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#include "r600_formats.h"
24#include "r600_shader.h"
25#include "r600d.h"
26
27#include "pipe/p_shader_tokens.h"
28#include "util/u_pack_color.h"
29#include "util/u_memory.h"
30#include "util/u_framebuffer.h"
31#include "util/u_dual_blend.h"
32
33static uint32_t r600_translate_blend_function(int blend_func)
34{
35	switch (blend_func) {
36	case PIPE_BLEND_ADD:
37		return V_028804_COMB_DST_PLUS_SRC;
38	case PIPE_BLEND_SUBTRACT:
39		return V_028804_COMB_SRC_MINUS_DST;
40	case PIPE_BLEND_REVERSE_SUBTRACT:
41		return V_028804_COMB_DST_MINUS_SRC;
42	case PIPE_BLEND_MIN:
43		return V_028804_COMB_MIN_DST_SRC;
44	case PIPE_BLEND_MAX:
45		return V_028804_COMB_MAX_DST_SRC;
46	default:
47		R600_ERR("Unknown blend function %d\n", blend_func);
48		assert(0);
49		break;
50	}
51	return 0;
52}
53
54static uint32_t r600_translate_blend_factor(int blend_fact)
55{
56	switch (blend_fact) {
57	case PIPE_BLENDFACTOR_ONE:
58		return V_028804_BLEND_ONE;
59	case PIPE_BLENDFACTOR_SRC_COLOR:
60		return V_028804_BLEND_SRC_COLOR;
61	case PIPE_BLENDFACTOR_SRC_ALPHA:
62		return V_028804_BLEND_SRC_ALPHA;
63	case PIPE_BLENDFACTOR_DST_ALPHA:
64		return V_028804_BLEND_DST_ALPHA;
65	case PIPE_BLENDFACTOR_DST_COLOR:
66		return V_028804_BLEND_DST_COLOR;
67	case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
68		return V_028804_BLEND_SRC_ALPHA_SATURATE;
69	case PIPE_BLENDFACTOR_CONST_COLOR:
70		return V_028804_BLEND_CONST_COLOR;
71	case PIPE_BLENDFACTOR_CONST_ALPHA:
72		return V_028804_BLEND_CONST_ALPHA;
73	case PIPE_BLENDFACTOR_ZERO:
74		return V_028804_BLEND_ZERO;
75	case PIPE_BLENDFACTOR_INV_SRC_COLOR:
76		return V_028804_BLEND_ONE_MINUS_SRC_COLOR;
77	case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
78		return V_028804_BLEND_ONE_MINUS_SRC_ALPHA;
79	case PIPE_BLENDFACTOR_INV_DST_ALPHA:
80		return V_028804_BLEND_ONE_MINUS_DST_ALPHA;
81	case PIPE_BLENDFACTOR_INV_DST_COLOR:
82		return V_028804_BLEND_ONE_MINUS_DST_COLOR;
83	case PIPE_BLENDFACTOR_INV_CONST_COLOR:
84		return V_028804_BLEND_ONE_MINUS_CONST_COLOR;
85	case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
86		return V_028804_BLEND_ONE_MINUS_CONST_ALPHA;
87	case PIPE_BLENDFACTOR_SRC1_COLOR:
88		return V_028804_BLEND_SRC1_COLOR;
89	case PIPE_BLENDFACTOR_SRC1_ALPHA:
90		return V_028804_BLEND_SRC1_ALPHA;
91	case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
92		return V_028804_BLEND_INV_SRC1_COLOR;
93	case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
94		return V_028804_BLEND_INV_SRC1_ALPHA;
95	default:
96		R600_ERR("Bad blend factor %d not supported!\n", blend_fact);
97		assert(0);
98		break;
99	}
100	return 0;
101}
102
103static unsigned r600_tex_dim(unsigned dim, unsigned nr_samples)
104{
105	switch (dim) {
106	default:
107	case PIPE_TEXTURE_1D:
108		return V_038000_SQ_TEX_DIM_1D;
109	case PIPE_TEXTURE_1D_ARRAY:
110		return V_038000_SQ_TEX_DIM_1D_ARRAY;
111	case PIPE_TEXTURE_2D:
112	case PIPE_TEXTURE_RECT:
113		return nr_samples > 1 ? V_038000_SQ_TEX_DIM_2D_MSAA :
114					V_038000_SQ_TEX_DIM_2D;
115	case PIPE_TEXTURE_2D_ARRAY:
116		return nr_samples > 1 ? V_038000_SQ_TEX_DIM_2D_ARRAY_MSAA :
117					V_038000_SQ_TEX_DIM_2D_ARRAY;
118	case PIPE_TEXTURE_3D:
119		return V_038000_SQ_TEX_DIM_3D;
120	case PIPE_TEXTURE_CUBE:
121	case PIPE_TEXTURE_CUBE_ARRAY:
122		return V_038000_SQ_TEX_DIM_CUBEMAP;
123	}
124}
125
126static uint32_t r600_translate_dbformat(enum pipe_format format)
127{
128	switch (format) {
129	case PIPE_FORMAT_Z16_UNORM:
130		return V_028010_DEPTH_16;
131	case PIPE_FORMAT_Z24X8_UNORM:
132		return V_028010_DEPTH_X8_24;
133	case PIPE_FORMAT_Z24_UNORM_S8_UINT:
134		return V_028010_DEPTH_8_24;
135	case PIPE_FORMAT_Z32_FLOAT:
136		return V_028010_DEPTH_32_FLOAT;
137	case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
138		return V_028010_DEPTH_X24_8_32_FLOAT;
139	default:
140		return ~0U;
141	}
142}
143
144static bool r600_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
145{
146	return r600_translate_texformat(screen, format, NULL, NULL, NULL,
147                                   FALSE) != ~0U;
148}
149
150static bool r600_is_colorbuffer_format_supported(enum amd_gfx_level chip, enum pipe_format format)
151{
152	return r600_translate_colorformat(chip, format, FALSE) != ~0U &&
153	       r600_translate_colorswap(format, FALSE) != ~0U;
154}
155
156static bool r600_is_zs_format_supported(enum pipe_format format)
157{
158	return r600_translate_dbformat(format) != ~0U;
159}
160
161bool r600_is_format_supported(struct pipe_screen *screen,
162			      enum pipe_format format,
163			      enum pipe_texture_target target,
164			      unsigned sample_count,
165			      unsigned storage_sample_count,
166			      unsigned usage)
167{
168	struct r600_screen *rscreen = (struct r600_screen*)screen;
169	unsigned retval = 0;
170
171	if (target >= PIPE_MAX_TEXTURE_TYPES) {
172		R600_ERR("r600: unsupported texture type %d\n", target);
173		return false;
174	}
175
176	if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
177		return false;
178
179	if (sample_count > 1) {
180		if (!rscreen->has_msaa)
181			return false;
182
183		/* R11G11B10 is broken on R6xx. */
184		if (rscreen->b.gfx_level == R600 &&
185		    format == PIPE_FORMAT_R11G11B10_FLOAT)
186			return false;
187
188		/* MSAA integer colorbuffers hang. */
189		if (util_format_is_pure_integer(format) &&
190		    !util_format_is_depth_or_stencil(format))
191			return false;
192
193		switch (sample_count) {
194		case 2:
195		case 4:
196		case 8:
197			break;
198		default:
199			return false;
200		}
201	}
202
203	if (usage & PIPE_BIND_SAMPLER_VIEW) {
204		if (target == PIPE_BUFFER) {
205			if (r600_is_buffer_format_supported(format, false))
206				retval |= PIPE_BIND_SAMPLER_VIEW;
207		} else {
208			if (r600_is_sampler_format_supported(screen, format))
209				retval |= PIPE_BIND_SAMPLER_VIEW;
210		}
211	}
212
213	if ((usage & (PIPE_BIND_RENDER_TARGET |
214		      PIPE_BIND_DISPLAY_TARGET |
215		      PIPE_BIND_SCANOUT |
216		      PIPE_BIND_SHARED |
217		      PIPE_BIND_BLENDABLE)) &&
218	    r600_is_colorbuffer_format_supported(rscreen->b.gfx_level, format)) {
219		retval |= usage &
220			  (PIPE_BIND_RENDER_TARGET |
221			   PIPE_BIND_DISPLAY_TARGET |
222			   PIPE_BIND_SCANOUT |
223			   PIPE_BIND_SHARED);
224		if (!util_format_is_pure_integer(format) &&
225		    !util_format_is_depth_or_stencil(format))
226			retval |= usage & PIPE_BIND_BLENDABLE;
227	}
228
229	if ((usage & PIPE_BIND_DEPTH_STENCIL) &&
230	    r600_is_zs_format_supported(format)) {
231		retval |= PIPE_BIND_DEPTH_STENCIL;
232	}
233
234	if ((usage & PIPE_BIND_VERTEX_BUFFER) &&
235	    r600_is_buffer_format_supported(format, true)) {
236		retval |= PIPE_BIND_VERTEX_BUFFER;
237	}
238
239	if (usage & PIPE_BIND_INDEX_BUFFER &&
240	    r600_is_index_format_supported(format)) {
241		retval |= PIPE_BIND_INDEX_BUFFER;
242	}
243
244	if ((usage & PIPE_BIND_LINEAR) &&
245	    !util_format_is_compressed(format) &&
246	    !(usage & PIPE_BIND_DEPTH_STENCIL))
247		retval |= PIPE_BIND_LINEAR;
248
249	return retval == usage;
250}
251
252static void r600_emit_polygon_offset(struct r600_context *rctx, struct r600_atom *a)
253{
254	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
255	struct r600_poly_offset_state *state = (struct r600_poly_offset_state*)a;
256	float offset_units = state->offset_units;
257	float offset_scale = state->offset_scale;
258	uint32_t pa_su_poly_offset_db_fmt_cntl = 0;
259
260	if (!state->offset_units_unscaled) {
261		switch (state->zs_format) {
262		case PIPE_FORMAT_Z24X8_UNORM:
263		case PIPE_FORMAT_Z24_UNORM_S8_UINT:
264			offset_units *= 2.0f;
265			pa_su_poly_offset_db_fmt_cntl =
266				S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-24);
267			break;
268		case PIPE_FORMAT_Z16_UNORM:
269			offset_units *= 4.0f;
270			pa_su_poly_offset_db_fmt_cntl =
271				S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-16);
272			break;
273		default:
274			pa_su_poly_offset_db_fmt_cntl =
275				S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-23) |
276				S_028DF8_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
277		}
278	}
279
280	radeon_set_context_reg_seq(cs, R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 4);
281	radeon_emit(cs, fui(offset_scale));
282	radeon_emit(cs, fui(offset_units));
283	radeon_emit(cs, fui(offset_scale));
284	radeon_emit(cs, fui(offset_units));
285
286	radeon_set_context_reg(cs, R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
287			       pa_su_poly_offset_db_fmt_cntl);
288}
289
290static uint32_t r600_get_blend_control(const struct pipe_blend_state *state, unsigned i)
291{
292	int j = state->independent_blend_enable ? i : 0;
293
294	unsigned eqRGB = state->rt[j].rgb_func;
295	unsigned srcRGB = state->rt[j].rgb_src_factor;
296	unsigned dstRGB = state->rt[j].rgb_dst_factor;
297
298	unsigned eqA = state->rt[j].alpha_func;
299	unsigned srcA = state->rt[j].alpha_src_factor;
300	unsigned dstA = state->rt[j].alpha_dst_factor;
301	uint32_t bc = 0;
302
303	if (!state->rt[j].blend_enable)
304		return 0;
305
306	bc |= S_028804_COLOR_COMB_FCN(r600_translate_blend_function(eqRGB));
307	bc |= S_028804_COLOR_SRCBLEND(r600_translate_blend_factor(srcRGB));
308	bc |= S_028804_COLOR_DESTBLEND(r600_translate_blend_factor(dstRGB));
309
310	if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
311		bc |= S_028804_SEPARATE_ALPHA_BLEND(1);
312		bc |= S_028804_ALPHA_COMB_FCN(r600_translate_blend_function(eqA));
313		bc |= S_028804_ALPHA_SRCBLEND(r600_translate_blend_factor(srcA));
314		bc |= S_028804_ALPHA_DESTBLEND(r600_translate_blend_factor(dstA));
315	}
316	return bc;
317}
318
319static void *r600_create_blend_state_mode(struct pipe_context *ctx,
320					  const struct pipe_blend_state *state,
321					  int mode)
322{
323	struct r600_context *rctx = (struct r600_context *)ctx;
324	uint32_t color_control = 0, target_mask = 0;
325	struct r600_blend_state *blend = CALLOC_STRUCT(r600_blend_state);
326
327	if (!blend) {
328		return NULL;
329	}
330
331	r600_init_command_buffer(&blend->buffer, 20);
332	r600_init_command_buffer(&blend->buffer_no_blend, 20);
333
334	/* R600 does not support per-MRT blends */
335	if (rctx->b.family > CHIP_R600)
336		color_control |= S_028808_PER_MRT_BLEND(1);
337
338	if (state->logicop_enable) {
339		color_control |= (state->logicop_func << 16) | (state->logicop_func << 20);
340	} else {
341		color_control |= (0xcc << 16);
342	}
343	/* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
344	if (state->independent_blend_enable) {
345		for (int i = 0; i < 8; i++) {
346			if (state->rt[i].blend_enable) {
347				color_control |= S_028808_TARGET_BLEND_ENABLE(1 << i);
348			}
349			target_mask |= (state->rt[i].colormask << (4 * i));
350		}
351	} else {
352		for (int i = 0; i < 8; i++) {
353			if (state->rt[0].blend_enable) {
354				color_control |= S_028808_TARGET_BLEND_ENABLE(1 << i);
355			}
356			target_mask |= (state->rt[0].colormask << (4 * i));
357		}
358	}
359
360	if (target_mask)
361		color_control |= S_028808_SPECIAL_OP(mode);
362	else
363		color_control |= S_028808_SPECIAL_OP(V_028808_DISABLE);
364
365	/* only MRT0 has dual src blend */
366	blend->dual_src_blend = util_blend_state_is_dual(state, 0);
367	blend->cb_target_mask = target_mask;
368	blend->cb_color_control = color_control;
369	blend->cb_color_control_no_blend = color_control & C_028808_TARGET_BLEND_ENABLE;
370	blend->alpha_to_one = state->alpha_to_one;
371
372	r600_store_context_reg(&blend->buffer, R_028D44_DB_ALPHA_TO_MASK,
373			       S_028D44_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
374			       S_028D44_ALPHA_TO_MASK_OFFSET0(2) |
375			       S_028D44_ALPHA_TO_MASK_OFFSET1(2) |
376			       S_028D44_ALPHA_TO_MASK_OFFSET2(2) |
377			       S_028D44_ALPHA_TO_MASK_OFFSET3(2));
378
379	/* Copy over the registers set so far into buffer_no_blend. */
380	memcpy(blend->buffer_no_blend.buf, blend->buffer.buf, blend->buffer.num_dw * 4);
381	blend->buffer_no_blend.num_dw = blend->buffer.num_dw;
382
383	/* Only add blend registers if blending is enabled. */
384	if (!G_028808_TARGET_BLEND_ENABLE(color_control)) {
385		return blend;
386	}
387
388	/* The first R600 does not support per-MRT blends */
389	r600_store_context_reg(&blend->buffer, R_028804_CB_BLEND_CONTROL,
390			       r600_get_blend_control(state, 0));
391
392	if (rctx->b.family > CHIP_R600) {
393		r600_store_context_reg_seq(&blend->buffer, R_028780_CB_BLEND0_CONTROL, 8);
394		for (int i = 0; i < 8; i++) {
395			r600_store_value(&blend->buffer, r600_get_blend_control(state, i));
396		}
397	}
398	return blend;
399}
400
401static void *r600_create_blend_state(struct pipe_context *ctx,
402				     const struct pipe_blend_state *state)
403{
404	return r600_create_blend_state_mode(ctx, state, V_028808_SPECIAL_NORMAL);
405}
406
407static void *r600_create_dsa_state(struct pipe_context *ctx,
408				   const struct pipe_depth_stencil_alpha_state *state)
409{
410	unsigned db_depth_control, alpha_test_control, alpha_ref;
411	struct r600_dsa_state *dsa = CALLOC_STRUCT(r600_dsa_state);
412
413	if (!dsa) {
414		return NULL;
415	}
416
417	r600_init_command_buffer(&dsa->buffer, 3);
418
419	dsa->valuemask[0] = state->stencil[0].valuemask;
420	dsa->valuemask[1] = state->stencil[1].valuemask;
421	dsa->writemask[0] = state->stencil[0].writemask;
422	dsa->writemask[1] = state->stencil[1].writemask;
423	dsa->zwritemask = state->depth_writemask;
424
425	db_depth_control = S_028800_Z_ENABLE(state->depth_enabled) |
426		S_028800_Z_WRITE_ENABLE(state->depth_writemask) |
427		S_028800_ZFUNC(state->depth_func);
428
429	/* stencil */
430	if (state->stencil[0].enabled) {
431		db_depth_control |= S_028800_STENCIL_ENABLE(1);
432		db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func); /* translates straight */
433		db_depth_control |= S_028800_STENCILFAIL(r600_translate_stencil_op(state->stencil[0].fail_op));
434		db_depth_control |= S_028800_STENCILZPASS(r600_translate_stencil_op(state->stencil[0].zpass_op));
435		db_depth_control |= S_028800_STENCILZFAIL(r600_translate_stencil_op(state->stencil[0].zfail_op));
436
437		if (state->stencil[1].enabled) {
438			db_depth_control |= S_028800_BACKFACE_ENABLE(1);
439			db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func); /* translates straight */
440			db_depth_control |= S_028800_STENCILFAIL_BF(r600_translate_stencil_op(state->stencil[1].fail_op));
441			db_depth_control |= S_028800_STENCILZPASS_BF(r600_translate_stencil_op(state->stencil[1].zpass_op));
442			db_depth_control |= S_028800_STENCILZFAIL_BF(r600_translate_stencil_op(state->stencil[1].zfail_op));
443		}
444	}
445
446	/* alpha */
447	alpha_test_control = 0;
448	alpha_ref = 0;
449	if (state->alpha_enabled) {
450		alpha_test_control = S_028410_ALPHA_FUNC(state->alpha_func);
451		alpha_test_control |= S_028410_ALPHA_TEST_ENABLE(1);
452		alpha_ref = fui(state->alpha_ref_value);
453	}
454	dsa->sx_alpha_test_control = alpha_test_control & 0xff;
455	dsa->alpha_ref = alpha_ref;
456
457	r600_store_context_reg(&dsa->buffer, R_028800_DB_DEPTH_CONTROL, db_depth_control);
458	return dsa;
459}
460
461static void *r600_create_rs_state(struct pipe_context *ctx,
462				  const struct pipe_rasterizer_state *state)
463{
464	struct r600_context *rctx = (struct r600_context *)ctx;
465	unsigned tmp, sc_mode_cntl, spi_interp;
466	float psize_min, psize_max;
467	struct r600_rasterizer_state *rs = CALLOC_STRUCT(r600_rasterizer_state);
468
469	if (!rs) {
470		return NULL;
471	}
472
473	r600_init_command_buffer(&rs->buffer, 30);
474
475	rs->scissor_enable = state->scissor;
476	rs->clip_halfz = state->clip_halfz;
477	rs->flatshade = state->flatshade;
478	rs->sprite_coord_enable = state->sprite_coord_enable;
479	rs->rasterizer_discard = state->rasterizer_discard;
480	rs->two_side = state->light_twoside;
481	rs->clip_plane_enable = state->clip_plane_enable;
482	rs->pa_sc_line_stipple = state->line_stipple_enable ?
483				S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
484				S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
485	rs->pa_cl_clip_cntl =
486		S_028810_DX_CLIP_SPACE_DEF(state->clip_halfz) |
487		S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip_near) |
488		S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip_far) |
489		S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
490	if (rctx->b.gfx_level == R700) {
491		rs->pa_cl_clip_cntl |=
492			S_028810_DX_RASTERIZATION_KILL(state->rasterizer_discard);
493	}
494	rs->multisample_enable = state->multisample;
495
496	/* offset */
497	rs->offset_units = state->offset_units;
498	rs->offset_scale = state->offset_scale * 16.0f;
499	rs->offset_enable = state->offset_point || state->offset_line || state->offset_tri;
500	rs->offset_units_unscaled = state->offset_units_unscaled;
501
502	if (state->point_size_per_vertex) {
503		psize_min = util_get_min_point_size(state);
504		psize_max = 8192;
505	} else {
506		/* Force the point size to be as if the vertex output was disabled. */
507		psize_min = state->point_size;
508		psize_max = state->point_size;
509	}
510
511	sc_mode_cntl = S_028A4C_MSAA_ENABLE(state->multisample) |
512		       S_028A4C_LINE_STIPPLE_ENABLE(state->line_stipple_enable) |
513		       S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) |
514		       S_028A4C_PS_ITER_SAMPLE(state->multisample && rctx->ps_iter_samples > 1);
515	if (rctx->b.family == CHIP_RV770) {
516		/* workaround possible rendering corruption on RV770 with hyperz together with sample shading */
517		sc_mode_cntl |= S_028A4C_TILE_COVER_DISABLE(state->multisample && rctx->ps_iter_samples > 1);
518	}
519	if (rctx->b.gfx_level >= R700) {
520		sc_mode_cntl |= S_028A4C_FORCE_EOV_REZ_ENABLE(1) |
521				S_028A4C_R700_ZMM_LINE_OFFSET(1) |
522				S_028A4C_R700_VPORT_SCISSOR_ENABLE(1);
523	} else {
524		sc_mode_cntl |= S_028A4C_WALK_ALIGN8_PRIM_FITS_ST(1);
525	}
526
527	spi_interp = S_0286D4_FLAT_SHADE_ENA(1);
528	spi_interp |= S_0286D4_PNT_SPRITE_ENA(1) |
529		S_0286D4_PNT_SPRITE_OVRD_X(2) |
530		S_0286D4_PNT_SPRITE_OVRD_Y(3) |
531		S_0286D4_PNT_SPRITE_OVRD_Z(0) |
532		S_0286D4_PNT_SPRITE_OVRD_W(1);
533	if (state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT) {
534		spi_interp |= S_0286D4_PNT_SPRITE_TOP_1(1);
535	}
536
537	r600_store_context_reg_seq(&rs->buffer, R_028A00_PA_SU_POINT_SIZE, 3);
538	/* point size 12.4 fixed point (divide by two, because 0.5 = 1 pixel. */
539	tmp = r600_pack_float_12p4(state->point_size/2);
540	r600_store_value(&rs->buffer, /* R_028A00_PA_SU_POINT_SIZE */
541			 S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
542	r600_store_value(&rs->buffer, /* R_028A04_PA_SU_POINT_MINMAX */
543			 S_028A04_MIN_SIZE(r600_pack_float_12p4(psize_min/2)) |
544			 S_028A04_MAX_SIZE(r600_pack_float_12p4(psize_max/2)));
545	r600_store_value(&rs->buffer, /* R_028A08_PA_SU_LINE_CNTL */
546			 S_028A08_WIDTH(r600_pack_float_12p4(state->line_width/2)));
547
548	r600_store_context_reg(&rs->buffer, R_0286D4_SPI_INTERP_CONTROL_0, spi_interp);
549	r600_store_context_reg(&rs->buffer, R_028A4C_PA_SC_MODE_CNTL, sc_mode_cntl);
550	r600_store_context_reg(&rs->buffer, R_028C08_PA_SU_VTX_CNTL,
551			       S_028C08_PIX_CENTER_HALF(state->half_pixel_center) |
552			       S_028C08_QUANT_MODE(V_028C08_X_1_256TH));
553	r600_store_context_reg(&rs->buffer, R_028DFC_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp));
554
555	rs->pa_su_sc_mode_cntl = S_028814_PROVOKING_VTX_LAST(!state->flatshade_first) |
556				 S_028814_CULL_FRONT(state->cull_face & PIPE_FACE_FRONT ? 1 : 0) |
557				 S_028814_CULL_BACK(state->cull_face & PIPE_FACE_BACK ? 1 : 0) |
558				 S_028814_FACE(!state->front_ccw) |
559				 S_028814_POLY_OFFSET_FRONT_ENABLE(util_get_offset(state, state->fill_front)) |
560				 S_028814_POLY_OFFSET_BACK_ENABLE(util_get_offset(state, state->fill_back)) |
561				 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_point || state->offset_line) |
562				 S_028814_POLY_MODE(state->fill_front != PIPE_POLYGON_MODE_FILL ||
563									 state->fill_back != PIPE_POLYGON_MODE_FILL) |
564				 S_028814_POLYMODE_FRONT_PTYPE(r600_translate_fill(state->fill_front)) |
565				 S_028814_POLYMODE_BACK_PTYPE(r600_translate_fill(state->fill_back));
566	if (rctx->b.gfx_level == R700) {
567		r600_store_context_reg(&rs->buffer, R_028814_PA_SU_SC_MODE_CNTL, rs->pa_su_sc_mode_cntl);
568	}
569	if (rctx->b.gfx_level == R600) {
570		r600_store_context_reg(&rs->buffer, R_028350_SX_MISC,
571				       S_028350_MULTIPASS(state->rasterizer_discard));
572	}
573	return rs;
574}
575
576static unsigned r600_tex_filter(unsigned filter, unsigned max_aniso)
577{
578	if (filter == PIPE_TEX_FILTER_LINEAR)
579		return max_aniso > 1 ? V_03C000_SQ_TEX_XY_FILTER_ANISO_BILINEAR
580				     : V_03C000_SQ_TEX_XY_FILTER_BILINEAR;
581	else
582		return max_aniso > 1 ? V_03C000_SQ_TEX_XY_FILTER_ANISO_POINT
583				     : V_03C000_SQ_TEX_XY_FILTER_POINT;
584}
585
586static void *r600_create_sampler_state(struct pipe_context *ctx,
587					const struct pipe_sampler_state *state)
588{
589	struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen;
590	struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state);
591	unsigned max_aniso = rscreen->force_aniso >= 0 ? rscreen->force_aniso
592						       : state->max_anisotropy;
593	unsigned max_aniso_ratio = r600_tex_aniso_filter(max_aniso);
594
595	if (!ss) {
596		return NULL;
597	}
598
599	ss->seamless_cube_map = state->seamless_cube_map;
600	ss->border_color_use = sampler_state_needs_border_color(state);
601
602	/* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */
603	ss->tex_sampler_words[0] =
604		S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
605		S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
606		S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
607		S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter, max_aniso)) |
608		S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter, max_aniso)) |
609		S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
610		S_03C000_MAX_ANISO_RATIO(max_aniso_ratio) |
611		S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
612		S_03C000_BORDER_COLOR_TYPE(ss->border_color_use ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0);
613	/* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */
614	ss->tex_sampler_words[1] =
615		S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 6)) |
616		S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 6)) |
617		S_03C004_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 6));
618	/* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */
619	ss->tex_sampler_words[2] = S_03C008_TYPE(1);
620
621	if (ss->border_color_use) {
622		memcpy(&ss->border_color, &state->border_color, sizeof(state->border_color));
623	}
624	return ss;
625}
626
627static struct pipe_sampler_view *
628texture_buffer_sampler_view(struct r600_pipe_sampler_view *view,
629			    unsigned width0, unsigned height0)
630
631{
632	struct r600_texture *tmp = (struct r600_texture*)view->base.texture;
633	int stride = util_format_get_blocksize(view->base.format);
634	unsigned format, num_format, format_comp, endian;
635	uint64_t offset = view->base.u.buf.offset;
636	unsigned size = view->base.u.buf.size;
637
638	r600_vertex_data_type(view->base.format,
639			      &format, &num_format, &format_comp,
640			      &endian);
641
642	view->tex_resource = &tmp->resource;
643	view->skip_mip_address_reloc = true;
644
645	view->tex_resource_words[0] = offset;
646	view->tex_resource_words[1] = size - 1;
647	view->tex_resource_words[2] = S_038008_BASE_ADDRESS_HI(offset >> 32UL) |
648		S_038008_STRIDE(stride) |
649		S_038008_DATA_FORMAT(format) |
650		S_038008_NUM_FORMAT_ALL(num_format) |
651		S_038008_FORMAT_COMP_ALL(format_comp) |
652		S_038008_ENDIAN_SWAP(endian);
653	view->tex_resource_words[3] = 0;
654	/*
655	 * in theory dword 4 is for number of elements, for use with resinfo,
656	 * but it seems to utterly fail to work, the amd gpu shader analyser
657	 * uses a const buffer to store the element sizes for buffer txq
658	 */
659	view->tex_resource_words[4] = 0;
660	view->tex_resource_words[5] = 0;
661	view->tex_resource_words[6] = S_038018_TYPE(V_038010_SQ_TEX_VTX_VALID_BUFFER);
662	return &view->base;
663}
664
665struct pipe_sampler_view *
666r600_create_sampler_view_custom(struct pipe_context *ctx,
667				struct pipe_resource *texture,
668				const struct pipe_sampler_view *state,
669				unsigned width_first_level, unsigned height_first_level)
670{
671	struct r600_pipe_sampler_view *view = CALLOC_STRUCT(r600_pipe_sampler_view);
672	struct r600_texture *tmp = (struct r600_texture*)texture;
673	unsigned format, endian;
674	uint32_t word4 = 0, yuv_format = 0, pitch = 0;
675	unsigned char swizzle[4], array_mode = 0;
676	unsigned width, height, depth, offset_level, last_level;
677	bool do_endian_swap = FALSE;
678
679	if (!view)
680		return NULL;
681
682	/* initialize base object */
683	view->base = *state;
684	view->base.texture = NULL;
685	pipe_reference(NULL, &texture->reference);
686	view->base.texture = texture;
687	view->base.reference.count = 1;
688	view->base.context = ctx;
689
690	if (texture->target == PIPE_BUFFER)
691		return texture_buffer_sampler_view(view, texture->width0, 1);
692
693	swizzle[0] = state->swizzle_r;
694	swizzle[1] = state->swizzle_g;
695	swizzle[2] = state->swizzle_b;
696	swizzle[3] = state->swizzle_a;
697
698	if (R600_BIG_ENDIAN)
699		do_endian_swap = !tmp->db_compatible;
700
701	format = r600_translate_texformat(ctx->screen, state->format,
702					  swizzle,
703					  &word4, &yuv_format, do_endian_swap);
704	assert(format != ~0);
705	if (format == ~0) {
706		FREE(view);
707		return NULL;
708	}
709
710	if (state->format == PIPE_FORMAT_X24S8_UINT ||
711	    state->format == PIPE_FORMAT_S8X24_UINT ||
712	    state->format == PIPE_FORMAT_X32_S8X24_UINT ||
713	    state->format == PIPE_FORMAT_S8_UINT)
714		view->is_stencil_sampler = true;
715
716	if (tmp->is_depth && !r600_can_sample_zs(tmp, view->is_stencil_sampler)) {
717		if (!r600_init_flushed_depth_texture(ctx, texture, NULL)) {
718			FREE(view);
719			return NULL;
720		}
721		tmp = tmp->flushed_depth_texture;
722	}
723
724	endian = r600_colorformat_endian_swap(format, do_endian_swap);
725
726	offset_level = state->u.tex.first_level;
727	last_level = state->u.tex.last_level - offset_level;
728	width = width_first_level;
729	height = height_first_level;
730        depth = u_minify(texture->depth0, offset_level);
731	pitch = tmp->surface.u.legacy.level[offset_level].nblk_x * util_format_get_blockwidth(state->format);
732
733	if (texture->target == PIPE_TEXTURE_1D_ARRAY) {
734		height = 1;
735		depth = texture->array_size;
736	} else if (texture->target == PIPE_TEXTURE_2D_ARRAY) {
737		depth = texture->array_size;
738	} else if (texture->target == PIPE_TEXTURE_CUBE_ARRAY)
739		depth = texture->array_size / 6;
740
741	switch (tmp->surface.u.legacy.level[offset_level].mode) {
742	default:
743	case RADEON_SURF_MODE_LINEAR_ALIGNED:
744		array_mode = V_038000_ARRAY_LINEAR_ALIGNED;
745		break;
746	case RADEON_SURF_MODE_1D:
747		array_mode = V_038000_ARRAY_1D_TILED_THIN1;
748		break;
749	case RADEON_SURF_MODE_2D:
750		array_mode = V_038000_ARRAY_2D_TILED_THIN1;
751		break;
752	}
753
754	view->tex_resource = &tmp->resource;
755	view->tex_resource_words[0] = (S_038000_DIM(r600_tex_dim(texture->target, texture->nr_samples)) |
756				       S_038000_TILE_MODE(array_mode) |
757				       S_038000_TILE_TYPE(tmp->non_disp_tiling) |
758				       S_038000_PITCH((pitch / 8) - 1) |
759				       S_038000_TEX_WIDTH(width - 1));
760	view->tex_resource_words[1] = (S_038004_TEX_HEIGHT(height - 1) |
761				       S_038004_TEX_DEPTH(depth - 1) |
762				       S_038004_DATA_FORMAT(format));
763	view->tex_resource_words[2] = tmp->surface.u.legacy.level[offset_level].offset_256B;
764	if (offset_level >= tmp->resource.b.b.last_level) {
765		view->tex_resource_words[3] = tmp->surface.u.legacy.level[offset_level].offset_256B;
766	} else {
767		view->tex_resource_words[3] = tmp->surface.u.legacy.level[offset_level + 1].offset_256B;
768	}
769	view->tex_resource_words[4] = (word4 |
770				       S_038010_REQUEST_SIZE(1) |
771				       S_038010_ENDIAN_SWAP(endian) |
772				       S_038010_BASE_LEVEL(0));
773	view->tex_resource_words[5] = (S_038014_BASE_ARRAY(state->u.tex.first_layer) |
774				       S_038014_LAST_ARRAY(state->u.tex.last_layer));
775	if (texture->nr_samples > 1) {
776		/* LAST_LEVEL holds log2(nr_samples) for multisample textures */
777		view->tex_resource_words[5] |= S_038014_LAST_LEVEL(util_logbase2(texture->nr_samples));
778	} else {
779		view->tex_resource_words[5] |= S_038014_LAST_LEVEL(last_level);
780	}
781	view->tex_resource_words[6] = (S_038018_TYPE(V_038010_SQ_TEX_VTX_VALID_TEXTURE) |
782				       S_038018_MAX_ANISO(4 /* max 16 samples */));
783	return &view->base;
784}
785
786static struct pipe_sampler_view *
787r600_create_sampler_view(struct pipe_context *ctx,
788			 struct pipe_resource *tex,
789			 const struct pipe_sampler_view *state)
790{
791	return r600_create_sampler_view_custom(ctx, tex, state,
792                                               u_minify(tex->width0, state->u.tex.first_level),
793                                               u_minify(tex->height0, state->u.tex.first_level));
794}
795
796static void r600_emit_clip_state(struct r600_context *rctx, struct r600_atom *atom)
797{
798	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
799	struct pipe_clip_state *state = &rctx->clip_state.state;
800
801	radeon_set_context_reg_seq(cs, R_028E20_PA_CL_UCP0_X, 6*4);
802	radeon_emit_array(cs, (unsigned*)state, 6*4);
803}
804
805static void r600_set_polygon_stipple(struct pipe_context *ctx,
806					 const struct pipe_poly_stipple *state)
807{
808}
809
810static void r600_init_color_surface(struct r600_context *rctx,
811				    struct r600_surface *surf,
812				    bool force_cmask_fmask)
813{
814	struct r600_screen *rscreen = rctx->screen;
815	struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
816	unsigned level = surf->base.u.tex.level;
817	unsigned pitch, slice;
818	unsigned color_info;
819	unsigned color_view;
820	unsigned format, swap, ntype, endian;
821	unsigned offset;
822	const struct util_format_description *desc;
823	int i;
824	bool blend_bypass = 0, blend_clamp = 0, do_endian_swap = FALSE;
825
826	if (rtex->db_compatible && !r600_can_sample_zs(rtex, false)) {
827		r600_init_flushed_depth_texture(&rctx->b.b, surf->base.texture, NULL);
828		rtex = rtex->flushed_depth_texture;
829		assert(rtex);
830	}
831
832	offset = (uint64_t)rtex->surface.u.legacy.level[level].offset_256B * 256;
833	color_view = S_028080_SLICE_START(surf->base.u.tex.first_layer) |
834		     S_028080_SLICE_MAX(surf->base.u.tex.last_layer);
835
836	pitch = rtex->surface.u.legacy.level[level].nblk_x / 8 - 1;
837	slice = (rtex->surface.u.legacy.level[level].nblk_x * rtex->surface.u.legacy.level[level].nblk_y) / 64;
838	if (slice) {
839		slice = slice - 1;
840	}
841	color_info = 0;
842	switch (rtex->surface.u.legacy.level[level].mode) {
843	default:
844	case RADEON_SURF_MODE_LINEAR_ALIGNED:
845		color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_LINEAR_ALIGNED);
846		break;
847	case RADEON_SURF_MODE_1D:
848		color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_1D_TILED_THIN1);
849		break;
850	case RADEON_SURF_MODE_2D:
851		color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_2D_TILED_THIN1);
852		break;
853	}
854
855	desc = util_format_description(surf->base.format);
856
857	for (i = 0; i < 4; i++) {
858		if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
859			break;
860		}
861	}
862
863	ntype = V_0280A0_NUMBER_UNORM;
864	if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
865		ntype = V_0280A0_NUMBER_SRGB;
866	else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
867		if (desc->channel[i].normalized)
868			ntype = V_0280A0_NUMBER_SNORM;
869		else if (desc->channel[i].pure_integer)
870			ntype = V_0280A0_NUMBER_SINT;
871	} else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
872		if (desc->channel[i].normalized)
873			ntype = V_0280A0_NUMBER_UNORM;
874		else if (desc->channel[i].pure_integer)
875			ntype = V_0280A0_NUMBER_UINT;
876	} else if (desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT) {
877		ntype = V_0280A0_NUMBER_FLOAT;
878	}
879
880	if (R600_BIG_ENDIAN)
881		do_endian_swap = !rtex->db_compatible;
882
883	format = r600_translate_colorformat(rctx->b.gfx_level, surf->base.format,
884			                              do_endian_swap);
885	assert(format != ~0);
886
887	swap = r600_translate_colorswap(surf->base.format, do_endian_swap);
888	assert(swap != ~0);
889
890	endian = r600_colorformat_endian_swap(format, do_endian_swap);
891
892	/* blend clamp should be set for all NORM/SRGB types */
893	if (ntype == V_0280A0_NUMBER_UNORM || ntype == V_0280A0_NUMBER_SNORM ||
894	    ntype == V_0280A0_NUMBER_SRGB)
895		blend_clamp = 1;
896
897	/* set blend bypass according to docs if SINT/UINT or
898	   8/24 COLOR variants */
899	if (ntype == V_0280A0_NUMBER_UINT || ntype == V_0280A0_NUMBER_SINT ||
900	    format == V_0280A0_COLOR_8_24 || format == V_0280A0_COLOR_24_8 ||
901	    format == V_0280A0_COLOR_X24_8_32_FLOAT) {
902		blend_clamp = 0;
903		blend_bypass = 1;
904	}
905
906	surf->alphatest_bypass = ntype == V_0280A0_NUMBER_UINT || ntype == V_0280A0_NUMBER_SINT;
907
908	color_info |= S_0280A0_FORMAT(format) |
909		S_0280A0_COMP_SWAP(swap) |
910		S_0280A0_BLEND_BYPASS(blend_bypass) |
911		S_0280A0_BLEND_CLAMP(blend_clamp) |
912		S_0280A0_SIMPLE_FLOAT(1) |
913		S_0280A0_NUMBER_TYPE(ntype) |
914		S_0280A0_ENDIAN(endian);
915
916	/* EXPORT_NORM is an optimization that can be enabled for better
917	 * performance in certain cases
918	 */
919	if (rctx->b.gfx_level == R600) {
920		/* EXPORT_NORM can be enabled if:
921		 * - 11-bit or smaller UNORM/SNORM/SRGB
922		 * - BLEND_CLAMP is enabled
923		 * - BLEND_FLOAT32 is disabled
924		 */
925		if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
926		    (desc->channel[i].size < 12 &&
927		     desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
928		     ntype != V_0280A0_NUMBER_UINT &&
929		     ntype != V_0280A0_NUMBER_SINT) &&
930		    G_0280A0_BLEND_CLAMP(color_info) &&
931		    /* XXX this condition is always true since BLEND_FLOAT32 is never set (bug?). */
932		    !G_0280A0_BLEND_FLOAT32(color_info)) {
933			color_info |= S_0280A0_SOURCE_FORMAT(V_0280A0_EXPORT_NORM);
934			surf->export_16bpc = true;
935		}
936	} else {
937		/* EXPORT_NORM can be enabled if:
938		 * - 11-bit or smaller UNORM/SNORM/SRGB
939		 * - 16-bit or smaller FLOAT
940		 */
941		if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
942		    ((desc->channel[i].size < 12 &&
943		      desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
944		      ntype != V_0280A0_NUMBER_UINT && ntype != V_0280A0_NUMBER_SINT) ||
945		    (desc->channel[i].size < 17 &&
946		     desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT))) {
947			color_info |= S_0280A0_SOURCE_FORMAT(V_0280A0_EXPORT_NORM);
948			surf->export_16bpc = true;
949		}
950	}
951
952	/* These might not always be initialized to zero. */
953	surf->cb_color_base = offset >> 8;
954	surf->cb_color_size = S_028060_PITCH_TILE_MAX(pitch) |
955			      S_028060_SLICE_TILE_MAX(slice);
956	surf->cb_color_fmask = surf->cb_color_base;
957	surf->cb_color_cmask = surf->cb_color_base;
958	surf->cb_color_mask = 0;
959
960	r600_resource_reference(&surf->cb_buffer_cmask, &rtex->resource);
961	r600_resource_reference(&surf->cb_buffer_fmask, &rtex->resource);
962
963	if (rtex->cmask.size) {
964		surf->cb_color_cmask = rtex->cmask.offset >> 8;
965		surf->cb_color_mask |= S_028100_CMASK_BLOCK_MAX(rtex->cmask.slice_tile_max);
966
967		if (rtex->fmask.size) {
968			color_info |= S_0280A0_TILE_MODE(V_0280A0_FRAG_ENABLE);
969			surf->cb_color_fmask = rtex->fmask.offset >> 8;
970			surf->cb_color_mask |= S_028100_FMASK_TILE_MAX(rtex->fmask.slice_tile_max);
971		} else { /* cmask only */
972			color_info |= S_0280A0_TILE_MODE(V_0280A0_CLEAR_ENABLE);
973		}
974	} else if (force_cmask_fmask) {
975		/* Allocate dummy FMASK and CMASK if they aren't allocated already.
976		 *
977		 * R6xx needs FMASK and CMASK for the destination buffer of color resolve,
978		 * otherwise it hangs. We don't have FMASK and CMASK pre-allocated,
979		 * because it's not an MSAA buffer.
980		 */
981		struct r600_cmask_info cmask;
982		struct r600_fmask_info fmask;
983
984		r600_texture_get_cmask_info(&rscreen->b, rtex, &cmask);
985		r600_texture_get_fmask_info(&rscreen->b, rtex, 8, &fmask);
986
987		/* CMASK. */
988		if (!rctx->dummy_cmask ||
989		    rctx->dummy_cmask->b.b.width0 < cmask.size ||
990		    (1 << rctx->dummy_cmask->buf->alignment_log2) % cmask.alignment != 0) {
991			struct pipe_transfer *transfer;
992			void *ptr;
993
994			r600_resource_reference(&rctx->dummy_cmask, NULL);
995			rctx->dummy_cmask = (struct r600_resource*)
996				r600_aligned_buffer_create(&rscreen->b.b, 0,
997							   PIPE_USAGE_DEFAULT,
998							   cmask.size, cmask.alignment);
999
1000			if (unlikely(!rctx->dummy_cmask)) {
1001				surf->color_initialized = false;
1002				return;
1003			}
1004
1005			/* Set the contents to 0xCC. */
1006			ptr = pipe_buffer_map(&rctx->b.b, &rctx->dummy_cmask->b.b, PIPE_MAP_WRITE, &transfer);
1007			memset(ptr, 0xCC, cmask.size);
1008			pipe_buffer_unmap(&rctx->b.b, transfer);
1009		}
1010		r600_resource_reference(&surf->cb_buffer_cmask, rctx->dummy_cmask);
1011
1012		/* FMASK. */
1013		if (!rctx->dummy_fmask ||
1014		    rctx->dummy_fmask->b.b.width0 < fmask.size ||
1015		    (1 << rctx->dummy_fmask->buf->alignment_log2) % fmask.alignment != 0) {
1016			r600_resource_reference(&rctx->dummy_fmask, NULL);
1017			rctx->dummy_fmask = (struct r600_resource*)
1018				r600_aligned_buffer_create(&rscreen->b.b, 0,
1019							   PIPE_USAGE_DEFAULT,
1020							   fmask.size, fmask.alignment);
1021
1022			if (unlikely(!rctx->dummy_fmask)) {
1023				surf->color_initialized = false;
1024				return;
1025			}
1026		}
1027		r600_resource_reference(&surf->cb_buffer_fmask, rctx->dummy_fmask);
1028
1029		/* Init the registers. */
1030		color_info |= S_0280A0_TILE_MODE(V_0280A0_FRAG_ENABLE);
1031		surf->cb_color_cmask = 0;
1032		surf->cb_color_fmask = 0;
1033		surf->cb_color_mask = S_028100_CMASK_BLOCK_MAX(cmask.slice_tile_max) |
1034				      S_028100_FMASK_TILE_MAX(fmask.slice_tile_max);
1035	}
1036
1037	surf->cb_color_info = color_info;
1038	surf->cb_color_view = color_view;
1039	surf->color_initialized = true;
1040}
1041
1042static void r600_init_depth_surface(struct r600_context *rctx,
1043				    struct r600_surface *surf)
1044{
1045	struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1046	unsigned level, pitch, slice, format, offset, array_mode;
1047
1048	level = surf->base.u.tex.level;
1049	offset = (uint64_t)rtex->surface.u.legacy.level[level].offset_256B * 256;
1050	pitch = rtex->surface.u.legacy.level[level].nblk_x / 8 - 1;
1051	slice = (rtex->surface.u.legacy.level[level].nblk_x * rtex->surface.u.legacy.level[level].nblk_y) / 64;
1052	if (slice) {
1053		slice = slice - 1;
1054	}
1055	switch (rtex->surface.u.legacy.level[level].mode) {
1056	case RADEON_SURF_MODE_2D:
1057		array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
1058		break;
1059	case RADEON_SURF_MODE_1D:
1060	case RADEON_SURF_MODE_LINEAR_ALIGNED:
1061	default:
1062		array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
1063		break;
1064	}
1065
1066	format = r600_translate_dbformat(surf->base.format);
1067	assert(format != ~0);
1068
1069	surf->db_depth_info = S_028010_ARRAY_MODE(array_mode) | S_028010_FORMAT(format);
1070	surf->db_depth_base = offset >> 8;
1071	surf->db_depth_view = S_028004_SLICE_START(surf->base.u.tex.first_layer) |
1072			      S_028004_SLICE_MAX(surf->base.u.tex.last_layer);
1073	surf->db_depth_size = S_028000_PITCH_TILE_MAX(pitch) | S_028000_SLICE_TILE_MAX(slice);
1074	surf->db_prefetch_limit = (rtex->surface.u.legacy.level[level].nblk_y / 8) - 1;
1075
1076	if (r600_htile_enabled(rtex, level)) {
1077		surf->db_htile_data_base = rtex->htile_offset >> 8;
1078		surf->db_htile_surface = S_028D24_HTILE_WIDTH(1) |
1079					 S_028D24_HTILE_HEIGHT(1) |
1080					 S_028D24_FULL_CACHE(1);
1081		/* preload is not working properly on r6xx/r7xx */
1082		surf->db_depth_info |= S_028010_TILE_SURFACE_ENABLE(1);
1083	}
1084
1085	surf->depth_initialized = true;
1086}
1087
1088static void r600_set_framebuffer_state(struct pipe_context *ctx,
1089					const struct pipe_framebuffer_state *state)
1090{
1091	struct r600_context *rctx = (struct r600_context *)ctx;
1092	struct r600_surface *surf;
1093	struct r600_texture *rtex;
1094	unsigned i;
1095	uint32_t target_mask = 0;
1096
1097	/* Flush TC when changing the framebuffer state, because the only
1098	 * client not using TC that can change textures is the framebuffer.
1099	 * Other places don't typically have to flush TC.
1100	 */
1101	rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE |
1102			 R600_CONTEXT_FLUSH_AND_INV |
1103			 R600_CONTEXT_FLUSH_AND_INV_CB |
1104			 R600_CONTEXT_FLUSH_AND_INV_CB_META |
1105			 R600_CONTEXT_FLUSH_AND_INV_DB |
1106			 R600_CONTEXT_FLUSH_AND_INV_DB_META |
1107			 R600_CONTEXT_INV_TEX_CACHE;
1108
1109	/* Set the new state. */
1110	util_copy_framebuffer_state(&rctx->framebuffer.state, state);
1111
1112	rctx->framebuffer.export_16bpc = state->nr_cbufs != 0;
1113	rctx->framebuffer.cb0_is_integer = state->nr_cbufs && state->cbufs[0] &&
1114			       util_format_is_pure_integer(state->cbufs[0]->format);
1115	rctx->framebuffer.compressed_cb_mask = 0;
1116	rctx->framebuffer.is_msaa_resolve = state->nr_cbufs == 2 &&
1117					    state->cbufs[0] && state->cbufs[1] &&
1118					    state->cbufs[0]->texture->nr_samples > 1 &&
1119				            state->cbufs[1]->texture->nr_samples <= 1;
1120	rctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
1121
1122	/* Colorbuffers. */
1123	for (i = 0; i < state->nr_cbufs; i++) {
1124		/* The resolve buffer must have CMASK and FMASK to prevent hardlocks on R6xx. */
1125		bool force_cmask_fmask = rctx->b.gfx_level == R600 &&
1126					 rctx->framebuffer.is_msaa_resolve &&
1127					 i == 1;
1128
1129		surf = (struct r600_surface*)state->cbufs[i];
1130		if (!surf)
1131			continue;
1132
1133		rtex = (struct r600_texture*)surf->base.texture;
1134		r600_context_add_resource_size(ctx, state->cbufs[i]->texture);
1135
1136		target_mask |= (0xf << (i * 4));
1137
1138		if (!surf->color_initialized || force_cmask_fmask) {
1139			r600_init_color_surface(rctx, surf, force_cmask_fmask);
1140			if (force_cmask_fmask) {
1141				/* re-initialize later without compression */
1142				surf->color_initialized = false;
1143			}
1144		}
1145
1146		if (!surf->export_16bpc) {
1147			rctx->framebuffer.export_16bpc = false;
1148		}
1149
1150		if (rtex->fmask.size) {
1151			rctx->framebuffer.compressed_cb_mask |= 1 << i;
1152		}
1153	}
1154
1155	/* Update alpha-test state dependencies.
1156	 * Alpha-test is done on the first colorbuffer only. */
1157	if (state->nr_cbufs) {
1158		bool alphatest_bypass = false;
1159
1160		surf = (struct r600_surface*)state->cbufs[0];
1161		if (surf) {
1162			alphatest_bypass = surf->alphatest_bypass;
1163		}
1164
1165		if (rctx->alphatest_state.bypass != alphatest_bypass) {
1166			rctx->alphatest_state.bypass = alphatest_bypass;
1167			r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1168		}
1169	}
1170
1171	/* ZS buffer. */
1172	if (state->zsbuf) {
1173		surf = (struct r600_surface*)state->zsbuf;
1174
1175		r600_context_add_resource_size(ctx, state->zsbuf->texture);
1176
1177		if (!surf->depth_initialized) {
1178			r600_init_depth_surface(rctx, surf);
1179		}
1180
1181		if (state->zsbuf->format != rctx->poly_offset_state.zs_format) {
1182			rctx->poly_offset_state.zs_format = state->zsbuf->format;
1183			r600_mark_atom_dirty(rctx, &rctx->poly_offset_state.atom);
1184		}
1185
1186		if (rctx->db_state.rsurf != surf) {
1187			rctx->db_state.rsurf = surf;
1188			r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1189			r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1190		}
1191	} else if (rctx->db_state.rsurf) {
1192		rctx->db_state.rsurf = NULL;
1193		r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1194		r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1195	}
1196
1197	if (rctx->cb_misc_state.nr_cbufs != state->nr_cbufs ||
1198	    rctx->cb_misc_state.bound_cbufs_target_mask != target_mask) {
1199		rctx->cb_misc_state.bound_cbufs_target_mask = target_mask;
1200		rctx->cb_misc_state.nr_cbufs = state->nr_cbufs;
1201		r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom);
1202	}
1203
1204	if (state->nr_cbufs == 0 && rctx->alphatest_state.bypass) {
1205		rctx->alphatest_state.bypass = false;
1206		r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1207	}
1208
1209	/* Calculate the CS size. */
1210	rctx->framebuffer.atom.num_dw =
1211		10 /*COLOR_INFO*/ + 4 /*SCISSOR*/ + 3 /*SHADER_CONTROL*/ + 8 /*MSAA*/;
1212
1213	if (rctx->framebuffer.state.nr_cbufs) {
1214		rctx->framebuffer.atom.num_dw += 15 * rctx->framebuffer.state.nr_cbufs;
1215		rctx->framebuffer.atom.num_dw += 3 * (2 + rctx->framebuffer.state.nr_cbufs);
1216	}
1217	if (rctx->framebuffer.state.zsbuf) {
1218		rctx->framebuffer.atom.num_dw += 16;
1219	} else {
1220		rctx->framebuffer.atom.num_dw += 3;
1221	}
1222	if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770) {
1223		rctx->framebuffer.atom.num_dw += 2;
1224	}
1225
1226	r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom);
1227
1228	r600_set_sample_locations_constant_buffer(rctx);
1229	rctx->framebuffer.do_update_surf_dirtiness = true;
1230}
1231
1232static const uint32_t sample_locs_2x[] = {
1233	FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1234	FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1235};
1236static const unsigned max_dist_2x = 4;
1237
1238static const uint32_t sample_locs_4x[] = {
1239	FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1240	FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1241};
1242static const unsigned max_dist_4x = 6;
1243static const uint32_t sample_locs_8x[] = {
1244	FILL_SREG(-1,  1,  1,  5,  3, -5,  5,  3),
1245	FILL_SREG(-7, -1, -3, -7,  7, -3, -5,  7),
1246};
1247static const unsigned max_dist_8x = 7;
1248
1249static void r600_get_sample_position(struct pipe_context *ctx,
1250				     unsigned sample_count,
1251				     unsigned sample_index,
1252				     float *out_value)
1253{
1254	int offset, index;
1255	struct {
1256		int idx:4;
1257	} val;
1258	switch (sample_count) {
1259	case 1:
1260	default:
1261		out_value[0] = out_value[1] = 0.5;
1262		break;
1263	case 2:
1264		offset = 4 * (sample_index * 2);
1265		val.idx = (sample_locs_2x[0] >> offset) & 0xf;
1266		out_value[0] = (float)(val.idx + 8) / 16.0f;
1267		val.idx = (sample_locs_2x[0] >> (offset + 4)) & 0xf;
1268		out_value[1] = (float)(val.idx + 8) / 16.0f;
1269		break;
1270	case 4:
1271		offset = 4 * (sample_index * 2);
1272		val.idx = (sample_locs_4x[0] >> offset) & 0xf;
1273		out_value[0] = (float)(val.idx + 8) / 16.0f;
1274		val.idx = (sample_locs_4x[0] >> (offset + 4)) & 0xf;
1275		out_value[1] = (float)(val.idx + 8) / 16.0f;
1276		break;
1277	case 8:
1278		offset = 4 * (sample_index % 4 * 2);
1279		index = (sample_index / 4);
1280		val.idx = (sample_locs_8x[index] >> offset) & 0xf;
1281		out_value[0] = (float)(val.idx + 8) / 16.0f;
1282		val.idx = (sample_locs_8x[index] >> (offset + 4)) & 0xf;
1283		out_value[1] = (float)(val.idx + 8) / 16.0f;
1284		break;
1285	}
1286}
1287
1288static void r600_emit_msaa_state(struct r600_context *rctx, int nr_samples)
1289{
1290	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1291	unsigned max_dist = 0;
1292
1293	if (rctx->b.family == CHIP_R600) {
1294		switch (nr_samples) {
1295		default:
1296			nr_samples = 0;
1297			break;
1298		case 2:
1299			radeon_set_config_reg(cs, R_008B40_PA_SC_AA_SAMPLE_LOCS_2S, sample_locs_2x[0]);
1300			max_dist = max_dist_2x;
1301			break;
1302		case 4:
1303			radeon_set_config_reg(cs, R_008B44_PA_SC_AA_SAMPLE_LOCS_4S, sample_locs_4x[0]);
1304			max_dist = max_dist_4x;
1305			break;
1306		case 8:
1307			radeon_set_config_reg_seq(cs, R_008B48_PA_SC_AA_SAMPLE_LOCS_8S_WD0, 2);
1308			radeon_emit(cs, sample_locs_8x[0]); /* R_008B48_PA_SC_AA_SAMPLE_LOCS_8S_WD0 */
1309			radeon_emit(cs, sample_locs_8x[1]); /* R_008B4C_PA_SC_AA_SAMPLE_LOCS_8S_WD1 */
1310			max_dist = max_dist_8x;
1311			break;
1312		}
1313	} else {
1314		switch (nr_samples) {
1315		default:
1316			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1317			radeon_emit(cs, 0); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1318			radeon_emit(cs, 0); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1319			nr_samples = 0;
1320			break;
1321		case 2:
1322			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1323			radeon_emit(cs, sample_locs_2x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1324			radeon_emit(cs, sample_locs_2x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1325			max_dist = max_dist_2x;
1326			break;
1327		case 4:
1328			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1329			radeon_emit(cs, sample_locs_4x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1330			radeon_emit(cs, sample_locs_4x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1331			max_dist = max_dist_4x;
1332			break;
1333		case 8:
1334			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1335			radeon_emit(cs, sample_locs_8x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1336			radeon_emit(cs, sample_locs_8x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1337			max_dist = max_dist_8x;
1338			break;
1339		}
1340	}
1341
1342	if (nr_samples > 1) {
1343		radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1344		radeon_emit(cs, S_028C00_LAST_PIXEL(1) |
1345				     S_028C00_EXPAND_LINE_WIDTH(1)); /* R_028C00_PA_SC_LINE_CNTL */
1346		radeon_emit(cs, S_028C04_MSAA_NUM_SAMPLES(util_logbase2(nr_samples)) |
1347				     S_028C04_MAX_SAMPLE_DIST(max_dist)); /* R_028C04_PA_SC_AA_CONFIG */
1348	} else {
1349		radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1350		radeon_emit(cs, S_028C00_LAST_PIXEL(1)); /* R_028C00_PA_SC_LINE_CNTL */
1351		radeon_emit(cs, 0); /* R_028C04_PA_SC_AA_CONFIG */
1352	}
1353}
1354
1355static void r600_emit_framebuffer_state(struct r600_context *rctx, struct r600_atom *atom)
1356{
1357	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1358	struct pipe_framebuffer_state *state = &rctx->framebuffer.state;
1359	unsigned nr_cbufs = state->nr_cbufs;
1360	struct r600_surface **cb = (struct r600_surface**)&state->cbufs[0];
1361	unsigned i, sbu = 0;
1362
1363	/* Colorbuffers. */
1364	radeon_set_context_reg_seq(cs, R_0280A0_CB_COLOR0_INFO, 8);
1365	for (i = 0; i < nr_cbufs; i++) {
1366		radeon_emit(cs, cb[i] ? cb[i]->cb_color_info : 0);
1367	}
1368	/* set CB_COLOR1_INFO for possible dual-src blending */
1369	if (rctx->framebuffer.dual_src_blend && i == 1 && cb[0]) {
1370		radeon_emit(cs, cb[0]->cb_color_info);
1371		i++;
1372	}
1373	for (; i < 8; i++) {
1374		radeon_emit(cs, 0);
1375	}
1376
1377	if (nr_cbufs) {
1378		for (i = 0; i < nr_cbufs; i++) {
1379			unsigned reloc;
1380
1381			if (!cb[i])
1382				continue;
1383
1384			/* COLOR_BASE */
1385			radeon_set_context_reg(cs, R_028040_CB_COLOR0_BASE + i*4, cb[i]->cb_color_base);
1386
1387			reloc = radeon_add_to_buffer_list(&rctx->b,
1388						      &rctx->b.gfx,
1389						      (struct r600_resource*)cb[i]->base.texture,
1390						      RADEON_USAGE_READWRITE |
1391						      (cb[i]->base.texture->nr_samples > 1 ?
1392							      RADEON_PRIO_COLOR_BUFFER_MSAA :
1393							      RADEON_PRIO_COLOR_BUFFER));
1394			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1395			radeon_emit(cs, reloc);
1396
1397			/* FMASK */
1398			radeon_set_context_reg(cs, R_0280E0_CB_COLOR0_FRAG + i*4, cb[i]->cb_color_fmask);
1399
1400			reloc = radeon_add_to_buffer_list(&rctx->b,
1401						      &rctx->b.gfx,
1402						      cb[i]->cb_buffer_fmask,
1403						      RADEON_USAGE_READWRITE |
1404						      (cb[i]->base.texture->nr_samples > 1 ?
1405							      RADEON_PRIO_COLOR_BUFFER_MSAA :
1406							      RADEON_PRIO_COLOR_BUFFER));
1407			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1408			radeon_emit(cs, reloc);
1409
1410			/* CMASK */
1411			radeon_set_context_reg(cs, R_0280C0_CB_COLOR0_TILE + i*4, cb[i]->cb_color_cmask);
1412
1413			reloc = radeon_add_to_buffer_list(&rctx->b,
1414						      &rctx->b.gfx,
1415						      cb[i]->cb_buffer_cmask,
1416						      RADEON_USAGE_READWRITE |
1417						      (cb[i]->base.texture->nr_samples > 1 ?
1418							      RADEON_PRIO_COLOR_BUFFER_MSAA :
1419							      RADEON_PRIO_COLOR_BUFFER));
1420			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1421			radeon_emit(cs, reloc);
1422		}
1423
1424		radeon_set_context_reg_seq(cs, R_028060_CB_COLOR0_SIZE, nr_cbufs);
1425		for (i = 0; i < nr_cbufs; i++) {
1426			radeon_emit(cs, cb[i] ? cb[i]->cb_color_size : 0);
1427		}
1428
1429		radeon_set_context_reg_seq(cs, R_028080_CB_COLOR0_VIEW, nr_cbufs);
1430		for (i = 0; i < nr_cbufs; i++) {
1431			radeon_emit(cs, cb[i] ? cb[i]->cb_color_view : 0);
1432		}
1433
1434		radeon_set_context_reg_seq(cs, R_028100_CB_COLOR0_MASK, nr_cbufs);
1435		for (i = 0; i < nr_cbufs; i++) {
1436			radeon_emit(cs, cb[i] ? cb[i]->cb_color_mask : 0);
1437		}
1438
1439		sbu |= SURFACE_BASE_UPDATE_COLOR_NUM(nr_cbufs);
1440	}
1441
1442	/* SURFACE_BASE_UPDATE */
1443	if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770 && sbu) {
1444		radeon_emit(cs, PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0));
1445		radeon_emit(cs, sbu);
1446		sbu = 0;
1447	}
1448
1449	/* Zbuffer. */
1450	if (state->zsbuf) {
1451		struct r600_surface *surf = (struct r600_surface*)state->zsbuf;
1452		unsigned reloc = radeon_add_to_buffer_list(&rctx->b,
1453						       &rctx->b.gfx,
1454						       (struct r600_resource*)state->zsbuf->texture,
1455						       RADEON_USAGE_READWRITE |
1456						       (surf->base.texture->nr_samples > 1 ?
1457							       RADEON_PRIO_DEPTH_BUFFER_MSAA :
1458							       RADEON_PRIO_DEPTH_BUFFER));
1459
1460		radeon_set_context_reg_seq(cs, R_028000_DB_DEPTH_SIZE, 2);
1461		radeon_emit(cs, surf->db_depth_size); /* R_028000_DB_DEPTH_SIZE */
1462		radeon_emit(cs, surf->db_depth_view); /* R_028004_DB_DEPTH_VIEW */
1463		radeon_set_context_reg_seq(cs, R_02800C_DB_DEPTH_BASE, 2);
1464		radeon_emit(cs, surf->db_depth_base); /* R_02800C_DB_DEPTH_BASE */
1465		radeon_emit(cs, surf->db_depth_info); /* R_028010_DB_DEPTH_INFO */
1466
1467		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1468		radeon_emit(cs, reloc);
1469
1470		radeon_set_context_reg(cs, R_028D34_DB_PREFETCH_LIMIT, surf->db_prefetch_limit);
1471
1472		sbu |= SURFACE_BASE_UPDATE_DEPTH;
1473	} else {
1474		radeon_set_context_reg(cs, R_028010_DB_DEPTH_INFO, S_028010_FORMAT(V_028010_DEPTH_INVALID));
1475	}
1476
1477	/* SURFACE_BASE_UPDATE */
1478	if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770 && sbu) {
1479		radeon_emit(cs, PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0));
1480		radeon_emit(cs, sbu);
1481		sbu = 0;
1482	}
1483
1484	/* Framebuffer dimensions. */
1485	radeon_set_context_reg_seq(cs, R_028204_PA_SC_WINDOW_SCISSOR_TL, 2);
1486	radeon_emit(cs, S_028240_TL_X(0) | S_028240_TL_Y(0) |
1487			     S_028240_WINDOW_OFFSET_DISABLE(1)); /* R_028204_PA_SC_WINDOW_SCISSOR_TL */
1488	radeon_emit(cs, S_028244_BR_X(state->width) |
1489			     S_028244_BR_Y(state->height)); /* R_028208_PA_SC_WINDOW_SCISSOR_BR */
1490
1491	if (rctx->framebuffer.is_msaa_resolve) {
1492		radeon_set_context_reg(cs, R_0287A0_CB_SHADER_CONTROL, 1);
1493	} else {
1494		/* Always enable the first colorbuffer in CB_SHADER_CONTROL. This
1495		 * will assure that the alpha-test will work even if there is
1496		 * no colorbuffer bound. */
1497		radeon_set_context_reg(cs, R_0287A0_CB_SHADER_CONTROL,
1498				       (1ull << MAX2(nr_cbufs, 1)) - 1);
1499	}
1500
1501	r600_emit_msaa_state(rctx, rctx->framebuffer.nr_samples);
1502}
1503
1504static void r600_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
1505{
1506	struct r600_context *rctx = (struct r600_context *)ctx;
1507
1508	if (rctx->ps_iter_samples == min_samples)
1509		return;
1510
1511	rctx->ps_iter_samples = min_samples;
1512	if (rctx->framebuffer.nr_samples > 1) {
1513		r600_mark_atom_dirty(rctx, &rctx->rasterizer_state.atom);
1514		if (rctx->b.gfx_level == R600)
1515			r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1516	}
1517}
1518
1519static void r600_emit_cb_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1520{
1521	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1522	struct r600_cb_misc_state *a = (struct r600_cb_misc_state*)atom;
1523
1524	if (G_028808_SPECIAL_OP(a->cb_color_control) == V_028808_SPECIAL_RESOLVE_BOX) {
1525		radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
1526		if (rctx->b.gfx_level == R600) {
1527			radeon_emit(cs, 0xff); /* R_028238_CB_TARGET_MASK */
1528			radeon_emit(cs, 0xff); /* R_02823C_CB_SHADER_MASK */
1529		} else {
1530			radeon_emit(cs, 0xf); /* R_028238_CB_TARGET_MASK */
1531			radeon_emit(cs, 0xf); /* R_02823C_CB_SHADER_MASK */
1532		}
1533		radeon_set_context_reg(cs, R_028808_CB_COLOR_CONTROL, a->cb_color_control);
1534	} else {
1535		unsigned fb_colormask = a->bound_cbufs_target_mask;
1536		unsigned ps_colormask = a->ps_color_export_mask;
1537		unsigned multiwrite = a->multiwrite && a->nr_cbufs > 1;
1538
1539		radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
1540		radeon_emit(cs, a->blend_colormask & fb_colormask); /* R_028238_CB_TARGET_MASK */
1541		/* Always enable the first color output to make sure alpha-test works even without one. */
1542		radeon_emit(cs, 0xf | (multiwrite ? fb_colormask : ps_colormask)); /* R_02823C_CB_SHADER_MASK */
1543		radeon_set_context_reg(cs, R_028808_CB_COLOR_CONTROL,
1544				       a->cb_color_control |
1545				       S_028808_MULTIWRITE_ENABLE(multiwrite));
1546	}
1547}
1548
1549static void r600_emit_db_state(struct r600_context *rctx, struct r600_atom *atom)
1550{
1551	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1552	struct r600_db_state *a = (struct r600_db_state*)atom;
1553
1554	if (a->rsurf && a->rsurf->db_htile_surface) {
1555		struct r600_texture *rtex = (struct r600_texture *)a->rsurf->base.texture;
1556		unsigned reloc_idx;
1557
1558		radeon_set_context_reg(cs, R_02802C_DB_DEPTH_CLEAR, fui(rtex->depth_clear_value));
1559		radeon_set_context_reg(cs, R_028D24_DB_HTILE_SURFACE, a->rsurf->db_htile_surface);
1560		radeon_set_context_reg(cs, R_028014_DB_HTILE_DATA_BASE, a->rsurf->db_htile_data_base);
1561		reloc_idx = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, &rtex->resource,
1562						  RADEON_USAGE_READWRITE | RADEON_PRIO_SEPARATE_META);
1563		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1564		radeon_emit(cs, reloc_idx);
1565	} else {
1566		radeon_set_context_reg(cs, R_028D24_DB_HTILE_SURFACE, 0);
1567	}
1568}
1569
1570static void r600_emit_db_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1571{
1572	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1573	struct r600_db_misc_state *a = (struct r600_db_misc_state*)atom;
1574	unsigned db_render_control = 0;
1575	unsigned db_render_override =
1576		S_028D10_FORCE_HIS_ENABLE0(V_028D10_FORCE_DISABLE) |
1577		S_028D10_FORCE_HIS_ENABLE1(V_028D10_FORCE_DISABLE);
1578
1579	if (rctx->b.gfx_level >= R700) {
1580		switch (a->ps_conservative_z) {
1581		default: /* fall through */
1582		case TGSI_FS_DEPTH_LAYOUT_ANY:
1583			db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_ANY_Z);
1584			break;
1585		case TGSI_FS_DEPTH_LAYOUT_GREATER:
1586			db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_GREATER_THAN_Z);
1587			break;
1588		case TGSI_FS_DEPTH_LAYOUT_LESS:
1589			db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_LESS_THAN_Z);
1590			break;
1591		}
1592	}
1593
1594	if (rctx->b.num_occlusion_queries > 0 &&
1595	    !a->occlusion_queries_disabled) {
1596		if (rctx->b.gfx_level >= R700) {
1597			db_render_control |= S_028D0C_R700_PERFECT_ZPASS_COUNTS(1);
1598		}
1599		db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1600	} else {
1601		db_render_control |= S_028D0C_ZPASS_INCREMENT_DISABLE(1);
1602	}
1603
1604	if (rctx->db_state.rsurf && rctx->db_state.rsurf->db_htile_surface) {
1605		/* FORCE_OFF means HiZ/HiS are determined by DB_SHADER_CONTROL */
1606		db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_OFF);
1607		/* This is to fix a lockup when hyperz and alpha test are enabled at
1608		 * the same time somehow GPU get confuse on which order to pick for
1609		 * z test
1610		 */
1611		if (rctx->alphatest_state.sx_alpha_test_control) {
1612			db_render_override |= S_028D10_FORCE_SHADER_Z_ORDER(1);
1613		}
1614	} else {
1615		db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1616	}
1617	if (rctx->b.gfx_level == R600 && rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0) {
1618		/* sample shading and hyperz causes lockups on R6xx chips */
1619		db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1620	}
1621	if (a->flush_depthstencil_through_cb) {
1622		assert(a->copy_depth || a->copy_stencil);
1623
1624		db_render_control |= S_028D0C_DEPTH_COPY_ENABLE(a->copy_depth) |
1625				     S_028D0C_STENCIL_COPY_ENABLE(a->copy_stencil) |
1626				     S_028D0C_COPY_CENTROID(1) |
1627				     S_028D0C_COPY_SAMPLE(a->copy_sample);
1628
1629		if (rctx->b.gfx_level == R600)
1630			db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1631
1632		if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
1633		    rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
1634			db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1635	} else if (a->flush_depth_inplace || a->flush_stencil_inplace) {
1636		db_render_control |= S_028D0C_DEPTH_COMPRESS_DISABLE(a->flush_depth_inplace) |
1637				     S_028D0C_STENCIL_COMPRESS_DISABLE(a->flush_stencil_inplace);
1638		db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1639	}
1640	if (a->htile_clear) {
1641		db_render_control |= S_028D0C_DEPTH_CLEAR_ENABLE(1);
1642	}
1643
1644	/* RV770 workaround for a hang with 8x MSAA. */
1645	if (rctx->b.family == CHIP_RV770 && a->log_samples == 3) {
1646		db_render_override |= S_028D10_MAX_TILES_IN_DTT(6);
1647	}
1648
1649	radeon_set_context_reg_seq(cs, R_028D0C_DB_RENDER_CONTROL, 2);
1650	radeon_emit(cs, db_render_control); /* R_028D0C_DB_RENDER_CONTROL */
1651	radeon_emit(cs, db_render_override); /* R_028D10_DB_RENDER_OVERRIDE */
1652	radeon_set_context_reg(cs, R_02880C_DB_SHADER_CONTROL, a->db_shader_control);
1653}
1654
1655static void r600_emit_config_state(struct r600_context *rctx, struct r600_atom *atom)
1656{
1657	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1658	struct r600_config_state *a = (struct r600_config_state*)atom;
1659
1660	radeon_set_config_reg(cs, R_008C04_SQ_GPR_RESOURCE_MGMT_1, a->sq_gpr_resource_mgmt_1);
1661	radeon_set_config_reg(cs, R_008C08_SQ_GPR_RESOURCE_MGMT_2, a->sq_gpr_resource_mgmt_2);
1662}
1663
1664static void r600_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom *atom)
1665{
1666	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1667	uint32_t dirty_mask = rctx->vertex_buffer_state.dirty_mask;
1668
1669	while (dirty_mask) {
1670		struct pipe_vertex_buffer *vb;
1671		struct r600_resource *rbuffer;
1672		unsigned offset;
1673		unsigned buffer_index = u_bit_scan(&dirty_mask);
1674
1675		vb = &rctx->vertex_buffer_state.vb[buffer_index];
1676		rbuffer = (struct r600_resource*)vb->buffer.resource;
1677		assert(rbuffer);
1678
1679		offset = vb->buffer_offset;
1680
1681		/* fetch resources start at index 320 (OFFSET_FS) */
1682		radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1683		radeon_emit(cs, (R600_FETCH_CONSTANTS_OFFSET_FS + buffer_index) * 7);
1684		radeon_emit(cs, offset); /* RESOURCEi_WORD0 */
1685		radeon_emit(cs, rbuffer->b.b.width0 - offset - 1); /* RESOURCEi_WORD1 */
1686		radeon_emit(cs, /* RESOURCEi_WORD2 */
1687				 S_038008_ENDIAN_SWAP(r600_endian_swap(32)) |
1688				 S_038008_STRIDE(vb->stride));
1689		radeon_emit(cs, 0); /* RESOURCEi_WORD3 */
1690		radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1691		radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1692		radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD6 */
1693
1694		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1695		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1696						      RADEON_USAGE_READ | RADEON_PRIO_VERTEX_BUFFER));
1697	}
1698}
1699
1700static void r600_emit_constant_buffers(struct r600_context *rctx,
1701				       struct r600_constbuf_state *state,
1702				       unsigned buffer_id_base,
1703				       unsigned reg_alu_constbuf_size,
1704				       unsigned reg_alu_const_cache)
1705{
1706	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1707	uint32_t dirty_mask = state->dirty_mask;
1708
1709	while (dirty_mask) {
1710		struct pipe_constant_buffer *cb;
1711		struct r600_resource *rbuffer;
1712		unsigned offset;
1713		unsigned buffer_index = ffs(dirty_mask) - 1;
1714		unsigned gs_ring_buffer = (buffer_index == R600_GS_RING_CONST_BUFFER);
1715		cb = &state->cb[buffer_index];
1716		rbuffer = (struct r600_resource*)cb->buffer;
1717		assert(rbuffer);
1718
1719		offset = cb->buffer_offset;
1720
1721		if (!gs_ring_buffer) {
1722			assert(buffer_index < R600_MAX_HW_CONST_BUFFERS);
1723			radeon_set_context_reg(cs, reg_alu_constbuf_size + buffer_index * 4,
1724					       DIV_ROUND_UP(cb->buffer_size, 256));
1725			radeon_set_context_reg(cs, reg_alu_const_cache + buffer_index * 4, offset >> 8);
1726			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1727			radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1728								  RADEON_USAGE_READ | RADEON_PRIO_CONST_BUFFER));
1729		}
1730
1731		radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1732		radeon_emit(cs, (buffer_id_base + buffer_index) * 7);
1733		radeon_emit(cs, offset); /* RESOURCEi_WORD0 */
1734		radeon_emit(cs, cb->buffer_size - 1); /* RESOURCEi_WORD1 */
1735		radeon_emit(cs, /* RESOURCEi_WORD2 */
1736			    S_038008_ENDIAN_SWAP(gs_ring_buffer ? ENDIAN_NONE : r600_endian_swap(32)) |
1737			    S_038008_STRIDE(gs_ring_buffer ? 4 : 16));
1738		radeon_emit(cs, 0); /* RESOURCEi_WORD3 */
1739		radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1740		radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1741		radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD6 */
1742
1743		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1744		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1745						      RADEON_USAGE_READ | RADEON_PRIO_CONST_BUFFER));
1746
1747		dirty_mask &= ~(1 << buffer_index);
1748	}
1749	state->dirty_mask = 0;
1750}
1751
1752static void r600_emit_vs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1753{
1754	r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX],
1755				   R600_FETCH_CONSTANTS_OFFSET_VS,
1756				   R_028180_ALU_CONST_BUFFER_SIZE_VS_0,
1757				   R_028980_ALU_CONST_CACHE_VS_0);
1758}
1759
1760static void r600_emit_gs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1761{
1762	r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY],
1763				   R600_FETCH_CONSTANTS_OFFSET_GS,
1764				   R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0,
1765				   R_0289C0_ALU_CONST_CACHE_GS_0);
1766}
1767
1768static void r600_emit_ps_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1769{
1770	r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT],
1771				   R600_FETCH_CONSTANTS_OFFSET_PS,
1772				   R_028140_ALU_CONST_BUFFER_SIZE_PS_0,
1773				   R_028940_ALU_CONST_CACHE_PS_0);
1774}
1775
1776static void r600_emit_sampler_views(struct r600_context *rctx,
1777				    struct r600_samplerview_state *state,
1778				    unsigned resource_id_base)
1779{
1780	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1781	uint32_t dirty_mask = state->dirty_mask;
1782
1783	while (dirty_mask) {
1784		struct r600_pipe_sampler_view *rview;
1785		unsigned resource_index = u_bit_scan(&dirty_mask);
1786		unsigned reloc;
1787
1788		rview = state->views[resource_index];
1789		assert(rview);
1790
1791		radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1792		radeon_emit(cs, (resource_id_base + resource_index) * 7);
1793		radeon_emit_array(cs, rview->tex_resource_words, 7);
1794
1795		reloc = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rview->tex_resource,
1796					      RADEON_USAGE_READ |
1797					      r600_get_sampler_view_priority(rview->tex_resource));
1798		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1799		radeon_emit(cs, reloc);
1800		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1801		radeon_emit(cs, reloc);
1802	}
1803	state->dirty_mask = 0;
1804}
1805
1806
1807static void r600_emit_vs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1808{
1809	r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views, R600_FETCH_CONSTANTS_OFFSET_VS + R600_MAX_CONST_BUFFERS);
1810}
1811
1812static void r600_emit_gs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1813{
1814	r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views, R600_FETCH_CONSTANTS_OFFSET_GS + R600_MAX_CONST_BUFFERS);
1815}
1816
1817static void r600_emit_ps_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1818{
1819	r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views, R600_FETCH_CONSTANTS_OFFSET_PS + R600_MAX_CONST_BUFFERS);
1820}
1821
1822static void r600_emit_sampler_states(struct r600_context *rctx,
1823				struct r600_textures_info *texinfo,
1824				unsigned resource_id_base,
1825				unsigned border_color_reg)
1826{
1827	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1828	uint32_t dirty_mask = texinfo->states.dirty_mask;
1829
1830	while (dirty_mask) {
1831		struct r600_pipe_sampler_state *rstate;
1832		struct r600_pipe_sampler_view *rview;
1833		unsigned i = u_bit_scan(&dirty_mask);
1834
1835		rstate = texinfo->states.states[i];
1836		assert(rstate);
1837		rview = texinfo->views.views[i];
1838
1839		/* TEX_ARRAY_OVERRIDE must be set for array textures to disable
1840		 * filtering between layers.
1841		 */
1842		enum pipe_texture_target target = PIPE_BUFFER;
1843		if (rview)
1844			target = rview->base.texture->target;
1845		if (target == PIPE_TEXTURE_1D_ARRAY ||
1846		    target == PIPE_TEXTURE_2D_ARRAY) {
1847			rstate->tex_sampler_words[0] |= S_03C000_TEX_ARRAY_OVERRIDE(1);
1848			texinfo->is_array_sampler[i] = true;
1849		} else {
1850			rstate->tex_sampler_words[0] &= C_03C000_TEX_ARRAY_OVERRIDE;
1851			texinfo->is_array_sampler[i] = false;
1852		}
1853
1854		radeon_emit(cs, PKT3(PKT3_SET_SAMPLER, 3, 0));
1855		radeon_emit(cs, (resource_id_base + i) * 3);
1856		radeon_emit_array(cs, rstate->tex_sampler_words, 3);
1857
1858		if (rstate->border_color_use) {
1859			unsigned offset;
1860
1861			offset = border_color_reg;
1862			offset += i * 16;
1863			radeon_set_config_reg_seq(cs, offset, 4);
1864			radeon_emit_array(cs, rstate->border_color.ui, 4);
1865		}
1866	}
1867	texinfo->states.dirty_mask = 0;
1868}
1869
1870static void r600_emit_vs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1871{
1872	r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 18, R_00A600_TD_VS_SAMPLER0_BORDER_RED);
1873}
1874
1875static void r600_emit_gs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1876{
1877	r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY], 36, R_00A800_TD_GS_SAMPLER0_BORDER_RED);
1878}
1879
1880static void r600_emit_ps_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1881{
1882	r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT], 0, R_00A400_TD_PS_SAMPLER0_BORDER_RED);
1883}
1884
1885static void r600_emit_seamless_cube_map(struct r600_context *rctx, struct r600_atom *atom)
1886{
1887	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1888	unsigned tmp;
1889
1890	tmp = S_009508_DISABLE_CUBE_ANISO(1) |
1891		S_009508_SYNC_GRADIENT(1) |
1892		S_009508_SYNC_WALKER(1) |
1893		S_009508_SYNC_ALIGNER(1);
1894	if (!rctx->seamless_cube_map.enabled) {
1895		tmp |= S_009508_DISABLE_CUBE_WRAP(1);
1896	}
1897	radeon_set_config_reg(cs, R_009508_TA_CNTL_AUX, tmp);
1898}
1899
1900static void r600_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a)
1901{
1902	struct r600_sample_mask *s = (struct r600_sample_mask*)a;
1903	uint8_t mask = s->sample_mask;
1904
1905	radeon_set_context_reg(&rctx->b.gfx.cs, R_028C48_PA_SC_AA_MASK,
1906			       mask | (mask << 8) | (mask << 16) | (mask << 24));
1907}
1908
1909static void r600_emit_vertex_fetch_shader(struct r600_context *rctx, struct r600_atom *a)
1910{
1911	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1912	struct r600_cso_state *state = (struct r600_cso_state*)a;
1913	struct r600_fetch_shader *shader = (struct r600_fetch_shader*)state->cso;
1914
1915	if (!shader)
1916		return;
1917
1918	radeon_set_context_reg(cs, R_028894_SQ_PGM_START_FS, shader->offset >> 8);
1919	radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1920	radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, shader->buffer,
1921                                                  RADEON_USAGE_READ |
1922                                                  RADEON_PRIO_SHADER_BINARY));
1923}
1924
1925static void r600_emit_shader_stages(struct r600_context *rctx, struct r600_atom *a)
1926{
1927	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1928	struct r600_shader_stages_state *state = (struct r600_shader_stages_state*)a;
1929
1930	uint32_t v2 = 0, primid = 0;
1931
1932	if (rctx->vs_shader->current->shader.vs_as_gs_a) {
1933		v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_A);
1934		primid = 1;
1935	}
1936
1937	if (state->geom_enable) {
1938		uint32_t cut_val;
1939
1940		if (rctx->gs_shader->gs_max_out_vertices <= 128)
1941			cut_val = V_028A40_GS_CUT_128;
1942		else if (rctx->gs_shader->gs_max_out_vertices <= 256)
1943			cut_val = V_028A40_GS_CUT_256;
1944		else if (rctx->gs_shader->gs_max_out_vertices <= 512)
1945			cut_val = V_028A40_GS_CUT_512;
1946		else
1947			cut_val = V_028A40_GS_CUT_1024;
1948
1949		v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_G) |
1950			S_028A40_CUT_MODE(cut_val);
1951
1952		if (rctx->gs_shader->current->shader.gs_prim_id_input)
1953			primid = 1;
1954	}
1955
1956	radeon_set_context_reg(cs, R_028A40_VGT_GS_MODE, v2);
1957	radeon_set_context_reg(cs, R_028A84_VGT_PRIMITIVEID_EN, primid);
1958}
1959
1960static void r600_emit_gs_rings(struct r600_context *rctx, struct r600_atom *a)
1961{
1962	struct radeon_cmdbuf *cs = &rctx->b.gfx.cs;
1963	struct r600_gs_rings_state *state = (struct r600_gs_rings_state*)a;
1964	struct r600_resource *rbuffer;
1965
1966	radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
1967	radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
1968	radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
1969
1970	if (state->enable) {
1971		rbuffer =(struct r600_resource*)state->esgs_ring.buffer;
1972		radeon_set_config_reg(cs, R_008C40_SQ_ESGS_RING_BASE, 0);
1973		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1974		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1975						      RADEON_USAGE_READWRITE |
1976						      RADEON_PRIO_SHADER_RINGS));
1977		radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE,
1978				state->esgs_ring.buffer_size >> 8);
1979
1980		rbuffer =(struct r600_resource*)state->gsvs_ring.buffer;
1981		radeon_set_config_reg(cs, R_008C48_SQ_GSVS_RING_BASE, 0);
1982		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1983		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1984						      RADEON_USAGE_READWRITE |
1985						      RADEON_PRIO_SHADER_RINGS));
1986		radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE,
1987				state->gsvs_ring.buffer_size >> 8);
1988	} else {
1989		radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE, 0);
1990		radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE, 0);
1991	}
1992
1993	radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
1994	radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
1995	radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
1996}
1997
1998/* Adjust GPR allocation on R6xx/R7xx */
1999bool r600_adjust_gprs(struct r600_context *rctx)
2000{
2001	unsigned num_gprs[R600_NUM_HW_STAGES];
2002	unsigned new_gprs[R600_NUM_HW_STAGES];
2003	unsigned cur_gprs[R600_NUM_HW_STAGES];
2004	unsigned def_gprs[R600_NUM_HW_STAGES];
2005	unsigned def_num_clause_temp_gprs = rctx->r6xx_num_clause_temp_gprs;
2006	unsigned max_gprs;
2007	unsigned tmp, tmp2;
2008	unsigned i;
2009	bool need_recalc = false, use_default = true;
2010
2011	/* hardware will reserve twice num_clause_temp_gprs */
2012	max_gprs = def_num_clause_temp_gprs * 2;
2013	for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2014		def_gprs[i] = rctx->default_gprs[i];
2015		max_gprs += def_gprs[i];
2016	}
2017
2018	cur_gprs[R600_HW_STAGE_PS] = G_008C04_NUM_PS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
2019	cur_gprs[R600_HW_STAGE_VS] = G_008C04_NUM_VS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
2020	cur_gprs[R600_HW_STAGE_GS] = G_008C08_NUM_GS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
2021	cur_gprs[R600_HW_STAGE_ES] = G_008C08_NUM_ES_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
2022
2023	num_gprs[R600_HW_STAGE_PS] = rctx->ps_shader->current->shader.bc.ngpr;
2024	if (rctx->gs_shader) {
2025		num_gprs[R600_HW_STAGE_ES] = rctx->vs_shader->current->shader.bc.ngpr;
2026		num_gprs[R600_HW_STAGE_GS] = rctx->gs_shader->current->shader.bc.ngpr;
2027		num_gprs[R600_HW_STAGE_VS] = rctx->gs_shader->current->gs_copy_shader->shader.bc.ngpr;
2028	} else {
2029		num_gprs[R600_HW_STAGE_ES] = 0;
2030		num_gprs[R600_HW_STAGE_GS] = 0;
2031		num_gprs[R600_HW_STAGE_VS] = rctx->vs_shader->current->shader.bc.ngpr;
2032	}
2033
2034	for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2035		new_gprs[i] = num_gprs[i];
2036		if (new_gprs[i] > cur_gprs[i])
2037			need_recalc = true;
2038		if (new_gprs[i] > def_gprs[i])
2039			use_default = false;
2040	}
2041
2042	/* the sum of all SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS must <= to max_gprs */
2043	if (!need_recalc)
2044		return true;
2045
2046	/* try to use switch back to default */
2047	if (!use_default) {
2048		/* always privilege vs stage so that at worst we have the
2049		 * pixel stage producing wrong output (not the vertex
2050		 * stage) */
2051		new_gprs[R600_HW_STAGE_PS] = max_gprs - def_num_clause_temp_gprs * 2;
2052		for (i = R600_HW_STAGE_VS; i < R600_NUM_HW_STAGES; i++)
2053			new_gprs[R600_HW_STAGE_PS] -= new_gprs[i];
2054	} else {
2055		for (i = 0; i < R600_NUM_HW_STAGES; i++)
2056			new_gprs[i] = def_gprs[i];
2057	}
2058
2059	/* SQ_PGM_RESOURCES_*.NUM_GPRS must always be program to a value <=
2060	 * SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS otherwise the GPU will lockup
2061	 * Also if a shader use more gpr than SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS
2062	 * it will lockup. So in this case just discard the draw command
2063	 * and don't change the current gprs repartitions.
2064	 */
2065	for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2066		if (num_gprs[i] > new_gprs[i]) {
2067			R600_ERR("shaders require too many register (%d + %d + %d + %d) "
2068				 "for a combined maximum of %d\n",
2069				 num_gprs[R600_HW_STAGE_PS], num_gprs[R600_HW_STAGE_VS], num_gprs[R600_HW_STAGE_ES], num_gprs[R600_HW_STAGE_GS], max_gprs);
2070			return false;
2071		}
2072	}
2073
2074	/* in some case we endup recomputing the current value */
2075	tmp = S_008C04_NUM_PS_GPRS(new_gprs[R600_HW_STAGE_PS]) |
2076		S_008C04_NUM_VS_GPRS(new_gprs[R600_HW_STAGE_VS]) |
2077		S_008C04_NUM_CLAUSE_TEMP_GPRS(def_num_clause_temp_gprs);
2078
2079	tmp2 = S_008C08_NUM_ES_GPRS(new_gprs[R600_HW_STAGE_ES]) |
2080		S_008C08_NUM_GS_GPRS(new_gprs[R600_HW_STAGE_GS]);
2081	if (rctx->config_state.sq_gpr_resource_mgmt_1 != tmp || rctx->config_state.sq_gpr_resource_mgmt_2 != tmp2) {
2082		rctx->config_state.sq_gpr_resource_mgmt_1 = tmp;
2083		rctx->config_state.sq_gpr_resource_mgmt_2 = tmp2;
2084		r600_mark_atom_dirty(rctx, &rctx->config_state.atom);
2085		rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE;
2086	}
2087	return true;
2088}
2089
2090void r600_init_atom_start_cs(struct r600_context *rctx)
2091{
2092	int ps_prio;
2093	int vs_prio;
2094	int gs_prio;
2095	int es_prio;
2096	int num_ps_gprs;
2097	int num_vs_gprs;
2098	int num_gs_gprs;
2099	int num_es_gprs;
2100	int num_temp_gprs;
2101	int num_ps_threads;
2102	int num_vs_threads;
2103	int num_gs_threads;
2104	int num_es_threads;
2105	int num_ps_stack_entries;
2106	int num_vs_stack_entries;
2107	int num_gs_stack_entries;
2108	int num_es_stack_entries;
2109	enum radeon_family family;
2110	struct r600_command_buffer *cb = &rctx->start_cs_cmd;
2111	uint32_t tmp, i;
2112
2113	r600_init_command_buffer(cb, 256);
2114
2115	/* R6xx requires this packet at the start of each command buffer */
2116	if (rctx->b.gfx_level == R600) {
2117		r600_store_value(cb, PKT3(PKT3_START_3D_CMDBUF, 0, 0));
2118		r600_store_value(cb, 0);
2119	}
2120	/* All asics require this one */
2121	r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
2122	r600_store_value(cb, 0x80000000);
2123	r600_store_value(cb, 0x80000000);
2124
2125	/* We're setting config registers here. */
2126	r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2127	r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
2128
2129	/* This enables pipeline stat & streamout queries.
2130	 * They are only disabled by blits.
2131	 */
2132	r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2133	r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PIPELINESTAT_START) | EVENT_INDEX(0));
2134
2135	family = rctx->b.family;
2136	ps_prio = 0;
2137	vs_prio = 1;
2138	gs_prio = 2;
2139	es_prio = 3;
2140	switch (family) {
2141	case CHIP_R600:
2142		num_ps_gprs = 192;
2143		num_vs_gprs = 56;
2144		num_temp_gprs = 4;
2145		num_gs_gprs = 0;
2146		num_es_gprs = 0;
2147		num_ps_threads = 136;
2148		num_vs_threads = 48;
2149		num_gs_threads = 4;
2150		num_es_threads = 4;
2151		num_ps_stack_entries = 128;
2152		num_vs_stack_entries = 128;
2153		num_gs_stack_entries = 0;
2154		num_es_stack_entries = 0;
2155		break;
2156	case CHIP_RV630:
2157	case CHIP_RV635:
2158		num_ps_gprs = 84;
2159		num_vs_gprs = 36;
2160		num_temp_gprs = 4;
2161		num_gs_gprs = 0;
2162		num_es_gprs = 0;
2163		num_ps_threads = 144;
2164		num_vs_threads = 40;
2165		num_gs_threads = 4;
2166		num_es_threads = 4;
2167		num_ps_stack_entries = 40;
2168		num_vs_stack_entries = 40;
2169		num_gs_stack_entries = 32;
2170		num_es_stack_entries = 16;
2171		break;
2172	case CHIP_RV610:
2173	case CHIP_RV620:
2174	case CHIP_RS780:
2175	case CHIP_RS880:
2176	default:
2177		num_ps_gprs = 84;
2178		num_vs_gprs = 36;
2179		num_temp_gprs = 4;
2180		num_gs_gprs = 0;
2181		num_es_gprs = 0;
2182		/* use limits 40 VS and at least 16 ES/GS */
2183		num_ps_threads = 120;
2184		num_vs_threads = 40;
2185		num_gs_threads = 16;
2186		num_es_threads = 16;
2187		num_ps_stack_entries = 40;
2188		num_vs_stack_entries = 40;
2189		num_gs_stack_entries = 32;
2190		num_es_stack_entries = 16;
2191		break;
2192	case CHIP_RV670:
2193		num_ps_gprs = 144;
2194		num_vs_gprs = 40;
2195		num_temp_gprs = 4;
2196		num_gs_gprs = 0;
2197		num_es_gprs = 0;
2198		num_ps_threads = 136;
2199		num_vs_threads = 48;
2200		num_gs_threads = 4;
2201		num_es_threads = 4;
2202		num_ps_stack_entries = 40;
2203		num_vs_stack_entries = 40;
2204		num_gs_stack_entries = 32;
2205		num_es_stack_entries = 16;
2206		break;
2207	case CHIP_RV770:
2208		num_ps_gprs = 130;
2209		num_vs_gprs = 56;
2210		num_temp_gprs = 4;
2211		num_gs_gprs = 31;
2212		num_es_gprs = 31;
2213		num_ps_threads = 180;
2214		num_vs_threads = 60;
2215		num_gs_threads = 4;
2216		num_es_threads = 4;
2217		num_ps_stack_entries = 128;
2218		num_vs_stack_entries = 128;
2219		num_gs_stack_entries = 128;
2220		num_es_stack_entries = 128;
2221		break;
2222	case CHIP_RV730:
2223	case CHIP_RV740:
2224		num_ps_gprs = 84;
2225		num_vs_gprs = 36;
2226		num_temp_gprs = 4;
2227		num_gs_gprs = 0;
2228		num_es_gprs = 0;
2229		num_ps_threads = 180;
2230		num_vs_threads = 60;
2231		num_gs_threads = 4;
2232		num_es_threads = 4;
2233		num_ps_stack_entries = 128;
2234		num_vs_stack_entries = 128;
2235		num_gs_stack_entries = 0;
2236		num_es_stack_entries = 0;
2237		break;
2238	case CHIP_RV710:
2239		num_ps_gprs = 192;
2240		num_vs_gprs = 56;
2241		num_temp_gprs = 4;
2242		num_gs_gprs = 0;
2243		num_es_gprs = 0;
2244		num_ps_threads = 136;
2245		num_vs_threads = 48;
2246		num_gs_threads = 4;
2247		num_es_threads = 4;
2248		num_ps_stack_entries = 128;
2249		num_vs_stack_entries = 128;
2250		num_gs_stack_entries = 0;
2251		num_es_stack_entries = 0;
2252		break;
2253	}
2254
2255	rctx->default_gprs[R600_HW_STAGE_PS] = num_ps_gprs;
2256	rctx->default_gprs[R600_HW_STAGE_VS] = num_vs_gprs;
2257	rctx->default_gprs[R600_HW_STAGE_GS] = 0;
2258	rctx->default_gprs[R600_HW_STAGE_ES] = 0;
2259
2260	rctx->r6xx_num_clause_temp_gprs = num_temp_gprs;
2261
2262	/* SQ_CONFIG */
2263	tmp = 0;
2264	switch (family) {
2265	case CHIP_RV610:
2266	case CHIP_RV620:
2267	case CHIP_RS780:
2268	case CHIP_RS880:
2269	case CHIP_RV710:
2270		break;
2271	default:
2272		tmp |= S_008C00_VC_ENABLE(1);
2273		break;
2274	}
2275	tmp |= S_008C00_DX9_CONSTS(0);
2276	tmp |= S_008C00_ALU_INST_PREFER_VECTOR(1);
2277	tmp |= S_008C00_PS_PRIO(ps_prio);
2278	tmp |= S_008C00_VS_PRIO(vs_prio);
2279	tmp |= S_008C00_GS_PRIO(gs_prio);
2280	tmp |= S_008C00_ES_PRIO(es_prio);
2281	r600_store_config_reg(cb, R_008C00_SQ_CONFIG, tmp);
2282
2283	/* SQ_GPR_RESOURCE_MGMT_2 */
2284	tmp = S_008C08_NUM_GS_GPRS(num_gs_gprs);
2285	tmp |= S_008C08_NUM_ES_GPRS(num_es_gprs);
2286	r600_store_config_reg_seq(cb, R_008C08_SQ_GPR_RESOURCE_MGMT_2, 4);
2287	r600_store_value(cb, tmp);
2288
2289	/* SQ_THREAD_RESOURCE_MGMT */
2290	tmp = S_008C0C_NUM_PS_THREADS(num_ps_threads);
2291	tmp |= S_008C0C_NUM_VS_THREADS(num_vs_threads);
2292	tmp |= S_008C0C_NUM_GS_THREADS(num_gs_threads);
2293	tmp |= S_008C0C_NUM_ES_THREADS(num_es_threads);
2294	r600_store_value(cb, tmp); /* R_008C0C_SQ_THREAD_RESOURCE_MGMT */
2295
2296	/* SQ_STACK_RESOURCE_MGMT_1 */
2297	tmp = S_008C10_NUM_PS_STACK_ENTRIES(num_ps_stack_entries);
2298	tmp |= S_008C10_NUM_VS_STACK_ENTRIES(num_vs_stack_entries);
2299	r600_store_value(cb, tmp); /* R_008C10_SQ_STACK_RESOURCE_MGMT_1 */
2300
2301	/* SQ_STACK_RESOURCE_MGMT_2 */
2302	tmp = S_008C14_NUM_GS_STACK_ENTRIES(num_gs_stack_entries);
2303	tmp |= S_008C14_NUM_ES_STACK_ENTRIES(num_es_stack_entries);
2304	r600_store_value(cb, tmp); /* R_008C14_SQ_STACK_RESOURCE_MGMT_2 */
2305
2306	r600_store_config_reg(cb, R_009714_VC_ENHANCE, 0);
2307
2308	if (rctx->b.gfx_level >= R700) {
2309		r600_store_context_reg(cb, R_028A50_VGT_ENHANCE, 4);
2310		r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0x00004000);
2311		r600_store_config_reg(cb, R_009830_DB_DEBUG, 0);
2312		r600_store_config_reg(cb, R_009838_DB_WATERMARKS, 0x00420204);
2313		r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0);
2314	} else {
2315		r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0);
2316		r600_store_config_reg(cb, R_009830_DB_DEBUG, 0x82000000);
2317		r600_store_config_reg(cb, R_009838_DB_WATERMARKS, 0x01020204);
2318		r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 1);
2319	}
2320	r600_store_context_reg_seq(cb, R_0288A8_SQ_ESGS_RING_ITEMSIZE, 9);
2321	r600_store_value(cb, 0); /* R_0288A8_SQ_ESGS_RING_ITEMSIZE */
2322	r600_store_value(cb, 0); /* R_0288AC_SQ_GSVS_RING_ITEMSIZE */
2323	r600_store_value(cb, 0); /* R_0288B0_SQ_ESTMP_RING_ITEMSIZE */
2324	r600_store_value(cb, 0); /* R_0288B4_SQ_GSTMP_RING_ITEMSIZE */
2325	r600_store_value(cb, 0); /* R_0288B8_SQ_VSTMP_RING_ITEMSIZE */
2326	r600_store_value(cb, 0); /* R_0288BC_SQ_PSTMP_RING_ITEMSIZE */
2327	r600_store_value(cb, 0); /* R_0288C0_SQ_FBUF_RING_ITEMSIZE */
2328	r600_store_value(cb, 0); /* R_0288C4_SQ_REDUC_RING_ITEMSIZE */
2329	r600_store_value(cb, 0); /* R_0288C8_SQ_GS_VERT_ITEMSIZE */
2330
2331	/* to avoid GPU doing any preloading of constant from random address */
2332	r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16);
2333	for (i = 0; i < 16; i++)
2334		r600_store_value(cb, 0);
2335
2336	r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16);
2337	for (i = 0; i < 16; i++)
2338		r600_store_value(cb, 0);
2339
2340	r600_store_context_reg_seq(cb, R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 16);
2341	for (i = 0; i < 16; i++)
2342		r600_store_value(cb, 0);
2343
2344	r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13);
2345	r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */
2346	r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */
2347	r600_store_value(cb, 0); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */
2348	r600_store_value(cb, 0); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */
2349	r600_store_value(cb, 0); /* R_028A20_VGT_HOS_REUSE_DEPTH */
2350	r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */
2351	r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */
2352	r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */
2353	r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */
2354	r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */
2355	r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */
2356	r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */
2357	r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE, 0); */
2358
2359	r600_store_context_reg(cb, R_028A84_VGT_PRIMITIVEID_EN, 0);
2360	r600_store_context_reg(cb, R_028AA0_VGT_INSTANCE_STEP_RATE_0, 0);
2361	r600_store_context_reg(cb, R_028AA4_VGT_INSTANCE_STEP_RATE_1, 0);
2362
2363	r600_store_context_reg_seq(cb, R_028AB4_VGT_REUSE_OFF, 2);
2364	r600_store_value(cb, 1); /* R_028AB4_VGT_REUSE_OFF */
2365	r600_store_value(cb, 0); /* R_028AB8_VGT_VTX_CNT_EN */
2366
2367	r600_store_context_reg(cb, R_028B20_VGT_STRMOUT_BUFFER_EN, 0);
2368
2369	r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0);
2370
2371	r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0);
2372
2373	r600_store_context_reg_seq(cb, R_0286DC_SPI_FOG_CNTL, 3);
2374	r600_store_value(cb, 0); /* R_0286DC_SPI_FOG_CNTL */
2375	r600_store_value(cb, 0); /* R_0286E0_SPI_FOG_FUNC_SCALE */
2376	r600_store_value(cb, 0); /* R_0286E4_SPI_FOG_FUNC_BIAS */
2377
2378	r600_store_context_reg_seq(cb, R_028D28_DB_SRESULTS_COMPARE_STATE0, 3);
2379	r600_store_value(cb, 0); /* R_028D28_DB_SRESULTS_COMPARE_STATE0 */
2380	r600_store_value(cb, 0); /* R_028D2C_DB_SRESULTS_COMPARE_STATE1 */
2381	r600_store_value(cb, 0); /* R_028D30_DB_PRELOAD_CONTROL */
2382
2383	r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0);
2384	r600_store_context_reg(cb, R_028A48_PA_SC_MPASS_PS_CNTL, 0);
2385
2386	r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0);
2387	r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
2388
2389	if (rctx->b.gfx_level >= R700) {
2390		r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
2391	}
2392
2393	r600_store_context_reg_seq(cb, R_028C30_CB_CLRCMP_CONTROL, 4);
2394	r600_store_value(cb, 0x1000000);  /* R_028C30_CB_CLRCMP_CONTROL */
2395	r600_store_value(cb, 0);          /* R_028C34_CB_CLRCMP_SRC */
2396	r600_store_value(cb, 0xFF);       /* R_028C38_CB_CLRCMP_DST */
2397	r600_store_value(cb, 0xFFFFFFFF); /* R_028C3C_CB_CLRCMP_MSK */
2398
2399	r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2);
2400	r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */
2401	r600_store_value(cb, S_028034_BR_X(8192) | S_028034_BR_Y(8192)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */
2402
2403	r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2);
2404	r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */
2405	r600_store_value(cb, S_028244_BR_X(8192) | S_028244_BR_Y(8192)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */
2406
2407	r600_store_context_reg_seq(cb, R_0288CC_SQ_PGM_CF_OFFSET_PS, 5);
2408	r600_store_value(cb, 0); /* R_0288CC_SQ_PGM_CF_OFFSET_PS */
2409	r600_store_value(cb, 0); /* R_0288D0_SQ_PGM_CF_OFFSET_VS */
2410	r600_store_value(cb, 0); /* R_0288D4_SQ_PGM_CF_OFFSET_GS */
2411	r600_store_value(cb, 0); /* R_0288D8_SQ_PGM_CF_OFFSET_ES */
2412	r600_store_value(cb, 0); /* R_0288DC_SQ_PGM_CF_OFFSET_FS */
2413
2414        r600_store_context_reg(cb, R_0288E0_SQ_VTX_SEMANTIC_CLEAR, ~0);
2415
2416        r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2);
2417	r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */
2418	r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */
2419
2420	r600_store_context_reg(cb, R_0288A4_SQ_PGM_RESOURCES_FS, 0);
2421
2422	if (rctx->b.gfx_level == R700)
2423		r600_store_context_reg(cb, R_028350_SX_MISC, 0);
2424	if (rctx->b.gfx_level == R700 && rctx->screen->b.has_streamout)
2425		r600_store_context_reg(cb, R_028354_SX_SURFACE_SYNC, S_028354_SURFACE_SYNC_MASK(0xf));
2426
2427	r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0);
2428	if (rctx->screen->b.has_streamout) {
2429		r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
2430	}
2431
2432	r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0, 0x1000FFF);
2433	r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0 + (32 * 4), 0x1000FFF);
2434	r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0 + (64 * 4), 0x1000FFF);
2435}
2436
2437void r600_update_ps_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2438{
2439	struct r600_context *rctx = (struct r600_context *)ctx;
2440	struct r600_command_buffer *cb = &shader->command_buffer;
2441	struct r600_shader *rshader = &shader->shader;
2442	unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1, db_shader_control;
2443	int pos_index = -1, face_index = -1, fixed_pt_position_index = -1;
2444	unsigned tmp, sid, ufi = 0;
2445	int need_linear = 0;
2446	unsigned z_export = 0, stencil_export = 0, mask_export = 0;
2447
2448	/* Pull any state we use out of rctx.  Make sure that any additional
2449	 * state added to this list is also checked in the caller in
2450	 * r600_update_derived_state().
2451	 */
2452	bool sprite_coord_enable = rctx->rasterizer ? rctx->rasterizer->sprite_coord_enable : 0;
2453	bool flatshade = rctx->rasterizer ? rctx->rasterizer->flatshade : 0;
2454	bool msaa = rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0;
2455
2456	if (!cb->buf) {
2457		r600_init_command_buffer(cb, 64);
2458	} else {
2459		cb->num_dw = 0;
2460	}
2461
2462	r600_store_context_reg_seq(cb, R_028644_SPI_PS_INPUT_CNTL_0, rshader->ninput);
2463	for (i = 0; i < rshader->ninput; i++) {
2464		if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
2465			pos_index = i;
2466		if (rshader->input[i].name == TGSI_SEMANTIC_FACE && face_index == -1)
2467			face_index = i;
2468		if (rshader->input[i].name == TGSI_SEMANTIC_SAMPLEID)
2469			fixed_pt_position_index = i;
2470
2471		sid = rshader->input[i].spi_sid;
2472
2473		tmp = S_028644_SEMANTIC(sid);
2474
2475		/* D3D 9 behaviour. GL is undefined */
2476		if (rshader->input[i].name == TGSI_SEMANTIC_COLOR && rshader->input[i].sid == 0)
2477			tmp |= S_028644_DEFAULT_VAL(3);
2478
2479		if (rshader->input[i].name == TGSI_SEMANTIC_POSITION ||
2480			rshader->input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
2481			(rshader->input[i].interpolate == TGSI_INTERPOLATE_COLOR && flatshade))
2482			tmp |= S_028644_FLAT_SHADE(1);
2483
2484		if (rshader->input[i].name == TGSI_SEMANTIC_PCOORD ||
2485		    (rshader->input[i].name == TGSI_SEMANTIC_TEXCOORD &&
2486		     sprite_coord_enable & (1 << rshader->input[i].sid))) {
2487			tmp |= S_028644_PT_SPRITE_TEX(1);
2488		}
2489
2490		if (rshader->input[i].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID)
2491			tmp |= S_028644_SEL_CENTROID(1);
2492
2493		if (rshader->input[i].interpolate_location == TGSI_INTERPOLATE_LOC_SAMPLE)
2494			tmp |= S_028644_SEL_SAMPLE(1);
2495
2496		if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR) {
2497			need_linear = 1;
2498			tmp |= S_028644_SEL_LINEAR(1);
2499		}
2500
2501		r600_store_value(cb, tmp);
2502	}
2503
2504	db_shader_control = 0;
2505	for (i = 0; i < rshader->noutput; i++) {
2506		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
2507			z_export = 1;
2508		if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
2509			stencil_export = 1;
2510		if (rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK && msaa)
2511			mask_export = 1;
2512	}
2513	db_shader_control |= S_02880C_Z_EXPORT_ENABLE(z_export);
2514	db_shader_control |= S_02880C_STENCIL_REF_EXPORT_ENABLE(stencil_export);
2515	db_shader_control |= S_02880C_MASK_EXPORT_ENABLE(mask_export);
2516	if (rshader->uses_kill)
2517		db_shader_control |= S_02880C_KILL_ENABLE(1);
2518
2519	exports_ps = 0;
2520	for (i = 0; i < rshader->noutput; i++) {
2521		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||
2522		    rshader->output[i].name == TGSI_SEMANTIC_STENCIL ||
2523		    rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK) {
2524			exports_ps |= 1;
2525		}
2526	}
2527	num_cout = rshader->nr_ps_color_exports;
2528	exports_ps |= S_028854_EXPORT_COLORS(num_cout);
2529	if (!exports_ps) {
2530		/* always at least export 1 component per pixel */
2531		exports_ps = 2;
2532	}
2533
2534	shader->nr_ps_color_outputs = num_cout;
2535	shader->ps_color_export_mask = rshader->ps_color_export_mask;
2536
2537	spi_ps_in_control_0 = S_0286CC_NUM_INTERP(rshader->ninput) |
2538				S_0286CC_PERSP_GRADIENT_ENA(1)|
2539				S_0286CC_LINEAR_GRADIENT_ENA(need_linear);
2540	spi_input_z = 0;
2541	if (pos_index != -1) {
2542		spi_ps_in_control_0 |= (S_0286CC_POSITION_ENA(1) |
2543					S_0286CC_POSITION_CENTROID(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID) |
2544					S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr) |
2545					S_0286CC_BARYC_SAMPLE_CNTL(1)) |
2546					S_0286CC_POSITION_SAMPLE(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_SAMPLE);
2547		spi_input_z |= S_0286D8_PROVIDE_Z_TO_SPI(1);
2548	}
2549
2550	spi_ps_in_control_1 = 0;
2551	if (face_index != -1) {
2552		spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) |
2553			S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr);
2554	}
2555	if (fixed_pt_position_index != -1) {
2556		spi_ps_in_control_1 |= S_0286D0_FIXED_PT_POSITION_ENA(1) |
2557			S_0286D0_FIXED_PT_POSITION_ADDR(rshader->input[fixed_pt_position_index].gpr);
2558	}
2559
2560	/* HW bug in original R600 */
2561	if (rctx->b.family == CHIP_R600)
2562		ufi = 1;
2563
2564	r600_store_context_reg_seq(cb, R_0286CC_SPI_PS_IN_CONTROL_0, 2);
2565	r600_store_value(cb, spi_ps_in_control_0); /* R_0286CC_SPI_PS_IN_CONTROL_0 */
2566	r600_store_value(cb, spi_ps_in_control_1); /* R_0286D0_SPI_PS_IN_CONTROL_1 */
2567
2568	r600_store_context_reg(cb, R_0286D8_SPI_INPUT_Z, spi_input_z);
2569
2570	r600_store_context_reg_seq(cb, R_028850_SQ_PGM_RESOURCES_PS, 2);
2571	r600_store_value(cb, /* R_028850_SQ_PGM_RESOURCES_PS*/
2572			 S_028850_NUM_GPRS(rshader->bc.ngpr) |
2573	/*
2574	 * docs are misleading about the dx10_clamp bit. This only affects
2575	 * instructions using CLAMP dst modifier, in which case they will
2576	 * return 0 with this set for a NaN (otherwise NaN).
2577	 */
2578			 S_028850_DX10_CLAMP(1) |
2579			 S_028850_STACK_SIZE(rshader->bc.nstack) |
2580			 S_028850_UNCACHED_FIRST_INST(ufi));
2581	r600_store_value(cb, exports_ps); /* R_028854_SQ_PGM_EXPORTS_PS */
2582
2583	r600_store_context_reg(cb, R_028840_SQ_PGM_START_PS, 0);
2584	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2585
2586	/* only set some bits here, the other bits are set in the dsa state */
2587	shader->db_shader_control = db_shader_control;
2588	shader->ps_depth_export = z_export | stencil_export | mask_export;
2589
2590	shader->sprite_coord_enable = sprite_coord_enable;
2591	shader->flatshade = flatshade;
2592	shader->msaa = msaa;
2593}
2594
2595void r600_update_vs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2596{
2597	struct r600_command_buffer *cb = &shader->command_buffer;
2598	struct r600_shader *rshader = &shader->shader;
2599	unsigned spi_vs_out_id[10] = {};
2600	unsigned i, tmp, nparams = 0;
2601
2602	for (i = 0; i < rshader->noutput; i++) {
2603		if (rshader->output[i].spi_sid) {
2604			tmp = rshader->output[i].spi_sid << ((nparams & 3) * 8);
2605			spi_vs_out_id[nparams / 4] |= tmp;
2606			nparams++;
2607		}
2608	}
2609
2610	r600_init_command_buffer(cb, 32);
2611
2612	r600_store_context_reg_seq(cb, R_028614_SPI_VS_OUT_ID_0, 10);
2613	for (i = 0; i < 10; i++) {
2614		r600_store_value(cb, spi_vs_out_id[i]);
2615	}
2616
2617	/* Certain attributes (position, psize, etc.) don't count as params.
2618	 * VS is required to export at least one param and r600_shader_from_tgsi()
2619	 * takes care of adding a dummy export.
2620	 */
2621	if (nparams < 1)
2622		nparams = 1;
2623
2624	r600_store_context_reg(cb, R_0286C4_SPI_VS_OUT_CONFIG,
2625			       S_0286C4_VS_EXPORT_COUNT(nparams - 1));
2626	r600_store_context_reg(cb, R_028868_SQ_PGM_RESOURCES_VS,
2627			       S_028868_NUM_GPRS(rshader->bc.ngpr) |
2628			       S_028868_DX10_CLAMP(1) |
2629			       S_028868_STACK_SIZE(rshader->bc.nstack));
2630	if (rshader->vs_position_window_space) {
2631		r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
2632			S_028818_VTX_XY_FMT(1) | S_028818_VTX_Z_FMT(1));
2633	} else {
2634		r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
2635			S_028818_VTX_W0_FMT(1) |
2636			S_028818_VPORT_X_SCALE_ENA(1) | S_028818_VPORT_X_OFFSET_ENA(1) |
2637			S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
2638			S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1));
2639
2640	}
2641	r600_store_context_reg(cb, R_028858_SQ_PGM_START_VS, 0);
2642	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2643
2644	shader->pa_cl_vs_out_cntl =
2645		S_02881C_VS_OUT_CCDIST0_VEC_ENA((rshader->clip_dist_write & 0x0F) != 0) |
2646		S_02881C_VS_OUT_CCDIST1_VEC_ENA((rshader->clip_dist_write & 0xF0) != 0) |
2647		S_02881C_VS_OUT_MISC_VEC_ENA(rshader->vs_out_misc_write) |
2648		S_02881C_USE_VTX_POINT_SIZE(rshader->vs_out_point_size) |
2649		S_02881C_USE_VTX_EDGE_FLAG(rshader->vs_out_edgeflag) |
2650		S_02881C_USE_VTX_RENDER_TARGET_INDX(rshader->vs_out_layer) |
2651		S_02881C_USE_VTX_VIEWPORT_INDX(rshader->vs_out_viewport);
2652}
2653
2654#define RV610_GSVS_ALIGN 32
2655#define R600_GSVS_ALIGN 16
2656
2657void r600_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2658{
2659	struct r600_context *rctx = (struct r600_context *)ctx;
2660	struct r600_command_buffer *cb = &shader->command_buffer;
2661	struct r600_shader *rshader = &shader->shader;
2662	struct r600_shader *cp_shader = &shader->gs_copy_shader->shader;
2663	unsigned gsvs_itemsize =
2664			(cp_shader->ring_item_sizes[0] * shader->selector->gs_max_out_vertices) >> 2;
2665
2666	/* some r600s needs gsvs itemsize aligned to cacheline size
2667	   this was fixed in rs780 and above. */
2668	switch (rctx->b.family) {
2669	case CHIP_RV610:
2670		gsvs_itemsize = align(gsvs_itemsize, RV610_GSVS_ALIGN);
2671		break;
2672	case CHIP_R600:
2673	case CHIP_RV630:
2674	case CHIP_RV670:
2675	case CHIP_RV620:
2676	case CHIP_RV635:
2677		gsvs_itemsize = align(gsvs_itemsize, R600_GSVS_ALIGN);
2678		break;
2679	default:
2680		break;
2681	}
2682
2683	r600_init_command_buffer(cb, 64);
2684
2685	/* VGT_GS_MODE is written by r600_emit_shader_stages */
2686	r600_store_context_reg(cb, R_028AB8_VGT_VTX_CNT_EN, 1);
2687
2688	if (rctx->b.gfx_level >= R700) {
2689		r600_store_context_reg(cb, R_028B38_VGT_GS_MAX_VERT_OUT,
2690				       S_028B38_MAX_VERT_OUT(shader->selector->gs_max_out_vertices));
2691	}
2692	r600_store_context_reg(cb, R_028A6C_VGT_GS_OUT_PRIM_TYPE,
2693			       r600_conv_prim_to_gs_out(shader->selector->gs_output_prim));
2694
2695	r600_store_context_reg(cb, R_0288C8_SQ_GS_VERT_ITEMSIZE,
2696	                       cp_shader->ring_item_sizes[0] >> 2);
2697
2698	r600_store_context_reg(cb, R_0288A8_SQ_ESGS_RING_ITEMSIZE,
2699			       (rshader->ring_item_sizes[0]) >> 2);
2700
2701	r600_store_context_reg(cb, R_0288AC_SQ_GSVS_RING_ITEMSIZE,
2702			       gsvs_itemsize);
2703
2704	/* FIXME calculate these values somehow ??? */
2705	r600_store_config_reg_seq(cb, R_0088C8_VGT_GS_PER_ES, 2);
2706	r600_store_value(cb, 0x80); /* GS_PER_ES */
2707	r600_store_value(cb, 0x100); /* ES_PER_GS */
2708	r600_store_config_reg_seq(cb, R_0088E8_VGT_GS_PER_VS, 1);
2709	r600_store_value(cb, 0x2); /* GS_PER_VS */
2710
2711	r600_store_context_reg(cb, R_02887C_SQ_PGM_RESOURCES_GS,
2712			       S_02887C_NUM_GPRS(rshader->bc.ngpr) |
2713			       S_02887C_DX10_CLAMP(1) |
2714			       S_02887C_STACK_SIZE(rshader->bc.nstack));
2715	r600_store_context_reg(cb, R_02886C_SQ_PGM_START_GS, 0);
2716	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2717}
2718
2719void r600_update_es_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2720{
2721	struct r600_command_buffer *cb = &shader->command_buffer;
2722	struct r600_shader *rshader = &shader->shader;
2723
2724	r600_init_command_buffer(cb, 32);
2725
2726	r600_store_context_reg(cb, R_028890_SQ_PGM_RESOURCES_ES,
2727			       S_028890_NUM_GPRS(rshader->bc.ngpr) |
2728			       S_028890_DX10_CLAMP(1) |
2729			       S_028890_STACK_SIZE(rshader->bc.nstack));
2730	r600_store_context_reg(cb, R_028880_SQ_PGM_START_ES, 0);
2731	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2732}
2733
2734
2735void *r600_create_resolve_blend(struct r600_context *rctx)
2736{
2737	struct pipe_blend_state blend;
2738	unsigned i;
2739
2740	memset(&blend, 0, sizeof(blend));
2741	blend.independent_blend_enable = true;
2742	for (i = 0; i < 2; i++) {
2743		blend.rt[i].colormask = 0xf;
2744		blend.rt[i].blend_enable = 1;
2745		blend.rt[i].rgb_func = PIPE_BLEND_ADD;
2746		blend.rt[i].alpha_func = PIPE_BLEND_ADD;
2747		blend.rt[i].rgb_src_factor = PIPE_BLENDFACTOR_ZERO;
2748		blend.rt[i].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
2749		blend.rt[i].alpha_src_factor = PIPE_BLENDFACTOR_ZERO;
2750		blend.rt[i].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
2751	}
2752	return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_RESOLVE_BOX);
2753}
2754
2755void *r700_create_resolve_blend(struct r600_context *rctx)
2756{
2757	struct pipe_blend_state blend;
2758
2759	memset(&blend, 0, sizeof(blend));
2760	blend.independent_blend_enable = true;
2761	blend.rt[0].colormask = 0xf;
2762	return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_RESOLVE_BOX);
2763}
2764
2765void *r600_create_decompress_blend(struct r600_context *rctx)
2766{
2767	struct pipe_blend_state blend;
2768
2769	memset(&blend, 0, sizeof(blend));
2770	blend.independent_blend_enable = true;
2771	blend.rt[0].colormask = 0xf;
2772	return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_EXPAND_SAMPLES);
2773}
2774
2775void *r600_create_db_flush_dsa(struct r600_context *rctx)
2776{
2777	struct pipe_depth_stencil_alpha_state dsa;
2778	boolean quirk = false;
2779
2780	if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
2781		rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
2782		quirk = true;
2783
2784	memset(&dsa, 0, sizeof(dsa));
2785
2786	if (quirk) {
2787		dsa.depth_enabled = 1;
2788		dsa.depth_func = PIPE_FUNC_LEQUAL;
2789		dsa.stencil[0].enabled = 1;
2790		dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
2791		dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_KEEP;
2792		dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_INCR;
2793		dsa.stencil[0].writemask = 0xff;
2794	}
2795
2796	return rctx->b.b.create_depth_stencil_alpha_state(&rctx->b.b, &dsa);
2797}
2798
2799void r600_update_db_shader_control(struct r600_context * rctx)
2800{
2801	bool dual_export;
2802	unsigned db_shader_control;
2803	uint8_t ps_conservative_z;
2804
2805	if (!rctx->ps_shader) {
2806		return;
2807	}
2808
2809	dual_export = rctx->framebuffer.export_16bpc &&
2810		      !rctx->ps_shader->current->ps_depth_export;
2811
2812	db_shader_control = rctx->ps_shader->current->db_shader_control |
2813			    S_02880C_DUAL_EXPORT_ENABLE(dual_export);
2814
2815	ps_conservative_z = rctx->ps_shader->current->shader.ps_conservative_z;
2816
2817	/* When alpha test is enabled we can't trust the hw to make the proper
2818	 * decision on the order in which ztest should be run related to fragment
2819	 * shader execution.
2820	 *
2821	 * If alpha test is enabled perform z test after fragment. RE_Z (early
2822	 * z test but no write to the zbuffer) seems to cause lockup on r6xx/r7xx
2823	 */
2824	if (rctx->alphatest_state.sx_alpha_test_control) {
2825		db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z);
2826	} else {
2827		db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
2828	}
2829
2830	if (db_shader_control != rctx->db_misc_state.db_shader_control ||
2831		ps_conservative_z != rctx->db_misc_state.ps_conservative_z) {
2832		rctx->db_misc_state.db_shader_control = db_shader_control;
2833		rctx->db_misc_state.ps_conservative_z = ps_conservative_z;
2834		r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
2835	}
2836}
2837
2838static inline unsigned r600_array_mode(unsigned mode)
2839{
2840	switch (mode) {
2841	default:
2842	case RADEON_SURF_MODE_LINEAR_ALIGNED:	return V_0280A0_ARRAY_LINEAR_ALIGNED;
2843		break;
2844	case RADEON_SURF_MODE_1D:		return V_0280A0_ARRAY_1D_TILED_THIN1;
2845		break;
2846	case RADEON_SURF_MODE_2D:		return V_0280A0_ARRAY_2D_TILED_THIN1;
2847	}
2848}
2849
2850static boolean r600_dma_copy_tile(struct r600_context *rctx,
2851				struct pipe_resource *dst,
2852				unsigned dst_level,
2853				unsigned dst_x,
2854				unsigned dst_y,
2855				unsigned dst_z,
2856				struct pipe_resource *src,
2857				unsigned src_level,
2858				unsigned src_x,
2859				unsigned src_y,
2860				unsigned src_z,
2861				unsigned copy_height,
2862				unsigned pitch,
2863				unsigned bpp)
2864{
2865	struct radeon_cmdbuf *cs = &rctx->b.dma.cs;
2866	struct r600_texture *rsrc = (struct r600_texture*)src;
2867	struct r600_texture *rdst = (struct r600_texture*)dst;
2868	unsigned array_mode, lbpp, pitch_tile_max, slice_tile_max, size;
2869	unsigned ncopy, height, cheight, detile, i, x, y, z, src_mode, dst_mode;
2870	uint64_t base, addr;
2871
2872	dst_mode = rdst->surface.u.legacy.level[dst_level].mode;
2873	src_mode = rsrc->surface.u.legacy.level[src_level].mode;
2874	assert(dst_mode != src_mode);
2875
2876	y = 0;
2877	lbpp = util_logbase2(bpp);
2878	pitch_tile_max = ((pitch / bpp) / 8) - 1;
2879
2880	if (dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED) {
2881		/* T2L */
2882		array_mode = r600_array_mode(src_mode);
2883		slice_tile_max = (rsrc->surface.u.legacy.level[src_level].nblk_x * rsrc->surface.u.legacy.level[src_level].nblk_y) / (8*8);
2884		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
2885		/* linear height must be the same as the slice tile max height, it's ok even
2886		 * if the linear destination/source have smaller heigh as the size of the
2887		 * dma packet will be using the copy_height which is always smaller or equal
2888		 * to the linear height
2889		 */
2890		height = u_minify(rsrc->resource.b.b.height0, src_level);
2891		detile = 1;
2892		x = src_x;
2893		y = src_y;
2894		z = src_z;
2895		base = (uint64_t)rsrc->surface.u.legacy.level[src_level].offset_256B * 256;
2896		addr = (uint64_t)rdst->surface.u.legacy.level[dst_level].offset_256B * 256;
2897		addr += (uint64_t)rdst->surface.u.legacy.level[dst_level].slice_size_dw * 4 * dst_z;
2898		addr += dst_y * pitch + dst_x * bpp;
2899	} else {
2900		/* L2T */
2901		array_mode = r600_array_mode(dst_mode);
2902		slice_tile_max = (rdst->surface.u.legacy.level[dst_level].nblk_x * rdst->surface.u.legacy.level[dst_level].nblk_y) / (8*8);
2903		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
2904		/* linear height must be the same as the slice tile max height, it's ok even
2905		 * if the linear destination/source have smaller heigh as the size of the
2906		 * dma packet will be using the copy_height which is always smaller or equal
2907		 * to the linear height
2908		 */
2909		height = u_minify(rdst->resource.b.b.height0, dst_level);
2910		detile = 0;
2911		x = dst_x;
2912		y = dst_y;
2913		z = dst_z;
2914		base = (uint64_t)rdst->surface.u.legacy.level[dst_level].offset_256B * 256;
2915		addr = (uint64_t)rsrc->surface.u.legacy.level[src_level].offset_256B * 256;
2916		addr += (uint64_t)rsrc->surface.u.legacy.level[src_level].slice_size_dw * 4 * src_z;
2917		addr += src_y * pitch + src_x * bpp;
2918	}
2919	/* check that we are in dw/base alignment constraint */
2920	if (addr % 4 || base % 256) {
2921		return FALSE;
2922	}
2923
2924	/* It's a r6xx/r7xx limitation, the blit must be on 8 boundary for number
2925	 * line in the blit. Compute max 8 line we can copy in the size limit
2926	 */
2927	cheight = ((R600_DMA_COPY_MAX_SIZE_DW * 4) / pitch) & 0xfffffff8;
2928	ncopy = (copy_height / cheight) + !!(copy_height % cheight);
2929	r600_need_dma_space(&rctx->b, ncopy * 7, &rdst->resource, &rsrc->resource);
2930
2931	for (i = 0; i < ncopy; i++) {
2932		cheight = cheight > copy_height ? copy_height : cheight;
2933		size = (cheight * pitch) / 4;
2934		/* emit reloc before writing cs so that cs is always in consistent state */
2935		radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rsrc->resource, RADEON_USAGE_READ);
2936		radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rdst->resource, RADEON_USAGE_WRITE);
2937		radeon_emit(cs, DMA_PACKET(DMA_PACKET_COPY, 1, 0, size));
2938		radeon_emit(cs, base >> 8);
2939		radeon_emit(cs, (detile << 31) | (array_mode << 27) |
2940				(lbpp << 24) | ((height - 1) << 10) |
2941				pitch_tile_max);
2942		radeon_emit(cs, (slice_tile_max << 12) | (z << 0));
2943		radeon_emit(cs, (x << 3) | (y << 17));
2944		radeon_emit(cs, addr & 0xfffffffc);
2945		radeon_emit(cs, (addr >> 32UL) & 0xff);
2946		copy_height -= cheight;
2947		addr += cheight * pitch;
2948		y += cheight;
2949	}
2950	return TRUE;
2951}
2952
2953static void r600_dma_copy(struct pipe_context *ctx,
2954			  struct pipe_resource *dst,
2955			  unsigned dst_level,
2956			  unsigned dstx, unsigned dsty, unsigned dstz,
2957			  struct pipe_resource *src,
2958			  unsigned src_level,
2959			  const struct pipe_box *src_box)
2960{
2961	struct r600_context *rctx = (struct r600_context *)ctx;
2962	struct r600_texture *rsrc = (struct r600_texture*)src;
2963	struct r600_texture *rdst = (struct r600_texture*)dst;
2964	unsigned dst_pitch, src_pitch, bpp, dst_mode, src_mode, copy_height;
2965	unsigned src_w, dst_w;
2966	unsigned src_x, src_y;
2967	unsigned dst_x = dstx, dst_y = dsty, dst_z = dstz;
2968
2969	if (rctx->b.dma.cs.priv == NULL) {
2970		goto fallback;
2971	}
2972
2973	if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
2974		if (dst_x % 4 || src_box->x % 4 || src_box->width % 4)
2975			goto fallback;
2976
2977		r600_dma_copy_buffer(rctx, dst, src, dst_x, src_box->x, src_box->width);
2978		return;
2979	}
2980
2981	if (src_box->depth > 1 ||
2982	    !r600_prepare_for_dma_blit(&rctx->b, rdst, dst_level, dstx, dsty,
2983					dstz, rsrc, src_level, src_box))
2984		goto fallback;
2985
2986	src_x = util_format_get_nblocksx(src->format, src_box->x);
2987	dst_x = util_format_get_nblocksx(src->format, dst_x);
2988	src_y = util_format_get_nblocksy(src->format, src_box->y);
2989	dst_y = util_format_get_nblocksy(src->format, dst_y);
2990
2991	bpp = rdst->surface.bpe;
2992	dst_pitch = rdst->surface.u.legacy.level[dst_level].nblk_x * rdst->surface.bpe;
2993	src_pitch = rsrc->surface.u.legacy.level[src_level].nblk_x * rsrc->surface.bpe;
2994	src_w = u_minify(rsrc->resource.b.b.width0, src_level);
2995	dst_w = u_minify(rdst->resource.b.b.width0, dst_level);
2996	copy_height = src_box->height / rsrc->surface.blk_h;
2997
2998	dst_mode = rdst->surface.u.legacy.level[dst_level].mode;
2999	src_mode = rsrc->surface.u.legacy.level[src_level].mode;
3000
3001	if (src_pitch != dst_pitch || src_box->x || dst_x || src_w != dst_w) {
3002		/* strict requirement on r6xx/r7xx */
3003		goto fallback;
3004	}
3005	/* lot of constraint on alignment this should capture them all */
3006	if (src_pitch % 8 || src_box->y % 8 || dst_y % 8) {
3007		goto fallback;
3008	}
3009
3010	if (src_mode == dst_mode) {
3011		uint64_t dst_offset, src_offset, size;
3012
3013		/* simple dma blit would do NOTE code here assume :
3014		 *   src_box.x/y == 0
3015		 *   dst_x/y == 0
3016		 *   dst_pitch == src_pitch
3017		 */
3018		src_offset= (uint64_t)rsrc->surface.u.legacy.level[src_level].offset_256B * 256;
3019		src_offset += (uint64_t)rsrc->surface.u.legacy.level[src_level].slice_size_dw * 4 * src_box->z;
3020		src_offset += src_y * src_pitch + src_x * bpp;
3021		dst_offset = (uint64_t)rdst->surface.u.legacy.level[dst_level].offset_256B * 256;
3022		dst_offset += (uint64_t)rdst->surface.u.legacy.level[dst_level].slice_size_dw * 4 * dst_z;
3023		dst_offset += dst_y * dst_pitch + dst_x * bpp;
3024		size = src_box->height * src_pitch;
3025		/* must be dw aligned */
3026		if (dst_offset % 4 || src_offset % 4 || size % 4) {
3027			goto fallback;
3028		}
3029		r600_dma_copy_buffer(rctx, dst, src, dst_offset, src_offset, size);
3030	} else {
3031		if (!r600_dma_copy_tile(rctx, dst, dst_level, dst_x, dst_y, dst_z,
3032					src, src_level, src_x, src_y, src_box->z,
3033					copy_height, dst_pitch, bpp)) {
3034			goto fallback;
3035		}
3036	}
3037	return;
3038
3039fallback:
3040	r600_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
3041				  src, src_level, src_box);
3042}
3043
3044void r600_init_state_functions(struct r600_context *rctx)
3045{
3046	unsigned id = 1;
3047	unsigned i;
3048	/* !!!
3049	 *  To avoid GPU lockup registers must be emited in a specific order
3050	 * (no kidding ...). The order below is important and have been
3051	 * partialy infered from analyzing fglrx command stream.
3052	 *
3053	 * Don't reorder atom without carefully checking the effect (GPU lockup
3054	 * or piglit regression).
3055	 * !!!
3056	 */
3057
3058	r600_init_atom(rctx, &rctx->framebuffer.atom, id++, r600_emit_framebuffer_state, 0);
3059
3060	/* shader const */
3061	r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX].atom, id++, r600_emit_vs_constant_buffers, 0);
3062	r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY].atom, id++, r600_emit_gs_constant_buffers, 0);
3063	r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT].atom, id++, r600_emit_ps_constant_buffers, 0);
3064
3065	/* sampler must be emited before TA_CNTL_AUX otherwise DISABLE_CUBE_WRAP change
3066	 * does not take effect (TA_CNTL_AUX emited by r600_emit_seamless_cube_map)
3067	 */
3068	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].states.atom, id++, r600_emit_vs_sampler_states, 0);
3069	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].states.atom, id++, r600_emit_gs_sampler_states, 0);
3070	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].states.atom, id++, r600_emit_ps_sampler_states, 0);
3071	/* resource */
3072	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views.atom, id++, r600_emit_vs_sampler_views, 0);
3073	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views.atom, id++, r600_emit_gs_sampler_views, 0);
3074	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views.atom, id++, r600_emit_ps_sampler_views, 0);
3075	r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, r600_emit_vertex_buffers, 0);
3076
3077	r600_init_atom(rctx, &rctx->vgt_state.atom, id++, r600_emit_vgt_state, 10);
3078
3079	r600_init_atom(rctx, &rctx->seamless_cube_map.atom, id++, r600_emit_seamless_cube_map, 3);
3080	r600_init_atom(rctx, &rctx->sample_mask.atom, id++, r600_emit_sample_mask, 3);
3081	rctx->sample_mask.sample_mask = ~0;
3082
3083	r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6);
3084	r600_init_atom(rctx, &rctx->blend_color.atom, id++, r600_emit_blend_color, 6);
3085	r600_init_atom(rctx, &rctx->blend_state.atom, id++, r600_emit_cso_state, 0);
3086	r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, r600_emit_cb_misc_state, 7);
3087	r600_init_atom(rctx, &rctx->clip_misc_state.atom, id++, r600_emit_clip_misc_state, 6);
3088	r600_init_atom(rctx, &rctx->clip_state.atom, id++, r600_emit_clip_state, 26);
3089	r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, r600_emit_db_misc_state, 7);
3090	r600_init_atom(rctx, &rctx->db_state.atom, id++, r600_emit_db_state, 11);
3091	r600_init_atom(rctx, &rctx->dsa_state.atom, id++, r600_emit_cso_state, 0);
3092	r600_init_atom(rctx, &rctx->poly_offset_state.atom, id++, r600_emit_polygon_offset, 9);
3093	r600_init_atom(rctx, &rctx->rasterizer_state.atom, id++, r600_emit_cso_state, 0);
3094	r600_add_atom(rctx, &rctx->b.scissors.atom, id++);
3095	r600_add_atom(rctx, &rctx->b.viewports.atom, id++);
3096	r600_init_atom(rctx, &rctx->config_state.atom, id++, r600_emit_config_state, 3);
3097	r600_init_atom(rctx, &rctx->stencil_ref.atom, id++, r600_emit_stencil_ref, 4);
3098	r600_init_atom(rctx, &rctx->vertex_fetch_shader.atom, id++, r600_emit_vertex_fetch_shader, 5);
3099	r600_add_atom(rctx, &rctx->b.render_cond_atom, id++);
3100	r600_add_atom(rctx, &rctx->b.streamout.begin_atom, id++);
3101	r600_add_atom(rctx, &rctx->b.streamout.enable_atom, id++);
3102	for (i = 0; i < R600_NUM_HW_STAGES; i++)
3103		r600_init_atom(rctx, &rctx->hw_shader_stages[i].atom, id++, r600_emit_shader, 0);
3104	r600_init_atom(rctx, &rctx->shader_stages.atom, id++, r600_emit_shader_stages, 0);
3105	r600_init_atom(rctx, &rctx->gs_rings.atom, id++, r600_emit_gs_rings, 0);
3106
3107	rctx->b.b.create_blend_state = r600_create_blend_state;
3108	rctx->b.b.create_depth_stencil_alpha_state = r600_create_dsa_state;
3109	rctx->b.b.create_rasterizer_state = r600_create_rs_state;
3110	rctx->b.b.create_sampler_state = r600_create_sampler_state;
3111	rctx->b.b.create_sampler_view = r600_create_sampler_view;
3112	rctx->b.b.set_framebuffer_state = r600_set_framebuffer_state;
3113	rctx->b.b.set_polygon_stipple = r600_set_polygon_stipple;
3114	rctx->b.b.set_min_samples = r600_set_min_samples;
3115	rctx->b.b.get_sample_position = r600_get_sample_position;
3116	rctx->b.dma_copy = r600_dma_copy;
3117}
3118/* this function must be last */
3119