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 #include "brw_context.h"
28 #include "brw_state.h"
29 #include "brw_shader.h"
30 #include "main/enums.h"
31 #include "main/formats.h"
32 #include "main/fbobject.h"
33 #include "main/samplerobj.h"
34 #include "main/framebuffer.h"
35 #include "program/prog_parameter.h"
36 #include "program/program.h"
37 #include "intel_mipmap_tree.h"
38 #include "intel_image.h"
40 #include "brw_program.h"
42 #include "util/ralloc.h"
45 assign_fs_binding_table_offsets(const struct gen_device_info
*devinfo
,
46 const struct gl_shader_program
*shader_prog
,
47 const struct gl_program
*prog
,
48 const struct brw_wm_prog_key
*key
,
49 struct brw_wm_prog_data
*prog_data
)
51 uint32_t next_binding_table_offset
= 0;
53 /* If there are no color regions, we still perform an FB write to a null
54 * renderbuffer, which we place at surface index 0.
56 prog_data
->binding_table
.render_target_start
= next_binding_table_offset
;
57 next_binding_table_offset
+= MAX2(key
->nr_color_regions
, 1);
59 next_binding_table_offset
=
60 brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT
, devinfo
,
61 shader_prog
, prog
, &prog_data
->base
,
62 next_binding_table_offset
);
64 if (prog
->nir
->info
.outputs_read
&& !key
->coherent_fb_fetch
) {
65 prog_data
->binding_table
.render_target_read_start
=
66 next_binding_table_offset
;
67 next_binding_table_offset
+= key
->nr_color_regions
;
72 * All Mesa program -> GPU code generation goes through this function.
73 * Depending on the instructions used (i.e. flow control instructions)
74 * we'll use one of two code generators.
77 brw_codegen_wm_prog(struct brw_context
*brw
,
78 struct gl_shader_program
*prog
,
79 struct brw_fragment_program
*fp
,
80 struct brw_wm_prog_key
*key
)
82 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
83 struct gl_context
*ctx
= &brw
->ctx
;
84 void *mem_ctx
= ralloc_context(NULL
);
85 struct brw_wm_prog_data prog_data
;
86 const GLuint
*program
;
87 struct brw_shader
*fs
= NULL
;
89 bool start_busy
= false;
90 double start_time
= 0;
93 fs
= (struct brw_shader
*)prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
95 memset(&prog_data
, 0, sizeof(prog_data
));
97 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
99 prog_data
.base
.use_alt_mode
= true;
101 assign_fs_binding_table_offsets(devinfo
, prog
,
102 &fp
->program
.Base
, key
, &prog_data
);
104 /* Allocate the references to the uniforms that will end up in the
105 * prog_data associated with the compiled program, and which will be freed
106 * by the state cache.
108 int param_count
= fp
->program
.Base
.nir
->num_uniforms
/ 4;
110 prog_data
.base
.nr_image_params
= fs
->base
.NumImages
;
111 /* The backend also sometimes adds params for texture size. */
112 param_count
+= 2 * ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
;
113 prog_data
.base
.param
=
114 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
115 prog_data
.base
.pull_param
=
116 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
117 prog_data
.base
.image_param
=
118 rzalloc_array(NULL
, struct brw_image_param
,
119 prog_data
.base
.nr_image_params
);
120 prog_data
.base
.nr_params
= param_count
;
123 brw_nir_setup_glsl_uniforms(fp
->program
.Base
.nir
, prog
, &fp
->program
.Base
,
124 &prog_data
.base
, true);
126 brw_nir_setup_arb_uniforms(fp
->program
.Base
.nir
, &fp
->program
.Base
,
130 if (unlikely(brw
->perf_debug
)) {
131 start_busy
= (brw
->batch
.last_bo
&&
132 drm_intel_bo_busy(brw
->batch
.last_bo
));
133 start_time
= get_time();
136 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
137 brw_dump_ir("fragment", prog
, fs
? &fs
->base
: NULL
, &fp
->program
.Base
);
139 int st_index8
= -1, st_index16
= -1;
140 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
141 st_index8
= brw_get_shader_time_index(brw
, prog
, &fp
->program
.Base
, ST_FS8
);
142 st_index16
= brw_get_shader_time_index(brw
, prog
, &fp
->program
.Base
, ST_FS16
);
145 char *error_str
= NULL
;
146 program
= brw_compile_fs(brw
->screen
->compiler
, brw
, mem_ctx
,
147 key
, &prog_data
, fp
->program
.Base
.nir
,
148 &fp
->program
.Base
, st_index8
, st_index16
,
149 true, brw
->use_rep_send
,
150 &program_size
, &error_str
);
151 if (program
== NULL
) {
153 prog
->LinkStatus
= false;
154 ralloc_strcat(&prog
->InfoLog
, error_str
);
157 _mesa_problem(NULL
, "Failed to compile fragment shader: %s\n", error_str
);
159 ralloc_free(mem_ctx
);
163 if (unlikely(brw
->perf_debug
) && fs
) {
164 if (fs
->compiled_once
)
165 brw_wm_debug_recompile(brw
, prog
, key
);
166 fs
->compiled_once
= true;
168 if (start_busy
&& !drm_intel_bo_busy(brw
->batch
.last_bo
)) {
169 perf_debug("FS compile took %.03f ms and stalled the GPU\n",
170 (get_time() - start_time
) * 1000);
174 brw_alloc_stage_scratch(brw
, &brw
->wm
.base
,
175 prog_data
.base
.total_scratch
,
176 devinfo
->max_wm_threads
);
178 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
179 fprintf(stderr
, "\n");
181 brw_upload_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
182 key
, sizeof(struct brw_wm_prog_key
),
183 program
, program_size
,
184 &prog_data
, sizeof(prog_data
),
185 &brw
->wm
.base
.prog_offset
, &brw
->wm
.base
.prog_data
);
187 ralloc_free(mem_ctx
);
193 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
194 const struct brw_sampler_prog_key_data
*old_key
,
195 const struct brw_sampler_prog_key_data
*key
)
199 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
200 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
201 old_key
->swizzles
[i
], key
->swizzles
[i
]);
203 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
204 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
205 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
206 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
207 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
208 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
209 found
|= key_debug(brw
, "gather channel quirk on any texture unit",
210 old_key
->gather_channel_quirk_mask
, key
->gather_channel_quirk_mask
);
211 found
|= key_debug(brw
, "compressed multisample layout",
212 old_key
->compressed_multisample_layout_mask
,
213 key
->compressed_multisample_layout_mask
);
214 found
|= key_debug(brw
, "16x msaa",
218 found
|= key_debug(brw
, "y_uv image bound",
219 old_key
->y_uv_image_mask
,
220 key
->y_uv_image_mask
);
221 found
|= key_debug(brw
, "y_u_v image bound",
222 old_key
->y_u_v_image_mask
,
223 key
->y_u_v_image_mask
);
224 found
|= key_debug(brw
, "yx_xuxv image bound",
225 old_key
->yx_xuxv_image_mask
,
226 key
->yx_xuxv_image_mask
);
228 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
229 found
|= key_debug(brw
, "textureGather workarounds",
230 old_key
->gen6_gather_wa
[i
], key
->gen6_gather_wa
[i
]);
237 brw_wm_debug_recompile(struct brw_context
*brw
,
238 struct gl_shader_program
*prog
,
239 const struct brw_wm_prog_key
*key
)
241 struct brw_cache_item
*c
= NULL
;
242 const struct brw_wm_prog_key
*old_key
= NULL
;
245 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
247 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
248 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
249 if (c
->cache_id
== BRW_CACHE_FS_PROG
) {
252 if (old_key
->program_string_id
== key
->program_string_id
)
261 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
265 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
267 old_key
->iz_lookup
, key
->iz_lookup
);
268 found
|= key_debug(brw
, "depth statistics",
269 old_key
->stats_wm
, key
->stats_wm
);
270 found
|= key_debug(brw
, "flat shading",
271 old_key
->flat_shade
, key
->flat_shade
);
272 found
|= key_debug(brw
, "per-sample interpolation",
273 old_key
->persample_interp
, key
->persample_interp
);
274 found
|= key_debug(brw
, "number of color buffers",
275 old_key
->nr_color_regions
, key
->nr_color_regions
);
276 found
|= key_debug(brw
, "MRT alpha test or alpha-to-coverage",
277 old_key
->replicate_alpha
, key
->replicate_alpha
);
278 found
|= key_debug(brw
, "fragment color clamping",
279 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
280 found
|= key_debug(brw
, "multisampled FBO",
281 old_key
->multisample_fbo
, key
->multisample_fbo
);
282 found
|= key_debug(brw
, "line smoothing",
283 old_key
->line_aa
, key
->line_aa
);
284 found
|= key_debug(brw
, "input slots valid",
285 old_key
->input_slots_valid
, key
->input_slots_valid
);
286 found
|= key_debug(brw
, "mrt alpha test function",
287 old_key
->alpha_test_func
, key
->alpha_test_func
);
288 found
|= key_debug(brw
, "mrt alpha test reference value",
289 old_key
->alpha_test_ref
, key
->alpha_test_ref
);
291 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
294 perf_debug(" Something else\n");
299 gen6_gather_workaround(GLenum internalformat
)
301 switch (internalformat
) {
302 case GL_R8I
: return WA_SIGN
| WA_8BIT
;
303 case GL_R8UI
: return WA_8BIT
;
304 case GL_R16I
: return WA_SIGN
| WA_16BIT
;
305 case GL_R16UI
: return WA_16BIT
;
307 /* Note that even though GL_R32I and GL_R32UI have format overrides in
308 * the surface state, there is no shader w/a required.
315 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
316 const struct gl_program
*prog
,
317 struct brw_sampler_prog_key_data
*key
)
319 struct brw_context
*brw
= brw_context(ctx
);
320 GLbitfield mask
= prog
->SamplersUsed
;
323 const int s
= u_bit_scan(&mask
);
325 key
->swizzles
[s
] = SWIZZLE_NOOP
;
327 int unit_id
= prog
->SamplerUnits
[s
];
328 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
330 if (unit
->_Current
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
331 const struct gl_texture_object
*t
= unit
->_Current
;
332 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
333 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
335 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
336 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
337 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
339 /* Haswell handles texture swizzling as surface format overrides
340 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
342 if (alpha_depth
|| (brw
->gen
< 8 && !brw
->is_haswell
))
343 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
346 sampler
->MinFilter
!= GL_NEAREST
&&
347 sampler
->MagFilter
!= GL_NEAREST
) {
348 if (sampler
->WrapS
== GL_CLAMP
)
349 key
->gl_clamp_mask
[0] |= 1 << s
;
350 if (sampler
->WrapT
== GL_CLAMP
)
351 key
->gl_clamp_mask
[1] |= 1 << s
;
352 if (sampler
->WrapR
== GL_CLAMP
)
353 key
->gl_clamp_mask
[2] |= 1 << s
;
356 /* gather4's channel select for green from RG32F is broken; requires
357 * a shader w/a on IVB; fixable with just SCS on HSW.
359 if (brw
->gen
== 7 && !brw
->is_haswell
&&
360 prog
->nir
->info
.uses_texture_gather
) {
361 if (img
->InternalFormat
== GL_RG32F
)
362 key
->gather_channel_quirk_mask
|= 1 << s
;
365 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
366 * UNORM/FLOAT instead and fix it in the shader.
368 if (brw
->gen
== 6 && prog
->nir
->info
.uses_texture_gather
) {
369 key
->gen6_gather_wa
[s
] = gen6_gather_workaround(img
->InternalFormat
);
372 /* If this is a multisample sampler, and uses the CMS MSAA layout,
373 * then we need to emit slightly different code to first sample the
376 struct intel_texture_object
*intel_tex
=
377 intel_texture_object((struct gl_texture_object
*)t
);
379 /* From gen9 onwards some single sampled buffers can also be
380 * compressed. These don't need ld2dms sampling along with mcs fetch.
383 intel_tex
->mt
->msaa_layout
== INTEL_MSAA_LAYOUT_CMS
&&
384 intel_tex
->mt
->num_samples
> 1) {
385 key
->compressed_multisample_layout_mask
|= 1 << s
;
387 if (intel_tex
->mt
->num_samples
>= 16) {
388 assert(brw
->gen
>= 9);
389 key
->msaa_16
|= 1 << s
;
393 if (t
->Target
== GL_TEXTURE_EXTERNAL_OES
&& intel_tex
->planar_format
) {
394 switch (intel_tex
->planar_format
->components
) {
395 case __DRI_IMAGE_COMPONENTS_Y_UV
:
396 key
->y_uv_image_mask
|= 1 << s
;
398 case __DRI_IMAGE_COMPONENTS_Y_U_V
:
399 key
->y_u_v_image_mask
|= 1 << s
;
401 case __DRI_IMAGE_COMPONENTS_Y_XUXV
:
402 key
->yx_xuxv_image_mask
|= 1 << s
;
414 brw_wm_state_dirty(const struct brw_context
*brw
)
416 return brw_state_dirty(brw
,
428 BRW_NEW_FRAGMENT_PROGRAM
|
429 BRW_NEW_REDUCED_PRIMITIVE
|
431 BRW_NEW_VUE_MAP_GEOM_OUT
);
435 brw_wm_populate_key(struct brw_context
*brw
, struct brw_wm_prog_key
*key
)
437 struct gl_context
*ctx
= &brw
->ctx
;
438 /* BRW_NEW_FRAGMENT_PROGRAM */
439 const struct brw_fragment_program
*fp
=
440 (struct brw_fragment_program
*) brw
->fragment_program
;
441 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
445 memset(key
, 0, sizeof(*key
));
447 /* Build the index for table lookup
451 if (fp
->program
.Base
.nir
->info
.fs
.uses_discard
||
452 ctx
->Color
.AlphaEnabled
) {
453 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
456 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
457 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
461 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
463 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
464 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
466 /* _NEW_STENCIL | _NEW_BUFFERS */
467 if (ctx
->Stencil
._Enabled
) {
468 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
470 if (ctx
->Stencil
.WriteMask
[0] ||
471 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
472 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
474 key
->iz_lookup
= lookup
;
479 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
480 if (ctx
->Line
.SmoothFlag
) {
481 if (brw
->reduced_primitive
== GL_LINES
) {
484 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
485 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
486 line_aa
= AA_SOMETIMES
;
488 if (ctx
->Polygon
.BackMode
== GL_LINE
||
489 (ctx
->Polygon
.CullFlag
&&
490 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
493 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
494 line_aa
= AA_SOMETIMES
;
496 if ((ctx
->Polygon
.CullFlag
&&
497 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
503 key
->line_aa
= line_aa
;
506 key
->high_quality_derivatives
=
507 ctx
->Hint
.FragmentShaderDerivative
== GL_NICEST
;
510 key
->stats_wm
= brw
->stats_wm
;
513 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
515 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
516 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
519 brw_populate_sampler_prog_key_data(ctx
, prog
, &key
->tex
);
522 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
525 key
->force_dual_color_blend
= brw
->dual_color_blend_by_location
&&
526 (ctx
->Color
.BlendEnabled
& 1) && ctx
->Color
.Blend
[0]._UsesDualSrc
;
528 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
529 key
->replicate_alpha
= ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
530 (ctx
->Multisample
.SampleAlphaToCoverage
|| ctx
->Color
.AlphaEnabled
);
532 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
533 /* Ignore sample qualifier while computing this flag. */
534 if (ctx
->Multisample
.Enabled
) {
535 key
->persample_interp
=
536 ctx
->Multisample
.SampleShading
&&
537 (ctx
->Multisample
.MinSampleShadingValue
*
538 _mesa_geometric_samples(ctx
->DrawBuffer
) > 1);
540 key
->multisample_fbo
= _mesa_geometric_samples(ctx
->DrawBuffer
) > 1;
543 /* BRW_NEW_VUE_MAP_GEOM_OUT */
544 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->program
.Base
.InputsRead
&
545 BRW_FS_VARYING_INPUT_MASK
) > 16)
546 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
549 /* _NEW_COLOR | _NEW_BUFFERS */
550 /* Pre-gen6, the hardware alpha test always used each render
551 * target's alpha to do alpha test, as opposed to render target 0's alpha
552 * like GL requires. Fix that by building the alpha test into the
553 * shader, and we'll skip enabling the fixed function alpha test.
555 if (brw
->gen
< 6 && ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
556 ctx
->Color
.AlphaEnabled
) {
557 key
->alpha_test_func
= ctx
->Color
.AlphaFunc
;
558 key
->alpha_test_ref
= ctx
->Color
.AlphaRef
;
561 /* The unique fragment program ID */
562 key
->program_string_id
= fp
->id
;
564 /* Whether reads from the framebuffer should behave coherently. */
565 key
->coherent_fb_fetch
= ctx
->Extensions
.MESA_shader_framebuffer_fetch
;
569 brw_upload_wm_prog(struct brw_context
*brw
)
571 struct gl_context
*ctx
= &brw
->ctx
;
572 struct gl_shader_program
*current
= ctx
->_Shader
->_CurrentFragmentProgram
;
573 struct brw_wm_prog_key key
;
574 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
575 brw
->fragment_program
;
577 if (!brw_wm_state_dirty(brw
))
580 brw_wm_populate_key(brw
, &key
);
582 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
584 &brw
->wm
.base
.prog_offset
,
585 &brw
->wm
.base
.prog_data
)) {
586 bool success
= brw_codegen_wm_prog(brw
, current
, fp
, &key
);
593 brw_fs_precompile(struct gl_context
*ctx
,
594 struct gl_shader_program
*shader_prog
,
595 struct gl_program
*prog
)
597 struct brw_context
*brw
= brw_context(ctx
);
598 struct brw_wm_prog_key key
;
600 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*) prog
;
601 struct brw_fragment_program
*bfp
= brw_fragment_program(fp
);
603 memset(&key
, 0, sizeof(key
));
606 if (fp
->Base
.nir
->info
.fs
.uses_discard
)
607 key
.iz_lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
609 if (fp
->Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
610 key
.iz_lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
612 /* Just assume depth testing. */
613 key
.iz_lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
614 key
.iz_lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
617 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->Base
.InputsRead
&
618 BRW_FS_VARYING_INPUT_MASK
) > 16)
619 key
.input_slots_valid
= fp
->Base
.InputsRead
| VARYING_BIT_POS
;
621 brw_setup_tex_for_precompile(brw
, &key
.tex
, &fp
->Base
);
623 key
.nr_color_regions
= _mesa_bitcount_64(fp
->Base
.OutputsWritten
&
624 ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH
) |
625 BITFIELD64_BIT(FRAG_RESULT_STENCIL
) |
626 BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)));
628 key
.program_string_id
= bfp
->id
;
630 /* Whether reads from the framebuffer should behave coherently. */
631 key
.coherent_fb_fetch
= ctx
->Extensions
.MESA_shader_framebuffer_fetch
;
633 uint32_t old_prog_offset
= brw
->wm
.base
.prog_offset
;
634 struct brw_stage_prog_data
*old_prog_data
= brw
->wm
.base
.prog_data
;
636 bool success
= brw_codegen_wm_prog(brw
, shader_prog
, bfp
, &key
);
638 brw
->wm
.base
.prog_offset
= old_prog_offset
;
639 brw
->wm
.base
.prog_data
= old_prog_data
;