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 "util/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 computed_depth_mode(struct gl_fragment_program
*fp
)
122 if (fp
->Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
123 switch (fp
->FragDepthLayout
) {
124 case FRAG_DEPTH_LAYOUT_NONE
:
125 case FRAG_DEPTH_LAYOUT_ANY
:
126 return BRW_PSCDEPTH_ON
;
127 case FRAG_DEPTH_LAYOUT_GREATER
:
128 return BRW_PSCDEPTH_ON_GE
;
129 case FRAG_DEPTH_LAYOUT_LESS
:
130 return BRW_PSCDEPTH_ON_LE
;
131 case FRAG_DEPTH_LAYOUT_UNCHANGED
:
132 return BRW_PSCDEPTH_OFF
;
135 return BRW_PSCDEPTH_OFF
;
139 brw_wm_prog_data_compare(const void *in_a
, const void *in_b
)
141 const struct brw_wm_prog_data
*a
= in_a
;
142 const struct brw_wm_prog_data
*b
= in_b
;
144 /* Compare the base structure. */
145 if (!brw_stage_prog_data_compare(&a
->base
, &b
->base
))
148 /* Compare the rest of the structure. */
149 const unsigned offset
= sizeof(struct brw_stage_prog_data
);
150 if (memcmp(((char *) a
) + offset
, ((char *) b
) + offset
,
151 sizeof(struct brw_wm_prog_data
) - offset
))
158 * All Mesa program -> GPU code generation goes through this function.
159 * Depending on the instructions used (i.e. flow control instructions)
160 * we'll use one of two code generators.
163 brw_compile_wm_prog(struct brw_context
*brw
,
164 struct gl_shader_program
*prog
,
165 struct brw_fragment_program
*fp
,
166 struct brw_wm_prog_key
*key
)
168 struct gl_context
*ctx
= &brw
->ctx
;
169 void *mem_ctx
= ralloc_context(NULL
);
170 struct brw_wm_prog_data prog_data
;
171 const GLuint
*program
;
172 struct gl_shader
*fs
= NULL
;
176 fs
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
178 memset(&prog_data
, 0, sizeof(prog_data
));
179 /* key->alpha_test_func means simulating alpha testing via discards,
180 * so the shader definitely kills pixels.
182 prog_data
.uses_kill
= fp
->program
.UsesKill
|| key
->alpha_test_func
;
184 prog_data
.computed_depth_mode
= computed_depth_mode(&fp
->program
);
186 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
188 prog_data
.base
.use_alt_mode
= true;
190 /* Allocate the references to the uniforms that will end up in the
191 * prog_data associated with the compiled program, and which will be freed
192 * by the state cache.
196 param_count
= fs
->num_uniform_components
;
198 param_count
= fp
->program
.Base
.Parameters
->NumParameters
* 4;
200 /* The backend also sometimes adds params for texture size. */
201 param_count
+= 2 * ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
;
202 prog_data
.base
.param
=
203 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
204 prog_data
.base
.pull_param
=
205 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
206 prog_data
.base
.nr_params
= param_count
;
208 prog_data
.barycentric_interp_modes
=
209 brw_compute_barycentric_interp_modes(brw
, key
->flat_shade
,
210 key
->persample_shading
,
213 program
= brw_wm_fs_emit(brw
, mem_ctx
, key
, &prog_data
,
214 &fp
->program
, prog
, &program_size
);
215 if (program
== NULL
) {
216 ralloc_free(mem_ctx
);
220 if (prog_data
.base
.total_scratch
) {
221 brw_get_scratch_bo(brw
, &brw
->wm
.base
.scratch_bo
,
222 prog_data
.base
.total_scratch
* brw
->max_wm_threads
);
225 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
226 fprintf(stderr
, "\n");
228 brw_upload_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
229 key
, sizeof(struct brw_wm_prog_key
),
230 program
, program_size
,
231 &prog_data
, sizeof(prog_data
),
232 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
);
234 ralloc_free(mem_ctx
);
240 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
243 perf_debug(" %s %d->%d\n", name
, a
, b
);
251 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
252 const struct brw_sampler_prog_key_data
*old_key
,
253 const struct brw_sampler_prog_key_data
*key
)
257 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
258 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
259 old_key
->swizzles
[i
], key
->swizzles
[i
]);
261 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
262 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
263 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
264 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
265 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
266 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
267 found
|= key_debug(brw
, "gather channel quirk on any texture unit",
268 old_key
->gather_channel_quirk_mask
, key
->gather_channel_quirk_mask
);
274 brw_wm_debug_recompile(struct brw_context
*brw
,
275 struct gl_shader_program
*prog
,
276 const struct brw_wm_prog_key
*key
)
278 struct brw_cache_item
*c
= NULL
;
279 const struct brw_wm_prog_key
*old_key
= NULL
;
282 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
284 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
285 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
286 if (c
->cache_id
== BRW_CACHE_FS_PROG
) {
289 if (old_key
->program_string_id
== key
->program_string_id
)
298 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
302 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
304 old_key
->iz_lookup
, key
->iz_lookup
);
305 found
|= key_debug(brw
, "depth statistics",
306 old_key
->stats_wm
, key
->stats_wm
);
307 found
|= key_debug(brw
, "flat shading",
308 old_key
->flat_shade
, key
->flat_shade
);
309 found
|= key_debug(brw
, "per-sample shading",
310 old_key
->persample_shading
, key
->persample_shading
);
311 found
|= key_debug(brw
, "per-sample shading and 2x MSAA",
312 old_key
->persample_2x
, key
->persample_2x
);
313 found
|= key_debug(brw
, "number of color buffers",
314 old_key
->nr_color_regions
, key
->nr_color_regions
);
315 found
|= key_debug(brw
, "MRT alpha test or alpha-to-coverage",
316 old_key
->replicate_alpha
, key
->replicate_alpha
);
317 found
|= key_debug(brw
, "rendering to FBO",
318 old_key
->render_to_fbo
, key
->render_to_fbo
);
319 found
|= key_debug(brw
, "fragment color clamping",
320 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
321 found
|= key_debug(brw
, "line smoothing",
322 old_key
->line_aa
, key
->line_aa
);
323 found
|= key_debug(brw
, "renderbuffer height",
324 old_key
->drawable_height
, key
->drawable_height
);
325 found
|= key_debug(brw
, "input slots valid",
326 old_key
->input_slots_valid
, key
->input_slots_valid
);
327 found
|= key_debug(brw
, "mrt alpha test function",
328 old_key
->alpha_test_func
, key
->alpha_test_func
);
329 found
|= key_debug(brw
, "mrt alpha test reference value",
330 old_key
->alpha_test_ref
, key
->alpha_test_ref
);
332 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
335 perf_debug(" Something else\n");
340 gen6_gather_workaround(GLenum internalformat
)
342 switch (internalformat
) {
343 case GL_R8I
: return WA_SIGN
| WA_8BIT
;
344 case GL_R8UI
: return WA_8BIT
;
345 case GL_R16I
: return WA_SIGN
| WA_16BIT
;
346 case GL_R16UI
: return WA_16BIT
;
347 /* note that even though GL_R32I and GL_R32UI have format overrides
348 * in the surface state, there is no shader w/a required */
354 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
355 const struct gl_program
*prog
,
356 unsigned sampler_count
,
357 struct brw_sampler_prog_key_data
*key
)
359 struct brw_context
*brw
= brw_context(ctx
);
361 for (int s
= 0; s
< sampler_count
; s
++) {
362 key
->swizzles
[s
] = SWIZZLE_NOOP
;
364 if (!(prog
->SamplersUsed
& (1 << s
)))
367 int unit_id
= prog
->SamplerUnits
[s
];
368 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
370 if (unit
->_Current
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
371 const struct gl_texture_object
*t
= unit
->_Current
;
372 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
373 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
375 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
376 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
377 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
379 /* Haswell handles texture swizzling as surface format overrides
380 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
382 if (alpha_depth
|| (brw
->gen
< 8 && !brw
->is_haswell
))
383 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
386 sampler
->MinFilter
!= GL_NEAREST
&&
387 sampler
->MagFilter
!= GL_NEAREST
) {
388 if (sampler
->WrapS
== GL_CLAMP
)
389 key
->gl_clamp_mask
[0] |= 1 << s
;
390 if (sampler
->WrapT
== GL_CLAMP
)
391 key
->gl_clamp_mask
[1] |= 1 << s
;
392 if (sampler
->WrapR
== GL_CLAMP
)
393 key
->gl_clamp_mask
[2] |= 1 << s
;
396 /* gather4's channel select for green from RG32F is broken;
397 * requires a shader w/a on IVB; fixable with just SCS on HSW. */
398 if (brw
->gen
== 7 && !brw
->is_haswell
&& prog
->UsesGather
) {
399 if (img
->InternalFormat
== GL_RG32F
)
400 key
->gather_channel_quirk_mask
|= 1 << s
;
403 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
404 * UNORM/FLOAT instead and fix it in the shader.
406 if (brw
->gen
== 6 && prog
->UsesGather
) {
407 key
->gen6_gather_wa
[s
] = gen6_gather_workaround(img
->InternalFormat
);
410 /* If this is a multisample sampler, and uses the CMS MSAA layout,
411 * then we need to emit slightly different code to first sample the
414 struct intel_texture_object
*intel_tex
=
415 intel_texture_object((struct gl_texture_object
*)t
);
418 intel_tex
->mt
->msaa_layout
== INTEL_MSAA_LAYOUT_CMS
) {
419 key
->compressed_multisample_layout_mask
|= 1 << s
;
426 brw_wm_state_dirty (struct brw_context
*brw
)
428 return brw_state_dirty(brw
,
440 BRW_NEW_FRAGMENT_PROGRAM
|
441 BRW_NEW_REDUCED_PRIMITIVE
|
443 BRW_NEW_VUE_MAP_GEOM_OUT
);
446 static void brw_wm_populate_key( struct brw_context
*brw
,
447 struct brw_wm_prog_key
*key
)
449 struct gl_context
*ctx
= &brw
->ctx
;
450 /* BRW_NEW_FRAGMENT_PROGRAM */
451 const struct brw_fragment_program
*fp
=
452 (struct brw_fragment_program
*)brw
->fragment_program
;
453 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
456 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
457 bool multisample_fbo
= ctx
->DrawBuffer
->Visual
.samples
> 1;
459 memset(key
, 0, sizeof(*key
));
461 /* Build the index for table lookup
465 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
466 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
468 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
469 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
473 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
475 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
476 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
478 /* _NEW_STENCIL | _NEW_BUFFERS */
479 if (ctx
->Stencil
._Enabled
) {
480 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
482 if (ctx
->Stencil
.WriteMask
[0] ||
483 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
484 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
486 key
->iz_lookup
= lookup
;
491 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
492 if (ctx
->Line
.SmoothFlag
) {
493 if (brw
->reduced_primitive
== GL_LINES
) {
496 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
497 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
498 line_aa
= AA_SOMETIMES
;
500 if (ctx
->Polygon
.BackMode
== GL_LINE
||
501 (ctx
->Polygon
.CullFlag
&&
502 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
505 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
506 line_aa
= AA_SOMETIMES
;
508 if ((ctx
->Polygon
.CullFlag
&&
509 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
515 key
->line_aa
= line_aa
;
518 key
->high_quality_derivatives
=
519 ctx
->Hint
.FragmentShaderDerivative
== GL_NICEST
;
522 key
->stats_wm
= brw
->stats_wm
;
525 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
527 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
528 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
531 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->wm
.base
.sampler_count
,
536 * Include the draw buffer origin and height so that we can calculate
537 * fragment position values relative to the bottom left of the drawable,
538 * from the incoming screen origin relative position we get as part of our
541 * This is only needed for the WM_WPOSXY opcode when the fragment program
542 * uses the gl_FragCoord input.
544 * We could avoid recompiling by including this as a constant referenced by
545 * our program, but if we were to do that it would also be nice to handle
546 * getting that constant updated at batchbuffer submit time (when we
547 * hold the lock and know where the buffer really is) rather than at emit
548 * time when we don't hold the lock and are just guessing. We could also
549 * just avoid using this as key data if the program doesn't use
552 * For DRI2 the origin_x/y will always be (0,0) but we still need the
553 * drawable height in order to invert the Y axis.
555 if (fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) {
556 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
559 if ((fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) || program_uses_dfdy
) {
560 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
564 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
566 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
567 key
->replicate_alpha
= ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
568 (ctx
->Multisample
.SampleAlphaToCoverage
|| ctx
->Color
.AlphaEnabled
);
570 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
571 /* Ignore sample qualifier while computing this flag. */
572 key
->persample_shading
=
573 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, true) > 1;
574 if (key
->persample_shading
)
575 key
->persample_2x
= ctx
->DrawBuffer
->Visual
.samples
== 2;
577 key
->compute_pos_offset
=
578 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, false) > 1 &&
579 fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_POS
;
581 key
->compute_sample_id
=
583 ctx
->Multisample
.Enabled
&&
584 (fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_ID
);
586 /* BRW_NEW_VUE_MAP_GEOM_OUT */
587 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->program
.Base
.InputsRead
&
588 BRW_FS_VARYING_INPUT_MASK
) > 16)
589 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
592 /* _NEW_COLOR | _NEW_BUFFERS */
593 /* Pre-gen6, the hardware alpha test always used each render
594 * target's alpha to do alpha test, as opposed to render target 0's alpha
595 * like GL requires. Fix that by building the alpha test into the
596 * shader, and we'll skip enabling the fixed function alpha test.
598 if (brw
->gen
< 6 && ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 && ctx
->Color
.AlphaEnabled
) {
599 key
->alpha_test_func
= ctx
->Color
.AlphaFunc
;
600 key
->alpha_test_ref
= ctx
->Color
.AlphaRef
;
603 /* The unique fragment program ID */
604 key
->program_string_id
= fp
->id
;
608 brw_upload_wm_prog(struct brw_context
*brw
)
610 struct gl_context
*ctx
= &brw
->ctx
;
611 struct gl_shader_program
*current
= ctx
->_Shader
->_CurrentFragmentProgram
;
612 struct brw_wm_prog_key key
;
613 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
614 brw
->fragment_program
;
616 if (!brw_wm_state_dirty(brw
))
619 brw_wm_populate_key(brw
, &key
);
621 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
623 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
)) {
624 bool success
= brw_compile_wm_prog(brw
, current
, fp
, &key
);
628 brw
->wm
.base
.prog_data
= &brw
->wm
.prog_data
->base
;