(enable_block, &ctx->enable_list, enable_list) {
1125 if(!(enable_block->status & R600_BLOCK_STATUS_DIRTY)) {
1126 LIST_ADDTAIL(&enable_block->list,&ctx->dirty);
1127 enable_block->status |= R600_BLOCK_STATUS_DIRTY;
1128 }
1129 ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords;
1130 enable_block->nreg_dirty = enable_block->nreg;
1131 }
1132 }
1133
1134 void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value)
1135 {
1136 struct radeon_winsys_cs *cs = ctx->cs;
1137 uint64_t va;
1138
1139 r600_need_cs_space(ctx, 10, FALSE);
1140
1141 va = r600_resource_va(&ctx->screen->screen, (void*)fence_bo);
1142 va = va + (offset << 2);
1143
1144 r600_context_ps_partial_flush(ctx);
1145 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1146 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1147 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* ADDRESS_LO */
1148 /* DATA_SEL | INT_EN | ADDRESS_HI */
1149 cs->buf[cs->cdw++] = (1 << 29) | (0 << 24) | ((va >> 32UL) & 0xFF);
1150 cs->buf[cs->cdw++] = value; /* DATA_LO */
1151 cs->buf[cs->cdw++] = 0; /* DATA_HI */
1152 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1153 cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, fence_bo, RADEON_USAGE_WRITE);
1154 }
1155
1156 static void r600_flush_vgt_streamout(struct r600_context *ctx)
1157 {
1158 struct radeon_winsys_cs *cs = ctx->cs;
1159
1160 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0);
1161 cs->buf[cs->cdw++] = (R_008490_CP_STRMOUT_CNTL - R600_CONFIG_REG_OFFSET) >> 2;
1162 cs->buf[cs->cdw++] = 0;
1163
1164 cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
1165 cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0);
1166
1167 cs->buf[cs->cdw++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0);
1168 cs->buf[cs->cdw++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */
1169 cs->buf[cs->cdw++] = R_008490_CP_STRMOUT_CNTL >> 2; /* register */
1170 cs->buf[cs->cdw++] = 0;
1171 cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* reference value */
1172 cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* mask */
1173 cs->buf[cs->cdw++] = 4; /* poll interval */
1174 }
1175
1176 static void r600_set_streamout_enable(struct r600_context *ctx, unsigned buffer_enable_bit)
1177 {
1178 struct radeon_winsys_cs *cs = ctx->cs;
1179
1180 if (buffer_enable_bit) {
1181 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1182 cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1183 cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(1);
1184
1185 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1186 cs->buf[cs->cdw++] = (R_028B20_VGT_STRMOUT_BUFFER_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1187 cs->buf[cs->cdw++] = buffer_enable_bit;
1188 } else {
1189 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1190 cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1191 cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(0);
1192 }
1193 }
1194
1195 void r600_context_streamout_begin(struct r600_context *ctx)
1196 {
1197 struct radeon_winsys_cs *cs = ctx->cs;
1198 struct r600_so_target **t = ctx->so_targets;
1199 unsigned *stride_in_dw = ctx->vs_shader->so.stride;
1200 unsigned buffer_en, i, update_flags = 0;
1201 uint64_t va;
1202
1203 buffer_en = (ctx->num_so_targets >= 1 && t[0] ? 1 : 0) |
1204 (ctx->num_so_targets >= 2 && t[1] ? 2 : 0) |
1205 (ctx->num_so_targets >= 3 && t[2] ? 4 : 0) |
1206 (ctx->num_so_targets >= 4 && t[3] ? 8 : 0);
1207
1208 ctx->num_cs_dw_streamout_end =
1209 12 + /* flush_vgt_streamout */
1210 util_bitcount(buffer_en) * 8 + /* STRMOUT_BUFFER_UPDATE */
1211 3 /* set_streamout_enable(0) */;
1212
1213 r600_need_cs_space(ctx,
1214 12 + /* flush_vgt_streamout */
1215 6 + /* set_streamout_enable */
1216 util_bitcount(buffer_en) * 7 + /* SET_CONTEXT_REG */
1217 (ctx->chip_class == R700 ? util_bitcount(buffer_en) * 5 : 0) + /* STRMOUT_BASE_UPDATE */
1218 util_bitcount(buffer_en & ctx->streamout_append_bitmask) * 8 + /* STRMOUT_BUFFER_UPDATE */
1219 util_bitcount(buffer_en & ~ctx->streamout_append_bitmask) * 6 + /* STRMOUT_BUFFER_UPDATE */
1220 (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770 ? 2 : 0) + /* SURFACE_BASE_UPDATE */
1221 ctx->num_cs_dw_streamout_end, TRUE);
1222
1223 if (ctx->chip_class >= EVERGREEN) {
1224 evergreen_flush_vgt_streamout(ctx);
1225 evergreen_set_streamout_enable(ctx, buffer_en);
1226 } else {
1227 r600_flush_vgt_streamout(ctx);
1228 r600_set_streamout_enable(ctx, buffer_en);
1229 }
1230
1231 for (i = 0; i < ctx->num_so_targets; i++) {
1232 if (t[i]) {
1233 t[i]->stride_in_dw = stride_in_dw[i];
1234 t[i]->so_index = i;
1235 va = r600_resource_va(&ctx->screen->screen,
1236 (void*)t[i]->b.buffer);
1237
1238 update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i);
1239
1240 cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 3, 0);
1241 cs->buf[cs->cdw++] = (R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 +
1242 16*i - R600_CONTEXT_REG_OFFSET) >> 2;
1243 cs->buf[cs->cdw++] = (t[i]->b.buffer_offset +
1244 t[i]->b.buffer_size) >> 2; /* BUFFER_SIZE (in DW) */
1245 cs->buf[cs->cdw++] = stride_in_dw[i]; /* VTX_STRIDE (in DW) */
1246 cs->buf[cs->cdw++] = va >> 8; /* BUFFER_BASE */
1247
1248 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1249 cs->buf[cs->cdw++] =
1250 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1251 RADEON_USAGE_WRITE);
1252
1253 /* R7xx requires this packet after updating BUFFER_BASE.
1254 * Without this, R7xx locks up. */
1255 if (ctx->chip_class == R700) {
1256 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BASE_UPDATE, 1, 0);
1257 cs->buf[cs->cdw++] = i;
1258 cs->buf[cs->cdw++] = va >> 8;
1259
1260 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1261 cs->buf[cs->cdw++] =
1262 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1263 RADEON_USAGE_WRITE);
1264 }
1265
1266 if (ctx->streamout_append_bitmask & (1 << i)) {
1267 va = r600_resource_va(&ctx->screen->screen,
1268 (void*)t[i]->filled_size);
1269 /* Append. */
1270 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1271 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1272 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_MEM); /* control */
1273 cs->buf[cs->cdw++] = 0; /* unused */
1274 cs->buf[cs->cdw++] = 0; /* unused */
1275 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* src address lo */
1276 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* src address hi */
1277
1278 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1279 cs->buf[cs->cdw++] =
1280 r600_context_bo_reloc(ctx, t[i]->filled_size,
1281 RADEON_USAGE_READ);
1282 } else {
1283 /* Start from the beginning. */
1284 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1285 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1286 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_PACKET); /* control */
1287 cs->buf[cs->cdw++] = 0; /* unused */
1288 cs->buf[cs->cdw++] = 0; /* unused */
1289 cs->buf[cs->cdw++] = t[i]->b.buffer_offset >> 2; /* buffer offset in DW */
1290 cs->buf[cs->cdw++] = 0; /* unused */
1291 }
1292 }
1293 }
1294
1295 if (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770) {
1296 cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
1297 cs->buf[cs->cdw++] = update_flags;
1298 }
1299 }
1300
1301 void r600_context_streamout_end(struct r600_context *ctx)
1302 {
1303 struct radeon_winsys_cs *cs = ctx->cs;
1304 struct r600_so_target **t = ctx->so_targets;
1305 unsigned i, flush_flags = 0;
1306 uint64_t va;
1307
1308 if (ctx->chip_class >= EVERGREEN) {
1309 evergreen_flush_vgt_streamout(ctx);
1310 } else {
1311 r600_flush_vgt_streamout(ctx);
1312 }
1313
1314 for (i = 0; i < ctx->num_so_targets; i++) {
1315 if (t[i]) {
1316 va = r600_resource_va(&ctx->screen->screen,
1317 (void*)t[i]->filled_size);
1318 cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1319 cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1320 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_NONE) |
1321 STRMOUT_STORE_BUFFER_FILLED_SIZE; /* control */
1322 cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* dst address lo */
1323 cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* dst address hi */
1324 cs->buf[cs->cdw++] = 0; /* unused */
1325 cs->buf[cs->cdw++] = 0; /* unused */
1326
1327 cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1328 cs->buf[cs->cdw++] =
1329 r600_context_bo_reloc(ctx, t[i]->filled_size,
1330 RADEON_USAGE_WRITE);
1331
1332 flush_flags |= S_0085F0_SO0_DEST_BASE_ENA(1) << i;
1333 }
1334 }
1335
1336 if (ctx->chip_class >= EVERGREEN) {
1337 evergreen_set_streamout_enable(ctx, 0);
1338 } else {
1339 r600_set_streamout_enable(ctx, 0);
1340 }
1341
1342 /* This is needed to fix cache flushes on r600. */
1343 if (ctx->chip_class == R600) {
1344 if (ctx->family == CHIP_RV670 ||
1345 ctx->family == CHIP_RS780 ||
1346 ctx->family == CHIP_RS880) {
1347 flush_flags |= S_0085F0_DEST_BASE_0_ENA(1);
1348 }
1349
1350 r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
1351 }
1352
1353 /* Flush streamout caches. */
1354 ctx->surface_sync_cmd.flush_flags |=
1355 S_0085F0_SMX_ACTION_ENA(1) | flush_flags;
1356 r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1357
1358 ctx->num_cs_dw_streamout_end = 0;
1359 }