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 #include "main/compiler.h"
34 #include "brw_context.h"
37 #include "brw_state.h"
38 #include "program/prog_print.h"
39 #include "program/prog_parameter.h"
41 #include "glsl/ralloc.h"
43 static inline void assign_vue_slot(struct brw_vue_map
*vue_map
,
46 /* Make sure this varying hasn't been assigned a slot already */
47 assert (vue_map
->varying_to_slot
[varying
] == -1);
49 vue_map
->varying_to_slot
[varying
] = vue_map
->num_slots
;
50 vue_map
->slot_to_varying
[vue_map
->num_slots
++] = varying
;
54 * Compute the VUE map for vertex shader program.
57 brw_compute_vue_map(struct brw_context
*brw
, struct brw_vue_map
*vue_map
,
58 GLbitfield64 slots_valid
)
60 vue_map
->slots_valid
= slots_valid
;
63 /* gl_Layer and gl_ViewportIndex don't get their own varying slots -- they
64 * are stored in the first VUE slot (VARYING_SLOT_PSIZ).
66 slots_valid
&= ~(VARYING_BIT_LAYER
| VARYING_BIT_VIEWPORT
);
68 /* Make sure that the values we store in vue_map->varying_to_slot and
69 * vue_map->slot_to_varying won't overflow the signed chars that are used
70 * to store them. Note that since vue_map->slot_to_varying sometimes holds
71 * values equal to BRW_VARYING_SLOT_COUNT, we need to ensure that
72 * BRW_VARYING_SLOT_COUNT is <= 127, not 128.
74 STATIC_ASSERT(BRW_VARYING_SLOT_COUNT
<= 127);
76 vue_map
->num_slots
= 0;
77 for (i
= 0; i
< BRW_VARYING_SLOT_COUNT
; ++i
) {
78 vue_map
->varying_to_slot
[i
] = -1;
79 vue_map
->slot_to_varying
[i
] = BRW_VARYING_SLOT_COUNT
;
82 /* VUE header: format depends on chip generation and whether clipping is
88 /* There are 8 dwords in VUE header pre-Ironlake:
89 * dword 0-3 is indices, point width, clip flags.
90 * dword 4-7 is ndc position
91 * dword 8-11 is the first vertex data.
93 * On Ironlake the VUE header is nominally 20 dwords, but the hardware
94 * will accept the same header layout as Gen4 [and should be a bit faster]
96 assign_vue_slot(vue_map
, VARYING_SLOT_PSIZ
);
97 assign_vue_slot(vue_map
, BRW_VARYING_SLOT_NDC
);
98 assign_vue_slot(vue_map
, VARYING_SLOT_POS
);
103 /* There are 8 or 16 DWs (D0-D15) in VUE header on Sandybridge:
104 * dword 0-3 of the header is indices, point width, clip flags.
105 * dword 4-7 is the 4D space position
106 * dword 8-15 of the vertex header is the user clip distance if
108 * dword 8-11 or 16-19 is the first vertex element data we fill.
110 assign_vue_slot(vue_map
, VARYING_SLOT_PSIZ
);
111 assign_vue_slot(vue_map
, VARYING_SLOT_POS
);
112 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0
))
113 assign_vue_slot(vue_map
, VARYING_SLOT_CLIP_DIST0
);
114 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1
))
115 assign_vue_slot(vue_map
, VARYING_SLOT_CLIP_DIST1
);
117 /* front and back colors need to be consecutive so that we can use
118 * ATTRIBUTE_SWIZZLE_INPUTATTR_FACING to swizzle them when doing
121 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_COL0
))
122 assign_vue_slot(vue_map
, VARYING_SLOT_COL0
);
123 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
124 assign_vue_slot(vue_map
, VARYING_SLOT_BFC0
);
125 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_COL1
))
126 assign_vue_slot(vue_map
, VARYING_SLOT_COL1
);
127 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
128 assign_vue_slot(vue_map
, VARYING_SLOT_BFC1
);
131 assert (!"VUE map not known for this chip generation");
135 /* The hardware doesn't care about the rest of the vertex outputs, so just
136 * assign them contiguously. Don't reassign outputs that already have a
139 * We generally don't need to assign a slot for VARYING_SLOT_CLIP_VERTEX,
140 * since it's encoded as the clip distances by emit_clip_distances().
141 * However, it may be output by transform feedback, and we'd rather not
142 * recompute state when TF changes, so we just always include it.
144 for (int i
= 0; i
< VARYING_SLOT_MAX
; ++i
) {
145 if ((slots_valid
& BITFIELD64_BIT(i
)) &&
146 vue_map
->varying_to_slot
[i
] == -1) {
147 assign_vue_slot(vue_map
, i
);
154 * Decide which set of clip planes should be used when clipping via
155 * gl_Position or gl_ClipVertex.
157 gl_clip_plane
*brw_select_clip_planes(struct gl_context
*ctx
)
159 if (ctx
->Shader
.CurrentProgram
[MESA_SHADER_VERTEX
]) {
160 /* There is currently a GLSL vertex shader, so clip according to GLSL
161 * rules, which means compare gl_ClipVertex (or gl_Position, if
162 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
163 * that were stored in EyeUserPlane at the time the clip planes were
166 return ctx
->Transform
.EyeUserPlane
;
168 /* Either we are using fixed function or an ARB vertex program. In
169 * either case the clip planes are going to be compared against
170 * gl_Position (which is in clip coordinates) so we have to clip using
171 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
174 return ctx
->Transform
._ClipUserPlane
;
180 brw_vs_prog_data_compare(const void *in_a
, const void *in_b
)
182 const struct brw_vs_prog_data
*a
= in_a
;
183 const struct brw_vs_prog_data
*b
= in_b
;
185 /* Compare the base vec4 structure. */
186 if (!brw_vec4_prog_data_compare(&a
->base
, &b
->base
))
189 /* Compare the rest of the struct. */
190 const unsigned offset
= sizeof(struct brw_vec4_prog_data
);
191 if (memcmp(((char *) a
) + offset
, ((char *) b
) + offset
,
192 sizeof(struct brw_vs_prog_data
) - offset
)) {
200 do_vs_prog(struct brw_context
*brw
,
201 struct gl_shader_program
*prog
,
202 struct brw_vertex_program
*vp
,
203 struct brw_vs_prog_key
*key
)
206 const GLuint
*program
;
207 struct brw_vs_compile c
;
208 struct brw_vs_prog_data prog_data
;
211 struct gl_shader
*vs
= NULL
;
214 vs
= prog
->_LinkedShaders
[MESA_SHADER_VERTEX
];
216 memset(&c
, 0, sizeof(c
));
217 memcpy(&c
.key
, key
, sizeof(*key
));
218 memset(&prog_data
, 0, sizeof(prog_data
));
220 mem_ctx
= ralloc_context(NULL
);
224 /* Allocate the references to the uniforms that will end up in the
225 * prog_data associated with the compiled program, and which will be freed
226 * by the state cache.
230 /* We add padding around uniform values below vec4 size, with the worst
231 * case being a float value that gets blown up to a vec4, so be
234 param_count
= vs
->num_uniform_components
* 4;
237 param_count
= vp
->program
.Base
.Parameters
->NumParameters
* 4;
239 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
240 * planes as uniforms.
242 param_count
+= c
.key
.base
.nr_userclip_plane_consts
* 4;
244 prog_data
.base
.param
= rzalloc_array(NULL
, const float *, param_count
);
245 prog_data
.base
.pull_param
= rzalloc_array(NULL
, const float *, param_count
);
247 GLbitfield64 outputs_written
= vp
->program
.Base
.OutputsWritten
;
248 prog_data
.inputs_read
= vp
->program
.Base
.InputsRead
;
250 if (c
.key
.copy_edgeflag
) {
251 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_EDGE
);
252 prog_data
.inputs_read
|= VERT_BIT_EDGEFLAG
;
256 /* Put dummy slots into the VUE for the SF to put the replaced
257 * point sprite coords in. We shouldn't need these dummy slots,
258 * which take up precious URB space, but it would mean that the SF
259 * doesn't get nice aligned pairs of input coords into output
260 * coords, which would be a pain to handle.
262 for (i
= 0; i
< 8; i
++) {
263 if (c
.key
.point_coord_replace
& (1 << i
))
264 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_TEX0
+ i
);
267 /* if back colors are written, allocate slots for front colors too */
268 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
269 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL0
);
270 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
271 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL1
);
274 /* In order for legacy clipping to work, we need to populate the clip
275 * distance varying slots whenever clipping is enabled, even if the vertex
276 * shader doesn't write to gl_ClipDistance.
278 if (c
.key
.base
.userclip_active
) {
279 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0
);
280 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1
);
283 brw_compute_vue_map(brw
, &prog_data
.base
.vue_map
, outputs_written
);
286 _mesa_fprint_program_opt(stdout
, &c
.vp
->program
.Base
, PROG_PRINT_DEBUG
,
292 program
= brw_vs_emit(brw
, prog
, &c
, &prog_data
, mem_ctx
, &program_size
);
293 if (program
== NULL
) {
294 ralloc_free(mem_ctx
);
298 /* Scratch space is used for register spilling */
299 if (c
.base
.last_scratch
) {
300 perf_debug("Vertex shader triggered register spilling. "
301 "Try reducing the number of live vec4 values to "
302 "improve performance.\n");
304 prog_data
.base
.total_scratch
305 = brw_get_scratch_size(c
.base
.last_scratch
*REG_SIZE
);
307 brw_get_scratch_bo(brw
, &brw
->vs
.base
.scratch_bo
,
308 prog_data
.base
.total_scratch
* brw
->max_vs_threads
);
311 brw_upload_cache(&brw
->cache
, BRW_VS_PROG
,
312 &c
.key
, sizeof(c
.key
),
313 program
, program_size
,
314 &prog_data
, sizeof(prog_data
),
315 &brw
->vs
.base
.prog_offset
, &brw
->vs
.prog_data
);
316 ralloc_free(mem_ctx
);
322 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
325 perf_debug(" %s %d->%d\n", name
, a
, b
);
332 brw_vs_debug_recompile(struct brw_context
*brw
,
333 struct gl_shader_program
*prog
,
334 const struct brw_vs_prog_key
*key
)
336 struct brw_cache_item
*c
= NULL
;
337 const struct brw_vs_prog_key
*old_key
= NULL
;
340 perf_debug("Recompiling vertex shader for program %d\n", prog
->Name
);
342 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
343 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
344 if (c
->cache_id
== BRW_VS_PROG
) {
347 if (old_key
->base
.program_string_id
== key
->base
.program_string_id
)
356 perf_debug(" Didn't find previous compile in the shader cache for "
361 for (unsigned int i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
362 found
|= key_debug(brw
, "Vertex attrib w/a flags",
363 old_key
->gl_attrib_wa_flags
[i
],
364 key
->gl_attrib_wa_flags
[i
]);
367 found
|= key_debug(brw
, "user clip flags",
368 old_key
->base
.userclip_active
, key
->base
.userclip_active
);
370 found
|= key_debug(brw
, "user clipping planes as push constants",
371 old_key
->base
.nr_userclip_plane_consts
,
372 key
->base
.nr_userclip_plane_consts
);
374 found
|= key_debug(brw
, "copy edgeflag",
375 old_key
->copy_edgeflag
, key
->copy_edgeflag
);
376 found
|= key_debug(brw
, "PointCoord replace",
377 old_key
->point_coord_replace
, key
->point_coord_replace
);
378 found
|= key_debug(brw
, "vertex color clamping",
379 old_key
->base
.clamp_vertex_color
, key
->base
.clamp_vertex_color
);
381 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->base
.tex
,
385 perf_debug(" Something else\n");
391 brw_setup_vec4_key_clip_info(struct brw_context
*brw
,
392 struct brw_vec4_prog_key
*key
,
393 bool program_uses_clip_distance
)
395 struct gl_context
*ctx
= &brw
->ctx
;
397 key
->userclip_active
= (ctx
->Transform
.ClipPlanesEnabled
!= 0);
398 if (key
->userclip_active
&& !program_uses_clip_distance
) {
399 key
->nr_userclip_plane_consts
400 = _mesa_logbase2(ctx
->Transform
.ClipPlanesEnabled
) + 1;
405 static void brw_upload_vs_prog(struct brw_context
*brw
)
407 struct gl_context
*ctx
= &brw
->ctx
;
408 struct brw_vs_prog_key key
;
409 /* BRW_NEW_VERTEX_PROGRAM */
410 struct brw_vertex_program
*vp
=
411 (struct brw_vertex_program
*)brw
->vertex_program
;
412 struct gl_program
*prog
= (struct gl_program
*) brw
->vertex_program
;
415 memset(&key
, 0, sizeof(key
));
417 /* Just upload the program verbatim for now. Always send it all
418 * the inputs it asks for, whether they are varying or not.
420 key
.base
.program_string_id
= vp
->id
;
421 brw_setup_vec4_key_clip_info(brw
, &key
.base
,
422 vp
->program
.Base
.UsesClipDistanceOut
);
426 key
.copy_edgeflag
= (ctx
->Polygon
.FrontMode
!= GL_FILL
||
427 ctx
->Polygon
.BackMode
!= GL_FILL
);
430 /* _NEW_LIGHT | _NEW_BUFFERS */
431 key
.base
.clamp_vertex_color
= ctx
->Light
._ClampVertexColor
;
434 if (brw
->gen
< 6 && ctx
->Point
.PointSprite
) {
435 for (i
= 0; i
< 8; i
++) {
436 if (ctx
->Point
.CoordReplace
[i
])
437 key
.point_coord_replace
|= (1 << i
);
442 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->vs
.base
.sampler_count
,
445 /* BRW_NEW_VERTICES */
446 if (brw
->gen
< 8 && !brw
->is_haswell
) {
447 /* Prior to Haswell, the hardware can't natively support GL_FIXED or
448 * 2_10_10_10_REV vertex formats. Set appropriate workaround flags.
450 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
451 if (!(vp
->program
.Base
.InputsRead
& BITFIELD64_BIT(i
)))
454 uint8_t wa_flags
= 0;
456 switch (brw
->vb
.inputs
[i
].glarray
->Type
) {
459 wa_flags
= brw
->vb
.inputs
[i
].glarray
->Size
;
462 case GL_INT_2_10_10_10_REV
:
463 wa_flags
|= BRW_ATTRIB_WA_SIGN
;
466 case GL_UNSIGNED_INT_2_10_10_10_REV
:
467 if (brw
->vb
.inputs
[i
].glarray
->Format
== GL_BGRA
)
468 wa_flags
|= BRW_ATTRIB_WA_BGRA
;
470 if (brw
->vb
.inputs
[i
].glarray
->Normalized
)
471 wa_flags
|= BRW_ATTRIB_WA_NORMALIZE
;
472 else if (!brw
->vb
.inputs
[i
].glarray
->Integer
)
473 wa_flags
|= BRW_ATTRIB_WA_SCALE
;
478 key
.gl_attrib_wa_flags
[i
] = wa_flags
;
482 if (!brw_search_cache(&brw
->cache
, BRW_VS_PROG
,
484 &brw
->vs
.base
.prog_offset
, &brw
->vs
.prog_data
)) {
486 do_vs_prog(brw
, ctx
->Shader
.CurrentProgram
[MESA_SHADER_VERTEX
], vp
,
491 brw
->vs
.base
.prog_data
= &brw
->vs
.prog_data
->base
.base
;
493 if (memcmp(&brw
->vs
.prog_data
->base
.vue_map
, &brw
->vue_map_geom_out
,
494 sizeof(brw
->vue_map_geom_out
)) != 0) {
495 brw
->vue_map_vs
= brw
->vs
.prog_data
->base
.vue_map
;
496 brw
->state
.dirty
.brw
|= BRW_NEW_VUE_MAP_VS
;
498 /* No geometry shader support, so the VS VUE map is the VUE map for
499 * the output of the "geometry" portion of the pipeline.
501 brw
->vue_map_geom_out
= brw
->vue_map_vs
;
502 brw
->state
.dirty
.brw
|= BRW_NEW_VUE_MAP_GEOM_OUT
;
509 const struct brw_tracked_state brw_vs_prog
= {
511 .mesa
= (_NEW_TRANSFORM
| _NEW_POLYGON
| _NEW_POINT
| _NEW_LIGHT
|
514 .brw
= (BRW_NEW_VERTEX_PROGRAM
|
518 .emit
= brw_upload_vs_prog
522 brw_vs_precompile(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
524 struct brw_context
*brw
= brw_context(ctx
);
525 struct brw_vs_prog_key key
;
526 uint32_t old_prog_offset
= brw
->vs
.base
.prog_offset
;
527 struct brw_vs_prog_data
*old_prog_data
= brw
->vs
.prog_data
;
530 if (!prog
->_LinkedShaders
[MESA_SHADER_VERTEX
])
533 struct gl_vertex_program
*vp
= (struct gl_vertex_program
*)
534 prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->Program
;
535 struct brw_vertex_program
*bvp
= brw_vertex_program(vp
);
537 memset(&key
, 0, sizeof(key
));
539 brw_vec4_setup_prog_key_for_precompile(ctx
, &key
.base
, bvp
->id
, &vp
->Base
);
541 success
= do_vs_prog(brw
, prog
, bvp
, &key
);
543 brw
->vs
.base
.prog_offset
= old_prog_offset
;
544 brw
->vs
.prog_data
= old_prog_data
;
551 brw_vs_prog_data_free(const void *in_prog_data
)
553 const struct brw_vs_prog_data
*prog_data
= in_prog_data
;
555 brw_vec4_prog_data_free(&prog_data
->base
);