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
< FRAG_ATTRIB_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
== FRAG_ATTRIB_COL0
|| attr
== FRAG_ATTRIB_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
== FRAG_ATTRIB_WPOS
|| attr
== FRAG_ATTRIB_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
;
106 brw_wm_payload_setup(struct brw_context
*brw
,
107 struct brw_wm_compile
*c
)
109 struct intel_context
*intel
= &brw
->intel
;
110 bool uses_depth
= (c
->fp
->program
.Base
.InputsRead
&
111 (1 << FRAG_ATTRIB_WPOS
)) != 0;
112 unsigned barycentric_interp_modes
= c
->prog_data
.barycentric_interp_modes
;
115 if (intel
->gen
>= 6) {
116 /* R0-1: masks, pixel X/Y coordinates. */
117 c
->nr_payload_regs
= 2;
118 /* R2: only for 32-pixel dispatch.*/
120 /* R3-26: barycentric interpolation coordinates. These appear in the
121 * same order that they appear in the brw_wm_barycentric_interp_mode
122 * enum. Each set of coordinates occupies 2 registers if dispatch width
123 * == 8 and 4 registers if dispatch width == 16. Coordinates only
124 * appear if they were enabled using the "Barycentric Interpolation
125 * Mode" bits in WM_STATE.
127 for (i
= 0; i
< BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT
; ++i
) {
128 if (barycentric_interp_modes
& (1 << i
)) {
129 c
->barycentric_coord_reg
[i
] = c
->nr_payload_regs
;
130 c
->nr_payload_regs
+= 2;
131 if (c
->dispatch_width
== 16) {
132 c
->nr_payload_regs
+= 2;
137 /* R27: interpolated depth if uses source depth */
139 c
->source_depth_reg
= c
->nr_payload_regs
;
140 c
->nr_payload_regs
++;
141 if (c
->dispatch_width
== 16) {
142 /* R28: interpolated depth if not 8-wide. */
143 c
->nr_payload_regs
++;
146 /* R29: interpolated W set if GEN6_WM_USES_SOURCE_W.
149 c
->source_w_reg
= c
->nr_payload_regs
;
150 c
->nr_payload_regs
++;
151 if (c
->dispatch_width
== 16) {
152 /* R30: interpolated W if not 8-wide. */
153 c
->nr_payload_regs
++;
156 /* R31: MSAA position offsets. */
157 /* R32-: bary for 32-pixel. */
158 /* R58-59: interp W for 32-pixel. */
160 if (c
->fp
->program
.Base
.OutputsWritten
&
161 BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
162 c
->source_depth_to_render_target
= true;
163 c
->computes_depth
= true;
166 brw_wm_lookup_iz(intel
, c
);
171 brw_wm_prog_data_compare(const void *in_a
, const void *in_b
,
172 int aux_size
, const void *in_key
)
174 const struct brw_wm_prog_data
*a
= in_a
;
175 const struct brw_wm_prog_data
*b
= in_b
;
177 /* Compare all the struct up to the pointers. */
178 if (memcmp(a
, b
, offsetof(struct brw_wm_prog_data
, param
)))
181 if (memcmp(a
->param
, b
->param
, a
->nr_params
* sizeof(void *)))
184 if (memcmp(a
->pull_param
, b
->pull_param
, a
->nr_pull_params
* sizeof(void *)))
191 brw_wm_prog_data_free(const void *in_prog_data
)
193 const struct brw_wm_prog_data
*prog_data
= in_prog_data
;
195 ralloc_free((void *)prog_data
->param
);
196 ralloc_free((void *)prog_data
->pull_param
);
200 * All Mesa program -> GPU code generation goes through this function.
201 * Depending on the instructions used (i.e. flow control instructions)
202 * we'll use one of two code generators.
204 bool do_wm_prog(struct brw_context
*brw
,
205 struct gl_shader_program
*prog
,
206 struct brw_fragment_program
*fp
,
207 struct brw_wm_prog_key
*key
)
209 struct intel_context
*intel
= &brw
->intel
;
210 struct brw_wm_compile
*c
;
211 const GLuint
*program
;
212 struct gl_shader
*fs
= NULL
;
216 fs
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
218 c
= rzalloc(NULL
, struct brw_wm_compile
);
220 /* Allocate the references to the uniforms that will end up in the
221 * prog_data associated with the compiled program, and which will be freed
222 * by the state cache.
226 param_count
= fs
->num_uniform_components
;
228 param_count
= fp
->program
.Base
.Parameters
->NumParameters
* 4;
230 /* The backend also sometimes adds params for texture size. */
231 param_count
+= 2 * BRW_MAX_TEX_UNIT
;
232 c
->prog_data
.param
= rzalloc_array(NULL
, const float *, param_count
);
233 c
->prog_data
.pull_param
= rzalloc_array(NULL
, const float *, param_count
);
235 memcpy(&c
->key
, key
, sizeof(*key
));
239 brw_init_compile(brw
, &c
->func
, c
);
241 c
->prog_data
.barycentric_interp_modes
=
242 brw_compute_barycentric_interp_modes(brw
, c
->key
.flat_shade
,
245 brw_wm_fs_emit(brw
, c
, prog
);
247 /* Scratch space is used for register spilling */
248 if (c
->last_scratch
) {
249 perf_debug("Fragment shader triggered register spilling. "
250 "Try reducing the number of live scalar values to "
251 "improve performance.\n");
253 c
->prog_data
.total_scratch
= brw_get_scratch_size(c
->last_scratch
);
255 brw_get_scratch_bo(intel
, &brw
->wm
.scratch_bo
,
256 c
->prog_data
.total_scratch
* brw
->max_wm_threads
);
259 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
260 fprintf(stderr
, "\n");
264 program
= brw_get_program(&c
->func
, &program_size
);
266 brw_upload_cache(&brw
->cache
, BRW_WM_PROG
,
267 &c
->key
, sizeof(c
->key
),
268 program
, program_size
,
269 &c
->prog_data
, sizeof(c
->prog_data
),
270 &brw
->wm
.prog_offset
, &brw
->wm
.prog_data
);
278 key_debug(const char *name
, int a
, int b
)
281 perf_debug(" %s %d->%d\n", name
, a
, b
);
289 brw_debug_recompile_sampler_key(const struct brw_sampler_prog_key_data
*old_key
,
290 const struct brw_sampler_prog_key_data
*key
)
294 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
295 found
|= key_debug("EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
296 old_key
->swizzles
[i
], key
->swizzles
[i
]);
298 found
|= key_debug("GL_CLAMP enabled on any texture unit's 1st coordinate",
299 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
300 found
|= key_debug("GL_CLAMP enabled on any texture unit's 2nd coordinate",
301 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
302 found
|= key_debug("GL_CLAMP enabled on any texture unit's 3rd coordinate",
303 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
304 found
|= key_debug("GL_MESA_ycbcr texturing\n",
305 old_key
->yuvtex_mask
, key
->yuvtex_mask
);
306 found
|= key_debug("GL_MESA_ycbcr UV swapping\n",
307 old_key
->yuvtex_swap_mask
, key
->yuvtex_swap_mask
);
313 brw_wm_debug_recompile(struct brw_context
*brw
,
314 struct gl_shader_program
*prog
,
315 const struct brw_wm_prog_key
*key
)
317 struct brw_cache_item
*c
= NULL
;
318 const struct brw_wm_prog_key
*old_key
= NULL
;
321 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
323 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
324 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
325 if (c
->cache_id
== BRW_WM_PROG
) {
328 if (old_key
->program_string_id
== key
->program_string_id
)
337 perf_debug(" Didn't find previous compile in the shader cache for "
342 found
|= key_debug("alphatest, computed depth, depth test, or depth write",
343 old_key
->iz_lookup
, key
->iz_lookup
);
344 found
|= key_debug("depth statistics", old_key
->stats_wm
, key
->stats_wm
);
345 found
|= key_debug("flat shading", old_key
->flat_shade
, key
->flat_shade
);
346 found
|= key_debug("number of color buffers", old_key
->nr_color_regions
, key
->nr_color_regions
);
347 found
|= key_debug("sample alpha to coverage", old_key
->sample_alpha_to_coverage
, key
->sample_alpha_to_coverage
);
348 found
|= key_debug("rendering to FBO", old_key
->render_to_fbo
, key
->render_to_fbo
);
349 found
|= key_debug("fragment color clamping", old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
350 found
|= key_debug("line smoothing", old_key
->line_aa
, key
->line_aa
);
351 found
|= key_debug("proj_attrib_mask", old_key
->proj_attrib_mask
, key
->proj_attrib_mask
);
352 found
|= key_debug("renderbuffer height", old_key
->drawable_height
, key
->drawable_height
);
353 found
|= key_debug("vertex shader outputs", old_key
->vp_outputs_written
, key
->vp_outputs_written
);
355 found
|= brw_debug_recompile_sampler_key(&old_key
->tex
, &key
->tex
);
358 perf_debug(" Something else\n");
363 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
364 const struct gl_program
*prog
,
365 struct brw_sampler_prog_key_data
*key
)
367 struct intel_context
*intel
= intel_context(ctx
);
369 for (int s
= 0; s
< MAX_SAMPLERS
; s
++) {
370 key
->swizzles
[s
] = SWIZZLE_NOOP
;
372 if (!(prog
->SamplersUsed
& (1 << s
)))
375 int unit_id
= prog
->SamplerUnits
[s
];
376 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
378 if (unit
->_ReallyEnabled
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
379 const struct gl_texture_object
*t
= unit
->_Current
;
380 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
381 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
383 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
384 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
385 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
387 /* Haswell handles texture swizzling as surface format overrides
388 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
390 if (!intel
->is_haswell
|| alpha_depth
)
391 key
->swizzles
[s
] = brw_get_texture_swizzle(t
);
393 if (img
->InternalFormat
== GL_YCBCR_MESA
) {
394 key
->yuvtex_mask
|= 1 << s
;
395 if (img
->TexFormat
== MESA_FORMAT_YCBCR
)
396 key
->yuvtex_swap_mask
|= 1 << s
;
399 if (sampler
->MinFilter
!= GL_NEAREST
&&
400 sampler
->MagFilter
!= GL_NEAREST
) {
401 if (sampler
->WrapS
== GL_CLAMP
)
402 key
->gl_clamp_mask
[0] |= 1 << s
;
403 if (sampler
->WrapT
== GL_CLAMP
)
404 key
->gl_clamp_mask
[1] |= 1 << s
;
405 if (sampler
->WrapR
== GL_CLAMP
)
406 key
->gl_clamp_mask
[2] |= 1 << s
;
412 static void brw_wm_populate_key( struct brw_context
*brw
,
413 struct brw_wm_prog_key
*key
)
415 struct gl_context
*ctx
= &brw
->intel
.ctx
;
416 struct intel_context
*intel
= &brw
->intel
;
417 /* BRW_NEW_FRAGMENT_PROGRAM */
418 const struct brw_fragment_program
*fp
=
419 (struct brw_fragment_program
*)brw
->fragment_program
;
420 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
423 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
425 memset(key
, 0, sizeof(*key
));
427 /* Build the index for table lookup
429 if (intel
->gen
< 6) {
431 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
432 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
434 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
435 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
439 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
441 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
442 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
445 if (ctx
->Stencil
._Enabled
) {
446 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
448 if (ctx
->Stencil
.WriteMask
[0] ||
449 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
450 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
452 key
->iz_lookup
= lookup
;
457 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
458 if (ctx
->Line
.SmoothFlag
) {
459 if (brw
->intel
.reduced_primitive
== GL_LINES
) {
462 else if (brw
->intel
.reduced_primitive
== GL_TRIANGLES
) {
463 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
464 line_aa
= AA_SOMETIMES
;
466 if (ctx
->Polygon
.BackMode
== GL_LINE
||
467 (ctx
->Polygon
.CullFlag
&&
468 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
471 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
472 line_aa
= AA_SOMETIMES
;
474 if ((ctx
->Polygon
.CullFlag
&&
475 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
481 key
->line_aa
= line_aa
;
484 key
->stats_wm
= brw
->intel
.stats_wm
;
486 /* BRW_NEW_WM_INPUT_DIMENSIONS */
487 /* Only set this for fixed function. The optimization it enables isn't
488 * useful for programs using shaders.
490 if (ctx
->Shader
.CurrentFragmentProgram
)
491 key
->proj_attrib_mask
= 0xffffffff;
493 key
->proj_attrib_mask
= brw
->wm
.input_size_masks
[4-1];
496 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
498 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
499 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
502 brw_populate_sampler_prog_key_data(ctx
, prog
, &key
->tex
);
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
& FRAG_BIT_WPOS
) {
526 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
529 if ((fp
->program
.Base
.InputsRead
& FRAG_BIT_WPOS
) || program_uses_dfdy
) {
530 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
534 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
535 /* _NEW_MULTISAMPLE */
536 key
->sample_alpha_to_coverage
= ctx
->Multisample
.SampleAlphaToCoverage
;
538 /* CACHE_NEW_VS_PROG */
540 key
->vp_outputs_written
= brw
->vs
.prog_data
->outputs_written
;
542 /* The unique fragment program ID */
543 key
->program_string_id
= fp
->id
;
548 brw_upload_wm_prog(struct brw_context
*brw
)
550 struct intel_context
*intel
= &brw
->intel
;
551 struct gl_context
*ctx
= &intel
->ctx
;
552 struct brw_wm_prog_key key
;
553 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
554 brw
->fragment_program
;
556 brw_wm_populate_key(brw
, &key
);
558 if (!brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
560 &brw
->wm
.prog_offset
, &brw
->wm
.prog_data
)) {
561 bool success
= do_wm_prog(brw
, ctx
->Shader
._CurrentFragmentProgram
, fp
,
569 const struct brw_tracked_state brw_wm_prog
= {
571 .mesa
= (_NEW_COLOR
|
581 .brw
= (BRW_NEW_FRAGMENT_PROGRAM
|
582 BRW_NEW_WM_INPUT_DIMENSIONS
|
583 BRW_NEW_REDUCED_PRIMITIVE
),
584 .cache
= CACHE_NEW_VS_PROG
,
586 .emit
= brw_upload_wm_prog