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>
32 #include "brw_context.h"
34 #include "brw_state.h"
35 #include "main/formats.h"
36 #include "main/fbobject.h"
37 #include "main/samplerobj.h"
38 #include "program/prog_parameter.h"
40 #include "glsl/ralloc.h"
42 /** Return number of src args for given instruction */
43 GLuint
brw_wm_nr_args( GLuint opcode
)
60 assert(opcode
< MAX_OPCODE
);
61 return _mesa_num_inst_src_regs(opcode
);
66 GLuint
brw_wm_is_scalar_result( GLuint opcode
)
90 * Do GPU code generation for non-GLSL shader. non-GLSL shaders have
91 * no flow control instructions so we can more readily do SSA-style
95 brw_wm_non_glsl_emit(struct brw_context
*brw
, struct brw_wm_compile
*c
)
97 /* Augment fragment program. Add instructions for pre- and
98 * post-fragment-program tasks such as interpolation and fogging.
102 /* Translate to intermediate representation. Build register usage
107 /* Dead code removal.
111 /* Register allocation.
112 * Divide by two because we operate on 16 pixels at a time and require
113 * two GRF entries for each logical shader register.
115 c
->grf_limit
= BRW_WM_MAX_GRF
/ 2;
119 /* how many general-purpose registers are used */
120 c
->prog_data
.reg_blocks
= brw_register_blocks(c
->max_wm_grf
);
129 * Return a bitfield where bit n is set if barycentric interpolation mode n
130 * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
133 brw_compute_barycentric_interp_modes(struct brw_context
*brw
,
134 bool shade_model_flat
,
135 const struct gl_fragment_program
*fprog
)
137 unsigned barycentric_interp_modes
= 0;
140 /* Loop through all fragment shader inputs to figure out what interpolation
141 * modes are in use, and set the appropriate bits in
142 * barycentric_interp_modes.
144 for (attr
= 0; attr
< FRAG_ATTRIB_MAX
; ++attr
) {
145 enum glsl_interp_qualifier interp_qualifier
=
146 fprog
->InterpQualifier
[attr
];
147 bool is_centroid
= fprog
->IsCentroid
& BITFIELD64_BIT(attr
);
148 bool is_gl_Color
= attr
== FRAG_ATTRIB_COL0
|| attr
== FRAG_ATTRIB_COL1
;
150 /* Ignore unused inputs. */
151 if (!(fprog
->Base
.InputsRead
& BITFIELD64_BIT(attr
)))
154 /* Ignore WPOS and FACE, because they don't require interpolation. */
155 if (attr
== FRAG_ATTRIB_WPOS
|| attr
== FRAG_ATTRIB_FACE
)
158 /* Determine the set (or sets) of barycentric coordinates needed to
159 * interpolate this variable. Note that when
160 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
161 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
162 * for lit pixels, so we need both sets of barycentric coordinates.
164 if (interp_qualifier
== INTERP_QUALIFIER_NOPERSPECTIVE
) {
166 barycentric_interp_modes
|=
167 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC
;
169 if (!is_centroid
|| brw
->needs_unlit_centroid_workaround
) {
170 barycentric_interp_modes
|=
171 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC
;
173 } else if (interp_qualifier
== INTERP_QUALIFIER_SMOOTH
||
174 (!(shade_model_flat
&& is_gl_Color
) &&
175 interp_qualifier
== INTERP_QUALIFIER_NONE
)) {
177 barycentric_interp_modes
|=
178 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC
;
180 if (!is_centroid
|| brw
->needs_unlit_centroid_workaround
) {
181 barycentric_interp_modes
|=
182 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC
;
187 return barycentric_interp_modes
;
192 brw_wm_payload_setup(struct brw_context
*brw
,
193 struct brw_wm_compile
*c
)
195 struct intel_context
*intel
= &brw
->intel
;
196 bool uses_depth
= (c
->fp
->program
.Base
.InputsRead
&
197 (1 << FRAG_ATTRIB_WPOS
)) != 0;
198 unsigned barycentric_interp_modes
= c
->prog_data
.barycentric_interp_modes
;
201 if (intel
->gen
>= 6) {
202 /* R0-1: masks, pixel X/Y coordinates. */
203 c
->nr_payload_regs
= 2;
204 /* R2: only for 32-pixel dispatch.*/
206 /* R3-26: barycentric interpolation coordinates. These appear in the
207 * same order that they appear in the brw_wm_barycentric_interp_mode
208 * enum. Each set of coordinates occupies 2 registers if dispatch width
209 * == 8 and 4 registers if dispatch width == 16. Coordinates only
210 * appear if they were enabled using the "Barycentric Interpolation
211 * Mode" bits in WM_STATE.
213 for (i
= 0; i
< BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT
; ++i
) {
214 if (barycentric_interp_modes
& (1 << i
)) {
215 c
->barycentric_coord_reg
[i
] = c
->nr_payload_regs
;
216 c
->nr_payload_regs
+= 2;
217 if (c
->dispatch_width
== 16) {
218 c
->nr_payload_regs
+= 2;
223 /* R27: interpolated depth if uses source depth */
225 c
->source_depth_reg
= c
->nr_payload_regs
;
226 c
->nr_payload_regs
++;
227 if (c
->dispatch_width
== 16) {
228 /* R28: interpolated depth if not 8-wide. */
229 c
->nr_payload_regs
++;
232 /* R29: interpolated W set if GEN6_WM_USES_SOURCE_W.
235 c
->source_w_reg
= c
->nr_payload_regs
;
236 c
->nr_payload_regs
++;
237 if (c
->dispatch_width
== 16) {
238 /* R30: interpolated W if not 8-wide. */
239 c
->nr_payload_regs
++;
242 /* R31: MSAA position offsets. */
243 /* R32-: bary for 32-pixel. */
244 /* R58-59: interp W for 32-pixel. */
246 if (c
->fp
->program
.Base
.OutputsWritten
&
247 BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
248 c
->source_depth_to_render_target
= true;
249 c
->computes_depth
= true;
252 brw_wm_lookup_iz(intel
, c
);
257 * All Mesa program -> GPU code generation goes through this function.
258 * Depending on the instructions used (i.e. flow control instructions)
259 * we'll use one of two code generators.
261 bool do_wm_prog(struct brw_context
*brw
,
262 struct gl_shader_program
*prog
,
263 struct brw_fragment_program
*fp
,
264 struct brw_wm_prog_key
*key
)
266 struct intel_context
*intel
= &brw
->intel
;
267 struct brw_wm_compile
*c
;
268 const GLuint
*program
;
271 c
= brw
->wm
.compile_data
;
273 brw
->wm
.compile_data
= rzalloc(NULL
, struct brw_wm_compile
);
274 c
= brw
->wm
.compile_data
;
276 /* Ouch - big out of memory problem. Can't continue
277 * without triggering a segfault, no way to signal,
283 void *instruction
= c
->instruction
;
284 void *prog_instructions
= c
->prog_instructions
;
285 void *vreg
= c
->vreg
;
286 void *refs
= c
->refs
;
287 memset(c
, 0, sizeof(*brw
->wm
.compile_data
));
288 c
->instruction
= instruction
;
289 c
->prog_instructions
= prog_instructions
;
293 memcpy(&c
->key
, key
, sizeof(*key
));
296 c
->env_param
= brw
->intel
.ctx
.FragmentProgram
.Parameters
;
298 brw_init_compile(brw
, &c
->func
, c
);
300 c
->prog_data
.barycentric_interp_modes
=
301 brw_compute_barycentric_interp_modes(brw
, c
->key
.flat_shade
,
304 if (prog
&& prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]) {
305 if (!brw_wm_fs_emit(brw
, c
, prog
))
308 if (!c
->instruction
) {
309 c
->instruction
= rzalloc_array(c
, struct brw_wm_instruction
, BRW_WM_MAX_INSN
);
310 c
->prog_instructions
= rzalloc_array(c
, struct prog_instruction
, BRW_WM_MAX_INSN
);
311 c
->vreg
= rzalloc_array(c
, struct brw_wm_value
, BRW_WM_MAX_VREG
);
312 c
->refs
= rzalloc_array(c
, struct brw_wm_ref
, BRW_WM_MAX_REF
);
315 /* Fallback for fixed function and ARB_fp shaders. */
316 c
->dispatch_width
= 16;
317 brw_wm_payload_setup(brw
, c
);
318 brw_wm_non_glsl_emit(brw
, c
);
319 c
->prog_data
.dispatch_width
= 16;
322 /* Scratch space is used for register spilling */
323 if (c
->last_scratch
) {
324 perf_debug("Fragment shader triggered register spilling. "
325 "Try reducing the number of live scalar values to "
326 "improve performance.\n");
328 c
->prog_data
.total_scratch
= brw_get_scratch_size(c
->last_scratch
);
330 brw_get_scratch_bo(intel
, &brw
->wm
.scratch_bo
,
331 c
->prog_data
.total_scratch
* brw
->max_wm_threads
);
334 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
335 fprintf(stderr
, "\n");
339 program
= brw_get_program(&c
->func
, &program_size
);
341 brw_upload_cache(&brw
->cache
, BRW_WM_PROG
,
342 &c
->key
, sizeof(c
->key
),
343 program
, program_size
,
344 &c
->prog_data
, sizeof(c
->prog_data
),
345 &brw
->wm
.prog_offset
, &brw
->wm
.prog_data
);
351 key_debug(const char *name
, int a
, int b
)
354 perf_debug(" %s %d->%d\n", name
, a
, b
);
362 brw_debug_recompile_sampler_key(const struct brw_sampler_prog_key_data
*old_key
,
363 const struct brw_sampler_prog_key_data
*key
)
367 for (unsigned int i
= 0; i
< BRW_MAX_TEX_UNIT
; i
++) {
368 found
|= key_debug("EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
369 old_key
->swizzles
[i
], key
->swizzles
[i
]);
371 found
|= key_debug("GL_CLAMP enabled on any texture unit's 1st coordinate",
372 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
373 found
|= key_debug("GL_CLAMP enabled on any texture unit's 2nd coordinate",
374 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
375 found
|= key_debug("GL_CLAMP enabled on any texture unit's 3rd coordinate",
376 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
377 found
|= key_debug("GL_MESA_ycbcr texturing\n",
378 old_key
->yuvtex_mask
, key
->yuvtex_mask
);
379 found
|= key_debug("GL_MESA_ycbcr UV swapping\n",
380 old_key
->yuvtex_swap_mask
, key
->yuvtex_swap_mask
);
386 brw_wm_debug_recompile(struct brw_context
*brw
,
387 struct gl_shader_program
*prog
,
388 const struct brw_wm_prog_key
*key
)
390 struct brw_cache_item
*c
= NULL
;
391 const struct brw_wm_prog_key
*old_key
= NULL
;
394 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
396 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
397 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
398 if (c
->cache_id
== BRW_WM_PROG
) {
401 if (old_key
->program_string_id
== key
->program_string_id
)
410 perf_debug(" Didn't find previous compile in the shader cache for "
415 found
|= key_debug("alphatest, computed depth, depth test, or depth write",
416 old_key
->iz_lookup
, key
->iz_lookup
);
417 found
|= key_debug("depth statistics", old_key
->stats_wm
, key
->stats_wm
);
418 found
|= key_debug("flat shading", old_key
->flat_shade
, key
->flat_shade
);
419 found
|= key_debug("number of color buffers", old_key
->nr_color_regions
, key
->nr_color_regions
);
420 found
|= key_debug("rendering to FBO", old_key
->render_to_fbo
, key
->render_to_fbo
);
421 found
|= key_debug("fragment color clamping", old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
422 found
|= key_debug("line smoothing", old_key
->line_aa
, key
->line_aa
);
423 found
|= key_debug("proj_attrib_mask", old_key
->proj_attrib_mask
, key
->proj_attrib_mask
);
424 found
|= key_debug("renderbuffer height", old_key
->drawable_height
, key
->drawable_height
);
425 found
|= key_debug("vertex shader outputs", old_key
->vp_outputs_written
, key
->vp_outputs_written
);
427 found
|= brw_debug_recompile_sampler_key(&old_key
->tex
, &key
->tex
);
430 perf_debug(" Something else\n");
435 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
436 const struct gl_program
*prog
,
437 struct brw_sampler_prog_key_data
*key
)
439 for (int i
= 0; i
< BRW_MAX_TEX_UNIT
; i
++) {
440 key
->swizzles
[i
] = SWIZZLE_NOOP
;
442 if (!prog
->TexturesUsed
[i
])
445 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[i
];
447 if (unit
->_ReallyEnabled
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
448 const struct gl_texture_object
*t
= unit
->_Current
;
449 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
450 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, i
);
451 int swizzles
[SWIZZLE_NIL
+ 1] = {
461 if (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
462 img
->_BaseFormat
== GL_DEPTH_STENCIL
) {
463 /* We handle GL_DEPTH_TEXTURE_MODE here instead of as surface
464 * format overrides because shadow comparison always returns the
465 * result of the comparison in all channels anyway.
467 switch (t
->DepthMode
) {
469 swizzles
[0] = SWIZZLE_ZERO
;
470 swizzles
[1] = SWIZZLE_ZERO
;
471 swizzles
[2] = SWIZZLE_ZERO
;
472 swizzles
[3] = SWIZZLE_X
;
475 swizzles
[0] = SWIZZLE_X
;
476 swizzles
[1] = SWIZZLE_X
;
477 swizzles
[2] = SWIZZLE_X
;
478 swizzles
[3] = SWIZZLE_ONE
;
481 swizzles
[0] = SWIZZLE_X
;
482 swizzles
[1] = SWIZZLE_X
;
483 swizzles
[2] = SWIZZLE_X
;
484 swizzles
[3] = SWIZZLE_X
;
487 swizzles
[0] = SWIZZLE_X
;
488 swizzles
[1] = SWIZZLE_ZERO
;
489 swizzles
[2] = SWIZZLE_ZERO
;
490 swizzles
[3] = SWIZZLE_ONE
;
495 if (img
->InternalFormat
== GL_YCBCR_MESA
) {
496 key
->yuvtex_mask
|= 1 << i
;
497 if (img
->TexFormat
== MESA_FORMAT_YCBCR
)
498 key
->yuvtex_swap_mask
|= 1 << i
;
502 MAKE_SWIZZLE4(swizzles
[GET_SWZ(t
->_Swizzle
, 0)],
503 swizzles
[GET_SWZ(t
->_Swizzle
, 1)],
504 swizzles
[GET_SWZ(t
->_Swizzle
, 2)],
505 swizzles
[GET_SWZ(t
->_Swizzle
, 3)]);
507 if (sampler
->MinFilter
!= GL_NEAREST
&&
508 sampler
->MagFilter
!= GL_NEAREST
) {
509 if (sampler
->WrapS
== GL_CLAMP
)
510 key
->gl_clamp_mask
[0] |= 1 << i
;
511 if (sampler
->WrapT
== GL_CLAMP
)
512 key
->gl_clamp_mask
[1] |= 1 << i
;
513 if (sampler
->WrapR
== GL_CLAMP
)
514 key
->gl_clamp_mask
[2] |= 1 << i
;
520 static void brw_wm_populate_key( struct brw_context
*brw
,
521 struct brw_wm_prog_key
*key
)
523 struct gl_context
*ctx
= &brw
->intel
.ctx
;
524 struct intel_context
*intel
= &brw
->intel
;
525 /* BRW_NEW_FRAGMENT_PROGRAM */
526 const struct brw_fragment_program
*fp
=
527 (struct brw_fragment_program
*)brw
->fragment_program
;
528 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
531 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
533 memset(key
, 0, sizeof(*key
));
535 /* Build the index for table lookup
537 if (intel
->gen
< 6) {
539 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
540 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
542 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
543 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
547 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
549 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
550 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
553 if (ctx
->Stencil
._Enabled
) {
554 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
556 if (ctx
->Stencil
.WriteMask
[0] ||
557 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
558 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
560 key
->iz_lookup
= lookup
;
565 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
566 if (ctx
->Line
.SmoothFlag
) {
567 if (brw
->intel
.reduced_primitive
== GL_LINES
) {
570 else if (brw
->intel
.reduced_primitive
== GL_TRIANGLES
) {
571 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
572 line_aa
= AA_SOMETIMES
;
574 if (ctx
->Polygon
.BackMode
== GL_LINE
||
575 (ctx
->Polygon
.CullFlag
&&
576 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
579 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
580 line_aa
= AA_SOMETIMES
;
582 if ((ctx
->Polygon
.CullFlag
&&
583 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
589 key
->line_aa
= line_aa
;
592 key
->stats_wm
= brw
->intel
.stats_wm
;
594 /* BRW_NEW_WM_INPUT_DIMENSIONS */
595 /* Only set this for fixed function. The optimization it enables isn't
596 * useful for programs using shaders.
598 if (ctx
->Shader
.CurrentFragmentProgram
)
599 key
->proj_attrib_mask
= 0xffffffff;
601 key
->proj_attrib_mask
= brw
->wm
.input_size_masks
[4-1];
604 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
606 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
607 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
610 brw_populate_sampler_prog_key_data(ctx
, prog
, &key
->tex
);
614 * Include the draw buffer origin and height so that we can calculate
615 * fragment position values relative to the bottom left of the drawable,
616 * from the incoming screen origin relative position we get as part of our
619 * This is only needed for the WM_WPOSXY opcode when the fragment program
620 * uses the gl_FragCoord input.
622 * We could avoid recompiling by including this as a constant referenced by
623 * our program, but if we were to do that it would also be nice to handle
624 * getting that constant updated at batchbuffer submit time (when we
625 * hold the lock and know where the buffer really is) rather than at emit
626 * time when we don't hold the lock and are just guessing. We could also
627 * just avoid using this as key data if the program doesn't use
630 * For DRI2 the origin_x/y will always be (0,0) but we still need the
631 * drawable height in order to invert the Y axis.
633 if (fp
->program
.Base
.InputsRead
& FRAG_BIT_WPOS
) {
634 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
637 if ((fp
->program
.Base
.InputsRead
& FRAG_BIT_WPOS
) || program_uses_dfdy
) {
638 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
642 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
643 /* _NEW_MULTISAMPLE */
644 key
->sample_alpha_to_coverage
= ctx
->Multisample
.SampleAlphaToCoverage
;
646 /* CACHE_NEW_VS_PROG */
648 key
->vp_outputs_written
= brw
->vs
.prog_data
->outputs_written
;
650 /* The unique fragment program ID */
651 key
->program_string_id
= fp
->id
;
656 brw_upload_wm_prog(struct brw_context
*brw
)
658 struct intel_context
*intel
= &brw
->intel
;
659 struct gl_context
*ctx
= &intel
->ctx
;
660 struct brw_wm_prog_key key
;
661 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
662 brw
->fragment_program
;
664 brw_wm_populate_key(brw
, &key
);
666 if (!brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
668 &brw
->wm
.prog_offset
, &brw
->wm
.prog_data
)) {
669 bool success
= do_wm_prog(brw
, ctx
->Shader
._CurrentFragmentProgram
, fp
,
677 const struct brw_tracked_state brw_wm_prog
= {
679 .mesa
= (_NEW_COLOR
|
688 .brw
= (BRW_NEW_FRAGMENT_PROGRAM
|
689 BRW_NEW_WM_INPUT_DIMENSIONS
|
690 BRW_NEW_REDUCED_PRIMITIVE
),
691 .cache
= CACHE_NEW_VS_PROG
,
693 .emit
= brw_upload_wm_prog