r600_hw_context.c revision a01791add08fbcb5386e0e9209ba21ed58fbdc42
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 */
32void 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
111err:
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
117void 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
130static 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
184int 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 */
241static 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};
245
246static const struct r600_reg r600_ctl_const_list[] = {
247	{R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0},
248};
249
250static const struct r600_reg r600_context_reg_list[] = {
251	{R_028A4C_PA_SC_MODE_CNTL, 0, 0},
252	{GROUP_FORCE_NEW_BLOCK, 0, 0},
253	{R_028040_CB_COLOR0_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(0)},
254	{GROUP_FORCE_NEW_BLOCK, 0, 0},
255	{R_0280A0_CB_COLOR0_INFO, REG_FLAG_NEED_BO, 0},
256	{R_028060_CB_COLOR0_SIZE, 0, 0},
257	{R_028080_CB_COLOR0_VIEW, 0, 0},
258	{GROUP_FORCE_NEW_BLOCK, 0, 0},
259	{R_0280E0_CB_COLOR0_FRAG, REG_FLAG_NEED_BO, 0},
260	{GROUP_FORCE_NEW_BLOCK, 0, 0},
261	{R_0280C0_CB_COLOR0_TILE, REG_FLAG_NEED_BO, 0},
262	{GROUP_FORCE_NEW_BLOCK, 0, 0},
263	{R_028044_CB_COLOR1_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(1)},
264	{GROUP_FORCE_NEW_BLOCK, 0, 0},
265	{R_0280A4_CB_COLOR1_INFO, REG_FLAG_NEED_BO, 0},
266	{R_028064_CB_COLOR1_SIZE, 0, 0},
267	{R_028084_CB_COLOR1_VIEW, 0, 0},
268	{GROUP_FORCE_NEW_BLOCK, 0, 0},
269	{R_0280E4_CB_COLOR1_FRAG, REG_FLAG_NEED_BO, 0},
270	{GROUP_FORCE_NEW_BLOCK, 0, 0},
271	{R_0280C4_CB_COLOR1_TILE, REG_FLAG_NEED_BO, 0},
272	{GROUP_FORCE_NEW_BLOCK, 0, 0},
273	{R_028048_CB_COLOR2_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(2)},
274	{GROUP_FORCE_NEW_BLOCK, 0, 0},
275	{R_0280A8_CB_COLOR2_INFO, REG_FLAG_NEED_BO, 0},
276	{R_028068_CB_COLOR2_SIZE, 0, 0},
277	{R_028088_CB_COLOR2_VIEW, 0, 0},
278	{GROUP_FORCE_NEW_BLOCK, 0, 0},
279	{R_0280E8_CB_COLOR2_FRAG, REG_FLAG_NEED_BO, 0},
280	{GROUP_FORCE_NEW_BLOCK, 0, 0},
281	{R_0280C8_CB_COLOR2_TILE, REG_FLAG_NEED_BO, 0},
282	{GROUP_FORCE_NEW_BLOCK, 0, 0},
283	{R_02804C_CB_COLOR3_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(3)},
284	{GROUP_FORCE_NEW_BLOCK, 0, 0},
285	{R_0280AC_CB_COLOR3_INFO, REG_FLAG_NEED_BO, 0},
286	{R_02806C_CB_COLOR3_SIZE, 0, 0},
287	{R_02808C_CB_COLOR3_VIEW, 0, 0},
288	{GROUP_FORCE_NEW_BLOCK, 0, 0},
289	{R_0280EC_CB_COLOR3_FRAG, REG_FLAG_NEED_BO, 0},
290	{GROUP_FORCE_NEW_BLOCK, 0, 0},
291	{R_0280CC_CB_COLOR3_TILE, REG_FLAG_NEED_BO, 0},
292	{GROUP_FORCE_NEW_BLOCK, 0, 0},
293	{R_028050_CB_COLOR4_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(4)},
294	{GROUP_FORCE_NEW_BLOCK, 0, 0},
295	{R_0280B0_CB_COLOR4_INFO, REG_FLAG_NEED_BO, 0},
296	{R_028070_CB_COLOR4_SIZE, 0, 0},
297	{R_028090_CB_COLOR4_VIEW, 0, 0},
298	{GROUP_FORCE_NEW_BLOCK, 0, 0},
299	{R_0280F0_CB_COLOR4_FRAG, REG_FLAG_NEED_BO, 0},
300	{GROUP_FORCE_NEW_BLOCK, 0, 0},
301	{R_0280D0_CB_COLOR4_TILE, REG_FLAG_NEED_BO, 0},
302	{GROUP_FORCE_NEW_BLOCK, 0, 0},
303	{R_028054_CB_COLOR5_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(5)},
304	{GROUP_FORCE_NEW_BLOCK, 0, 0},
305	{R_0280B4_CB_COLOR5_INFO, REG_FLAG_NEED_BO, 0},
306	{R_028074_CB_COLOR5_SIZE, 0, 0},
307	{R_028094_CB_COLOR5_VIEW, 0, 0},
308	{GROUP_FORCE_NEW_BLOCK, 0, 0},
309	{R_0280F4_CB_COLOR5_FRAG, REG_FLAG_NEED_BO, 0},
310	{GROUP_FORCE_NEW_BLOCK, 0, 0},
311	{R_0280D4_CB_COLOR5_TILE, REG_FLAG_NEED_BO, 0},
312	{R_028058_CB_COLOR6_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(6)},
313	{R_0280B8_CB_COLOR6_INFO, REG_FLAG_NEED_BO, 0},
314	{R_028078_CB_COLOR6_SIZE, 0, 0},
315	{R_028098_CB_COLOR6_VIEW, 0, 0},
316	{GROUP_FORCE_NEW_BLOCK, 0, 0},
317	{R_0280F8_CB_COLOR6_FRAG, REG_FLAG_NEED_BO, 0},
318	{GROUP_FORCE_NEW_BLOCK, 0, 0},
319	{R_0280D8_CB_COLOR6_TILE, REG_FLAG_NEED_BO, 0},
320	{GROUP_FORCE_NEW_BLOCK, 0, 0},
321	{R_02805C_CB_COLOR7_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(7)},
322	{GROUP_FORCE_NEW_BLOCK, 0, 0},
323	{R_0280BC_CB_COLOR7_INFO, REG_FLAG_NEED_BO, 0},
324	{R_02807C_CB_COLOR7_SIZE, 0, 0},
325	{R_02809C_CB_COLOR7_VIEW, 0, 0},
326	{R_0280FC_CB_COLOR7_FRAG, REG_FLAG_NEED_BO, 0},
327	{R_0280DC_CB_COLOR7_TILE, REG_FLAG_NEED_BO, 0},
328	{R_028120_CB_CLEAR_RED, 0, 0},
329	{R_028124_CB_CLEAR_GREEN, 0, 0},
330	{R_028128_CB_CLEAR_BLUE, 0, 0},
331	{R_02812C_CB_CLEAR_ALPHA, 0, 0},
332	{R_028414_CB_BLEND_RED, 0, 0},
333	{R_028418_CB_BLEND_GREEN, 0, 0},
334	{R_02841C_CB_BLEND_BLUE, 0, 0},
335	{R_028420_CB_BLEND_ALPHA, 0, 0},
336	{R_028424_CB_FOG_RED, 0, 0},
337	{R_028428_CB_FOG_GREEN, 0, 0},
338	{R_02842C_CB_FOG_BLUE, 0, 0},
339	{R_028430_DB_STENCILREFMASK, 0, 0},
340	{R_028434_DB_STENCILREFMASK_BF, 0, 0},
341	{R_028780_CB_BLEND0_CONTROL, REG_FLAG_NOT_R600, 0},
342	{R_028784_CB_BLEND1_CONTROL, REG_FLAG_NOT_R600, 0},
343	{R_028788_CB_BLEND2_CONTROL, REG_FLAG_NOT_R600, 0},
344	{R_02878C_CB_BLEND3_CONTROL, REG_FLAG_NOT_R600, 0},
345	{R_028790_CB_BLEND4_CONTROL, REG_FLAG_NOT_R600, 0},
346	{R_028794_CB_BLEND5_CONTROL, REG_FLAG_NOT_R600, 0},
347	{R_028798_CB_BLEND6_CONTROL, REG_FLAG_NOT_R600, 0},
348	{R_02879C_CB_BLEND7_CONTROL, REG_FLAG_NOT_R600, 0},
349	{R_0287A0_CB_SHADER_CONTROL, 0, 0},
350	{R_028800_DB_DEPTH_CONTROL, 0, 0},
351	{R_028804_CB_BLEND_CONTROL, 0, 0},
352	{R_02880C_DB_SHADER_CONTROL, 0, 0},
353	{R_02800C_DB_DEPTH_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_DEPTH},
354	{R_028000_DB_DEPTH_SIZE, 0, 0},
355	{R_028004_DB_DEPTH_VIEW, 0, 0},
356	{GROUP_FORCE_NEW_BLOCK, 0, 0},
357	{R_028010_DB_DEPTH_INFO, REG_FLAG_NEED_BO, 0},
358	{R_028A6C_VGT_GS_OUT_PRIM_TYPE, 0, 0},
359	{R_028D24_DB_HTILE_SURFACE, 0, 0},
360	{R_028D34_DB_PREFETCH_LIMIT, 0, 0},
361	{R_028D44_DB_ALPHA_TO_MASK, 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
509static int r600_loop_const_init(struct r600_context *ctx, uint32_t offset)
510{
511	unsigned nreg = 32;
512	struct r600_reg r600_loop_consts[32];
513	int i;
514
515	for (i = 0; i < nreg; i++) {
516		r600_loop_consts[i].offset = R600_LOOP_CONST_OFFSET + ((offset + i) * 4);
517		r600_loop_consts[i].flags = REG_FLAG_DIRTY_ALWAYS;
518		r600_loop_consts[i].sbu_flags = 0;
519	}
520	return r600_context_add_block(ctx, r600_loop_consts, nreg, PKT3_SET_LOOP_CONST, R600_LOOP_CONST_OFFSET);
521}
522
523/* initialize */
524void r600_context_fini(struct r600_context *ctx)
525{
526	struct r600_block *block;
527	struct r600_range *range;
528
529	if (ctx->range) {
530		for (int i = 0; i < NUM_RANGES; i++) {
531			if (!ctx->range[i].blocks)
532				continue;
533			for (int j = 0; j < (1 << HASH_SHIFT); j++) {
534				block = ctx->range[i].blocks[j];
535				if (block) {
536					for (int k = 0, offset = block->start_offset; k < block->nreg; k++, offset += 4) {
537						range = &ctx->range[CTX_RANGE_ID(offset)];
538						range->blocks[CTX_BLOCK_ID(offset)] = NULL;
539					}
540					for (int k = 1; k <= block->nbo; k++) {
541						pipe_resource_reference((struct pipe_resource**)&block->reloc[k].bo, NULL);
542					}
543					free(block);
544				}
545			}
546			free(ctx->range[i].blocks);
547		}
548	}
549	free(ctx->blocks);
550}
551
552int r600_setup_block_table(struct r600_context *ctx)
553{
554	/* setup block table */
555	int c = 0;
556	ctx->blocks = calloc(ctx->nblocks, sizeof(void*));
557	if (!ctx->blocks)
558		return -ENOMEM;
559	for (int i = 0; i < NUM_RANGES; i++) {
560		if (!ctx->range[i].blocks)
561			continue;
562		for (int j = 0, add; j < (1 << HASH_SHIFT); j++) {
563			if (!ctx->range[i].blocks[j])
564				continue;
565
566			add = 1;
567			for (int k = 0; k < c; k++) {
568				if (ctx->blocks[k] == ctx->range[i].blocks[j]) {
569					add = 0;
570					break;
571				}
572			}
573			if (add) {
574				assert(c < ctx->nblocks);
575				ctx->blocks[c++] = ctx->range[i].blocks[j];
576				j += (ctx->range[i].blocks[j]->nreg) - 1;
577			}
578		}
579	}
580	return 0;
581}
582
583int r600_context_init(struct r600_context *ctx)
584{
585	int r;
586
587	/* add blocks */
588	r = r600_context_add_block(ctx, r600_config_reg_list,
589				   Elements(r600_config_reg_list), PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
590	if (r)
591		goto out_err;
592	r = r600_context_add_block(ctx, r600_context_reg_list,
593				   Elements(r600_context_reg_list), PKT3_SET_CONTEXT_REG, R600_CONTEXT_REG_OFFSET);
594	if (r)
595		goto out_err;
596	r = r600_context_add_block(ctx, r600_ctl_const_list,
597				   Elements(r600_ctl_const_list), PKT3_SET_CTL_CONST, R600_CTL_CONST_OFFSET);
598	if (r)
599		goto out_err;
600
601	/* PS loop const */
602	r600_loop_const_init(ctx, 0);
603	/* VS loop const */
604	r600_loop_const_init(ctx, 32);
605
606	r = r600_setup_block_table(ctx);
607	if (r)
608		goto out_err;
609
610	ctx->max_db = 4;
611	return 0;
612out_err:
613	r600_context_fini(ctx);
614	return r;
615}
616
617void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
618			boolean count_draw_in)
619{
620	struct r600_atom *state;
621
622	/* The number of dwords we already used in the CS so far. */
623	num_dw += ctx->cs->cdw;
624
625	if (count_draw_in) {
626		/* The number of dwords all the dirty states would take. */
627		LIST_FOR_EACH_ENTRY(state, &ctx->dirty_states, head) {
628			num_dw += state->num_dw;
629		}
630
631		num_dw += ctx->pm4_dirty_cdwords;
632
633		/* The upper-bound of how much a draw command would take. */
634		num_dw += R600_MAX_DRAW_CS_DWORDS;
635	}
636
637	/* Count in queries_suspend. */
638	num_dw += ctx->num_cs_dw_nontimer_queries_suspend;
639	num_dw += ctx->num_cs_dw_timer_queries_suspend;
640
641	/* Count in streamout_end at the end of CS. */
642	num_dw += ctx->num_cs_dw_streamout_end;
643
644	/* Count in render_condition(NULL) at the end of CS. */
645	if (ctx->predicate_drawing) {
646		num_dw += 3;
647	}
648
649	/* Count in framebuffer cache flushes at the end of CS. */
650	num_dw += 7; /* one SURFACE_SYNC and CACHE_FLUSH_AND_INV (r6xx-only) */
651
652	/* Save 16 dwords for the fence mechanism. */
653	num_dw += 16;
654
655	/* Flush if there's not enough space. */
656	if (num_dw > RADEON_MAX_CMDBUF_DWORDS) {
657		r600_flush(&ctx->context, NULL, RADEON_FLUSH_ASYNC);
658	}
659}
660
661void r600_context_dirty_block(struct r600_context *ctx,
662			      struct r600_block *block,
663			      int dirty, int index)
664{
665	if ((index + 1) > block->nreg_dirty)
666		block->nreg_dirty = index + 1;
667
668	if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) {
669		block->status |= R600_BLOCK_STATUS_DIRTY;
670		ctx->pm4_dirty_cdwords += block->pm4_ndwords;
671		if (!(block->status & R600_BLOCK_STATUS_ENABLED)) {
672			block->status |= R600_BLOCK_STATUS_ENABLED;
673			LIST_ADDTAIL(&block->enable_list, &ctx->enable_list);
674		}
675		LIST_ADDTAIL(&block->list,&ctx->dirty);
676
677		if (block->flags & REG_FLAG_FLUSH_CHANGE) {
678			r600_context_ps_partial_flush(ctx);
679		}
680	}
681}
682
683/**
684 * If reg needs a reloc, this function will add it to its block's reloc list.
685 * @return true if reg needs a reloc, false otherwise
686 */
687static bool r600_reg_set_block_reloc(struct r600_pipe_reg *reg)
688{
689	unsigned reloc_id;
690
691	if (!reg->block->pm4_bo_index[reg->id]) {
692		return false;
693	}
694	/* find relocation */
695	reloc_id = reg->block->pm4_bo_index[reg->id];
696	pipe_resource_reference(
697		(struct pipe_resource**)&reg->block->reloc[reloc_id].bo,
698		&reg->bo->b.b);
699	reg->block->reloc[reloc_id].bo_usage = reg->bo_usage;
700	return true;
701}
702
703/**
704 * This function will emit all the registers in state directly to the command
705 * stream allowing you to bypass the r600_context dirty list.
706 *
707 * This is used for dispatching compute shaders to avoid mixing compute and
708 * 3D states in the context's dirty list.
709 *
710 * @param pkt_flags Should be either 0 or RADEON_CP_PACKET3_COMPUTE_MODE.  This
711 * value will be passed on to r600_context_block_emit_dirty an or'd against
712 * the PKT3 headers.
713 */
714void r600_context_pipe_state_emit(struct r600_context *ctx,
715                          struct r600_pipe_state *state,
716                          unsigned pkt_flags)
717{
718	unsigned i;
719
720	/* Mark all blocks as dirty:
721	 * Since two registers can be in the same block, we need to make sure
722	 * we mark all the blocks dirty before we emit any of them.  If we were
723	 * to mark blocks dirty and emit them in the same loop, like this:
724	 *
725	 * foreach (reg in state->regs) {
726	 *     mark_dirty(reg->block)
727	 *     emit_block(reg->block)
728	 * }
729	 *
730	 * Then if we have two registers in this state that are in the same
731	 * block, we would end up emitting that block twice.
732	 */
733	for (i = 0; i < state->nregs; i++) {
734		struct r600_pipe_reg *reg = &state->regs[i];
735		/* Mark all the registers in the block as dirty */
736		reg->block->nreg_dirty = reg->block->nreg;
737		reg->block->status |= R600_BLOCK_STATUS_DIRTY;
738		/* Update the reloc for this register if necessary. */
739		r600_reg_set_block_reloc(reg);
740	}
741
742	/* Emit the registers writes */
743	for (i = 0; i < state->nregs; i++) {
744		struct r600_pipe_reg *reg = &state->regs[i];
745		if (reg->block->status & R600_BLOCK_STATUS_DIRTY) {
746			r600_context_block_emit_dirty(ctx, reg->block, pkt_flags);
747		}
748	}
749}
750
751void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state)
752{
753	struct r600_block *block;
754	int dirty;
755	for (int i = 0; i < state->nregs; i++) {
756		unsigned id;
757		struct r600_pipe_reg *reg = &state->regs[i];
758
759		block = reg->block;
760		id = reg->id;
761
762		dirty = block->status & R600_BLOCK_STATUS_DIRTY;
763
764		if (reg->value != block->reg[id]) {
765			block->reg[id] = reg->value;
766			dirty |= R600_BLOCK_STATUS_DIRTY;
767		}
768		if (block->flags & REG_FLAG_DIRTY_ALWAYS)
769			dirty |= R600_BLOCK_STATUS_DIRTY;
770		if (r600_reg_set_block_reloc(reg)) {
771			/* always force dirty for relocs for now */
772			dirty |= R600_BLOCK_STATUS_DIRTY;
773		}
774
775		if (dirty)
776			r600_context_dirty_block(ctx, block, dirty, id);
777	}
778}
779
780/**
781 * @param pkt_flags should be set to RADEON_CP_PACKET3_COMPUTE_MODE if this
782 * block will be used for compute shaders.
783 */
784void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block,
785	unsigned pkt_flags)
786{
787	struct radeon_winsys_cs *cs = ctx->cs;
788	int optional = block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS);
789	int cp_dwords = block->pm4_ndwords, start_dword = 0;
790	int new_dwords = 0;
791	int nbo = block->nbo;
792
793	if (block->nreg_dirty == 0 && optional) {
794		goto out;
795	}
796
797	if (nbo) {
798		for (int j = 0; j < block->nreg; j++) {
799			if (block->pm4_bo_index[j]) {
800				/* find relocation */
801				struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]];
802				if (reloc->bo) {
803					block->pm4[reloc->bo_pm4_index] =
804							r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage);
805				} else {
806					block->pm4[reloc->bo_pm4_index] = 0;
807				}
808				nbo--;
809				if (nbo == 0)
810					break;
811
812			}
813		}
814	}
815
816	optional &= (block->nreg_dirty != block->nreg);
817	if (optional) {
818		new_dwords = block->nreg_dirty;
819		start_dword = cs->cdw;
820		cp_dwords = new_dwords + 2;
821	}
822	memcpy(&cs->buf[cs->cdw], block->pm4, cp_dwords * 4);
823
824	/* We are applying the pkt_flags after copying the register block to
825	 * the the command stream, because it is possible this block will be
826	 * emitted with a different pkt_flags, and we don't want to store the
827	 * pkt_flags in the block.
828	 */
829	cs->buf[cs->cdw] |= pkt_flags;
830	cs->cdw += cp_dwords;
831
832	if (optional) {
833		uint32_t newword;
834
835		newword = cs->buf[start_dword];
836		newword &= PKT_COUNT_C;
837		newword |= PKT_COUNT_S(new_dwords);
838		cs->buf[start_dword] = newword;
839	}
840out:
841	block->status ^= R600_BLOCK_STATUS_DIRTY;
842	block->nreg_dirty = 0;
843	LIST_DELINIT(&block->list);
844}
845
846void r600_inval_shader_cache(struct r600_context *ctx)
847{
848	ctx->surface_sync_cmd.flush_flags |= S_0085F0_SH_ACTION_ENA(1);
849	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
850}
851
852void r600_inval_texture_cache(struct r600_context *ctx)
853{
854	ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
855	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
856}
857
858void r600_inval_vertex_cache(struct r600_context *ctx)
859{
860	if (ctx->has_vertex_cache) {
861		ctx->surface_sync_cmd.flush_flags |= S_0085F0_VC_ACTION_ENA(1);
862	} else {
863		/* Some GPUs don't have the vertex cache and must use the texture cache instead. */
864		ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
865	}
866	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
867}
868
869void r600_flush_framebuffer(struct r600_context *ctx, bool flush_now)
870{
871	if (!(ctx->flags & R600_CONTEXT_DST_CACHES_DIRTY))
872		return;
873
874	ctx->surface_sync_cmd.flush_flags |=
875		r600_get_cb_flush_flags(ctx) |
876		(ctx->framebuffer.zsbuf ? S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1) : 0);
877
878	if (flush_now) {
879		r600_emit_atom(ctx, &ctx->surface_sync_cmd.atom);
880	} else {
881		r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
882	}
883
884	/* Also add a complete cache flush to work around broken flushing on R6xx. */
885	if (ctx->chip_class == R600) {
886		if (flush_now) {
887			r600_emit_atom(ctx, &ctx->r6xx_flush_and_inv_cmd);
888		} else {
889			r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
890		}
891	}
892
893	ctx->flags &= ~R600_CONTEXT_DST_CACHES_DIRTY;
894}
895
896void r600_context_flush(struct r600_context *ctx, unsigned flags)
897{
898	struct radeon_winsys_cs *cs = ctx->cs;
899	struct r600_block *enable_block = NULL;
900	bool timer_queries_suspended = false;
901	bool nontimer_queries_suspended = false;
902	bool streamout_suspended = false;
903
904	if (cs->cdw == ctx->start_cs_cmd.atom.num_dw)
905		return;
906
907	/* suspend queries */
908	if (ctx->num_cs_dw_timer_queries_suspend) {
909		r600_suspend_timer_queries(ctx);
910		timer_queries_suspended = true;
911	}
912	if (ctx->num_cs_dw_nontimer_queries_suspend) {
913		r600_suspend_nontimer_queries(ctx);
914		nontimer_queries_suspended = true;
915	}
916
917	if (ctx->num_cs_dw_streamout_end) {
918		r600_context_streamout_end(ctx);
919		streamout_suspended = true;
920	}
921
922	r600_flush_framebuffer(ctx, true);
923
924	/* partial flush is needed to avoid lockups on some chips with user fences */
925	r600_context_ps_partial_flush(ctx);
926
927	/* old kernels and userspace don't set SX_MISC, so we must reset it to 0 here */
928	if (ctx->chip_class <= R700) {
929		r600_write_context_reg(cs, R_028350_SX_MISC, 0);
930	}
931
932	/* force to keep tiling flags */
933	flags |= RADEON_FLUSH_KEEP_TILING_FLAGS;
934
935	/* Flush the CS. */
936	ctx->ws->cs_flush(ctx->cs, flags);
937
938	ctx->pm4_dirty_cdwords = 0;
939	ctx->flags = 0;
940
941	/* Begin a new CS. */
942	r600_emit_atom(ctx, &ctx->start_cs_cmd.atom);
943
944	/* Invalidate caches. */
945	r600_inval_texture_cache(ctx);
946	r600_flush_framebuffer(ctx, false);
947
948	/* Re-emit states. */
949	r600_atom_dirty(ctx, &ctx->alphatest_state.atom);
950	r600_atom_dirty(ctx, &ctx->cb_misc_state.atom);
951	r600_atom_dirty(ctx, &ctx->db_misc_state.atom);
952	/* reemit sampler, will only matter if atom_sampler.num_dw != 0 */
953	r600_atom_dirty(ctx, &ctx->vs_samplers.atom_sampler);
954	r600_atom_dirty(ctx, &ctx->ps_samplers.atom_sampler);
955	if (ctx->chip_class <= R700) {
956		r600_atom_dirty(ctx, &ctx->seamless_cube_map.atom);
957	}
958	r600_atom_dirty(ctx, &ctx->sample_mask.atom);
959
960	ctx->vertex_buffer_state.dirty_mask = ctx->vertex_buffer_state.enabled_mask;
961	r600_vertex_buffers_dirty(ctx);
962
963	ctx->vs_constbuf_state.dirty_mask = ctx->vs_constbuf_state.enabled_mask;
964	ctx->ps_constbuf_state.dirty_mask = ctx->ps_constbuf_state.enabled_mask;
965	r600_constant_buffers_dirty(ctx, &ctx->vs_constbuf_state);
966	r600_constant_buffers_dirty(ctx, &ctx->ps_constbuf_state);
967
968	ctx->vs_samplers.views.dirty_mask = ctx->vs_samplers.views.enabled_mask;
969	ctx->ps_samplers.views.dirty_mask = ctx->ps_samplers.views.enabled_mask;
970	r600_sampler_views_dirty(ctx, &ctx->vs_samplers.views);
971	r600_sampler_views_dirty(ctx, &ctx->ps_samplers.views);
972
973	if (streamout_suspended) {
974		ctx->streamout_start = TRUE;
975		ctx->streamout_append_bitmask = ~0;
976	}
977
978	/* resume queries */
979	if (timer_queries_suspended) {
980		r600_resume_timer_queries(ctx);
981	}
982	if (nontimer_queries_suspended) {
983		r600_resume_nontimer_queries(ctx);
984	}
985
986	/* set all valid group as dirty so they get reemited on
987	 * next draw command
988	 */
989	LIST_FOR_EACH_ENTRY(enable_block, &ctx->enable_list, enable_list) {
990		if(!(enable_block->status & R600_BLOCK_STATUS_DIRTY)) {
991			LIST_ADDTAIL(&enable_block->list,&ctx->dirty);
992			enable_block->status |= R600_BLOCK_STATUS_DIRTY;
993		}
994		ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords;
995		enable_block->nreg_dirty = enable_block->nreg;
996	}
997}
998
999void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value)
1000{
1001	struct radeon_winsys_cs *cs = ctx->cs;
1002	uint64_t va;
1003
1004	r600_need_cs_space(ctx, 10, FALSE);
1005
1006	va = r600_resource_va(&ctx->screen->screen, (void*)fence_bo);
1007	va = va + (offset << 2);
1008
1009	r600_context_ps_partial_flush(ctx);
1010	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
1011	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
1012	cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL;       /* ADDRESS_LO */
1013	/* DATA_SEL | INT_EN | ADDRESS_HI */
1014	cs->buf[cs->cdw++] = (1 << 29) | (0 << 24) | ((va >> 32UL) & 0xFF);
1015	cs->buf[cs->cdw++] = value;                   /* DATA_LO */
1016	cs->buf[cs->cdw++] = 0;                       /* DATA_HI */
1017	cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1018	cs->buf[cs->cdw++] = r600_context_bo_reloc(ctx, fence_bo, RADEON_USAGE_WRITE);
1019}
1020
1021static void r600_flush_vgt_streamout(struct r600_context *ctx)
1022{
1023	struct radeon_winsys_cs *cs = ctx->cs;
1024
1025	cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0);
1026	cs->buf[cs->cdw++] = (R_008490_CP_STRMOUT_CNTL - R600_CONFIG_REG_OFFSET) >> 2;
1027	cs->buf[cs->cdw++] = 0;
1028
1029	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
1030	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0);
1031
1032	cs->buf[cs->cdw++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0);
1033	cs->buf[cs->cdw++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */
1034	cs->buf[cs->cdw++] = R_008490_CP_STRMOUT_CNTL >> 2;  /* register */
1035	cs->buf[cs->cdw++] = 0;
1036	cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* reference value */
1037	cs->buf[cs->cdw++] = S_008490_OFFSET_UPDATE_DONE(1); /* mask */
1038	cs->buf[cs->cdw++] = 4; /* poll interval */
1039}
1040
1041static void r600_set_streamout_enable(struct r600_context *ctx, unsigned buffer_enable_bit)
1042{
1043	struct radeon_winsys_cs *cs = ctx->cs;
1044
1045	if (buffer_enable_bit) {
1046		cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1047		cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1048		cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(1);
1049
1050		cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1051		cs->buf[cs->cdw++] = (R_028B20_VGT_STRMOUT_BUFFER_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1052		cs->buf[cs->cdw++] = buffer_enable_bit;
1053	} else {
1054		cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
1055		cs->buf[cs->cdw++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2;
1056		cs->buf[cs->cdw++] = S_028AB0_STREAMOUT(0);
1057	}
1058}
1059
1060void r600_context_streamout_begin(struct r600_context *ctx)
1061{
1062	struct radeon_winsys_cs *cs = ctx->cs;
1063	struct r600_so_target **t = ctx->so_targets;
1064	unsigned *stride_in_dw = ctx->vs_shader->so.stride;
1065	unsigned buffer_en, i, update_flags = 0;
1066	uint64_t va;
1067
1068	buffer_en = (ctx->num_so_targets >= 1 && t[0] ? 1 : 0) |
1069		    (ctx->num_so_targets >= 2 && t[1] ? 2 : 0) |
1070		    (ctx->num_so_targets >= 3 && t[2] ? 4 : 0) |
1071		    (ctx->num_so_targets >= 4 && t[3] ? 8 : 0);
1072
1073	ctx->num_cs_dw_streamout_end =
1074		12 + /* flush_vgt_streamout */
1075		util_bitcount(buffer_en) * 8 + /* STRMOUT_BUFFER_UPDATE */
1076		3 /* set_streamout_enable(0) */;
1077
1078	r600_need_cs_space(ctx,
1079			   12 + /* flush_vgt_streamout */
1080			   6 + /* set_streamout_enable */
1081			   util_bitcount(buffer_en) * 7 + /* SET_CONTEXT_REG */
1082			   (ctx->chip_class == R700 ? util_bitcount(buffer_en) * 5 : 0) + /* STRMOUT_BASE_UPDATE */
1083			   util_bitcount(buffer_en & ctx->streamout_append_bitmask) * 8 + /* STRMOUT_BUFFER_UPDATE */
1084			   util_bitcount(buffer_en & ~ctx->streamout_append_bitmask) * 6 + /* STRMOUT_BUFFER_UPDATE */
1085			   (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770 ? 2 : 0) + /* SURFACE_BASE_UPDATE */
1086			   ctx->num_cs_dw_streamout_end, TRUE);
1087
1088	if (ctx->chip_class >= EVERGREEN) {
1089		evergreen_flush_vgt_streamout(ctx);
1090		evergreen_set_streamout_enable(ctx, buffer_en);
1091	} else {
1092		r600_flush_vgt_streamout(ctx);
1093		r600_set_streamout_enable(ctx, buffer_en);
1094	}
1095
1096	for (i = 0; i < ctx->num_so_targets; i++) {
1097		if (t[i]) {
1098			t[i]->stride_in_dw = stride_in_dw[i];
1099			t[i]->so_index = i;
1100			va = r600_resource_va(&ctx->screen->screen,
1101					      (void*)t[i]->b.buffer);
1102
1103			update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i);
1104
1105			cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, 3, 0);
1106			cs->buf[cs->cdw++] = (R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 +
1107							16*i - R600_CONTEXT_REG_OFFSET) >> 2;
1108			cs->buf[cs->cdw++] = (t[i]->b.buffer_offset +
1109							t[i]->b.buffer_size) >> 2; /* BUFFER_SIZE (in DW) */
1110			cs->buf[cs->cdw++] = stride_in_dw[i];		   /* VTX_STRIDE (in DW) */
1111			cs->buf[cs->cdw++] = va >> 8;			   /* BUFFER_BASE */
1112
1113			cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1114			cs->buf[cs->cdw++] =
1115				r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1116						      RADEON_USAGE_WRITE);
1117
1118			/* R7xx requires this packet after updating BUFFER_BASE.
1119			 * Without this, R7xx locks up. */
1120			if (ctx->chip_class == R700) {
1121				cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BASE_UPDATE, 1, 0);
1122				cs->buf[cs->cdw++] = i;
1123				cs->buf[cs->cdw++] = va >> 8;
1124
1125				cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1126				cs->buf[cs->cdw++] =
1127					r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer),
1128							      RADEON_USAGE_WRITE);
1129			}
1130
1131			if (ctx->streamout_append_bitmask & (1 << i)) {
1132				va = r600_resource_va(&ctx->screen->screen,
1133						      (void*)t[i]->filled_size);
1134				/* Append. */
1135				cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1136				cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1137							       STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_MEM); /* control */
1138				cs->buf[cs->cdw++] = 0; /* unused */
1139				cs->buf[cs->cdw++] = 0; /* unused */
1140				cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL; /* src address lo */
1141				cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* src address hi */
1142
1143				cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1144				cs->buf[cs->cdw++] =
1145					r600_context_bo_reloc(ctx,  t[i]->filled_size,
1146							      RADEON_USAGE_READ);
1147			} else {
1148				/* Start from the beginning. */
1149				cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1150				cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1151							       STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_PACKET); /* control */
1152				cs->buf[cs->cdw++] = 0; /* unused */
1153				cs->buf[cs->cdw++] = 0; /* unused */
1154				cs->buf[cs->cdw++] = t[i]->b.buffer_offset >> 2; /* buffer offset in DW */
1155				cs->buf[cs->cdw++] = 0; /* unused */
1156			}
1157		}
1158	}
1159
1160	if (ctx->family > CHIP_R600 && ctx->family < CHIP_RV770) {
1161		cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0);
1162		cs->buf[cs->cdw++] = update_flags;
1163	}
1164}
1165
1166void r600_context_streamout_end(struct r600_context *ctx)
1167{
1168	struct radeon_winsys_cs *cs = ctx->cs;
1169	struct r600_so_target **t = ctx->so_targets;
1170	unsigned i, flush_flags = 0;
1171	uint64_t va;
1172
1173	if (ctx->chip_class >= EVERGREEN) {
1174		evergreen_flush_vgt_streamout(ctx);
1175	} else {
1176		r600_flush_vgt_streamout(ctx);
1177	}
1178
1179	for (i = 0; i < ctx->num_so_targets; i++) {
1180		if (t[i]) {
1181			va = r600_resource_va(&ctx->screen->screen,
1182					      (void*)t[i]->filled_size);
1183			cs->buf[cs->cdw++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0);
1184			cs->buf[cs->cdw++] = STRMOUT_SELECT_BUFFER(i) |
1185						       STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_NONE) |
1186						       STRMOUT_STORE_BUFFER_FILLED_SIZE; /* control */
1187			cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL;     /* dst address lo */
1188			cs->buf[cs->cdw++] = (va >> 32UL) & 0xFFUL; /* dst address hi */
1189			cs->buf[cs->cdw++] = 0; /* unused */
1190			cs->buf[cs->cdw++] = 0; /* unused */
1191
1192			cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
1193			cs->buf[cs->cdw++] =
1194				r600_context_bo_reloc(ctx,  t[i]->filled_size,
1195						      RADEON_USAGE_WRITE);
1196
1197			flush_flags |= S_0085F0_SO0_DEST_BASE_ENA(1) << i;
1198		}
1199	}
1200
1201	if (ctx->chip_class >= EVERGREEN) {
1202		evergreen_set_streamout_enable(ctx, 0);
1203	} else {
1204		r600_set_streamout_enable(ctx, 0);
1205	}
1206
1207	/* This is needed to fix cache flushes on r600. */
1208	if (ctx->chip_class == R600) {
1209		if (ctx->family == CHIP_RV670 ||
1210		    ctx->family == CHIP_RS780 ||
1211		    ctx->family == CHIP_RS880) {
1212			flush_flags |= S_0085F0_DEST_BASE_0_ENA(1);
1213		}
1214
1215		r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
1216	}
1217
1218	/* Flush streamout caches. */
1219	ctx->surface_sync_cmd.flush_flags |=
1220		S_0085F0_SMX_ACTION_ENA(1) | flush_flags;
1221	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
1222
1223	ctx->num_cs_dw_streamout_end = 0;
1224}
1225