• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
r600_translate_blend_function(int blend_func)33 static 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 
r600_translate_blend_factor(int blend_fact)54 static 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 
r600_tex_dim(unsigned dim,unsigned nr_samples)103 static 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 
r600_translate_dbformat(enum pipe_format format)126 static 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 
r600_is_sampler_format_supported(struct pipe_screen * screen,enum pipe_format format)144 static 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 
r600_is_colorbuffer_format_supported(enum chip_class chip,enum pipe_format format)150 static bool r600_is_colorbuffer_format_supported(enum chip_class chip, enum pipe_format format)
151 {
152 	return r600_translate_colorformat(chip, format, FALSE) != ~0U &&
153 	       r600_translate_colorswap(format, FALSE) != ~0U;
154 }
155 
r600_is_zs_format_supported(enum pipe_format format)156 static bool r600_is_zs_format_supported(enum pipe_format format)
157 {
158 	return r600_translate_dbformat(format) != ~0U;
159 }
160 
r600_is_format_supported(struct pipe_screen * screen,enum pipe_format format,enum pipe_texture_target target,unsigned sample_count,unsigned storage_sample_count,unsigned usage)161 bool 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.chip_class == 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_vertex_format_supported(format))
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.chip_class, 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_vertex_format_supported(format)) {
236 		retval |= PIPE_BIND_VERTEX_BUFFER;
237 	}
238 
239 	if ((usage & PIPE_BIND_LINEAR) &&
240 	    !util_format_is_compressed(format) &&
241 	    !(usage & PIPE_BIND_DEPTH_STENCIL))
242 		retval |= PIPE_BIND_LINEAR;
243 
244 	return retval == usage;
245 }
246 
r600_emit_polygon_offset(struct r600_context * rctx,struct r600_atom * a)247 static void r600_emit_polygon_offset(struct r600_context *rctx, struct r600_atom *a)
248 {
249 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
250 	struct r600_poly_offset_state *state = (struct r600_poly_offset_state*)a;
251 	float offset_units = state->offset_units;
252 	float offset_scale = state->offset_scale;
253 	uint32_t pa_su_poly_offset_db_fmt_cntl = 0;
254 
255 	if (!state->offset_units_unscaled) {
256 		switch (state->zs_format) {
257 		case PIPE_FORMAT_Z24X8_UNORM:
258 		case PIPE_FORMAT_Z24_UNORM_S8_UINT:
259 			offset_units *= 2.0f;
260 			pa_su_poly_offset_db_fmt_cntl =
261 				S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-24);
262 			break;
263 		case PIPE_FORMAT_Z16_UNORM:
264 			offset_units *= 4.0f;
265 			pa_su_poly_offset_db_fmt_cntl =
266 				S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-16);
267 			break;
268 		default:
269 			pa_su_poly_offset_db_fmt_cntl =
270 				S_028DF8_POLY_OFFSET_NEG_NUM_DB_BITS((char)-23) |
271 				S_028DF8_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
272 		}
273 	}
274 
275 	radeon_set_context_reg_seq(cs, R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 4);
276 	radeon_emit(cs, fui(offset_scale));
277 	radeon_emit(cs, fui(offset_units));
278 	radeon_emit(cs, fui(offset_scale));
279 	radeon_emit(cs, fui(offset_units));
280 
281 	radeon_set_context_reg(cs, R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL,
282 			       pa_su_poly_offset_db_fmt_cntl);
283 }
284 
r600_get_blend_control(const struct pipe_blend_state * state,unsigned i)285 static uint32_t r600_get_blend_control(const struct pipe_blend_state *state, unsigned i)
286 {
287 	int j = state->independent_blend_enable ? i : 0;
288 
289 	unsigned eqRGB = state->rt[j].rgb_func;
290 	unsigned srcRGB = state->rt[j].rgb_src_factor;
291 	unsigned dstRGB = state->rt[j].rgb_dst_factor;
292 
293 	unsigned eqA = state->rt[j].alpha_func;
294 	unsigned srcA = state->rt[j].alpha_src_factor;
295 	unsigned dstA = state->rt[j].alpha_dst_factor;
296 	uint32_t bc = 0;
297 
298 	if (!state->rt[j].blend_enable)
299 		return 0;
300 
301 	bc |= S_028804_COLOR_COMB_FCN(r600_translate_blend_function(eqRGB));
302 	bc |= S_028804_COLOR_SRCBLEND(r600_translate_blend_factor(srcRGB));
303 	bc |= S_028804_COLOR_DESTBLEND(r600_translate_blend_factor(dstRGB));
304 
305 	if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
306 		bc |= S_028804_SEPARATE_ALPHA_BLEND(1);
307 		bc |= S_028804_ALPHA_COMB_FCN(r600_translate_blend_function(eqA));
308 		bc |= S_028804_ALPHA_SRCBLEND(r600_translate_blend_factor(srcA));
309 		bc |= S_028804_ALPHA_DESTBLEND(r600_translate_blend_factor(dstA));
310 	}
311 	return bc;
312 }
313 
r600_create_blend_state_mode(struct pipe_context * ctx,const struct pipe_blend_state * state,int mode)314 static void *r600_create_blend_state_mode(struct pipe_context *ctx,
315 					  const struct pipe_blend_state *state,
316 					  int mode)
317 {
318 	struct r600_context *rctx = (struct r600_context *)ctx;
319 	uint32_t color_control = 0, target_mask = 0;
320 	struct r600_blend_state *blend = CALLOC_STRUCT(r600_blend_state);
321 
322 	if (!blend) {
323 		return NULL;
324 	}
325 
326 	r600_init_command_buffer(&blend->buffer, 20);
327 	r600_init_command_buffer(&blend->buffer_no_blend, 20);
328 
329 	/* R600 does not support per-MRT blends */
330 	if (rctx->b.family > CHIP_R600)
331 		color_control |= S_028808_PER_MRT_BLEND(1);
332 
333 	if (state->logicop_enable) {
334 		color_control |= (state->logicop_func << 16) | (state->logicop_func << 20);
335 	} else {
336 		color_control |= (0xcc << 16);
337 	}
338 	/* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
339 	if (state->independent_blend_enable) {
340 		for (int i = 0; i < 8; i++) {
341 			if (state->rt[i].blend_enable) {
342 				color_control |= S_028808_TARGET_BLEND_ENABLE(1 << i);
343 			}
344 			target_mask |= (state->rt[i].colormask << (4 * i));
345 		}
346 	} else {
347 		for (int i = 0; i < 8; i++) {
348 			if (state->rt[0].blend_enable) {
349 				color_control |= S_028808_TARGET_BLEND_ENABLE(1 << i);
350 			}
351 			target_mask |= (state->rt[0].colormask << (4 * i));
352 		}
353 	}
354 
355 	if (target_mask)
356 		color_control |= S_028808_SPECIAL_OP(mode);
357 	else
358 		color_control |= S_028808_SPECIAL_OP(V_028808_DISABLE);
359 
360 	/* only MRT0 has dual src blend */
361 	blend->dual_src_blend = util_blend_state_is_dual(state, 0);
362 	blend->cb_target_mask = target_mask;
363 	blend->cb_color_control = color_control;
364 	blend->cb_color_control_no_blend = color_control & C_028808_TARGET_BLEND_ENABLE;
365 	blend->alpha_to_one = state->alpha_to_one;
366 
367 	r600_store_context_reg(&blend->buffer, R_028D44_DB_ALPHA_TO_MASK,
368 			       S_028D44_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
369 			       S_028D44_ALPHA_TO_MASK_OFFSET0(2) |
370 			       S_028D44_ALPHA_TO_MASK_OFFSET1(2) |
371 			       S_028D44_ALPHA_TO_MASK_OFFSET2(2) |
372 			       S_028D44_ALPHA_TO_MASK_OFFSET3(2));
373 
374 	/* Copy over the registers set so far into buffer_no_blend. */
375 	memcpy(blend->buffer_no_blend.buf, blend->buffer.buf, blend->buffer.num_dw * 4);
376 	blend->buffer_no_blend.num_dw = blend->buffer.num_dw;
377 
378 	/* Only add blend registers if blending is enabled. */
379 	if (!G_028808_TARGET_BLEND_ENABLE(color_control)) {
380 		return blend;
381 	}
382 
383 	/* The first R600 does not support per-MRT blends */
384 	r600_store_context_reg(&blend->buffer, R_028804_CB_BLEND_CONTROL,
385 			       r600_get_blend_control(state, 0));
386 
387 	if (rctx->b.family > CHIP_R600) {
388 		r600_store_context_reg_seq(&blend->buffer, R_028780_CB_BLEND0_CONTROL, 8);
389 		for (int i = 0; i < 8; i++) {
390 			r600_store_value(&blend->buffer, r600_get_blend_control(state, i));
391 		}
392 	}
393 	return blend;
394 }
395 
r600_create_blend_state(struct pipe_context * ctx,const struct pipe_blend_state * state)396 static void *r600_create_blend_state(struct pipe_context *ctx,
397 				     const struct pipe_blend_state *state)
398 {
399 	return r600_create_blend_state_mode(ctx, state, V_028808_SPECIAL_NORMAL);
400 }
401 
r600_create_dsa_state(struct pipe_context * ctx,const struct pipe_depth_stencil_alpha_state * state)402 static void *r600_create_dsa_state(struct pipe_context *ctx,
403 				   const struct pipe_depth_stencil_alpha_state *state)
404 {
405 	unsigned db_depth_control, alpha_test_control, alpha_ref;
406 	struct r600_dsa_state *dsa = CALLOC_STRUCT(r600_dsa_state);
407 
408 	if (!dsa) {
409 		return NULL;
410 	}
411 
412 	r600_init_command_buffer(&dsa->buffer, 3);
413 
414 	dsa->valuemask[0] = state->stencil[0].valuemask;
415 	dsa->valuemask[1] = state->stencil[1].valuemask;
416 	dsa->writemask[0] = state->stencil[0].writemask;
417 	dsa->writemask[1] = state->stencil[1].writemask;
418 	dsa->zwritemask = state->depth.writemask;
419 
420 	db_depth_control = S_028800_Z_ENABLE(state->depth.enabled) |
421 		S_028800_Z_WRITE_ENABLE(state->depth.writemask) |
422 		S_028800_ZFUNC(state->depth.func);
423 
424 	/* stencil */
425 	if (state->stencil[0].enabled) {
426 		db_depth_control |= S_028800_STENCIL_ENABLE(1);
427 		db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func); /* translates straight */
428 		db_depth_control |= S_028800_STENCILFAIL(r600_translate_stencil_op(state->stencil[0].fail_op));
429 		db_depth_control |= S_028800_STENCILZPASS(r600_translate_stencil_op(state->stencil[0].zpass_op));
430 		db_depth_control |= S_028800_STENCILZFAIL(r600_translate_stencil_op(state->stencil[0].zfail_op));
431 
432 		if (state->stencil[1].enabled) {
433 			db_depth_control |= S_028800_BACKFACE_ENABLE(1);
434 			db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func); /* translates straight */
435 			db_depth_control |= S_028800_STENCILFAIL_BF(r600_translate_stencil_op(state->stencil[1].fail_op));
436 			db_depth_control |= S_028800_STENCILZPASS_BF(r600_translate_stencil_op(state->stencil[1].zpass_op));
437 			db_depth_control |= S_028800_STENCILZFAIL_BF(r600_translate_stencil_op(state->stencil[1].zfail_op));
438 		}
439 	}
440 
441 	/* alpha */
442 	alpha_test_control = 0;
443 	alpha_ref = 0;
444 	if (state->alpha.enabled) {
445 		alpha_test_control = S_028410_ALPHA_FUNC(state->alpha.func);
446 		alpha_test_control |= S_028410_ALPHA_TEST_ENABLE(1);
447 		alpha_ref = fui(state->alpha.ref_value);
448 	}
449 	dsa->sx_alpha_test_control = alpha_test_control & 0xff;
450 	dsa->alpha_ref = alpha_ref;
451 
452 	r600_store_context_reg(&dsa->buffer, R_028800_DB_DEPTH_CONTROL, db_depth_control);
453 	return dsa;
454 }
455 
r600_create_rs_state(struct pipe_context * ctx,const struct pipe_rasterizer_state * state)456 static void *r600_create_rs_state(struct pipe_context *ctx,
457 				  const struct pipe_rasterizer_state *state)
458 {
459 	struct r600_context *rctx = (struct r600_context *)ctx;
460 	unsigned tmp, sc_mode_cntl, spi_interp;
461 	float psize_min, psize_max;
462 	struct r600_rasterizer_state *rs = CALLOC_STRUCT(r600_rasterizer_state);
463 
464 	if (!rs) {
465 		return NULL;
466 	}
467 
468 	r600_init_command_buffer(&rs->buffer, 30);
469 
470 	rs->scissor_enable = state->scissor;
471 	rs->clip_halfz = state->clip_halfz;
472 	rs->flatshade = state->flatshade;
473 	rs->sprite_coord_enable = state->sprite_coord_enable;
474 	rs->rasterizer_discard = state->rasterizer_discard;
475 	rs->two_side = state->light_twoside;
476 	rs->clip_plane_enable = state->clip_plane_enable;
477 	rs->pa_sc_line_stipple = state->line_stipple_enable ?
478 				S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
479 				S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
480 	rs->pa_cl_clip_cntl =
481 		S_028810_DX_CLIP_SPACE_DEF(state->clip_halfz) |
482 		S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip_near) |
483 		S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip_far) |
484 		S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
485 	if (rctx->b.chip_class == R700) {
486 		rs->pa_cl_clip_cntl |=
487 			S_028810_DX_RASTERIZATION_KILL(state->rasterizer_discard);
488 	}
489 	rs->multisample_enable = state->multisample;
490 
491 	/* offset */
492 	rs->offset_units = state->offset_units;
493 	rs->offset_scale = state->offset_scale * 16.0f;
494 	rs->offset_enable = state->offset_point || state->offset_line || state->offset_tri;
495 	rs->offset_units_unscaled = state->offset_units_unscaled;
496 
497 	if (state->point_size_per_vertex) {
498 		psize_min = util_get_min_point_size(state);
499 		psize_max = 8192;
500 	} else {
501 		/* Force the point size to be as if the vertex output was disabled. */
502 		psize_min = state->point_size;
503 		psize_max = state->point_size;
504 	}
505 
506 	sc_mode_cntl = S_028A4C_MSAA_ENABLE(state->multisample) |
507 		       S_028A4C_LINE_STIPPLE_ENABLE(state->line_stipple_enable) |
508 		       S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) |
509 		       S_028A4C_PS_ITER_SAMPLE(state->multisample && rctx->ps_iter_samples > 1);
510 	if (rctx->b.family == CHIP_RV770) {
511 		/* workaround possible rendering corruption on RV770 with hyperz together with sample shading */
512 		sc_mode_cntl |= S_028A4C_TILE_COVER_DISABLE(state->multisample && rctx->ps_iter_samples > 1);
513 	}
514 	if (rctx->b.chip_class >= R700) {
515 		sc_mode_cntl |= S_028A4C_FORCE_EOV_REZ_ENABLE(1) |
516 				S_028A4C_R700_ZMM_LINE_OFFSET(1) |
517 				S_028A4C_R700_VPORT_SCISSOR_ENABLE(1);
518 	} else {
519 		sc_mode_cntl |= S_028A4C_WALK_ALIGN8_PRIM_FITS_ST(1);
520 	}
521 
522 	spi_interp = S_0286D4_FLAT_SHADE_ENA(1);
523 	spi_interp |= S_0286D4_PNT_SPRITE_ENA(1) |
524 		S_0286D4_PNT_SPRITE_OVRD_X(2) |
525 		S_0286D4_PNT_SPRITE_OVRD_Y(3) |
526 		S_0286D4_PNT_SPRITE_OVRD_Z(0) |
527 		S_0286D4_PNT_SPRITE_OVRD_W(1);
528 	if (state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT) {
529 		spi_interp |= S_0286D4_PNT_SPRITE_TOP_1(1);
530 	}
531 
532 	r600_store_context_reg_seq(&rs->buffer, R_028A00_PA_SU_POINT_SIZE, 3);
533 	/* point size 12.4 fixed point (divide by two, because 0.5 = 1 pixel. */
534 	tmp = r600_pack_float_12p4(state->point_size/2);
535 	r600_store_value(&rs->buffer, /* R_028A00_PA_SU_POINT_SIZE */
536 			 S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
537 	r600_store_value(&rs->buffer, /* R_028A04_PA_SU_POINT_MINMAX */
538 			 S_028A04_MIN_SIZE(r600_pack_float_12p4(psize_min/2)) |
539 			 S_028A04_MAX_SIZE(r600_pack_float_12p4(psize_max/2)));
540 	r600_store_value(&rs->buffer, /* R_028A08_PA_SU_LINE_CNTL */
541 			 S_028A08_WIDTH(r600_pack_float_12p4(state->line_width/2)));
542 
543 	r600_store_context_reg(&rs->buffer, R_0286D4_SPI_INTERP_CONTROL_0, spi_interp);
544 	r600_store_context_reg(&rs->buffer, R_028A4C_PA_SC_MODE_CNTL, sc_mode_cntl);
545 	r600_store_context_reg(&rs->buffer, R_028C08_PA_SU_VTX_CNTL,
546 			       S_028C08_PIX_CENTER_HALF(state->half_pixel_center) |
547 			       S_028C08_QUANT_MODE(V_028C08_X_1_256TH));
548 	r600_store_context_reg(&rs->buffer, R_028DFC_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp));
549 
550 	rs->pa_su_sc_mode_cntl = S_028814_PROVOKING_VTX_LAST(!state->flatshade_first) |
551 				 S_028814_CULL_FRONT(state->cull_face & PIPE_FACE_FRONT ? 1 : 0) |
552 				 S_028814_CULL_BACK(state->cull_face & PIPE_FACE_BACK ? 1 : 0) |
553 				 S_028814_FACE(!state->front_ccw) |
554 				 S_028814_POLY_OFFSET_FRONT_ENABLE(util_get_offset(state, state->fill_front)) |
555 				 S_028814_POLY_OFFSET_BACK_ENABLE(util_get_offset(state, state->fill_back)) |
556 				 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_point || state->offset_line) |
557 				 S_028814_POLY_MODE(state->fill_front != PIPE_POLYGON_MODE_FILL ||
558 									 state->fill_back != PIPE_POLYGON_MODE_FILL) |
559 				 S_028814_POLYMODE_FRONT_PTYPE(r600_translate_fill(state->fill_front)) |
560 				 S_028814_POLYMODE_BACK_PTYPE(r600_translate_fill(state->fill_back));
561 	if (rctx->b.chip_class == R700) {
562 		r600_store_context_reg(&rs->buffer, R_028814_PA_SU_SC_MODE_CNTL, rs->pa_su_sc_mode_cntl);
563 	}
564 	if (rctx->b.chip_class == R600) {
565 		r600_store_context_reg(&rs->buffer, R_028350_SX_MISC,
566 				       S_028350_MULTIPASS(state->rasterizer_discard));
567 	}
568 	return rs;
569 }
570 
r600_tex_filter(unsigned filter,unsigned max_aniso)571 static unsigned r600_tex_filter(unsigned filter, unsigned max_aniso)
572 {
573 	if (filter == PIPE_TEX_FILTER_LINEAR)
574 		return max_aniso > 1 ? V_03C000_SQ_TEX_XY_FILTER_ANISO_BILINEAR
575 				     : V_03C000_SQ_TEX_XY_FILTER_BILINEAR;
576 	else
577 		return max_aniso > 1 ? V_03C000_SQ_TEX_XY_FILTER_ANISO_POINT
578 				     : V_03C000_SQ_TEX_XY_FILTER_POINT;
579 }
580 
r600_create_sampler_state(struct pipe_context * ctx,const struct pipe_sampler_state * state)581 static void *r600_create_sampler_state(struct pipe_context *ctx,
582 					const struct pipe_sampler_state *state)
583 {
584 	struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen;
585 	struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state);
586 	unsigned max_aniso = rscreen->force_aniso >= 0 ? rscreen->force_aniso
587 						       : state->max_anisotropy;
588 	unsigned max_aniso_ratio = r600_tex_aniso_filter(max_aniso);
589 
590 	if (!ss) {
591 		return NULL;
592 	}
593 
594 	ss->seamless_cube_map = state->seamless_cube_map;
595 	ss->border_color_use = sampler_state_needs_border_color(state);
596 
597 	/* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */
598 	ss->tex_sampler_words[0] =
599 		S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
600 		S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
601 		S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
602 		S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter, max_aniso)) |
603 		S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter, max_aniso)) |
604 		S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
605 		S_03C000_MAX_ANISO_RATIO(max_aniso_ratio) |
606 		S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
607 		S_03C000_BORDER_COLOR_TYPE(ss->border_color_use ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0);
608 	/* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */
609 	ss->tex_sampler_words[1] =
610 		S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 6)) |
611 		S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 6)) |
612 		S_03C004_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 6));
613 	/* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */
614 	ss->tex_sampler_words[2] = S_03C008_TYPE(1);
615 
616 	if (ss->border_color_use) {
617 		memcpy(&ss->border_color, &state->border_color, sizeof(state->border_color));
618 	}
619 	return ss;
620 }
621 
622 static struct pipe_sampler_view *
texture_buffer_sampler_view(struct r600_pipe_sampler_view * view,unsigned width0,unsigned height0)623 texture_buffer_sampler_view(struct r600_pipe_sampler_view *view,
624 			    unsigned width0, unsigned height0)
625 
626 {
627 	struct r600_texture *tmp = (struct r600_texture*)view->base.texture;
628 	int stride = util_format_get_blocksize(view->base.format);
629 	unsigned format, num_format, format_comp, endian;
630 	uint64_t offset = view->base.u.buf.offset;
631 	unsigned size = view->base.u.buf.size;
632 
633 	r600_vertex_data_type(view->base.format,
634 			      &format, &num_format, &format_comp,
635 			      &endian);
636 
637 	view->tex_resource = &tmp->resource;
638 	view->skip_mip_address_reloc = true;
639 
640 	view->tex_resource_words[0] = offset;
641 	view->tex_resource_words[1] = size - 1;
642 	view->tex_resource_words[2] = S_038008_BASE_ADDRESS_HI(offset >> 32UL) |
643 		S_038008_STRIDE(stride) |
644 		S_038008_DATA_FORMAT(format) |
645 		S_038008_NUM_FORMAT_ALL(num_format) |
646 		S_038008_FORMAT_COMP_ALL(format_comp) |
647 		S_038008_ENDIAN_SWAP(endian);
648 	view->tex_resource_words[3] = 0;
649 	/*
650 	 * in theory dword 4 is for number of elements, for use with resinfo,
651 	 * but it seems to utterly fail to work, the amd gpu shader analyser
652 	 * uses a const buffer to store the element sizes for buffer txq
653 	 */
654 	view->tex_resource_words[4] = 0;
655 	view->tex_resource_words[5] = 0;
656 	view->tex_resource_words[6] = S_038018_TYPE(V_038010_SQ_TEX_VTX_VALID_BUFFER);
657 	return &view->base;
658 }
659 
660 struct pipe_sampler_view *
r600_create_sampler_view_custom(struct pipe_context * ctx,struct pipe_resource * texture,const struct pipe_sampler_view * state,unsigned width_first_level,unsigned height_first_level)661 r600_create_sampler_view_custom(struct pipe_context *ctx,
662 				struct pipe_resource *texture,
663 				const struct pipe_sampler_view *state,
664 				unsigned width_first_level, unsigned height_first_level)
665 {
666 	struct r600_pipe_sampler_view *view = CALLOC_STRUCT(r600_pipe_sampler_view);
667 	struct r600_texture *tmp = (struct r600_texture*)texture;
668 	unsigned format, endian;
669 	uint32_t word4 = 0, yuv_format = 0, pitch = 0;
670 	unsigned char swizzle[4], array_mode = 0;
671 	unsigned width, height, depth, offset_level, last_level;
672 	bool do_endian_swap = FALSE;
673 
674 	if (!view)
675 		return NULL;
676 
677 	/* initialize base object */
678 	view->base = *state;
679 	view->base.texture = NULL;
680 	pipe_reference(NULL, &texture->reference);
681 	view->base.texture = texture;
682 	view->base.reference.count = 1;
683 	view->base.context = ctx;
684 
685 	if (texture->target == PIPE_BUFFER)
686 		return texture_buffer_sampler_view(view, texture->width0, 1);
687 
688 	swizzle[0] = state->swizzle_r;
689 	swizzle[1] = state->swizzle_g;
690 	swizzle[2] = state->swizzle_b;
691 	swizzle[3] = state->swizzle_a;
692 
693 	if (R600_BIG_ENDIAN)
694 		do_endian_swap = !tmp->db_compatible;
695 
696 	format = r600_translate_texformat(ctx->screen, state->format,
697 					  swizzle,
698 					  &word4, &yuv_format, do_endian_swap);
699 	assert(format != ~0);
700 	if (format == ~0) {
701 		FREE(view);
702 		return NULL;
703 	}
704 
705 	if (state->format == PIPE_FORMAT_X24S8_UINT ||
706 	    state->format == PIPE_FORMAT_S8X24_UINT ||
707 	    state->format == PIPE_FORMAT_X32_S8X24_UINT ||
708 	    state->format == PIPE_FORMAT_S8_UINT)
709 		view->is_stencil_sampler = true;
710 
711 	if (tmp->is_depth && !r600_can_sample_zs(tmp, view->is_stencil_sampler)) {
712 		if (!r600_init_flushed_depth_texture(ctx, texture, NULL)) {
713 			FREE(view);
714 			return NULL;
715 		}
716 		tmp = tmp->flushed_depth_texture;
717 	}
718 
719 	endian = r600_colorformat_endian_swap(format, do_endian_swap);
720 
721 	offset_level = state->u.tex.first_level;
722 	last_level = state->u.tex.last_level - offset_level;
723 	width = width_first_level;
724 	height = height_first_level;
725         depth = u_minify(texture->depth0, offset_level);
726 	pitch = tmp->surface.u.legacy.level[offset_level].nblk_x * util_format_get_blockwidth(state->format);
727 
728 	if (texture->target == PIPE_TEXTURE_1D_ARRAY) {
729 		height = 1;
730 		depth = texture->array_size;
731 	} else if (texture->target == PIPE_TEXTURE_2D_ARRAY) {
732 		depth = texture->array_size;
733 	} else if (texture->target == PIPE_TEXTURE_CUBE_ARRAY)
734 		depth = texture->array_size / 6;
735 
736 	switch (tmp->surface.u.legacy.level[offset_level].mode) {
737 	default:
738 	case RADEON_SURF_MODE_LINEAR_ALIGNED:
739 		array_mode = V_038000_ARRAY_LINEAR_ALIGNED;
740 		break;
741 	case RADEON_SURF_MODE_1D:
742 		array_mode = V_038000_ARRAY_1D_TILED_THIN1;
743 		break;
744 	case RADEON_SURF_MODE_2D:
745 		array_mode = V_038000_ARRAY_2D_TILED_THIN1;
746 		break;
747 	}
748 
749 	view->tex_resource = &tmp->resource;
750 	view->tex_resource_words[0] = (S_038000_DIM(r600_tex_dim(texture->target, texture->nr_samples)) |
751 				       S_038000_TILE_MODE(array_mode) |
752 				       S_038000_TILE_TYPE(tmp->non_disp_tiling) |
753 				       S_038000_PITCH((pitch / 8) - 1) |
754 				       S_038000_TEX_WIDTH(width - 1));
755 	view->tex_resource_words[1] = (S_038004_TEX_HEIGHT(height - 1) |
756 				       S_038004_TEX_DEPTH(depth - 1) |
757 				       S_038004_DATA_FORMAT(format));
758 	view->tex_resource_words[2] = tmp->surface.u.legacy.level[offset_level].offset >> 8;
759 	if (offset_level >= tmp->resource.b.b.last_level) {
760 		view->tex_resource_words[3] = tmp->surface.u.legacy.level[offset_level].offset >> 8;
761 	} else {
762 		view->tex_resource_words[3] = tmp->surface.u.legacy.level[offset_level + 1].offset >> 8;
763 	}
764 	view->tex_resource_words[4] = (word4 |
765 				       S_038010_REQUEST_SIZE(1) |
766 				       S_038010_ENDIAN_SWAP(endian) |
767 				       S_038010_BASE_LEVEL(0));
768 	view->tex_resource_words[5] = (S_038014_BASE_ARRAY(state->u.tex.first_layer) |
769 				       S_038014_LAST_ARRAY(state->u.tex.last_layer));
770 	if (texture->nr_samples > 1) {
771 		/* LAST_LEVEL holds log2(nr_samples) for multisample textures */
772 		view->tex_resource_words[5] |= S_038014_LAST_LEVEL(util_logbase2(texture->nr_samples));
773 	} else {
774 		view->tex_resource_words[5] |= S_038014_LAST_LEVEL(last_level);
775 	}
776 	view->tex_resource_words[6] = (S_038018_TYPE(V_038010_SQ_TEX_VTX_VALID_TEXTURE) |
777 				       S_038018_MAX_ANISO(4 /* max 16 samples */));
778 	return &view->base;
779 }
780 
781 static struct pipe_sampler_view *
r600_create_sampler_view(struct pipe_context * ctx,struct pipe_resource * tex,const struct pipe_sampler_view * state)782 r600_create_sampler_view(struct pipe_context *ctx,
783 			 struct pipe_resource *tex,
784 			 const struct pipe_sampler_view *state)
785 {
786 	return r600_create_sampler_view_custom(ctx, tex, state,
787                                                u_minify(tex->width0, state->u.tex.first_level),
788                                                u_minify(tex->height0, state->u.tex.first_level));
789 }
790 
r600_emit_clip_state(struct r600_context * rctx,struct r600_atom * atom)791 static void r600_emit_clip_state(struct r600_context *rctx, struct r600_atom *atom)
792 {
793 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
794 	struct pipe_clip_state *state = &rctx->clip_state.state;
795 
796 	radeon_set_context_reg_seq(cs, R_028E20_PA_CL_UCP0_X, 6*4);
797 	radeon_emit_array(cs, (unsigned*)state, 6*4);
798 }
799 
r600_set_polygon_stipple(struct pipe_context * ctx,const struct pipe_poly_stipple * state)800 static void r600_set_polygon_stipple(struct pipe_context *ctx,
801 					 const struct pipe_poly_stipple *state)
802 {
803 }
804 
r600_init_color_surface(struct r600_context * rctx,struct r600_surface * surf,bool force_cmask_fmask)805 static void r600_init_color_surface(struct r600_context *rctx,
806 				    struct r600_surface *surf,
807 				    bool force_cmask_fmask)
808 {
809 	struct r600_screen *rscreen = rctx->screen;
810 	struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
811 	unsigned level = surf->base.u.tex.level;
812 	unsigned pitch, slice;
813 	unsigned color_info;
814 	unsigned color_view;
815 	unsigned format, swap, ntype, endian;
816 	unsigned offset;
817 	const struct util_format_description *desc;
818 	int i;
819 	bool blend_bypass = 0, blend_clamp = 0, do_endian_swap = FALSE;
820 
821 	if (rtex->db_compatible && !r600_can_sample_zs(rtex, false)) {
822 		r600_init_flushed_depth_texture(&rctx->b.b, surf->base.texture, NULL);
823 		rtex = rtex->flushed_depth_texture;
824 		assert(rtex);
825 	}
826 
827 	offset = rtex->surface.u.legacy.level[level].offset;
828 	color_view = S_028080_SLICE_START(surf->base.u.tex.first_layer) |
829 		     S_028080_SLICE_MAX(surf->base.u.tex.last_layer);
830 
831 	pitch = rtex->surface.u.legacy.level[level].nblk_x / 8 - 1;
832 	slice = (rtex->surface.u.legacy.level[level].nblk_x * rtex->surface.u.legacy.level[level].nblk_y) / 64;
833 	if (slice) {
834 		slice = slice - 1;
835 	}
836 	color_info = 0;
837 	switch (rtex->surface.u.legacy.level[level].mode) {
838 	default:
839 	case RADEON_SURF_MODE_LINEAR_ALIGNED:
840 		color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_LINEAR_ALIGNED);
841 		break;
842 	case RADEON_SURF_MODE_1D:
843 		color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_1D_TILED_THIN1);
844 		break;
845 	case RADEON_SURF_MODE_2D:
846 		color_info = S_0280A0_ARRAY_MODE(V_038000_ARRAY_2D_TILED_THIN1);
847 		break;
848 	}
849 
850 	desc = util_format_description(surf->base.format);
851 
852 	for (i = 0; i < 4; i++) {
853 		if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
854 			break;
855 		}
856 	}
857 
858 	ntype = V_0280A0_NUMBER_UNORM;
859 	if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
860 		ntype = V_0280A0_NUMBER_SRGB;
861 	else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
862 		if (desc->channel[i].normalized)
863 			ntype = V_0280A0_NUMBER_SNORM;
864 		else if (desc->channel[i].pure_integer)
865 			ntype = V_0280A0_NUMBER_SINT;
866 	} else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) {
867 		if (desc->channel[i].normalized)
868 			ntype = V_0280A0_NUMBER_UNORM;
869 		else if (desc->channel[i].pure_integer)
870 			ntype = V_0280A0_NUMBER_UINT;
871 	} else if (desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT) {
872 		ntype = V_0280A0_NUMBER_FLOAT;
873 	}
874 
875 	if (R600_BIG_ENDIAN)
876 		do_endian_swap = !rtex->db_compatible;
877 
878 	format = r600_translate_colorformat(rctx->b.chip_class, surf->base.format,
879 			                              do_endian_swap);
880 	assert(format != ~0);
881 
882 	swap = r600_translate_colorswap(surf->base.format, do_endian_swap);
883 	assert(swap != ~0);
884 
885 	endian = r600_colorformat_endian_swap(format, do_endian_swap);
886 
887 	/* blend clamp should be set for all NORM/SRGB types */
888 	if (ntype == V_0280A0_NUMBER_UNORM || ntype == V_0280A0_NUMBER_SNORM ||
889 	    ntype == V_0280A0_NUMBER_SRGB)
890 		blend_clamp = 1;
891 
892 	/* set blend bypass according to docs if SINT/UINT or
893 	   8/24 COLOR variants */
894 	if (ntype == V_0280A0_NUMBER_UINT || ntype == V_0280A0_NUMBER_SINT ||
895 	    format == V_0280A0_COLOR_8_24 || format == V_0280A0_COLOR_24_8 ||
896 	    format == V_0280A0_COLOR_X24_8_32_FLOAT) {
897 		blend_clamp = 0;
898 		blend_bypass = 1;
899 	}
900 
901 	surf->alphatest_bypass = ntype == V_0280A0_NUMBER_UINT || ntype == V_0280A0_NUMBER_SINT;
902 
903 	color_info |= S_0280A0_FORMAT(format) |
904 		S_0280A0_COMP_SWAP(swap) |
905 		S_0280A0_BLEND_BYPASS(blend_bypass) |
906 		S_0280A0_BLEND_CLAMP(blend_clamp) |
907 		S_0280A0_SIMPLE_FLOAT(1) |
908 		S_0280A0_NUMBER_TYPE(ntype) |
909 		S_0280A0_ENDIAN(endian);
910 
911 	/* EXPORT_NORM is an optimzation that can be enabled for better
912 	 * performance in certain cases
913 	 */
914 	if (rctx->b.chip_class == R600) {
915 		/* EXPORT_NORM can be enabled if:
916 		 * - 11-bit or smaller UNORM/SNORM/SRGB
917 		 * - BLEND_CLAMP is enabled
918 		 * - BLEND_FLOAT32 is disabled
919 		 */
920 		if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
921 		    (desc->channel[i].size < 12 &&
922 		     desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
923 		     ntype != V_0280A0_NUMBER_UINT &&
924 		     ntype != V_0280A0_NUMBER_SINT) &&
925 		    G_0280A0_BLEND_CLAMP(color_info) &&
926 		    /* XXX this condition is always true since BLEND_FLOAT32 is never set (bug?). */
927 		    !G_0280A0_BLEND_FLOAT32(color_info)) {
928 			color_info |= S_0280A0_SOURCE_FORMAT(V_0280A0_EXPORT_NORM);
929 			surf->export_16bpc = true;
930 		}
931 	} else {
932 		/* EXPORT_NORM can be enabled if:
933 		 * - 11-bit or smaller UNORM/SNORM/SRGB
934 		 * - 16-bit or smaller FLOAT
935 		 */
936 		if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS &&
937 		    ((desc->channel[i].size < 12 &&
938 		      desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT &&
939 		      ntype != V_0280A0_NUMBER_UINT && ntype != V_0280A0_NUMBER_SINT) ||
940 		    (desc->channel[i].size < 17 &&
941 		     desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT))) {
942 			color_info |= S_0280A0_SOURCE_FORMAT(V_0280A0_EXPORT_NORM);
943 			surf->export_16bpc = true;
944 		}
945 	}
946 
947 	/* These might not always be initialized to zero. */
948 	surf->cb_color_base = offset >> 8;
949 	surf->cb_color_size = S_028060_PITCH_TILE_MAX(pitch) |
950 			      S_028060_SLICE_TILE_MAX(slice);
951 	surf->cb_color_fmask = surf->cb_color_base;
952 	surf->cb_color_cmask = surf->cb_color_base;
953 	surf->cb_color_mask = 0;
954 
955 	r600_resource_reference(&surf->cb_buffer_cmask, &rtex->resource);
956 	r600_resource_reference(&surf->cb_buffer_fmask, &rtex->resource);
957 
958 	if (rtex->cmask.size) {
959 		surf->cb_color_cmask = rtex->cmask.offset >> 8;
960 		surf->cb_color_mask |= S_028100_CMASK_BLOCK_MAX(rtex->cmask.slice_tile_max);
961 
962 		if (rtex->fmask.size) {
963 			color_info |= S_0280A0_TILE_MODE(V_0280A0_FRAG_ENABLE);
964 			surf->cb_color_fmask = rtex->fmask.offset >> 8;
965 			surf->cb_color_mask |= S_028100_FMASK_TILE_MAX(rtex->fmask.slice_tile_max);
966 		} else { /* cmask only */
967 			color_info |= S_0280A0_TILE_MODE(V_0280A0_CLEAR_ENABLE);
968 		}
969 	} else if (force_cmask_fmask) {
970 		/* Allocate dummy FMASK and CMASK if they aren't allocated already.
971 		 *
972 		 * R6xx needs FMASK and CMASK for the destination buffer of color resolve,
973 		 * otherwise it hangs. We don't have FMASK and CMASK pre-allocated,
974 		 * because it's not an MSAA buffer.
975 		 */
976 		struct r600_cmask_info cmask;
977 		struct r600_fmask_info fmask;
978 
979 		r600_texture_get_cmask_info(&rscreen->b, rtex, &cmask);
980 		r600_texture_get_fmask_info(&rscreen->b, rtex, 8, &fmask);
981 
982 		/* CMASK. */
983 		if (!rctx->dummy_cmask ||
984 		    rctx->dummy_cmask->b.b.width0 < cmask.size ||
985 		    rctx->dummy_cmask->buf->alignment % cmask.alignment != 0) {
986 			struct pipe_transfer *transfer;
987 			void *ptr;
988 
989 			r600_resource_reference(&rctx->dummy_cmask, NULL);
990 			rctx->dummy_cmask = (struct r600_resource*)
991 				r600_aligned_buffer_create(&rscreen->b.b, 0,
992 							   PIPE_USAGE_DEFAULT,
993 							   cmask.size, cmask.alignment);
994 
995 			if (unlikely(!rctx->dummy_cmask)) {
996 				surf->color_initialized = false;
997 				return;
998 			}
999 
1000 			/* Set the contents to 0xCC. */
1001 			ptr = pipe_buffer_map(&rctx->b.b, &rctx->dummy_cmask->b.b, PIPE_MAP_WRITE, &transfer);
1002 			memset(ptr, 0xCC, cmask.size);
1003 			pipe_buffer_unmap(&rctx->b.b, transfer);
1004 		}
1005 		r600_resource_reference(&surf->cb_buffer_cmask, rctx->dummy_cmask);
1006 
1007 		/* FMASK. */
1008 		if (!rctx->dummy_fmask ||
1009 		    rctx->dummy_fmask->b.b.width0 < fmask.size ||
1010 		    rctx->dummy_fmask->buf->alignment % fmask.alignment != 0) {
1011 			r600_resource_reference(&rctx->dummy_fmask, NULL);
1012 			rctx->dummy_fmask = (struct r600_resource*)
1013 				r600_aligned_buffer_create(&rscreen->b.b, 0,
1014 							   PIPE_USAGE_DEFAULT,
1015 							   fmask.size, fmask.alignment);
1016 
1017 			if (unlikely(!rctx->dummy_fmask)) {
1018 				surf->color_initialized = false;
1019 				return;
1020 			}
1021 		}
1022 		r600_resource_reference(&surf->cb_buffer_fmask, rctx->dummy_fmask);
1023 
1024 		/* Init the registers. */
1025 		color_info |= S_0280A0_TILE_MODE(V_0280A0_FRAG_ENABLE);
1026 		surf->cb_color_cmask = 0;
1027 		surf->cb_color_fmask = 0;
1028 		surf->cb_color_mask = S_028100_CMASK_BLOCK_MAX(cmask.slice_tile_max) |
1029 				      S_028100_FMASK_TILE_MAX(fmask.slice_tile_max);
1030 	}
1031 
1032 	surf->cb_color_info = color_info;
1033 	surf->cb_color_view = color_view;
1034 	surf->color_initialized = true;
1035 }
1036 
r600_init_depth_surface(struct r600_context * rctx,struct r600_surface * surf)1037 static void r600_init_depth_surface(struct r600_context *rctx,
1038 				    struct r600_surface *surf)
1039 {
1040 	struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
1041 	unsigned level, pitch, slice, format, offset, array_mode;
1042 
1043 	level = surf->base.u.tex.level;
1044 	offset = rtex->surface.u.legacy.level[level].offset;
1045 	pitch = rtex->surface.u.legacy.level[level].nblk_x / 8 - 1;
1046 	slice = (rtex->surface.u.legacy.level[level].nblk_x * rtex->surface.u.legacy.level[level].nblk_y) / 64;
1047 	if (slice) {
1048 		slice = slice - 1;
1049 	}
1050 	switch (rtex->surface.u.legacy.level[level].mode) {
1051 	case RADEON_SURF_MODE_2D:
1052 		array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
1053 		break;
1054 	case RADEON_SURF_MODE_1D:
1055 	case RADEON_SURF_MODE_LINEAR_ALIGNED:
1056 	default:
1057 		array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
1058 		break;
1059 	}
1060 
1061 	format = r600_translate_dbformat(surf->base.format);
1062 	assert(format != ~0);
1063 
1064 	surf->db_depth_info = S_028010_ARRAY_MODE(array_mode) | S_028010_FORMAT(format);
1065 	surf->db_depth_base = offset >> 8;
1066 	surf->db_depth_view = S_028004_SLICE_START(surf->base.u.tex.first_layer) |
1067 			      S_028004_SLICE_MAX(surf->base.u.tex.last_layer);
1068 	surf->db_depth_size = S_028000_PITCH_TILE_MAX(pitch) | S_028000_SLICE_TILE_MAX(slice);
1069 	surf->db_prefetch_limit = (rtex->surface.u.legacy.level[level].nblk_y / 8) - 1;
1070 
1071 	if (r600_htile_enabled(rtex, level)) {
1072 		surf->db_htile_data_base = rtex->htile_offset >> 8;
1073 		surf->db_htile_surface = S_028D24_HTILE_WIDTH(1) |
1074 					 S_028D24_HTILE_HEIGHT(1) |
1075 					 S_028D24_FULL_CACHE(1);
1076 		/* preload is not working properly on r6xx/r7xx */
1077 		surf->db_depth_info |= S_028010_TILE_SURFACE_ENABLE(1);
1078 	}
1079 
1080 	surf->depth_initialized = true;
1081 }
1082 
r600_set_framebuffer_state(struct pipe_context * ctx,const struct pipe_framebuffer_state * state)1083 static void r600_set_framebuffer_state(struct pipe_context *ctx,
1084 					const struct pipe_framebuffer_state *state)
1085 {
1086 	struct r600_context *rctx = (struct r600_context *)ctx;
1087 	struct r600_surface *surf;
1088 	struct r600_texture *rtex;
1089 	unsigned i;
1090 	uint32_t target_mask = 0;
1091 
1092 	/* Flush TC when changing the framebuffer state, because the only
1093 	 * client not using TC that can change textures is the framebuffer.
1094 	 * Other places don't typically have to flush TC.
1095 	 */
1096 	rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE |
1097 			 R600_CONTEXT_FLUSH_AND_INV |
1098 			 R600_CONTEXT_FLUSH_AND_INV_CB |
1099 			 R600_CONTEXT_FLUSH_AND_INV_CB_META |
1100 			 R600_CONTEXT_FLUSH_AND_INV_DB |
1101 			 R600_CONTEXT_FLUSH_AND_INV_DB_META |
1102 			 R600_CONTEXT_INV_TEX_CACHE;
1103 
1104 	/* Set the new state. */
1105 	util_copy_framebuffer_state(&rctx->framebuffer.state, state);
1106 
1107 	rctx->framebuffer.export_16bpc = state->nr_cbufs != 0;
1108 	rctx->framebuffer.cb0_is_integer = state->nr_cbufs && state->cbufs[0] &&
1109 			       util_format_is_pure_integer(state->cbufs[0]->format);
1110 	rctx->framebuffer.compressed_cb_mask = 0;
1111 	rctx->framebuffer.is_msaa_resolve = state->nr_cbufs == 2 &&
1112 					    state->cbufs[0] && state->cbufs[1] &&
1113 					    state->cbufs[0]->texture->nr_samples > 1 &&
1114 				            state->cbufs[1]->texture->nr_samples <= 1;
1115 	rctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
1116 
1117 	/* Colorbuffers. */
1118 	for (i = 0; i < state->nr_cbufs; i++) {
1119 		/* The resolve buffer must have CMASK and FMASK to prevent hardlocks on R6xx. */
1120 		bool force_cmask_fmask = rctx->b.chip_class == R600 &&
1121 					 rctx->framebuffer.is_msaa_resolve &&
1122 					 i == 1;
1123 
1124 		surf = (struct r600_surface*)state->cbufs[i];
1125 		if (!surf)
1126 			continue;
1127 
1128 		rtex = (struct r600_texture*)surf->base.texture;
1129 		r600_context_add_resource_size(ctx, state->cbufs[i]->texture);
1130 
1131 		target_mask |= (0xf << (i * 4));
1132 
1133 		if (!surf->color_initialized || force_cmask_fmask) {
1134 			r600_init_color_surface(rctx, surf, force_cmask_fmask);
1135 			if (force_cmask_fmask) {
1136 				/* re-initialize later without compression */
1137 				surf->color_initialized = false;
1138 			}
1139 		}
1140 
1141 		if (!surf->export_16bpc) {
1142 			rctx->framebuffer.export_16bpc = false;
1143 		}
1144 
1145 		if (rtex->fmask.size) {
1146 			rctx->framebuffer.compressed_cb_mask |= 1 << i;
1147 		}
1148 	}
1149 
1150 	/* Update alpha-test state dependencies.
1151 	 * Alpha-test is done on the first colorbuffer only. */
1152 	if (state->nr_cbufs) {
1153 		bool alphatest_bypass = false;
1154 
1155 		surf = (struct r600_surface*)state->cbufs[0];
1156 		if (surf) {
1157 			alphatest_bypass = surf->alphatest_bypass;
1158 		}
1159 
1160 		if (rctx->alphatest_state.bypass != alphatest_bypass) {
1161 			rctx->alphatest_state.bypass = alphatest_bypass;
1162 			r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1163 		}
1164 	}
1165 
1166 	/* ZS buffer. */
1167 	if (state->zsbuf) {
1168 		surf = (struct r600_surface*)state->zsbuf;
1169 
1170 		r600_context_add_resource_size(ctx, state->zsbuf->texture);
1171 
1172 		if (!surf->depth_initialized) {
1173 			r600_init_depth_surface(rctx, surf);
1174 		}
1175 
1176 		if (state->zsbuf->format != rctx->poly_offset_state.zs_format) {
1177 			rctx->poly_offset_state.zs_format = state->zsbuf->format;
1178 			r600_mark_atom_dirty(rctx, &rctx->poly_offset_state.atom);
1179 		}
1180 
1181 		if (rctx->db_state.rsurf != surf) {
1182 			rctx->db_state.rsurf = surf;
1183 			r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1184 			r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1185 		}
1186 	} else if (rctx->db_state.rsurf) {
1187 		rctx->db_state.rsurf = NULL;
1188 		r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
1189 		r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1190 	}
1191 
1192 	if (rctx->cb_misc_state.nr_cbufs != state->nr_cbufs ||
1193 	    rctx->cb_misc_state.bound_cbufs_target_mask != target_mask) {
1194 		rctx->cb_misc_state.bound_cbufs_target_mask = target_mask;
1195 		rctx->cb_misc_state.nr_cbufs = state->nr_cbufs;
1196 		r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom);
1197 	}
1198 
1199 	if (state->nr_cbufs == 0 && rctx->alphatest_state.bypass) {
1200 		rctx->alphatest_state.bypass = false;
1201 		r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom);
1202 	}
1203 
1204 	/* Calculate the CS size. */
1205 	rctx->framebuffer.atom.num_dw =
1206 		10 /*COLOR_INFO*/ + 4 /*SCISSOR*/ + 3 /*SHADER_CONTROL*/ + 8 /*MSAA*/;
1207 
1208 	if (rctx->framebuffer.state.nr_cbufs) {
1209 		rctx->framebuffer.atom.num_dw += 15 * rctx->framebuffer.state.nr_cbufs;
1210 		rctx->framebuffer.atom.num_dw += 3 * (2 + rctx->framebuffer.state.nr_cbufs);
1211 	}
1212 	if (rctx->framebuffer.state.zsbuf) {
1213 		rctx->framebuffer.atom.num_dw += 16;
1214 	} else if (rctx->screen->b.info.drm_minor >= 18) {
1215 		rctx->framebuffer.atom.num_dw += 3;
1216 	}
1217 	if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770) {
1218 		rctx->framebuffer.atom.num_dw += 2;
1219 	}
1220 
1221 	r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom);
1222 
1223 	r600_set_sample_locations_constant_buffer(rctx);
1224 	rctx->framebuffer.do_update_surf_dirtiness = true;
1225 }
1226 
1227 static const uint32_t sample_locs_2x[] = {
1228 	FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1229 	FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1230 };
1231 static const unsigned max_dist_2x = 4;
1232 
1233 static const uint32_t sample_locs_4x[] = {
1234 	FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1235 	FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1236 };
1237 static const unsigned max_dist_4x = 6;
1238 static const uint32_t sample_locs_8x[] = {
1239 	FILL_SREG(-1,  1,  1,  5,  3, -5,  5,  3),
1240 	FILL_SREG(-7, -1, -3, -7,  7, -3, -5,  7),
1241 };
1242 static const unsigned max_dist_8x = 7;
1243 
r600_get_sample_position(struct pipe_context * ctx,unsigned sample_count,unsigned sample_index,float * out_value)1244 static void r600_get_sample_position(struct pipe_context *ctx,
1245 				     unsigned sample_count,
1246 				     unsigned sample_index,
1247 				     float *out_value)
1248 {
1249 	int offset, index;
1250 	struct {
1251 		int idx:4;
1252 	} val;
1253 	switch (sample_count) {
1254 	case 1:
1255 	default:
1256 		out_value[0] = out_value[1] = 0.5;
1257 		break;
1258 	case 2:
1259 		offset = 4 * (sample_index * 2);
1260 		val.idx = (sample_locs_2x[0] >> offset) & 0xf;
1261 		out_value[0] = (float)(val.idx + 8) / 16.0f;
1262 		val.idx = (sample_locs_2x[0] >> (offset + 4)) & 0xf;
1263 		out_value[1] = (float)(val.idx + 8) / 16.0f;
1264 		break;
1265 	case 4:
1266 		offset = 4 * (sample_index * 2);
1267 		val.idx = (sample_locs_4x[0] >> offset) & 0xf;
1268 		out_value[0] = (float)(val.idx + 8) / 16.0f;
1269 		val.idx = (sample_locs_4x[0] >> (offset + 4)) & 0xf;
1270 		out_value[1] = (float)(val.idx + 8) / 16.0f;
1271 		break;
1272 	case 8:
1273 		offset = 4 * (sample_index % 4 * 2);
1274 		index = (sample_index / 4);
1275 		val.idx = (sample_locs_8x[index] >> offset) & 0xf;
1276 		out_value[0] = (float)(val.idx + 8) / 16.0f;
1277 		val.idx = (sample_locs_8x[index] >> (offset + 4)) & 0xf;
1278 		out_value[1] = (float)(val.idx + 8) / 16.0f;
1279 		break;
1280 	}
1281 }
1282 
r600_emit_msaa_state(struct r600_context * rctx,int nr_samples)1283 static void r600_emit_msaa_state(struct r600_context *rctx, int nr_samples)
1284 {
1285 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1286 	unsigned max_dist = 0;
1287 
1288 	if (rctx->b.family == CHIP_R600) {
1289 		switch (nr_samples) {
1290 		default:
1291 			nr_samples = 0;
1292 			break;
1293 		case 2:
1294 			radeon_set_config_reg(cs, R_008B40_PA_SC_AA_SAMPLE_LOCS_2S, sample_locs_2x[0]);
1295 			max_dist = max_dist_2x;
1296 			break;
1297 		case 4:
1298 			radeon_set_config_reg(cs, R_008B44_PA_SC_AA_SAMPLE_LOCS_4S, sample_locs_4x[0]);
1299 			max_dist = max_dist_4x;
1300 			break;
1301 		case 8:
1302 			radeon_set_config_reg_seq(cs, R_008B48_PA_SC_AA_SAMPLE_LOCS_8S_WD0, 2);
1303 			radeon_emit(cs, sample_locs_8x[0]); /* R_008B48_PA_SC_AA_SAMPLE_LOCS_8S_WD0 */
1304 			radeon_emit(cs, sample_locs_8x[1]); /* R_008B4C_PA_SC_AA_SAMPLE_LOCS_8S_WD1 */
1305 			max_dist = max_dist_8x;
1306 			break;
1307 		}
1308 	} else {
1309 		switch (nr_samples) {
1310 		default:
1311 			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1312 			radeon_emit(cs, 0); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1313 			radeon_emit(cs, 0); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1314 			nr_samples = 0;
1315 			break;
1316 		case 2:
1317 			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1318 			radeon_emit(cs, sample_locs_2x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1319 			radeon_emit(cs, sample_locs_2x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1320 			max_dist = max_dist_2x;
1321 			break;
1322 		case 4:
1323 			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1324 			radeon_emit(cs, sample_locs_4x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1325 			radeon_emit(cs, sample_locs_4x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1326 			max_dist = max_dist_4x;
1327 			break;
1328 		case 8:
1329 			radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 2);
1330 			radeon_emit(cs, sample_locs_8x[0]); /* R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX */
1331 			radeon_emit(cs, sample_locs_8x[1]); /* R_028C20_PA_SC_AA_SAMPLE_LOCS_8D_WD1_MCTX */
1332 			max_dist = max_dist_8x;
1333 			break;
1334 		}
1335 	}
1336 
1337 	if (nr_samples > 1) {
1338 		radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1339 		radeon_emit(cs, S_028C00_LAST_PIXEL(1) |
1340 				     S_028C00_EXPAND_LINE_WIDTH(1)); /* R_028C00_PA_SC_LINE_CNTL */
1341 		radeon_emit(cs, S_028C04_MSAA_NUM_SAMPLES(util_logbase2(nr_samples)) |
1342 				     S_028C04_MAX_SAMPLE_DIST(max_dist)); /* R_028C04_PA_SC_AA_CONFIG */
1343 	} else {
1344 		radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2);
1345 		radeon_emit(cs, S_028C00_LAST_PIXEL(1)); /* R_028C00_PA_SC_LINE_CNTL */
1346 		radeon_emit(cs, 0); /* R_028C04_PA_SC_AA_CONFIG */
1347 	}
1348 }
1349 
r600_emit_framebuffer_state(struct r600_context * rctx,struct r600_atom * atom)1350 static void r600_emit_framebuffer_state(struct r600_context *rctx, struct r600_atom *atom)
1351 {
1352 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1353 	struct pipe_framebuffer_state *state = &rctx->framebuffer.state;
1354 	unsigned nr_cbufs = state->nr_cbufs;
1355 	struct r600_surface **cb = (struct r600_surface**)&state->cbufs[0];
1356 	unsigned i, sbu = 0;
1357 
1358 	/* Colorbuffers. */
1359 	radeon_set_context_reg_seq(cs, R_0280A0_CB_COLOR0_INFO, 8);
1360 	for (i = 0; i < nr_cbufs; i++) {
1361 		radeon_emit(cs, cb[i] ? cb[i]->cb_color_info : 0);
1362 	}
1363 	/* set CB_COLOR1_INFO for possible dual-src blending */
1364 	if (rctx->framebuffer.dual_src_blend && i == 1 && cb[0]) {
1365 		radeon_emit(cs, cb[0]->cb_color_info);
1366 		i++;
1367 	}
1368 	for (; i < 8; i++) {
1369 		radeon_emit(cs, 0);
1370 	}
1371 
1372 	if (nr_cbufs) {
1373 		for (i = 0; i < nr_cbufs; i++) {
1374 			unsigned reloc;
1375 
1376 			if (!cb[i])
1377 				continue;
1378 
1379 			/* COLOR_BASE */
1380 			radeon_set_context_reg(cs, R_028040_CB_COLOR0_BASE + i*4, cb[i]->cb_color_base);
1381 
1382 			reloc = radeon_add_to_buffer_list(&rctx->b,
1383 						      &rctx->b.gfx,
1384 						      (struct r600_resource*)cb[i]->base.texture,
1385 						      RADEON_USAGE_READWRITE,
1386 						      cb[i]->base.texture->nr_samples > 1 ?
1387 							      RADEON_PRIO_COLOR_BUFFER_MSAA :
1388 							      RADEON_PRIO_COLOR_BUFFER);
1389 			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1390 			radeon_emit(cs, reloc);
1391 
1392 			/* FMASK */
1393 			radeon_set_context_reg(cs, R_0280E0_CB_COLOR0_FRAG + i*4, cb[i]->cb_color_fmask);
1394 
1395 			reloc = radeon_add_to_buffer_list(&rctx->b,
1396 						      &rctx->b.gfx,
1397 						      cb[i]->cb_buffer_fmask,
1398 						      RADEON_USAGE_READWRITE,
1399 						      cb[i]->base.texture->nr_samples > 1 ?
1400 							      RADEON_PRIO_COLOR_BUFFER_MSAA :
1401 							      RADEON_PRIO_COLOR_BUFFER);
1402 			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1403 			radeon_emit(cs, reloc);
1404 
1405 			/* CMASK */
1406 			radeon_set_context_reg(cs, R_0280C0_CB_COLOR0_TILE + i*4, cb[i]->cb_color_cmask);
1407 
1408 			reloc = radeon_add_to_buffer_list(&rctx->b,
1409 						      &rctx->b.gfx,
1410 						      cb[i]->cb_buffer_cmask,
1411 						      RADEON_USAGE_READWRITE,
1412 						      cb[i]->base.texture->nr_samples > 1 ?
1413 							      RADEON_PRIO_COLOR_BUFFER_MSAA :
1414 							      RADEON_PRIO_COLOR_BUFFER);
1415 			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1416 			radeon_emit(cs, reloc);
1417 		}
1418 
1419 		radeon_set_context_reg_seq(cs, R_028060_CB_COLOR0_SIZE, nr_cbufs);
1420 		for (i = 0; i < nr_cbufs; i++) {
1421 			radeon_emit(cs, cb[i] ? cb[i]->cb_color_size : 0);
1422 		}
1423 
1424 		radeon_set_context_reg_seq(cs, R_028080_CB_COLOR0_VIEW, nr_cbufs);
1425 		for (i = 0; i < nr_cbufs; i++) {
1426 			radeon_emit(cs, cb[i] ? cb[i]->cb_color_view : 0);
1427 		}
1428 
1429 		radeon_set_context_reg_seq(cs, R_028100_CB_COLOR0_MASK, nr_cbufs);
1430 		for (i = 0; i < nr_cbufs; i++) {
1431 			radeon_emit(cs, cb[i] ? cb[i]->cb_color_mask : 0);
1432 		}
1433 
1434 		sbu |= SURFACE_BASE_UPDATE_COLOR_NUM(nr_cbufs);
1435 	}
1436 
1437 	/* SURFACE_BASE_UPDATE */
1438 	if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770 && sbu) {
1439 		radeon_emit(cs, PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0));
1440 		radeon_emit(cs, sbu);
1441 		sbu = 0;
1442 	}
1443 
1444 	/* Zbuffer. */
1445 	if (state->zsbuf) {
1446 		struct r600_surface *surf = (struct r600_surface*)state->zsbuf;
1447 		unsigned reloc = radeon_add_to_buffer_list(&rctx->b,
1448 						       &rctx->b.gfx,
1449 						       (struct r600_resource*)state->zsbuf->texture,
1450 						       RADEON_USAGE_READWRITE,
1451 						       surf->base.texture->nr_samples > 1 ?
1452 							       RADEON_PRIO_DEPTH_BUFFER_MSAA :
1453 							       RADEON_PRIO_DEPTH_BUFFER);
1454 
1455 		radeon_set_context_reg_seq(cs, R_028000_DB_DEPTH_SIZE, 2);
1456 		radeon_emit(cs, surf->db_depth_size); /* R_028000_DB_DEPTH_SIZE */
1457 		radeon_emit(cs, surf->db_depth_view); /* R_028004_DB_DEPTH_VIEW */
1458 		radeon_set_context_reg_seq(cs, R_02800C_DB_DEPTH_BASE, 2);
1459 		radeon_emit(cs, surf->db_depth_base); /* R_02800C_DB_DEPTH_BASE */
1460 		radeon_emit(cs, surf->db_depth_info); /* R_028010_DB_DEPTH_INFO */
1461 
1462 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1463 		radeon_emit(cs, reloc);
1464 
1465 		radeon_set_context_reg(cs, R_028D34_DB_PREFETCH_LIMIT, surf->db_prefetch_limit);
1466 
1467 		sbu |= SURFACE_BASE_UPDATE_DEPTH;
1468 	} else if (rctx->screen->b.info.drm_minor >= 18) {
1469 		/* DRM 2.6.18 allows the INVALID format to disable depth/stencil.
1470 		 * Older kernels are out of luck. */
1471 		radeon_set_context_reg(cs, R_028010_DB_DEPTH_INFO, S_028010_FORMAT(V_028010_DEPTH_INVALID));
1472 	}
1473 
1474 	/* SURFACE_BASE_UPDATE */
1475 	if (rctx->b.family > CHIP_R600 && rctx->b.family < CHIP_RV770 && sbu) {
1476 		radeon_emit(cs, PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0));
1477 		radeon_emit(cs, sbu);
1478 		sbu = 0;
1479 	}
1480 
1481 	/* Framebuffer dimensions. */
1482 	radeon_set_context_reg_seq(cs, R_028204_PA_SC_WINDOW_SCISSOR_TL, 2);
1483 	radeon_emit(cs, S_028240_TL_X(0) | S_028240_TL_Y(0) |
1484 			     S_028240_WINDOW_OFFSET_DISABLE(1)); /* R_028204_PA_SC_WINDOW_SCISSOR_TL */
1485 	radeon_emit(cs, S_028244_BR_X(state->width) |
1486 			     S_028244_BR_Y(state->height)); /* R_028208_PA_SC_WINDOW_SCISSOR_BR */
1487 
1488 	if (rctx->framebuffer.is_msaa_resolve) {
1489 		radeon_set_context_reg(cs, R_0287A0_CB_SHADER_CONTROL, 1);
1490 	} else {
1491 		/* Always enable the first colorbuffer in CB_SHADER_CONTROL. This
1492 		 * will assure that the alpha-test will work even if there is
1493 		 * no colorbuffer bound. */
1494 		radeon_set_context_reg(cs, R_0287A0_CB_SHADER_CONTROL,
1495 				       (1ull << MAX2(nr_cbufs, 1)) - 1);
1496 	}
1497 
1498 	r600_emit_msaa_state(rctx, rctx->framebuffer.nr_samples);
1499 }
1500 
r600_set_min_samples(struct pipe_context * ctx,unsigned min_samples)1501 static void r600_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
1502 {
1503 	struct r600_context *rctx = (struct r600_context *)ctx;
1504 
1505 	if (rctx->ps_iter_samples == min_samples)
1506 		return;
1507 
1508 	rctx->ps_iter_samples = min_samples;
1509 	if (rctx->framebuffer.nr_samples > 1) {
1510 		r600_mark_atom_dirty(rctx, &rctx->rasterizer_state.atom);
1511 		if (rctx->b.chip_class == R600)
1512 			r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
1513 	}
1514 }
1515 
r600_emit_cb_misc_state(struct r600_context * rctx,struct r600_atom * atom)1516 static void r600_emit_cb_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1517 {
1518 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1519 	struct r600_cb_misc_state *a = (struct r600_cb_misc_state*)atom;
1520 
1521 	if (G_028808_SPECIAL_OP(a->cb_color_control) == V_028808_SPECIAL_RESOLVE_BOX) {
1522 		radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
1523 		if (rctx->b.chip_class == R600) {
1524 			radeon_emit(cs, 0xff); /* R_028238_CB_TARGET_MASK */
1525 			radeon_emit(cs, 0xff); /* R_02823C_CB_SHADER_MASK */
1526 		} else {
1527 			radeon_emit(cs, 0xf); /* R_028238_CB_TARGET_MASK */
1528 			radeon_emit(cs, 0xf); /* R_02823C_CB_SHADER_MASK */
1529 		}
1530 		radeon_set_context_reg(cs, R_028808_CB_COLOR_CONTROL, a->cb_color_control);
1531 	} else {
1532 		unsigned fb_colormask = a->bound_cbufs_target_mask;
1533 		unsigned ps_colormask = a->ps_color_export_mask;
1534 		unsigned multiwrite = a->multiwrite && a->nr_cbufs > 1;
1535 
1536 		radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2);
1537 		radeon_emit(cs, a->blend_colormask & fb_colormask); /* R_028238_CB_TARGET_MASK */
1538 		/* Always enable the first color output to make sure alpha-test works even without one. */
1539 		radeon_emit(cs, 0xf | (multiwrite ? fb_colormask : ps_colormask)); /* R_02823C_CB_SHADER_MASK */
1540 		radeon_set_context_reg(cs, R_028808_CB_COLOR_CONTROL,
1541 				       a->cb_color_control |
1542 				       S_028808_MULTIWRITE_ENABLE(multiwrite));
1543 	}
1544 }
1545 
r600_emit_db_state(struct r600_context * rctx,struct r600_atom * atom)1546 static void r600_emit_db_state(struct r600_context *rctx, struct r600_atom *atom)
1547 {
1548 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1549 	struct r600_db_state *a = (struct r600_db_state*)atom;
1550 
1551 	if (a->rsurf && a->rsurf->db_htile_surface) {
1552 		struct r600_texture *rtex = (struct r600_texture *)a->rsurf->base.texture;
1553 		unsigned reloc_idx;
1554 
1555 		radeon_set_context_reg(cs, R_02802C_DB_DEPTH_CLEAR, fui(rtex->depth_clear_value));
1556 		radeon_set_context_reg(cs, R_028D24_DB_HTILE_SURFACE, a->rsurf->db_htile_surface);
1557 		radeon_set_context_reg(cs, R_028014_DB_HTILE_DATA_BASE, a->rsurf->db_htile_data_base);
1558 		reloc_idx = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, &rtex->resource,
1559 						  RADEON_USAGE_READWRITE, RADEON_PRIO_SEPARATE_META);
1560 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1561 		radeon_emit(cs, reloc_idx);
1562 	} else {
1563 		radeon_set_context_reg(cs, R_028D24_DB_HTILE_SURFACE, 0);
1564 	}
1565 }
1566 
r600_emit_db_misc_state(struct r600_context * rctx,struct r600_atom * atom)1567 static void r600_emit_db_misc_state(struct r600_context *rctx, struct r600_atom *atom)
1568 {
1569 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1570 	struct r600_db_misc_state *a = (struct r600_db_misc_state*)atom;
1571 	unsigned db_render_control = 0;
1572 	unsigned db_render_override =
1573 		S_028D10_FORCE_HIS_ENABLE0(V_028D10_FORCE_DISABLE) |
1574 		S_028D10_FORCE_HIS_ENABLE1(V_028D10_FORCE_DISABLE);
1575 
1576 	if (rctx->b.chip_class >= R700) {
1577 		switch (a->ps_conservative_z) {
1578 		default: /* fall through */
1579 		case TGSI_FS_DEPTH_LAYOUT_ANY:
1580 			db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_ANY_Z);
1581 			break;
1582 		case TGSI_FS_DEPTH_LAYOUT_GREATER:
1583 			db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_GREATER_THAN_Z);
1584 			break;
1585 		case TGSI_FS_DEPTH_LAYOUT_LESS:
1586 			db_render_control |= S_028D0C_CONSERVATIVE_Z_EXPORT(V_028D0C_EXPORT_LESS_THAN_Z);
1587 			break;
1588 		}
1589 	}
1590 
1591 	if (rctx->b.num_occlusion_queries > 0 &&
1592 	    !a->occlusion_queries_disabled) {
1593 		if (rctx->b.chip_class >= R700) {
1594 			db_render_control |= S_028D0C_R700_PERFECT_ZPASS_COUNTS(1);
1595 		}
1596 		db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1597 	} else {
1598 		db_render_control |= S_028D0C_ZPASS_INCREMENT_DISABLE(1);
1599 	}
1600 
1601 	if (rctx->db_state.rsurf && rctx->db_state.rsurf->db_htile_surface) {
1602 		/* FORCE_OFF means HiZ/HiS are determined by DB_SHADER_CONTROL */
1603 		db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_OFF);
1604 		/* This is to fix a lockup when hyperz and alpha test are enabled at
1605 		 * the same time somehow GPU get confuse on which order to pick for
1606 		 * z test
1607 		 */
1608 		if (rctx->alphatest_state.sx_alpha_test_control) {
1609 			db_render_override |= S_028D10_FORCE_SHADER_Z_ORDER(1);
1610 		}
1611 	} else {
1612 		db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1613 	}
1614 	if (rctx->b.chip_class == R600 && rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0) {
1615 		/* sample shading and hyperz causes lockups on R6xx chips */
1616 		db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1617 	}
1618 	if (a->flush_depthstencil_through_cb) {
1619 		assert(a->copy_depth || a->copy_stencil);
1620 
1621 		db_render_control |= S_028D0C_DEPTH_COPY_ENABLE(a->copy_depth) |
1622 				     S_028D0C_STENCIL_COPY_ENABLE(a->copy_stencil) |
1623 				     S_028D0C_COPY_CENTROID(1) |
1624 				     S_028D0C_COPY_SAMPLE(a->copy_sample);
1625 
1626 		if (rctx->b.chip_class == R600)
1627 			db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1628 
1629 		if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
1630 		    rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
1631 			db_render_override |= S_028D10_FORCE_HIZ_ENABLE(V_028D10_FORCE_DISABLE);
1632 	} else if (a->flush_depth_inplace || a->flush_stencil_inplace) {
1633 		db_render_control |= S_028D0C_DEPTH_COMPRESS_DISABLE(a->flush_depth_inplace) |
1634 				     S_028D0C_STENCIL_COMPRESS_DISABLE(a->flush_stencil_inplace);
1635 		db_render_override |= S_028D10_NOOP_CULL_DISABLE(1);
1636 	}
1637 	if (a->htile_clear) {
1638 		db_render_control |= S_028D0C_DEPTH_CLEAR_ENABLE(1);
1639 	}
1640 
1641 	/* RV770 workaround for a hang with 8x MSAA. */
1642 	if (rctx->b.family == CHIP_RV770 && a->log_samples == 3) {
1643 		db_render_override |= S_028D10_MAX_TILES_IN_DTT(6);
1644 	}
1645 
1646 	radeon_set_context_reg_seq(cs, R_028D0C_DB_RENDER_CONTROL, 2);
1647 	radeon_emit(cs, db_render_control); /* R_028D0C_DB_RENDER_CONTROL */
1648 	radeon_emit(cs, db_render_override); /* R_028D10_DB_RENDER_OVERRIDE */
1649 	radeon_set_context_reg(cs, R_02880C_DB_SHADER_CONTROL, a->db_shader_control);
1650 }
1651 
r600_emit_config_state(struct r600_context * rctx,struct r600_atom * atom)1652 static void r600_emit_config_state(struct r600_context *rctx, struct r600_atom *atom)
1653 {
1654 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1655 	struct r600_config_state *a = (struct r600_config_state*)atom;
1656 
1657 	radeon_set_config_reg(cs, R_008C04_SQ_GPR_RESOURCE_MGMT_1, a->sq_gpr_resource_mgmt_1);
1658 	radeon_set_config_reg(cs, R_008C08_SQ_GPR_RESOURCE_MGMT_2, a->sq_gpr_resource_mgmt_2);
1659 }
1660 
r600_emit_vertex_buffers(struct r600_context * rctx,struct r600_atom * atom)1661 static void r600_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom *atom)
1662 {
1663 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1664 	uint32_t dirty_mask = rctx->vertex_buffer_state.dirty_mask;
1665 
1666 	while (dirty_mask) {
1667 		struct pipe_vertex_buffer *vb;
1668 		struct r600_resource *rbuffer;
1669 		unsigned offset;
1670 		unsigned buffer_index = u_bit_scan(&dirty_mask);
1671 
1672 		vb = &rctx->vertex_buffer_state.vb[buffer_index];
1673 		rbuffer = (struct r600_resource*)vb->buffer.resource;
1674 		assert(rbuffer);
1675 
1676 		offset = vb->buffer_offset;
1677 
1678 		/* fetch resources start at index 320 (OFFSET_FS) */
1679 		radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1680 		radeon_emit(cs, (R600_FETCH_CONSTANTS_OFFSET_FS + buffer_index) * 7);
1681 		radeon_emit(cs, offset); /* RESOURCEi_WORD0 */
1682 		radeon_emit(cs, rbuffer->b.b.width0 - offset - 1); /* RESOURCEi_WORD1 */
1683 		radeon_emit(cs, /* RESOURCEi_WORD2 */
1684 				 S_038008_ENDIAN_SWAP(r600_endian_swap(32)) |
1685 				 S_038008_STRIDE(vb->stride));
1686 		radeon_emit(cs, 0); /* RESOURCEi_WORD3 */
1687 		radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1688 		radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1689 		radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD6 */
1690 
1691 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1692 		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1693 						      RADEON_USAGE_READ, RADEON_PRIO_VERTEX_BUFFER));
1694 	}
1695 }
1696 
r600_emit_constant_buffers(struct r600_context * rctx,struct r600_constbuf_state * state,unsigned buffer_id_base,unsigned reg_alu_constbuf_size,unsigned reg_alu_const_cache)1697 static void r600_emit_constant_buffers(struct r600_context *rctx,
1698 				       struct r600_constbuf_state *state,
1699 				       unsigned buffer_id_base,
1700 				       unsigned reg_alu_constbuf_size,
1701 				       unsigned reg_alu_const_cache)
1702 {
1703 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1704 	uint32_t dirty_mask = state->dirty_mask;
1705 
1706 	while (dirty_mask) {
1707 		struct pipe_constant_buffer *cb;
1708 		struct r600_resource *rbuffer;
1709 		unsigned offset;
1710 		unsigned buffer_index = ffs(dirty_mask) - 1;
1711 		unsigned gs_ring_buffer = (buffer_index == R600_GS_RING_CONST_BUFFER);
1712 		cb = &state->cb[buffer_index];
1713 		rbuffer = (struct r600_resource*)cb->buffer;
1714 		assert(rbuffer);
1715 
1716 		offset = cb->buffer_offset;
1717 
1718 		if (!gs_ring_buffer) {
1719 			assert(buffer_index < R600_MAX_HW_CONST_BUFFERS);
1720 			radeon_set_context_reg(cs, reg_alu_constbuf_size + buffer_index * 4,
1721 					       DIV_ROUND_UP(cb->buffer_size, 256));
1722 			radeon_set_context_reg(cs, reg_alu_const_cache + buffer_index * 4, offset >> 8);
1723 			radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1724 			radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1725 								  RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER));
1726 		}
1727 
1728 		radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1729 		radeon_emit(cs, (buffer_id_base + buffer_index) * 7);
1730 		radeon_emit(cs, offset); /* RESOURCEi_WORD0 */
1731 		radeon_emit(cs, cb->buffer_size - 1); /* RESOURCEi_WORD1 */
1732 		radeon_emit(cs, /* RESOURCEi_WORD2 */
1733 			    S_038008_ENDIAN_SWAP(gs_ring_buffer ? ENDIAN_NONE : r600_endian_swap(32)) |
1734 			    S_038008_STRIDE(gs_ring_buffer ? 4 : 16));
1735 		radeon_emit(cs, 0); /* RESOURCEi_WORD3 */
1736 		radeon_emit(cs, 0); /* RESOURCEi_WORD4 */
1737 		radeon_emit(cs, 0); /* RESOURCEi_WORD5 */
1738 		radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD6 */
1739 
1740 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1741 		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1742 						      RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER));
1743 
1744 		dirty_mask &= ~(1 << buffer_index);
1745 	}
1746 	state->dirty_mask = 0;
1747 }
1748 
r600_emit_vs_constant_buffers(struct r600_context * rctx,struct r600_atom * atom)1749 static void r600_emit_vs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1750 {
1751 	r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX],
1752 				   R600_FETCH_CONSTANTS_OFFSET_VS,
1753 				   R_028180_ALU_CONST_BUFFER_SIZE_VS_0,
1754 				   R_028980_ALU_CONST_CACHE_VS_0);
1755 }
1756 
r600_emit_gs_constant_buffers(struct r600_context * rctx,struct r600_atom * atom)1757 static void r600_emit_gs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1758 {
1759 	r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY],
1760 				   R600_FETCH_CONSTANTS_OFFSET_GS,
1761 				   R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0,
1762 				   R_0289C0_ALU_CONST_CACHE_GS_0);
1763 }
1764 
r600_emit_ps_constant_buffers(struct r600_context * rctx,struct r600_atom * atom)1765 static void r600_emit_ps_constant_buffers(struct r600_context *rctx, struct r600_atom *atom)
1766 {
1767 	r600_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT],
1768 				   R600_FETCH_CONSTANTS_OFFSET_PS,
1769 				   R_028140_ALU_CONST_BUFFER_SIZE_PS_0,
1770 				   R_028940_ALU_CONST_CACHE_PS_0);
1771 }
1772 
r600_emit_sampler_views(struct r600_context * rctx,struct r600_samplerview_state * state,unsigned resource_id_base)1773 static void r600_emit_sampler_views(struct r600_context *rctx,
1774 				    struct r600_samplerview_state *state,
1775 				    unsigned resource_id_base)
1776 {
1777 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1778 	uint32_t dirty_mask = state->dirty_mask;
1779 
1780 	while (dirty_mask) {
1781 		struct r600_pipe_sampler_view *rview;
1782 		unsigned resource_index = u_bit_scan(&dirty_mask);
1783 		unsigned reloc;
1784 
1785 		rview = state->views[resource_index];
1786 		assert(rview);
1787 
1788 		radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 7, 0));
1789 		radeon_emit(cs, (resource_id_base + resource_index) * 7);
1790 		radeon_emit_array(cs, rview->tex_resource_words, 7);
1791 
1792 		reloc = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rview->tex_resource,
1793 					      RADEON_USAGE_READ,
1794 					      r600_get_sampler_view_priority(rview->tex_resource));
1795 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1796 		radeon_emit(cs, reloc);
1797 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1798 		radeon_emit(cs, reloc);
1799 	}
1800 	state->dirty_mask = 0;
1801 }
1802 
1803 
r600_emit_vs_sampler_views(struct r600_context * rctx,struct r600_atom * atom)1804 static void r600_emit_vs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1805 {
1806 	r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views, R600_FETCH_CONSTANTS_OFFSET_VS + R600_MAX_CONST_BUFFERS);
1807 }
1808 
r600_emit_gs_sampler_views(struct r600_context * rctx,struct r600_atom * atom)1809 static void r600_emit_gs_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1810 {
1811 	r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views, R600_FETCH_CONSTANTS_OFFSET_GS + R600_MAX_CONST_BUFFERS);
1812 }
1813 
r600_emit_ps_sampler_views(struct r600_context * rctx,struct r600_atom * atom)1814 static void r600_emit_ps_sampler_views(struct r600_context *rctx, struct r600_atom *atom)
1815 {
1816 	r600_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views, R600_FETCH_CONSTANTS_OFFSET_PS + R600_MAX_CONST_BUFFERS);
1817 }
1818 
r600_emit_sampler_states(struct r600_context * rctx,struct r600_textures_info * texinfo,unsigned resource_id_base,unsigned border_color_reg)1819 static void r600_emit_sampler_states(struct r600_context *rctx,
1820 				struct r600_textures_info *texinfo,
1821 				unsigned resource_id_base,
1822 				unsigned border_color_reg)
1823 {
1824 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1825 	uint32_t dirty_mask = texinfo->states.dirty_mask;
1826 
1827 	while (dirty_mask) {
1828 		struct r600_pipe_sampler_state *rstate;
1829 		struct r600_pipe_sampler_view *rview;
1830 		unsigned i = u_bit_scan(&dirty_mask);
1831 
1832 		rstate = texinfo->states.states[i];
1833 		assert(rstate);
1834 		rview = texinfo->views.views[i];
1835 
1836 		/* TEX_ARRAY_OVERRIDE must be set for array textures to disable
1837 		 * filtering between layers.
1838 		 */
1839 		enum pipe_texture_target target = PIPE_BUFFER;
1840 		if (rview)
1841 			target = rview->base.texture->target;
1842 		if (target == PIPE_TEXTURE_1D_ARRAY ||
1843 		    target == PIPE_TEXTURE_2D_ARRAY) {
1844 			rstate->tex_sampler_words[0] |= S_03C000_TEX_ARRAY_OVERRIDE(1);
1845 			texinfo->is_array_sampler[i] = true;
1846 		} else {
1847 			rstate->tex_sampler_words[0] &= C_03C000_TEX_ARRAY_OVERRIDE;
1848 			texinfo->is_array_sampler[i] = false;
1849 		}
1850 
1851 		radeon_emit(cs, PKT3(PKT3_SET_SAMPLER, 3, 0));
1852 		radeon_emit(cs, (resource_id_base + i) * 3);
1853 		radeon_emit_array(cs, rstate->tex_sampler_words, 3);
1854 
1855 		if (rstate->border_color_use) {
1856 			unsigned offset;
1857 
1858 			offset = border_color_reg;
1859 			offset += i * 16;
1860 			radeon_set_config_reg_seq(cs, offset, 4);
1861 			radeon_emit_array(cs, rstate->border_color.ui, 4);
1862 		}
1863 	}
1864 	texinfo->states.dirty_mask = 0;
1865 }
1866 
r600_emit_vs_sampler_states(struct r600_context * rctx,struct r600_atom * atom)1867 static void r600_emit_vs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1868 {
1869 	r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 18, R_00A600_TD_VS_SAMPLER0_BORDER_RED);
1870 }
1871 
r600_emit_gs_sampler_states(struct r600_context * rctx,struct r600_atom * atom)1872 static void r600_emit_gs_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1873 {
1874 	r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY], 36, R_00A800_TD_GS_SAMPLER0_BORDER_RED);
1875 }
1876 
r600_emit_ps_sampler_states(struct r600_context * rctx,struct r600_atom * atom)1877 static void r600_emit_ps_sampler_states(struct r600_context *rctx, struct r600_atom *atom)
1878 {
1879 	r600_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT], 0, R_00A400_TD_PS_SAMPLER0_BORDER_RED);
1880 }
1881 
r600_emit_seamless_cube_map(struct r600_context * rctx,struct r600_atom * atom)1882 static void r600_emit_seamless_cube_map(struct r600_context *rctx, struct r600_atom *atom)
1883 {
1884 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1885 	unsigned tmp;
1886 
1887 	tmp = S_009508_DISABLE_CUBE_ANISO(1) |
1888 		S_009508_SYNC_GRADIENT(1) |
1889 		S_009508_SYNC_WALKER(1) |
1890 		S_009508_SYNC_ALIGNER(1);
1891 	if (!rctx->seamless_cube_map.enabled) {
1892 		tmp |= S_009508_DISABLE_CUBE_WRAP(1);
1893 	}
1894 	radeon_set_config_reg(cs, R_009508_TA_CNTL_AUX, tmp);
1895 }
1896 
r600_emit_sample_mask(struct r600_context * rctx,struct r600_atom * a)1897 static void r600_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a)
1898 {
1899 	struct r600_sample_mask *s = (struct r600_sample_mask*)a;
1900 	uint8_t mask = s->sample_mask;
1901 
1902 	radeon_set_context_reg(rctx->b.gfx.cs, R_028C48_PA_SC_AA_MASK,
1903 			       mask | (mask << 8) | (mask << 16) | (mask << 24));
1904 }
1905 
r600_emit_vertex_fetch_shader(struct r600_context * rctx,struct r600_atom * a)1906 static void r600_emit_vertex_fetch_shader(struct r600_context *rctx, struct r600_atom *a)
1907 {
1908 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1909 	struct r600_cso_state *state = (struct r600_cso_state*)a;
1910 	struct r600_fetch_shader *shader = (struct r600_fetch_shader*)state->cso;
1911 
1912 	if (!shader)
1913 		return;
1914 
1915 	radeon_set_context_reg(cs, R_028894_SQ_PGM_START_FS, shader->offset >> 8);
1916 	radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1917 	radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, shader->buffer,
1918                                                   RADEON_USAGE_READ,
1919                                                   RADEON_PRIO_SHADER_BINARY));
1920 }
1921 
r600_emit_shader_stages(struct r600_context * rctx,struct r600_atom * a)1922 static void r600_emit_shader_stages(struct r600_context *rctx, struct r600_atom *a)
1923 {
1924 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1925 	struct r600_shader_stages_state *state = (struct r600_shader_stages_state*)a;
1926 
1927 	uint32_t v2 = 0, primid = 0;
1928 
1929 	if (rctx->vs_shader->current->shader.vs_as_gs_a) {
1930 		v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_A);
1931 		primid = 1;
1932 	}
1933 
1934 	if (state->geom_enable) {
1935 		uint32_t cut_val;
1936 
1937 		if (rctx->gs_shader->gs_max_out_vertices <= 128)
1938 			cut_val = V_028A40_GS_CUT_128;
1939 		else if (rctx->gs_shader->gs_max_out_vertices <= 256)
1940 			cut_val = V_028A40_GS_CUT_256;
1941 		else if (rctx->gs_shader->gs_max_out_vertices <= 512)
1942 			cut_val = V_028A40_GS_CUT_512;
1943 		else
1944 			cut_val = V_028A40_GS_CUT_1024;
1945 
1946 		v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_G) |
1947 			S_028A40_CUT_MODE(cut_val);
1948 
1949 		if (rctx->gs_shader->current->shader.gs_prim_id_input)
1950 			primid = 1;
1951 	}
1952 
1953 	radeon_set_context_reg(cs, R_028A40_VGT_GS_MODE, v2);
1954 	radeon_set_context_reg(cs, R_028A84_VGT_PRIMITIVEID_EN, primid);
1955 }
1956 
r600_emit_gs_rings(struct r600_context * rctx,struct r600_atom * a)1957 static void r600_emit_gs_rings(struct r600_context *rctx, struct r600_atom *a)
1958 {
1959 	struct radeon_cmdbuf *cs = rctx->b.gfx.cs;
1960 	struct r600_gs_rings_state *state = (struct r600_gs_rings_state*)a;
1961 	struct r600_resource *rbuffer;
1962 
1963 	radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
1964 	radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
1965 	radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
1966 
1967 	if (state->enable) {
1968 		rbuffer =(struct r600_resource*)state->esgs_ring.buffer;
1969 		radeon_set_config_reg(cs, R_008C40_SQ_ESGS_RING_BASE, 0);
1970 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1971 		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1972 						      RADEON_USAGE_READWRITE,
1973 						      RADEON_PRIO_SHADER_RINGS));
1974 		radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE,
1975 				state->esgs_ring.buffer_size >> 8);
1976 
1977 		rbuffer =(struct r600_resource*)state->gsvs_ring.buffer;
1978 		radeon_set_config_reg(cs, R_008C48_SQ_GSVS_RING_BASE, 0);
1979 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
1980 		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer,
1981 						      RADEON_USAGE_READWRITE,
1982 						      RADEON_PRIO_SHADER_RINGS));
1983 		radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE,
1984 				state->gsvs_ring.buffer_size >> 8);
1985 	} else {
1986 		radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE, 0);
1987 		radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE, 0);
1988 	}
1989 
1990 	radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
1991 	radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
1992 	radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH));
1993 }
1994 
1995 /* Adjust GPR allocation on R6xx/R7xx */
r600_adjust_gprs(struct r600_context * rctx)1996 bool r600_adjust_gprs(struct r600_context *rctx)
1997 {
1998 	unsigned num_gprs[R600_NUM_HW_STAGES];
1999 	unsigned new_gprs[R600_NUM_HW_STAGES];
2000 	unsigned cur_gprs[R600_NUM_HW_STAGES];
2001 	unsigned def_gprs[R600_NUM_HW_STAGES];
2002 	unsigned def_num_clause_temp_gprs = rctx->r6xx_num_clause_temp_gprs;
2003 	unsigned max_gprs;
2004 	unsigned tmp, tmp2;
2005 	unsigned i;
2006 	bool need_recalc = false, use_default = true;
2007 
2008 	/* hardware will reserve twice num_clause_temp_gprs */
2009 	max_gprs = def_num_clause_temp_gprs * 2;
2010 	for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2011 		def_gprs[i] = rctx->default_gprs[i];
2012 		max_gprs += def_gprs[i];
2013 	}
2014 
2015 	cur_gprs[R600_HW_STAGE_PS] = G_008C04_NUM_PS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
2016 	cur_gprs[R600_HW_STAGE_VS] = G_008C04_NUM_VS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1);
2017 	cur_gprs[R600_HW_STAGE_GS] = G_008C08_NUM_GS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
2018 	cur_gprs[R600_HW_STAGE_ES] = G_008C08_NUM_ES_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2);
2019 
2020 	num_gprs[R600_HW_STAGE_PS] = rctx->ps_shader->current->shader.bc.ngpr;
2021 	if (rctx->gs_shader) {
2022 		num_gprs[R600_HW_STAGE_ES] = rctx->vs_shader->current->shader.bc.ngpr;
2023 		num_gprs[R600_HW_STAGE_GS] = rctx->gs_shader->current->shader.bc.ngpr;
2024 		num_gprs[R600_HW_STAGE_VS] = rctx->gs_shader->current->gs_copy_shader->shader.bc.ngpr;
2025 	} else {
2026 		num_gprs[R600_HW_STAGE_ES] = 0;
2027 		num_gprs[R600_HW_STAGE_GS] = 0;
2028 		num_gprs[R600_HW_STAGE_VS] = rctx->vs_shader->current->shader.bc.ngpr;
2029 	}
2030 
2031 	for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2032 		new_gprs[i] = num_gprs[i];
2033 		if (new_gprs[i] > cur_gprs[i])
2034 			need_recalc = true;
2035 		if (new_gprs[i] > def_gprs[i])
2036 			use_default = false;
2037 	}
2038 
2039 	/* the sum of all SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS must <= to max_gprs */
2040 	if (!need_recalc)
2041 		return true;
2042 
2043 	/* try to use switch back to default */
2044 	if (!use_default) {
2045 		/* always privilege vs stage so that at worst we have the
2046 		 * pixel stage producing wrong output (not the vertex
2047 		 * stage) */
2048 		new_gprs[R600_HW_STAGE_PS] = max_gprs - def_num_clause_temp_gprs * 2;
2049 		for (i = R600_HW_STAGE_VS; i < R600_NUM_HW_STAGES; i++)
2050 			new_gprs[R600_HW_STAGE_PS] -= new_gprs[i];
2051 	} else {
2052 		for (i = 0; i < R600_NUM_HW_STAGES; i++)
2053 			new_gprs[i] = def_gprs[i];
2054 	}
2055 
2056 	/* SQ_PGM_RESOURCES_*.NUM_GPRS must always be program to a value <=
2057 	 * SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS otherwise the GPU will lockup
2058 	 * Also if a shader use more gpr than SQ_GPR_RESOURCE_MGMT*.NUM_*_GPRS
2059 	 * it will lockup. So in this case just discard the draw command
2060 	 * and don't change the current gprs repartitions.
2061 	 */
2062 	for (i = 0; i < R600_NUM_HW_STAGES; i++) {
2063 		if (num_gprs[i] > new_gprs[i]) {
2064 			R600_ERR("shaders require too many register (%d + %d + %d + %d) "
2065 				 "for a combined maximum of %d\n",
2066 				 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);
2067 			return false;
2068 		}
2069 	}
2070 
2071 	/* in some case we endup recomputing the current value */
2072 	tmp = S_008C04_NUM_PS_GPRS(new_gprs[R600_HW_STAGE_PS]) |
2073 		S_008C04_NUM_VS_GPRS(new_gprs[R600_HW_STAGE_VS]) |
2074 		S_008C04_NUM_CLAUSE_TEMP_GPRS(def_num_clause_temp_gprs);
2075 
2076 	tmp2 = S_008C08_NUM_ES_GPRS(new_gprs[R600_HW_STAGE_ES]) |
2077 		S_008C08_NUM_GS_GPRS(new_gprs[R600_HW_STAGE_GS]);
2078 	if (rctx->config_state.sq_gpr_resource_mgmt_1 != tmp || rctx->config_state.sq_gpr_resource_mgmt_2 != tmp2) {
2079 		rctx->config_state.sq_gpr_resource_mgmt_1 = tmp;
2080 		rctx->config_state.sq_gpr_resource_mgmt_2 = tmp2;
2081 		r600_mark_atom_dirty(rctx, &rctx->config_state.atom);
2082 		rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE;
2083 	}
2084 	return true;
2085 }
2086 
r600_init_atom_start_cs(struct r600_context * rctx)2087 void r600_init_atom_start_cs(struct r600_context *rctx)
2088 {
2089 	int ps_prio;
2090 	int vs_prio;
2091 	int gs_prio;
2092 	int es_prio;
2093 	int num_ps_gprs;
2094 	int num_vs_gprs;
2095 	int num_gs_gprs;
2096 	int num_es_gprs;
2097 	int num_temp_gprs;
2098 	int num_ps_threads;
2099 	int num_vs_threads;
2100 	int num_gs_threads;
2101 	int num_es_threads;
2102 	int num_ps_stack_entries;
2103 	int num_vs_stack_entries;
2104 	int num_gs_stack_entries;
2105 	int num_es_stack_entries;
2106 	enum radeon_family family;
2107 	struct r600_command_buffer *cb = &rctx->start_cs_cmd;
2108 	uint32_t tmp, i;
2109 
2110 	r600_init_command_buffer(cb, 256);
2111 
2112 	/* R6xx requires this packet at the start of each command buffer */
2113 	if (rctx->b.chip_class == R600) {
2114 		r600_store_value(cb, PKT3(PKT3_START_3D_CMDBUF, 0, 0));
2115 		r600_store_value(cb, 0);
2116 	}
2117 	/* All asics require this one */
2118 	r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
2119 	r600_store_value(cb, 0x80000000);
2120 	r600_store_value(cb, 0x80000000);
2121 
2122 	/* We're setting config registers here. */
2123 	r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2124 	r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
2125 
2126 	/* This enables pipeline stat & streamout queries.
2127 	 * They are only disabled by blits.
2128 	 */
2129 	r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0));
2130 	r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PIPELINESTAT_START) | EVENT_INDEX(0));
2131 
2132 	family = rctx->b.family;
2133 	ps_prio = 0;
2134 	vs_prio = 1;
2135 	gs_prio = 2;
2136 	es_prio = 3;
2137 	switch (family) {
2138 	case CHIP_R600:
2139 		num_ps_gprs = 192;
2140 		num_vs_gprs = 56;
2141 		num_temp_gprs = 4;
2142 		num_gs_gprs = 0;
2143 		num_es_gprs = 0;
2144 		num_ps_threads = 136;
2145 		num_vs_threads = 48;
2146 		num_gs_threads = 4;
2147 		num_es_threads = 4;
2148 		num_ps_stack_entries = 128;
2149 		num_vs_stack_entries = 128;
2150 		num_gs_stack_entries = 0;
2151 		num_es_stack_entries = 0;
2152 		break;
2153 	case CHIP_RV630:
2154 	case CHIP_RV635:
2155 		num_ps_gprs = 84;
2156 		num_vs_gprs = 36;
2157 		num_temp_gprs = 4;
2158 		num_gs_gprs = 0;
2159 		num_es_gprs = 0;
2160 		num_ps_threads = 144;
2161 		num_vs_threads = 40;
2162 		num_gs_threads = 4;
2163 		num_es_threads = 4;
2164 		num_ps_stack_entries = 40;
2165 		num_vs_stack_entries = 40;
2166 		num_gs_stack_entries = 32;
2167 		num_es_stack_entries = 16;
2168 		break;
2169 	case CHIP_RV610:
2170 	case CHIP_RV620:
2171 	case CHIP_RS780:
2172 	case CHIP_RS880:
2173 	default:
2174 		num_ps_gprs = 84;
2175 		num_vs_gprs = 36;
2176 		num_temp_gprs = 4;
2177 		num_gs_gprs = 0;
2178 		num_es_gprs = 0;
2179 		/* use limits 40 VS and at least 16 ES/GS */
2180 		num_ps_threads = 120;
2181 		num_vs_threads = 40;
2182 		num_gs_threads = 16;
2183 		num_es_threads = 16;
2184 		num_ps_stack_entries = 40;
2185 		num_vs_stack_entries = 40;
2186 		num_gs_stack_entries = 32;
2187 		num_es_stack_entries = 16;
2188 		break;
2189 	case CHIP_RV670:
2190 		num_ps_gprs = 144;
2191 		num_vs_gprs = 40;
2192 		num_temp_gprs = 4;
2193 		num_gs_gprs = 0;
2194 		num_es_gprs = 0;
2195 		num_ps_threads = 136;
2196 		num_vs_threads = 48;
2197 		num_gs_threads = 4;
2198 		num_es_threads = 4;
2199 		num_ps_stack_entries = 40;
2200 		num_vs_stack_entries = 40;
2201 		num_gs_stack_entries = 32;
2202 		num_es_stack_entries = 16;
2203 		break;
2204 	case CHIP_RV770:
2205 		num_ps_gprs = 130;
2206 		num_vs_gprs = 56;
2207 		num_temp_gprs = 4;
2208 		num_gs_gprs = 31;
2209 		num_es_gprs = 31;
2210 		num_ps_threads = 180;
2211 		num_vs_threads = 60;
2212 		num_gs_threads = 4;
2213 		num_es_threads = 4;
2214 		num_ps_stack_entries = 128;
2215 		num_vs_stack_entries = 128;
2216 		num_gs_stack_entries = 128;
2217 		num_es_stack_entries = 128;
2218 		break;
2219 	case CHIP_RV730:
2220 	case CHIP_RV740:
2221 		num_ps_gprs = 84;
2222 		num_vs_gprs = 36;
2223 		num_temp_gprs = 4;
2224 		num_gs_gprs = 0;
2225 		num_es_gprs = 0;
2226 		num_ps_threads = 180;
2227 		num_vs_threads = 60;
2228 		num_gs_threads = 4;
2229 		num_es_threads = 4;
2230 		num_ps_stack_entries = 128;
2231 		num_vs_stack_entries = 128;
2232 		num_gs_stack_entries = 0;
2233 		num_es_stack_entries = 0;
2234 		break;
2235 	case CHIP_RV710:
2236 		num_ps_gprs = 192;
2237 		num_vs_gprs = 56;
2238 		num_temp_gprs = 4;
2239 		num_gs_gprs = 0;
2240 		num_es_gprs = 0;
2241 		num_ps_threads = 136;
2242 		num_vs_threads = 48;
2243 		num_gs_threads = 4;
2244 		num_es_threads = 4;
2245 		num_ps_stack_entries = 128;
2246 		num_vs_stack_entries = 128;
2247 		num_gs_stack_entries = 0;
2248 		num_es_stack_entries = 0;
2249 		break;
2250 	}
2251 
2252 	rctx->default_gprs[R600_HW_STAGE_PS] = num_ps_gprs;
2253 	rctx->default_gprs[R600_HW_STAGE_VS] = num_vs_gprs;
2254 	rctx->default_gprs[R600_HW_STAGE_GS] = 0;
2255 	rctx->default_gprs[R600_HW_STAGE_ES] = 0;
2256 
2257 	rctx->r6xx_num_clause_temp_gprs = num_temp_gprs;
2258 
2259 	/* SQ_CONFIG */
2260 	tmp = 0;
2261 	switch (family) {
2262 	case CHIP_RV610:
2263 	case CHIP_RV620:
2264 	case CHIP_RS780:
2265 	case CHIP_RS880:
2266 	case CHIP_RV710:
2267 		break;
2268 	default:
2269 		tmp |= S_008C00_VC_ENABLE(1);
2270 		break;
2271 	}
2272 	tmp |= S_008C00_DX9_CONSTS(0);
2273 	tmp |= S_008C00_ALU_INST_PREFER_VECTOR(1);
2274 	tmp |= S_008C00_PS_PRIO(ps_prio);
2275 	tmp |= S_008C00_VS_PRIO(vs_prio);
2276 	tmp |= S_008C00_GS_PRIO(gs_prio);
2277 	tmp |= S_008C00_ES_PRIO(es_prio);
2278 	r600_store_config_reg(cb, R_008C00_SQ_CONFIG, tmp);
2279 
2280 	/* SQ_GPR_RESOURCE_MGMT_2 */
2281 	tmp = S_008C08_NUM_GS_GPRS(num_gs_gprs);
2282 	tmp |= S_008C08_NUM_ES_GPRS(num_es_gprs);
2283 	r600_store_config_reg_seq(cb, R_008C08_SQ_GPR_RESOURCE_MGMT_2, 4);
2284 	r600_store_value(cb, tmp);
2285 
2286 	/* SQ_THREAD_RESOURCE_MGMT */
2287 	tmp = S_008C0C_NUM_PS_THREADS(num_ps_threads);
2288 	tmp |= S_008C0C_NUM_VS_THREADS(num_vs_threads);
2289 	tmp |= S_008C0C_NUM_GS_THREADS(num_gs_threads);
2290 	tmp |= S_008C0C_NUM_ES_THREADS(num_es_threads);
2291 	r600_store_value(cb, tmp); /* R_008C0C_SQ_THREAD_RESOURCE_MGMT */
2292 
2293 	/* SQ_STACK_RESOURCE_MGMT_1 */
2294 	tmp = S_008C10_NUM_PS_STACK_ENTRIES(num_ps_stack_entries);
2295 	tmp |= S_008C10_NUM_VS_STACK_ENTRIES(num_vs_stack_entries);
2296 	r600_store_value(cb, tmp); /* R_008C10_SQ_STACK_RESOURCE_MGMT_1 */
2297 
2298 	/* SQ_STACK_RESOURCE_MGMT_2 */
2299 	tmp = S_008C14_NUM_GS_STACK_ENTRIES(num_gs_stack_entries);
2300 	tmp |= S_008C14_NUM_ES_STACK_ENTRIES(num_es_stack_entries);
2301 	r600_store_value(cb, tmp); /* R_008C14_SQ_STACK_RESOURCE_MGMT_2 */
2302 
2303 	r600_store_config_reg(cb, R_009714_VC_ENHANCE, 0);
2304 
2305 	if (rctx->b.chip_class >= R700) {
2306 		r600_store_context_reg(cb, R_028A50_VGT_ENHANCE, 4);
2307 		r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0x00004000);
2308 		r600_store_config_reg(cb, R_009830_DB_DEBUG, 0);
2309 		r600_store_config_reg(cb, R_009838_DB_WATERMARKS, 0x00420204);
2310 		r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0);
2311 	} else {
2312 		r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0);
2313 		r600_store_config_reg(cb, R_009830_DB_DEBUG, 0x82000000);
2314 		r600_store_config_reg(cb, R_009838_DB_WATERMARKS, 0x01020204);
2315 		r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 1);
2316 	}
2317 	r600_store_context_reg_seq(cb, R_0288A8_SQ_ESGS_RING_ITEMSIZE, 9);
2318 	r600_store_value(cb, 0); /* R_0288A8_SQ_ESGS_RING_ITEMSIZE */
2319 	r600_store_value(cb, 0); /* R_0288AC_SQ_GSVS_RING_ITEMSIZE */
2320 	r600_store_value(cb, 0); /* R_0288B0_SQ_ESTMP_RING_ITEMSIZE */
2321 	r600_store_value(cb, 0); /* R_0288B4_SQ_GSTMP_RING_ITEMSIZE */
2322 	r600_store_value(cb, 0); /* R_0288B8_SQ_VSTMP_RING_ITEMSIZE */
2323 	r600_store_value(cb, 0); /* R_0288BC_SQ_PSTMP_RING_ITEMSIZE */
2324 	r600_store_value(cb, 0); /* R_0288C0_SQ_FBUF_RING_ITEMSIZE */
2325 	r600_store_value(cb, 0); /* R_0288C4_SQ_REDUC_RING_ITEMSIZE */
2326 	r600_store_value(cb, 0); /* R_0288C8_SQ_GS_VERT_ITEMSIZE */
2327 
2328 	/* to avoid GPU doing any preloading of constant from random address */
2329 	r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16);
2330 	for (i = 0; i < 16; i++)
2331 		r600_store_value(cb, 0);
2332 
2333 	r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16);
2334 	for (i = 0; i < 16; i++)
2335 		r600_store_value(cb, 0);
2336 
2337 	r600_store_context_reg_seq(cb, R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 16);
2338 	for (i = 0; i < 16; i++)
2339 		r600_store_value(cb, 0);
2340 
2341 	r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13);
2342 	r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */
2343 	r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */
2344 	r600_store_value(cb, 0); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */
2345 	r600_store_value(cb, 0); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */
2346 	r600_store_value(cb, 0); /* R_028A20_VGT_HOS_REUSE_DEPTH */
2347 	r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */
2348 	r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */
2349 	r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */
2350 	r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */
2351 	r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */
2352 	r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */
2353 	r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */
2354 	r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE, 0); */
2355 
2356 	r600_store_context_reg(cb, R_028A84_VGT_PRIMITIVEID_EN, 0);
2357 	r600_store_context_reg(cb, R_028AA0_VGT_INSTANCE_STEP_RATE_0, 0);
2358 	r600_store_context_reg(cb, R_028AA4_VGT_INSTANCE_STEP_RATE_1, 0);
2359 
2360 	r600_store_context_reg_seq(cb, R_028AB4_VGT_REUSE_OFF, 2);
2361 	r600_store_value(cb, 1); /* R_028AB4_VGT_REUSE_OFF */
2362 	r600_store_value(cb, 0); /* R_028AB8_VGT_VTX_CNT_EN */
2363 
2364 	r600_store_context_reg(cb, R_028B20_VGT_STRMOUT_BUFFER_EN, 0);
2365 
2366 	r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0);
2367 
2368 	r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0);
2369 
2370 	r600_store_context_reg_seq(cb, R_0286DC_SPI_FOG_CNTL, 3);
2371 	r600_store_value(cb, 0); /* R_0286DC_SPI_FOG_CNTL */
2372 	r600_store_value(cb, 0); /* R_0286E0_SPI_FOG_FUNC_SCALE */
2373 	r600_store_value(cb, 0); /* R_0286E4_SPI_FOG_FUNC_BIAS */
2374 
2375 	r600_store_context_reg_seq(cb, R_028D28_DB_SRESULTS_COMPARE_STATE0, 3);
2376 	r600_store_value(cb, 0); /* R_028D28_DB_SRESULTS_COMPARE_STATE0 */
2377 	r600_store_value(cb, 0); /* R_028D2C_DB_SRESULTS_COMPARE_STATE1 */
2378 	r600_store_value(cb, 0); /* R_028D30_DB_PRELOAD_CONTROL */
2379 
2380 	r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0);
2381 	r600_store_context_reg(cb, R_028A48_PA_SC_MPASS_PS_CNTL, 0);
2382 
2383 	r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0);
2384 	r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
2385 
2386 	if (rctx->b.chip_class >= R700) {
2387 		r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
2388 	}
2389 
2390 	r600_store_context_reg_seq(cb, R_028C30_CB_CLRCMP_CONTROL, 4);
2391 	r600_store_value(cb, 0x1000000);  /* R_028C30_CB_CLRCMP_CONTROL */
2392 	r600_store_value(cb, 0);          /* R_028C34_CB_CLRCMP_SRC */
2393 	r600_store_value(cb, 0xFF);       /* R_028C38_CB_CLRCMP_DST */
2394 	r600_store_value(cb, 0xFFFFFFFF); /* R_028C3C_CB_CLRCMP_MSK */
2395 
2396 	r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2);
2397 	r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */
2398 	r600_store_value(cb, S_028034_BR_X(8192) | S_028034_BR_Y(8192)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */
2399 
2400 	r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2);
2401 	r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */
2402 	r600_store_value(cb, S_028244_BR_X(8192) | S_028244_BR_Y(8192)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */
2403 
2404 	r600_store_context_reg_seq(cb, R_0288CC_SQ_PGM_CF_OFFSET_PS, 5);
2405 	r600_store_value(cb, 0); /* R_0288CC_SQ_PGM_CF_OFFSET_PS */
2406 	r600_store_value(cb, 0); /* R_0288D0_SQ_PGM_CF_OFFSET_VS */
2407 	r600_store_value(cb, 0); /* R_0288D4_SQ_PGM_CF_OFFSET_GS */
2408 	r600_store_value(cb, 0); /* R_0288D8_SQ_PGM_CF_OFFSET_ES */
2409 	r600_store_value(cb, 0); /* R_0288DC_SQ_PGM_CF_OFFSET_FS */
2410 
2411         r600_store_context_reg(cb, R_0288E0_SQ_VTX_SEMANTIC_CLEAR, ~0);
2412 
2413         r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2);
2414 	r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */
2415 	r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */
2416 
2417 	r600_store_context_reg(cb, R_0288A4_SQ_PGM_RESOURCES_FS, 0);
2418 
2419 	if (rctx->b.chip_class == R700)
2420 		r600_store_context_reg(cb, R_028350_SX_MISC, 0);
2421 	if (rctx->b.chip_class == R700 && rctx->screen->b.has_streamout)
2422 		r600_store_context_reg(cb, R_028354_SX_SURFACE_SYNC, S_028354_SURFACE_SYNC_MASK(0xf));
2423 
2424 	r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0);
2425 	if (rctx->screen->b.has_streamout) {
2426 		r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
2427 	}
2428 
2429 	r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0, 0x1000FFF);
2430 	r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0 + (32 * 4), 0x1000FFF);
2431 	r600_store_loop_const(cb, R_03E200_SQ_LOOP_CONST_0 + (64 * 4), 0x1000FFF);
2432 }
2433 
r600_update_ps_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2434 void r600_update_ps_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2435 {
2436 	struct r600_context *rctx = (struct r600_context *)ctx;
2437 	struct r600_command_buffer *cb = &shader->command_buffer;
2438 	struct r600_shader *rshader = &shader->shader;
2439 	unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1, db_shader_control;
2440 	int pos_index = -1, face_index = -1, fixed_pt_position_index = -1;
2441 	unsigned tmp, sid, ufi = 0;
2442 	int need_linear = 0;
2443 	unsigned z_export = 0, stencil_export = 0, mask_export = 0;
2444 	unsigned sprite_coord_enable = rctx->rasterizer ? rctx->rasterizer->sprite_coord_enable : 0;
2445 
2446 	if (!cb->buf) {
2447 		r600_init_command_buffer(cb, 64);
2448 	} else {
2449 		cb->num_dw = 0;
2450 	}
2451 
2452 	r600_store_context_reg_seq(cb, R_028644_SPI_PS_INPUT_CNTL_0, rshader->ninput);
2453 	for (i = 0; i < rshader->ninput; i++) {
2454 		if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
2455 			pos_index = i;
2456 		if (rshader->input[i].name == TGSI_SEMANTIC_FACE && face_index == -1)
2457 			face_index = i;
2458 		if (rshader->input[i].name == TGSI_SEMANTIC_SAMPLEID)
2459 			fixed_pt_position_index = i;
2460 
2461 		sid = rshader->input[i].spi_sid;
2462 
2463 		tmp = S_028644_SEMANTIC(sid);
2464 
2465 		/* D3D 9 behaviour. GL is undefined */
2466 		if (rshader->input[i].name == TGSI_SEMANTIC_COLOR && rshader->input[i].sid == 0)
2467 			tmp |= S_028644_DEFAULT_VAL(3);
2468 
2469 		if (rshader->input[i].name == TGSI_SEMANTIC_POSITION ||
2470 			rshader->input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
2471 			(rshader->input[i].interpolate == TGSI_INTERPOLATE_COLOR &&
2472 				rctx->rasterizer && rctx->rasterizer->flatshade))
2473 			tmp |= S_028644_FLAT_SHADE(1);
2474 
2475 		if (rshader->input[i].name == TGSI_SEMANTIC_PCOORD ||
2476 		    (rshader->input[i].name == TGSI_SEMANTIC_TEXCOORD &&
2477 		     sprite_coord_enable & (1 << rshader->input[i].sid))) {
2478 			tmp |= S_028644_PT_SPRITE_TEX(1);
2479 		}
2480 
2481 		if (rshader->input[i].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID)
2482 			tmp |= S_028644_SEL_CENTROID(1);
2483 
2484 		if (rshader->input[i].interpolate_location == TGSI_INTERPOLATE_LOC_SAMPLE)
2485 			tmp |= S_028644_SEL_SAMPLE(1);
2486 
2487 		if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR) {
2488 			need_linear = 1;
2489 			tmp |= S_028644_SEL_LINEAR(1);
2490 		}
2491 
2492 		r600_store_value(cb, tmp);
2493 	}
2494 
2495 	db_shader_control = 0;
2496 	for (i = 0; i < rshader->noutput; i++) {
2497 		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
2498 			z_export = 1;
2499 		if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
2500 			stencil_export = 1;
2501 		if (rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK &&
2502 			rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0)
2503 			mask_export = 1;
2504 	}
2505 	db_shader_control |= S_02880C_Z_EXPORT_ENABLE(z_export);
2506 	db_shader_control |= S_02880C_STENCIL_REF_EXPORT_ENABLE(stencil_export);
2507 	db_shader_control |= S_02880C_MASK_EXPORT_ENABLE(mask_export);
2508 	if (rshader->uses_kill)
2509 		db_shader_control |= S_02880C_KILL_ENABLE(1);
2510 
2511 	exports_ps = 0;
2512 	for (i = 0; i < rshader->noutput; i++) {
2513 		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||
2514 		    rshader->output[i].name == TGSI_SEMANTIC_STENCIL ||
2515 		    rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK) {
2516 			exports_ps |= 1;
2517 		}
2518 	}
2519 	num_cout = rshader->nr_ps_color_exports;
2520 	exports_ps |= S_028854_EXPORT_COLORS(num_cout);
2521 	if (!exports_ps) {
2522 		/* always at least export 1 component per pixel */
2523 		exports_ps = 2;
2524 	}
2525 
2526 	shader->nr_ps_color_outputs = num_cout;
2527 	shader->ps_color_export_mask = rshader->ps_color_export_mask;
2528 
2529 	spi_ps_in_control_0 = S_0286CC_NUM_INTERP(rshader->ninput) |
2530 				S_0286CC_PERSP_GRADIENT_ENA(1)|
2531 				S_0286CC_LINEAR_GRADIENT_ENA(need_linear);
2532 	spi_input_z = 0;
2533 	if (pos_index != -1) {
2534 		spi_ps_in_control_0 |= (S_0286CC_POSITION_ENA(1) |
2535 					S_0286CC_POSITION_CENTROID(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID) |
2536 					S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr) |
2537 					S_0286CC_BARYC_SAMPLE_CNTL(1)) |
2538 					S_0286CC_POSITION_SAMPLE(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_SAMPLE);
2539 		spi_input_z |= S_0286D8_PROVIDE_Z_TO_SPI(1);
2540 	}
2541 
2542 	spi_ps_in_control_1 = 0;
2543 	if (face_index != -1) {
2544 		spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) |
2545 			S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr);
2546 	}
2547 	if (fixed_pt_position_index != -1) {
2548 		spi_ps_in_control_1 |= S_0286D0_FIXED_PT_POSITION_ENA(1) |
2549 			S_0286D0_FIXED_PT_POSITION_ADDR(rshader->input[fixed_pt_position_index].gpr);
2550 	}
2551 
2552 	/* HW bug in original R600 */
2553 	if (rctx->b.family == CHIP_R600)
2554 		ufi = 1;
2555 
2556 	r600_store_context_reg_seq(cb, R_0286CC_SPI_PS_IN_CONTROL_0, 2);
2557 	r600_store_value(cb, spi_ps_in_control_0); /* R_0286CC_SPI_PS_IN_CONTROL_0 */
2558 	r600_store_value(cb, spi_ps_in_control_1); /* R_0286D0_SPI_PS_IN_CONTROL_1 */
2559 
2560 	r600_store_context_reg(cb, R_0286D8_SPI_INPUT_Z, spi_input_z);
2561 
2562 	r600_store_context_reg_seq(cb, R_028850_SQ_PGM_RESOURCES_PS, 2);
2563 	r600_store_value(cb, /* R_028850_SQ_PGM_RESOURCES_PS*/
2564 			 S_028850_NUM_GPRS(rshader->bc.ngpr) |
2565 	/*
2566 	 * docs are misleading about the dx10_clamp bit. This only affects
2567 	 * instructions using CLAMP dst modifier, in which case they will
2568 	 * return 0 with this set for a NaN (otherwise NaN).
2569 	 */
2570 			 S_028850_DX10_CLAMP(1) |
2571 			 S_028850_STACK_SIZE(rshader->bc.nstack) |
2572 			 S_028850_UNCACHED_FIRST_INST(ufi));
2573 	r600_store_value(cb, exports_ps); /* R_028854_SQ_PGM_EXPORTS_PS */
2574 
2575 	r600_store_context_reg(cb, R_028840_SQ_PGM_START_PS, 0);
2576 	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2577 
2578 	/* only set some bits here, the other bits are set in the dsa state */
2579 	shader->db_shader_control = db_shader_control;
2580 	shader->ps_depth_export = z_export | stencil_export | mask_export;
2581 
2582 	shader->sprite_coord_enable = sprite_coord_enable;
2583 	if (rctx->rasterizer)
2584 		shader->flatshade = rctx->rasterizer->flatshade;
2585 }
2586 
r600_update_vs_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2587 void r600_update_vs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2588 {
2589 	struct r600_command_buffer *cb = &shader->command_buffer;
2590 	struct r600_shader *rshader = &shader->shader;
2591 	unsigned spi_vs_out_id[10] = {};
2592 	unsigned i, tmp, nparams = 0;
2593 
2594 	for (i = 0; i < rshader->noutput; i++) {
2595 		if (rshader->output[i].spi_sid) {
2596 			tmp = rshader->output[i].spi_sid << ((nparams & 3) * 8);
2597 			spi_vs_out_id[nparams / 4] |= tmp;
2598 			nparams++;
2599 		}
2600 	}
2601 
2602 	r600_init_command_buffer(cb, 32);
2603 
2604 	r600_store_context_reg_seq(cb, R_028614_SPI_VS_OUT_ID_0, 10);
2605 	for (i = 0; i < 10; i++) {
2606 		r600_store_value(cb, spi_vs_out_id[i]);
2607 	}
2608 
2609 	/* Certain attributes (position, psize, etc.) don't count as params.
2610 	 * VS is required to export at least one param and r600_shader_from_tgsi()
2611 	 * takes care of adding a dummy export.
2612 	 */
2613 	if (nparams < 1)
2614 		nparams = 1;
2615 
2616 	r600_store_context_reg(cb, R_0286C4_SPI_VS_OUT_CONFIG,
2617 			       S_0286C4_VS_EXPORT_COUNT(nparams - 1));
2618 	r600_store_context_reg(cb, R_028868_SQ_PGM_RESOURCES_VS,
2619 			       S_028868_NUM_GPRS(rshader->bc.ngpr) |
2620 			       S_028868_DX10_CLAMP(1) |
2621 			       S_028868_STACK_SIZE(rshader->bc.nstack));
2622 	if (rshader->vs_position_window_space) {
2623 		r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
2624 			S_028818_VTX_XY_FMT(1) | S_028818_VTX_Z_FMT(1));
2625 	} else {
2626 		r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL,
2627 			S_028818_VTX_W0_FMT(1) |
2628 			S_028818_VPORT_X_SCALE_ENA(1) | S_028818_VPORT_X_OFFSET_ENA(1) |
2629 			S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
2630 			S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1));
2631 
2632 	}
2633 	r600_store_context_reg(cb, R_028858_SQ_PGM_START_VS, 0);
2634 	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2635 
2636 	shader->pa_cl_vs_out_cntl =
2637 		S_02881C_VS_OUT_CCDIST0_VEC_ENA((rshader->clip_dist_write & 0x0F) != 0) |
2638 		S_02881C_VS_OUT_CCDIST1_VEC_ENA((rshader->clip_dist_write & 0xF0) != 0) |
2639 		S_02881C_VS_OUT_MISC_VEC_ENA(rshader->vs_out_misc_write) |
2640 		S_02881C_USE_VTX_POINT_SIZE(rshader->vs_out_point_size) |
2641 		S_02881C_USE_VTX_EDGE_FLAG(rshader->vs_out_edgeflag) |
2642 		S_02881C_USE_VTX_RENDER_TARGET_INDX(rshader->vs_out_layer) |
2643 		S_02881C_USE_VTX_VIEWPORT_INDX(rshader->vs_out_viewport);
2644 }
2645 
2646 #define RV610_GSVS_ALIGN 32
2647 #define R600_GSVS_ALIGN 16
2648 
r600_update_gs_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2649 void r600_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2650 {
2651 	struct r600_context *rctx = (struct r600_context *)ctx;
2652 	struct r600_command_buffer *cb = &shader->command_buffer;
2653 	struct r600_shader *rshader = &shader->shader;
2654 	struct r600_shader *cp_shader = &shader->gs_copy_shader->shader;
2655 	unsigned gsvs_itemsize =
2656 			(cp_shader->ring_item_sizes[0] * shader->selector->gs_max_out_vertices) >> 2;
2657 
2658 	/* some r600s needs gsvs itemsize aligned to cacheline size
2659 	   this was fixed in rs780 and above. */
2660 	switch (rctx->b.family) {
2661 	case CHIP_RV610:
2662 		gsvs_itemsize = align(gsvs_itemsize, RV610_GSVS_ALIGN);
2663 		break;
2664 	case CHIP_R600:
2665 	case CHIP_RV630:
2666 	case CHIP_RV670:
2667 	case CHIP_RV620:
2668 	case CHIP_RV635:
2669 		gsvs_itemsize = align(gsvs_itemsize, R600_GSVS_ALIGN);
2670 		break;
2671 	default:
2672 		break;
2673 	}
2674 
2675 	r600_init_command_buffer(cb, 64);
2676 
2677 	/* VGT_GS_MODE is written by r600_emit_shader_stages */
2678 	r600_store_context_reg(cb, R_028AB8_VGT_VTX_CNT_EN, 1);
2679 
2680 	if (rctx->b.chip_class >= R700) {
2681 		r600_store_context_reg(cb, R_028B38_VGT_GS_MAX_VERT_OUT,
2682 				       S_028B38_MAX_VERT_OUT(shader->selector->gs_max_out_vertices));
2683 	}
2684 	r600_store_context_reg(cb, R_028A6C_VGT_GS_OUT_PRIM_TYPE,
2685 			       r600_conv_prim_to_gs_out(shader->selector->gs_output_prim));
2686 
2687 	r600_store_context_reg(cb, R_0288C8_SQ_GS_VERT_ITEMSIZE,
2688 	                       cp_shader->ring_item_sizes[0] >> 2);
2689 
2690 	r600_store_context_reg(cb, R_0288A8_SQ_ESGS_RING_ITEMSIZE,
2691 			       (rshader->ring_item_sizes[0]) >> 2);
2692 
2693 	r600_store_context_reg(cb, R_0288AC_SQ_GSVS_RING_ITEMSIZE,
2694 			       gsvs_itemsize);
2695 
2696 	/* FIXME calculate these values somehow ??? */
2697 	r600_store_config_reg_seq(cb, R_0088C8_VGT_GS_PER_ES, 2);
2698 	r600_store_value(cb, 0x80); /* GS_PER_ES */
2699 	r600_store_value(cb, 0x100); /* ES_PER_GS */
2700 	r600_store_config_reg_seq(cb, R_0088E8_VGT_GS_PER_VS, 1);
2701 	r600_store_value(cb, 0x2); /* GS_PER_VS */
2702 
2703 	r600_store_context_reg(cb, R_02887C_SQ_PGM_RESOURCES_GS,
2704 			       S_02887C_NUM_GPRS(rshader->bc.ngpr) |
2705 			       S_02887C_DX10_CLAMP(1) |
2706 			       S_02887C_STACK_SIZE(rshader->bc.nstack));
2707 	r600_store_context_reg(cb, R_02886C_SQ_PGM_START_GS, 0);
2708 	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2709 }
2710 
r600_update_es_state(struct pipe_context * ctx,struct r600_pipe_shader * shader)2711 void r600_update_es_state(struct pipe_context *ctx, struct r600_pipe_shader *shader)
2712 {
2713 	struct r600_command_buffer *cb = &shader->command_buffer;
2714 	struct r600_shader *rshader = &shader->shader;
2715 
2716 	r600_init_command_buffer(cb, 32);
2717 
2718 	r600_store_context_reg(cb, R_028890_SQ_PGM_RESOURCES_ES,
2719 			       S_028890_NUM_GPRS(rshader->bc.ngpr) |
2720 			       S_028890_DX10_CLAMP(1) |
2721 			       S_028890_STACK_SIZE(rshader->bc.nstack));
2722 	r600_store_context_reg(cb, R_028880_SQ_PGM_START_ES, 0);
2723 	/* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */
2724 }
2725 
2726 
r600_create_resolve_blend(struct r600_context * rctx)2727 void *r600_create_resolve_blend(struct r600_context *rctx)
2728 {
2729 	struct pipe_blend_state blend;
2730 	unsigned i;
2731 
2732 	memset(&blend, 0, sizeof(blend));
2733 	blend.independent_blend_enable = true;
2734 	for (i = 0; i < 2; i++) {
2735 		blend.rt[i].colormask = 0xf;
2736 		blend.rt[i].blend_enable = 1;
2737 		blend.rt[i].rgb_func = PIPE_BLEND_ADD;
2738 		blend.rt[i].alpha_func = PIPE_BLEND_ADD;
2739 		blend.rt[i].rgb_src_factor = PIPE_BLENDFACTOR_ZERO;
2740 		blend.rt[i].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
2741 		blend.rt[i].alpha_src_factor = PIPE_BLENDFACTOR_ZERO;
2742 		blend.rt[i].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
2743 	}
2744 	return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_RESOLVE_BOX);
2745 }
2746 
r700_create_resolve_blend(struct r600_context * rctx)2747 void *r700_create_resolve_blend(struct r600_context *rctx)
2748 {
2749 	struct pipe_blend_state blend;
2750 
2751 	memset(&blend, 0, sizeof(blend));
2752 	blend.independent_blend_enable = true;
2753 	blend.rt[0].colormask = 0xf;
2754 	return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_RESOLVE_BOX);
2755 }
2756 
r600_create_decompress_blend(struct r600_context * rctx)2757 void *r600_create_decompress_blend(struct r600_context *rctx)
2758 {
2759 	struct pipe_blend_state blend;
2760 
2761 	memset(&blend, 0, sizeof(blend));
2762 	blend.independent_blend_enable = true;
2763 	blend.rt[0].colormask = 0xf;
2764 	return r600_create_blend_state_mode(&rctx->b.b, &blend, V_028808_SPECIAL_EXPAND_SAMPLES);
2765 }
2766 
r600_create_db_flush_dsa(struct r600_context * rctx)2767 void *r600_create_db_flush_dsa(struct r600_context *rctx)
2768 {
2769 	struct pipe_depth_stencil_alpha_state dsa;
2770 	boolean quirk = false;
2771 
2772 	if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
2773 		rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
2774 		quirk = true;
2775 
2776 	memset(&dsa, 0, sizeof(dsa));
2777 
2778 	if (quirk) {
2779 		dsa.depth.enabled = 1;
2780 		dsa.depth.func = PIPE_FUNC_LEQUAL;
2781 		dsa.stencil[0].enabled = 1;
2782 		dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
2783 		dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_KEEP;
2784 		dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_INCR;
2785 		dsa.stencil[0].writemask = 0xff;
2786 	}
2787 
2788 	return rctx->b.b.create_depth_stencil_alpha_state(&rctx->b.b, &dsa);
2789 }
2790 
r600_update_db_shader_control(struct r600_context * rctx)2791 void r600_update_db_shader_control(struct r600_context * rctx)
2792 {
2793 	bool dual_export;
2794 	unsigned db_shader_control;
2795 	uint8_t ps_conservative_z;
2796 
2797 	if (!rctx->ps_shader) {
2798 		return;
2799 	}
2800 
2801 	dual_export = rctx->framebuffer.export_16bpc &&
2802 		      !rctx->ps_shader->current->ps_depth_export;
2803 
2804 	db_shader_control = rctx->ps_shader->current->db_shader_control |
2805 			    S_02880C_DUAL_EXPORT_ENABLE(dual_export);
2806 
2807 	ps_conservative_z = rctx->ps_shader->current->shader.ps_conservative_z;
2808 
2809 	/* When alpha test is enabled we can't trust the hw to make the proper
2810 	 * decision on the order in which ztest should be run related to fragment
2811 	 * shader execution.
2812 	 *
2813 	 * If alpha test is enabled perform z test after fragment. RE_Z (early
2814 	 * z test but no write to the zbuffer) seems to cause lockup on r6xx/r7xx
2815 	 */
2816 	if (rctx->alphatest_state.sx_alpha_test_control) {
2817 		db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z);
2818 	} else {
2819 		db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
2820 	}
2821 
2822 	if (db_shader_control != rctx->db_misc_state.db_shader_control ||
2823 		ps_conservative_z != rctx->db_misc_state.ps_conservative_z) {
2824 		rctx->db_misc_state.db_shader_control = db_shader_control;
2825 		rctx->db_misc_state.ps_conservative_z = ps_conservative_z;
2826 		r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
2827 	}
2828 }
2829 
r600_array_mode(unsigned mode)2830 static inline unsigned r600_array_mode(unsigned mode)
2831 {
2832 	switch (mode) {
2833 	default:
2834 	case RADEON_SURF_MODE_LINEAR_ALIGNED:	return V_0280A0_ARRAY_LINEAR_ALIGNED;
2835 		break;
2836 	case RADEON_SURF_MODE_1D:		return V_0280A0_ARRAY_1D_TILED_THIN1;
2837 		break;
2838 	case RADEON_SURF_MODE_2D:		return V_0280A0_ARRAY_2D_TILED_THIN1;
2839 	}
2840 }
2841 
r600_dma_copy_tile(struct r600_context * rctx,struct pipe_resource * dst,unsigned dst_level,unsigned dst_x,unsigned dst_y,unsigned dst_z,struct pipe_resource * src,unsigned src_level,unsigned src_x,unsigned src_y,unsigned src_z,unsigned copy_height,unsigned pitch,unsigned bpp)2842 static boolean r600_dma_copy_tile(struct r600_context *rctx,
2843 				struct pipe_resource *dst,
2844 				unsigned dst_level,
2845 				unsigned dst_x,
2846 				unsigned dst_y,
2847 				unsigned dst_z,
2848 				struct pipe_resource *src,
2849 				unsigned src_level,
2850 				unsigned src_x,
2851 				unsigned src_y,
2852 				unsigned src_z,
2853 				unsigned copy_height,
2854 				unsigned pitch,
2855 				unsigned bpp)
2856 {
2857 	struct radeon_cmdbuf *cs = rctx->b.dma.cs;
2858 	struct r600_texture *rsrc = (struct r600_texture*)src;
2859 	struct r600_texture *rdst = (struct r600_texture*)dst;
2860 	unsigned array_mode, lbpp, pitch_tile_max, slice_tile_max, size;
2861 	unsigned ncopy, height, cheight, detile, i, x, y, z, src_mode, dst_mode;
2862 	uint64_t base, addr;
2863 
2864 	dst_mode = rdst->surface.u.legacy.level[dst_level].mode;
2865 	src_mode = rsrc->surface.u.legacy.level[src_level].mode;
2866 	assert(dst_mode != src_mode);
2867 
2868 	y = 0;
2869 	lbpp = util_logbase2(bpp);
2870 	pitch_tile_max = ((pitch / bpp) / 8) - 1;
2871 
2872 	if (dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED) {
2873 		/* T2L */
2874 		array_mode = r600_array_mode(src_mode);
2875 		slice_tile_max = (rsrc->surface.u.legacy.level[src_level].nblk_x * rsrc->surface.u.legacy.level[src_level].nblk_y) / (8*8);
2876 		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
2877 		/* linear height must be the same as the slice tile max height, it's ok even
2878 		 * if the linear destination/source have smaller heigh as the size of the
2879 		 * dma packet will be using the copy_height which is always smaller or equal
2880 		 * to the linear height
2881 		 */
2882 		height = u_minify(rsrc->resource.b.b.height0, src_level);
2883 		detile = 1;
2884 		x = src_x;
2885 		y = src_y;
2886 		z = src_z;
2887 		base = rsrc->surface.u.legacy.level[src_level].offset;
2888 		addr = rdst->surface.u.legacy.level[dst_level].offset;
2889 		addr += (uint64_t)rdst->surface.u.legacy.level[dst_level].slice_size_dw * 4 * dst_z;
2890 		addr += dst_y * pitch + dst_x * bpp;
2891 	} else {
2892 		/* L2T */
2893 		array_mode = r600_array_mode(dst_mode);
2894 		slice_tile_max = (rdst->surface.u.legacy.level[dst_level].nblk_x * rdst->surface.u.legacy.level[dst_level].nblk_y) / (8*8);
2895 		slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0;
2896 		/* linear height must be the same as the slice tile max height, it's ok even
2897 		 * if the linear destination/source have smaller heigh as the size of the
2898 		 * dma packet will be using the copy_height which is always smaller or equal
2899 		 * to the linear height
2900 		 */
2901 		height = u_minify(rdst->resource.b.b.height0, dst_level);
2902 		detile = 0;
2903 		x = dst_x;
2904 		y = dst_y;
2905 		z = dst_z;
2906 		base = rdst->surface.u.legacy.level[dst_level].offset;
2907 		addr = rsrc->surface.u.legacy.level[src_level].offset;
2908 		addr += (uint64_t)rsrc->surface.u.legacy.level[src_level].slice_size_dw * 4 * src_z;
2909 		addr += src_y * pitch + src_x * bpp;
2910 	}
2911 	/* check that we are in dw/base alignment constraint */
2912 	if (addr % 4 || base % 256) {
2913 		return FALSE;
2914 	}
2915 
2916 	/* It's a r6xx/r7xx limitation, the blit must be on 8 boundary for number
2917 	 * line in the blit. Compute max 8 line we can copy in the size limit
2918 	 */
2919 	cheight = ((R600_DMA_COPY_MAX_SIZE_DW * 4) / pitch) & 0xfffffff8;
2920 	ncopy = (copy_height / cheight) + !!(copy_height % cheight);
2921 	r600_need_dma_space(&rctx->b, ncopy * 7, &rdst->resource, &rsrc->resource);
2922 
2923 	for (i = 0; i < ncopy; i++) {
2924 		cheight = cheight > copy_height ? copy_height : cheight;
2925 		size = (cheight * pitch) / 4;
2926 		/* emit reloc before writing cs so that cs is always in consistent state */
2927 		radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rsrc->resource, RADEON_USAGE_READ, 0);
2928 		radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rdst->resource, RADEON_USAGE_WRITE, 0);
2929 		radeon_emit(cs, DMA_PACKET(DMA_PACKET_COPY, 1, 0, size));
2930 		radeon_emit(cs, base >> 8);
2931 		radeon_emit(cs, (detile << 31) | (array_mode << 27) |
2932 				(lbpp << 24) | ((height - 1) << 10) |
2933 				pitch_tile_max);
2934 		radeon_emit(cs, (slice_tile_max << 12) | (z << 0));
2935 		radeon_emit(cs, (x << 3) | (y << 17));
2936 		radeon_emit(cs, addr & 0xfffffffc);
2937 		radeon_emit(cs, (addr >> 32UL) & 0xff);
2938 		copy_height -= cheight;
2939 		addr += cheight * pitch;
2940 		y += cheight;
2941 	}
2942 	return TRUE;
2943 }
2944 
r600_dma_copy(struct pipe_context * ctx,struct pipe_resource * dst,unsigned dst_level,unsigned dstx,unsigned dsty,unsigned dstz,struct pipe_resource * src,unsigned src_level,const struct pipe_box * src_box)2945 static void r600_dma_copy(struct pipe_context *ctx,
2946 			  struct pipe_resource *dst,
2947 			  unsigned dst_level,
2948 			  unsigned dstx, unsigned dsty, unsigned dstz,
2949 			  struct pipe_resource *src,
2950 			  unsigned src_level,
2951 			  const struct pipe_box *src_box)
2952 {
2953 	struct r600_context *rctx = (struct r600_context *)ctx;
2954 	struct r600_texture *rsrc = (struct r600_texture*)src;
2955 	struct r600_texture *rdst = (struct r600_texture*)dst;
2956 	unsigned dst_pitch, src_pitch, bpp, dst_mode, src_mode, copy_height;
2957 	unsigned src_w, dst_w;
2958 	unsigned src_x, src_y;
2959 	unsigned dst_x = dstx, dst_y = dsty, dst_z = dstz;
2960 
2961 	if (rctx->b.dma.cs == NULL) {
2962 		goto fallback;
2963 	}
2964 
2965 	if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
2966 		if (dst_x % 4 || src_box->x % 4 || src_box->width % 4)
2967 			goto fallback;
2968 
2969 		r600_dma_copy_buffer(rctx, dst, src, dst_x, src_box->x, src_box->width);
2970 		return;
2971 	}
2972 
2973 	if (src_box->depth > 1 ||
2974 	    !r600_prepare_for_dma_blit(&rctx->b, rdst, dst_level, dstx, dsty,
2975 					dstz, rsrc, src_level, src_box))
2976 		goto fallback;
2977 
2978 	src_x = util_format_get_nblocksx(src->format, src_box->x);
2979 	dst_x = util_format_get_nblocksx(src->format, dst_x);
2980 	src_y = util_format_get_nblocksy(src->format, src_box->y);
2981 	dst_y = util_format_get_nblocksy(src->format, dst_y);
2982 
2983 	bpp = rdst->surface.bpe;
2984 	dst_pitch = rdst->surface.u.legacy.level[dst_level].nblk_x * rdst->surface.bpe;
2985 	src_pitch = rsrc->surface.u.legacy.level[src_level].nblk_x * rsrc->surface.bpe;
2986 	src_w = u_minify(rsrc->resource.b.b.width0, src_level);
2987 	dst_w = u_minify(rdst->resource.b.b.width0, dst_level);
2988 	copy_height = src_box->height / rsrc->surface.blk_h;
2989 
2990 	dst_mode = rdst->surface.u.legacy.level[dst_level].mode;
2991 	src_mode = rsrc->surface.u.legacy.level[src_level].mode;
2992 
2993 	if (src_pitch != dst_pitch || src_box->x || dst_x || src_w != dst_w) {
2994 		/* strict requirement on r6xx/r7xx */
2995 		goto fallback;
2996 	}
2997 	/* lot of constraint on alignment this should capture them all */
2998 	if (src_pitch % 8 || src_box->y % 8 || dst_y % 8) {
2999 		goto fallback;
3000 	}
3001 
3002 	if (src_mode == dst_mode) {
3003 		uint64_t dst_offset, src_offset, size;
3004 
3005 		/* simple dma blit would do NOTE code here assume :
3006 		 *   src_box.x/y == 0
3007 		 *   dst_x/y == 0
3008 		 *   dst_pitch == src_pitch
3009 		 */
3010 		src_offset= rsrc->surface.u.legacy.level[src_level].offset;
3011 		src_offset += (uint64_t)rsrc->surface.u.legacy.level[src_level].slice_size_dw * 4 * src_box->z;
3012 		src_offset += src_y * src_pitch + src_x * bpp;
3013 		dst_offset = rdst->surface.u.legacy.level[dst_level].offset;
3014 		dst_offset += (uint64_t)rdst->surface.u.legacy.level[dst_level].slice_size_dw * 4 * dst_z;
3015 		dst_offset += dst_y * dst_pitch + dst_x * bpp;
3016 		size = src_box->height * src_pitch;
3017 		/* must be dw aligned */
3018 		if (dst_offset % 4 || src_offset % 4 || size % 4) {
3019 			goto fallback;
3020 		}
3021 		r600_dma_copy_buffer(rctx, dst, src, dst_offset, src_offset, size);
3022 	} else {
3023 		if (!r600_dma_copy_tile(rctx, dst, dst_level, dst_x, dst_y, dst_z,
3024 					src, src_level, src_x, src_y, src_box->z,
3025 					copy_height, dst_pitch, bpp)) {
3026 			goto fallback;
3027 		}
3028 	}
3029 	return;
3030 
3031 fallback:
3032 	r600_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
3033 				  src, src_level, src_box);
3034 }
3035 
r600_init_state_functions(struct r600_context * rctx)3036 void r600_init_state_functions(struct r600_context *rctx)
3037 {
3038 	unsigned id = 1;
3039 	unsigned i;
3040 	/* !!!
3041 	 *  To avoid GPU lockup registers must be emited in a specific order
3042 	 * (no kidding ...). The order below is important and have been
3043 	 * partialy infered from analyzing fglrx command stream.
3044 	 *
3045 	 * Don't reorder atom without carefully checking the effect (GPU lockup
3046 	 * or piglit regression).
3047 	 * !!!
3048 	 */
3049 
3050 	r600_init_atom(rctx, &rctx->framebuffer.atom, id++, r600_emit_framebuffer_state, 0);
3051 
3052 	/* shader const */
3053 	r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX].atom, id++, r600_emit_vs_constant_buffers, 0);
3054 	r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY].atom, id++, r600_emit_gs_constant_buffers, 0);
3055 	r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT].atom, id++, r600_emit_ps_constant_buffers, 0);
3056 
3057 	/* sampler must be emited before TA_CNTL_AUX otherwise DISABLE_CUBE_WRAP change
3058 	 * does not take effect (TA_CNTL_AUX emited by r600_emit_seamless_cube_map)
3059 	 */
3060 	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].states.atom, id++, r600_emit_vs_sampler_states, 0);
3061 	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].states.atom, id++, r600_emit_gs_sampler_states, 0);
3062 	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].states.atom, id++, r600_emit_ps_sampler_states, 0);
3063 	/* resource */
3064 	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views.atom, id++, r600_emit_vs_sampler_views, 0);
3065 	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views.atom, id++, r600_emit_gs_sampler_views, 0);
3066 	r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views.atom, id++, r600_emit_ps_sampler_views, 0);
3067 	r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, r600_emit_vertex_buffers, 0);
3068 
3069 	r600_init_atom(rctx, &rctx->vgt_state.atom, id++, r600_emit_vgt_state, 10);
3070 
3071 	r600_init_atom(rctx, &rctx->seamless_cube_map.atom, id++, r600_emit_seamless_cube_map, 3);
3072 	r600_init_atom(rctx, &rctx->sample_mask.atom, id++, r600_emit_sample_mask, 3);
3073 	rctx->sample_mask.sample_mask = ~0;
3074 
3075 	r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6);
3076 	r600_init_atom(rctx, &rctx->blend_color.atom, id++, r600_emit_blend_color, 6);
3077 	r600_init_atom(rctx, &rctx->blend_state.atom, id++, r600_emit_cso_state, 0);
3078 	r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, r600_emit_cb_misc_state, 7);
3079 	r600_init_atom(rctx, &rctx->clip_misc_state.atom, id++, r600_emit_clip_misc_state, 6);
3080 	r600_init_atom(rctx, &rctx->clip_state.atom, id++, r600_emit_clip_state, 26);
3081 	r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, r600_emit_db_misc_state, 7);
3082 	r600_init_atom(rctx, &rctx->db_state.atom, id++, r600_emit_db_state, 11);
3083 	r600_init_atom(rctx, &rctx->dsa_state.atom, id++, r600_emit_cso_state, 0);
3084 	r600_init_atom(rctx, &rctx->poly_offset_state.atom, id++, r600_emit_polygon_offset, 9);
3085 	r600_init_atom(rctx, &rctx->rasterizer_state.atom, id++, r600_emit_cso_state, 0);
3086 	r600_add_atom(rctx, &rctx->b.scissors.atom, id++);
3087 	r600_add_atom(rctx, &rctx->b.viewports.atom, id++);
3088 	r600_init_atom(rctx, &rctx->config_state.atom, id++, r600_emit_config_state, 3);
3089 	r600_init_atom(rctx, &rctx->stencil_ref.atom, id++, r600_emit_stencil_ref, 4);
3090 	r600_init_atom(rctx, &rctx->vertex_fetch_shader.atom, id++, r600_emit_vertex_fetch_shader, 5);
3091 	r600_add_atom(rctx, &rctx->b.render_cond_atom, id++);
3092 	r600_add_atom(rctx, &rctx->b.streamout.begin_atom, id++);
3093 	r600_add_atom(rctx, &rctx->b.streamout.enable_atom, id++);
3094 	for (i = 0; i < R600_NUM_HW_STAGES; i++)
3095 		r600_init_atom(rctx, &rctx->hw_shader_stages[i].atom, id++, r600_emit_shader, 0);
3096 	r600_init_atom(rctx, &rctx->shader_stages.atom, id++, r600_emit_shader_stages, 0);
3097 	r600_init_atom(rctx, &rctx->gs_rings.atom, id++, r600_emit_gs_rings, 0);
3098 
3099 	rctx->b.b.create_blend_state = r600_create_blend_state;
3100 	rctx->b.b.create_depth_stencil_alpha_state = r600_create_dsa_state;
3101 	rctx->b.b.create_rasterizer_state = r600_create_rs_state;
3102 	rctx->b.b.create_sampler_state = r600_create_sampler_state;
3103 	rctx->b.b.create_sampler_view = r600_create_sampler_view;
3104 	rctx->b.b.set_framebuffer_state = r600_set_framebuffer_state;
3105 	rctx->b.b.set_polygon_stipple = r600_set_polygon_stipple;
3106 	rctx->b.b.set_min_samples = r600_set_min_samples;
3107 	rctx->b.b.get_sample_position = r600_get_sample_position;
3108 	rctx->b.dma_copy = r600_dma_copy;
3109 }
3110 /* this function must be last */
3111