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"
35 #include "r300_state_inlines.h"
38 void r300_emit_blend_state(struct r300_context
* r300
, 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
, void* state
)
64 struct r300_blend_color_state
* bc
= (struct r300_blend_color_state
*)state
;
65 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
68 if (r300screen
->caps
->is_r500
) {
70 OUT_CS_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR
, 2);
71 OUT_CS(bc
->blend_color_red_alpha
);
72 OUT_CS(bc
->blend_color_green_blue
);
76 OUT_CS_REG(R300_RB3D_BLEND_COLOR
, bc
->blend_color
);
81 void r300_emit_clip_state(struct r300_context
* r300
, void* state
)
83 struct pipe_clip_state
* clip
= (struct pipe_clip_state
*)state
;
85 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
88 if (r300screen
->caps
->has_tcl
) {
89 BEGIN_CS(5 + (6 * 4));
90 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG
,
91 (r300screen
->caps
->is_r500
?
92 R500_PVS_UCP_START
: R300_PVS_UCP_START
));
93 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, 6 * 4);
94 for (i
= 0; i
< 6; i
++) {
95 OUT_CS_32F(clip
->ucp
[i
][0]);
96 OUT_CS_32F(clip
->ucp
[i
][1]);
97 OUT_CS_32F(clip
->ucp
[i
][2]);
98 OUT_CS_32F(clip
->ucp
[i
][3]);
100 OUT_CS_REG(R300_VAP_CLIP_CNTL
, ((1 << clip
->nr
) - 1) |
101 R300_PS_UCP_MODE_CLIP_AS_TRIFAN
);
105 OUT_CS_REG(R300_VAP_CLIP_CNTL
, R300_CLIP_DISABLE
);
111 void r300_emit_dsa_state(struct r300_context
* r300
, void* state
)
113 struct r300_dsa_state
* dsa
= (struct r300_dsa_state
*)state
;
114 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
115 struct pipe_framebuffer_state
* fb
=
116 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
117 struct pipe_stencil_ref stencil_ref
= r300
->stencil_ref
;
120 BEGIN_CS(r300screen
->caps
->is_r500
? 8 : 6);
121 OUT_CS_REG(R300_FG_ALPHA_FUNC
, dsa
->alpha_function
);
122 OUT_CS_REG_SEQ(R300_ZB_CNTL
, 3);
125 OUT_CS(dsa
->z_buffer_control
);
126 OUT_CS(dsa
->z_stencil_control
);
132 OUT_CS(dsa
->stencil_ref_mask
| stencil_ref
.ref_value
[0]);
134 if (r300screen
->caps
->is_r500
) {
135 OUT_CS_REG(R500_ZB_STENCILREFMASK_BF
, dsa
->stencil_ref_bf
| stencil_ref
.ref_value
[1]);
140 static const float * get_shader_constant(
141 struct r300_context
* r300
,
142 struct rc_constant
* constant
,
143 struct r300_constant_buffer
* externals
)
145 struct r300_viewport_state
* viewport
=
146 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
147 static float vec
[4] = { 0.0, 0.0, 0.0, 1.0 };
148 struct pipe_texture
*tex
;
150 switch(constant
->Type
) {
151 case RC_CONSTANT_EXTERNAL
:
152 return externals
->constants
[constant
->u
.External
];
154 case RC_CONSTANT_IMMEDIATE
:
155 return constant
->u
.Immediate
;
157 case RC_CONSTANT_STATE
:
158 switch (constant
->u
.State
[0]) {
159 /* Factor for converting rectangle coords to
160 * normalized coords. Should only show up on non-r500. */
161 case RC_STATE_R300_TEXRECT_FACTOR
:
162 tex
= &r300
->textures
[constant
->u
.State
[1]]->tex
;
163 vec
[0] = 1.0 / tex
->width0
;
164 vec
[1] = 1.0 / tex
->height0
;
167 /* Texture compare-fail value. Shouldn't ever show up, but if
168 * it does, we'll be ready. */
169 case RC_STATE_SHADOW_AMBIENT
:
173 case RC_STATE_R300_VIEWPORT_SCALE
:
174 if (r300
->tcl_bypass
) {
179 vec
[0] = viewport
->xscale
;
180 vec
[1] = viewport
->yscale
;
181 vec
[2] = viewport
->zscale
;
185 case RC_STATE_R300_VIEWPORT_OFFSET
:
186 if (!r300
->tcl_bypass
) {
187 vec
[0] = viewport
->xoffset
;
188 vec
[1] = viewport
->yoffset
;
189 vec
[2] = viewport
->zoffset
;
194 debug_printf("r300: Implementation error: "
195 "Unknown RC_CONSTANT type %d\n", constant
->u
.State
[0]);
200 debug_printf("r300: Implementation error: "
201 "Unhandled constant type %d\n", constant
->Type
);
204 /* This should either be (0, 0, 0, 1), which should be a relatively safe
205 * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
210 /* Convert a normal single-precision float into the 7.16 format
211 * used by the R300 fragment shader.
213 static uint32_t pack_float24(float f
)
221 uint32_t float24
= 0;
228 mantissa
= frexpf(f
, &exponent
);
232 float24
|= (1 << 23);
233 mantissa
= mantissa
* -1.0;
235 /* Handle exponent, bias of 63 */
237 float24
|= (exponent
<< 16);
238 /* Kill 7 LSB of mantissa */
239 float24
|= (u
.u
& 0x7FFFFF) >> 7;
244 void r300_emit_fragment_program_code(struct r300_context
* r300
,
245 struct rX00_fragment_program_code
* generic_code
)
247 struct r300_fragment_program_code
* code
= &generic_code
->code
.r300
;
252 code
->alu
.length
* 4 +
253 (code
->tex
.length
? (1 + code
->tex
.length
) : 0));
255 OUT_CS_REG(R300_US_CONFIG
, code
->config
);
256 OUT_CS_REG(R300_US_PIXSIZE
, code
->pixsize
);
257 OUT_CS_REG(R300_US_CODE_OFFSET
, code
->code_offset
);
259 OUT_CS_REG_SEQ(R300_US_CODE_ADDR_0
, 4);
260 for(i
= 0; i
< 4; ++i
)
261 OUT_CS(code
->code_addr
[i
]);
263 OUT_CS_REG_SEQ(R300_US_ALU_RGB_INST_0
, code
->alu
.length
);
264 for (i
= 0; i
< code
->alu
.length
; i
++)
265 OUT_CS(code
->alu
.inst
[i
].rgb_inst
);
267 OUT_CS_REG_SEQ(R300_US_ALU_RGB_ADDR_0
, code
->alu
.length
);
268 for (i
= 0; i
< code
->alu
.length
; i
++)
269 OUT_CS(code
->alu
.inst
[i
].rgb_addr
);
271 OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_INST_0
, code
->alu
.length
);
272 for (i
= 0; i
< code
->alu
.length
; i
++)
273 OUT_CS(code
->alu
.inst
[i
].alpha_inst
);
275 OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0
, code
->alu
.length
);
276 for (i
= 0; i
< code
->alu
.length
; i
++)
277 OUT_CS(code
->alu
.inst
[i
].alpha_addr
);
279 if (code
->tex
.length
) {
280 OUT_CS_REG_SEQ(R300_US_TEX_INST_0
, code
->tex
.length
);
281 for(i
= 0; i
< code
->tex
.length
; ++i
)
282 OUT_CS(code
->tex
.inst
[i
]);
288 void r300_emit_fs_constant_buffer(struct r300_context
* r300
,
289 struct rc_constant_list
* constants
)
294 if (constants
->Count
== 0)
297 BEGIN_CS(constants
->Count
* 4 + 1);
298 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X
, constants
->Count
* 4);
299 for(i
= 0; i
< constants
->Count
; ++i
) {
300 const float * data
= get_shader_constant(r300
,
301 &constants
->Constants
[i
],
302 &r300
->shader_constants
[PIPE_SHADER_FRAGMENT
]);
303 OUT_CS(pack_float24(data
[0]));
304 OUT_CS(pack_float24(data
[1]));
305 OUT_CS(pack_float24(data
[2]));
306 OUT_CS(pack_float24(data
[3]));
311 static void r300_emit_fragment_depth_config(struct r300_context
* r300
,
312 struct r300_fragment_shader
* fs
)
317 if (r300_fragment_shader_writes_depth(fs
)) {
318 OUT_CS_REG(R300_FG_DEPTH_SRC
, R300_FG_DEPTH_SRC_SHADER
);
319 OUT_CS_REG(R300_US_W_FMT
, R300_W_FMT_W24
| R300_W_SRC_US
);
321 OUT_CS_REG(R300_FG_DEPTH_SRC
, R300_FG_DEPTH_SRC_SCAN
);
322 OUT_CS_REG(R300_US_W_FMT
, R300_W_FMT_W0
| R300_W_SRC_US
);
327 void r500_emit_fragment_program_code(struct r300_context
* r300
,
328 struct rX00_fragment_program_code
* generic_code
)
330 struct r500_fragment_program_code
* code
= &generic_code
->code
.r500
;
335 ((code
->inst_end
+ 1) * 6));
336 OUT_CS_REG(R500_US_CONFIG
, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO
);
337 OUT_CS_REG(R500_US_PIXSIZE
, code
->max_temp_idx
);
338 OUT_CS_REG(R500_US_CODE_RANGE
,
339 R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code
->inst_end
));
340 OUT_CS_REG(R500_US_CODE_OFFSET
, 0);
341 OUT_CS_REG(R500_US_CODE_ADDR
,
342 R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code
->inst_end
));
344 OUT_CS_REG(R500_GA_US_VECTOR_INDEX
, R500_GA_US_VECTOR_INDEX_TYPE_INSTR
);
345 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA
, (code
->inst_end
+ 1) * 6);
346 for (i
= 0; i
<= code
->inst_end
; i
++) {
347 OUT_CS(code
->inst
[i
].inst0
);
348 OUT_CS(code
->inst
[i
].inst1
);
349 OUT_CS(code
->inst
[i
].inst2
);
350 OUT_CS(code
->inst
[i
].inst3
);
351 OUT_CS(code
->inst
[i
].inst4
);
352 OUT_CS(code
->inst
[i
].inst5
);
358 void r500_emit_fs_constant_buffer(struct r300_context
* r300
,
359 struct rc_constant_list
* constants
)
364 if (constants
->Count
== 0)
367 BEGIN_CS(constants
->Count
* 4 + 3);
368 OUT_CS_REG(R500_GA_US_VECTOR_INDEX
, R500_GA_US_VECTOR_INDEX_TYPE_CONST
);
369 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA
, constants
->Count
* 4);
370 for (i
= 0; i
< constants
->Count
; i
++) {
371 const float * data
= get_shader_constant(r300
,
372 &constants
->Constants
[i
],
373 &r300
->shader_constants
[PIPE_SHADER_FRAGMENT
]);
382 void r300_emit_fb_state(struct r300_context
* r300
, void* state
)
384 struct pipe_framebuffer_state
* fb
= (struct pipe_framebuffer_state
*)state
;
385 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
386 struct r300_texture
* tex
;
387 struct pipe_surface
* surf
;
391 BEGIN_CS((10 * fb
->nr_cbufs
) + (2 * (4 - fb
->nr_cbufs
)) +
392 (fb
->zsbuf
? 10 : 0) + 6);
394 /* Flush and free renderbuffer caches. */
395 OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT
,
396 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS
|
397 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D
);
398 OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT
,
399 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE
|
400 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE
);
402 /* Set the number of colorbuffers. */
403 if (fb
->nr_cbufs
> 1) {
404 if (r300screen
->caps
->is_r500
) {
405 OUT_CS_REG(R300_RB3D_CCTL
,
406 R300_RB3D_CCTL_NUM_MULTIWRITES(fb
->nr_cbufs
) |
407 R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE
|
408 R300_RB3D_CCTL_INDEPENDENT_COLOR_CHANNEL_MASK_ENABLE
);
410 OUT_CS_REG(R300_RB3D_CCTL
,
411 R300_RB3D_CCTL_NUM_MULTIWRITES(fb
->nr_cbufs
) |
412 R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE
);
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
->pitch
[surf
->level
] |
429 r300_translate_colorformat(tex
->tex
.format
) |
430 R300_COLOR_TILE(tex
->macrotile
) |
431 R300_COLOR_MICROTILE(tex
->microtile
),
432 0, RADEON_GEM_DOMAIN_VRAM
, 0);
434 OUT_CS_REG(R300_US_OUT_FMT_0
+ (4 * i
),
435 r300_translate_out_fmt(surf
->format
));
438 /* Disable unused colorbuffers. */
440 OUT_CS_REG(R300_US_OUT_FMT_0
+ (4 * i
), R300_US_OUT_FMT_UNUSED
);
443 /* Set up a zbuffer. */
446 tex
= (struct r300_texture
*)surf
->texture
;
447 assert(tex
&& tex
->buffer
&& "zsbuf is marked, but NULL!");
449 OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET
, 1);
450 OUT_CS_RELOC(tex
->buffer
, surf
->offset
, 0, RADEON_GEM_DOMAIN_VRAM
, 0);
452 OUT_CS_REG(R300_ZB_FORMAT
, r300_translate_zsformat(tex
->tex
.format
));
454 OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH
, 1);
455 OUT_CS_RELOC(tex
->buffer
, tex
->pitch
[surf
->level
] |
456 R300_DEPTHMACROTILE(tex
->macrotile
) |
457 R300_DEPTHMICROTILE(tex
->microtile
),
458 0, RADEON_GEM_DOMAIN_VRAM
, 0);
464 static void r300_emit_query_start(struct r300_context
*r300
)
466 struct r300_capabilities
*caps
= r300_screen(r300
->context
.screen
)->caps
;
467 struct r300_query
*query
= r300
->query_current
;
474 if (caps
->family
== CHIP_FAMILY_RV530
) {
475 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL
);
477 OUT_CS_REG(R300_SU_REG_DEST
, R300_RASTER_PIPE_SELECT_ALL
);
479 OUT_CS_REG(R300_ZB_ZPASS_DATA
, 0);
481 query
->begin_emitted
= TRUE
;
485 static void r300_emit_query_finish(struct r300_context
*r300
,
486 struct r300_query
*query
)
488 struct r300_capabilities
* caps
= r300_screen(r300
->context
.screen
)->caps
;
491 assert(caps
->num_frag_pipes
);
493 BEGIN_CS(6 * caps
->num_frag_pipes
+ 2);
494 /* I'm not so sure I like this switch, but it's hard to be elegant
495 * when there's so many special cases...
497 * So here's the basic idea. For each pipe, enable writes to it only,
498 * then put out the relocation for ZPASS_ADDR, taking into account a
499 * 4-byte offset for each pipe. RV380 and older are special; they have
500 * only two pipes, and the second pipe's enable is on bit 3, not bit 1,
501 * so there's a chipset cap for that. */
502 switch (caps
->num_frag_pipes
) {
505 OUT_CS_REG(R300_SU_REG_DEST
, 1 << 3);
506 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
507 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 3),
508 0, RADEON_GEM_DOMAIN_GTT
, 0);
511 OUT_CS_REG(R300_SU_REG_DEST
, 1 << 2);
512 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
513 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 2),
514 0, RADEON_GEM_DOMAIN_GTT
, 0);
517 /* As mentioned above, accomodate RV380 and older. */
518 OUT_CS_REG(R300_SU_REG_DEST
,
519 1 << (caps
->high_second_pipe
? 3 : 1));
520 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
521 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 1),
522 0, RADEON_GEM_DOMAIN_GTT
, 0);
525 OUT_CS_REG(R300_SU_REG_DEST
, 1 << 0);
526 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
527 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ (sizeof(uint32_t) * 0),
528 0, RADEON_GEM_DOMAIN_GTT
, 0);
531 debug_printf("r300: Implementation error: Chipset reports %d"
532 " pixel pipes!\n", caps
->num_frag_pipes
);
536 /* And, finally, reset it to normal... */
537 OUT_CS_REG(R300_SU_REG_DEST
, 0xF);
541 static void rv530_emit_query_single(struct r300_context
*r300
,
542 struct r300_query
*query
)
547 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_0
);
548 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
549 OUT_CS_RELOC(r300
->oqbo
, query
->offset
, 0, RADEON_GEM_DOMAIN_GTT
, 0);
550 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL
);
554 static void rv530_emit_query_double(struct r300_context
*r300
,
555 struct r300_query
*query
)
560 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_0
);
561 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
562 OUT_CS_RELOC(r300
->oqbo
, query
->offset
, 0, RADEON_GEM_DOMAIN_GTT
, 0);
563 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_1
);
564 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR
, 1);
565 OUT_CS_RELOC(r300
->oqbo
, query
->offset
+ sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT
, 0);
566 OUT_CS_REG(RV530_FG_ZBREG_DEST
, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL
);
570 void r300_emit_query_end(struct r300_context
* r300
)
572 struct r300_capabilities
*caps
= r300_screen(r300
->context
.screen
)->caps
;
573 struct r300_query
*query
= r300
->query_current
;
578 if (query
->begin_emitted
== FALSE
)
581 if (caps
->family
== CHIP_FAMILY_RV530
) {
582 if (caps
->num_z_pipes
== 2)
583 rv530_emit_query_double(r300
, query
);
585 rv530_emit_query_single(r300
, query
);
587 r300_emit_query_finish(r300
, query
);
590 void r300_emit_rs_state(struct r300_context
* r300
, void* state
)
592 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
596 BEGIN_CS(18 + (rs
->polygon_offset_enable
? 5 : 0));
597 OUT_CS_REG(R300_VAP_CNTL_STATUS
, rs
->vap_control_status
);
599 OUT_CS_REG(R300_GB_AA_CONFIG
, rs
->antialiasing_config
);
601 OUT_CS_REG(R300_GA_POINT_SIZE
, rs
->point_size
);
602 OUT_CS_REG_SEQ(R300_GA_POINT_MINMAX
, 2);
603 OUT_CS(rs
->point_minmax
);
604 OUT_CS(rs
->line_control
);
606 if (rs
->polygon_offset_enable
) {
607 scale
= rs
->depth_scale
* 12;
608 offset
= rs
->depth_offset
;
610 switch (r300
->zbuffer_bpp
) {
619 OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE
, 4);
626 OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE
, 2);
627 OUT_CS(rs
->polygon_offset_enable
);
628 OUT_CS(rs
->cull_mode
);
629 OUT_CS_REG(R300_GA_LINE_STIPPLE_CONFIG
, rs
->line_stipple_config
);
630 OUT_CS_REG(R300_GA_LINE_STIPPLE_VALUE
, rs
->line_stipple_value
);
631 OUT_CS_REG(R300_GA_POLY_MODE
, rs
->polygon_mode
);
635 void r300_emit_rs_block_state(struct r300_context
* r300
, void* state
)
637 struct r300_rs_block
* rs
= (struct r300_rs_block
*)state
;
639 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
642 DBG(r300
, DBG_DRAW
, "r300: RS emit:\n");
645 if (r300screen
->caps
->is_r500
) {
646 OUT_CS_REG_SEQ(R500_RS_IP_0
, 8);
648 OUT_CS_REG_SEQ(R300_RS_IP_0
, 8);
650 for (i
= 0; i
< 8; i
++) {
652 DBG(r300
, DBG_DRAW
, " : ip %d: 0x%08x\n", i
, rs
->ip
[i
]);
655 OUT_CS_REG_SEQ(R300_RS_COUNT
, 2);
657 OUT_CS(rs
->inst_count
);
659 if (r300screen
->caps
->is_r500
) {
660 OUT_CS_REG_SEQ(R500_RS_INST_0
, 8);
662 OUT_CS_REG_SEQ(R300_RS_INST_0
, 8);
664 for (i
= 0; i
< 8; i
++) {
666 DBG(r300
, DBG_DRAW
, " : inst %d: 0x%08x\n", i
, rs
->inst
[i
]);
669 DBG(r300
, DBG_DRAW
, " : count: 0x%08x inst_count: 0x%08x\n",
670 rs
->count
, rs
->inst_count
);
675 void r300_emit_scissor_state(struct r300_context
* r300
, void* state
)
677 unsigned minx
, miny
, maxx
, maxy
;
678 uint32_t top_left
, bottom_right
;
679 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
680 struct pipe_scissor_state
* scissor
= (struct pipe_scissor_state
*)state
;
681 struct pipe_framebuffer_state
* fb
=
682 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
689 if (((struct r300_rs_state
*)r300
->rs_state
.state
)->rs
.scissor
) {
690 minx
= MAX2(minx
, scissor
->minx
);
691 miny
= MAX2(miny
, scissor
->miny
);
692 maxx
= MIN2(maxx
, scissor
->maxx
);
693 maxy
= MIN2(maxy
, scissor
->maxy
);
696 /* Special case for zero-area scissor.
698 * We can't allow the variables maxx and maxy to be zero because they are
699 * subtracted from later in the code, which would cause emitting ~0 and
700 * making the kernel checker angry.
702 * Let's consider we change maxx and maxy to 1, which is effectively
703 * a one-pixel area. We must then change minx and miny to a number which is
704 * greater than 1 to get the zero area back. */
705 if (!maxx
|| !maxy
) {
712 if (r300screen
->caps
->is_r500
) {
714 (minx
<< R300_SCISSORS_X_SHIFT
) |
715 (miny
<< R300_SCISSORS_Y_SHIFT
);
717 ((maxx
- 1) << R300_SCISSORS_X_SHIFT
) |
718 ((maxy
- 1) << R300_SCISSORS_Y_SHIFT
);
720 /* Offset of 1440 in non-R500 chipsets. */
722 ((minx
+ 1440) << R300_SCISSORS_X_SHIFT
) |
723 ((miny
+ 1440) << R300_SCISSORS_Y_SHIFT
);
725 (((maxx
- 1) + 1440) << R300_SCISSORS_X_SHIFT
) |
726 (((maxy
- 1) + 1440) << R300_SCISSORS_Y_SHIFT
);
730 OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL
, 2);
732 OUT_CS(bottom_right
);
736 void r300_emit_texture(struct r300_context
* r300
,
737 struct r300_sampler_state
* sampler
,
738 struct r300_texture
* tex
,
741 uint32_t filter0
= sampler
->filter0
;
742 uint32_t format0
= tex
->state
.format0
;
743 unsigned min_level
, max_level
;
746 /* to emulate 1D textures through 2D ones correctly */
747 if (tex
->tex
.target
== PIPE_TEXTURE_1D
) {
748 filter0
&= ~R300_TX_WRAP_T_MASK
;
749 filter0
|= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE
);
753 /* NPOT textures don't support mip filter, unfortunately.
754 * This prevents incorrect rendering. */
755 filter0
&= ~R300_TX_MIN_FILTER_MIP_MASK
;
757 /* determine min/max levels */
758 /* the MAX_MIP level is the largest (finest) one */
759 max_level
= MIN2(sampler
->max_lod
, tex
->tex
.last_level
);
760 min_level
= MIN2(sampler
->min_lod
, max_level
);
761 format0
|= R300_TX_NUM_LEVELS(max_level
);
762 filter0
|= R300_TX_MAX_MIP_LEVEL(min_level
);
766 OUT_CS_REG(R300_TX_FILTER0_0
+ (offset
* 4), filter0
|
768 OUT_CS_REG(R300_TX_FILTER1_0
+ (offset
* 4), sampler
->filter1
);
769 OUT_CS_REG(R300_TX_BORDER_COLOR_0
+ (offset
* 4), sampler
->border_color
);
771 OUT_CS_REG(R300_TX_FORMAT0_0
+ (offset
* 4), format0
);
772 OUT_CS_REG(R300_TX_FORMAT1_0
+ (offset
* 4), tex
->state
.format1
);
773 OUT_CS_REG(R300_TX_FORMAT2_0
+ (offset
* 4), tex
->state
.format2
);
774 OUT_CS_REG_SEQ(R300_TX_OFFSET_0
+ (offset
* 4), 1);
775 OUT_CS_RELOC(tex
->buffer
,
776 R300_TXO_MACRO_TILE(tex
->macrotile
) |
777 R300_TXO_MICRO_TILE(tex
->microtile
),
778 RADEON_GEM_DOMAIN_GTT
| RADEON_GEM_DOMAIN_VRAM
, 0, 0);
782 void r300_emit_aos(struct r300_context
* r300
, unsigned offset
)
784 struct pipe_vertex_buffer
*vb1
, *vb2
, *vbuf
= r300
->vertex_buffer
;
785 struct pipe_vertex_element
*velem
= r300
->vertex_element
;
787 unsigned size1
, size2
, aos_count
= r300
->vertex_element_count
;
788 unsigned packet_size
= (aos_count
* 3 + 1) / 2;
791 BEGIN_CS(2 + packet_size
+ aos_count
* 2);
792 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR
, packet_size
);
795 for (i
= 0; i
< aos_count
- 1; i
+= 2) {
796 vb1
= &vbuf
[velem
[i
].vertex_buffer_index
];
797 vb2
= &vbuf
[velem
[i
+1].vertex_buffer_index
];
798 size1
= util_format_get_blocksize(velem
[i
].src_format
);
799 size2
= util_format_get_blocksize(velem
[i
+1].src_format
);
801 OUT_CS(R300_VBPNTR_SIZE0(size1
) | R300_VBPNTR_STRIDE0(vb1
->stride
) |
802 R300_VBPNTR_SIZE1(size2
) | R300_VBPNTR_STRIDE1(vb2
->stride
));
803 OUT_CS(vb1
->buffer_offset
+ velem
[i
].src_offset
+ offset
* vb1
->stride
);
804 OUT_CS(vb2
->buffer_offset
+ velem
[i
+1].src_offset
+ offset
* vb2
->stride
);
808 vb1
= &vbuf
[velem
[i
].vertex_buffer_index
];
809 size1
= util_format_get_blocksize(velem
[i
].src_format
);
811 OUT_CS(R300_VBPNTR_SIZE0(size1
) | R300_VBPNTR_STRIDE0(vb1
->stride
));
812 OUT_CS(vb1
->buffer_offset
+ velem
[i
].src_offset
+ offset
* vb1
->stride
);
815 for (i
= 0; i
< aos_count
; i
++) {
816 OUT_CS_RELOC_NO_OFFSET(vbuf
[velem
[i
].vertex_buffer_index
].buffer
,
817 RADEON_GEM_DOMAIN_GTT
, 0, 0);
822 void r300_emit_vertex_format_state(struct r300_context
* r300
, void* state
)
824 struct r300_vertex_info
* vertex_info
= (struct r300_vertex_info
*)state
;
828 DBG(r300
, DBG_DRAW
, "r300: VAP/PSC emit:\n");
831 OUT_CS_REG(R300_VAP_VTX_SIZE
, vertex_info
->vinfo
.size
);
833 OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL
, 2);
834 OUT_CS(vertex_info
->vinfo
.hwfmt
[0]);
835 OUT_CS(vertex_info
->vinfo
.hwfmt
[1]);
836 OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0
, 2);
837 OUT_CS(vertex_info
->vinfo
.hwfmt
[2]);
838 OUT_CS(vertex_info
->vinfo
.hwfmt
[3]);
839 for (i
= 0; i
< 4; i
++) {
840 DBG(r300
, DBG_DRAW
, " : hwfmt%d: 0x%08x\n", i
,
841 vertex_info
->vinfo
.hwfmt
[i
]);
844 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0
, 8);
845 for (i
= 0; i
< 8; i
++) {
846 OUT_CS(vertex_info
->vap_prog_stream_cntl
[i
]);
847 DBG(r300
, DBG_DRAW
, " : prog_stream_cntl%d: 0x%08x\n", i
,
848 vertex_info
->vap_prog_stream_cntl
[i
]);
850 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0
, 8);
851 for (i
= 0; i
< 8; i
++) {
852 OUT_CS(vertex_info
->vap_prog_stream_cntl_ext
[i
]);
853 DBG(r300
, DBG_DRAW
, " : prog_stream_cntl_ext%d: 0x%08x\n", i
,
854 vertex_info
->vap_prog_stream_cntl_ext
[i
]);
860 void r300_emit_vertex_program_code(struct r300_context
* r300
,
861 struct r300_vertex_program_code
* code
)
864 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
865 unsigned instruction_count
= code
->length
/ 4;
867 int vtx_mem_size
= r300screen
->caps
->is_r500
? 128 : 72;
868 int input_count
= MAX2(util_bitcount(code
->InputsRead
), 1);
869 int output_count
= MAX2(util_bitcount(code
->OutputsWritten
), 1);
870 int temp_count
= MAX2(code
->num_temporaries
, 1);
871 int pvs_num_slots
= MIN3(vtx_mem_size
/ input_count
,
872 vtx_mem_size
/ output_count
, 10);
873 int pvs_num_controllers
= MIN2(vtx_mem_size
/ temp_count
, 6);
877 if (!r300screen
->caps
->has_tcl
) {
878 debug_printf("r300: Implementation error: emit_vertex_shader called,"
879 " but has_tcl is FALSE!\n");
883 BEGIN_CS(9 + code
->length
);
884 /* R300_VAP_PVS_CODE_CNTL_0
885 * R300_VAP_PVS_CONST_CNTL
886 * R300_VAP_PVS_CODE_CNTL_1
887 * See the r5xx docs for instructions on how to use these. */
888 OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0
, 3);
889 OUT_CS(R300_PVS_FIRST_INST(0) |
890 R300_PVS_XYZW_VALID_INST(instruction_count
- 1) |
891 R300_PVS_LAST_INST(instruction_count
- 1));
892 OUT_CS(R300_PVS_MAX_CONST_ADDR(code
->constants
.Count
- 1));
893 OUT_CS(instruction_count
- 1);
895 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG
, 0);
896 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, code
->length
);
897 for (i
= 0; i
< code
->length
; i
++)
898 OUT_CS(code
->body
.d
[i
]);
900 OUT_CS_REG(R300_VAP_CNTL
, R300_PVS_NUM_SLOTS(pvs_num_slots
) |
901 R300_PVS_NUM_CNTLRS(pvs_num_controllers
) |
902 R300_PVS_NUM_FPUS(r300screen
->caps
->num_vert_fpus
) |
903 R300_PVS_VF_MAX_VTX_NUM(12) |
904 (r300screen
->caps
->is_r500
? R500_TCL_STATE_OPTIMIZATION
: 0));
908 void r300_emit_vertex_shader(struct r300_context
* r300
,
909 struct r300_vertex_shader
* vs
)
911 r300_emit_vertex_program_code(r300
, &vs
->code
);
914 void r300_emit_vs_constant_buffer(struct r300_context
* r300
,
915 struct rc_constant_list
* constants
)
918 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
921 if (!r300screen
->caps
->has_tcl
) {
922 debug_printf("r300: Implementation error: emit_vertex_shader called,"
923 " but has_tcl is FALSE!\n");
927 if (constants
->Count
== 0)
930 BEGIN_CS(constants
->Count
* 4 + 3);
931 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG
,
932 (r300screen
->caps
->is_r500
?
933 R500_PVS_CONST_START
: R300_PVS_CONST_START
));
934 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, constants
->Count
* 4);
935 for (i
= 0; i
< constants
->Count
; i
++) {
936 const float * data
= get_shader_constant(r300
,
937 &constants
->Constants
[i
],
938 &r300
->shader_constants
[PIPE_SHADER_VERTEX
]);
947 void r300_emit_viewport_state(struct r300_context
* r300
, void* state
)
949 struct r300_viewport_state
* viewport
= (struct r300_viewport_state
*)state
;
952 if (r300
->tcl_bypass
) {
954 OUT_CS_REG(R300_VAP_VTE_CNTL
, 0);
958 OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE
, 6);
959 OUT_CS_32F(viewport
->xscale
);
960 OUT_CS_32F(viewport
->xoffset
);
961 OUT_CS_32F(viewport
->yscale
);
962 OUT_CS_32F(viewport
->yoffset
);
963 OUT_CS_32F(viewport
->zscale
);
964 OUT_CS_32F(viewport
->zoffset
);
965 OUT_CS_REG(R300_VAP_VTE_CNTL
, viewport
->vte_control
);
970 void r300_emit_texture_count(struct r300_context
* r300
)
972 uint32_t tx_enable
= 0;
976 /* Notice that texture_count and sampler_count are just sizes
977 * of the respective arrays. We still have to check for the individual
979 for (i
= 0; i
< MIN2(r300
->sampler_count
, r300
->texture_count
); i
++) {
980 if (r300
->textures
[i
]) {
986 OUT_CS_REG(R300_TX_ENABLE
, tx_enable
);
991 void r300_emit_ztop_state(struct r300_context
* r300
, void* state
)
993 struct r300_ztop_state
* ztop
= (struct r300_ztop_state
*)state
;
997 OUT_CS_REG(R300_ZB_ZTOP
, ztop
->z_buffer_top
);
1001 void r300_flush_textures(struct r300_context
* r300
)
1006 OUT_CS_REG(R300_TX_INVALTAGS
, 0);
1010 static void r300_flush_pvs(struct r300_context
* r300
)
1015 OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG
, 0x0);
1019 void r300_emit_buffer_validate(struct r300_context
*r300
)
1021 struct pipe_framebuffer_state
* fb
=
1022 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
1023 struct r300_texture
* tex
;
1025 boolean invalid
= FALSE
;
1027 /* Clean out BOs. */
1028 r300
->winsys
->reset_bos(r300
->winsys
);
1031 /* Color buffers... */
1032 for (i
= 0; i
< fb
->nr_cbufs
; i
++) {
1033 tex
= (struct r300_texture
*)fb
->cbufs
[i
]->texture
;
1034 assert(tex
&& tex
->buffer
&& "cbuf is marked, but NULL!");
1035 if (!r300
->winsys
->add_buffer(r300
->winsys
, tex
->buffer
,
1036 0, RADEON_GEM_DOMAIN_VRAM
)) {
1037 r300
->context
.flush(&r300
->context
, 0, NULL
);
1041 /* ...depth buffer... */
1043 tex
= (struct r300_texture
*)fb
->zsbuf
->texture
;
1044 assert(tex
&& tex
->buffer
&& "zsbuf is marked, but NULL!");
1045 if (!r300
->winsys
->add_buffer(r300
->winsys
, tex
->buffer
,
1046 0, RADEON_GEM_DOMAIN_VRAM
)) {
1047 r300
->context
.flush(&r300
->context
, 0, NULL
);
1051 /* ...textures... */
1052 for (i
= 0; i
< r300
->texture_count
; i
++) {
1053 tex
= r300
->textures
[i
];
1056 if (!r300
->winsys
->add_buffer(r300
->winsys
, tex
->buffer
,
1057 RADEON_GEM_DOMAIN_GTT
| RADEON_GEM_DOMAIN_VRAM
, 0)) {
1058 r300
->context
.flush(&r300
->context
, 0, NULL
);
1062 /* ...occlusion query buffer... */
1063 if (r300
->dirty_state
& R300_NEW_QUERY
) {
1064 if (!r300
->winsys
->add_buffer(r300
->winsys
, r300
->oqbo
,
1065 0, RADEON_GEM_DOMAIN_GTT
)) {
1066 r300
->context
.flush(&r300
->context
, 0, NULL
);
1070 /* ...and vertex buffer. */
1072 if (!r300
->winsys
->add_buffer(r300
->winsys
, r300
->vbo
,
1073 RADEON_GEM_DOMAIN_GTT
, 0)) {
1074 r300
->context
.flush(&r300
->context
, 0, NULL
);
1078 /* debug_printf("No VBO while emitting dirty state!\n"); */
1080 if (!r300
->winsys
->validate(r300
->winsys
)) {
1081 r300
->context
.flush(&r300
->context
, 0, NULL
);
1084 debug_printf("r300: Stuck in validation loop, gonna quit now.");
1092 /* Emit all dirty state. */
1093 void r300_emit_dirty_state(struct r300_context
* r300
)
1095 struct r300_screen
* r300screen
= r300_screen(r300
->context
.screen
);
1096 struct r300_atom
* atom
;
1097 unsigned i
, dwords
= 1024;
1100 /* Check the required number of dwords against the space remaining in the
1101 * current CS object. If we need more, then flush. */
1103 foreach(atom
, &r300
->atom_list
) {
1104 if (atom
->dirty
|| atom
->always_dirty
) {
1105 dwords
+= atom
->size
;
1109 /* Make sure we have at least 2*1024 spare dwords. */
1110 /* XXX It would be nice to know the number of dwords we really need to
1112 while (!r300
->winsys
->check_cs(r300
->winsys
, dwords
)) {
1113 r300
->context
.flush(&r300
->context
, 0, NULL
);
1116 if (r300
->dirty_state
& R300_NEW_QUERY
) {
1117 r300_emit_query_start(r300
);
1118 r300
->dirty_state
&= ~R300_NEW_QUERY
;
1121 foreach(atom
, &r300
->atom_list
) {
1122 if (atom
->dirty
|| atom
->always_dirty
) {
1123 atom
->emit(r300
, atom
->state
);
1124 atom
->dirty
= FALSE
;
1128 if (r300
->dirty_state
& R300_NEW_FRAGMENT_SHADER
) {
1129 r300_emit_fragment_depth_config(r300
, r300
->fs
);
1130 if (r300screen
->caps
->is_r500
) {
1131 r500_emit_fragment_program_code(r300
, &r300
->fs
->shader
->code
);
1133 r300_emit_fragment_program_code(r300
, &r300
->fs
->shader
->code
);
1135 r300
->dirty_state
&= ~R300_NEW_FRAGMENT_SHADER
;
1138 if (r300
->dirty_state
& R300_NEW_FRAGMENT_SHADER_CONSTANTS
) {
1139 if (r300screen
->caps
->is_r500
) {
1140 r500_emit_fs_constant_buffer(r300
,
1141 &r300
->fs
->shader
->code
.constants
);
1143 r300_emit_fs_constant_buffer(r300
,
1144 &r300
->fs
->shader
->code
.constants
);
1146 r300
->dirty_state
&= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1149 /* Samplers and textures are tracked separately but emitted together. */
1150 if (r300
->dirty_state
&
1151 (R300_ANY_NEW_SAMPLERS
| R300_ANY_NEW_TEXTURES
)) {
1152 r300_emit_texture_count(r300
);
1154 for (i
= 0; i
< MIN2(r300
->sampler_count
, r300
->texture_count
); i
++) {
1155 if (r300
->dirty_state
&
1156 ((R300_NEW_SAMPLER
<< i
) | (R300_NEW_TEXTURE
<< i
))) {
1157 if (r300
->textures
[i
])
1158 r300_emit_texture(r300
,
1159 r300
->sampler_states
[i
],
1162 r300
->dirty_state
&=
1163 ~((R300_NEW_SAMPLER
<< i
) | (R300_NEW_TEXTURE
<< i
));
1167 r300
->dirty_state
&= ~(R300_ANY_NEW_SAMPLERS
| R300_ANY_NEW_TEXTURES
);
1171 r300_flush_textures(r300
);
1174 if (r300
->dirty_state
& (R300_NEW_VERTEX_SHADER
| R300_NEW_VERTEX_SHADER_CONSTANTS
)) {
1175 r300_flush_pvs(r300
);
1178 if (r300
->dirty_state
& R300_NEW_VERTEX_SHADER
) {
1179 r300_emit_vertex_shader(r300
, r300
->vs
);
1180 r300
->dirty_state
&= ~R300_NEW_VERTEX_SHADER
;
1183 if (r300
->dirty_state
& R300_NEW_VERTEX_SHADER_CONSTANTS
) {
1184 r300_emit_vs_constant_buffer(r300
, &r300
->vs
->code
.constants
);
1185 r300
->dirty_state
&= ~R300_NEW_VERTEX_SHADER_CONSTANTS
;
1189 assert(r300->dirty_state == 0);
1192 /* Finally, emit the VBO. */
1193 /* r300_emit_vertex_buffer(r300); */