2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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:
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
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.
31 #include <pipe/p_compiler.h>
32 #include <util/u_inlines.h>
33 #include <util/u_memory.h>
34 #include <pipebuffer/pb_bufmgr.h>
36 #include "radeon_drm.h"
37 #include "r600_priv.h"
41 #define GROUP_FORCE_NEW_BLOCK 0
43 /* Get backends mask */
44 void r600_get_backend_mask(struct r600_context *ctx)
46 struct r600_bo * buffer;
48 unsigned num_backends = r600_get_num_backends(ctx->radeon);
51 /* if backend_map query is supported by the kernel */
52 if (ctx->radeon->backend_map_valid) {
53 unsigned num_tile_pipes = r600_get_num_tile_pipes(ctx->radeon);
54 unsigned backend_map = r600_get_backend_map(ctx->radeon);
55 unsigned item_width, item_mask;
57 if (ctx->radeon->chip_class >= EVERGREEN) {
65 while(num_tile_pipes--) {
66 i = backend_map & item_mask;
68 backend_map >>= item_width;
71 ctx->backend_mask = mask;
76 /* otherwise backup path for older kernels */
78 /* create buffer for event data */
79 buffer = r600_bo(ctx->radeon, ctx->max_db*16, 1, 0,
84 /* initialize buffer with zeroes */
85 results = r600_bo_map(ctx->radeon, buffer, PB_USAGE_CPU_WRITE, NULL);
87 memset(results, 0, ctx->max_db * 4 * 4);
88 r600_bo_unmap(ctx->radeon, buffer);
90 /* emit EVENT_WRITE for ZPASS_DONE */
91 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
92 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
93 ctx->pm4[ctx->pm4_cdwords++] = 0;
94 ctx->pm4[ctx->pm4_cdwords++] = 0;
96 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
97 ctx->pm4[ctx->pm4_cdwords++] = 0;
98 r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], buffer);
101 r600_context_flush(ctx);
103 /* analyze results */
104 results = r600_bo_map(ctx->radeon, buffer, PB_USAGE_CPU_READ, NULL);
106 for(i = 0; i < ctx->max_db; i++) {
107 /* at least highest bit will be set if backend is used */
108 if (results[i*4 + 1])
111 r600_bo_unmap(ctx->radeon, buffer);
115 r600_bo_reference(ctx->radeon, &buffer, NULL);
118 ctx->backend_mask = mask;
123 /* fallback to old method - set num_backends lower bits to 1 */
124 ctx->backend_mask = (~((u32)0))>>(32-num_backends);
128 static inline void r600_context_ps_partial_flush(struct r600_context *ctx)
130 if (!(ctx->flags & R600_CONTEXT_DRAW_PENDING))
133 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
134 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
136 ctx->flags &= ~R600_CONTEXT_DRAW_PENDING;
139 void r600_init_cs(struct r600_context *ctx)
141 /* R6xx requires this packet at the start of each command buffer */
142 if (ctx->radeon->family < CHIP_RV770) {
143 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_START_3D_CMDBUF, 0, 0);
144 ctx->pm4[ctx->pm4_cdwords++] = 0x00000000;
146 /* All asics require this one */
147 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_CONTEXT_CONTROL, 1, 0);
148 ctx->pm4[ctx->pm4_cdwords++] = 0x80000000;
149 ctx->pm4[ctx->pm4_cdwords++] = 0x80000000;
151 ctx->init_dwords = ctx->pm4_cdwords;
154 static void INLINE r600_context_update_fenced_list(struct r600_context *ctx)
156 for (int i = 0; i < ctx->creloc; i++) {
157 if (!LIST_IS_EMPTY(&ctx->bo[i]->fencedlist))
158 LIST_DELINIT(&ctx->bo[i]->fencedlist);
159 LIST_ADDTAIL(&ctx->bo[i]->fencedlist, &ctx->fenced_bo);
160 ctx->bo[i]->fence = ctx->radeon->fence;
161 ctx->bo[i]->ctx = ctx;
165 static void INLINE r600_context_fence_wraparound(struct r600_context *ctx, unsigned fence)
167 struct radeon_bo *bo = NULL;
168 struct radeon_bo *tmp;
170 LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &ctx->fenced_bo, fencedlist) {
171 if (bo->fence <= *ctx->radeon->cfence) {
172 LIST_DELINIT(&bo->fencedlist);
180 static void r600_init_block(struct r600_context *ctx,
181 struct r600_block *block,
182 const struct r600_reg *reg, int index, int nreg,
183 unsigned opcode, unsigned offset_base)
188 /* initialize block */
189 if (opcode == PKT3_SET_RESOURCE) {
190 block->flags = BLOCK_FLAG_RESOURCE;
191 block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY; /* dirty all blocks at start */
194 block->status |= R600_BLOCK_STATUS_DIRTY; /* dirty all blocks at start */
196 block->start_offset = reg[i].offset;
197 block->pm4[block->pm4_ndwords++] = PKT3(opcode, n, 0);
198 block->pm4[block->pm4_ndwords++] = (block->start_offset - offset_base) >> 2;
199 block->reg = &block->pm4[block->pm4_ndwords];
200 block->pm4_ndwords += n;
202 block->nreg_dirty = n;
203 LIST_INITHEAD(&block->list);
204 LIST_INITHEAD(&block->enable_list);
206 for (j = 0; j < n; j++) {
207 if (reg[i+j].flags & REG_FLAG_DIRTY_ALWAYS) {
208 block->flags |= REG_FLAG_DIRTY_ALWAYS;
210 if (reg[i+j].flags & REG_FLAG_ENABLE_ALWAYS) {
211 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
212 block->status |= R600_BLOCK_STATUS_ENABLED;
213 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
214 LIST_ADDTAIL(&block->list,&ctx->dirty);
217 if (reg[i+j].flags & REG_FLAG_FLUSH_CHANGE) {
218 block->flags |= REG_FLAG_FLUSH_CHANGE;
221 if (reg[i+j].flags & REG_FLAG_NEED_BO) {
223 assert(block->nbo < R600_BLOCK_MAX_BO);
224 block->pm4_bo_index[j] = block->nbo;
225 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0, 0);
226 block->pm4[block->pm4_ndwords++] = 0x00000000;
227 if (reg[i+j].flags & REG_FLAG_RV6XX_SBU) {
228 block->reloc[block->nbo].flush_flags = 0;
229 block->reloc[block->nbo].flush_mask = 0;
231 block->reloc[block->nbo].flush_flags = reg[i+j].flush_flags;
232 block->reloc[block->nbo].flush_mask = reg[i+j].flush_mask;
234 block->reloc[block->nbo].bo_pm4_index = block->pm4_ndwords - 1;
236 if ((ctx->radeon->family > CHIP_R600) &&
237 (ctx->radeon->family < CHIP_RV770) && reg[i+j].flags & REG_FLAG_RV6XX_SBU) {
238 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
239 block->pm4[block->pm4_ndwords++] = reg[i+j].flush_flags;
242 for (j = 0; j < n; j++) {
243 if (reg[i+j].flush_flags) {
244 block->pm4_flush_ndwords += 7;
247 /* check that we stay in limit */
248 assert(block->pm4_ndwords < R600_BLOCK_MAX_REG);
251 int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg,
252 unsigned opcode, unsigned offset_base)
254 struct r600_block *block;
255 struct r600_range *range;
258 for (unsigned i = 0, n = 0; i < nreg; i += n) {
259 /* ignore new block balise */
260 if (reg[i].offset == GROUP_FORCE_NEW_BLOCK) {
265 /* ignore regs not on R600 on R600 */
266 if ((reg[i].flags & REG_FLAG_NOT_R600) && ctx->radeon->family == CHIP_R600) {
271 /* register that need relocation are in their own group */
272 /* find number of consecutive registers */
274 offset = reg[i].offset;
275 while (reg[i + n].offset == offset) {
280 if (n >= (R600_BLOCK_MAX_REG - 2))
284 /* allocate new block */
285 block = calloc(1, sizeof(struct r600_block));
290 for (int j = 0; j < n; j++) {
291 range = &ctx->range[CTX_RANGE_ID(reg[i + j].offset)];
292 /* create block table if it doesn't exist */
294 range->blocks = calloc(1 << HASH_SHIFT, sizeof(void *));
298 range->blocks[CTX_BLOCK_ID(reg[i + j].offset)] = block;
301 r600_init_block(ctx, block, reg, i, n, opcode, offset_base);
307 /* R600/R700 configuration */
308 static const struct r600_reg r600_config_reg_list[] = {
309 {R_008958_VGT_PRIMITIVE_TYPE, 0, 0, 0},
310 {R_008C00_SQ_CONFIG, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
311 {R_008C04_SQ_GPR_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
312 {R_008C08_SQ_GPR_RESOURCE_MGMT_2, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
313 {R_008C0C_SQ_THREAD_RESOURCE_MGMT, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
314 {R_008C10_SQ_STACK_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
315 {R_008C14_SQ_STACK_RESOURCE_MGMT_2, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
316 {R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
317 {R_009508_TA_CNTL_AUX, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
318 {R_009714_VC_ENHANCE, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
319 {R_009830_DB_DEBUG, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
320 {R_009838_DB_WATERMARKS, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0},
323 static const struct r600_reg r600_ctl_const_list[] = {
324 {R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0, 0, 0},
325 {R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0, 0},
328 static const struct r600_reg r600_context_reg_list[] = {
329 {R_028350_SX_MISC, 0, 0, 0},
330 {R_0286C8_SPI_THREAD_GROUPING, 0, 0, 0},
331 {R_0288A8_SQ_ESGS_RING_ITEMSIZE, 0, 0, 0},
332 {R_0288AC_SQ_GSVS_RING_ITEMSIZE, 0, 0, 0},
333 {R_0288B0_SQ_ESTMP_RING_ITEMSIZE, 0, 0, 0},
334 {R_0288B4_SQ_GSTMP_RING_ITEMSIZE, 0, 0, 0},
335 {R_0288B8_SQ_VSTMP_RING_ITEMSIZE, 0, 0, 0},
336 {R_0288BC_SQ_PSTMP_RING_ITEMSIZE, 0, 0, 0},
337 {R_0288C0_SQ_FBUF_RING_ITEMSIZE, 0, 0, 0},
338 {R_0288C4_SQ_REDUC_RING_ITEMSIZE, 0, 0, 0},
339 {R_0288C8_SQ_GS_VERT_ITEMSIZE, 0, 0, 0},
340 {R_028A10_VGT_OUTPUT_PATH_CNTL, 0, 0, 0},
341 {R_028A14_VGT_HOS_CNTL, 0, 0, 0},
342 {R_028A18_VGT_HOS_MAX_TESS_LEVEL, 0, 0, 0},
343 {R_028A1C_VGT_HOS_MIN_TESS_LEVEL, 0, 0, 0},
344 {R_028A20_VGT_HOS_REUSE_DEPTH, 0, 0, 0},
345 {R_028A24_VGT_GROUP_PRIM_TYPE, 0, 0, 0},
346 {R_028A28_VGT_GROUP_FIRST_DECR, 0, 0, 0},
347 {R_028A2C_VGT_GROUP_DECR, 0, 0, 0},
348 {R_028A30_VGT_GROUP_VECT_0_CNTL, 0, 0, 0},
349 {R_028A34_VGT_GROUP_VECT_1_CNTL, 0, 0, 0},
350 {R_028A38_VGT_GROUP_VECT_0_FMT_CNTL, 0, 0, 0},
351 {R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL, 0, 0, 0},
352 {R_028A40_VGT_GS_MODE, 0, 0, 0},
353 {R_028A4C_PA_SC_MODE_CNTL, 0, 0, 0},
354 {R_028AB0_VGT_STRMOUT_EN, 0, 0, 0},
355 {R_028AB4_VGT_REUSE_OFF, 0, 0, 0},
356 {R_028AB8_VGT_VTX_CNT_EN, 0, 0, 0},
357 {R_028B20_VGT_STRMOUT_BUFFER_EN, 0, 0, 0},
358 {R_028028_DB_STENCIL_CLEAR, 0, 0, 0},
359 {R_02802C_DB_DEPTH_CLEAR, 0, 0, 0},
360 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
361 {R_028040_CB_COLOR0_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(0), 0},
362 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
363 {R_0280A0_CB_COLOR0_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
364 {R_028060_CB_COLOR0_SIZE, 0, 0, 0},
365 {R_028080_CB_COLOR0_VIEW, 0, 0, 0},
366 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
367 {R_0280E0_CB_COLOR0_FRAG, REG_FLAG_NEED_BO, 0, 0},
368 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
369 {R_0280C0_CB_COLOR0_TILE, REG_FLAG_NEED_BO, 0, 0},
370 {R_028100_CB_COLOR0_MASK, 0, 0, 0},
371 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
372 {R_028044_CB_COLOR1_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(1), 0},
373 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
374 {R_0280A4_CB_COLOR1_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
375 {R_028064_CB_COLOR1_SIZE, 0, 0, 0},
376 {R_028084_CB_COLOR1_VIEW, 0, 0, 0},
377 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
378 {R_0280E4_CB_COLOR1_FRAG, REG_FLAG_NEED_BO, 0, 0},
379 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
380 {R_0280C4_CB_COLOR1_TILE, REG_FLAG_NEED_BO, 0, 0},
381 {R_028104_CB_COLOR1_MASK, 0, 0, 0},
382 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
383 {R_028048_CB_COLOR2_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(2), 0},
384 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
385 {R_0280A8_CB_COLOR2_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
386 {R_028068_CB_COLOR2_SIZE, 0, 0, 0},
387 {R_028088_CB_COLOR2_VIEW, 0, 0, 0},
388 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
389 {R_0280E8_CB_COLOR2_FRAG, REG_FLAG_NEED_BO, 0, 0},
390 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
391 {R_0280C8_CB_COLOR2_TILE, REG_FLAG_NEED_BO, 0, 0},
392 {R_028108_CB_COLOR2_MASK, 0, 0, 0},
393 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
394 {R_02804C_CB_COLOR3_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(3), 0},
395 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
396 {R_0280AC_CB_COLOR3_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
397 {R_02806C_CB_COLOR3_SIZE, 0, 0, 0},
398 {R_02808C_CB_COLOR3_VIEW, 0, 0, 0},
399 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
400 {R_0280EC_CB_COLOR3_FRAG, REG_FLAG_NEED_BO, 0, 0},
401 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
402 {R_0280CC_CB_COLOR3_TILE, REG_FLAG_NEED_BO, 0, 0},
403 {R_02810C_CB_COLOR3_MASK, 0, 0, 0},
404 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
405 {R_028050_CB_COLOR4_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(4), 0},
406 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
407 {R_0280B0_CB_COLOR4_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
408 {R_028070_CB_COLOR4_SIZE, 0, 0, 0},
409 {R_028090_CB_COLOR4_VIEW, 0, 0, 0},
410 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
411 {R_0280F0_CB_COLOR4_FRAG, REG_FLAG_NEED_BO, 0, 0},
412 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
413 {R_0280D0_CB_COLOR4_TILE, REG_FLAG_NEED_BO, 0, 0},
414 {R_028110_CB_COLOR4_MASK, 0, 0, 0},
415 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
416 {R_028054_CB_COLOR5_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(5), 0},
417 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
418 {R_0280B4_CB_COLOR5_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
419 {R_028074_CB_COLOR5_SIZE, 0, 0, 0},
420 {R_028094_CB_COLOR5_VIEW, 0, 0, 0},
421 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
422 {R_0280F4_CB_COLOR5_FRAG, REG_FLAG_NEED_BO, 0, 0},
423 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
424 {R_0280D4_CB_COLOR5_TILE, REG_FLAG_NEED_BO, 0, 0},
425 {R_028114_CB_COLOR5_MASK, 0, 0, 0},
426 {R_028058_CB_COLOR6_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(6), 0},
427 {R_0280B8_CB_COLOR6_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
428 {R_028078_CB_COLOR6_SIZE, 0, 0, 0},
429 {R_028098_CB_COLOR6_VIEW, 0, 0, 0},
430 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
431 {R_0280F8_CB_COLOR6_FRAG, REG_FLAG_NEED_BO, 0, 0},
432 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
433 {R_0280D8_CB_COLOR6_TILE, REG_FLAG_NEED_BO, 0, 0},
434 {R_028118_CB_COLOR6_MASK, 0, 0, 0},
435 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
436 {R_02805C_CB_COLOR7_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(7), 0},
437 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
438 {R_0280BC_CB_COLOR7_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF},
439 {R_02807C_CB_COLOR7_SIZE, 0, 0, 0},
440 {R_02809C_CB_COLOR7_VIEW, 0, 0, 0},
441 {R_0280FC_CB_COLOR7_FRAG, REG_FLAG_NEED_BO, 0, 0},
442 {R_0280DC_CB_COLOR7_TILE, REG_FLAG_NEED_BO, 0, 0},
443 {R_02811C_CB_COLOR7_MASK, 0, 0, 0},
444 {R_028120_CB_CLEAR_RED, 0, 0, 0},
445 {R_028124_CB_CLEAR_GREEN, 0, 0, 0},
446 {R_028128_CB_CLEAR_BLUE, 0, 0, 0},
447 {R_02812C_CB_CLEAR_ALPHA, 0, 0, 0},
448 {R_028140_ALU_CONST_BUFFER_SIZE_PS_0, REG_FLAG_DIRTY_ALWAYS, 0, 0},
449 {R_028180_ALU_CONST_BUFFER_SIZE_VS_0, REG_FLAG_DIRTY_ALWAYS, 0, 0},
450 {R_028940_ALU_CONST_CACHE_PS_0, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF},
451 {R_028980_ALU_CONST_CACHE_VS_0, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF},
452 {R_02823C_CB_SHADER_MASK, 0, 0, 0},
453 {R_028238_CB_TARGET_MASK, 0, 0, 0},
454 {R_028410_SX_ALPHA_TEST_CONTROL, 0, 0, 0},
455 {R_028414_CB_BLEND_RED, 0, 0, 0},
456 {R_028418_CB_BLEND_GREEN, 0, 0, 0},
457 {R_02841C_CB_BLEND_BLUE, 0, 0, 0},
458 {R_028420_CB_BLEND_ALPHA, 0, 0, 0},
459 {R_028424_CB_FOG_RED, 0, 0, 0},
460 {R_028428_CB_FOG_GREEN, 0, 0, 0},
461 {R_02842C_CB_FOG_BLUE, 0, 0, 0},
462 {R_028430_DB_STENCILREFMASK, 0, 0, 0},
463 {R_028434_DB_STENCILREFMASK_BF, 0, 0, 0},
464 {R_028438_SX_ALPHA_REF, 0, 0, 0},
465 {R_0286DC_SPI_FOG_CNTL, 0, 0, 0},
466 {R_0286E0_SPI_FOG_FUNC_SCALE, 0, 0, 0},
467 {R_0286E4_SPI_FOG_FUNC_BIAS, 0, 0, 0},
468 {R_028780_CB_BLEND0_CONTROL, REG_FLAG_NOT_R600, 0, 0},
469 {R_028784_CB_BLEND1_CONTROL, REG_FLAG_NOT_R600, 0, 0},
470 {R_028788_CB_BLEND2_CONTROL, REG_FLAG_NOT_R600, 0, 0},
471 {R_02878C_CB_BLEND3_CONTROL, REG_FLAG_NOT_R600, 0, 0},
472 {R_028790_CB_BLEND4_CONTROL, REG_FLAG_NOT_R600, 0, 0},
473 {R_028794_CB_BLEND5_CONTROL, REG_FLAG_NOT_R600, 0, 0},
474 {R_028798_CB_BLEND6_CONTROL, REG_FLAG_NOT_R600, 0, 0},
475 {R_02879C_CB_BLEND7_CONTROL, REG_FLAG_NOT_R600, 0, 0},
476 {R_0287A0_CB_SHADER_CONTROL, 0, 0, 0},
477 {R_028800_DB_DEPTH_CONTROL, 0, 0, 0},
478 {R_028804_CB_BLEND_CONTROL, 0, 0, 0},
479 {R_028808_CB_COLOR_CONTROL, 0, 0, 0},
480 {R_02880C_DB_SHADER_CONTROL, 0, 0, 0},
481 {R_028C04_PA_SC_AA_CONFIG, 0, 0, 0},
482 {R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 0, 0, 0},
483 {R_028C20_PA_SC_AA_SAMPLE_LOCS_8S_WD1_MCTX, 0, 0, 0},
484 {R_028C30_CB_CLRCMP_CONTROL, 0, 0, 0},
485 {R_028C34_CB_CLRCMP_SRC, 0, 0, 0},
486 {R_028C38_CB_CLRCMP_DST, 0, 0, 0},
487 {R_028C3C_CB_CLRCMP_MSK, 0, 0, 0},
488 {R_028C48_PA_SC_AA_MASK, 0, 0, 0},
489 {R_028D2C_DB_SRESULTS_COMPARE_STATE1, 0, 0, 0},
490 {R_028D44_DB_ALPHA_TO_MASK, 0, 0, 0},
491 {R_02800C_DB_DEPTH_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_DEPTH, 0},
492 {R_028000_DB_DEPTH_SIZE, 0, 0, 0},
493 {R_028004_DB_DEPTH_VIEW, 0, 0, 0},
494 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
495 {R_028010_DB_DEPTH_INFO, REG_FLAG_NEED_BO, 0, 0},
496 {R_028D0C_DB_RENDER_CONTROL, 0, 0, 0},
497 {R_028D10_DB_RENDER_OVERRIDE, 0, 0, 0},
498 {R_028D24_DB_HTILE_SURFACE, 0, 0, 0},
499 {R_028D30_DB_PRELOAD_CONTROL, 0, 0, 0},
500 {R_028D34_DB_PREFETCH_LIMIT, 0, 0, 0},
501 {R_028030_PA_SC_SCREEN_SCISSOR_TL, 0, 0, 0},
502 {R_028034_PA_SC_SCREEN_SCISSOR_BR, 0, 0, 0},
503 {R_028200_PA_SC_WINDOW_OFFSET, 0, 0, 0},
504 {R_028204_PA_SC_WINDOW_SCISSOR_TL, 0, 0, 0},
505 {R_028208_PA_SC_WINDOW_SCISSOR_BR, 0, 0, 0},
506 {R_02820C_PA_SC_CLIPRECT_RULE, 0, 0, 0},
507 {R_028210_PA_SC_CLIPRECT_0_TL, 0, 0, 0},
508 {R_028214_PA_SC_CLIPRECT_0_BR, 0, 0, 0},
509 {R_028218_PA_SC_CLIPRECT_1_TL, 0, 0, 0},
510 {R_02821C_PA_SC_CLIPRECT_1_BR, 0, 0, 0},
511 {R_028220_PA_SC_CLIPRECT_2_TL, 0, 0, 0},
512 {R_028224_PA_SC_CLIPRECT_2_BR, 0, 0, 0},
513 {R_028228_PA_SC_CLIPRECT_3_TL, 0, 0, 0},
514 {R_02822C_PA_SC_CLIPRECT_3_BR, 0, 0, 0},
515 {R_028230_PA_SC_EDGERULE, 0, 0, 0},
516 {R_028240_PA_SC_GENERIC_SCISSOR_TL, 0, 0, 0},
517 {R_028244_PA_SC_GENERIC_SCISSOR_BR, 0, 0, 0},
518 {R_028250_PA_SC_VPORT_SCISSOR_0_TL, 0, 0, 0},
519 {R_028254_PA_SC_VPORT_SCISSOR_0_BR, 0, 0, 0},
520 {R_0282D0_PA_SC_VPORT_ZMIN_0, 0, 0, 0},
521 {R_0282D4_PA_SC_VPORT_ZMAX_0, 0, 0, 0},
522 {R_02843C_PA_CL_VPORT_XSCALE_0, 0, 0, 0},
523 {R_028440_PA_CL_VPORT_XOFFSET_0, 0, 0, 0},
524 {R_028444_PA_CL_VPORT_YSCALE_0, 0, 0, 0},
525 {R_028448_PA_CL_VPORT_YOFFSET_0, 0, 0, 0},
526 {R_02844C_PA_CL_VPORT_ZSCALE_0, 0, 0, 0},
527 {R_028450_PA_CL_VPORT_ZOFFSET_0, 0, 0, 0},
528 {R_0286D4_SPI_INTERP_CONTROL_0, 0, 0, 0},
529 {R_028810_PA_CL_CLIP_CNTL, 0, 0, 0},
530 {R_028814_PA_SU_SC_MODE_CNTL, 0, 0, 0},
531 {R_028818_PA_CL_VTE_CNTL, 0, 0, 0},
532 {R_02881C_PA_CL_VS_OUT_CNTL, 0, 0, 0},
533 {R_028820_PA_CL_NANINF_CNTL, 0, 0, 0},
534 {R_028A00_PA_SU_POINT_SIZE, 0, 0, 0},
535 {R_028A04_PA_SU_POINT_MINMAX, 0, 0, 0},
536 {R_028A08_PA_SU_LINE_CNTL, 0, 0, 0},
537 {R_028A0C_PA_SC_LINE_STIPPLE, 0, 0, 0},
538 {R_028A48_PA_SC_MPASS_PS_CNTL, 0, 0, 0},
539 {R_028C00_PA_SC_LINE_CNTL, 0, 0, 0},
540 {R_028C08_PA_SU_VTX_CNTL, 0, 0, 0},
541 {R_028C0C_PA_CL_GB_VERT_CLIP_ADJ, 0, 0, 0},
542 {R_028C10_PA_CL_GB_VERT_DISC_ADJ, 0, 0, 0},
543 {R_028C14_PA_CL_GB_HORZ_CLIP_ADJ, 0, 0, 0},
544 {R_028C18_PA_CL_GB_HORZ_DISC_ADJ, 0, 0, 0},
545 {R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL, 0, 0, 0},
546 {R_028DFC_PA_SU_POLY_OFFSET_CLAMP, 0, 0, 0},
547 {R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 0, 0, 0},
548 {R_028E04_PA_SU_POLY_OFFSET_FRONT_OFFSET, 0, 0, 0},
549 {R_028E08_PA_SU_POLY_OFFSET_BACK_SCALE, 0, 0, 0},
550 {R_028E0C_PA_SU_POLY_OFFSET_BACK_OFFSET, 0, 0, 0},
551 {R_028E20_PA_CL_UCP0_X, 0, 0, 0},
552 {R_028E24_PA_CL_UCP0_Y, 0, 0, 0},
553 {R_028E28_PA_CL_UCP0_Z, 0, 0, 0},
554 {R_028E2C_PA_CL_UCP0_W, 0, 0, 0},
555 {R_028E30_PA_CL_UCP1_X, 0, 0, 0},
556 {R_028E34_PA_CL_UCP1_Y, 0, 0, 0},
557 {R_028E38_PA_CL_UCP1_Z, 0, 0, 0},
558 {R_028E3C_PA_CL_UCP1_W, 0, 0, 0},
559 {R_028E40_PA_CL_UCP2_X, 0, 0, 0},
560 {R_028E44_PA_CL_UCP2_Y, 0, 0, 0},
561 {R_028E48_PA_CL_UCP2_Z, 0, 0, 0},
562 {R_028E4C_PA_CL_UCP2_W, 0, 0, 0},
563 {R_028E50_PA_CL_UCP3_X, 0, 0, 0},
564 {R_028E54_PA_CL_UCP3_Y, 0, 0, 0},
565 {R_028E58_PA_CL_UCP3_Z, 0, 0, 0},
566 {R_028E5C_PA_CL_UCP3_W, 0, 0, 0},
567 {R_028E60_PA_CL_UCP4_X, 0, 0, 0},
568 {R_028E64_PA_CL_UCP4_Y, 0, 0, 0},
569 {R_028E68_PA_CL_UCP4_Z, 0, 0, 0},
570 {R_028E6C_PA_CL_UCP4_W, 0, 0, 0},
571 {R_028E70_PA_CL_UCP5_X, 0, 0, 0},
572 {R_028E74_PA_CL_UCP5_Y, 0, 0, 0},
573 {R_028E78_PA_CL_UCP5_Z, 0, 0, 0},
574 {R_028E7C_PA_CL_UCP5_W, 0, 0, 0},
575 {R_028380_SQ_VTX_SEMANTIC_0, 0, 0, 0},
576 {R_028384_SQ_VTX_SEMANTIC_1, 0, 0, 0},
577 {R_028388_SQ_VTX_SEMANTIC_2, 0, 0, 0},
578 {R_02838C_SQ_VTX_SEMANTIC_3, 0, 0, 0},
579 {R_028390_SQ_VTX_SEMANTIC_4, 0, 0, 0},
580 {R_028394_SQ_VTX_SEMANTIC_5, 0, 0, 0},
581 {R_028398_SQ_VTX_SEMANTIC_6, 0, 0, 0},
582 {R_02839C_SQ_VTX_SEMANTIC_7, 0, 0, 0},
583 {R_0283A0_SQ_VTX_SEMANTIC_8, 0, 0, 0},
584 {R_0283A4_SQ_VTX_SEMANTIC_9, 0, 0, 0},
585 {R_0283A8_SQ_VTX_SEMANTIC_10, 0, 0, 0},
586 {R_0283AC_SQ_VTX_SEMANTIC_11, 0, 0, 0},
587 {R_0283B0_SQ_VTX_SEMANTIC_12, 0, 0, 0},
588 {R_0283B4_SQ_VTX_SEMANTIC_13, 0, 0, 0},
589 {R_0283B8_SQ_VTX_SEMANTIC_14, 0, 0, 0},
590 {R_0283BC_SQ_VTX_SEMANTIC_15, 0, 0, 0},
591 {R_0283C0_SQ_VTX_SEMANTIC_16, 0, 0, 0},
592 {R_0283C4_SQ_VTX_SEMANTIC_17, 0, 0, 0},
593 {R_0283C8_SQ_VTX_SEMANTIC_18, 0, 0, 0},
594 {R_0283CC_SQ_VTX_SEMANTIC_19, 0, 0, 0},
595 {R_0283D0_SQ_VTX_SEMANTIC_20, 0, 0, 0},
596 {R_0283D4_SQ_VTX_SEMANTIC_21, 0, 0, 0},
597 {R_0283D8_SQ_VTX_SEMANTIC_22, 0, 0, 0},
598 {R_0283DC_SQ_VTX_SEMANTIC_23, 0, 0, 0},
599 {R_0283E0_SQ_VTX_SEMANTIC_24, 0, 0, 0},
600 {R_0283E4_SQ_VTX_SEMANTIC_25, 0, 0, 0},
601 {R_0283E8_SQ_VTX_SEMANTIC_26, 0, 0, 0},
602 {R_0283EC_SQ_VTX_SEMANTIC_27, 0, 0, 0},
603 {R_0283F0_SQ_VTX_SEMANTIC_28, 0, 0, 0},
604 {R_0283F4_SQ_VTX_SEMANTIC_29, 0, 0, 0},
605 {R_0283F8_SQ_VTX_SEMANTIC_30, 0, 0, 0},
606 {R_0283FC_SQ_VTX_SEMANTIC_31, 0, 0, 0},
607 {R_028614_SPI_VS_OUT_ID_0, 0, 0, 0},
608 {R_028618_SPI_VS_OUT_ID_1, 0, 0, 0},
609 {R_02861C_SPI_VS_OUT_ID_2, 0, 0, 0},
610 {R_028620_SPI_VS_OUT_ID_3, 0, 0, 0},
611 {R_028624_SPI_VS_OUT_ID_4, 0, 0, 0},
612 {R_028628_SPI_VS_OUT_ID_5, 0, 0, 0},
613 {R_02862C_SPI_VS_OUT_ID_6, 0, 0, 0},
614 {R_028630_SPI_VS_OUT_ID_7, 0, 0, 0},
615 {R_028634_SPI_VS_OUT_ID_8, 0, 0, 0},
616 {R_028638_SPI_VS_OUT_ID_9, 0, 0, 0},
617 {R_0286C4_SPI_VS_OUT_CONFIG, 0, 0, 0},
618 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
619 {R_028858_SQ_PGM_START_VS, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF},
620 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
621 {R_028868_SQ_PGM_RESOURCES_VS, 0, 0, 0},
622 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
623 {R_028894_SQ_PGM_START_FS, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF},
624 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
625 {R_0288A4_SQ_PGM_RESOURCES_FS, 0, 0, 0},
626 {R_0288D0_SQ_PGM_CF_OFFSET_VS, 0, 0, 0},
627 {R_0288DC_SQ_PGM_CF_OFFSET_FS, 0, 0, 0},
628 {R_028644_SPI_PS_INPUT_CNTL_0, 0, 0, 0},
629 {R_028648_SPI_PS_INPUT_CNTL_1, 0, 0, 0},
630 {R_02864C_SPI_PS_INPUT_CNTL_2, 0, 0, 0},
631 {R_028650_SPI_PS_INPUT_CNTL_3, 0, 0, 0},
632 {R_028654_SPI_PS_INPUT_CNTL_4, 0, 0, 0},
633 {R_028658_SPI_PS_INPUT_CNTL_5, 0, 0, 0},
634 {R_02865C_SPI_PS_INPUT_CNTL_6, 0, 0, 0},
635 {R_028660_SPI_PS_INPUT_CNTL_7, 0, 0, 0},
636 {R_028664_SPI_PS_INPUT_CNTL_8, 0, 0, 0},
637 {R_028668_SPI_PS_INPUT_CNTL_9, 0, 0, 0},
638 {R_02866C_SPI_PS_INPUT_CNTL_10, 0, 0, 0},
639 {R_028670_SPI_PS_INPUT_CNTL_11, 0, 0, 0},
640 {R_028674_SPI_PS_INPUT_CNTL_12, 0, 0, 0},
641 {R_028678_SPI_PS_INPUT_CNTL_13, 0, 0, 0},
642 {R_02867C_SPI_PS_INPUT_CNTL_14, 0, 0, 0},
643 {R_028680_SPI_PS_INPUT_CNTL_15, 0, 0, 0},
644 {R_028684_SPI_PS_INPUT_CNTL_16, 0, 0, 0},
645 {R_028688_SPI_PS_INPUT_CNTL_17, 0, 0, 0},
646 {R_02868C_SPI_PS_INPUT_CNTL_18, 0, 0, 0},
647 {R_028690_SPI_PS_INPUT_CNTL_19, 0, 0, 0},
648 {R_028694_SPI_PS_INPUT_CNTL_20, 0, 0, 0},
649 {R_028698_SPI_PS_INPUT_CNTL_21, 0, 0, 0},
650 {R_02869C_SPI_PS_INPUT_CNTL_22, 0, 0, 0},
651 {R_0286A0_SPI_PS_INPUT_CNTL_23, 0, 0, 0},
652 {R_0286A4_SPI_PS_INPUT_CNTL_24, 0, 0, 0},
653 {R_0286A8_SPI_PS_INPUT_CNTL_25, 0, 0, 0},
654 {R_0286AC_SPI_PS_INPUT_CNTL_26, 0, 0, 0},
655 {R_0286B0_SPI_PS_INPUT_CNTL_27, 0, 0, 0},
656 {R_0286B4_SPI_PS_INPUT_CNTL_28, 0, 0, 0},
657 {R_0286B8_SPI_PS_INPUT_CNTL_29, 0, 0, 0},
658 {R_0286BC_SPI_PS_INPUT_CNTL_30, 0, 0, 0},
659 {R_0286C0_SPI_PS_INPUT_CNTL_31, 0, 0, 0},
660 {R_0286CC_SPI_PS_IN_CONTROL_0, 0, 0, 0},
661 {R_0286D0_SPI_PS_IN_CONTROL_1, 0, 0, 0},
662 {R_0286D8_SPI_INPUT_Z, 0, 0, 0},
663 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
664 {R_028840_SQ_PGM_START_PS, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF},
665 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0},
666 {R_028850_SQ_PGM_RESOURCES_PS, 0, 0, 0},
667 {R_028854_SQ_PGM_EXPORTS_PS, 0, 0, 0},
668 {R_0288CC_SQ_PGM_CF_OFFSET_PS, 0, 0, 0},
669 {R_028400_VGT_MAX_VTX_INDX, 0, 0, 0},
670 {R_028404_VGT_MIN_VTX_INDX, 0, 0, 0},
671 {R_028408_VGT_INDX_OFFSET, 0, 0, 0},
672 {R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, 0, 0, 0},
673 {R_028A84_VGT_PRIMITIVEID_EN, 0, 0, 0},
674 {R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, 0, 0, 0},
675 {R_028AA0_VGT_INSTANCE_STEP_RATE_0, 0, 0, 0},
676 {R_028AA4_VGT_INSTANCE_STEP_RATE_1, 0, 0, 0},
679 /* SHADER RESOURCE R600/R700 */
680 int r600_resource_init(struct r600_context *ctx, struct r600_range *range, unsigned offset, unsigned nblocks, unsigned stride, struct r600_reg *reg, int nreg, unsigned offset_base)
683 struct r600_block *block;
684 range->blocks = calloc(nblocks, sizeof(struct r600_block *));
685 if (range->blocks == NULL)
688 reg[0].offset += offset;
689 for (i = 0; i < nblocks; i++) {
690 block = calloc(1, sizeof(struct r600_block));
695 range->blocks[i] = block;
696 r600_init_block(ctx, block, reg, 0, nreg, PKT3_SET_RESOURCE, offset_base);
698 reg[0].offset += stride;
704 static int r600_resource_range_init(struct r600_context *ctx, struct r600_range *range, unsigned offset, unsigned nblocks, unsigned stride)
706 struct r600_reg r600_shader_resource[] = {
707 {R_038000_RESOURCE0_WORD0, REG_FLAG_NEED_BO, S_0085F0_TC_ACTION_ENA(1) | S_0085F0_VC_ACTION_ENA(1), 0xFFFFFFFF},
708 {R_038004_RESOURCE0_WORD1, REG_FLAG_NEED_BO, S_0085F0_TC_ACTION_ENA(1) | S_0085F0_VC_ACTION_ENA(1), 0xFFFFFFFF},
709 {R_038008_RESOURCE0_WORD2, 0, 0, 0},
710 {R_03800C_RESOURCE0_WORD3, 0, 0, 0},
711 {R_038010_RESOURCE0_WORD4, 0, 0, 0},
712 {R_038014_RESOURCE0_WORD5, 0, 0, 0},
713 {R_038018_RESOURCE0_WORD6, 0, 0, 0},
715 unsigned nreg = Elements(r600_shader_resource);
717 return r600_resource_init(ctx, range, offset, nblocks, stride, r600_shader_resource, nreg, R600_RESOURCE_OFFSET);
720 /* SHADER SAMPLER R600/R700 */
721 static int r600_state_sampler_init(struct r600_context *ctx, u32 offset)
723 struct r600_reg r600_shader_sampler[] = {
724 {R_03C000_SQ_TEX_SAMPLER_WORD0_0, 0, 0, 0},
725 {R_03C004_SQ_TEX_SAMPLER_WORD1_0, 0, 0, 0},
726 {R_03C008_SQ_TEX_SAMPLER_WORD2_0, 0, 0, 0},
728 unsigned nreg = Elements(r600_shader_sampler);
730 for (int i = 0; i < nreg; i++) {
731 r600_shader_sampler[i].offset += offset;
733 return r600_context_add_block(ctx, r600_shader_sampler, nreg, PKT3_SET_SAMPLER, R600_SAMPLER_OFFSET);
736 /* SHADER SAMPLER BORDER R600/R700 */
737 static int r600_state_sampler_border_init(struct r600_context *ctx, u32 offset)
739 struct r600_reg r600_shader_sampler_border[] = {
740 {R_00A400_TD_PS_SAMPLER0_BORDER_RED, 0, 0, 0},
741 {R_00A404_TD_PS_SAMPLER0_BORDER_GREEN, 0, 0, 0},
742 {R_00A408_TD_PS_SAMPLER0_BORDER_BLUE, 0, 0, 0},
743 {R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA, 0, 0, 0},
745 unsigned nreg = Elements(r600_shader_sampler_border);
747 for (int i = 0; i < nreg; i++) {
748 r600_shader_sampler_border[i].offset += offset;
750 return r600_context_add_block(ctx, r600_shader_sampler_border, nreg, PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
753 static int r600_loop_const_init(struct r600_context *ctx, u32 offset)
756 struct r600_reg r600_loop_consts[32];
759 for (i = 0; i < nreg; i++) {
760 r600_loop_consts[i].offset = R600_LOOP_CONST_OFFSET + ((offset + i) * 4);
761 r600_loop_consts[i].flags = REG_FLAG_DIRTY_ALWAYS;
762 r600_loop_consts[i].flush_flags = 0;
763 r600_loop_consts[i].flush_mask = 0;
765 return r600_context_add_block(ctx, r600_loop_consts, nreg, PKT3_SET_LOOP_CONST, R600_LOOP_CONST_OFFSET);
768 static void r600_context_clear_fenced_bo(struct r600_context *ctx)
770 struct radeon_bo *bo, *tmp;
772 LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &ctx->fenced_bo, fencedlist) {
773 LIST_DELINIT(&bo->fencedlist);
779 static void r600_free_resource_range(struct r600_context *ctx, struct r600_range *range, int nblocks)
781 struct r600_block *block;
783 for (i = 0; i < nblocks; i++) {
784 block = range->blocks[i];
786 for (int k = 1; k <= block->nbo; k++)
787 r600_bo_reference(ctx->radeon, &block->reloc[k].bo, NULL);
796 void r600_context_fini(struct r600_context *ctx)
798 struct r600_block *block;
799 struct r600_range *range;
801 for (int i = 0; i < NUM_RANGES; i++) {
802 if (!ctx->range[i].blocks)
804 for (int j = 0; j < (1 << HASH_SHIFT); j++) {
805 block = ctx->range[i].blocks[j];
807 for (int k = 0, offset = block->start_offset; k < block->nreg; k++, offset += 4) {
808 range = &ctx->range[CTX_RANGE_ID(offset)];
809 range->blocks[CTX_BLOCK_ID(offset)] = NULL;
811 for (int k = 1; k <= block->nbo; k++) {
812 r600_bo_reference(ctx->radeon, &block->reloc[k].bo, NULL);
817 free(ctx->range[i].blocks);
819 r600_free_resource_range(ctx, &ctx->ps_resources, ctx->num_ps_resources);
820 r600_free_resource_range(ctx, &ctx->vs_resources, ctx->num_vs_resources);
821 r600_free_resource_range(ctx, &ctx->fs_resources, ctx->num_fs_resources);
828 r600_context_clear_fenced_bo(ctx);
829 memset(ctx, 0, sizeof(struct r600_context));
832 static void r600_add_resource_block(struct r600_context *ctx, struct r600_range *range, int num_blocks, int *index)
835 for (int j = 0; j < num_blocks; j++) {
836 if (!range->blocks[j])
839 ctx->blocks[c++] = range->blocks[j];
844 int r600_setup_block_table(struct r600_context *ctx)
846 /* setup block table */
848 ctx->blocks = calloc(ctx->nblocks, sizeof(void*));
851 for (int i = 0; i < NUM_RANGES; i++) {
852 if (!ctx->range[i].blocks)
854 for (int j = 0, add; j < (1 << HASH_SHIFT); j++) {
855 if (!ctx->range[i].blocks[j])
859 for (int k = 0; k < c; k++) {
860 if (ctx->blocks[k] == ctx->range[i].blocks[j]) {
866 assert(c < ctx->nblocks);
867 ctx->blocks[c++] = ctx->range[i].blocks[j];
868 j += (ctx->range[i].blocks[j]->nreg) - 1;
873 r600_add_resource_block(ctx, &ctx->ps_resources, ctx->num_ps_resources, &c);
874 r600_add_resource_block(ctx, &ctx->vs_resources, ctx->num_vs_resources, &c);
875 r600_add_resource_block(ctx, &ctx->fs_resources, ctx->num_fs_resources, &c);
879 int r600_context_init(struct r600_context *ctx, struct radeon *radeon)
883 memset(ctx, 0, sizeof(struct r600_context));
884 ctx->radeon = radeon;
885 LIST_INITHEAD(&ctx->query_list);
887 /* init dirty list */
888 LIST_INITHEAD(&ctx->dirty);
889 LIST_INITHEAD(&ctx->resource_dirty);
890 LIST_INITHEAD(&ctx->enable_list);
892 ctx->range = calloc(NUM_RANGES, sizeof(struct r600_range));
899 r = r600_context_add_block(ctx, r600_config_reg_list,
900 Elements(r600_config_reg_list), PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
903 r = r600_context_add_block(ctx, r600_context_reg_list,
904 Elements(r600_context_reg_list), PKT3_SET_CONTEXT_REG, R600_CONTEXT_REG_OFFSET);
907 r = r600_context_add_block(ctx, r600_ctl_const_list,
908 Elements(r600_ctl_const_list), PKT3_SET_CTL_CONST, R600_CTL_CONST_OFFSET);
912 /* PS SAMPLER BORDER */
913 for (int j = 0, offset = 0; j < 18; j++, offset += 0x10) {
914 r = r600_state_sampler_border_init(ctx, offset);
919 /* VS SAMPLER BORDER */
920 for (int j = 0, offset = 0x200; j < 18; j++, offset += 0x10) {
921 r = r600_state_sampler_border_init(ctx, offset);
926 for (int j = 0, offset = 0; j < 18; j++, offset += 0xC) {
927 r = r600_state_sampler_init(ctx, offset);
932 for (int j = 0, offset = 0xD8; j < 18; j++, offset += 0xC) {
933 r = r600_state_sampler_init(ctx, offset);
938 ctx->num_ps_resources = 160;
939 ctx->num_vs_resources = 160;
940 ctx->num_fs_resources = 16;
941 r = r600_resource_range_init(ctx, &ctx->ps_resources, 0, 160, 0x1c);
944 r = r600_resource_range_init(ctx, &ctx->vs_resources, 0x1180, 160, 0x1c);
947 r = r600_resource_range_init(ctx, &ctx->fs_resources, 0x2300, 16, 0x1c);
952 r600_loop_const_init(ctx, 0);
954 r600_loop_const_init(ctx, 32);
956 r = r600_setup_block_table(ctx);
960 /* allocate cs variables */
961 ctx->nreloc = RADEON_CTX_MAX_PM4;
962 ctx->reloc = calloc(ctx->nreloc, sizeof(struct r600_reloc));
963 if (ctx->reloc == NULL) {
967 ctx->bo = calloc(ctx->nreloc, sizeof(void *));
968 if (ctx->bo == NULL) {
972 ctx->pm4_ndwords = RADEON_CTX_MAX_PM4;
973 ctx->pm4 = calloc(ctx->pm4_ndwords, 4);
974 if (ctx->pm4 == NULL) {
980 /* save 16dwords space for fence mecanism */
981 ctx->pm4_ndwords -= 16;
983 LIST_INITHEAD(&ctx->fenced_bo);
987 r600_get_backend_mask(ctx);
991 r600_context_fini(ctx);
995 /* Flushes all surfaces */
996 void r600_context_flush_all(struct r600_context *ctx, unsigned flush_flags)
998 unsigned ndwords = 5;
1000 if ((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) {
1002 r600_context_flush(ctx);
1005 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, ctx->predicate_drawing);
1006 ctx->pm4[ctx->pm4_cdwords++] = flush_flags; /* CP_COHER_CNTL */
1007 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* CP_COHER_SIZE */
1008 ctx->pm4[ctx->pm4_cdwords++] = 0; /* CP_COHER_BASE */
1009 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; /* POLL_INTERVAL */
1012 void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags,
1013 unsigned flush_mask, struct r600_bo *rbo)
1015 struct radeon_bo *bo;
1018 /* if bo has already been flushed */
1019 if (!(~bo->last_flush & flush_flags)) {
1020 bo->last_flush &= flush_mask;
1024 if ((ctx->radeon->family < CHIP_RV770) &&
1025 (G_0085F0_CB_ACTION_ENA(flush_flags) ||
1026 G_0085F0_DB_ACTION_ENA(flush_flags))) {
1027 if (ctx->flags & R600_CONTEXT_CHECK_EVENT_FLUSH) {
1028 /* the rv670 seems to fail fbo-generatemipmap unless we flush the CB1 dest base ena */
1029 if ((bo->binding & BO_BOUND_TEXTURE) &&
1030 (flush_flags & S_0085F0_CB_ACTION_ENA(1))) {
1031 if ((ctx->radeon->family == CHIP_RV670) ||
1032 (ctx->radeon->family == CHIP_RS780) ||
1033 (ctx->radeon->family == CHIP_RS880)) {
1034 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, ctx->predicate_drawing);
1035 ctx->pm4[ctx->pm4_cdwords++] = S_0085F0_CB1_DEST_BASE_ENA(1); /* CP_COHER_CNTL */
1036 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* CP_COHER_SIZE */
1037 ctx->pm4[ctx->pm4_cdwords++] = 0; /* CP_COHER_BASE */
1038 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; /* POLL_INTERVAL */
1042 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, ctx->predicate_drawing);
1043 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
1044 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH;
1047 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, ctx->predicate_drawing);
1048 ctx->pm4[ctx->pm4_cdwords++] = flush_flags;
1049 ctx->pm4[ctx->pm4_cdwords++] = (bo->size + 255) >> 8;
1050 ctx->pm4[ctx->pm4_cdwords++] = 0x00000000;
1051 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A;
1052 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing);
1053 ctx->pm4[ctx->pm4_cdwords++] = bo->reloc_id;
1055 bo->last_flush = (bo->last_flush | flush_flags) & flush_mask;
1058 void r600_context_get_reloc(struct r600_context *ctx, struct r600_bo *rbo)
1060 struct radeon_bo *bo = rbo->bo;
1061 bo->reloc = &ctx->reloc[ctx->creloc];
1062 bo->reloc_id = ctx->creloc * sizeof(struct r600_reloc) / 4;
1063 ctx->reloc[ctx->creloc].handle = bo->handle;
1064 ctx->reloc[ctx->creloc].read_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
1065 ctx->reloc[ctx->creloc].write_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
1066 ctx->reloc[ctx->creloc].flags = 0;
1067 radeon_bo_reference(ctx->radeon, &ctx->bo[ctx->creloc], bo);
1068 rbo->fence = ctx->radeon->fence;
1072 void r600_context_reg(struct r600_context *ctx,
1073 unsigned offset, unsigned value,
1076 struct r600_range *range;
1077 struct r600_block *block;
1082 range = &ctx->range[CTX_RANGE_ID(offset)];
1083 block = range->blocks[CTX_BLOCK_ID(offset)];
1084 id = (offset - block->start_offset) >> 2;
1086 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
1088 new_val = block->reg[id];
1091 if (new_val != block->reg[id]) {
1092 dirty |= R600_BLOCK_STATUS_DIRTY;
1093 block->reg[id] = new_val;
1096 r600_context_dirty_block(ctx, block, dirty, id);
1099 void r600_context_dirty_block(struct r600_context *ctx,
1100 struct r600_block *block,
1101 int dirty, int index)
1103 if ((index + 1) > block->nreg_dirty)
1104 block->nreg_dirty = index + 1;
1106 if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
1107 block->status |= R600_BLOCK_STATUS_DIRTY;
1108 ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords;
1109 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
1110 block->status |= R600_BLOCK_STATUS_ENABLED;
1111 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
1113 LIST_ADDTAIL(&block->list,&ctx->dirty);
1115 if (block->flags & REG_FLAG_FLUSH_CHANGE) {
1116 r600_context_ps_partial_flush(ctx);
1121 void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state)
1123 struct r600_block *block;
1126 for (int i = 0; i < state->nregs; i++) {
1127 unsigned id, reloc_id;
1128 struct r600_pipe_reg *reg = &state->regs[i];
1133 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
1135 new_val = block->reg[id];
1136 new_val &= ~reg->mask;
1137 new_val |= reg->value;
1138 if (new_val != block->reg[id]) {
1139 block->reg[id] = new_val;
1140 dirty |= R600_BLOCK_STATUS_DIRTY;
1142 if (block->flags & REG_FLAG_DIRTY_ALWAYS)
1143 dirty |= R600_BLOCK_STATUS_DIRTY;
1144 if (block->pm4_bo_index[id]) {
1145 /* find relocation */
1146 reloc_id = block->pm4_bo_index[id];
1147 r600_bo_reference(ctx->radeon, &block->reloc[reloc_id].bo, reg->bo);
1148 reg->bo->fence = ctx->radeon->fence;
1149 /* always force dirty for relocs for now */
1150 dirty |= R600_BLOCK_STATUS_DIRTY;
1154 r600_context_dirty_block(ctx, block, dirty, id);
1158 static void r600_context_dirty_resource_block(struct r600_context *ctx,
1159 struct r600_block *block,
1160 int dirty, int index)
1162 block->nreg_dirty = index + 1;
1164 if ((dirty != (block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
1165 block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1166 ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords;
1167 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
1168 block->status |= R600_BLOCK_STATUS_ENABLED;
1169 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
1171 LIST_ADDTAIL(&block->list,&ctx->resource_dirty);
1175 void r600_context_pipe_state_set_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, struct r600_block *block)
1178 int num_regs = ctx->radeon->chip_class >= EVERGREEN ? 8 : 7;
1181 if (state == NULL) {
1182 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_RESOURCE_DIRTY);
1183 if (block->reloc[1].bo)
1184 block->reloc[1].bo->bo->binding &= ~BO_BOUND_TEXTURE;
1186 r600_bo_reference(ctx->radeon, &block->reloc[1].bo, NULL);
1187 r600_bo_reference(ctx->radeon, &block->reloc[2].bo, NULL);
1188 LIST_DELINIT(&block->list);
1189 LIST_DELINIT(&block->enable_list);
1193 is_vertex = ((state->val[num_regs-1] & 0xc0000000) == 0xc0000000);
1194 dirty = block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY;
1196 if (memcmp(block->reg, state->val, num_regs*4)) {
1197 memcpy(block->reg, state->val, num_regs * 4);
1198 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1201 /* if no BOs on block, force dirty */
1202 if (!block->reloc[1].bo || !block->reloc[2].bo)
1203 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1207 if (block->reloc[1].bo->bo->handle != state->bo[0]->bo->handle)
1208 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1210 if ((block->reloc[1].bo->bo->handle != state->bo[0]->bo->handle) ||
1211 (block->reloc[2].bo->bo->handle != state->bo[1]->bo->handle))
1212 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1217 state->bo[0]->fence = ctx->radeon->fence;
1219 state->bo[0]->fence = ctx->radeon->fence;
1220 state->bo[1]->fence = ctx->radeon->fence;
1224 /* VERTEX RESOURCE, we preted there is 2 bo to relocate so
1225 * we have single case btw VERTEX & TEXTURE resource
1227 r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->bo[0]);
1228 r600_bo_reference(ctx->radeon, &block->reloc[2].bo, NULL);
1229 state->bo[0]->fence = ctx->radeon->fence;
1231 /* TEXTURE RESOURCE */
1232 r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->bo[0]);
1233 r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->bo[1]);
1234 state->bo[0]->fence = ctx->radeon->fence;
1235 state->bo[1]->fence = ctx->radeon->fence;
1236 state->bo[0]->bo->binding |= BO_BOUND_TEXTURE;
1241 block->status |= R600_BLOCK_STATUS_RESOURCE_VERTEX;
1243 block->status &= ~R600_BLOCK_STATUS_RESOURCE_VERTEX;
1245 r600_context_dirty_resource_block(ctx, block, dirty, num_regs - 1);
1249 void r600_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid)
1251 struct r600_block *block = ctx->ps_resources.blocks[rid];
1253 r600_context_pipe_state_set_resource(ctx, state, block);
1256 void r600_context_pipe_state_set_vs_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid)
1258 struct r600_block *block = ctx->vs_resources.blocks[rid];
1260 r600_context_pipe_state_set_resource(ctx, state, block);
1263 void r600_context_pipe_state_set_fs_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid)
1265 struct r600_block *block = ctx->fs_resources.blocks[rid];
1267 r600_context_pipe_state_set_resource(ctx, state, block);
1270 static inline void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
1272 struct r600_range *range;
1273 struct r600_block *block;
1277 range = &ctx->range[CTX_RANGE_ID(offset)];
1278 block = range->blocks[CTX_BLOCK_ID(offset)];
1279 if (state == NULL) {
1280 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
1281 LIST_DELINIT(&block->list);
1282 LIST_DELINIT(&block->enable_list);
1285 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
1286 for (i = 0; i < 3; i++) {
1287 if (block->reg[i] != state->regs[i].value) {
1288 block->reg[i] = state->regs[i].value;
1289 dirty |= R600_BLOCK_STATUS_DIRTY;
1294 r600_context_dirty_block(ctx, block, dirty, 2);
1298 static inline void r600_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
1300 struct r600_range *range;
1301 struct r600_block *block;
1305 range = &ctx->range[CTX_RANGE_ID(offset)];
1306 block = range->blocks[CTX_BLOCK_ID(offset)];
1307 if (state == NULL) {
1308 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
1309 LIST_DELINIT(&block->list);
1310 LIST_DELINIT(&block->enable_list);
1313 if (state->nregs <= 3) {
1316 dirty = block->status & R600_BLOCK_STATUS_DIRTY;
1317 for (i = 0; i < 4; i++) {
1318 if (block->reg[i] != state->regs[i + 3].value) {
1319 block->reg[i] = state->regs[i + 3].value;
1320 dirty |= R600_BLOCK_STATUS_DIRTY;
1324 /* We have to flush the shaders before we change the border color
1325 * registers, or previous draw commands that haven't completed yet
1326 * will end up using the new border color. */
1327 if (dirty & R600_BLOCK_STATUS_DIRTY)
1328 r600_context_ps_partial_flush(ctx);
1330 r600_context_dirty_block(ctx, block, dirty, 3);
1333 void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
1337 offset = 0x0003C000 + id * 0xc;
1338 r600_context_pipe_state_set_sampler(ctx, state, offset);
1339 offset = 0x0000A400 + id * 0x10;
1340 r600_context_pipe_state_set_sampler_border(ctx, state, offset);
1343 void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
1347 offset = 0x0003C0D8 + id * 0xc;
1348 r600_context_pipe_state_set_sampler(ctx, state, offset);
1349 offset = 0x0000A600 + id * 0x10;
1350 r600_context_pipe_state_set_sampler_border(ctx, state, offset);
1353 struct r600_bo *r600_context_reg_bo(struct r600_context *ctx, unsigned offset)
1355 struct r600_range *range;
1356 struct r600_block *block;
1359 range = &ctx->range[CTX_RANGE_ID(offset)];
1360 block = range->blocks[CTX_BLOCK_ID(offset)];
1361 offset -= block->start_offset;
1362 id = block->pm4_bo_index[offset >> 2];
1363 if (block->reloc[id].bo) {
1364 return block->reloc[id].bo;
1369 void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block)
1372 int optional = block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS);
1373 int cp_dwords = block->pm4_ndwords, start_dword = 0;
1375 int nbo = block->nbo;
1377 if (block->nreg_dirty == 0 && optional) {
1382 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH;
1384 for (int j = 0; j < block->nreg; j++) {
1385 if (block->pm4_bo_index[j]) {
1386 /* find relocation */
1387 id = block->pm4_bo_index[j];
1388 r600_context_bo_reloc(ctx,
1389 &block->pm4[block->reloc[id].bo_pm4_index],
1390 block->reloc[id].bo);
1391 r600_context_bo_flush(ctx,
1392 block->reloc[id].flush_flags,
1393 block->reloc[id].flush_mask,
1394 block->reloc[id].bo);
1400 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH;
1403 optional &= (block->nreg_dirty != block->nreg);
1405 new_dwords = block->nreg_dirty;
1406 start_dword = ctx->pm4_cdwords;
1407 cp_dwords = new_dwords + 2;
1409 memcpy(&ctx->pm4[ctx->pm4_cdwords], block->pm4, cp_dwords * 4);
1410 ctx->pm4_cdwords += cp_dwords;
1415 newword = ctx->pm4[start_dword];
1416 newword &= PKT_COUNT_C;
1417 newword |= PKT_COUNT_S(new_dwords);
1418 ctx->pm4[start_dword] = newword;
1421 block->status ^= R600_BLOCK_STATUS_DIRTY;
1422 block->nreg_dirty = 0;
1423 LIST_DELINIT(&block->list);
1426 void r600_context_block_resource_emit_dirty(struct r600_context *ctx, struct r600_block *block)
1429 int cp_dwords = block->pm4_ndwords;
1430 int nbo = block->nbo;
1432 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH;
1434 if (block->status & R600_BLOCK_STATUS_RESOURCE_VERTEX) {
1436 cp_dwords -= 2; /* don't copy the second NOP */
1439 for (int j = 0; j < nbo; j++) {
1440 if (block->pm4_bo_index[j]) {
1441 /* find relocation */
1442 id = block->pm4_bo_index[j];
1443 r600_context_bo_reloc(ctx,
1444 &block->pm4[block->reloc[id].bo_pm4_index],
1445 block->reloc[id].bo);
1446 r600_context_bo_flush(ctx,
1447 block->reloc[id].flush_flags,
1448 block->reloc[id].flush_mask,
1449 block->reloc[id].bo);
1452 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH;
1454 memcpy(&ctx->pm4[ctx->pm4_cdwords], block->pm4, cp_dwords * 4);
1455 ctx->pm4_cdwords += cp_dwords;
1457 block->status ^= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1458 block->nreg_dirty = 0;
1459 LIST_DELINIT(&block->list);
1462 void r600_context_flush_dest_caches(struct r600_context *ctx)
1464 struct r600_bo *cb[8];
1468 if (!(ctx->flags & R600_CONTEXT_DST_CACHES_DIRTY))
1471 db = r600_context_reg_bo(ctx, R_02800C_DB_DEPTH_BASE);
1472 cb[0] = r600_context_reg_bo(ctx, R_028040_CB_COLOR0_BASE);
1473 cb[1] = r600_context_reg_bo(ctx, R_028044_CB_COLOR1_BASE);
1474 cb[2] = r600_context_reg_bo(ctx, R_028048_CB_COLOR2_BASE);
1475 cb[3] = r600_context_reg_bo(ctx, R_02804C_CB_COLOR3_BASE);
1476 cb[4] = r600_context_reg_bo(ctx, R_028050_CB_COLOR4_BASE);
1477 cb[5] = r600_context_reg_bo(ctx, R_028054_CB_COLOR5_BASE);
1478 cb[6] = r600_context_reg_bo(ctx, R_028058_CB_COLOR6_BASE);
1479 cb[7] = r600_context_reg_bo(ctx, R_02805C_CB_COLOR7_BASE);
1481 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH;
1482 /* flush the color buffers */
1483 for (i = 0; i < 8; i++) {
1487 r600_context_bo_flush(ctx,
1488 (S_0085F0_CB0_DEST_BASE_ENA(1) << i) |
1489 S_0085F0_CB_ACTION_ENA(1),
1493 r600_context_bo_flush(ctx, S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1), 0, db);
1495 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH;
1496 ctx->flags &= ~R600_CONTEXT_DST_CACHES_DIRTY;
1499 void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
1501 unsigned ndwords = 7;
1502 struct r600_block *dirty_block = NULL;
1503 struct r600_block *next_block;
1506 if (draw->indices) {
1508 /* make sure there is enough relocation space before scheduling draw */
1509 if (ctx->creloc >= (ctx->nreloc - 1)) {
1510 r600_context_flush(ctx);
1514 /* queries need some special values */
1515 if (ctx->num_query_running) {
1516 if (ctx->radeon->family >= CHIP_RV770) {
1517 r600_context_reg(ctx,
1518 R_028D0C_DB_RENDER_CONTROL,
1519 S_028D0C_R700_PERFECT_ZPASS_COUNTS(1),
1520 S_028D0C_R700_PERFECT_ZPASS_COUNTS(1));
1522 r600_context_reg(ctx,
1523 R_028D10_DB_RENDER_OVERRIDE,
1524 S_028D10_NOOP_CULL_DISABLE(1),
1525 S_028D10_NOOP_CULL_DISABLE(1));
1528 /* update the max dword count to make sure we have enough space
1529 * reserved for flushing the destination caches */
1530 ctx->pm4_ndwords = RADEON_CTX_MAX_PM4 - ctx->num_dest_buffers * 7 - 16;
1532 if ((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) {
1534 r600_context_flush(ctx);
1536 /* at that point everythings is flushed and ctx->pm4_cdwords = 0 */
1537 if ((ctx->pm4_dirty_cdwords + ndwords) > ctx->pm4_ndwords) {
1538 R600_ERR("context is too big to be scheduled\n");
1541 /* enough room to copy packet */
1542 LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->dirty, list) {
1543 r600_context_block_emit_dirty(ctx, dirty_block);
1546 LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->resource_dirty, list) {
1547 r600_context_block_resource_emit_dirty(ctx, dirty_block);
1551 pm4 = &ctx->pm4[ctx->pm4_cdwords];
1553 pm4[0] = PKT3(PKT3_INDEX_TYPE, 0, ctx->predicate_drawing);
1554 pm4[1] = draw->vgt_index_type;
1555 pm4[2] = PKT3(PKT3_NUM_INSTANCES, 0, ctx->predicate_drawing);
1556 pm4[3] = draw->vgt_num_instances;
1557 if (draw->indices) {
1558 pm4[4] = PKT3(PKT3_DRAW_INDEX, 3, ctx->predicate_drawing);
1559 pm4[5] = draw->indices_bo_offset + r600_bo_offset(draw->indices);
1561 pm4[7] = draw->vgt_num_indices;
1562 pm4[8] = draw->vgt_draw_initiator;
1563 pm4[9] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing);
1565 r600_context_bo_reloc(ctx, &pm4[10], draw->indices);
1567 pm4[4] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, ctx->predicate_drawing);
1568 pm4[5] = draw->vgt_num_indices;
1569 pm4[6] = draw->vgt_draw_initiator;
1571 ctx->pm4_cdwords += ndwords;
1573 ctx->flags |= (R600_CONTEXT_DST_CACHES_DIRTY | R600_CONTEXT_DRAW_PENDING);
1575 /* all dirty state have been scheduled in current cs */
1576 ctx->pm4_dirty_cdwords = 0;
1579 void r600_context_flush(struct r600_context *ctx)
1581 struct drm_radeon_cs drmib = {};
1582 struct drm_radeon_cs_chunk chunks[2];
1583 uint64_t chunk_array[2];
1586 struct r600_block *enable_block = NULL;
1588 if (ctx->pm4_cdwords == ctx->init_dwords)
1591 /* suspend queries */
1592 r600_context_queries_suspend(ctx);
1594 if (ctx->radeon->family >= CHIP_CEDAR)
1595 evergreen_context_flush_dest_caches(ctx);
1597 r600_context_flush_dest_caches(ctx);
1599 /* partial flush is needed to avoid lockups on some chips with user fences */
1600 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
1601 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
1603 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1604 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1605 ctx->pm4[ctx->pm4_cdwords++] = 0;
1606 ctx->pm4[ctx->pm4_cdwords++] = (1 << 29) | (0 << 24);
1607 ctx->pm4[ctx->pm4_cdwords++] = ctx->radeon->fence;
1608 ctx->pm4[ctx->pm4_cdwords++] = 0;
1609 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
1610 ctx->pm4[ctx->pm4_cdwords++] = 0;
1611 r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], ctx->radeon->fence_bo);
1615 drmib.num_chunks = 2;
1616 drmib.chunks = (uint64_t)(uintptr_t)chunk_array;
1617 chunks[0].chunk_id = RADEON_CHUNK_ID_IB;
1618 chunks[0].length_dw = ctx->pm4_cdwords;
1619 chunks[0].chunk_data = (uint64_t)(uintptr_t)ctx->pm4;
1620 chunks[1].chunk_id = RADEON_CHUNK_ID_RELOCS;
1621 chunks[1].length_dw = ctx->creloc * sizeof(struct r600_reloc) / 4;
1622 chunks[1].chunk_data = (uint64_t)(uintptr_t)ctx->reloc;
1623 chunk_array[0] = (uint64_t)(uintptr_t)&chunks[0];
1624 chunk_array[1] = (uint64_t)(uintptr_t)&chunks[1];
1625 r = drmCommandWriteRead(ctx->radeon->fd, DRM_RADEON_CS, &drmib,
1626 sizeof(struct drm_radeon_cs));
1628 *ctx->radeon->cfence = ctx->radeon->fence;
1631 r600_context_update_fenced_list(ctx);
1633 fence = ctx->radeon->fence + 1;
1634 if (fence < ctx->radeon->fence) {
1637 r600_context_fence_wraparound(ctx, fence);
1639 ctx->radeon->fence = fence;
1642 for (int i = 0; i < ctx->creloc; i++) {
1643 ctx->bo[i]->reloc = NULL;
1644 ctx->bo[i]->last_flush = 0;
1645 radeon_bo_reference(ctx->radeon, &ctx->bo[i], NULL);
1648 ctx->pm4_dirty_cdwords = 0;
1649 ctx->pm4_cdwords = 0;
1654 /* resume queries */
1655 r600_context_queries_resume(ctx, TRUE);
1657 /* set all valid group as dirty so they get reemited on
1660 LIST_FOR_EACH_ENTRY(enable_block, &ctx->enable_list, enable_list) {
1661 if (!(enable_block->flags & BLOCK_FLAG_RESOURCE)) {
1662 if(!(enable_block->status & R600_BLOCK_STATUS_DIRTY)) {
1663 LIST_ADDTAIL(&enable_block->list,&ctx->dirty);
1664 enable_block->status |= R600_BLOCK_STATUS_DIRTY;
1667 if(!(enable_block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) {
1668 LIST_ADDTAIL(&enable_block->list,&ctx->resource_dirty);
1669 enable_block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY;
1672 ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords +
1673 enable_block->pm4_flush_ndwords;
1674 enable_block->nreg_dirty = enable_block->nreg;
1678 void r600_context_emit_fence(struct r600_context *ctx, struct r600_bo *fence_bo, unsigned offset, unsigned value)
1680 unsigned ndwords = 10;
1682 if (((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) ||
1683 (ctx->creloc >= (ctx->nreloc - 1))) {
1685 r600_context_flush(ctx);
1688 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
1689 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
1690 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1691 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1692 ctx->pm4[ctx->pm4_cdwords++] = offset << 2; /* ADDRESS_LO */
1693 ctx->pm4[ctx->pm4_cdwords++] = (1 << 29) | (0 << 24); /* DATA_SEL | INT_EN | ADDRESS_HI */
1694 ctx->pm4[ctx->pm4_cdwords++] = value; /* DATA_LO */
1695 ctx->pm4[ctx->pm4_cdwords++] = 0; /* DATA_HI */
1696 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
1697 ctx->pm4[ctx->pm4_cdwords++] = 0;
1698 r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], fence_bo);
1701 void r600_context_dump_bof(struct r600_context *ctx, const char *file)
1703 bof_t *bcs, *blob, *array, *bo, *size, *handle, *device_id, *root;
1706 root = device_id = bcs = blob = array = bo = size = handle = NULL;
1707 root = bof_object();
1710 device_id = bof_int32(ctx->radeon->device);
1711 if (device_id == NULL)
1713 if (bof_object_set(root, "device_id", device_id))
1715 bof_decref(device_id);
1718 blob = bof_blob(ctx->creloc * 16, ctx->reloc);
1721 if (bof_object_set(root, "reloc", blob))
1726 blob = bof_blob(ctx->pm4_cdwords * 4, ctx->pm4);
1729 if (bof_object_set(root, "pm4", blob))
1734 array = bof_array();
1737 for (i = 0; i < ctx->creloc; i++) {
1738 struct radeon_bo *rbo = ctx->bo[i];
1742 size = bof_int32(rbo->size);
1745 if (bof_object_set(bo, "size", size))
1749 handle = bof_int32(rbo->handle);
1752 if (bof_object_set(bo, "handle", handle))
1756 radeon_bo_map(ctx->radeon, rbo);
1757 blob = bof_blob(rbo->size, rbo->data);
1758 radeon_bo_unmap(ctx->radeon, rbo);
1761 if (bof_object_set(bo, "data", blob))
1765 if (bof_array_append(array, bo))
1770 if (bof_object_set(root, "bo", array))
1772 bof_dump_file(root, file);
1779 bof_decref(device_id);
1783 static boolean r600_query_result(struct r600_context *ctx, struct r600_query *query, boolean wait)
1785 unsigned results_base = query->results_start;
1787 u32 *results, *current_result;
1790 results = r600_bo_map(ctx->radeon, query->buffer, PB_USAGE_CPU_READ, NULL);
1792 results = r600_bo_map(ctx->radeon, query->buffer, PB_USAGE_DONTBLOCK | PB_USAGE_CPU_READ, NULL);
1797 /* count all results across all data blocks */
1798 while (results_base != query->results_end) {
1799 current_result = (u32*)((char*)results + results_base);
1801 start = (u64)current_result[0] | (u64)current_result[1] << 32;
1802 end = (u64)current_result[2] | (u64)current_result[3] << 32;
1803 if (((start & 0x8000000000000000UL) && (end & 0x8000000000000000UL))
1804 || query->type == PIPE_QUERY_TIME_ELAPSED) {
1805 query->result += end - start;
1808 results_base += 4 * 4;
1809 if (results_base >= query->buffer_size)
1813 query->results_start = query->results_end;
1814 r600_bo_unmap(ctx->radeon, query->buffer);
1818 void r600_query_begin(struct r600_context *ctx, struct r600_query *query)
1820 unsigned required_space, new_results_end;
1822 /* query request needs 6/8 dwords for begin + 6/8 dwords for end */
1823 if (query->type == PIPE_QUERY_TIME_ELAPSED)
1824 required_space = 16;
1826 required_space = 12;
1828 if ((required_space + ctx->pm4_cdwords) > ctx->pm4_ndwords) {
1830 r600_context_flush(ctx);
1833 if (query->type == PIPE_QUERY_OCCLUSION_COUNTER) {
1834 /* Count queries emitted without flushes, and flush if more than
1835 * half of buffer used, to avoid overwriting results which may be
1837 if (query->state & R600_QUERY_STATE_FLUSHED) {
1838 query->queries_emitted = 1;
1840 if (++query->queries_emitted > query->buffer_size / query->result_size / 2)
1841 r600_context_flush(ctx);
1845 new_results_end = query->results_end + query->result_size;
1846 if (new_results_end >= query->buffer_size)
1847 new_results_end = 0;
1849 /* collect current results if query buffer is full */
1850 if (new_results_end == query->results_start) {
1851 if (!(query->state & R600_QUERY_STATE_FLUSHED))
1852 r600_context_flush(ctx);
1853 r600_query_result(ctx, query, TRUE);
1856 if (query->type == PIPE_QUERY_OCCLUSION_COUNTER) {
1860 results = r600_bo_map(ctx->radeon, query->buffer, PB_USAGE_CPU_WRITE, NULL);
1862 results = (u32*)((char*)results + query->results_end);
1863 memset(results, 0, query->result_size);
1865 /* Set top bits for unused backends */
1866 for (i = 0; i < ctx->max_db; i++) {
1867 if (!(ctx->backend_mask & (1<<i))) {
1868 results[(i * 4)+1] = 0x80000000;
1869 results[(i * 4)+3] = 0x80000000;
1872 r600_bo_unmap(ctx->radeon, query->buffer);
1876 /* emit begin query */
1877 if (query->type == PIPE_QUERY_TIME_ELAPSED) {
1878 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1879 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1880 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + r600_bo_offset(query->buffer);
1881 ctx->pm4[ctx->pm4_cdwords++] = (3 << 29);
1882 ctx->pm4[ctx->pm4_cdwords++] = 0;
1883 ctx->pm4[ctx->pm4_cdwords++] = 0;
1885 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
1886 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
1887 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + r600_bo_offset(query->buffer);
1888 ctx->pm4[ctx->pm4_cdwords++] = 0;
1890 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
1891 ctx->pm4[ctx->pm4_cdwords++] = 0;
1892 r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer);
1894 query->state |= R600_QUERY_STATE_STARTED;
1895 query->state ^= R600_QUERY_STATE_ENDED;
1896 ctx->num_query_running++;
1899 void r600_query_end(struct r600_context *ctx, struct r600_query *query)
1901 /* emit end query */
1902 if (query->type == PIPE_QUERY_TIME_ELAPSED) {
1903 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1904 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1905 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + 8 + r600_bo_offset(query->buffer);
1906 ctx->pm4[ctx->pm4_cdwords++] = (3 << 29);
1907 ctx->pm4[ctx->pm4_cdwords++] = 0;
1908 ctx->pm4[ctx->pm4_cdwords++] = 0;
1910 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
1911 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
1912 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + 8 + r600_bo_offset(query->buffer);
1913 ctx->pm4[ctx->pm4_cdwords++] = 0;
1915 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
1916 ctx->pm4[ctx->pm4_cdwords++] = 0;
1917 r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer);
1919 query->results_end += query->result_size;
1920 if (query->results_end >= query->buffer_size)
1921 query->results_end = 0;
1923 query->state ^= R600_QUERY_STATE_STARTED;
1924 query->state |= R600_QUERY_STATE_ENDED;
1925 query->state &= ~R600_QUERY_STATE_FLUSHED;
1927 ctx->num_query_running--;
1930 void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int operation,
1933 if (operation == PREDICATION_OP_CLEAR) {
1934 if (ctx->pm4_cdwords + 3 > ctx->pm4_ndwords)
1935 r600_context_flush(ctx);
1937 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0);
1938 ctx->pm4[ctx->pm4_cdwords++] = 0;
1939 ctx->pm4[ctx->pm4_cdwords++] = PRED_OP(PREDICATION_OP_CLEAR);
1941 unsigned results_base = query->results_start;
1945 /* find count of the query data blocks */
1946 count = query->buffer_size + query->results_end - query->results_start;
1947 if (count >= query->buffer_size) count-=query->buffer_size;
1948 count /= query->result_size;
1950 if (ctx->pm4_cdwords + 5 * count > ctx->pm4_ndwords)
1951 r600_context_flush(ctx);
1953 op = PRED_OP(operation) | PREDICATION_DRAW_VISIBLE |
1954 (flag_wait ? PREDICATION_HINT_WAIT : PREDICATION_HINT_NOWAIT_DRAW);
1956 /* emit predicate packets for all data blocks */
1957 while (results_base != query->results_end) {
1958 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0);
1959 ctx->pm4[ctx->pm4_cdwords++] = results_base + r600_bo_offset(query->buffer);
1960 ctx->pm4[ctx->pm4_cdwords++] = op;
1961 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
1962 ctx->pm4[ctx->pm4_cdwords++] = 0;
1963 r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer);
1964 results_base += query->result_size;
1965 if (results_base >= query->buffer_size)
1967 /* set CONTINUE bit for all packets except the first */
1968 op |= PREDICATION_CONTINUE;
1973 struct r600_query *r600_context_query_create(struct r600_context *ctx, unsigned query_type)
1975 struct r600_query *query;
1977 if (query_type != PIPE_QUERY_OCCLUSION_COUNTER && query_type != PIPE_QUERY_TIME_ELAPSED)
1980 query = calloc(1, sizeof(struct r600_query));
1984 query->type = query_type;
1985 query->buffer_size = 4096;
1987 if (query_type == PIPE_QUERY_OCCLUSION_COUNTER)
1988 query->result_size = 4 * 4 * ctx->max_db;
1990 query->result_size = 4 * 4;
1992 /* adjust buffer size to simplify offsets wrapping math */
1993 query->buffer_size -= query->buffer_size % query->result_size;
1995 /* As of GL4, query buffers are normally read by the CPU after
1996 * being written by the gpu, hence staging is probably a good
1999 query->buffer = r600_bo(ctx->radeon, query->buffer_size, 1, 0,
2000 PIPE_USAGE_STAGING);
2001 if (!query->buffer) {
2006 LIST_ADDTAIL(&query->list, &ctx->query_list);
2011 void r600_context_query_destroy(struct r600_context *ctx, struct r600_query *query)
2013 r600_bo_reference(ctx->radeon, &query->buffer, NULL);
2014 LIST_DELINIT(&query->list);
2018 boolean r600_context_query_result(struct r600_context *ctx,
2019 struct r600_query *query,
2020 boolean wait, void *vresult)
2022 uint64_t *result = (uint64_t*)vresult;
2024 if (!(query->state & R600_QUERY_STATE_FLUSHED)) {
2025 r600_context_flush(ctx);
2027 if (!r600_query_result(ctx, query, wait))
2029 if (query->type == PIPE_QUERY_TIME_ELAPSED)
2030 *result = (1000000*query->result)/r600_get_clock_crystal_freq(ctx->radeon);
2032 *result = query->result;
2037 void r600_context_queries_suspend(struct r600_context *ctx)
2039 struct r600_query *query;
2041 LIST_FOR_EACH_ENTRY(query, &ctx->query_list, list) {
2042 if (query->state & R600_QUERY_STATE_STARTED) {
2043 r600_query_end(ctx, query);
2044 query->state |= R600_QUERY_STATE_SUSPENDED;
2049 void r600_context_queries_resume(struct r600_context *ctx, boolean flushed)
2051 struct r600_query *query;
2053 LIST_FOR_EACH_ENTRY(query, &ctx->query_list, list) {
2055 query->state |= R600_QUERY_STATE_FLUSHED;
2057 if (query->state & R600_QUERY_STATE_SUSPENDED) {
2058 r600_query_begin(ctx, query);
2059 query->state ^= R600_QUERY_STATE_SUSPENDED;