2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics 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 <keithw@vmware.com>
33 #ifndef BRWCONTEXT_INC
34 #define BRWCONTEXT_INC
37 #include "main/macros.h"
38 #include "main/mtypes.h"
39 #include "brw_structs.h"
40 #include "brw_compiler.h"
41 #include "intel_aub.h"
45 /* Evil hack for using libdrm in a c++ compiler. */
49 #include <intel_bufmgr.h>
58 #include "intel_debug.h"
59 #include "intel_screen.h"
60 #include "intel_tex_obj.h"
61 #include "intel_resolve_map.h"
65 * URB - uniform resource buffer. A mid-sized buffer which is
66 * partitioned between the fixed function units and used for passing
67 * values (vertices, primitives, constants) between them.
69 * CURBE - constant URB entry. An urb region (entry) used to hold
70 * constant values which the fixed function units can be instructed to
71 * preload into the GRF when spawning a thread.
73 * VUE - vertex URB entry. An urb entry holding a vertex and usually
74 * a vertex header. The header contains control information and
75 * things like primitive type, Begin/end flags and clip codes.
77 * PUE - primitive URB entry. An urb entry produced by the setup (SF)
78 * unit holding rasterization and interpolation parameters.
80 * GRF - general register file. One of several register files
81 * addressable by programmed threads. The inputs (r0, payload, curbe,
82 * urb) of the thread are preloaded to this area before the thread is
83 * spawned. The registers are individually 8 dwords wide and suitable
84 * for general usage. Registers holding thread input values are not
85 * special and may be overwritten.
87 * MRF - message register file. Threads communicate (and terminate)
88 * by sending messages. Message parameters are placed in contiguous
89 * MRF registers. All program output is via these messages. URB
90 * entries are populated by sending a message to the shared URB
91 * function containing the new data, together with a control word,
92 * often an unmodified copy of R0.
94 * R0 - GRF register 0. Typically holds control information used when
95 * sending messages to other threads.
97 * EU or GEN4 EU: The name of the programmable subsystem of the
98 * i965 hardware. Threads are executed by the EU, the registers
99 * described above are part of the EU architecture.
101 * Fixed function units:
103 * CS - Command streamer. Notional first unit, little software
104 * interaction. Holds the URB entries used for constant data, ie the
107 * VF/VS - Vertex Fetch / Vertex Shader. The fixed function part of
108 * this unit is responsible for pulling vertices out of vertex buffers
109 * in vram and injecting them into the processing pipe as VUEs. If
110 * enabled, it first passes them to a VS thread which is a good place
111 * for the driver to implement any active vertex shader.
113 * HS - Hull Shader (Tessellation Control Shader)
115 * TE - Tessellation Engine (Tessellation Primitive Generation)
117 * DS - Domain Shader (Tessellation Evaluation Shader)
119 * GS - Geometry Shader. This corresponds to a new DX10 concept. If
120 * enabled, incoming strips etc are passed to GS threads in individual
121 * line/triangle/point units. The GS thread may perform arbitary
122 * computation and emit whatever primtives with whatever vertices it
123 * chooses. This makes GS an excellent place to implement GL's
124 * unfilled polygon modes, though of course it is capable of much
125 * more. Additionally, GS is used to translate away primitives not
126 * handled by latter units, including Quads and Lineloops.
128 * CS - Clipper. Mesa's clipping algorithms are imported to run on
129 * this unit. The fixed function part performs cliptesting against
130 * the 6 fixed clipplanes and makes descisions on whether or not the
131 * incoming primitive needs to be passed to a thread for clipping.
132 * User clip planes are handled via cooperation with the VS thread.
134 * SF - Strips Fans or Setup: Triangles are prepared for
135 * rasterization. Interpolation coefficients are calculated.
136 * Flatshading and two-side lighting usually performed here.
138 * WM - Windower. Interpolation of vertex attributes performed here.
139 * Fragment shader implemented here. SIMD aspects of EU taken full
140 * advantage of, as pixels are processed in blocks of 16.
142 * CC - Color Calculator. No EU threads associated with this unit.
143 * Handles blending and (presumably) depth and stencil testing.
148 struct brw_vs_prog_key
;
149 struct brw_vue_prog_key
;
150 struct brw_wm_prog_key
;
151 struct brw_wm_prog_data
;
152 struct brw_cs_prog_key
;
153 struct brw_cs_prog_data
;
157 BRW_COMPUTE_PIPELINE
,
164 BRW_CACHE_BLORP_BLIT_PROG
,
167 BRW_CACHE_FF_GS_PROG
,
178 /* brw_cache_ids must come first - see brw_state_cache.c */
179 BRW_STATE_URB_FENCE
= BRW_MAX_CACHE
,
180 BRW_STATE_FRAGMENT_PROGRAM
,
181 BRW_STATE_GEOMETRY_PROGRAM
,
182 BRW_STATE_TESS_PROGRAMS
,
183 BRW_STATE_VERTEX_PROGRAM
,
184 BRW_STATE_CURBE_OFFSETS
,
185 BRW_STATE_REDUCED_PRIMITIVE
,
186 BRW_STATE_PATCH_PRIMITIVE
,
191 BRW_STATE_BINDING_TABLE_POINTERS
,
194 BRW_STATE_DEFAULT_TESS_LEVELS
,
196 BRW_STATE_INDEX_BUFFER
,
197 BRW_STATE_VS_CONSTBUF
,
198 BRW_STATE_TCS_CONSTBUF
,
199 BRW_STATE_TES_CONSTBUF
,
200 BRW_STATE_GS_CONSTBUF
,
201 BRW_STATE_PROGRAM_CACHE
,
202 BRW_STATE_STATE_BASE_ADDRESS
,
203 BRW_STATE_VUE_MAP_GEOM_OUT
,
204 BRW_STATE_TRANSFORM_FEEDBACK
,
205 BRW_STATE_RASTERIZER_DISCARD
,
207 BRW_STATE_UNIFORM_BUFFER
,
208 BRW_STATE_ATOMIC_BUFFER
,
209 BRW_STATE_IMAGE_UNITS
,
210 BRW_STATE_META_IN_PROGRESS
,
211 BRW_STATE_INTERPOLATION_MAP
,
212 BRW_STATE_PUSH_CONSTANT_ALLOCATION
,
213 BRW_STATE_NUM_SAMPLES
,
214 BRW_STATE_TEXTURE_BUFFER
,
215 BRW_STATE_GEN4_UNIT_STATE
,
219 BRW_STATE_SAMPLER_STATE_TABLE
,
220 BRW_STATE_VS_ATTRIB_WORKAROUNDS
,
221 BRW_STATE_COMPUTE_PROGRAM
,
222 BRW_STATE_CS_WORK_GROUPS
,
228 * BRW_NEW_*_PROG_DATA and BRW_NEW_*_PROGRAM are similar, but distinct.
230 * BRW_NEW_*_PROGRAM relates to the gl_shader_program/gl_program structures.
231 * When the currently bound shader program differs from the previous draw
232 * call, these will be flagged. They cover brw->{stage}_program and
233 * ctx->{Stage}Program->_Current.
235 * BRW_NEW_*_PROG_DATA is flagged when the effective shaders change, from a
236 * driver perspective. Even if the same shader is bound at the API level,
237 * we may need to switch between multiple versions of that shader to handle
238 * changes in non-orthagonal state.
240 * Additionally, multiple shader programs may have identical vertex shaders
241 * (for example), or compile down to the same code in the backend. We combine
242 * those into a single program cache entry.
244 * BRW_NEW_*_PROG_DATA occurs when switching program cache entries, which
245 * covers the brw_*_prog_data structures, and brw->*.prog_offset.
247 #define BRW_NEW_FS_PROG_DATA (1ull << BRW_CACHE_FS_PROG)
248 /* XXX: The BRW_NEW_BLORP_BLIT_PROG_DATA dirty bit is unused (as BLORP doesn't
249 * use the normal state upload paths), but the cache is still used. To avoid
250 * polluting the brw_state_cache code with special cases, we retain the dirty
251 * bit for now. It should eventually be removed.
253 #define BRW_NEW_BLORP_BLIT_PROG_DATA (1ull << BRW_CACHE_BLORP_BLIT_PROG)
254 #define BRW_NEW_SF_PROG_DATA (1ull << BRW_CACHE_SF_PROG)
255 #define BRW_NEW_VS_PROG_DATA (1ull << BRW_CACHE_VS_PROG)
256 #define BRW_NEW_FF_GS_PROG_DATA (1ull << BRW_CACHE_FF_GS_PROG)
257 #define BRW_NEW_GS_PROG_DATA (1ull << BRW_CACHE_GS_PROG)
258 #define BRW_NEW_TCS_PROG_DATA (1ull << BRW_CACHE_TCS_PROG)
259 #define BRW_NEW_TES_PROG_DATA (1ull << BRW_CACHE_TES_PROG)
260 #define BRW_NEW_CLIP_PROG_DATA (1ull << BRW_CACHE_CLIP_PROG)
261 #define BRW_NEW_CS_PROG_DATA (1ull << BRW_CACHE_CS_PROG)
262 #define BRW_NEW_URB_FENCE (1ull << BRW_STATE_URB_FENCE)
263 #define BRW_NEW_FRAGMENT_PROGRAM (1ull << BRW_STATE_FRAGMENT_PROGRAM)
264 #define BRW_NEW_GEOMETRY_PROGRAM (1ull << BRW_STATE_GEOMETRY_PROGRAM)
265 #define BRW_NEW_TESS_PROGRAMS (1ull << BRW_STATE_TESS_PROGRAMS)
266 #define BRW_NEW_VERTEX_PROGRAM (1ull << BRW_STATE_VERTEX_PROGRAM)
267 #define BRW_NEW_CURBE_OFFSETS (1ull << BRW_STATE_CURBE_OFFSETS)
268 #define BRW_NEW_REDUCED_PRIMITIVE (1ull << BRW_STATE_REDUCED_PRIMITIVE)
269 #define BRW_NEW_PATCH_PRIMITIVE (1ull << BRW_STATE_PATCH_PRIMITIVE)
270 #define BRW_NEW_PRIMITIVE (1ull << BRW_STATE_PRIMITIVE)
271 #define BRW_NEW_CONTEXT (1ull << BRW_STATE_CONTEXT)
272 #define BRW_NEW_PSP (1ull << BRW_STATE_PSP)
273 #define BRW_NEW_SURFACES (1ull << BRW_STATE_SURFACES)
274 #define BRW_NEW_BINDING_TABLE_POINTERS (1ull << BRW_STATE_BINDING_TABLE_POINTERS)
275 #define BRW_NEW_INDICES (1ull << BRW_STATE_INDICES)
276 #define BRW_NEW_VERTICES (1ull << BRW_STATE_VERTICES)
277 #define BRW_NEW_DEFAULT_TESS_LEVELS (1ull << BRW_STATE_DEFAULT_TESS_LEVELS)
279 * Used for any batch entry with a relocated pointer that will be used
280 * by any 3D rendering.
282 #define BRW_NEW_BATCH (1ull << BRW_STATE_BATCH)
283 /** \see brw.state.depth_region */
284 #define BRW_NEW_INDEX_BUFFER (1ull << BRW_STATE_INDEX_BUFFER)
285 #define BRW_NEW_VS_CONSTBUF (1ull << BRW_STATE_VS_CONSTBUF)
286 #define BRW_NEW_TCS_CONSTBUF (1ull << BRW_STATE_TCS_CONSTBUF)
287 #define BRW_NEW_TES_CONSTBUF (1ull << BRW_STATE_TES_CONSTBUF)
288 #define BRW_NEW_GS_CONSTBUF (1ull << BRW_STATE_GS_CONSTBUF)
289 #define BRW_NEW_PROGRAM_CACHE (1ull << BRW_STATE_PROGRAM_CACHE)
290 #define BRW_NEW_STATE_BASE_ADDRESS (1ull << BRW_STATE_STATE_BASE_ADDRESS)
291 #define BRW_NEW_VUE_MAP_GEOM_OUT (1ull << BRW_STATE_VUE_MAP_GEOM_OUT)
292 #define BRW_NEW_TRANSFORM_FEEDBACK (1ull << BRW_STATE_TRANSFORM_FEEDBACK)
293 #define BRW_NEW_RASTERIZER_DISCARD (1ull << BRW_STATE_RASTERIZER_DISCARD)
294 #define BRW_NEW_STATS_WM (1ull << BRW_STATE_STATS_WM)
295 #define BRW_NEW_UNIFORM_BUFFER (1ull << BRW_STATE_UNIFORM_BUFFER)
296 #define BRW_NEW_ATOMIC_BUFFER (1ull << BRW_STATE_ATOMIC_BUFFER)
297 #define BRW_NEW_IMAGE_UNITS (1ull << BRW_STATE_IMAGE_UNITS)
298 #define BRW_NEW_META_IN_PROGRESS (1ull << BRW_STATE_META_IN_PROGRESS)
299 #define BRW_NEW_INTERPOLATION_MAP (1ull << BRW_STATE_INTERPOLATION_MAP)
300 #define BRW_NEW_PUSH_CONSTANT_ALLOCATION (1ull << BRW_STATE_PUSH_CONSTANT_ALLOCATION)
301 #define BRW_NEW_NUM_SAMPLES (1ull << BRW_STATE_NUM_SAMPLES)
302 #define BRW_NEW_TEXTURE_BUFFER (1ull << BRW_STATE_TEXTURE_BUFFER)
303 #define BRW_NEW_GEN4_UNIT_STATE (1ull << BRW_STATE_GEN4_UNIT_STATE)
304 #define BRW_NEW_CC_VP (1ull << BRW_STATE_CC_VP)
305 #define BRW_NEW_SF_VP (1ull << BRW_STATE_SF_VP)
306 #define BRW_NEW_CLIP_VP (1ull << BRW_STATE_CLIP_VP)
307 #define BRW_NEW_SAMPLER_STATE_TABLE (1ull << BRW_STATE_SAMPLER_STATE_TABLE)
308 #define BRW_NEW_VS_ATTRIB_WORKAROUNDS (1ull << BRW_STATE_VS_ATTRIB_WORKAROUNDS)
309 #define BRW_NEW_COMPUTE_PROGRAM (1ull << BRW_STATE_COMPUTE_PROGRAM)
310 #define BRW_NEW_CS_WORK_GROUPS (1ull << BRW_STATE_CS_WORK_GROUPS)
311 #define BRW_NEW_URB_SIZE (1ull << BRW_STATE_URB_SIZE)
313 struct brw_state_flags
{
314 /** State update flags signalled by mesa internals */
317 * State update flags signalled as the result of brw_tracked_state updates
322 /** Subclass of Mesa vertex program */
323 struct brw_vertex_program
{
324 struct gl_vertex_program program
;
329 /** Subclass of Mesa tessellation control program */
330 struct brw_tess_ctrl_program
{
331 struct gl_tess_ctrl_program program
;
332 unsigned id
; /**< serial no. to identify tess ctrl progs, never re-used */
336 /** Subclass of Mesa tessellation evaluation program */
337 struct brw_tess_eval_program
{
338 struct gl_tess_eval_program program
;
339 unsigned id
; /**< serial no. to identify tess eval progs, never re-used */
343 /** Subclass of Mesa geometry program */
344 struct brw_geometry_program
{
345 struct gl_geometry_program program
;
346 unsigned id
; /**< serial no. to identify geom progs, never re-used */
350 /** Subclass of Mesa fragment program */
351 struct brw_fragment_program
{
352 struct gl_fragment_program program
;
353 GLuint id
; /**< serial no. to identify frag progs, never re-used */
357 /** Subclass of Mesa compute program */
358 struct brw_compute_program
{
359 struct gl_compute_program program
;
360 unsigned id
; /**< serial no. to identify compute progs, never re-used */
365 struct gl_shader base
;
371 * Bitmask indicating which fragment shader inputs represent varyings (and
372 * hence have to be delivered to the fragment shader by the SF/SBE stage).
374 #define BRW_FS_VARYING_INPUT_MASK \
375 (BITFIELD64_RANGE(0, VARYING_SLOT_MAX) & \
376 ~VARYING_BIT_POS & ~VARYING_BIT_FACE)
380 * Mapping of VUE map slots to interpolation modes.
382 struct interpolation_mode_map
{
383 unsigned char mode
[BRW_VARYING_SLOT_COUNT
];
386 static inline bool brw_any_flat_varyings(struct interpolation_mode_map
*map
)
388 for (int i
= 0; i
< BRW_VARYING_SLOT_COUNT
; i
++)
389 if (map
->mode
[i
] == INTERP_QUALIFIER_FLAT
)
395 static inline bool brw_any_noperspective_varyings(struct interpolation_mode_map
*map
)
397 for (int i
= 0; i
< BRW_VARYING_SLOT_COUNT
; i
++)
398 if (map
->mode
[i
] == INTERP_QUALIFIER_NOPERSPECTIVE
)
405 struct brw_sf_prog_data
{
406 GLuint urb_read_length
;
409 /* Each vertex may have upto 12 attributes, 4 components each,
410 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
413 * Actually we use 4 for each, so call it 12 rows.
415 GLuint urb_entry_size
;
420 * We always program SF to start reading at an offset of 1 (2 varying slots)
421 * from the start of the vertex URB entry. This causes it to skip:
422 * - VARYING_SLOT_PSIZ and BRW_VARYING_SLOT_NDC on gen4-5
423 * - VARYING_SLOT_PSIZ and VARYING_SLOT_POS on gen6+
425 #define BRW_SF_URB_ENTRY_READ_OFFSET 1
428 struct brw_clip_prog_data
{
429 GLuint curb_read_length
; /* user planes? */
431 GLuint urb_read_length
;
435 struct brw_ff_gs_prog_data
{
436 GLuint urb_read_length
;
440 * Gen6 transform feedback: Amount by which the streaming vertex buffer
441 * indices should be incremented each time the GS is invoked.
443 unsigned svbi_postincrement_value
;
446 /** Number of texture sampler units */
447 #define BRW_MAX_TEX_UNIT 32
449 /** Max number of render targets in a shader */
450 #define BRW_MAX_DRAW_BUFFERS 8
452 /** Max number of UBOs in a shader */
453 #define BRW_MAX_UBO 14
455 /** Max number of SSBOs in a shader */
456 #define BRW_MAX_SSBO 12
458 /** Max number of atomic counter buffer objects in a shader */
459 #define BRW_MAX_ABO 16
461 /** Max number of image uniforms in a shader */
462 #define BRW_MAX_IMAGES 32
465 * Max number of binding table entries used for stream output.
467 * From the OpenGL 3.0 spec, table 6.44 (Transform Feedback State), the
468 * minimum value of MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS is 64.
470 * On Gen6, the size of transform feedback data is limited not by the number
471 * of components but by the number of binding table entries we set aside. We
472 * use one binding table entry for a float, one entry for a vector, and one
473 * entry per matrix column. Since the only way we can communicate our
474 * transform feedback capabilities to the client is via
475 * MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS, we need to plan for the
476 * worst case, in which all the varyings are floats, so we use up one binding
477 * table entry per component. Therefore we need to set aside at least 64
478 * binding table entries for use by transform feedback.
480 * Note: since we don't currently pack varyings, it is currently impossible
481 * for the client to actually use up all of these binding table entries--if
482 * all of their varyings were floats, they would run out of varying slots and
483 * fail to link. But that's a bug, so it seems prudent to go ahead and
484 * allocate the number of binding table entries we will need once the bug is
487 #define BRW_MAX_SOL_BINDINGS 64
489 /** Maximum number of actual buffers used for stream output */
490 #define BRW_MAX_SOL_BUFFERS 4
492 #define BRW_MAX_SURFACES (BRW_MAX_DRAW_BUFFERS + \
493 BRW_MAX_TEX_UNIT * 2 + /* normal, gather */ \
498 2 + /* shader time, pull constants */ \
499 1 /* cs num work groups */)
501 #define SURF_INDEX_GEN6_SOL_BINDING(t) (t)
504 * Stride in bytes between shader_time entries.
506 * We separate entries by a cacheline to reduce traffic between EUs writing to
509 #define SHADER_TIME_STRIDE 64
511 struct brw_cache_item
{
513 * Effectively part of the key, cache_id identifies what kind of state
514 * buffer is involved, and also which dirty flag should set.
516 enum brw_cache_id cache_id
;
517 /** 32-bit hash of the key data */
519 GLuint key_size
; /* for variable-sized keys */
526 struct brw_cache_item
*next
;
531 struct brw_context
*brw
;
533 struct brw_cache_item
**items
;
535 GLuint size
, n_items
;
537 uint32_t next_offset
;
542 /* Considered adding a member to this struct to document which flags
543 * an update might raise so that ordering of the state atoms can be
544 * checked or derived at runtime. Dropped the idea in favor of having
545 * a debug mode where the state is monitored for flags which are
546 * raised that have already been tested against.
548 struct brw_tracked_state
{
549 struct brw_state_flags dirty
;
550 void (*emit
)( struct brw_context
*brw
);
553 enum shader_time_shader_type
{
564 struct brw_vertex_buffer
{
565 /** Buffer object containing the uploaded vertex data */
568 /** Byte stride between elements in the uploaded array */
572 struct brw_vertex_element
{
573 const struct gl_client_array
*glarray
;
577 /** Offset of the first element within the buffer object */
581 struct brw_query_object
{
582 struct gl_query_object Base
;
584 /** Last query BO associated with this query. */
587 /** Last index in bo with query data for this object. */
590 /** True if we know the batch has been flushed since we ended the query. */
600 struct intel_batchbuffer
{
601 /** Current batchbuffer being queued up. */
603 /** Last BO submitted to the hardware. Used for glFinish(). */
604 drm_intel_bo
*last_bo
;
607 uint16_t emit
, total
;
609 uint16_t reserved_space
;
613 #define BATCH_SZ (8192*sizeof(uint32_t))
615 uint32_t state_batch_offset
;
616 enum brw_gpu_ring ring
;
617 bool needs_sol_reset
;
625 #define MAX_GS_INPUT_VERTICES 6
627 #define BRW_MAX_XFB_STREAMS 4
629 struct brw_transform_feedback_object
{
630 struct gl_transform_feedback_object base
;
632 /** A buffer to hold SO_WRITE_OFFSET(n) values while paused. */
633 drm_intel_bo
*offset_bo
;
635 /** If true, SO_WRITE_OFFSET(n) should be reset to zero at next use. */
638 /** The most recent primitive mode (GL_TRIANGLES/GL_POINTS/GL_LINES). */
639 GLenum primitive_mode
;
642 * Count of primitives generated during this transform feedback operation.
645 uint64_t prims_generated
[BRW_MAX_XFB_STREAMS
];
646 drm_intel_bo
*prim_count_bo
;
647 unsigned prim_count_buffer_index
; /**< in number of uint64_t units */
651 * Number of vertices written between last Begin/EndTransformFeedback().
653 * Used to implement DrawTransformFeedback().
655 uint64_t vertices_written
[BRW_MAX_XFB_STREAMS
];
656 bool vertices_written_valid
;
660 * Data shared between each programmable stage in the pipeline (vs, gs, and
663 struct brw_stage_state
665 gl_shader_stage stage
;
666 struct brw_stage_prog_data
*prog_data
;
669 * Optional scratch buffer used to store spilled register values and
670 * variably-indexed GRF arrays.
672 drm_intel_bo
*scratch_bo
;
674 /** Offset in the program cache to the program */
675 uint32_t prog_offset
;
677 /** Offset in the batchbuffer to Gen4-5 pipelined state (VS/WM/GS_STATE). */
678 uint32_t state_offset
;
680 uint32_t push_const_offset
; /* Offset in the batchbuffer */
681 int push_const_size
; /* in 256-bit register increments */
683 /* Binding table: pointers to SURFACE_STATE entries. */
684 uint32_t bind_bo_offset
;
685 uint32_t surf_offset
[BRW_MAX_SURFACES
];
687 /** SAMPLER_STATE count and table offset */
688 uint32_t sampler_count
;
689 uint32_t sampler_offset
;
692 enum brw_predicate_state
{
693 /* The first two states are used if we can determine whether to draw
694 * without having to look at the values in the query object buffer. This
695 * will happen if there is no conditional render in progress, if the query
696 * object is already completed or if something else has already added
697 * samples to the preliminary result such as via a BLT command.
699 BRW_PREDICATE_STATE_RENDER
,
700 BRW_PREDICATE_STATE_DONT_RENDER
,
701 /* In this case whether to draw or not depends on the result of an
702 * MI_PREDICATE command so the predicate enable bit needs to be checked.
704 BRW_PREDICATE_STATE_USE_BIT
709 struct brw_l3_config
;
712 * brw_context is derived from gl_context.
716 struct gl_context ctx
; /**< base class, must be first field */
720 void (*update_texture_surface
)(struct gl_context
*ctx
,
722 uint32_t *surf_offset
,
724 uint32_t (*update_renderbuffer_surface
)(struct brw_context
*brw
,
725 struct gl_renderbuffer
*rb
,
726 bool layered
, unsigned unit
,
727 uint32_t surf_index
);
729 void (*emit_texture_surface_state
)(struct brw_context
*brw
,
730 struct intel_mipmap_tree
*mt
,
738 uint32_t *surf_offset
,
739 bool rw
, bool for_gather
);
740 void (*emit_buffer_surface_state
)(struct brw_context
*brw
,
741 uint32_t *out_offset
,
743 unsigned buffer_offset
,
744 unsigned surface_format
,
745 unsigned buffer_size
,
748 void (*emit_null_surface_state
)(struct brw_context
*brw
,
752 uint32_t *out_offset
);
755 * Send the appropriate state packets to configure depth, stencil, and
756 * HiZ buffers (i965+ only)
758 void (*emit_depth_stencil_hiz
)(struct brw_context
*brw
,
759 struct intel_mipmap_tree
*depth_mt
,
760 uint32_t depth_offset
,
761 uint32_t depthbuffer_format
,
762 uint32_t depth_surface_type
,
763 struct intel_mipmap_tree
*stencil_mt
,
764 bool hiz
, bool separate_stencil
,
765 uint32_t width
, uint32_t height
,
766 uint32_t tile_x
, uint32_t tile_y
);
772 drm_intel_context
*hw_ctx
;
774 /** BO for post-sync nonzero writes for gen6 workaround. */
775 drm_intel_bo
*workaround_bo
;
776 uint8_t pipe_controls_since_last_cs_stall
;
779 * Set of drm_intel_bo * that have been rendered to within this batchbuffer
780 * and would need flushing before being used from another cache domain that
781 * isn't coherent with it (i.e. the sampler).
783 struct set
*render_cache
;
786 * Number of resets observed in the system at context creation.
788 * This is tracked in the context so that we can determine that another
789 * reset has occurred.
791 uint32_t reset_count
;
793 struct intel_batchbuffer batch
;
798 uint32_t next_offset
;
802 * Set if rendering has occurred to the drawable's front buffer.
804 * This is used in the DRI2 case to detect that glFlush should also copy
805 * the contents of the fake front buffer to the real front buffer.
807 bool front_buffer_dirty
;
809 /** Framerate throttling: @{ */
810 drm_intel_bo
*throttle_batch
[2];
812 /* Limit the number of outstanding SwapBuffers by waiting for an earlier
813 * frame of rendering to complete. This gives a very precise cap to the
814 * latency between input and output such that rendering never gets more
815 * than a frame behind the user. (With the caveat that we technically are
816 * not using the SwapBuffers itself as a barrier but the first batch
817 * submitted afterwards, which may be immediately prior to the next
820 bool need_swap_throttle
;
822 /** General throttling, not caught by throttling between SwapBuffers */
823 bool need_flush_throttle
;
833 bool always_flush_batch
;
834 bool always_flush_cache
;
835 bool disable_throttling
;
838 driOptionCache optionCache
;
841 GLuint primitive
; /**< Hardware primitive, such as _3DPRIM_TRILIST. */
843 GLenum reduced_primitive
;
846 * Set if we're either a debug context or the INTEL_DEBUG=perf environment
847 * variable is set, this is the flag indicating to do expensive work that
848 * might lead to a perf_debug() call.
852 uint32_t max_gtt_map_object_size
;
864 bool has_separate_stencil
;
865 bool must_use_separate_stencil
;
868 bool has_surface_tile_offset
;
870 bool has_negative_rhw_bug
;
874 bool use_resource_streamer
;
877 * Whether LRI can be used to write register values from the batch buffer.
879 bool can_do_pipelined_register_writes
;
882 * Some versions of Gen hardware don't do centroid interpolation correctly
883 * on unlit pixels, causing incorrect values for derivatives near triangle
884 * edges. Enabling this flag causes the fragment shader to use
885 * non-centroid interpolation for unlit pixels, at the expense of two extra
886 * fragment shader instructions.
888 bool needs_unlit_centroid_workaround
;
892 struct brw_state_flags pipelines
[BRW_NUM_PIPELINES
];
895 enum brw_pipeline last_pipeline
;
897 struct brw_cache cache
;
899 /** IDs for meta stencil blit shader programs. */
900 unsigned meta_stencil_blit_programs
[2];
902 /* Whether a meta-operation is in progress. */
903 bool meta_in_progress
;
905 /* Whether the last depth/stencil packets were both NULL. */
906 bool no_depth_or_stencil
;
908 /* The last PMA stall bits programmed. */
909 uint32_t pma_stall_bits
;
913 /** The value of gl_BaseVertex for the current _mesa_prim. */
916 /** The value of gl_BaseInstance for the current _mesa_prim. */
921 * Buffer and offset used for GL_ARB_shader_draw_parameters
922 * (for now, only gl_BaseVertex).
924 drm_intel_bo
*draw_params_bo
;
925 uint32_t draw_params_offset
;
928 * The value of gl_DrawID for the current _mesa_prim. This always comes
929 * in from it's own vertex buffer since it's not part of the indirect
933 drm_intel_bo
*draw_id_bo
;
934 uint32_t draw_id_offset
;
939 * For gl_NumWorkGroups: If num_work_groups_bo is non NULL, then it is
940 * an indirect call, and num_work_groups_offset is valid. Otherwise,
941 * num_work_groups is set based on glDispatchCompute.
943 drm_intel_bo
*num_work_groups_bo
;
944 GLintptr num_work_groups_offset
;
945 const GLuint
*num_work_groups
;
949 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
950 struct brw_vertex_buffer buffers
[VERT_ATTRIB_MAX
];
952 struct brw_vertex_element
*enabled
[VERT_ATTRIB_MAX
];
956 /* Summary of size and varying of active arrays, so we can check
957 * for changes to this state:
959 unsigned int min_index
, max_index
;
961 /* Offset from start of vertex buffer so we can avoid redefining
962 * the same VB packed over and over again.
964 unsigned int start_vertex_bias
;
967 * Certain vertex attribute formats aren't natively handled by the
968 * hardware and require special VS code to fix up their values.
970 * These bitfields indicate which workarounds are needed.
972 uint8_t attrib_wa_flags
[VERT_ATTRIB_MAX
];
977 * Index buffer for this draw_prims call.
979 * Updates are signaled by BRW_NEW_INDICES.
981 const struct _mesa_index_buffer
*ib
;
983 /* Updates are signaled by BRW_NEW_INDEX_BUFFER. */
987 /* Offset to index buffer index to use in CMD_3D_PRIM so that we can
988 * avoid re-uploading the IB packet over and over if we're actually
989 * referencing the same index buffer.
991 unsigned int start_vertex_offset
;
994 /* Active vertex program:
996 const struct gl_vertex_program
*vertex_program
;
997 const struct gl_geometry_program
*geometry_program
;
998 const struct gl_tess_ctrl_program
*tess_ctrl_program
;
999 const struct gl_tess_eval_program
*tess_eval_program
;
1000 const struct gl_fragment_program
*fragment_program
;
1001 const struct gl_compute_program
*compute_program
;
1004 * Number of samples in ctx->DrawBuffer, updated by BRW_NEW_NUM_SAMPLES so
1005 * that we don't have to reemit that state every time we change FBOs.
1010 * Platform specific constants containing the maximum number of threads
1011 * for each pipeline stage.
1013 unsigned max_vs_threads
;
1014 unsigned max_hs_threads
;
1015 unsigned max_ds_threads
;
1016 unsigned max_gs_threads
;
1017 unsigned max_wm_threads
;
1018 unsigned max_cs_threads
;
1020 /* BRW_NEW_URB_ALLOCATIONS:
1023 GLuint vsize
; /* vertex size plus header in urb registers */
1024 GLuint gsize
; /* GS output size in urb registers */
1025 GLuint hsize
; /* Tessellation control output size in urb registers */
1026 GLuint dsize
; /* Tessellation evaluation output size in urb registers */
1027 GLuint csize
; /* constant buffer size in urb registers */
1028 GLuint sfsize
; /* setup data size in urb registers */
1032 GLuint min_vs_entries
; /* Minimum number of VS entries */
1033 GLuint max_vs_entries
; /* Maximum number of VS entries */
1034 GLuint max_hs_entries
; /* Maximum number of HS entries */
1035 GLuint max_ds_entries
; /* Maximum number of DS entries */
1036 GLuint max_gs_entries
; /* Maximum number of GS entries */
1038 GLuint nr_vs_entries
;
1039 GLuint nr_hs_entries
;
1040 GLuint nr_ds_entries
;
1041 GLuint nr_gs_entries
;
1042 GLuint nr_clip_entries
;
1043 GLuint nr_sf_entries
;
1044 GLuint nr_cs_entries
;
1054 * URB size in the current configuration. The units this is expressed
1055 * in are somewhat inconsistent, see brw_device_info::urb::size.
1057 * FINISHME: Represent the URB size consistently in KB on all platforms.
1061 /* True if the most recently sent _3DSTATE_URB message allocated
1062 * URB space for the GS.
1066 /* True if the most recently sent _3DSTATE_URB message allocated
1067 * URB space for the HS and DS.
1073 /* BRW_NEW_CURBE_OFFSETS:
1076 GLuint wm_start
; /**< pos of first wm const in CURBE buffer */
1077 GLuint wm_size
; /**< number of float[4] consts, multiple of 16 */
1085 * Pointer to the (intel_upload.c-generated) BO containing the uniforms
1086 * for upload to the CURBE.
1088 drm_intel_bo
*curbe_bo
;
1089 /** Offset within curbe_bo of space for current curbe entry */
1090 GLuint curbe_offset
;
1094 * Layout of vertex data exiting the geometry portion of the pipleine.
1095 * This comes from the last enabled shader stage (GS, DS, or VS).
1097 * BRW_NEW_VUE_MAP_GEOM_OUT is flagged when the VUE map changes.
1099 struct brw_vue_map vue_map_geom_out
;
1102 struct brw_stage_state base
;
1103 struct brw_vs_prog_data
*prog_data
;
1107 struct brw_stage_state base
;
1108 struct brw_tcs_prog_data
*prog_data
;
1111 * True if the 3DSTATE_HS command most recently emitted to the 3D
1112 * pipeline enabled the HS; false otherwise.
1118 struct brw_stage_state base
;
1119 struct brw_tes_prog_data
*prog_data
;
1122 * True if the 3DSTATE_DS command most recently emitted to the 3D
1123 * pipeline enabled the DS; false otherwise.
1129 struct brw_stage_state base
;
1130 struct brw_gs_prog_data
*prog_data
;
1133 * True if the 3DSTATE_GS command most recently emitted to the 3D
1134 * pipeline enabled the GS; false otherwise.
1140 struct brw_ff_gs_prog_data
*prog_data
;
1143 /** Offset in the program cache to the CLIP program pre-gen6 */
1144 uint32_t prog_offset
;
1145 uint32_t state_offset
;
1147 uint32_t bind_bo_offset
;
1149 * Surface offsets for the binding table. We only need surfaces to
1150 * implement transform feedback so BRW_MAX_SOL_BINDINGS is all that we
1151 * need in this case.
1153 uint32_t surf_offset
[BRW_MAX_SOL_BINDINGS
];
1157 struct brw_clip_prog_data
*prog_data
;
1159 /** Offset in the program cache to the CLIP program pre-gen6 */
1160 uint32_t prog_offset
;
1162 /* Offset in the batch to the CLIP state on pre-gen6. */
1163 uint32_t state_offset
;
1165 /* As of gen6, this is the offset in the batch to the CLIP VP,
1173 struct brw_sf_prog_data
*prog_data
;
1175 /** Offset in the program cache to the CLIP program pre-gen6 */
1176 uint32_t prog_offset
;
1177 uint32_t state_offset
;
1179 bool viewport_transform_enable
;
1183 struct brw_stage_state base
;
1184 struct brw_wm_prog_data
*prog_data
;
1189 * Buffer object used in place of multisampled null render targets on
1190 * Gen6. See brw_emit_null_surface_state().
1192 drm_intel_bo
*multisampled_null_render_target_bo
;
1193 uint32_t fast_clear_op
;
1199 struct brw_stage_state base
;
1200 struct brw_cs_prog_data
*prog_data
;
1203 /* RS hardware binding table */
1206 uint32_t next_offset
;
1210 uint32_t state_offset
;
1211 uint32_t blend_state_offset
;
1212 uint32_t depth_stencil_state_offset
;
1217 struct brw_query_object
*obj
;
1222 enum brw_predicate_state state
;
1227 /** A map from pipeline statistics counter IDs to MMIO addresses. */
1228 const int *statistics_registers
;
1230 /** The number of active monitors using OA counters. */
1234 * A buffer object storing OA counter snapshots taken at the start and
1235 * end of each batch (creating "bookends" around the batch).
1237 drm_intel_bo
*bookend_bo
;
1239 /** The number of snapshots written to bookend_bo. */
1240 int bookend_snapshots
;
1243 * An array of monitors whose results haven't yet been assembled based on
1244 * the data in buffer objects.
1246 * These may be active, or have already ended. However, the results
1247 * have not been requested.
1249 struct brw_perf_monitor_object
**unresolved
;
1250 int unresolved_elements
;
1251 int unresolved_array_size
;
1254 * Mapping from a uint32_t offset within an OA snapshot to the ID of
1255 * the counter which MI_REPORT_PERF_COUNT stores there.
1257 const int *oa_snapshot_layout
;
1259 /** Number of 32-bit entries in a hardware counter snapshot. */
1260 int entries_per_oa_snapshot
;
1263 int num_atoms
[BRW_NUM_PIPELINES
];
1264 const struct brw_tracked_state render_atoms
[76];
1265 const struct brw_tracked_state compute_atoms
[10];
1267 /* If (INTEL_DEBUG & DEBUG_BATCH) */
1271 enum aub_state_struct_type type
;
1273 } *state_batch_list
;
1274 int state_batch_count
;
1276 uint32_t render_target_format
[MESA_FORMAT_COUNT
];
1277 bool format_supported_as_render_target
[MESA_FORMAT_COUNT
];
1279 /* Interpolation modes, one byte per vue slot.
1280 * Used Gen4/5 by the clip|sf|wm stages. Ignored on Gen6+.
1282 struct interpolation_mode_map interpolation_mode
;
1284 /* PrimitiveRestart */
1287 bool enable_cut_index
;
1290 /** Computed depth/stencil/hiz state from the current attached
1291 * renderbuffers, valid only during the drawing state upload loop after
1292 * brw_workaround_depthstencil_alignment().
1295 struct intel_mipmap_tree
*depth_mt
;
1296 struct intel_mipmap_tree
*stencil_mt
;
1298 /* Inter-tile (page-aligned) byte offsets. */
1299 uint32_t depth_offset
, hiz_offset
, stencil_offset
;
1300 /* Intra-tile x,y offsets for drawing to depth/stencil/hiz */
1301 uint32_t tile_x
, tile_y
;
1304 uint32_t num_instances
;
1308 const struct brw_l3_config
*config
;
1315 enum shader_time_shader_type
*types
;
1316 struct shader_times
*cumulative
;
1322 struct brw_fast_clear_state
*fast_clear_state
;
1324 __DRIcontext
*driContext
;
1325 struct intel_screen
*intelScreen
;
1328 /*======================================================================
1331 void brwInitVtbl( struct brw_context
*brw
);
1334 extern void intelInitClearFuncs(struct dd_function_table
*functions
);
1336 /*======================================================================
1339 extern const char *const brw_vendor_string
;
1341 extern const char *brw_get_renderer_string(unsigned deviceID
);
1344 DRI_CONF_BO_REUSE_DISABLED
,
1345 DRI_CONF_BO_REUSE_ALL
1348 void intel_update_renderbuffers(__DRIcontext
*context
,
1349 __DRIdrawable
*drawable
);
1350 void intel_prepare_render(struct brw_context
*brw
);
1352 void intel_resolve_for_dri2_flush(struct brw_context
*brw
,
1353 __DRIdrawable
*drawable
);
1355 GLboolean
brwCreateContext(gl_api api
,
1356 const struct gl_config
*mesaVis
,
1357 __DRIcontext
*driContextPriv
,
1358 unsigned major_version
,
1359 unsigned minor_version
,
1363 void *sharedContextPrivate
);
1365 /*======================================================================
1368 GLuint
brw_get_rb_for_slice(struct brw_context
*brw
,
1369 struct intel_mipmap_tree
*mt
,
1370 unsigned level
, unsigned layer
, bool flat
);
1372 void brw_meta_updownsample(struct brw_context
*brw
,
1373 struct intel_mipmap_tree
*src
,
1374 struct intel_mipmap_tree
*dst
);
1376 void brw_meta_fbo_stencil_blit(struct brw_context
*brw
,
1377 struct gl_framebuffer
*read_fb
,
1378 struct gl_framebuffer
*draw_fb
,
1379 GLfloat srcX0
, GLfloat srcY0
,
1380 GLfloat srcX1
, GLfloat srcY1
,
1381 GLfloat dstX0
, GLfloat dstY0
,
1382 GLfloat dstX1
, GLfloat dstY1
);
1384 void brw_meta_stencil_updownsample(struct brw_context
*brw
,
1385 struct intel_mipmap_tree
*src
,
1386 struct intel_mipmap_tree
*dst
);
1388 bool brw_meta_fast_clear(struct brw_context
*brw
,
1389 struct gl_framebuffer
*fb
,
1391 bool partial_clear
);
1394 brw_meta_resolve_color(struct brw_context
*brw
,
1395 struct intel_mipmap_tree
*mt
);
1397 brw_meta_fast_clear_free(struct brw_context
*brw
);
1400 /*======================================================================
1403 void brw_get_depthstencil_tile_masks(struct intel_mipmap_tree
*depth_mt
,
1404 uint32_t depth_level
,
1405 uint32_t depth_layer
,
1406 struct intel_mipmap_tree
*stencil_mt
,
1407 uint32_t *out_tile_mask_x
,
1408 uint32_t *out_tile_mask_y
);
1409 void brw_workaround_depthstencil_alignment(struct brw_context
*brw
,
1410 GLbitfield clear_mask
);
1412 /* brw_object_purgeable.c */
1413 void brw_init_object_purgeable_functions(struct dd_function_table
*functions
);
1415 /*======================================================================
1418 void brw_init_common_queryobj_functions(struct dd_function_table
*functions
);
1419 void gen4_init_queryobj_functions(struct dd_function_table
*functions
);
1420 void brw_emit_query_begin(struct brw_context
*brw
);
1421 void brw_emit_query_end(struct brw_context
*brw
);
1423 /** gen6_queryobj.c */
1424 void gen6_init_queryobj_functions(struct dd_function_table
*functions
);
1425 void brw_write_timestamp(struct brw_context
*brw
, drm_intel_bo
*bo
, int idx
);
1426 void brw_write_depth_count(struct brw_context
*brw
, drm_intel_bo
*bo
, int idx
);
1427 void brw_store_register_mem64(struct brw_context
*brw
,
1428 drm_intel_bo
*bo
, uint32_t reg
, int idx
);
1430 /** brw_conditional_render.c */
1431 void brw_init_conditional_render_functions(struct dd_function_table
*functions
);
1432 bool brw_check_conditional_render(struct brw_context
*brw
);
1434 /** intel_batchbuffer.c */
1435 void brw_load_register_mem(struct brw_context
*brw
,
1438 uint32_t read_domains
, uint32_t write_domain
,
1440 void brw_load_register_mem64(struct brw_context
*brw
,
1443 uint32_t read_domains
, uint32_t write_domain
,
1446 /*======================================================================
1449 void brw_debug_batch(struct brw_context
*brw
);
1450 void brw_annotate_aub(struct brw_context
*brw
);
1452 /*======================================================================
1455 void brw_validate_textures( struct brw_context
*brw
);
1458 /*======================================================================
1462 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
1465 perf_debug(" %s %d->%d\n", name
, a
, b
);
1471 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
1473 /* Per-thread scratch space is a power-of-two multiple of 1KB. */
1475 brw_get_scratch_size(int size
)
1477 return util_next_power_of_two(size
| 1023);
1479 void brw_get_scratch_bo(struct brw_context
*brw
,
1480 drm_intel_bo
**scratch_bo
, int size
);
1481 void brw_init_shader_time(struct brw_context
*brw
);
1482 int brw_get_shader_time_index(struct brw_context
*brw
,
1483 struct gl_shader_program
*shader_prog
,
1484 struct gl_program
*prog
,
1485 enum shader_time_shader_type type
);
1486 void brw_collect_and_report_shader_time(struct brw_context
*brw
);
1487 void brw_destroy_shader_time(struct brw_context
*brw
);
1491 void brw_upload_urb_fence(struct brw_context
*brw
);
1495 void brw_upload_cs_urb_state(struct brw_context
*brw
);
1497 /* brw_fs_reg_allocate.cpp
1499 void brw_fs_alloc_reg_sets(struct brw_compiler
*compiler
);
1501 /* brw_vec4_reg_allocate.cpp */
1502 void brw_vec4_alloc_reg_set(struct brw_compiler
*compiler
);
1505 int brw_disassemble_inst(FILE *file
, const struct brw_device_info
*devinfo
,
1506 struct brw_inst
*inst
, bool is_compacted
);
1509 gl_clip_plane
*brw_select_clip_planes(struct gl_context
*ctx
);
1511 /* brw_draw_upload.c */
1512 unsigned brw_get_vertex_surface_type(struct brw_context
*brw
,
1513 const struct gl_client_array
*glarray
);
1515 static inline unsigned
1516 brw_get_index_type(GLenum type
)
1518 assert((type
== GL_UNSIGNED_BYTE
)
1519 || (type
== GL_UNSIGNED_SHORT
)
1520 || (type
== GL_UNSIGNED_INT
));
1522 /* The possible values for type are GL_UNSIGNED_BYTE (0x1401),
1523 * GL_UNSIGNED_SHORT (0x1403), and GL_UNSIGNED_INT (0x1405) which we want
1524 * to map to scale factors of 0, 1, and 2, respectively. These scale
1525 * factors are then left-shfited by 8 to be in the correct position in the
1526 * CMD_INDEX_BUFFER packet.
1528 * Subtracting 0x1401 gives 0, 2, and 4. Shifting left by 7 afterwards
1529 * gives 0x00000000, 0x00000100, and 0x00000200. These just happen to be
1530 * the values the need to be written in the CMD_INDEX_BUFFER packet.
1532 return (type
- 0x1401) << 7;
1535 void brw_prepare_vertices(struct brw_context
*brw
);
1537 /* brw_wm_surface_state.c */
1538 void brw_init_surface_formats(struct brw_context
*brw
);
1539 void brw_create_constant_surface(struct brw_context
*brw
,
1543 uint32_t *out_offset
);
1544 void brw_create_buffer_surface(struct brw_context
*brw
,
1548 uint32_t *out_offset
);
1549 void brw_update_buffer_texture_surface(struct gl_context
*ctx
,
1551 uint32_t *surf_offset
);
1553 brw_update_sol_surface(struct brw_context
*brw
,
1554 struct gl_buffer_object
*buffer_obj
,
1555 uint32_t *out_offset
, unsigned num_vector_components
,
1556 unsigned stride_dwords
, unsigned offset_dwords
);
1557 void brw_upload_ubo_surfaces(struct brw_context
*brw
,
1558 struct gl_shader
*shader
,
1559 struct brw_stage_state
*stage_state
,
1560 struct brw_stage_prog_data
*prog_data
);
1561 void brw_upload_abo_surfaces(struct brw_context
*brw
,
1562 struct gl_shader
*shader
,
1563 struct brw_stage_state
*stage_state
,
1564 struct brw_stage_prog_data
*prog_data
);
1565 void brw_upload_image_surfaces(struct brw_context
*brw
,
1566 struct gl_shader
*shader
,
1567 struct brw_stage_state
*stage_state
,
1568 struct brw_stage_prog_data
*prog_data
);
1570 /* brw_surface_formats.c */
1571 bool brw_render_target_supported(struct brw_context
*brw
,
1572 struct gl_renderbuffer
*rb
);
1573 bool brw_losslessly_compressible_format(struct brw_context
*brw
,
1574 uint32_t brw_format
);
1575 uint32_t brw_depth_format(struct brw_context
*brw
, mesa_format format
);
1576 mesa_format
brw_lower_mesa_image_format(const struct brw_device_info
*devinfo
,
1577 mesa_format format
);
1579 /* brw_performance_monitor.c */
1580 void brw_init_performance_monitors(struct brw_context
*brw
);
1581 void brw_dump_perf_monitors(struct brw_context
*brw
);
1582 void brw_perf_monitor_new_batch(struct brw_context
*brw
);
1583 void brw_perf_monitor_finish_batch(struct brw_context
*brw
);
1585 /* intel_buffer_objects.c */
1586 int brw_bo_map(struct brw_context
*brw
, drm_intel_bo
*bo
, int write_enable
,
1587 const char *bo_name
);
1588 int brw_bo_map_gtt(struct brw_context
*brw
, drm_intel_bo
*bo
,
1589 const char *bo_name
);
1591 /* intel_extensions.c */
1592 extern void intelInitExtensions(struct gl_context
*ctx
);
1595 extern int intel_translate_shadow_compare_func(GLenum func
);
1596 extern int intel_translate_compare_func(GLenum func
);
1597 extern int intel_translate_stencil_op(GLenum op
);
1598 extern int intel_translate_logic_op(GLenum opcode
);
1600 /* intel_syncobj.c */
1601 void intel_init_syncobj_functions(struct dd_function_table
*functions
);
1604 struct gl_transform_feedback_object
*
1605 brw_new_transform_feedback(struct gl_context
*ctx
, GLuint name
);
1607 brw_delete_transform_feedback(struct gl_context
*ctx
,
1608 struct gl_transform_feedback_object
*obj
);
1610 brw_begin_transform_feedback(struct gl_context
*ctx
, GLenum mode
,
1611 struct gl_transform_feedback_object
*obj
);
1613 brw_end_transform_feedback(struct gl_context
*ctx
,
1614 struct gl_transform_feedback_object
*obj
);
1616 brw_get_transform_feedback_vertex_count(struct gl_context
*ctx
,
1617 struct gl_transform_feedback_object
*obj
,
1620 /* gen7_sol_state.c */
1622 gen7_begin_transform_feedback(struct gl_context
*ctx
, GLenum mode
,
1623 struct gl_transform_feedback_object
*obj
);
1625 gen7_end_transform_feedback(struct gl_context
*ctx
,
1626 struct gl_transform_feedback_object
*obj
);
1628 gen7_pause_transform_feedback(struct gl_context
*ctx
,
1629 struct gl_transform_feedback_object
*obj
);
1631 gen7_resume_transform_feedback(struct gl_context
*ctx
,
1632 struct gl_transform_feedback_object
*obj
);
1634 /* brw_blorp_blit.cpp */
1636 brw_blorp_framebuffer(struct brw_context
*brw
,
1637 struct gl_framebuffer
*readFb
,
1638 struct gl_framebuffer
*drawFb
,
1639 GLint srcX0
, GLint srcY0
, GLint srcX1
, GLint srcY1
,
1640 GLint dstX0
, GLint dstY0
, GLint dstX1
, GLint dstY1
,
1641 GLbitfield mask
, GLenum filter
);
1644 brw_blorp_copytexsubimage(struct brw_context
*brw
,
1645 struct gl_renderbuffer
*src_rb
,
1646 struct gl_texture_image
*dst_image
,
1648 int srcX0
, int srcY0
,
1649 int dstX0
, int dstY0
,
1650 int width
, int height
);
1652 /* gen6_multisample_state.c */
1654 gen6_determine_sample_mask(struct brw_context
*brw
);
1657 gen6_emit_3dstate_multisample(struct brw_context
*brw
,
1658 unsigned num_samples
);
1660 gen6_emit_3dstate_sample_mask(struct brw_context
*brw
, unsigned mask
);
1662 gen6_get_sample_position(struct gl_context
*ctx
,
1663 struct gl_framebuffer
*fb
,
1667 gen6_set_sample_maps(struct gl_context
*ctx
);
1669 /* gen8_multisample_state.c */
1670 void gen8_emit_3dstate_multisample(struct brw_context
*brw
, unsigned num_samp
);
1671 void gen8_emit_3dstate_sample_pattern(struct brw_context
*brw
);
1675 gen7_emit_push_constant_state(struct brw_context
*brw
, unsigned vs_size
,
1676 unsigned hs_size
, unsigned ds_size
,
1677 unsigned gs_size
, unsigned fs_size
);
1680 gen7_emit_urb_state(struct brw_context
*brw
,
1681 unsigned nr_vs_entries
,
1682 unsigned vs_size
, unsigned vs_start
,
1683 unsigned nr_hs_entries
,
1684 unsigned hs_size
, unsigned hs_start
,
1685 unsigned nr_ds_entries
,
1686 unsigned ds_size
, unsigned ds_start
,
1687 unsigned nr_gs_entries
,
1688 unsigned gs_size
, unsigned gs_start
);
1693 brw_get_graphics_reset_status(struct gl_context
*ctx
);
1697 brw_init_compute_functions(struct dd_function_table
*functions
);
1699 /*======================================================================
1700 * Inline conversion functions. These are better-typed than the
1701 * macros used previously:
1703 static inline struct brw_context
*
1704 brw_context( struct gl_context
*ctx
)
1706 return (struct brw_context
*)ctx
;
1709 static inline struct brw_vertex_program
*
1710 brw_vertex_program(struct gl_vertex_program
*p
)
1712 return (struct brw_vertex_program
*) p
;
1715 static inline const struct brw_vertex_program
*
1716 brw_vertex_program_const(const struct gl_vertex_program
*p
)
1718 return (const struct brw_vertex_program
*) p
;
1721 static inline struct brw_tess_ctrl_program
*
1722 brw_tess_ctrl_program(struct gl_tess_ctrl_program
*p
)
1724 return (struct brw_tess_ctrl_program
*) p
;
1727 static inline struct brw_tess_eval_program
*
1728 brw_tess_eval_program(struct gl_tess_eval_program
*p
)
1730 return (struct brw_tess_eval_program
*) p
;
1733 static inline struct brw_geometry_program
*
1734 brw_geometry_program(struct gl_geometry_program
*p
)
1736 return (struct brw_geometry_program
*) p
;
1739 static inline struct brw_fragment_program
*
1740 brw_fragment_program(struct gl_fragment_program
*p
)
1742 return (struct brw_fragment_program
*) p
;
1745 static inline const struct brw_fragment_program
*
1746 brw_fragment_program_const(const struct gl_fragment_program
*p
)
1748 return (const struct brw_fragment_program
*) p
;
1751 static inline struct brw_compute_program
*
1752 brw_compute_program(struct gl_compute_program
*p
)
1754 return (struct brw_compute_program
*) p
;
1758 * Pre-gen6, the register file of the EUs was shared between threads,
1759 * and each thread used some subset allocated on a 16-register block
1760 * granularity. The unit states wanted these block counts.
1763 brw_register_blocks(int reg_count
)
1765 return ALIGN(reg_count
, 16) / 16 - 1;
1768 static inline uint32_t
1769 brw_program_reloc(struct brw_context
*brw
, uint32_t state_offset
,
1770 uint32_t prog_offset
)
1772 if (brw
->gen
>= 5) {
1773 /* Using state base address. */
1777 drm_intel_bo_emit_reloc(brw
->batch
.bo
,
1781 I915_GEM_DOMAIN_INSTRUCTION
, 0);
1783 return brw
->cache
.bo
->offset64
+ prog_offset
;
1786 bool brw_do_cubemap_normalize(struct exec_list
*instructions
);
1787 bool brw_lower_texture_gradients(struct brw_context
*brw
,
1788 struct exec_list
*instructions
);
1789 bool brw_do_lower_unnormalized_offset(struct exec_list
*instructions
);
1791 struct opcode_desc
{
1797 extern const struct opcode_desc opcode_descs
[128];
1798 extern const char * const conditional_modifier
[16];
1799 extern const char *const pred_ctrl_align16
[16];
1802 brw_emit_depthbuffer(struct brw_context
*brw
);
1805 brw_emit_depth_stencil_hiz(struct brw_context
*brw
,
1806 struct intel_mipmap_tree
*depth_mt
,
1807 uint32_t depth_offset
, uint32_t depthbuffer_format
,
1808 uint32_t depth_surface_type
,
1809 struct intel_mipmap_tree
*stencil_mt
,
1810 bool hiz
, bool separate_stencil
,
1811 uint32_t width
, uint32_t height
,
1812 uint32_t tile_x
, uint32_t tile_y
);
1815 gen6_emit_depth_stencil_hiz(struct brw_context
*brw
,
1816 struct intel_mipmap_tree
*depth_mt
,
1817 uint32_t depth_offset
, uint32_t depthbuffer_format
,
1818 uint32_t depth_surface_type
,
1819 struct intel_mipmap_tree
*stencil_mt
,
1820 bool hiz
, bool separate_stencil
,
1821 uint32_t width
, uint32_t height
,
1822 uint32_t tile_x
, uint32_t tile_y
);
1825 gen7_emit_depth_stencil_hiz(struct brw_context
*brw
,
1826 struct intel_mipmap_tree
*depth_mt
,
1827 uint32_t depth_offset
, uint32_t depthbuffer_format
,
1828 uint32_t depth_surface_type
,
1829 struct intel_mipmap_tree
*stencil_mt
,
1830 bool hiz
, bool separate_stencil
,
1831 uint32_t width
, uint32_t height
,
1832 uint32_t tile_x
, uint32_t tile_y
);
1834 gen8_emit_depth_stencil_hiz(struct brw_context
*brw
,
1835 struct intel_mipmap_tree
*depth_mt
,
1836 uint32_t depth_offset
, uint32_t depthbuffer_format
,
1837 uint32_t depth_surface_type
,
1838 struct intel_mipmap_tree
*stencil_mt
,
1839 bool hiz
, bool separate_stencil
,
1840 uint32_t width
, uint32_t height
,
1841 uint32_t tile_x
, uint32_t tile_y
);
1843 void gen8_hiz_exec(struct brw_context
*brw
, struct intel_mipmap_tree
*mt
,
1844 unsigned int level
, unsigned int layer
, enum gen6_hiz_op op
);
1846 uint32_t get_hw_prim_for_gl_prim(int mode
);
1849 gen6_upload_push_constants(struct brw_context
*brw
,
1850 const struct gl_program
*prog
,
1851 const struct brw_stage_prog_data
*prog_data
,
1852 struct brw_stage_state
*stage_state
,
1853 enum aub_state_struct_type type
);
1856 gen9_use_linear_1d_layout(const struct brw_context
*brw
,
1857 const struct intel_mipmap_tree
*mt
);
1859 /* brw_pipe_control.c */
1860 int brw_init_pipe_control(struct brw_context
*brw
,
1861 const struct brw_device_info
*info
);
1862 void brw_fini_pipe_control(struct brw_context
*brw
);
1864 void brw_emit_pipe_control_flush(struct brw_context
*brw
, uint32_t flags
);
1865 void brw_emit_pipe_control_write(struct brw_context
*brw
, uint32_t flags
,
1866 drm_intel_bo
*bo
, uint32_t offset
,
1867 uint32_t imm_lower
, uint32_t imm_upper
);
1868 void brw_emit_mi_flush(struct brw_context
*brw
);
1869 void brw_emit_post_sync_nonzero_flush(struct brw_context
*brw
);
1870 void brw_emit_depth_stall_flushes(struct brw_context
*brw
);
1871 void gen7_emit_vs_workaround_flush(struct brw_context
*brw
);
1872 void gen7_emit_cs_stall_flush(struct brw_context
*brw
);