2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
34 #include "intel_batchbuffer.h"
35 #include "intel_fbo.h"
36 #include "intel_mipmap_tree.h"
37 #include "intel_regions.h"
39 #include "brw_context.h"
40 #include "brw_state.h"
41 #include "brw_defines.h"
43 /* Constant single cliprect for framebuffer object or DRI2 drawing */
44 static void upload_drawing_rect(struct brw_context
*brw
)
46 struct intel_context
*intel
= &brw
->intel
;
47 struct gl_context
*ctx
= &intel
->ctx
;
50 OUT_BATCH(_3DSTATE_DRAWING_RECTANGLE
<< 16 | (4 - 2));
51 OUT_BATCH(0); /* xmin, ymin */
52 OUT_BATCH(((ctx
->DrawBuffer
->Width
- 1) & 0xffff) |
53 ((ctx
->DrawBuffer
->Height
- 1) << 16));
58 const struct brw_tracked_state brw_drawing_rect
= {
61 .brw
= BRW_NEW_CONTEXT
,
64 .emit
= upload_drawing_rect
68 * Upload the binding table pointers, which point each stage's array of surface
71 * The binding table pointers are relative to the surface state base address,
72 * which points at the batchbuffer containing the streamed batch state.
74 static void upload_binding_table_pointers(struct brw_context
*brw
)
76 struct intel_context
*intel
= &brw
->intel
;
79 OUT_BATCH(_3DSTATE_BINDING_TABLE_POINTERS
<< 16 | (6 - 2));
80 OUT_BATCH(brw
->bind
.bo_offset
);
81 OUT_BATCH(0); /* gs */
82 OUT_BATCH(0); /* clip */
83 OUT_BATCH(0); /* sf */
84 OUT_BATCH(brw
->bind
.bo_offset
);
88 const struct brw_tracked_state brw_binding_table_pointers
= {
91 .brw
= (BRW_NEW_BATCH
|
92 BRW_NEW_STATE_BASE_ADDRESS
|
93 BRW_NEW_VS_BINDING_TABLE
|
94 BRW_NEW_GS_BINDING_TABLE
|
95 BRW_NEW_PS_BINDING_TABLE
),
98 .emit
= upload_binding_table_pointers
,
102 * Upload the binding table pointers, which point each stage's array of surface
105 * The binding table pointers are relative to the surface state base address,
106 * which points at the batchbuffer containing the streamed batch state.
108 static void upload_gen6_binding_table_pointers(struct brw_context
*brw
)
110 struct intel_context
*intel
= &brw
->intel
;
113 OUT_BATCH(_3DSTATE_BINDING_TABLE_POINTERS
<< 16 |
114 GEN6_BINDING_TABLE_MODIFY_VS
|
115 GEN6_BINDING_TABLE_MODIFY_GS
|
116 GEN6_BINDING_TABLE_MODIFY_PS
|
118 OUT_BATCH(brw
->bind
.bo_offset
); /* vs */
119 OUT_BATCH(0); /* gs */
120 OUT_BATCH(brw
->bind
.bo_offset
); /* wm/ps */
124 const struct brw_tracked_state gen6_binding_table_pointers
= {
127 .brw
= (BRW_NEW_BATCH
|
128 BRW_NEW_STATE_BASE_ADDRESS
|
129 BRW_NEW_VS_BINDING_TABLE
|
130 BRW_NEW_GS_BINDING_TABLE
|
131 BRW_NEW_PS_BINDING_TABLE
),
134 .emit
= upload_gen6_binding_table_pointers
,
138 * Upload pointers to the per-stage state.
140 * The state pointers in this packet are all relative to the general state
141 * base address set by CMD_STATE_BASE_ADDRESS, which is 0.
143 static void upload_pipelined_state_pointers(struct brw_context
*brw
)
145 struct intel_context
*intel
= &brw
->intel
;
147 if (intel
->gen
== 5) {
148 /* Need to flush before changing clip max threads for errata. */
155 OUT_BATCH(_3DSTATE_PIPELINED_POINTERS
<< 16 | (7 - 2));
156 OUT_RELOC(intel
->batch
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
157 brw
->vs
.state_offset
);
158 if (brw
->gs
.prog_active
)
159 OUT_RELOC(brw
->intel
.batch
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
160 brw
->gs
.state_offset
| 1);
163 OUT_RELOC(brw
->intel
.batch
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
164 brw
->clip
.state_offset
| 1);
165 OUT_RELOC(brw
->intel
.batch
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
166 brw
->sf
.state_offset
);
167 OUT_RELOC(brw
->intel
.batch
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
168 brw
->wm
.state_offset
);
169 OUT_RELOC(brw
->intel
.batch
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
170 brw
->cc
.state_offset
);
173 brw
->state
.dirty
.brw
|= BRW_NEW_PSP
;
176 static void upload_psp_urb_cbs(struct brw_context
*brw
)
178 upload_pipelined_state_pointers(brw
);
179 brw_upload_urb_fence(brw
);
180 brw_upload_cs_urb_state(brw
);
183 const struct brw_tracked_state brw_psp_urb_cbs
= {
186 .brw
= (BRW_NEW_URB_FENCE
|
188 BRW_NEW_STATE_BASE_ADDRESS
),
189 .cache
= (CACHE_NEW_VS_UNIT
|
192 CACHE_NEW_CLIP_UNIT
|
197 .emit
= upload_psp_urb_cbs
,
201 brw_depthbuffer_format(struct brw_context
*brw
)
203 struct intel_context
*intel
= &brw
->intel
;
204 struct gl_context
*ctx
= &intel
->ctx
;
205 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
206 struct intel_renderbuffer
*drb
= intel_get_renderbuffer(fb
, BUFFER_DEPTH
);
207 struct intel_renderbuffer
*srb
;
210 (srb
= intel_get_renderbuffer(fb
, BUFFER_STENCIL
)) &&
211 !srb
->mt
->stencil_mt
&&
212 srb
->Base
.Format
== MESA_FORMAT_S8_Z24
) {
217 return BRW_DEPTHFORMAT_D32_FLOAT
;
219 switch (drb
->Base
.Format
) {
220 case MESA_FORMAT_Z16
:
221 return BRW_DEPTHFORMAT_D16_UNORM
;
222 case MESA_FORMAT_Z32_FLOAT
:
223 return BRW_DEPTHFORMAT_D32_FLOAT
;
224 case MESA_FORMAT_X8_Z24
:
225 return BRW_DEPTHFORMAT_D24_UNORM_X8_UINT
;
226 case MESA_FORMAT_S8_Z24
:
227 return BRW_DEPTHFORMAT_D24_UNORM_S8_UINT
;
229 _mesa_problem(ctx
, "Unexpected depth format %s\n",
230 _mesa_get_format_name(drb
->Base
.Format
));
231 return BRW_DEPTHFORMAT_D16_UNORM
;
235 static void emit_depthbuffer(struct brw_context
*brw
)
237 struct intel_context
*intel
= &brw
->intel
;
238 struct gl_context
*ctx
= &intel
->ctx
;
239 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
241 struct intel_renderbuffer
*depth_irb
= intel_get_renderbuffer(fb
, BUFFER_DEPTH
);
242 struct intel_renderbuffer
*stencil_irb
= intel_get_renderbuffer(fb
, BUFFER_STENCIL
);
243 struct intel_mipmap_tree
*stencil_mt
= NULL
;
244 struct intel_region
*hiz_region
= NULL
;
246 bool separate_stencil
= false;
250 depth_irb
->mt
->hiz_mt
) {
251 hiz_region
= depth_irb
->mt
->hiz_mt
->region
;
254 /* 3DSTATE_DEPTH_BUFFER, 3DSTATE_STENCIL_BUFFER are both
255 * non-pipelined state that will need the PIPE_CONTROL workaround.
257 if (intel
->gen
== 6) {
258 intel_emit_post_sync_nonzero_flush(intel
);
259 intel_emit_depth_stall_flushes(intel
);
262 /* Find the real separate stencil mt if present. */
264 stencil_mt
= stencil_irb
->mt
;
265 if (stencil_mt
->stencil_mt
)
266 stencil_mt
= stencil_mt
->stencil_mt
;
268 if (stencil_mt
->format
== MESA_FORMAT_S8
)
269 separate_stencil
= true;
272 /* If there's a packed depth/stencil bound to stencil only, we need to
273 * emit the packed depth/stencil buffer packet.
275 if (!depth_irb
&& stencil_irb
&& !separate_stencil
)
276 depth_irb
= stencil_irb
;
280 else if (intel
->is_g4x
|| intel
->gen
== 5)
285 if (!depth_irb
&& !separate_stencil
) {
287 OUT_BATCH(_3DSTATE_DEPTH_BUFFER
<< 16 | (len
- 2));
288 OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT
<< 18) |
289 (BRW_SURFACE_NULL
<< 29));
294 if (intel
->is_g4x
|| intel
->gen
>= 5)
302 } else if (!depth_irb
&& separate_stencil
) {
304 * There exists a separate stencil buffer but no depth buffer.
306 * The stencil buffer inherits most of its fields from
307 * 3DSTATE_DEPTH_BUFFER: namely the tile walk, surface type, width, and
310 * Since the stencil buffer has quirky pitch requirements, its region
311 * was allocated with half height and double cpp. So we need
312 * a multiplier of 2 to obtain the surface's real height.
314 * Enable the hiz bit because it and the separate stencil bit must have
315 * the same value. From Section 2.11.5.6.1.1 3DSTATE_DEPTH_BUFFER, Bit
316 * 1.21 "Separate Stencil Enable":
317 * [DevIL]: If this field is enabled, Hierarchical Depth Buffer
318 * Enable must also be enabled.
320 * [DevGT]: This field must be set to the same value (enabled or
321 * disabled) as Hierarchical Depth Buffer Enable
323 * The tiled bit must be set. From the Sandybridge PRM, Volume 2, Part 1,
324 * Section 7.5.5.1.1 3DSTATE_DEPTH_BUFFER, Bit 1.27 Tiled Surface:
325 * [DevGT+]: This field must be set to TRUE.
327 struct intel_region
*region
= stencil_mt
->region
;
329 assert(intel
->has_separate_stencil
);
332 OUT_BATCH(_3DSTATE_DEPTH_BUFFER
<< 16 | (len
- 2));
333 OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT
<< 18) |
334 (1 << 21) | /* separate stencil enable */
335 (1 << 22) | /* hiz enable */
336 (BRW_TILEWALK_YMAJOR
<< 26) |
337 (1 << 27) | /* tiled surface */
338 (BRW_SURFACE_2D
<< 29));
340 OUT_BATCH(((region
->width
- 1) << 6) |
341 (2 * region
->height
- 1) << 19);
351 struct intel_region
*region
= depth_irb
->mt
->region
;
352 uint32_t tile_x
, tile_y
, offset
;
354 /* If using separate stencil, hiz must be enabled. */
355 assert(!separate_stencil
|| hiz_region
);
357 offset
= intel_renderbuffer_tile_offsets(depth_irb
, &tile_x
, &tile_y
);
359 assert(intel
->gen
< 6 || region
->tiling
== I915_TILING_Y
);
360 assert(!hiz_region
|| region
->tiling
== I915_TILING_Y
);
363 OUT_BATCH(_3DSTATE_DEPTH_BUFFER
<< 16 | (len
- 2));
364 OUT_BATCH(((region
->pitch
* region
->cpp
) - 1) |
365 (brw_depthbuffer_format(brw
) << 18) |
366 ((hiz_region
? 1 : 0) << 21) | /* separate stencil enable */
367 ((hiz_region
? 1 : 0) << 22) | /* hiz enable */
368 (BRW_TILEWALK_YMAJOR
<< 26) |
369 ((region
->tiling
!= I915_TILING_NONE
) << 27) |
370 (BRW_SURFACE_2D
<< 29));
371 OUT_RELOC(region
->bo
,
372 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
374 OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW
<< 1) |
375 ((region
->width
- 1) << 6) |
376 ((region
->height
- 1) << 19));
379 if (intel
->is_g4x
|| intel
->gen
>= 5)
380 OUT_BATCH(tile_x
| (tile_y
<< 16));
382 assert(tile_x
== 0 && tile_y
== 0);
390 if (hiz_region
|| separate_stencil
) {
392 * In the 3DSTATE_DEPTH_BUFFER batch emitted above, the 'separate
393 * stencil enable' and 'hiz enable' bits were set. Therefore we must
394 * emit 3DSTATE_HIER_DEPTH_BUFFER and 3DSTATE_STENCIL_BUFFER. Even if
395 * there is no stencil buffer, 3DSTATE_STENCIL_BUFFER must be emitted;
396 * failure to do so causes hangs on gen5 and a stall on gen6.
399 /* Emit hiz buffer. */
402 OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER
<< 16) | (3 - 2));
403 OUT_BATCH(hiz_region
->pitch
* hiz_region
->cpp
- 1);
404 OUT_RELOC(hiz_region
->bo
,
405 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
410 OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER
<< 16) | (3 - 2));
416 /* Emit stencil buffer. */
417 if (separate_stencil
) {
418 struct intel_region
*region
= stencil_mt
->region
;
420 OUT_BATCH((_3DSTATE_STENCIL_BUFFER
<< 16) | (3 - 2));
421 OUT_BATCH(region
->pitch
* region
->cpp
- 1);
422 OUT_RELOC(region
->bo
,
423 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
428 OUT_BATCH((_3DSTATE_STENCIL_BUFFER
<< 16) | (3 - 2));
436 * On Gen >= 6, emit clear params for safety. If using hiz, then clear
437 * params must be emitted.
439 * From Section 2.11.5.6.4.1 3DSTATE_CLEAR_PARAMS:
440 * 3DSTATE_CLEAR_PARAMS packet must follow the DEPTH_BUFFER_STATE packet
441 * when HiZ is enabled and the DEPTH_BUFFER_STATE changes.
443 if (intel
->gen
>= 6 || hiz_region
) {
445 intel_emit_post_sync_nonzero_flush(intel
);
448 OUT_BATCH(_3DSTATE_CLEAR_PARAMS
<< 16 | (2 - 2));
454 const struct brw_tracked_state brw_depthbuffer
= {
456 .mesa
= _NEW_BUFFERS
,
457 .brw
= BRW_NEW_BATCH
,
460 .emit
= emit_depthbuffer
,
465 /***********************************************************************
466 * Polygon stipple packet
469 static void upload_polygon_stipple(struct brw_context
*brw
)
471 struct intel_context
*intel
= &brw
->intel
;
472 struct gl_context
*ctx
= &brw
->intel
.ctx
;
476 if (!ctx
->Polygon
.StippleFlag
)
480 intel_emit_post_sync_nonzero_flush(intel
);
483 OUT_BATCH(_3DSTATE_POLY_STIPPLE_PATTERN
<< 16 | (33 - 2));
485 /* Polygon stipple is provided in OpenGL order, i.e. bottom
486 * row first. If we're rendering to a window (i.e. the
487 * default frame buffer object, 0), then we need to invert
488 * it to match our pixel layout. But if we're rendering
489 * to a FBO (i.e. any named frame buffer object), we *don't*
490 * need to invert - we already match the layout.
492 if (ctx
->DrawBuffer
->Name
== 0) {
493 for (i
= 0; i
< 32; i
++)
494 OUT_BATCH(ctx
->PolygonStipple
[31 - i
]); /* invert */
497 for (i
= 0; i
< 32; i
++)
498 OUT_BATCH(ctx
->PolygonStipple
[i
]);
503 const struct brw_tracked_state brw_polygon_stipple
= {
505 .mesa
= (_NEW_POLYGONSTIPPLE
|
507 .brw
= BRW_NEW_CONTEXT
,
510 .emit
= upload_polygon_stipple
514 /***********************************************************************
515 * Polygon stipple offset packet
518 static void upload_polygon_stipple_offset(struct brw_context
*brw
)
520 struct intel_context
*intel
= &brw
->intel
;
521 struct gl_context
*ctx
= &brw
->intel
.ctx
;
524 if (!ctx
->Polygon
.StippleFlag
)
528 intel_emit_post_sync_nonzero_flush(intel
);
531 OUT_BATCH(_3DSTATE_POLY_STIPPLE_OFFSET
<< 16 | (2-2));
535 * If we're drawing to a system window (ctx->DrawBuffer->Name == 0),
536 * we have to invert the Y axis in order to match the OpenGL
537 * pixel coordinate system, and our offset must be matched
538 * to the window position. If we're drawing to a FBO
539 * (ctx->DrawBuffer->Name != 0), then our native pixel coordinate
540 * system works just fine, and there's no window system to
543 if (brw
->intel
.ctx
.DrawBuffer
->Name
== 0)
544 OUT_BATCH((32 - (ctx
->DrawBuffer
->Height
& 31)) & 31);
550 const struct brw_tracked_state brw_polygon_stipple_offset
= {
552 .mesa
= (_NEW_BUFFERS
|
554 .brw
= BRW_NEW_CONTEXT
,
557 .emit
= upload_polygon_stipple_offset
560 /**********************************************************************
563 static void upload_aa_line_parameters(struct brw_context
*brw
)
565 struct intel_context
*intel
= &brw
->intel
;
566 struct gl_context
*ctx
= &brw
->intel
.ctx
;
568 if (!ctx
->Line
.SmoothFlag
|| !brw
->has_aa_line_parameters
)
572 intel_emit_post_sync_nonzero_flush(intel
);
574 OUT_BATCH(_3DSTATE_AA_LINE_PARAMETERS
<< 16 | (3 - 2));
575 /* use legacy aa line coverage computation */
581 const struct brw_tracked_state brw_aa_line_parameters
= {
584 .brw
= BRW_NEW_CONTEXT
,
587 .emit
= upload_aa_line_parameters
590 /***********************************************************************
591 * Line stipple packet
594 static void upload_line_stipple(struct brw_context
*brw
)
596 struct intel_context
*intel
= &brw
->intel
;
597 struct gl_context
*ctx
= &brw
->intel
.ctx
;
601 if (!ctx
->Line
.StippleFlag
)
605 intel_emit_post_sync_nonzero_flush(intel
);
608 OUT_BATCH(_3DSTATE_LINE_STIPPLE_PATTERN
<< 16 | (3 - 2));
609 OUT_BATCH(ctx
->Line
.StipplePattern
);
610 tmp
= 1.0 / (GLfloat
) ctx
->Line
.StippleFactor
;
611 tmpi
= tmp
* (1<<13);
612 OUT_BATCH(tmpi
<< 16 | ctx
->Line
.StippleFactor
);
616 const struct brw_tracked_state brw_line_stipple
= {
619 .brw
= BRW_NEW_CONTEXT
,
622 .emit
= upload_line_stipple
626 /***********************************************************************
627 * Misc invarient state packets
630 static void upload_invarient_state( struct brw_context
*brw
)
632 struct intel_context
*intel
= &brw
->intel
;
634 /* 3DSTATE_SIP, 3DSTATE_MULTISAMPLE, etc. are nonpipelined. */
636 intel_emit_post_sync_nonzero_flush(intel
);
638 /* Select the 3D pipeline (as opposed to media) */
640 OUT_BATCH(brw
->CMD_PIPELINE_SELECT
<< 16 | 0);
643 if (intel
->gen
< 6) {
644 /* Disable depth offset clamping. */
646 OUT_BATCH(_3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP
<< 16 | (2 - 2));
651 if (intel
->gen
>= 6) {
653 int len
= intel
->gen
>= 7 ? 4 : 3;
656 OUT_BATCH(_3DSTATE_MULTISAMPLE
<< 16 | (len
- 2));
657 OUT_BATCH(MS_PIXEL_LOCATION_CENTER
|
659 OUT_BATCH(0); /* positions for 4/8-sample */
665 OUT_BATCH(_3DSTATE_SAMPLE_MASK
<< 16 | (2 - 2));
669 if (intel
->gen
< 7) {
670 for (i
= 0; i
< 4; i
++) {
672 OUT_BATCH(_3DSTATE_GS_SVB_INDEX
<< 16 | (4 - 2));
673 OUT_BATCH(i
<< SVB_INDEX_SHIFT
);
675 OUT_BATCH(0xffffffff);
682 OUT_BATCH(CMD_STATE_SIP
<< 16 | (2 - 2));
687 OUT_BATCH(brw
->CMD_VF_STATISTICS
<< 16 |
688 (unlikely(INTEL_DEBUG
& DEBUG_STATS
) ? 1 : 0));
692 const struct brw_tracked_state brw_invarient_state
= {
695 .brw
= BRW_NEW_CONTEXT
,
698 .emit
= upload_invarient_state
702 * Define the base addresses which some state is referenced from.
704 * This allows us to avoid having to emit relocations for the objects,
705 * and is actually required for binding table pointers on gen6.
707 * Surface state base address covers binding table pointers and
708 * surface state objects, but not the surfaces that the surface state
711 static void upload_state_base_address( struct brw_context
*brw
)
713 struct intel_context
*intel
= &brw
->intel
;
715 /* FINISHME: According to section 3.6.1 "STATE_BASE_ADDRESS" of
716 * vol1a of the G45 PRM, MI_FLUSH with the ISC invalidate should be
717 * programmed prior to STATE_BASE_ADDRESS.
719 * However, given that the instruction SBA (general state base
720 * address) on this chipset is always set to 0 across X and GL,
721 * maybe this isn't required for us in particular.
724 if (intel
->gen
>= 6) {
726 intel_emit_post_sync_nonzero_flush(intel
);
729 OUT_BATCH(CMD_STATE_BASE_ADDRESS
<< 16 | (10 - 2));
730 /* General state base address: stateless DP read/write requests */
732 /* Surface state base address:
733 * BINDING_TABLE_STATE
736 OUT_RELOC(intel
->batch
.bo
, I915_GEM_DOMAIN_SAMPLER
, 0, 1);
737 /* Dynamic state base address:
739 * SAMPLER_BORDER_COLOR_STATE
740 * CLIP, SF, WM/CC viewport state
742 * DEPTH_STENCIL_STATE
744 * Push constants (when INSTPM: CONSTANT_BUFFER Address Offset
745 * Disable is clear, which we rely on)
747 OUT_RELOC(intel
->batch
.bo
, (I915_GEM_DOMAIN_RENDER
|
748 I915_GEM_DOMAIN_INSTRUCTION
), 0, 1);
750 OUT_BATCH(1); /* Indirect object base address: MEDIA_OBJECT data */
751 OUT_RELOC(brw
->cache
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
752 1); /* Instruction base address: shader kernels (incl. SIP) */
754 OUT_BATCH(1); /* General state upper bound */
755 OUT_BATCH(1); /* Dynamic state upper bound */
756 OUT_BATCH(1); /* Indirect object upper bound */
757 OUT_BATCH(1); /* Instruction access upper bound */
759 } else if (intel
->gen
== 5) {
761 OUT_BATCH(CMD_STATE_BASE_ADDRESS
<< 16 | (8 - 2));
762 OUT_BATCH(1); /* General state base address */
763 OUT_RELOC(intel
->batch
.bo
, I915_GEM_DOMAIN_SAMPLER
, 0,
764 1); /* Surface state base address */
765 OUT_BATCH(1); /* Indirect object base address */
766 OUT_RELOC(brw
->cache
.bo
, I915_GEM_DOMAIN_INSTRUCTION
, 0,
767 1); /* Instruction base address */
768 OUT_BATCH(1); /* General state upper bound */
769 OUT_BATCH(1); /* Indirect object upper bound */
770 OUT_BATCH(1); /* Instruction access upper bound */
774 OUT_BATCH(CMD_STATE_BASE_ADDRESS
<< 16 | (6 - 2));
775 OUT_BATCH(1); /* General state base address */
776 OUT_RELOC(intel
->batch
.bo
, I915_GEM_DOMAIN_SAMPLER
, 0,
777 1); /* Surface state base address */
778 OUT_BATCH(1); /* Indirect object base address */
779 OUT_BATCH(1); /* General state upper bound */
780 OUT_BATCH(1); /* Indirect object upper bound */
784 /* According to section 3.6.1 of VOL1 of the 965 PRM,
785 * STATE_BASE_ADDRESS updates require a reissue of:
787 * 3DSTATE_PIPELINE_POINTERS
788 * 3DSTATE_BINDING_TABLE_POINTERS
789 * MEDIA_STATE_POINTERS
791 * and this continues through Ironlake. The Sandy Bridge PRM, vol
792 * 1 part 1 says that the folowing packets must be reissued:
794 * 3DSTATE_CC_POINTERS
795 * 3DSTATE_BINDING_TABLE_POINTERS
796 * 3DSTATE_SAMPLER_STATE_POINTERS
797 * 3DSTATE_VIEWPORT_STATE_POINTERS
798 * MEDIA_STATE_POINTERS
800 * Those are always reissued following SBA updates anyway (new
801 * batch time), except in the case of the program cache BO
802 * changing. Having a separate state flag makes the sequence more
806 brw
->state
.dirty
.brw
|= BRW_NEW_STATE_BASE_ADDRESS
;
809 const struct brw_tracked_state brw_state_base_address
= {
812 .brw
= (BRW_NEW_BATCH
|
813 BRW_NEW_PROGRAM_CACHE
),
816 .emit
= upload_state_base_address