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>
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.
56 * Note that consumers of this map using cache keys must include
57 * prog_data->userclip and prog_data->outputs_written in their key
58 * (generated by CACHE_NEW_VS_PROG).
61 brw_compute_vue_map(struct brw_context
*brw
, struct brw_vue_map
*vue_map
,
62 GLbitfield64 slots_valid
, bool userclip_active
)
64 vue_map
->slots_valid
= slots_valid
;
67 /* Make sure that the values we store in vue_map->varying_to_slot and
68 * vue_map->slot_to_varying won't overflow the signed chars that are used
69 * to store them. Note that since vue_map->slot_to_varying sometimes holds
70 * values equal to BRW_VARYING_SLOT_COUNT, we need to ensure that
71 * BRW_VARYING_SLOT_COUNT is <= 127, not 128.
73 STATIC_ASSERT(BRW_VARYING_SLOT_COUNT
<= 127);
75 vue_map
->num_slots
= 0;
76 for (i
= 0; i
< BRW_VARYING_SLOT_COUNT
; ++i
) {
77 vue_map
->varying_to_slot
[i
] = -1;
78 vue_map
->slot_to_varying
[i
] = BRW_VARYING_SLOT_COUNT
;
81 /* VUE header: format depends on chip generation and whether clipping is
87 /* There are 8 dwords in VUE header pre-Ironlake:
88 * dword 0-3 is indices, point width, clip flags.
89 * dword 4-7 is ndc position
90 * dword 8-11 is the first vertex data.
92 * On Ironlake the VUE header is nominally 20 dwords, but the hardware
93 * will accept the same header layout as Gen4 [and should be a bit faster]
95 assign_vue_slot(vue_map
, VARYING_SLOT_PSIZ
);
96 assign_vue_slot(vue_map
, BRW_VARYING_SLOT_NDC
);
97 assign_vue_slot(vue_map
, VARYING_SLOT_POS
);
101 /* There are 8 or 16 DWs (D0-D15) in VUE header on Sandybridge:
102 * dword 0-3 of the header is indices, point width, clip flags.
103 * dword 4-7 is the 4D space position
104 * dword 8-15 of the vertex header is the user clip distance if
106 * dword 8-11 or 16-19 is the first vertex element data we fill.
108 assign_vue_slot(vue_map
, VARYING_SLOT_PSIZ
);
109 assign_vue_slot(vue_map
, VARYING_SLOT_POS
);
110 if (userclip_active
) {
111 assign_vue_slot(vue_map
, VARYING_SLOT_CLIP_DIST0
);
112 assign_vue_slot(vue_map
, VARYING_SLOT_CLIP_DIST1
);
114 /* front and back colors need to be consecutive so that we can use
115 * ATTRIBUTE_SWIZZLE_INPUTATTR_FACING to swizzle them when doing
118 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_COL0
))
119 assign_vue_slot(vue_map
, VARYING_SLOT_COL0
);
120 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
121 assign_vue_slot(vue_map
, VARYING_SLOT_BFC0
);
122 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_COL1
))
123 assign_vue_slot(vue_map
, VARYING_SLOT_COL1
);
124 if (slots_valid
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
125 assign_vue_slot(vue_map
, VARYING_SLOT_BFC1
);
128 assert (!"VUE map not known for this chip generation");
132 /* The hardware doesn't care about the rest of the vertex outputs, so just
133 * assign them contiguously. Don't reassign outputs that already have a
136 * We generally don't need to assign a slot for VARYING_SLOT_CLIP_VERTEX,
137 * since it's encoded as the clip distances by emit_clip_distances().
138 * However, it may be output by transform feedback, and we'd rather not
139 * recompute state when TF changes, so we just always include it.
141 for (int i
= 0; i
< VARYING_SLOT_MAX
; ++i
) {
142 if ((slots_valid
& BITFIELD64_BIT(i
)) &&
143 vue_map
->varying_to_slot
[i
] == -1) {
144 assign_vue_slot(vue_map
, i
);
151 * Decide which set of clip planes should be used when clipping via
152 * gl_Position or gl_ClipVertex.
154 gl_clip_plane
*brw_select_clip_planes(struct gl_context
*ctx
)
156 if (ctx
->Shader
.CurrentVertexProgram
) {
157 /* There is currently a GLSL vertex shader, so clip according to GLSL
158 * rules, which means compare gl_ClipVertex (or gl_Position, if
159 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
160 * that were stored in EyeUserPlane at the time the clip planes were
163 return ctx
->Transform
.EyeUserPlane
;
165 /* Either we are using fixed function or an ARB vertex program. In
166 * either case the clip planes are going to be compared against
167 * gl_Position (which is in clip coordinates) so we have to clip using
168 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
171 return ctx
->Transform
._ClipUserPlane
;
177 brw_vec4_prog_data_compare(const struct brw_vec4_prog_data
*a
,
178 const struct brw_vec4_prog_data
*b
)
180 /* Compare all the struct up to the pointers. */
181 if (memcmp(a
, b
, offsetof(struct brw_vec4_prog_data
, param
)))
184 if (memcmp(a
->param
, b
->param
, a
->nr_params
* sizeof(void *)))
187 if (memcmp(a
->pull_param
, b
->pull_param
, a
->nr_pull_params
* sizeof(void *)))
195 brw_vs_prog_data_compare(const void *in_a
, const void *in_b
,
196 int aux_size
, const void *in_key
)
198 const struct brw_vs_prog_data
*a
= in_a
;
199 const struct brw_vs_prog_data
*b
= in_b
;
201 /* Compare the base vec4 structure. */
202 if (!brw_vec4_prog_data_compare(&a
->base
, &b
->base
))
205 /* Compare the rest of the struct. */
206 const unsigned offset
= sizeof(struct brw_vec4_prog_data
);
207 if (memcmp(((char *) &a
) + offset
, ((char *) &b
) + offset
,
208 sizeof(struct brw_vs_prog_data
) - offset
)) {
216 do_vs_prog(struct brw_context
*brw
,
217 struct gl_shader_program
*prog
,
218 struct brw_vertex_program
*vp
,
219 struct brw_vs_prog_key
*key
)
222 const GLuint
*program
;
223 struct brw_vs_compile c
;
224 struct brw_vs_prog_data prog_data
;
227 struct gl_shader
*vs
= NULL
;
230 vs
= prog
->_LinkedShaders
[MESA_SHADER_VERTEX
];
232 memset(&c
, 0, sizeof(c
));
233 memcpy(&c
.key
, key
, sizeof(*key
));
234 memset(&prog_data
, 0, sizeof(prog_data
));
236 mem_ctx
= ralloc_context(NULL
);
240 /* Allocate the references to the uniforms that will end up in the
241 * prog_data associated with the compiled program, and which will be freed
242 * by the state cache.
246 /* We add padding around uniform values below vec4 size, with the worst
247 * case being a float value that gets blown up to a vec4, so be
250 param_count
= vs
->num_uniform_components
* 4;
253 param_count
= vp
->program
.Base
.Parameters
->NumParameters
* 4;
255 /* We also upload clip plane data as uniforms */
256 param_count
+= MAX_CLIP_PLANES
* 4;
258 prog_data
.base
.param
= rzalloc_array(NULL
, const float *, param_count
);
259 prog_data
.base
.pull_param
= rzalloc_array(NULL
, const float *, param_count
);
261 GLbitfield64 outputs_written
= vp
->program
.Base
.OutputsWritten
;
262 prog_data
.inputs_read
= vp
->program
.Base
.InputsRead
;
264 if (c
.key
.copy_edgeflag
) {
265 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_EDGE
);
266 prog_data
.inputs_read
|= VERT_BIT_EDGEFLAG
;
270 /* Put dummy slots into the VUE for the SF to put the replaced
271 * point sprite coords in. We shouldn't need these dummy slots,
272 * which take up precious URB space, but it would mean that the SF
273 * doesn't get nice aligned pairs of input coords into output
274 * coords, which would be a pain to handle.
276 for (i
= 0; i
< 8; i
++) {
277 if (c
.key
.point_coord_replace
& (1 << i
))
278 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_TEX0
+ i
);
281 /* if back colors are written, allocate slots for front colors too */
282 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
283 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL0
);
284 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
285 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL1
);
287 if (c
.key
.base
.userclip_active
) {
288 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0
);
289 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1
);
293 brw_compute_vue_map(brw
, &prog_data
.base
.vue_map
, outputs_written
,
294 c
.key
.base
.userclip_active
);
297 _mesa_fprint_program_opt(stdout
, &c
.vp
->program
.Base
, PROG_PRINT_DEBUG
,
303 program
= brw_vs_emit(brw
, prog
, &c
, &prog_data
, mem_ctx
, &program_size
);
304 if (program
== NULL
) {
305 ralloc_free(mem_ctx
);
309 /* Scratch space is used for register spilling */
310 if (c
.base
.last_scratch
) {
311 perf_debug("Vertex shader triggered register spilling. "
312 "Try reducing the number of live vec4 values to "
313 "improve performance.\n");
315 prog_data
.base
.total_scratch
316 = brw_get_scratch_size(c
.base
.last_scratch
*REG_SIZE
);
318 brw_get_scratch_bo(brw
, &brw
->vs
.scratch_bo
,
319 prog_data
.base
.total_scratch
* brw
->max_vs_threads
);
322 brw_upload_cache(&brw
->cache
, BRW_VS_PROG
,
323 &c
.key
, sizeof(c
.key
),
324 program
, program_size
,
325 &prog_data
, sizeof(prog_data
),
326 &brw
->vs
.prog_offset
, &brw
->vs
.prog_data
);
327 ralloc_free(mem_ctx
);
333 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
336 perf_debug(" %s %d->%d\n", name
, a
, b
);
343 brw_vs_debug_recompile(struct brw_context
*brw
,
344 struct gl_shader_program
*prog
,
345 const struct brw_vs_prog_key
*key
)
347 struct brw_cache_item
*c
= NULL
;
348 const struct brw_vs_prog_key
*old_key
= NULL
;
351 perf_debug("Recompiling vertex shader for program %d\n", prog
->Name
);
353 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
354 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
355 if (c
->cache_id
== BRW_VS_PROG
) {
358 if (old_key
->base
.program_string_id
== key
->base
.program_string_id
)
367 perf_debug(" Didn't find previous compile in the shader cache for "
372 for (unsigned int i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
373 found
|= key_debug(brw
, "Vertex attrib w/a flags",
374 old_key
->gl_attrib_wa_flags
[i
],
375 key
->gl_attrib_wa_flags
[i
]);
378 found
|= key_debug(brw
, "user clip flags",
379 old_key
->base
.userclip_active
, key
->base
.userclip_active
);
381 found
|= key_debug(brw
, "user clipping planes as push constants",
382 old_key
->base
.nr_userclip_plane_consts
,
383 key
->base
.nr_userclip_plane_consts
);
385 found
|= key_debug(brw
, "clip distance enable",
386 old_key
->base
.uses_clip_distance
, key
->base
.uses_clip_distance
);
387 found
|= key_debug(brw
, "copy edgeflag",
388 old_key
->copy_edgeflag
, key
->copy_edgeflag
);
389 found
|= key_debug(brw
, "PointCoord replace",
390 old_key
->point_coord_replace
, key
->point_coord_replace
);
391 found
|= key_debug(brw
, "vertex color clamping",
392 old_key
->base
.clamp_vertex_color
, key
->base
.clamp_vertex_color
);
394 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->base
.tex
,
398 perf_debug(" Something else\n");
402 static void brw_upload_vs_prog(struct brw_context
*brw
)
404 struct gl_context
*ctx
= &brw
->ctx
;
405 struct brw_vs_prog_key key
;
406 /* BRW_NEW_VERTEX_PROGRAM */
407 struct brw_vertex_program
*vp
=
408 (struct brw_vertex_program
*)brw
->vertex_program
;
409 struct gl_program
*prog
= (struct gl_program
*) brw
->vertex_program
;
412 memset(&key
, 0, sizeof(key
));
414 /* Just upload the program verbatim for now. Always send it all
415 * the inputs it asks for, whether they are varying or not.
417 key
.base
.program_string_id
= vp
->id
;
418 key
.base
.userclip_active
= (ctx
->Transform
.ClipPlanesEnabled
!= 0);
419 key
.base
.uses_clip_distance
= vp
->program
.UsesClipDistance
;
420 if (key
.base
.userclip_active
&& !key
.base
.uses_clip_distance
) {
421 key
.base
.nr_userclip_plane_consts
422 = _mesa_logbase2(ctx
->Transform
.ClipPlanesEnabled
) + 1;
427 key
.copy_edgeflag
= (ctx
->Polygon
.FrontMode
!= GL_FILL
||
428 ctx
->Polygon
.BackMode
!= GL_FILL
);
431 /* _NEW_LIGHT | _NEW_BUFFERS */
432 key
.base
.clamp_vertex_color
= ctx
->Light
._ClampVertexColor
;
435 if (brw
->gen
< 6 && ctx
->Point
.PointSprite
) {
436 for (i
= 0; i
< 8; i
++) {
437 if (ctx
->Point
.CoordReplace
[i
])
438 key
.point_coord_replace
|= (1 << i
);
443 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->vs
.sampler_count
,
446 /* BRW_NEW_VERTICES */
447 if (brw
->gen
< 8 && !brw
->is_haswell
) {
448 /* Prior to Haswell, the hardware can't natively support GL_FIXED or
449 * 2_10_10_10_REV vertex formats. Set appropriate workaround flags.
451 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
452 if (!(vp
->program
.Base
.InputsRead
& BITFIELD64_BIT(i
)))
455 uint8_t wa_flags
= 0;
457 switch (brw
->vb
.inputs
[i
].glarray
->Type
) {
460 wa_flags
= brw
->vb
.inputs
[i
].glarray
->Size
;
463 case GL_INT_2_10_10_10_REV
:
464 wa_flags
|= BRW_ATTRIB_WA_SIGN
;
467 case GL_UNSIGNED_INT_2_10_10_10_REV
:
468 if (brw
->vb
.inputs
[i
].glarray
->Format
== GL_BGRA
)
469 wa_flags
|= BRW_ATTRIB_WA_BGRA
;
471 if (brw
->vb
.inputs
[i
].glarray
->Normalized
)
472 wa_flags
|= BRW_ATTRIB_WA_NORMALIZE
;
473 else if (!brw
->vb
.inputs
[i
].glarray
->Integer
)
474 wa_flags
|= BRW_ATTRIB_WA_SCALE
;
479 key
.gl_attrib_wa_flags
[i
] = wa_flags
;
483 if (!brw_search_cache(&brw
->cache
, BRW_VS_PROG
,
485 &brw
->vs
.prog_offset
, &brw
->vs
.prog_data
)) {
486 bool success
= do_vs_prog(brw
, ctx
->Shader
.CurrentVertexProgram
,
491 if (memcmp(&brw
->vs
.prog_data
->base
.vue_map
, &brw
->vue_map_geom_out
,
492 sizeof(brw
->vue_map_geom_out
)) != 0) {
493 brw
->vue_map_geom_out
= brw
->vs
.prog_data
->base
.vue_map
;
494 brw
->state
.dirty
.brw
|= BRW_NEW_VUE_MAP_GEOM_OUT
;
500 const struct brw_tracked_state brw_vs_prog
= {
502 .mesa
= (_NEW_TRANSFORM
| _NEW_POLYGON
| _NEW_POINT
| _NEW_LIGHT
|
505 .brw
= (BRW_NEW_VERTEX_PROGRAM
|
509 .emit
= brw_upload_vs_prog
513 brw_vs_precompile(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
515 struct brw_context
*brw
= brw_context(ctx
);
516 struct brw_vs_prog_key key
;
517 uint32_t old_prog_offset
= brw
->vs
.prog_offset
;
518 struct brw_vs_prog_data
*old_prog_data
= brw
->vs
.prog_data
;
521 if (!prog
->_LinkedShaders
[MESA_SHADER_VERTEX
])
524 struct gl_vertex_program
*vp
= (struct gl_vertex_program
*)
525 prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->Program
;
526 struct brw_vertex_program
*bvp
= brw_vertex_program(vp
);
528 memset(&key
, 0, sizeof(key
));
530 key
.base
.program_string_id
= bvp
->id
;
531 key
.base
.clamp_vertex_color
= ctx
->API
== API_OPENGL_COMPAT
;
533 unsigned sampler_count
= _mesa_fls(vp
->Base
.SamplersUsed
);
534 for (unsigned i
= 0; i
< sampler_count
; i
++) {
535 if (vp
->Base
.ShadowSamplers
& (1 << i
)) {
536 /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
537 key
.base
.tex
.swizzles
[i
] =
538 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_ONE
);
540 /* Color sampler: assume no swizzling. */
541 key
.base
.tex
.swizzles
[i
] = SWIZZLE_XYZW
;
545 success
= do_vs_prog(brw
, prog
, bvp
, &key
);
547 brw
->vs
.prog_offset
= old_prog_offset
;
548 brw
->vs
.prog_data
= old_prog_data
;
555 brw_vec4_prog_data_free(const struct brw_vec4_prog_data
*prog_data
)
557 ralloc_free((void *)prog_data
->param
);
558 ralloc_free((void *)prog_data
->pull_param
);
563 brw_vs_prog_data_free(const void *in_prog_data
)
565 const struct brw_vs_prog_data
*prog_data
= in_prog_data
;
567 brw_vec4_prog_data_free(&prog_data
->base
);