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 the base structure. */
126 if (!brw_stage_prog_data_compare(&a
->base
, &b
->base
))
129 /* Compare the rest of the structure. */
130 const unsigned offset
= sizeof(struct brw_stage_prog_data
);
131 if (memcmp(((char *) a
) + offset
, ((char *) b
) + offset
,
132 sizeof(struct brw_wm_prog_data
) - offset
))
139 * All Mesa program -> GPU code generation goes through this function.
140 * Depending on the instructions used (i.e. flow control instructions)
141 * we'll use one of two code generators.
143 bool do_wm_prog(struct brw_context
*brw
,
144 struct gl_shader_program
*prog
,
145 struct brw_fragment_program
*fp
,
146 struct brw_wm_prog_key
*key
)
148 struct gl_context
*ctx
= &brw
->ctx
;
149 void *mem_ctx
= ralloc_context(NULL
);
150 struct brw_wm_compile
*c
;
151 const GLuint
*program
;
152 struct gl_shader
*fs
= NULL
;
156 fs
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
158 c
= rzalloc(mem_ctx
, struct brw_wm_compile
);
160 /* Allocate the references to the uniforms that will end up in the
161 * prog_data associated with the compiled program, and which will be freed
162 * by the state cache.
166 param_count
= fs
->num_uniform_components
;
168 param_count
= fp
->program
.Base
.Parameters
->NumParameters
* 4;
170 /* The backend also sometimes adds params for texture size. */
171 param_count
+= 2 * ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
;
172 c
->prog_data
.base
.param
= rzalloc_array(NULL
, const float *, param_count
);
173 c
->prog_data
.base
.pull_param
=
174 rzalloc_array(NULL
, const float *, param_count
);
175 c
->prog_data
.base
.nr_params
= param_count
;
177 c
->prog_data
.barycentric_interp_modes
=
178 brw_compute_barycentric_interp_modes(brw
, key
->flat_shade
,
179 key
->persample_shading
,
182 program
= brw_wm_fs_emit(brw
, mem_ctx
, key
, &c
->prog_data
,
183 &fp
->program
, prog
, &program_size
);
184 if (program
== NULL
) {
185 ralloc_free(mem_ctx
);
189 if (c
->prog_data
.total_scratch
) {
190 brw_get_scratch_bo(brw
, &brw
->wm
.base
.scratch_bo
,
191 c
->prog_data
.total_scratch
* brw
->max_wm_threads
);
194 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
195 fprintf(stderr
, "\n");
197 brw_upload_cache(&brw
->cache
, BRW_WM_PROG
,
198 key
, sizeof(struct brw_wm_prog_key
),
199 program
, program_size
,
200 &c
->prog_data
, sizeof(c
->prog_data
),
201 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
);
203 ralloc_free(mem_ctx
);
209 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
212 perf_debug(" %s %d->%d\n", name
, a
, b
);
220 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
221 const struct brw_sampler_prog_key_data
*old_key
,
222 const struct brw_sampler_prog_key_data
*key
)
226 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
227 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
228 old_key
->swizzles
[i
], key
->swizzles
[i
]);
230 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
231 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
232 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
233 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
234 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
235 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
236 found
|= key_debug(brw
, "gather channel quirk on any texture unit",
237 old_key
->gather_channel_quirk_mask
, key
->gather_channel_quirk_mask
);
243 brw_wm_debug_recompile(struct brw_context
*brw
,
244 struct gl_shader_program
*prog
,
245 const struct brw_wm_prog_key
*key
)
247 struct brw_cache_item
*c
= NULL
;
248 const struct brw_wm_prog_key
*old_key
= NULL
;
251 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
253 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
254 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
255 if (c
->cache_id
== BRW_WM_PROG
) {
258 if (old_key
->program_string_id
== key
->program_string_id
)
267 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
271 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
273 old_key
->iz_lookup
, key
->iz_lookup
);
274 found
|= key_debug(brw
, "depth statistics",
275 old_key
->stats_wm
, key
->stats_wm
);
276 found
|= key_debug(brw
, "flat shading",
277 old_key
->flat_shade
, key
->flat_shade
);
278 found
|= key_debug(brw
, "number of color buffers",
279 old_key
->nr_color_regions
, key
->nr_color_regions
);
280 found
|= key_debug(brw
, "MRT alpha test or alpha-to-coverage",
281 old_key
->replicate_alpha
, key
->replicate_alpha
);
282 found
|= key_debug(brw
, "rendering to FBO",
283 old_key
->render_to_fbo
, key
->render_to_fbo
);
284 found
|= key_debug(brw
, "fragment color clamping",
285 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
286 found
|= key_debug(brw
, "line smoothing",
287 old_key
->line_aa
, key
->line_aa
);
288 found
|= key_debug(brw
, "renderbuffer height",
289 old_key
->drawable_height
, key
->drawable_height
);
290 found
|= key_debug(brw
, "input slots valid",
291 old_key
->input_slots_valid
, key
->input_slots_valid
);
292 found
|= key_debug(brw
, "mrt alpha test function",
293 old_key
->alpha_test_func
, key
->alpha_test_func
);
294 found
|= key_debug(brw
, "mrt alpha test reference value",
295 old_key
->alpha_test_ref
, key
->alpha_test_ref
);
297 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
300 perf_debug(" Something else\n");
305 gen6_gather_workaround(GLenum internalformat
)
307 switch (internalformat
) {
308 case GL_R8I
: return WA_SIGN
| WA_8BIT
;
309 case GL_R8UI
: return WA_8BIT
;
310 case GL_R16I
: return WA_SIGN
| WA_16BIT
;
311 case GL_R16UI
: return WA_16BIT
;
312 /* note that even though GL_R32I and GL_R32UI have format overrides
313 * in the surface state, there is no shader w/a required */
319 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
320 const struct gl_program
*prog
,
321 unsigned sampler_count
,
322 struct brw_sampler_prog_key_data
*key
)
324 struct brw_context
*brw
= brw_context(ctx
);
326 for (int s
= 0; s
< sampler_count
; s
++) {
327 key
->swizzles
[s
] = SWIZZLE_NOOP
;
329 if (!(prog
->SamplersUsed
& (1 << s
)))
332 int unit_id
= prog
->SamplerUnits
[s
];
333 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
335 if (unit
->_Current
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
336 const struct gl_texture_object
*t
= unit
->_Current
;
337 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
338 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
340 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
341 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
342 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
344 /* Haswell handles texture swizzling as surface format overrides
345 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
347 if (alpha_depth
|| (brw
->gen
< 8 && !brw
->is_haswell
))
348 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
350 if (sampler
->MinFilter
!= GL_NEAREST
&&
351 sampler
->MagFilter
!= GL_NEAREST
) {
352 if (sampler
->WrapS
== GL_CLAMP
)
353 key
->gl_clamp_mask
[0] |= 1 << s
;
354 if (sampler
->WrapT
== GL_CLAMP
)
355 key
->gl_clamp_mask
[1] |= 1 << s
;
356 if (sampler
->WrapR
== GL_CLAMP
)
357 key
->gl_clamp_mask
[2] |= 1 << s
;
360 /* gather4's channel select for green from RG32F is broken;
361 * requires a shader w/a on IVB; fixable with just SCS on HSW. */
362 if (brw
->gen
== 7 && !brw
->is_haswell
&& prog
->UsesGather
) {
363 if (img
->InternalFormat
== GL_RG32F
)
364 key
->gather_channel_quirk_mask
|= 1 << s
;
367 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
368 * UNORM/FLOAT instead and fix it in the shader.
370 if (brw
->gen
== 6 && prog
->UsesGather
) {
371 key
->gen6_gather_wa
[s
] = gen6_gather_workaround(img
->InternalFormat
);
374 /* If this is a multisample sampler, and uses the CMS MSAA layout,
375 * then we need to emit slightly different code to first sample the
378 struct intel_texture_object
*intel_tex
=
379 intel_texture_object((struct gl_texture_object
*)t
);
382 intel_tex
->mt
->msaa_layout
== INTEL_MSAA_LAYOUT_CMS
) {
383 key
->compressed_multisample_layout_mask
|= 1 << s
;
389 static void brw_wm_populate_key( struct brw_context
*brw
,
390 struct brw_wm_prog_key
*key
)
392 struct gl_context
*ctx
= &brw
->ctx
;
393 /* BRW_NEW_FRAGMENT_PROGRAM */
394 const struct brw_fragment_program
*fp
=
395 (struct brw_fragment_program
*)brw
->fragment_program
;
396 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
399 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
400 bool multisample_fbo
= ctx
->DrawBuffer
->Visual
.samples
> 1;
402 memset(key
, 0, sizeof(*key
));
404 /* Build the index for table lookup
408 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
409 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
411 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
412 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
416 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
418 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
419 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
421 /* _NEW_STENCIL | _NEW_BUFFERS */
422 if (ctx
->Stencil
._Enabled
) {
423 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
425 if (ctx
->Stencil
.WriteMask
[0] ||
426 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
427 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
429 key
->iz_lookup
= lookup
;
434 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
435 if (ctx
->Line
.SmoothFlag
) {
436 if (brw
->reduced_primitive
== GL_LINES
) {
439 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
440 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
441 line_aa
= AA_SOMETIMES
;
443 if (ctx
->Polygon
.BackMode
== GL_LINE
||
444 (ctx
->Polygon
.CullFlag
&&
445 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
448 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
449 line_aa
= AA_SOMETIMES
;
451 if ((ctx
->Polygon
.CullFlag
&&
452 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
458 key
->line_aa
= line_aa
;
461 if (brw
->disable_derivative_optimization
) {
462 key
->high_quality_derivatives
=
463 ctx
->Hint
.FragmentShaderDerivative
!= GL_FASTEST
;
465 key
->high_quality_derivatives
=
466 ctx
->Hint
.FragmentShaderDerivative
== GL_NICEST
;
470 key
->stats_wm
= brw
->stats_wm
;
473 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
475 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
476 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
479 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->wm
.base
.sampler_count
,
484 * Include the draw buffer origin and height so that we can calculate
485 * fragment position values relative to the bottom left of the drawable,
486 * from the incoming screen origin relative position we get as part of our
489 * This is only needed for the WM_WPOSXY opcode when the fragment program
490 * uses the gl_FragCoord input.
492 * We could avoid recompiling by including this as a constant referenced by
493 * our program, but if we were to do that it would also be nice to handle
494 * getting that constant updated at batchbuffer submit time (when we
495 * hold the lock and know where the buffer really is) rather than at emit
496 * time when we don't hold the lock and are just guessing. We could also
497 * just avoid using this as key data if the program doesn't use
500 * For DRI2 the origin_x/y will always be (0,0) but we still need the
501 * drawable height in order to invert the Y axis.
503 if (fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) {
504 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
507 if ((fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) || program_uses_dfdy
) {
508 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
512 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
514 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
515 key
->replicate_alpha
= ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
516 (ctx
->Multisample
.SampleAlphaToCoverage
|| ctx
->Color
.AlphaEnabled
);
518 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
519 /* Ignore sample qualifier while computing this flag. */
520 key
->persample_shading
=
521 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, true) > 1;
523 key
->compute_pos_offset
=
524 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, false) > 1 &&
525 fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_POS
;
527 key
->compute_sample_id
=
529 ctx
->Multisample
.Enabled
&&
530 (fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_ID
);
532 /* BRW_NEW_VUE_MAP_GEOM_OUT */
533 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->program
.Base
.InputsRead
&
534 BRW_FS_VARYING_INPUT_MASK
) > 16)
535 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
538 /* _NEW_COLOR | _NEW_BUFFERS */
539 /* Pre-gen6, the hardware alpha test always used each render
540 * target's alpha to do alpha test, as opposed to render target 0's alpha
541 * like GL requires. Fix that by building the alpha test into the
542 * shader, and we'll skip enabling the fixed function alpha test.
544 if (brw
->gen
< 6 && ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 && ctx
->Color
.AlphaEnabled
) {
545 key
->alpha_test_func
= ctx
->Color
.AlphaFunc
;
546 key
->alpha_test_ref
= ctx
->Color
.AlphaRef
;
549 /* The unique fragment program ID */
550 key
->program_string_id
= fp
->id
;
555 brw_upload_wm_prog(struct brw_context
*brw
)
557 struct gl_context
*ctx
= &brw
->ctx
;
558 struct brw_wm_prog_key key
;
559 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
560 brw
->fragment_program
;
562 brw_wm_populate_key(brw
, &key
);
564 if (!brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
566 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
)) {
567 bool success
= do_wm_prog(brw
, ctx
->_Shader
->_CurrentFragmentProgram
, fp
,
572 brw
->wm
.base
.prog_data
= &brw
->wm
.prog_data
->base
;
576 const struct brw_tracked_state brw_wm_prog
= {
578 .mesa
= (_NEW_COLOR
|
589 .brw
= (BRW_NEW_FRAGMENT_PROGRAM
|
590 BRW_NEW_REDUCED_PRIMITIVE
|
591 BRW_NEW_VUE_MAP_GEOM_OUT
|
594 .emit
= brw_upload_wm_prog