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>
32 #include "brw_context.h"
34 #include "brw_state.h"
35 #include "main/enums.h"
36 #include "main/formats.h"
37 #include "main/fbobject.h"
38 #include "main/samplerobj.h"
39 #include "program/prog_parameter.h"
40 #include "program/program.h"
41 #include "intel_mipmap_tree.h"
43 #include "glsl/ralloc.h"
46 * Return a bitfield where bit n is set if barycentric interpolation mode n
47 * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
50 brw_compute_barycentric_interp_modes(struct brw_context
*brw
,
51 bool shade_model_flat
,
52 bool persample_shading
,
53 const struct gl_fragment_program
*fprog
)
55 unsigned barycentric_interp_modes
= 0;
58 /* Loop through all fragment shader inputs to figure out what interpolation
59 * modes are in use, and set the appropriate bits in
60 * barycentric_interp_modes.
62 for (attr
= 0; attr
< VARYING_SLOT_MAX
; ++attr
) {
63 enum glsl_interp_qualifier interp_qualifier
=
64 fprog
->InterpQualifier
[attr
];
65 bool is_centroid
= (fprog
->IsCentroid
& BITFIELD64_BIT(attr
)) &&
67 bool is_sample
= (fprog
->IsSample
& BITFIELD64_BIT(attr
)) ||
69 bool is_gl_Color
= attr
== VARYING_SLOT_COL0
|| attr
== VARYING_SLOT_COL1
;
71 /* Ignore unused inputs. */
72 if (!(fprog
->Base
.InputsRead
& BITFIELD64_BIT(attr
)))
75 /* Ignore WPOS and FACE, because they don't require interpolation. */
76 if (attr
== VARYING_SLOT_POS
|| attr
== VARYING_SLOT_FACE
)
79 /* Determine the set (or sets) of barycentric coordinates needed to
80 * interpolate this variable. Note that when
81 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
82 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
83 * for lit pixels, so we need both sets of barycentric coordinates.
85 if (interp_qualifier
== INTERP_QUALIFIER_NOPERSPECTIVE
) {
87 barycentric_interp_modes
|=
88 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC
;
89 } else if (is_sample
) {
90 barycentric_interp_modes
|=
91 1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC
;
93 if ((!is_centroid
&& !is_sample
) ||
94 brw
->needs_unlit_centroid_workaround
) {
95 barycentric_interp_modes
|=
96 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC
;
98 } else if (interp_qualifier
== INTERP_QUALIFIER_SMOOTH
||
99 (!(shade_model_flat
&& is_gl_Color
) &&
100 interp_qualifier
== INTERP_QUALIFIER_NONE
)) {
102 barycentric_interp_modes
|=
103 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC
;
104 } else if (is_sample
) {
105 barycentric_interp_modes
|=
106 1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC
;
108 if ((!is_centroid
&& !is_sample
) ||
109 brw
->needs_unlit_centroid_workaround
) {
110 barycentric_interp_modes
|=
111 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC
;
116 return barycentric_interp_modes
;
120 brw_wm_prog_data_compare(const void *in_a
, const void *in_b
)
122 const struct brw_wm_prog_data
*a
= in_a
;
123 const struct brw_wm_prog_data
*b
= in_b
;
125 /* Compare all the struct (including the base) up to the pointers. */
126 if (memcmp(a
, b
, offsetof(struct brw_wm_prog_data
, param
)))
129 if (memcmp(a
->param
, b
->param
, a
->nr_params
* sizeof(void *)))
132 if (memcmp(a
->pull_param
, b
->pull_param
, a
->nr_pull_params
* sizeof(void *)))
139 brw_wm_prog_data_free(const void *in_prog_data
)
141 const struct brw_wm_prog_data
*prog_data
= in_prog_data
;
143 ralloc_free((void *)prog_data
->param
);
144 ralloc_free((void *)prog_data
->pull_param
);
148 * All Mesa program -> GPU code generation goes through this function.
149 * Depending on the instructions used (i.e. flow control instructions)
150 * we'll use one of two code generators.
152 bool do_wm_prog(struct brw_context
*brw
,
153 struct gl_shader_program
*prog
,
154 struct brw_fragment_program
*fp
,
155 struct brw_wm_prog_key
*key
)
157 struct gl_context
*ctx
= &brw
->ctx
;
158 struct brw_wm_compile
*c
;
159 const GLuint
*program
;
160 struct gl_shader
*fs
= NULL
;
164 fs
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
166 c
= rzalloc(NULL
, struct brw_wm_compile
);
168 /* Allocate the references to the uniforms that will end up in the
169 * prog_data associated with the compiled program, and which will be freed
170 * by the state cache.
174 param_count
= fs
->num_uniform_components
;
176 param_count
= fp
->program
.Base
.Parameters
->NumParameters
* 4;
178 /* The backend also sometimes adds params for texture size. */
179 param_count
+= 2 * ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
;
180 c
->prog_data
.param
= rzalloc_array(NULL
, const float *, param_count
);
181 c
->prog_data
.pull_param
= rzalloc_array(NULL
, const float *, param_count
);
183 memcpy(&c
->key
, key
, sizeof(*key
));
185 c
->prog_data
.barycentric_interp_modes
=
186 brw_compute_barycentric_interp_modes(brw
, c
->key
.flat_shade
,
187 c
->key
.persample_shading
,
190 program
= brw_wm_fs_emit(brw
, c
, &fp
->program
, prog
, &program_size
);
194 /* Scratch space is used for register spilling */
195 if (c
->last_scratch
) {
196 perf_debug("Fragment shader triggered register spilling. "
197 "Try reducing the number of live scalar values to "
198 "improve performance.\n");
200 c
->prog_data
.total_scratch
= brw_get_scratch_size(c
->last_scratch
);
202 brw_get_scratch_bo(brw
, &brw
->wm
.base
.scratch_bo
,
203 c
->prog_data
.total_scratch
* brw
->max_wm_threads
);
206 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
207 fprintf(stderr
, "\n");
209 brw_upload_cache(&brw
->cache
, BRW_WM_PROG
,
210 &c
->key
, sizeof(c
->key
),
211 program
, program_size
,
212 &c
->prog_data
, sizeof(c
->prog_data
),
213 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
);
221 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
224 perf_debug(" %s %d->%d\n", name
, a
, b
);
232 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
233 const struct brw_sampler_prog_key_data
*old_key
,
234 const struct brw_sampler_prog_key_data
*key
)
238 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
239 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
240 old_key
->swizzles
[i
], key
->swizzles
[i
]);
242 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
243 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
244 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
245 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
246 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
247 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
248 found
|= key_debug(brw
, "GL_MESA_ycbcr texturing\n",
249 old_key
->yuvtex_mask
, key
->yuvtex_mask
);
250 found
|= key_debug(brw
, "GL_MESA_ycbcr UV swapping\n",
251 old_key
->yuvtex_swap_mask
, key
->yuvtex_swap_mask
);
252 found
|= key_debug(brw
, "gather channel quirk on any texture unit",
253 old_key
->gather_channel_quirk_mask
, key
->gather_channel_quirk_mask
);
259 brw_wm_debug_recompile(struct brw_context
*brw
,
260 struct gl_shader_program
*prog
,
261 const struct brw_wm_prog_key
*key
)
263 struct brw_cache_item
*c
= NULL
;
264 const struct brw_wm_prog_key
*old_key
= NULL
;
267 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
269 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
270 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
271 if (c
->cache_id
== BRW_WM_PROG
) {
274 if (old_key
->program_string_id
== key
->program_string_id
)
283 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
287 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
289 old_key
->iz_lookup
, key
->iz_lookup
);
290 found
|= key_debug(brw
, "depth statistics",
291 old_key
->stats_wm
, key
->stats_wm
);
292 found
|= key_debug(brw
, "flat shading",
293 old_key
->flat_shade
, key
->flat_shade
);
294 found
|= key_debug(brw
, "number of color buffers",
295 old_key
->nr_color_regions
, key
->nr_color_regions
);
296 found
|= key_debug(brw
, "MRT alpha test or alpha-to-coverage",
297 old_key
->replicate_alpha
, key
->replicate_alpha
);
298 found
|= key_debug(brw
, "rendering to FBO",
299 old_key
->render_to_fbo
, key
->render_to_fbo
);
300 found
|= key_debug(brw
, "fragment color clamping",
301 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
302 found
|= key_debug(brw
, "line smoothing",
303 old_key
->line_aa
, key
->line_aa
);
304 found
|= key_debug(brw
, "renderbuffer height",
305 old_key
->drawable_height
, key
->drawable_height
);
306 found
|= key_debug(brw
, "input slots valid",
307 old_key
->input_slots_valid
, key
->input_slots_valid
);
308 found
|= key_debug(brw
, "mrt alpha test function",
309 old_key
->alpha_test_func
, key
->alpha_test_func
);
310 found
|= key_debug(brw
, "mrt alpha test reference value",
311 old_key
->alpha_test_ref
, key
->alpha_test_ref
);
313 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
316 perf_debug(" Something else\n");
321 gen6_gather_workaround(GLenum internalformat
)
323 switch (internalformat
) {
324 case GL_R8I
: return WA_SIGN
| WA_8BIT
;
325 case GL_R8UI
: return WA_8BIT
;
326 case GL_R16I
: return WA_SIGN
| WA_16BIT
;
327 case GL_R16UI
: return WA_16BIT
;
328 /* note that even though GL_R32I and GL_R32UI have format overrides
329 * in the surface state, there is no shader w/a required */
335 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
336 const struct gl_program
*prog
,
337 unsigned sampler_count
,
338 struct brw_sampler_prog_key_data
*key
)
340 struct brw_context
*brw
= brw_context(ctx
);
342 for (int s
= 0; s
< sampler_count
; s
++) {
343 key
->swizzles
[s
] = SWIZZLE_NOOP
;
345 if (!(prog
->SamplersUsed
& (1 << s
)))
348 int unit_id
= prog
->SamplerUnits
[s
];
349 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
351 if (unit
->_ReallyEnabled
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
352 const struct gl_texture_object
*t
= unit
->_Current
;
353 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
354 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
356 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
357 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
358 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
360 /* Haswell handles texture swizzling as surface format overrides
361 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
363 if (alpha_depth
|| (brw
->gen
< 8 && !brw
->is_haswell
))
364 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
366 if (img
->InternalFormat
== GL_YCBCR_MESA
) {
367 key
->yuvtex_mask
|= 1 << s
;
368 if (img
->TexFormat
== MESA_FORMAT_YCBCR
)
369 key
->yuvtex_swap_mask
|= 1 << s
;
372 if (sampler
->MinFilter
!= GL_NEAREST
&&
373 sampler
->MagFilter
!= GL_NEAREST
) {
374 if (sampler
->WrapS
== GL_CLAMP
)
375 key
->gl_clamp_mask
[0] |= 1 << s
;
376 if (sampler
->WrapT
== GL_CLAMP
)
377 key
->gl_clamp_mask
[1] |= 1 << s
;
378 if (sampler
->WrapR
== GL_CLAMP
)
379 key
->gl_clamp_mask
[2] |= 1 << s
;
382 /* gather4's channel select for green from RG32F is broken;
383 * requires a shader w/a on IVB; fixable with just SCS on HSW. */
384 if (brw
->gen
== 7 && !brw
->is_haswell
&& prog
->UsesGather
) {
385 if (img
->InternalFormat
== GL_RG32F
)
386 key
->gather_channel_quirk_mask
|= 1 << s
;
389 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
390 * UNORM/FLOAT instead and fix it in the shader.
392 if (brw
->gen
== 6 && prog
->UsesGather
) {
393 key
->gen6_gather_wa
[s
] = gen6_gather_workaround(img
->InternalFormat
);
396 /* If this is a multisample sampler, and uses the CMS MSAA layout,
397 * then we need to emit slightly different code to first sample the
400 struct intel_texture_object
*intel_tex
=
401 intel_texture_object((struct gl_texture_object
*)t
);
404 intel_tex
->mt
->msaa_layout
== INTEL_MSAA_LAYOUT_CMS
) {
405 key
->compressed_multisample_layout_mask
|= 1 << s
;
411 static void brw_wm_populate_key( struct brw_context
*brw
,
412 struct brw_wm_prog_key
*key
)
414 struct gl_context
*ctx
= &brw
->ctx
;
415 /* BRW_NEW_FRAGMENT_PROGRAM */
416 const struct brw_fragment_program
*fp
=
417 (struct brw_fragment_program
*)brw
->fragment_program
;
418 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
421 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
422 bool multisample_fbo
= ctx
->DrawBuffer
->Visual
.samples
> 1;
424 memset(key
, 0, sizeof(*key
));
426 /* Build the index for table lookup
430 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
431 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
433 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
434 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
438 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
440 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
441 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
443 /* _NEW_STENCIL | _NEW_BUFFERS */
444 if (ctx
->Stencil
._Enabled
) {
445 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
447 if (ctx
->Stencil
.WriteMask
[0] ||
448 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
449 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
451 key
->iz_lookup
= lookup
;
456 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
457 if (ctx
->Line
.SmoothFlag
) {
458 if (brw
->reduced_primitive
== GL_LINES
) {
461 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
462 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
463 line_aa
= AA_SOMETIMES
;
465 if (ctx
->Polygon
.BackMode
== GL_LINE
||
466 (ctx
->Polygon
.CullFlag
&&
467 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
470 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
471 line_aa
= AA_SOMETIMES
;
473 if ((ctx
->Polygon
.CullFlag
&&
474 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
480 key
->line_aa
= line_aa
;
483 if (brw
->disable_derivative_optimization
) {
484 key
->high_quality_derivatives
=
485 ctx
->Hint
.FragmentShaderDerivative
!= GL_FASTEST
;
487 key
->high_quality_derivatives
=
488 ctx
->Hint
.FragmentShaderDerivative
== GL_NICEST
;
492 key
->stats_wm
= brw
->stats_wm
;
495 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
497 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
498 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
501 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->wm
.base
.sampler_count
,
506 * Include the draw buffer origin and height so that we can calculate
507 * fragment position values relative to the bottom left of the drawable,
508 * from the incoming screen origin relative position we get as part of our
511 * This is only needed for the WM_WPOSXY opcode when the fragment program
512 * uses the gl_FragCoord input.
514 * We could avoid recompiling by including this as a constant referenced by
515 * our program, but if we were to do that it would also be nice to handle
516 * getting that constant updated at batchbuffer submit time (when we
517 * hold the lock and know where the buffer really is) rather than at emit
518 * time when we don't hold the lock and are just guessing. We could also
519 * just avoid using this as key data if the program doesn't use
522 * For DRI2 the origin_x/y will always be (0,0) but we still need the
523 * drawable height in order to invert the Y axis.
525 if (fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) {
526 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
529 if ((fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) || program_uses_dfdy
) {
530 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
534 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
536 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
537 key
->replicate_alpha
= ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
538 (ctx
->Multisample
.SampleAlphaToCoverage
|| ctx
->Color
.AlphaEnabled
);
540 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
541 /* Ignore sample qualifier while computing this flag. */
542 key
->persample_shading
=
543 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, true) > 1;
545 key
->compute_pos_offset
=
546 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, false) > 1 &&
547 fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_POS
;
549 key
->compute_sample_id
=
551 ctx
->Multisample
.Enabled
&&
552 (fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_ID
);
554 /* BRW_NEW_VUE_MAP_GEOM_OUT */
555 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->program
.Base
.InputsRead
&
556 BRW_FS_VARYING_INPUT_MASK
) > 16)
557 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
560 /* _NEW_COLOR | _NEW_BUFFERS */
561 /* Pre-gen6, the hardware alpha test always used each render
562 * target's alpha to do alpha test, as opposed to render target 0's alpha
563 * like GL requires. Fix that by building the alpha test into the
564 * shader, and we'll skip enabling the fixed function alpha test.
566 if (brw
->gen
< 6 && ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 && ctx
->Color
.AlphaEnabled
) {
567 key
->alpha_test_func
= ctx
->Color
.AlphaFunc
;
568 key
->alpha_test_ref
= ctx
->Color
.AlphaRef
;
571 /* The unique fragment program ID */
572 key
->program_string_id
= fp
->id
;
577 brw_upload_wm_prog(struct brw_context
*brw
)
579 struct gl_context
*ctx
= &brw
->ctx
;
580 struct brw_wm_prog_key key
;
581 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
582 brw
->fragment_program
;
584 brw_wm_populate_key(brw
, &key
);
586 if (!brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
588 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
)) {
589 bool success
= do_wm_prog(brw
, ctx
->Shader
._CurrentFragmentProgram
, fp
,
594 brw
->wm
.base
.prog_data
= &brw
->wm
.prog_data
->base
;
598 const struct brw_tracked_state brw_wm_prog
= {
600 .mesa
= (_NEW_COLOR
|
611 .brw
= (BRW_NEW_FRAGMENT_PROGRAM
|
612 BRW_NEW_REDUCED_PRIMITIVE
|
613 BRW_NEW_VUE_MAP_GEOM_OUT
|
616 .emit
= brw_upload_wm_prog