2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
24 /* r300_emit: Functions for emitting state. */
26 #include "util/u_format.h"
27 #include "util/u_math.h"
28 #include "util/u_simple_list.h"
30 #include "r300_context.h"
32 #include "r300_emit.h"
34 #include "r300_screen.h"
37 void r300_emit_blend_state(struct r300_context
* r300
,
38 unsigned size
, void* state
)
40 struct r300_blend_state
* blend
= (struct r300_blend_state
*)state
;
41 struct pipe_framebuffer_state
* fb
=
42 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
46 OUT_CS_REG(R300_RB3D_ROPCNTL
, blend
->rop
);
47 OUT_CS_REG_SEQ(R300_RB3D_CBLEND
, 3);
49 OUT_CS(blend
->blend_control
);
50 OUT_CS(blend
->alpha_blend_control
);
51 OUT_CS(blend
->color_channel_mask
);
56 /* XXX also disable fastfill here once it's supported */
58 OUT_CS_REG(R300_RB3D_DITHER_CTL
, blend
->dither
);
62 void r300_emit_blend_color_state(struct r300_context
* r300
,
63 unsigned size
, void* state
)
65 struct r300_blend_color_state
* bc
= (struct r300_blend_color_state
*)state
;
66 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
69 if (r300screen
->caps
->is_r500
) {
71 OUT_CS_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR
, 2);
72 OUT_CS(bc
->blend_color_red_alpha
);
73 OUT_CS(bc
->blend_color_green_blue
);
77 OUT_CS_REG(R300_RB3D_BLEND_COLOR
, bc
->blend_color
);
82 void r300_emit_clip_state(struct r300_context
* r300
,
83 unsigned size
, void* state
)
85 struct pipe_clip_state
* clip
= (struct pipe_clip_state
*)state
;
87 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
90 if (r300screen
->caps
->has_tcl
) {
91 BEGIN_CS(5 + (6 * 4));
92 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG
,
93 (r300screen
->caps
->is_r500
?
94 R500_PVS_UCP_START
: R300_PVS_UCP_START
));
95 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, 6 * 4);
96 for (i
= 0; i
< 6; i
++) {
97 OUT_CS_32F(clip
->ucp
[i
][0]);
98 OUT_CS_32F(clip
->ucp
[i
][1]);
99 OUT_CS_32F(clip
->ucp
[i
][2]);
100 OUT_CS_32F(clip
->ucp
[i
][3]);
102 OUT_CS_REG(R300_VAP_CLIP_CNTL
, ((1 << clip
->nr
) - 1) |
103 R300_PS_UCP_MODE_CLIP_AS_TRIFAN
);
107 OUT_CS_REG(R300_VAP_CLIP_CNTL
, R300_CLIP_DISABLE
);
113 void r300_emit_dsa_state(struct r300_context
* r300
, unsigned size
, void* state
)
115 struct r300_dsa_state
* dsa
= (struct r300_dsa_state
*)state
;
116 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
117 struct pipe_framebuffer_state
* fb
=
118 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
119 struct pipe_stencil_ref stencil_ref
= r300
->stencil_ref
;
122 BEGIN_CS(r300screen
->caps
->is_r500
? 8 : 6);
123 OUT_CS_REG(R300_FG_ALPHA_FUNC
, dsa
->alpha_function
);
124 OUT_CS_REG_SEQ(R300_ZB_CNTL
, 3);
127 OUT_CS(dsa
->z_buffer_control
);
128 OUT_CS(dsa
->z_stencil_control
);
134 OUT_CS(dsa
->stencil_ref_mask
| stencil_ref
.ref_value
[0]);
136 if (r300screen
->caps
->is_r500
) {
137 OUT_CS_REG(R500_ZB_STENCILREFMASK_BF
, dsa
->stencil_ref_bf
| stencil_ref
.ref_value
[1]);
142 static const float * get_shader_constant(
143 struct r300_context
* r300
,
144 struct rc_constant
* constant
,
145 struct r300_constant_buffer
* externals
)
147 struct r300_viewport_state
* viewport
=
148 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
149 static float vec
[4] = { 0.0, 0.0, 0.0, 1.0 };
150 struct pipe_texture
*tex
;
152 switch(constant
->Type
) {
153 case RC_CONSTANT_EXTERNAL
:
154 return externals
->constants
[constant
->u
.External
];
156 case RC_CONSTANT_IMMEDIATE
:
157 return constant
->u
.Immediate
;
159 case RC_CONSTANT_STATE
:
160 switch (constant
->u
.State
[0]) {
161 /* Factor for converting rectangle coords to
162 * normalized coords. Should only show up on non-r500. */
163 case RC_STATE_R300_TEXRECT_FACTOR
:
164 tex
= &r300
->textures
[constant
->u
.State
[1]]->tex
;
165 vec
[0] = 1.0 / tex
->width0
;
166 vec
[1] = 1.0 / tex
->height0
;
169 /* Texture compare-fail value. Shouldn't ever show up, but if
170 * it does, we'll be ready. */
171 case RC_STATE_SHADOW_AMBIENT
:
175 case RC_STATE_R300_VIEWPORT_SCALE
:
176 if (r300
->tcl_bypass
) {
181 vec
[0] = viewport
->xscale
;
182 vec
[1] = viewport
->yscale
;
183 vec
[2] = viewport
->zscale
;
187 case RC_STATE_R300_VIEWPORT_OFFSET
:
188 if (!r300
->tcl_bypass
) {
189 vec
[0] = viewport
->xoffset
;
190 vec
[1] = viewport
->yoffset
;
191 vec
[2] = viewport
->zoffset
;
196 debug_printf("r300: Implementation error: "
197 "Unknown RC_CONSTANT type %d\n", constant
->u
.State
[0]);
202 debug_printf("r300: Implementation error: "
203 "Unhandled constant type %d\n", constant
->Type
);
206 /* This should either be (0, 0, 0, 1), which should be a relatively safe
207 * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
212 /* Convert a normal single-precision float into the 7.16 format
213 * used by the R300 fragment shader.
215 static uint32_t pack_float24(float f
)
223 uint32_t float24
= 0;
230 mantissa
= frexpf(f
, &exponent
);
234 float24
|= (1 << 23);
235 mantissa
= mantissa
* -1.0;
237 /* Handle exponent, bias of 63 */
239 float24
|= (exponent
<< 16);
240 /* Kill 7 LSB of mantissa */
241 float24
|= (u
.u
& 0x7FFFFF) >> 7;
246 void r300_emit_fragment_program_code(struct r300_context
* r300
,
247 struct rX00_fragment_program_code
* generic_code
)
249 struct r300_fragment_program_code
* code
= &generic_code
->code
.r300
;
254 code
->alu
.length
* 4 +
255 (code
->tex
.length
? (1 + code
->tex
.length
) : 0));
257 OUT_CS_REG(R300_US_CONFIG
, code
->config
);
258 OUT_CS_REG(R300_US_PIXSIZE
, code
->pixsize
);
259 OUT_CS_REG(R300_US_CODE_OFFSET
, code
->code_offset
);
261 OUT_CS_REG_SEQ(R300_US_CODE_ADDR_0
, 4);
262 for(i
= 0; i
< 4; ++i
)
263 OUT_CS(code
->code_addr
[i
]);
265 OUT_CS_REG_SEQ(R300_US_ALU_RGB_INST_0
, code
->alu
.length
);
266 for (i
= 0; i
< code
->alu
.length
; i
++)
267 OUT_CS(code
->alu
.inst
[i
].rgb_inst
);
269 OUT_CS_REG_SEQ(R300_US_ALU_RGB_ADDR_0
, code
->alu
.length
);
270 for (i
= 0; i
< code
->alu
.length
; i
++)
271 OUT_CS(code
->alu
.inst
[i
].rgb_addr
);
273 OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_INST_0
, code
->alu
.length
);
274 for (i
= 0; i
< code
->alu
.length
; i
++)
275 OUT_CS(code
->alu
.inst
[i
].alpha_inst
);
277 OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0
, code
->alu
.length
);
278 for (i
= 0; i
< code
->alu
.length
; i
++)
279 OUT_CS(code
->alu
.inst
[i
].alpha_addr
);
281 if (code
->tex
.length
) {
282 OUT_CS_REG_SEQ(R300_US_TEX_INST_0
, code
->tex
.length
);
283 for(i
= 0; i
< code
->tex
.length
; ++i
)
284 OUT_CS(code
->tex
.inst
[i
]);
290 void r300_emit_fs_constant_buffer(struct r300_context
* r300
,
291 struct rc_constant_list
* constants
)
296 if (constants
->Count
== 0)
299 BEGIN_CS(constants
->Count
* 4 + 1);
300 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X
, constants
->Count
* 4);
301 for(i
= 0; i
< constants
->Count
; ++i
) {
302 const float * data
= get_shader_constant(r300
,
303 &constants
->Constants
[i
],
304 &r300
->shader_constants
[PIPE_SHADER_FRAGMENT
]);
305 OUT_CS(pack_float24(data
[0]));
306 OUT_CS(pack_float24(data
[1]));
307 OUT_CS(pack_float24(data
[2]));
308 OUT_CS(pack_float24(data
[3]));
313 static void r300_emit_fragment_depth_config(struct r300_context
* r300
,
314 struct r300_fragment_shader
* fs
)
319 if (r300_fragment_shader_writes_depth(fs
)) {
320 OUT_CS_REG(R300_FG_DEPTH_SRC
, R300_FG_DEPTH_SRC_SHADER
);
321 OUT_CS_REG(R300_US_W_FMT
, R300_W_FMT_W24
| R300_W_SRC_US
);
323 OUT_CS_REG(R300_FG_DEPTH_SRC
, R300_FG_DEPTH_SRC_SCAN
);
324 OUT_CS_REG(R300_US_W_FMT
, R300_W_FMT_W0
| R300_W_SRC_US
);
329 void r500_emit_fragment_program_code(struct r300_context
* r300
,
330 struct rX00_fragment_program_code
* generic_code
)
332 struct r500_fragment_program_code
* code
= &generic_code
->code
.r500
;
337 ((code
->inst_end
+ 1) * 6));
338 OUT_CS_REG(R500_US_CONFIG
, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO
);
339 OUT_CS_REG(R500_US_PIXSIZE
, code
->max_temp_idx
);
340 OUT_CS_REG(R500_US_CODE_RANGE
,
341 R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code
->inst_end
));
342 OUT_CS_REG(R500_US_CODE_OFFSET
, 0);
343 OUT_CS_REG(R500_US_CODE_ADDR
,
344 R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code
->inst_end
));
346 OUT_CS_REG(R500_GA_US_VECTOR_INDEX
, R500_GA_US_VECTOR_INDEX_TYPE_INSTR
);
347 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA
, (code
->inst_end
+ 1) * 6);
348 for (i
= 0; i
<= code
->inst_end
; i
++) {
349 OUT_CS(code
->inst
[i
].inst0
);
350 OUT_CS(code
->inst
[i
].inst1
);
351 OUT_CS(code
->inst
[i
].inst2
);
352 OUT_CS(code
->inst
[i
].inst3
);
353 OUT_CS(code
->inst
[i
].inst4
);
354 OUT_CS(code
->inst
[i
].inst5
);
360 void r500_emit_fs_constant_buffer(struct r300_context
* r300
,
361 struct rc_constant_list
* constants
)
366 if (constants
->Count
== 0)
369 BEGIN_CS(constants
->Count
* 4 + 3);
370 OUT_CS_REG(R500_GA_US_VECTOR_INDEX
, R500_GA_US_VECTOR_INDEX_TYPE_CONST
);
371 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA
, constants
->Count
* 4);
372 for (i
= 0; i
< constants
->Count
; i
++) {
373 const float * data
= get_shader_constant(r300
,
374 &constants
->Constants
[i
],
375 &r300
->shader_constants
[PIPE_SHADER_FRAGMENT
]);
384 void r300_emit_fb_state(struct r300_context
* r300
, unsigned size
, void* state
)
386 struct pipe_framebuffer_state
* fb
= (struct pipe_framebuffer_state
*)state
;
387 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
388 struct r300_texture
* tex
;
389 struct pipe_surface
* surf
;
393 BEGIN_CS((10 * fb
->nr_cbufs
) + (2 * (4 - fb
->nr_cbufs
)) +
394 (fb
->zsbuf
? 10 : 0) + 8);
396 /* Flush and free renderbuffer caches. */
397 OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT
,
398 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS
|
399 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D
);
400 OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT
,
401 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE
|
402 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE
);
404 /* Set the number of colorbuffers. */
405 if (fb
->nr_cbufs
> 1) {
406 if (r300screen
->caps
->is_r500
) {
407 OUT_CS_REG(R300_RB3D_CCTL
,
408 R300_RB3D_CCTL_NUM_MULTIWRITES(fb
->nr_cbufs
) |
409 R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE
);
411 OUT_CS_REG(R300_RB3D_CCTL
,
412 R300_RB3D_CCTL_NUM_MULTIWRITES(fb
->nr_cbufs
));
415 OUT_CS_REG(R300_RB3D_CCTL
, 0x0);
418 /* Set up colorbuffers. */
419 for (i
= 0; i
< fb
->nr_cbufs
; i
++) {
421 tex
= (struct r300_texture
*)surf
->texture
;
422 assert(tex
&& tex
->buffer
&& "cbuf is marked, but NULL!");
424 OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0
+ (4 * i
), 1);
425 OUT_CS_RELOC(tex
->buffer
, surf
->offset
, 0, RADEON_GEM_DOMAIN_VRAM
, 0);
427 OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0
+ (4 * i
), 1);
428 OUT_CS_RELOC(tex
->buffer
, tex
->fb_state
.colorpitch
[surf
->level
],
429 0, RADEON_GEM_DOMAIN_VRAM
, 0);
431 OUT_CS_REG(R300_US_OUT_FMT_0
+ (4 * i
), tex
->fb_state
.us_out_fmt
);
434 OUT_CS_REG(R300_US_OUT_FMT_0
+ (4 * i
), R300_US_OUT_FMT_UNUSED
);
437 /* Set up a zbuffer. */
440 tex
= (struct r300_texture
*)surf
->texture
;
441 assert(tex
&& tex
->buffer
&& "zsbuf is marked, but NULL!");
443 OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET
, 1);
444 OUT_CS_RELOC(tex
->buffer
, surf
->offset
, 0, RADEON_GEM_DOMAIN_VRAM
, 0);
446 OUT_CS_REG(R300_ZB_FORMAT
, tex
->fb_state
.zb_format
);
448 OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH
, 1);
449 OUT_CS_RELOC(tex
->buffer
, tex
->fb_state
.depthpitch
[surf
->level
],
450 0, RADEON_GEM_DOMAIN_VRAM
, 0);
453 OUT_CS_REG(R300_GA_POINT_MINMAX
,
454 (MAX2(fb
->width
, fb
->height
) * 6) << R300_GA_POINT_MINMAX_MAX_SHIFT
);
458 static void r300_emit_query_start(struct r300_context
*r300
)
460 struct r300_capabilities
*caps
= r300_screen(r300
->context
.screen
)->caps
;
461 struct r300_query
*query
= r300
->query_current
;
468 if (caps
->family
== CHIP_FAMILY_RV530
) {
469 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL
);
471 OUT_CS_REG(R300_SU_REG_DEST
, R300_RASTER_PIPE_SELECT_ALL
);
473 OUT_CS_REG(R300_ZB_ZPASS_DATA
, 0);
475 query
->begin_emitted
= TRUE
;
479 static void r300_emit_query_finish(struct r300_context
*r300
,
480 struct r300_query
*query
)
482 struct r300_capabilities
* caps
= r300_screen(r300
->context
.screen
)->caps
;
485 assert(caps
->num_frag_pipes
);
487 BEGIN_CS(6 * caps
->num_frag_pipes
+ 2);
488 /* I'm not so sure I like this switch, but it's hard to be elegant
489 * when there's so many special cases...
491 * So here's the basic idea. For each pipe, enable writes to it only,
492 * then put out the relocation for ZPASS_ADDR, taking into account a
493 * 4-byte offset for each pipe. RV380 and older are special; they have
494 * only two pipes, and the second pipe's enable is on bit 3, not bit 1,
495 * so there's a chipset cap for that. */
496 switch (caps
->num_frag_pipes
) {
499 OUT_CS_REG(R300_SU_REG_DEST
, 1 << 3);
500 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
501 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 3),
502 0, RADEON_GEM_DOMAIN_GTT
, 0);
505 OUT_CS_REG(R300_SU_REG_DEST
, 1 << 2);
506 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
507 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 2),
508 0, RADEON_GEM_DOMAIN_GTT
, 0);
511 /* As mentioned above, accomodate RV380 and older. */
512 OUT_CS_REG(R300_SU_REG_DEST
,
513 1 << (caps
->high_second_pipe
? 3 : 1));
514 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
515 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 1),
516 0, RADEON_GEM_DOMAIN_GTT
, 0);
519 OUT_CS_REG(R300_SU_REG_DEST
, 1 << 0);
520 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
521 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 0),
522 0, RADEON_GEM_DOMAIN_GTT
, 0);
525 debug_printf("r300: Implementation error: Chipset reports %d"
526 " pixel pipes!\n", caps
->num_frag_pipes
);
530 /* And, finally, reset it to normal... */
531 OUT_CS_REG(R300_SU_REG_DEST
, 0xF);
535 static void rv530_emit_query_single(struct r300_context
*r300
,
536 struct r300_query
*query
)
541 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_0
);
542 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
543 OUT_CS_RELOC(r300
->oqbo
, query
->offset
, 0, RADEON_GEM_DOMAIN_GTT
, 0);
544 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL
);
548 static void rv530_emit_query_double(struct r300_context
*r300
,
549 struct r300_query
*query
)
554 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_0
);
555 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
556 OUT_CS_RELOC(r300
->oqbo
, query
->offset
, 0, RADEON_GEM_DOMAIN_GTT
, 0);
557 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_1
);
558 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
559 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT
, 0);
560 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL
);
564 void r300_emit_query_end(struct r300_context
* r300
)
566 struct r300_capabilities
*caps
= r300_screen(r300
->context
.screen
)->caps
;
567 struct r300_query
*query
= r300
->query_current
;
572 if (query
->begin_emitted
== FALSE
)
575 if (caps
->family
== CHIP_FAMILY_RV530
) {
576 if (caps
->num_z_pipes
== 2)
577 rv530_emit_query_double(r300
, query
);
579 rv530_emit_query_single(r300
, query
);
581 r300_emit_query_finish(r300
, query
);
584 void r300_emit_rs_state(struct r300_context
* r300
, unsigned size
, void* state
)
586 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
590 BEGIN_CS(17 + (rs
->polygon_offset_enable
? 5 : 0));
591 OUT_CS_REG(R300_VAP_CNTL_STATUS
, rs
->vap_control_status
);
593 OUT_CS_REG(R300_GB_AA_CONFIG
, rs
->antialiasing_config
);
595 OUT_CS_REG(R300_GA_POINT_SIZE
, rs
->point_size
);
596 OUT_CS_REG(R300_GA_LINE_CNTL
, rs
->line_control
);
598 if (rs
->polygon_offset_enable
) {
599 scale
= rs
->depth_scale
* 12;
600 offset
= rs
->depth_offset
;
602 switch (r300
->zbuffer_bpp
) {
611 OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE
, 4);
618 OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE
, 2);
619 OUT_CS(rs
->polygon_offset_enable
);
620 OUT_CS(rs
->cull_mode
);
621 OUT_CS_REG(R300_GA_LINE_STIPPLE_CONFIG
, rs
->line_stipple_config
);
622 OUT_CS_REG(R300_GA_LINE_STIPPLE_VALUE
, rs
->line_stipple_value
);
623 OUT_CS_REG(R300_GA_POLY_MODE
, rs
->polygon_mode
);
627 void r300_emit_rs_block_state(struct r300_context
* r300
,
628 unsigned size
, void* state
)
630 struct r300_rs_block
* rs
= (struct r300_rs_block
*)state
;
632 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
633 /* It's the same for both INST and IP tables */
634 unsigned count
= (rs
->inst_count
& R300_RS_INST_COUNT_MASK
) + 1;
637 DBG(r300
, DBG_DRAW
, "r300: RS emit:\n");
639 BEGIN_CS(5 + count
*2);
640 if (r300screen
->caps
->is_r500
) {
641 OUT_CS_REG_SEQ(R500_RS_IP_0
, count
);
643 OUT_CS_REG_SEQ(R300_RS_IP_0
, count
);
645 for (i
= 0; i
< count
; i
++) {
647 DBG(r300
, DBG_DRAW
, " : ip %d: 0x%08x\n", i
, rs
->ip
[i
]);
650 OUT_CS_REG_SEQ(R300_RS_COUNT
, 2);
652 OUT_CS(rs
->inst_count
);
654 if (r300screen
->caps
->is_r500
) {
655 OUT_CS_REG_SEQ(R500_RS_INST_0
, count
);
657 OUT_CS_REG_SEQ(R300_RS_INST_0
, count
);
659 for (i
= 0; i
< count
; i
++) {
661 DBG(r300
, DBG_DRAW
, " : inst %d: 0x%08x\n", i
, rs
->inst
[i
]);
664 DBG(r300
, DBG_DRAW
, " : count: 0x%08x inst_count: 0x%08x\n",
665 rs
->count
, rs
->inst_count
);
670 void r300_emit_scissor_state(struct r300_context
* r300
,
671 unsigned size
, void* state
)
673 unsigned minx
, miny
, maxx
, maxy
;
674 uint32_t top_left
, bottom_right
;
675 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
676 struct pipe_scissor_state
* scissor
= (struct pipe_scissor_state
*)state
;
677 struct pipe_framebuffer_state
* fb
=
678 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
685 if (((struct r300_rs_state
*)r300
->rs_state
.state
)->rs
.scissor
) {
686 minx
= MAX2(minx
, scissor
->minx
);
687 miny
= MAX2(miny
, scissor
->miny
);
688 maxx
= MIN2(maxx
, scissor
->maxx
);
689 maxy
= MIN2(maxy
, scissor
->maxy
);
692 /* Special case for zero-area scissor.
694 * We can't allow the variables maxx and maxy to be zero because they are
695 * subtracted from later in the code, which would cause emitting ~0 and
696 * making the kernel checker angry.
698 * Let's consider we change maxx and maxy to 1, which is effectively
699 * a one-pixel area. We must then change minx and miny to a number which is
700 * greater than 1 to get the zero area back. */
701 if (!maxx
|| !maxy
) {
708 if (r300screen
->caps
->is_r500
) {
710 (minx
<< R300_SCISSORS_X_SHIFT
) |
711 (miny
<< R300_SCISSORS_Y_SHIFT
);
713 ((maxx
- 1) << R300_SCISSORS_X_SHIFT
) |
714 ((maxy
- 1) << R300_SCISSORS_Y_SHIFT
);
716 /* Offset of 1440 in non-R500 chipsets. */
718 ((minx
+ 1440) << R300_SCISSORS_X_SHIFT
) |
719 ((miny
+ 1440) << R300_SCISSORS_Y_SHIFT
);
721 (((maxx
- 1) + 1440) << R300_SCISSORS_X_SHIFT
) |
722 (((maxy
- 1) + 1440) << R300_SCISSORS_Y_SHIFT
);
726 OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL
, 2);
728 OUT_CS(bottom_right
);
732 void r300_emit_texture(struct r300_context
* r300
,
733 struct r300_sampler_state
* sampler
,
734 struct r300_texture
* tex
,
737 uint32_t filter0
= sampler
->filter0
;
738 uint32_t format0
= tex
->state
.format0
;
739 unsigned min_level
, max_level
;
742 /* to emulate 1D textures through 2D ones correctly */
743 if (tex
->tex
.target
== PIPE_TEXTURE_1D
) {
744 filter0
&= ~R300_TX_WRAP_T_MASK
;
745 filter0
|= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE
);
749 /* NPOT textures don't support mip filter, unfortunately.
750 * This prevents incorrect rendering. */
751 filter0
&= ~R300_TX_MIN_FILTER_MIP_MASK
;
753 /* determine min/max levels */
754 /* the MAX_MIP level is the largest (finest) one */
755 max_level
= MIN2(sampler
->max_lod
, tex
->tex
.last_level
);
756 min_level
= MIN2(sampler
->min_lod
, max_level
);
757 format0
|= R300_TX_NUM_LEVELS(max_level
);
758 filter0
|= R300_TX_MAX_MIP_LEVEL(min_level
);
762 OUT_CS_REG(R300_TX_FILTER0_0
+ (offset
* 4), filter0
|
764 OUT_CS_REG(R300_TX_FILTER1_0
+ (offset
* 4), sampler
->filter1
);
765 OUT_CS_REG(R300_TX_BORDER_COLOR_0
+ (offset
* 4), sampler
->border_color
);
767 OUT_CS_REG(R300_TX_FORMAT0_0
+ (offset
* 4), format0
);
768 OUT_CS_REG(R300_TX_FORMAT1_0
+ (offset
* 4), tex
->state
.format1
);
769 OUT_CS_REG(R300_TX_FORMAT2_0
+ (offset
* 4), tex
->state
.format2
);
770 OUT_CS_REG_SEQ(R300_TX_OFFSET_0
+ (offset
* 4), 1);
771 OUT_CS_RELOC(tex
->buffer
,
772 R300_TXO_MACRO_TILE(tex
->macrotile
) |
773 R300_TXO_MICRO_TILE(tex
->microtile
),
774 RADEON_GEM_DOMAIN_GTT
| RADEON_GEM_DOMAIN_VRAM
, 0, 0);
778 void r300_emit_aos(struct r300_context
* r300
, unsigned offset
)
780 struct pipe_vertex_buffer
*vb1
, *vb2
, *vbuf
= r300
->vertex_buffer
;
781 struct pipe_vertex_element
*velem
= r300
->vertex_element
;
783 unsigned size1
, size2
, aos_count
= r300
->vertex_element_count
;
784 unsigned packet_size
= (aos_count
* 3 + 1) / 2;
787 BEGIN_CS(2 + packet_size
+ aos_count
* 2);
788 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR
, packet_size
);
791 for (i
= 0; i
< aos_count
- 1; i
+= 2) {
792 vb1
= &vbuf
[velem
[i
].vertex_buffer_index
];
793 vb2
= &vbuf
[velem
[i
+1].vertex_buffer_index
];
794 size1
= util_format_get_blocksize(velem
[i
].src_format
);
795 size2
= util_format_get_blocksize(velem
[i
+1].src_format
);
797 OUT_CS(R300_VBPNTR_SIZE0(size1
) | R300_VBPNTR_STRIDE0(vb1
->stride
) |
798 R300_VBPNTR_SIZE1(size2
) | R300_VBPNTR_STRIDE1(vb2
->stride
));
799 OUT_CS(vb1
->buffer_offset
+ velem
[i
].src_offset
+ offset
* vb1
->stride
);
800 OUT_CS(vb2
->buffer_offset
+ velem
[i
+1].src_offset
+ offset
* vb2
->stride
);
804 vb1
= &vbuf
[velem
[i
].vertex_buffer_index
];
805 size1
= util_format_get_blocksize(velem
[i
].src_format
);
807 OUT_CS(R300_VBPNTR_SIZE0(size1
) | R300_VBPNTR_STRIDE0(vb1
->stride
));
808 OUT_CS(vb1
->buffer_offset
+ velem
[i
].src_offset
+ offset
* vb1
->stride
);
811 for (i
= 0; i
< aos_count
; i
++) {
812 OUT_CS_RELOC_NO_OFFSET(vbuf
[velem
[i
].vertex_buffer_index
].buffer
,
813 RADEON_GEM_DOMAIN_GTT
, 0, 0);
818 void r300_emit_vertex_format_state(struct r300_context
* r300
,
819 unsigned size
, void* state
)
821 struct r300_vertex_info
* vertex_info
= (struct r300_vertex_info
*)state
;
825 DBG(r300
, DBG_DRAW
, "r300: VAP/PSC emit:\n");
828 OUT_CS_REG(R300_VAP_VTX_SIZE
, vertex_info
->vinfo
.size
);
830 OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL
, 2);
831 OUT_CS(vertex_info
->vinfo
.hwfmt
[0]);
832 OUT_CS(vertex_info
->vinfo
.hwfmt
[1]);
833 OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0
, 2);
834 OUT_CS(vertex_info
->vinfo
.hwfmt
[2]);
835 OUT_CS(vertex_info
->vinfo
.hwfmt
[3]);
836 for (i
= 0; i
< 4; i
++) {
837 DBG(r300
, DBG_DRAW
, " : hwfmt%d: 0x%08x\n", i
,
838 vertex_info
->vinfo
.hwfmt
[i
]);
841 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0
, 8);
842 for (i
= 0; i
< 8; i
++) {
843 OUT_CS(vertex_info
->vap_prog_stream_cntl
[i
]);
844 DBG(r300
, DBG_DRAW
, " : prog_stream_cntl%d: 0x%08x\n", i
,
845 vertex_info
->vap_prog_stream_cntl
[i
]);
847 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0
, 8);
848 for (i
= 0; i
< 8; i
++) {
849 OUT_CS(vertex_info
->vap_prog_stream_cntl_ext
[i
]);
850 DBG(r300
, DBG_DRAW
, " : prog_stream_cntl_ext%d: 0x%08x\n", i
,
851 vertex_info
->vap_prog_stream_cntl_ext
[i
]);
856 void r300_emit_pvs_flush(struct r300_context
* r300
, unsigned size
, void* state
)
861 OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG
, 0x0);
865 void r300_emit_vs_state(struct r300_context
* r300
, unsigned size
, void* state
)
867 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)state
;
868 struct r300_vertex_program_code
* code
= &vs
->code
;
869 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
870 unsigned instruction_count
= code
->length
/ 4;
873 unsigned vtx_mem_size
= r300screen
->caps
->is_r500
? 128 : 72;
874 unsigned input_count
= MAX2(util_bitcount(code
->InputsRead
), 1);
875 unsigned output_count
= MAX2(util_bitcount(code
->OutputsWritten
), 1);
876 unsigned temp_count
= MAX2(code
->num_temporaries
, 1);
878 unsigned pvs_num_slots
= MIN3(vtx_mem_size
/ input_count
,
879 vtx_mem_size
/ output_count
, 10);
880 unsigned pvs_num_controllers
= MIN2(vtx_mem_size
/ temp_count
, 6);
884 if (!r300screen
->caps
->has_tcl
) {
885 debug_printf("r300: Implementation error: emit_vertex_shader called,"
886 " but has_tcl is FALSE!\n");
890 BEGIN_CS(9 + code
->length
);
891 /* R300_VAP_PVS_CODE_CNTL_0
892 * R300_VAP_PVS_CONST_CNTL
893 * R300_VAP_PVS_CODE_CNTL_1
894 * See the r5xx docs for instructions on how to use these. */
895 OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0
, 3);
896 OUT_CS(R300_PVS_FIRST_INST(0) |
897 R300_PVS_XYZW_VALID_INST(instruction_count
- 1) |
898 R300_PVS_LAST_INST(instruction_count
- 1));
899 OUT_CS(R300_PVS_MAX_CONST_ADDR(code
->constants
.Count
- 1));
900 OUT_CS(instruction_count
- 1);
902 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG
, 0);
903 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, code
->length
);
904 for (i
= 0; i
< code
->length
; i
++) {
905 OUT_CS(code
->body
.d
[i
]);
908 OUT_CS_REG(R300_VAP_CNTL
, R300_PVS_NUM_SLOTS(pvs_num_slots
) |
909 R300_PVS_NUM_CNTLRS(pvs_num_controllers
) |
910 R300_PVS_NUM_FPUS(r300screen
->caps
->num_vert_fpus
) |
911 R300_PVS_VF_MAX_VTX_NUM(12) |
912 (r300screen
->caps
->is_r500
? R500_TCL_STATE_OPTIMIZATION
: 0));
916 void r300_emit_vs_constant_buffer(struct r300_context
* r300
,
917 struct rc_constant_list
* constants
)
920 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
923 if (!r300screen
->caps
->has_tcl
) {
924 debug_printf("r300: Implementation error: emit_vertex_shader called,"
925 " but has_tcl is FALSE!\n");
929 if (constants
->Count
== 0)
932 BEGIN_CS(constants
->Count
* 4 + 3);
933 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG
,
934 (r300screen
->caps
->is_r500
?
935 R500_PVS_CONST_START
: R300_PVS_CONST_START
));
936 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, constants
->Count
* 4);
937 for (i
= 0; i
< constants
->Count
; i
++) {
938 const float * data
= get_shader_constant(r300
,
939 &constants
->Constants
[i
],
940 &r300
->shader_constants
[PIPE_SHADER_VERTEX
]);
949 void r300_emit_viewport_state(struct r300_context
* r300
,
950 unsigned size
, void* state
)
952 struct r300_viewport_state
* viewport
= (struct r300_viewport_state
*)state
;
955 if (r300
->tcl_bypass
) {
957 OUT_CS_REG(R300_VAP_VTE_CNTL
, 0);
961 OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE
, 6);
962 OUT_CS_32F(viewport
->xscale
);
963 OUT_CS_32F(viewport
->xoffset
);
964 OUT_CS_32F(viewport
->yscale
);
965 OUT_CS_32F(viewport
->yoffset
);
966 OUT_CS_32F(viewport
->zscale
);
967 OUT_CS_32F(viewport
->zoffset
);
968 OUT_CS_REG(R300_VAP_VTE_CNTL
, viewport
->vte_control
);
973 void r300_emit_texture_count(struct r300_context
* r300
)
975 uint32_t tx_enable
= 0;
979 /* Notice that texture_count and sampler_count are just sizes
980 * of the respective arrays. We still have to check for the individual
982 for (i
= 0; i
< MIN2(r300
->sampler_count
, r300
->texture_count
); i
++) {
983 if (r300
->textures
[i
]) {
989 OUT_CS_REG(R300_TX_ENABLE
, tx_enable
);
994 void r300_emit_ztop_state(struct r300_context
* r300
,
995 unsigned size
, void* state
)
997 struct r300_ztop_state
* ztop
= (struct r300_ztop_state
*)state
;
1001 OUT_CS_REG(R300_ZB_ZTOP
, ztop
->z_buffer_top
);
1005 void r300_flush_textures(struct r300_context
* r300
)
1010 OUT_CS_REG(R300_TX_INVALTAGS
, 0);
1014 void r300_emit_buffer_validate(struct r300_context
*r300
)
1016 struct pipe_framebuffer_state
* fb
=
1017 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
1018 struct r300_texture
* tex
;
1020 boolean invalid
= FALSE
;
1022 /* Clean out BOs. */
1023 r300
->winsys
->reset_bos(r300
->winsys
);
1026 /* Color buffers... */
1027 for (i
= 0; i
< fb
->nr_cbufs
; i
++) {
1028 tex
= (struct r300_texture
*)fb
->cbufs
[i
]->texture
;
1029 assert(tex
&& tex
->buffer
&& "cbuf is marked, but NULL!");
1030 if (!r300
->winsys
->add_buffer(r300
->winsys
, tex
->buffer
,
1031 0, RADEON_GEM_DOMAIN_VRAM
)) {
1032 r300
->context
.flush(&r300
->context
, 0, NULL
);
1036 /* ...depth buffer... */
1038 tex
= (struct r300_texture
*)fb
->zsbuf
->texture
;
1039 assert(tex
&& tex
->buffer
&& "zsbuf is marked, but NULL!");
1040 if (!r300
->winsys
->add_buffer(r300
->winsys
, tex
->buffer
,
1041 0, RADEON_GEM_DOMAIN_VRAM
)) {
1042 r300
->context
.flush(&r300
->context
, 0, NULL
);
1046 /* ...textures... */
1047 for (i
= 0; i
< r300
->texture_count
; i
++) {
1048 tex
= r300
->textures
[i
];
1051 if (!r300
->winsys
->add_buffer(r300
->winsys
, tex
->buffer
,
1052 RADEON_GEM_DOMAIN_GTT
| RADEON_GEM_DOMAIN_VRAM
, 0)) {
1053 r300
->context
.flush(&r300
->context
, 0, NULL
);
1057 /* ...occlusion query buffer... */
1058 if (r300
->dirty_state
& R300_NEW_QUERY
) {
1059 if (!r300
->winsys
->add_buffer(r300
->winsys
, r300
->oqbo
,
1060 0, RADEON_GEM_DOMAIN_GTT
)) {
1061 r300
->context
.flush(&r300
->context
, 0, NULL
);
1065 /* ...and vertex buffer. */
1067 if (!r300
->winsys
->add_buffer(r300
->winsys
, r300
->vbo
,
1068 RADEON_GEM_DOMAIN_GTT
, 0)) {
1069 r300
->context
.flush(&r300
->context
, 0, NULL
);
1073 /* debug_printf("No VBO while emitting dirty state!\n"); */
1075 if (!r300
->winsys
->validate(r300
->winsys
)) {
1076 r300
->context
.flush(&r300
->context
, 0, NULL
);
1079 debug_printf("r300: Stuck in validation loop, gonna quit now.");
1087 unsigned r300_get_num_dirty_dwords(struct r300_context
*r300
)
1089 struct r300_atom
* atom
;
1090 unsigned dwords
= 0;
1092 foreach(atom
, &r300
->atom_list
) {
1093 if (atom
->dirty
|| atom
->always_dirty
) {
1094 dwords
+= atom
->size
;
1098 /* XXX This is the compensation for the non-atomized states. */
1104 /* Emit all dirty state. */
1105 void r300_emit_dirty_state(struct r300_context
* r300
)
1107 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
1108 struct r300_atom
* atom
;
1112 if (r300
->dirty_state
& R300_NEW_QUERY
) {
1113 r300_emit_query_start(r300
);
1114 r300
->dirty_state
&= ~R300_NEW_QUERY
;
1117 foreach(atom
, &r300
->atom_list
) {
1118 if (atom
->dirty
|| atom
->always_dirty
) {
1119 atom
->emit(r300
, atom
->size
, atom
->state
);
1120 atom
->dirty
= FALSE
;
1124 if (r300
->dirty_state
& R300_NEW_FRAGMENT_SHADER
) {
1125 r300_emit_fragment_depth_config(r300
, r300
->fs
);
1126 if (r300screen
->caps
->is_r500
) {
1127 r500_emit_fragment_program_code(r300
, &r300
->fs
->shader
->code
);
1129 r300_emit_fragment_program_code(r300
, &r300
->fs
->shader
->code
);
1131 r300
->dirty_state
&= ~R300_NEW_FRAGMENT_SHADER
;
1134 if (r300
->dirty_state
& R300_NEW_FRAGMENT_SHADER_CONSTANTS
) {
1135 if (r300screen
->caps
->is_r500
) {
1136 r500_emit_fs_constant_buffer(r300
,
1137 &r300
->fs
->shader
->code
.constants
);
1139 r300_emit_fs_constant_buffer(r300
,
1140 &r300
->fs
->shader
->code
.constants
);
1142 r300
->dirty_state
&= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1145 /* Samplers and textures are tracked separately but emitted together. */
1146 if (r300
->dirty_state
&
1147 (R300_ANY_NEW_SAMPLERS
| R300_ANY_NEW_TEXTURES
)) {
1148 r300_emit_texture_count(r300
);
1150 for (i
= 0; i
< MIN2(r300
->sampler_count
, r300
->texture_count
); i
++) {
1151 if (r300
->dirty_state
&
1152 ((R300_NEW_SAMPLER
<< i
) | (R300_NEW_TEXTURE
<< i
))) {
1153 if (r300
->textures
[i
]) {
1154 r300_emit_texture(r300
,
1155 r300
->sampler_states
[i
],
1158 dirty_tex
|= r300
->dirty_state
& (R300_NEW_TEXTURE
<< i
);
1160 r300
->dirty_state
&=
1161 ~((R300_NEW_SAMPLER
<< i
) | (R300_NEW_TEXTURE
<< i
));
1164 r300
->dirty_state
&= ~(R300_ANY_NEW_SAMPLERS
| R300_ANY_NEW_TEXTURES
);
1168 r300_flush_textures(r300
);
1171 if (r300
->dirty_state
& R300_NEW_VERTEX_SHADER_CONSTANTS
) {
1172 struct r300_vertex_shader
* vs
= r300
->vs_state
.state
;
1173 r300_emit_vs_constant_buffer(r300
, &vs
->code
.constants
);
1174 r300
->dirty_state
&= ~R300_NEW_VERTEX_SHADER_CONSTANTS
;
1178 assert(r300->dirty_state == 0);
1181 /* Finally, emit the VBO. */
1182 /* r300_emit_vertex_buffer(r300); */