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"
43 * Return a bitfield where bit n is set if barycentric interpolation mode n
44 * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
47 brw_compute_barycentric_interp_modes(struct brw_context
*brw
,
48 bool shade_model_flat
,
49 const struct gl_fragment_program
*fprog
)
51 unsigned barycentric_interp_modes
= 0;
54 /* Loop through all fragment shader inputs to figure out what interpolation
55 * modes are in use, and set the appropriate bits in
56 * barycentric_interp_modes.
58 for (attr
= 0; attr
< VARYING_SLOT_MAX
; ++attr
) {
59 enum glsl_interp_qualifier interp_qualifier
=
60 fprog
->InterpQualifier
[attr
];
61 bool is_centroid
= fprog
->IsCentroid
& BITFIELD64_BIT(attr
);
62 bool is_gl_Color
= attr
== VARYING_SLOT_COL0
|| attr
== VARYING_SLOT_COL1
;
64 /* Ignore unused inputs. */
65 if (!(fprog
->Base
.InputsRead
& BITFIELD64_BIT(attr
)))
68 /* Ignore WPOS and FACE, because they don't require interpolation. */
69 if (attr
== VARYING_SLOT_POS
|| attr
== VARYING_SLOT_FACE
)
72 /* Determine the set (or sets) of barycentric coordinates needed to
73 * interpolate this variable. Note that when
74 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
75 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
76 * for lit pixels, so we need both sets of barycentric coordinates.
78 if (interp_qualifier
== INTERP_QUALIFIER_NOPERSPECTIVE
) {
80 barycentric_interp_modes
|=
81 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC
;
83 if (!is_centroid
|| brw
->needs_unlit_centroid_workaround
) {
84 barycentric_interp_modes
|=
85 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC
;
87 } else if (interp_qualifier
== INTERP_QUALIFIER_SMOOTH
||
88 (!(shade_model_flat
&& is_gl_Color
) &&
89 interp_qualifier
== INTERP_QUALIFIER_NONE
)) {
91 barycentric_interp_modes
|=
92 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC
;
94 if (!is_centroid
|| brw
->needs_unlit_centroid_workaround
) {
95 barycentric_interp_modes
|=
96 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC
;
101 return barycentric_interp_modes
;
105 brw_wm_prog_data_compare(const void *in_a
, const void *in_b
,
106 int aux_size
, const void *in_key
)
108 const struct brw_wm_prog_data
*a
= in_a
;
109 const struct brw_wm_prog_data
*b
= in_b
;
111 /* Compare all the struct up to the pointers. */
112 if (memcmp(a
, b
, offsetof(struct brw_wm_prog_data
, param
)))
115 if (memcmp(a
->param
, b
->param
, a
->nr_params
* sizeof(void *)))
118 if (memcmp(a
->pull_param
, b
->pull_param
, a
->nr_pull_params
* sizeof(void *)))
125 brw_wm_prog_data_free(const void *in_prog_data
)
127 const struct brw_wm_prog_data
*prog_data
= in_prog_data
;
129 ralloc_free((void *)prog_data
->param
);
130 ralloc_free((void *)prog_data
->pull_param
);
134 * All Mesa program -> GPU code generation goes through this function.
135 * Depending on the instructions used (i.e. flow control instructions)
136 * we'll use one of two code generators.
138 bool do_wm_prog(struct brw_context
*brw
,
139 struct gl_shader_program
*prog
,
140 struct brw_fragment_program
*fp
,
141 struct brw_wm_prog_key
*key
)
143 struct brw_wm_compile
*c
;
144 const GLuint
*program
;
145 struct gl_shader
*fs
= NULL
;
149 fs
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
151 c
= rzalloc(NULL
, struct brw_wm_compile
);
153 /* Allocate the references to the uniforms that will end up in the
154 * prog_data associated with the compiled program, and which will be freed
155 * by the state cache.
159 param_count
= fs
->num_uniform_components
;
161 param_count
= fp
->program
.Base
.Parameters
->NumParameters
* 4;
163 /* The backend also sometimes adds params for texture size. */
164 param_count
+= 2 * BRW_MAX_TEX_UNIT
;
165 c
->prog_data
.param
= rzalloc_array(NULL
, const float *, param_count
);
166 c
->prog_data
.pull_param
= rzalloc_array(NULL
, const float *, param_count
);
168 memcpy(&c
->key
, key
, sizeof(*key
));
170 c
->prog_data
.barycentric_interp_modes
=
171 brw_compute_barycentric_interp_modes(brw
, c
->key
.flat_shade
,
174 program
= brw_wm_fs_emit(brw
, c
, &fp
->program
, prog
, &program_size
);
178 /* Scratch space is used for register spilling */
179 if (c
->last_scratch
) {
180 perf_debug("Fragment shader triggered register spilling. "
181 "Try reducing the number of live scalar values to "
182 "improve performance.\n");
184 c
->prog_data
.total_scratch
= brw_get_scratch_size(c
->last_scratch
);
186 brw_get_scratch_bo(brw
, &brw
->wm
.scratch_bo
,
187 c
->prog_data
.total_scratch
* brw
->max_wm_threads
);
190 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
191 fprintf(stderr
, "\n");
193 brw_upload_cache(&brw
->cache
, BRW_WM_PROG
,
194 &c
->key
, sizeof(c
->key
),
195 program
, program_size
,
196 &c
->prog_data
, sizeof(c
->prog_data
),
197 &brw
->wm
.prog_offset
, &brw
->wm
.prog_data
);
205 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
208 perf_debug(" %s %d->%d\n", name
, a
, b
);
216 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
217 const struct brw_sampler_prog_key_data
*old_key
,
218 const struct brw_sampler_prog_key_data
*key
)
222 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
223 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
224 old_key
->swizzles
[i
], key
->swizzles
[i
]);
226 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
227 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
228 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
229 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
230 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
231 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
232 found
|= key_debug(brw
, "GL_MESA_ycbcr texturing\n",
233 old_key
->yuvtex_mask
, key
->yuvtex_mask
);
234 found
|= key_debug(brw
, "GL_MESA_ycbcr UV swapping\n",
235 old_key
->yuvtex_swap_mask
, key
->yuvtex_swap_mask
);
241 brw_wm_debug_recompile(struct brw_context
*brw
,
242 struct gl_shader_program
*prog
,
243 const struct brw_wm_prog_key
*key
)
245 struct brw_cache_item
*c
= NULL
;
246 const struct brw_wm_prog_key
*old_key
= NULL
;
249 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
251 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
252 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
253 if (c
->cache_id
== BRW_WM_PROG
) {
256 if (old_key
->program_string_id
== key
->program_string_id
)
265 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
269 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
271 old_key
->iz_lookup
, key
->iz_lookup
);
272 found
|= key_debug(brw
, "depth statistics",
273 old_key
->stats_wm
, key
->stats_wm
);
274 found
|= key_debug(brw
, "flat shading",
275 old_key
->flat_shade
, key
->flat_shade
);
276 found
|= key_debug(brw
, "number of color buffers",
277 old_key
->nr_color_regions
, key
->nr_color_regions
);
278 found
|= key_debug(brw
, "MRT alpha test or alpha-to-coverage",
279 old_key
->replicate_alpha
, key
->replicate_alpha
);
280 found
|= key_debug(brw
, "rendering to FBO",
281 old_key
->render_to_fbo
, key
->render_to_fbo
);
282 found
|= key_debug(brw
, "fragment color clamping",
283 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
284 found
|= key_debug(brw
, "line smoothing",
285 old_key
->line_aa
, key
->line_aa
);
286 found
|= key_debug(brw
, "renderbuffer height",
287 old_key
->drawable_height
, key
->drawable_height
);
288 found
|= key_debug(brw
, "input slots valid",
289 old_key
->input_slots_valid
, key
->input_slots_valid
);
291 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
294 perf_debug(" Something else\n");
299 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
300 const struct gl_program
*prog
,
301 unsigned sampler_count
,
302 struct brw_sampler_prog_key_data
*key
)
304 struct brw_context
*brw
= brw_context(ctx
);
306 for (int s
= 0; s
< sampler_count
; s
++) {
307 key
->swizzles
[s
] = SWIZZLE_NOOP
;
309 if (!(prog
->SamplersUsed
& (1 << s
)))
312 int unit_id
= prog
->SamplerUnits
[s
];
313 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
315 if (unit
->_ReallyEnabled
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
316 const struct gl_texture_object
*t
= unit
->_Current
;
317 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
318 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
320 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
321 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
322 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
324 /* Haswell handles texture swizzling as surface format overrides
325 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
327 if (!brw
->is_haswell
|| alpha_depth
)
328 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
330 if (img
->InternalFormat
== GL_YCBCR_MESA
) {
331 key
->yuvtex_mask
|= 1 << s
;
332 if (img
->TexFormat
== MESA_FORMAT_YCBCR
)
333 key
->yuvtex_swap_mask
|= 1 << s
;
336 if (sampler
->MinFilter
!= GL_NEAREST
&&
337 sampler
->MagFilter
!= GL_NEAREST
) {
338 if (sampler
->WrapS
== GL_CLAMP
)
339 key
->gl_clamp_mask
[0] |= 1 << s
;
340 if (sampler
->WrapT
== GL_CLAMP
)
341 key
->gl_clamp_mask
[1] |= 1 << s
;
342 if (sampler
->WrapR
== GL_CLAMP
)
343 key
->gl_clamp_mask
[2] |= 1 << s
;
349 static void brw_wm_populate_key( struct brw_context
*brw
,
350 struct brw_wm_prog_key
*key
)
352 struct gl_context
*ctx
= &brw
->ctx
;
353 /* BRW_NEW_FRAGMENT_PROGRAM */
354 const struct brw_fragment_program
*fp
=
355 (struct brw_fragment_program
*)brw
->fragment_program
;
356 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
359 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
361 memset(key
, 0, sizeof(*key
));
363 /* Build the index for table lookup
367 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
368 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
370 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
371 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
375 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
377 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
378 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
380 /* _NEW_STENCIL | _NEW_BUFFERS */
381 if (ctx
->Stencil
._Enabled
) {
382 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
384 if (ctx
->Stencil
.WriteMask
[0] ||
385 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
386 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
388 key
->iz_lookup
= lookup
;
393 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
394 if (ctx
->Line
.SmoothFlag
) {
395 if (brw
->reduced_primitive
== GL_LINES
) {
398 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
399 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
400 line_aa
= AA_SOMETIMES
;
402 if (ctx
->Polygon
.BackMode
== GL_LINE
||
403 (ctx
->Polygon
.CullFlag
&&
404 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
407 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
408 line_aa
= AA_SOMETIMES
;
410 if ((ctx
->Polygon
.CullFlag
&&
411 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
417 key
->line_aa
= line_aa
;
420 key
->stats_wm
= brw
->stats_wm
;
423 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
425 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
426 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
429 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->wm
.sampler_count
,
434 * Include the draw buffer origin and height so that we can calculate
435 * fragment position values relative to the bottom left of the drawable,
436 * from the incoming screen origin relative position we get as part of our
439 * This is only needed for the WM_WPOSXY opcode when the fragment program
440 * uses the gl_FragCoord input.
442 * We could avoid recompiling by including this as a constant referenced by
443 * our program, but if we were to do that it would also be nice to handle
444 * getting that constant updated at batchbuffer submit time (when we
445 * hold the lock and know where the buffer really is) rather than at emit
446 * time when we don't hold the lock and are just guessing. We could also
447 * just avoid using this as key data if the program doesn't use
450 * For DRI2 the origin_x/y will always be (0,0) but we still need the
451 * drawable height in order to invert the Y axis.
453 if (fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) {
454 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
457 if ((fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) || program_uses_dfdy
) {
458 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
462 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
464 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
465 key
->replicate_alpha
= ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
466 (ctx
->Multisample
.SampleAlphaToCoverage
|| ctx
->Color
.AlphaEnabled
);
468 /* BRW_NEW_VUE_MAP_GEOM_OUT */
470 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
472 /* The unique fragment program ID */
473 key
->program_string_id
= fp
->id
;
478 brw_upload_wm_prog(struct brw_context
*brw
)
480 struct gl_context
*ctx
= &brw
->ctx
;
481 struct brw_wm_prog_key key
;
482 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
483 brw
->fragment_program
;
485 brw_wm_populate_key(brw
, &key
);
487 if (!brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
489 &brw
->wm
.prog_offset
, &brw
->wm
.prog_data
)) {
490 bool success
= do_wm_prog(brw
, ctx
->Shader
._CurrentFragmentProgram
, fp
,
498 const struct brw_tracked_state brw_wm_prog
= {
500 .mesa
= (_NEW_COLOR
|
510 .brw
= (BRW_NEW_FRAGMENT_PROGRAM
|
511 BRW_NEW_REDUCED_PRIMITIVE
|
512 BRW_NEW_VUE_MAP_GEOM_OUT
|
515 .emit
= brw_upload_wm_prog