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/enums.h"
36 #include "main/formats.h"
37 #include "main/fbobject.h"
38 #include "main/samplerobj.h"
39 #include "program/prog_parameter.h"
41 #include "glsl/ralloc.h"
44 * Return a bitfield where bit n is set if barycentric interpolation mode n
45 * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
48 brw_compute_barycentric_interp_modes(struct brw_context
*brw
,
49 bool shade_model_flat
,
50 const struct gl_fragment_program
*fprog
)
52 unsigned barycentric_interp_modes
= 0;
55 /* Loop through all fragment shader inputs to figure out what interpolation
56 * modes are in use, and set the appropriate bits in
57 * barycentric_interp_modes.
59 for (attr
= 0; attr
< VARYING_SLOT_MAX
; ++attr
) {
60 enum glsl_interp_qualifier interp_qualifier
=
61 fprog
->InterpQualifier
[attr
];
62 bool is_centroid
= fprog
->IsCentroid
& BITFIELD64_BIT(attr
);
63 bool is_gl_Color
= attr
== VARYING_SLOT_COL0
|| attr
== VARYING_SLOT_COL1
;
65 /* Ignore unused inputs. */
66 if (!(fprog
->Base
.InputsRead
& BITFIELD64_BIT(attr
)))
69 /* Ignore WPOS and FACE, because they don't require interpolation. */
70 if (attr
== VARYING_SLOT_POS
|| attr
== VARYING_SLOT_FACE
)
73 /* Determine the set (or sets) of barycentric coordinates needed to
74 * interpolate this variable. Note that when
75 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
76 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
77 * for lit pixels, so we need both sets of barycentric coordinates.
79 if (interp_qualifier
== INTERP_QUALIFIER_NOPERSPECTIVE
) {
81 barycentric_interp_modes
|=
82 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC
;
84 if (!is_centroid
|| brw
->needs_unlit_centroid_workaround
) {
85 barycentric_interp_modes
|=
86 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC
;
88 } else if (interp_qualifier
== INTERP_QUALIFIER_SMOOTH
||
89 (!(shade_model_flat
&& is_gl_Color
) &&
90 interp_qualifier
== INTERP_QUALIFIER_NONE
)) {
92 barycentric_interp_modes
|=
93 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC
;
95 if (!is_centroid
|| brw
->needs_unlit_centroid_workaround
) {
96 barycentric_interp_modes
|=
97 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC
;
102 return barycentric_interp_modes
;
106 brw_wm_prog_data_compare(const void *in_a
, const void *in_b
,
107 int aux_size
, const void *in_key
)
109 const struct brw_wm_prog_data
*a
= in_a
;
110 const struct brw_wm_prog_data
*b
= in_b
;
112 /* Compare all the struct up to the pointers. */
113 if (memcmp(a
, b
, offsetof(struct brw_wm_prog_data
, param
)))
116 if (memcmp(a
->param
, b
->param
, a
->nr_params
* sizeof(void *)))
119 if (memcmp(a
->pull_param
, b
->pull_param
, a
->nr_pull_params
* sizeof(void *)))
126 brw_wm_prog_data_free(const void *in_prog_data
)
128 const struct brw_wm_prog_data
*prog_data
= in_prog_data
;
130 ralloc_free((void *)prog_data
->param
);
131 ralloc_free((void *)prog_data
->pull_param
);
135 * All Mesa program -> GPU code generation goes through this function.
136 * Depending on the instructions used (i.e. flow control instructions)
137 * we'll use one of two code generators.
139 bool do_wm_prog(struct brw_context
*brw
,
140 struct gl_shader_program
*prog
,
141 struct brw_fragment_program
*fp
,
142 struct brw_wm_prog_key
*key
)
144 struct brw_wm_compile
*c
;
145 const GLuint
*program
;
146 struct gl_shader
*fs
= NULL
;
150 fs
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
152 c
= rzalloc(NULL
, struct brw_wm_compile
);
154 /* Allocate the references to the uniforms that will end up in the
155 * prog_data associated with the compiled program, and which will be freed
156 * by the state cache.
160 param_count
= fs
->num_uniform_components
;
162 param_count
= fp
->program
.Base
.Parameters
->NumParameters
* 4;
164 /* The backend also sometimes adds params for texture size. */
165 param_count
+= 2 * BRW_MAX_TEX_UNIT
;
166 c
->prog_data
.param
= rzalloc_array(NULL
, const float *, param_count
);
167 c
->prog_data
.pull_param
= rzalloc_array(NULL
, const float *, param_count
);
169 memcpy(&c
->key
, key
, sizeof(*key
));
171 c
->prog_data
.barycentric_interp_modes
=
172 brw_compute_barycentric_interp_modes(brw
, c
->key
.flat_shade
,
175 program
= brw_wm_fs_emit(brw
, c
, &fp
->program
, prog
, &program_size
);
179 /* Scratch space is used for register spilling */
180 if (c
->last_scratch
) {
181 perf_debug("Fragment shader triggered register spilling. "
182 "Try reducing the number of live scalar values to "
183 "improve performance.\n");
185 c
->prog_data
.total_scratch
= brw_get_scratch_size(c
->last_scratch
);
187 brw_get_scratch_bo(brw
, &brw
->wm
.base
.scratch_bo
,
188 c
->prog_data
.total_scratch
* brw
->max_wm_threads
);
191 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
192 fprintf(stderr
, "\n");
194 brw_upload_cache(&brw
->cache
, BRW_WM_PROG
,
195 &c
->key
, sizeof(c
->key
),
196 program
, program_size
,
197 &c
->prog_data
, sizeof(c
->prog_data
),
198 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
);
206 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
209 perf_debug(" %s %d->%d\n", name
, a
, b
);
217 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
218 const struct brw_sampler_prog_key_data
*old_key
,
219 const struct brw_sampler_prog_key_data
*key
)
223 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
224 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
225 old_key
->swizzles
[i
], key
->swizzles
[i
]);
227 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
228 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
229 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
230 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
231 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
232 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
233 found
|= key_debug(brw
, "GL_MESA_ycbcr texturing\n",
234 old_key
->yuvtex_mask
, key
->yuvtex_mask
);
235 found
|= key_debug(brw
, "GL_MESA_ycbcr UV swapping\n",
236 old_key
->yuvtex_swap_mask
, key
->yuvtex_swap_mask
);
237 found
|= key_debug(brw
, "gather channel quirk on any texture unit",
238 old_key
->gather_channel_quirk_mask
, key
->gather_channel_quirk_mask
);
244 brw_wm_debug_recompile(struct brw_context
*brw
,
245 struct gl_shader_program
*prog
,
246 const struct brw_wm_prog_key
*key
)
248 struct brw_cache_item
*c
= NULL
;
249 const struct brw_wm_prog_key
*old_key
= NULL
;
252 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
254 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
255 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
256 if (c
->cache_id
== BRW_WM_PROG
) {
259 if (old_key
->program_string_id
== key
->program_string_id
)
268 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
272 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
274 old_key
->iz_lookup
, key
->iz_lookup
);
275 found
|= key_debug(brw
, "depth statistics",
276 old_key
->stats_wm
, key
->stats_wm
);
277 found
|= key_debug(brw
, "flat shading",
278 old_key
->flat_shade
, key
->flat_shade
);
279 found
|= key_debug(brw
, "number of color buffers",
280 old_key
->nr_color_regions
, key
->nr_color_regions
);
281 found
|= key_debug(brw
, "MRT alpha test or alpha-to-coverage",
282 old_key
->replicate_alpha
, key
->replicate_alpha
);
283 found
|= key_debug(brw
, "rendering to FBO",
284 old_key
->render_to_fbo
, key
->render_to_fbo
);
285 found
|= key_debug(brw
, "fragment color clamping",
286 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
287 found
|= key_debug(brw
, "line smoothing",
288 old_key
->line_aa
, key
->line_aa
);
289 found
|= key_debug(brw
, "renderbuffer height",
290 old_key
->drawable_height
, key
->drawable_height
);
291 found
|= key_debug(brw
, "input slots valid",
292 old_key
->input_slots_valid
, key
->input_slots_valid
);
294 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
297 perf_debug(" Something else\n");
302 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
303 const struct gl_program
*prog
,
304 unsigned sampler_count
,
305 struct brw_sampler_prog_key_data
*key
)
307 struct brw_context
*brw
= brw_context(ctx
);
309 for (int s
= 0; s
< sampler_count
; s
++) {
310 key
->swizzles
[s
] = SWIZZLE_NOOP
;
312 if (!(prog
->SamplersUsed
& (1 << s
)))
315 int unit_id
= prog
->SamplerUnits
[s
];
316 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
318 if (unit
->_ReallyEnabled
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
319 const struct gl_texture_object
*t
= unit
->_Current
;
320 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
321 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
323 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
324 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
325 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
327 /* Haswell handles texture swizzling as surface format overrides
328 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
330 if (!brw
->is_haswell
|| alpha_depth
)
331 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
333 if (img
->InternalFormat
== GL_YCBCR_MESA
) {
334 key
->yuvtex_mask
|= 1 << s
;
335 if (img
->TexFormat
== MESA_FORMAT_YCBCR
)
336 key
->yuvtex_swap_mask
|= 1 << s
;
339 if (sampler
->MinFilter
!= GL_NEAREST
&&
340 sampler
->MagFilter
!= GL_NEAREST
) {
341 if (sampler
->WrapS
== GL_CLAMP
)
342 key
->gl_clamp_mask
[0] |= 1 << s
;
343 if (sampler
->WrapT
== GL_CLAMP
)
344 key
->gl_clamp_mask
[1] |= 1 << s
;
345 if (sampler
->WrapR
== GL_CLAMP
)
346 key
->gl_clamp_mask
[2] |= 1 << s
;
349 /* gather4's channel select for green from RG32F is broken;
350 * requires a shader w/a on IVB; fixable with just SCS on HSW. */
351 if (brw
->gen
>= 7 && !brw
->is_haswell
&& prog
->UsesGather
) {
352 if (img
->InternalFormat
== GL_RG32F
)
353 key
->gather_channel_quirk_mask
|= 1 << s
;
359 static void brw_wm_populate_key( struct brw_context
*brw
,
360 struct brw_wm_prog_key
*key
)
362 struct gl_context
*ctx
= &brw
->ctx
;
363 /* BRW_NEW_FRAGMENT_PROGRAM */
364 const struct brw_fragment_program
*fp
=
365 (struct brw_fragment_program
*)brw
->fragment_program
;
366 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
369 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
371 memset(key
, 0, sizeof(*key
));
373 /* Build the index for table lookup
377 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
378 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
380 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
381 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
385 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
387 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
388 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
390 /* _NEW_STENCIL | _NEW_BUFFERS */
391 if (ctx
->Stencil
._Enabled
) {
392 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
394 if (ctx
->Stencil
.WriteMask
[0] ||
395 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
396 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
398 key
->iz_lookup
= lookup
;
403 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
404 if (ctx
->Line
.SmoothFlag
) {
405 if (brw
->reduced_primitive
== GL_LINES
) {
408 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
409 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
410 line_aa
= AA_SOMETIMES
;
412 if (ctx
->Polygon
.BackMode
== GL_LINE
||
413 (ctx
->Polygon
.CullFlag
&&
414 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
417 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
418 line_aa
= AA_SOMETIMES
;
420 if ((ctx
->Polygon
.CullFlag
&&
421 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
427 key
->line_aa
= line_aa
;
430 if (brw
->disable_derivative_optimization
) {
431 key
->high_quality_derivatives
=
432 ctx
->Hint
.FragmentShaderDerivative
!= GL_FASTEST
;
434 key
->high_quality_derivatives
=
435 ctx
->Hint
.FragmentShaderDerivative
== GL_NICEST
;
439 key
->stats_wm
= brw
->stats_wm
;
442 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
444 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
445 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
448 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->wm
.base
.sampler_count
,
453 * Include the draw buffer origin and height so that we can calculate
454 * fragment position values relative to the bottom left of the drawable,
455 * from the incoming screen origin relative position we get as part of our
458 * This is only needed for the WM_WPOSXY opcode when the fragment program
459 * uses the gl_FragCoord input.
461 * We could avoid recompiling by including this as a constant referenced by
462 * our program, but if we were to do that it would also be nice to handle
463 * getting that constant updated at batchbuffer submit time (when we
464 * hold the lock and know where the buffer really is) rather than at emit
465 * time when we don't hold the lock and are just guessing. We could also
466 * just avoid using this as key data if the program doesn't use
469 * For DRI2 the origin_x/y will always be (0,0) but we still need the
470 * drawable height in order to invert the Y axis.
472 if (fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) {
473 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
476 if ((fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) || program_uses_dfdy
) {
477 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
481 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
483 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
484 key
->replicate_alpha
= ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
485 (ctx
->Multisample
.SampleAlphaToCoverage
|| ctx
->Color
.AlphaEnabled
);
487 /* BRW_NEW_VUE_MAP_GEOM_OUT */
488 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->program
.Base
.InputsRead
&
489 BRW_FS_VARYING_INPUT_MASK
) > 16)
490 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
492 /* The unique fragment program ID */
493 key
->program_string_id
= fp
->id
;
498 brw_upload_wm_prog(struct brw_context
*brw
)
500 struct gl_context
*ctx
= &brw
->ctx
;
501 struct brw_wm_prog_key key
;
502 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
503 brw
->fragment_program
;
505 brw_wm_populate_key(brw
, &key
);
507 if (!brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
509 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
)) {
510 bool success
= do_wm_prog(brw
, ctx
->Shader
._CurrentFragmentProgram
, fp
,
518 const struct brw_tracked_state brw_wm_prog
= {
520 .mesa
= (_NEW_COLOR
|
531 .brw
= (BRW_NEW_FRAGMENT_PROGRAM
|
532 BRW_NEW_REDUCED_PRIMITIVE
|
533 BRW_NEW_VUE_MAP_GEOM_OUT
|
536 .emit
= brw_upload_wm_prog