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
,
81 struct brw_vue_map
*vue_map
)
83 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
84 struct gl_context
*ctx
= &brw
->ctx
;
85 void *mem_ctx
= ralloc_context(NULL
);
86 struct brw_wm_prog_data prog_data
;
87 const GLuint
*program
;
88 struct brw_shader
*fs
= NULL
;
90 bool start_busy
= false;
91 double start_time
= 0;
94 fs
= (struct brw_shader
*)prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
96 memset(&prog_data
, 0, sizeof(prog_data
));
98 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
100 prog_data
.base
.use_alt_mode
= true;
102 assign_fs_binding_table_offsets(devinfo
, prog
, &fp
->program
, key
,
105 /* Allocate the references to the uniforms that will end up in the
106 * prog_data associated with the compiled program, and which will be freed
107 * by the state cache.
109 int param_count
= fp
->program
.nir
->num_uniforms
/ 4;
111 prog_data
.base
.nr_image_params
= fs
->base
.NumImages
;
112 /* The backend also sometimes adds params for texture size. */
113 param_count
+= 2 * ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
;
114 prog_data
.base
.param
=
115 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
116 prog_data
.base
.pull_param
=
117 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
118 prog_data
.base
.image_param
=
119 rzalloc_array(NULL
, struct brw_image_param
,
120 prog_data
.base
.nr_image_params
);
121 prog_data
.base
.nr_params
= param_count
;
124 brw_nir_setup_glsl_uniforms(fp
->program
.nir
, prog
, &fp
->program
,
125 &prog_data
.base
, true);
127 brw_nir_setup_arb_uniforms(fp
->program
.nir
, &fp
->program
,
131 if (unlikely(brw
->perf_debug
)) {
132 start_busy
= (brw
->batch
.last_bo
&&
133 drm_intel_bo_busy(brw
->batch
.last_bo
));
134 start_time
= get_time();
137 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
138 brw_dump_ir("fragment", prog
, fs
? &fs
->base
: NULL
, &fp
->program
);
140 int st_index8
= -1, st_index16
= -1;
141 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
142 st_index8
= brw_get_shader_time_index(brw
, prog
, &fp
->program
, ST_FS8
);
143 st_index16
= brw_get_shader_time_index(brw
, prog
, &fp
->program
, ST_FS16
);
146 char *error_str
= NULL
;
147 program
= brw_compile_fs(brw
->screen
->compiler
, brw
, mem_ctx
,
148 key
, &prog_data
, fp
->program
.nir
,
149 &fp
->program
, st_index8
, st_index16
,
150 true, brw
->use_rep_send
, vue_map
,
151 &program_size
, &error_str
);
153 if (program
== NULL
) {
155 prog
->LinkStatus
= false;
156 ralloc_strcat(&prog
->InfoLog
, error_str
);
159 _mesa_problem(NULL
, "Failed to compile fragment shader: %s\n", error_str
);
161 ralloc_free(mem_ctx
);
165 if (unlikely(brw
->perf_debug
) && fs
) {
166 if (fs
->compiled_once
)
167 brw_wm_debug_recompile(brw
, prog
, key
);
168 fs
->compiled_once
= true;
170 if (start_busy
&& !drm_intel_bo_busy(brw
->batch
.last_bo
)) {
171 perf_debug("FS compile took %.03f ms and stalled the GPU\n",
172 (get_time() - start_time
) * 1000);
176 brw_alloc_stage_scratch(brw
, &brw
->wm
.base
,
177 prog_data
.base
.total_scratch
,
178 devinfo
->max_wm_threads
);
180 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
181 fprintf(stderr
, "\n");
183 brw_upload_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
184 key
, sizeof(struct brw_wm_prog_key
),
185 program
, program_size
,
186 &prog_data
, sizeof(prog_data
),
187 &brw
->wm
.base
.prog_offset
, &brw
->wm
.base
.prog_data
);
189 ralloc_free(mem_ctx
);
195 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
196 const struct brw_sampler_prog_key_data
*old_key
,
197 const struct brw_sampler_prog_key_data
*key
)
201 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
202 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
203 old_key
->swizzles
[i
], key
->swizzles
[i
]);
205 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
206 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
207 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
208 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
209 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
210 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
211 found
|= key_debug(brw
, "gather channel quirk on any texture unit",
212 old_key
->gather_channel_quirk_mask
, key
->gather_channel_quirk_mask
);
213 found
|= key_debug(brw
, "compressed multisample layout",
214 old_key
->compressed_multisample_layout_mask
,
215 key
->compressed_multisample_layout_mask
);
216 found
|= key_debug(brw
, "16x msaa",
220 found
|= key_debug(brw
, "y_uv image bound",
221 old_key
->y_uv_image_mask
,
222 key
->y_uv_image_mask
);
223 found
|= key_debug(brw
, "y_u_v image bound",
224 old_key
->y_u_v_image_mask
,
225 key
->y_u_v_image_mask
);
226 found
|= key_debug(brw
, "yx_xuxv image bound",
227 old_key
->yx_xuxv_image_mask
,
228 key
->yx_xuxv_image_mask
);
230 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
231 found
|= key_debug(brw
, "textureGather workarounds",
232 old_key
->gen6_gather_wa
[i
], key
->gen6_gather_wa
[i
]);
239 brw_wm_debug_recompile(struct brw_context
*brw
,
240 struct gl_shader_program
*prog
,
241 const struct brw_wm_prog_key
*key
)
243 struct brw_cache_item
*c
= NULL
;
244 const struct brw_wm_prog_key
*old_key
= NULL
;
247 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
249 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
250 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
251 if (c
->cache_id
== BRW_CACHE_FS_PROG
) {
254 if (old_key
->program_string_id
== key
->program_string_id
)
263 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
267 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
269 old_key
->iz_lookup
, key
->iz_lookup
);
270 found
|= key_debug(brw
, "depth statistics",
271 old_key
->stats_wm
, key
->stats_wm
);
272 found
|= key_debug(brw
, "flat shading",
273 old_key
->flat_shade
, key
->flat_shade
);
274 found
|= key_debug(brw
, "per-sample interpolation",
275 old_key
->persample_interp
, key
->persample_interp
);
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
, "fragment color clamping",
281 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
282 found
|= key_debug(brw
, "multisampled FBO",
283 old_key
->multisample_fbo
, key
->multisample_fbo
);
284 found
|= key_debug(brw
, "line smoothing",
285 old_key
->line_aa
, key
->line_aa
);
286 found
|= key_debug(brw
, "input slots valid",
287 old_key
->input_slots_valid
, key
->input_slots_valid
);
288 found
|= key_debug(brw
, "mrt alpha test function",
289 old_key
->alpha_test_func
, key
->alpha_test_func
);
290 found
|= key_debug(brw
, "mrt alpha test reference value",
291 old_key
->alpha_test_ref
, key
->alpha_test_ref
);
293 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
296 perf_debug(" Something else\n");
301 gen6_gather_workaround(GLenum internalformat
)
303 switch (internalformat
) {
304 case GL_R8I
: return WA_SIGN
| WA_8BIT
;
305 case GL_R8UI
: return WA_8BIT
;
306 case GL_R16I
: return WA_SIGN
| WA_16BIT
;
307 case GL_R16UI
: return WA_16BIT
;
309 /* Note that even though GL_R32I and GL_R32UI have format overrides in
310 * the surface state, there is no shader w/a required.
317 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
318 const struct gl_program
*prog
,
319 struct brw_sampler_prog_key_data
*key
)
321 struct brw_context
*brw
= brw_context(ctx
);
322 GLbitfield mask
= prog
->SamplersUsed
;
325 const int s
= u_bit_scan(&mask
);
327 key
->swizzles
[s
] = SWIZZLE_NOOP
;
329 int unit_id
= prog
->SamplerUnits
[s
];
330 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
332 if (unit
->_Current
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
333 const struct gl_texture_object
*t
= unit
->_Current
;
334 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
335 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
337 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
338 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
339 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
341 /* Haswell handles texture swizzling as surface format overrides
342 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
344 if (alpha_depth
|| (brw
->gen
< 8 && !brw
->is_haswell
))
345 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
348 sampler
->MinFilter
!= GL_NEAREST
&&
349 sampler
->MagFilter
!= GL_NEAREST
) {
350 if (sampler
->WrapS
== GL_CLAMP
)
351 key
->gl_clamp_mask
[0] |= 1 << s
;
352 if (sampler
->WrapT
== GL_CLAMP
)
353 key
->gl_clamp_mask
[1] |= 1 << s
;
354 if (sampler
->WrapR
== GL_CLAMP
)
355 key
->gl_clamp_mask
[2] |= 1 << s
;
358 /* gather4's channel select for green from RG32F is broken; requires
359 * a shader w/a on IVB; fixable with just SCS on HSW.
361 if (brw
->gen
== 7 && !brw
->is_haswell
&&
362 prog
->nir
->info
->uses_texture_gather
) {
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
->nir
->info
->uses_texture_gather
) {
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
);
381 /* From gen9 onwards some single sampled buffers can also be
382 * compressed. These don't need ld2dms sampling along with mcs fetch.
385 intel_tex
->mt
->msaa_layout
== INTEL_MSAA_LAYOUT_CMS
&&
386 intel_tex
->mt
->num_samples
> 1) {
387 key
->compressed_multisample_layout_mask
|= 1 << s
;
389 if (intel_tex
->mt
->num_samples
>= 16) {
390 assert(brw
->gen
>= 9);
391 key
->msaa_16
|= 1 << s
;
395 if (t
->Target
== GL_TEXTURE_EXTERNAL_OES
&& intel_tex
->planar_format
) {
396 switch (intel_tex
->planar_format
->components
) {
397 case __DRI_IMAGE_COMPONENTS_Y_UV
:
398 key
->y_uv_image_mask
|= 1 << s
;
400 case __DRI_IMAGE_COMPONENTS_Y_U_V
:
401 key
->y_u_v_image_mask
|= 1 << s
;
403 case __DRI_IMAGE_COMPONENTS_Y_XUXV
:
404 key
->yx_xuxv_image_mask
|= 1 << s
;
416 brw_wm_state_dirty(const struct brw_context
*brw
)
418 return brw_state_dirty(brw
,
430 BRW_NEW_FRAGMENT_PROGRAM
|
431 BRW_NEW_REDUCED_PRIMITIVE
|
433 BRW_NEW_VUE_MAP_GEOM_OUT
);
437 brw_wm_populate_key(struct brw_context
*brw
, struct brw_wm_prog_key
*key
)
439 struct gl_context
*ctx
= &brw
->ctx
;
440 /* BRW_NEW_FRAGMENT_PROGRAM */
441 const struct brw_fragment_program
*fp
=
442 (struct brw_fragment_program
*) brw
->fragment_program
;
443 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
447 memset(key
, 0, sizeof(*key
));
449 /* Build the index for table lookup
453 if (prog
->info
.fs
.uses_discard
|| ctx
->Color
.AlphaEnabled
) {
454 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
457 if (prog
->info
.outputs_written
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
458 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
463 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
465 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
466 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
468 /* _NEW_STENCIL | _NEW_BUFFERS */
469 if (ctx
->Stencil
._Enabled
) {
470 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
472 if (ctx
->Stencil
.WriteMask
[0] ||
473 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
474 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
476 key
->iz_lookup
= lookup
;
481 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
482 if (ctx
->Line
.SmoothFlag
) {
483 if (brw
->reduced_primitive
== GL_LINES
) {
486 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
487 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
488 line_aa
= AA_SOMETIMES
;
490 if (ctx
->Polygon
.BackMode
== GL_LINE
||
491 (ctx
->Polygon
.CullFlag
&&
492 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
495 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
496 line_aa
= AA_SOMETIMES
;
498 if ((ctx
->Polygon
.CullFlag
&&
499 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
505 key
->line_aa
= line_aa
;
508 key
->high_quality_derivatives
=
509 ctx
->Hint
.FragmentShaderDerivative
== GL_NICEST
;
512 key
->stats_wm
= brw
->stats_wm
;
515 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
517 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
518 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
521 brw_populate_sampler_prog_key_data(ctx
, prog
, &key
->tex
);
524 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
527 key
->force_dual_color_blend
= brw
->dual_color_blend_by_location
&&
528 (ctx
->Color
.BlendEnabled
& 1) && ctx
->Color
.Blend
[0]._UsesDualSrc
;
530 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
531 key
->replicate_alpha
= ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
532 (ctx
->Multisample
.SampleAlphaToCoverage
|| ctx
->Color
.AlphaEnabled
);
534 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
535 /* Ignore sample qualifier while computing this flag. */
536 if (ctx
->Multisample
.Enabled
) {
537 key
->persample_interp
=
538 ctx
->Multisample
.SampleShading
&&
539 (ctx
->Multisample
.MinSampleShadingValue
*
540 _mesa_geometric_samples(ctx
->DrawBuffer
) > 1);
542 key
->multisample_fbo
= _mesa_geometric_samples(ctx
->DrawBuffer
) > 1;
545 /* BRW_NEW_VUE_MAP_GEOM_OUT */
546 if (brw
->gen
< 6 || _mesa_bitcount_64(prog
->info
.inputs_read
&
547 BRW_FS_VARYING_INPUT_MASK
) > 16) {
548 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
552 /* _NEW_COLOR | _NEW_BUFFERS */
553 /* Pre-gen6, the hardware alpha test always used each render
554 * target's alpha to do alpha test, as opposed to render target 0's alpha
555 * like GL requires. Fix that by building the alpha test into the
556 * shader, and we'll skip enabling the fixed function alpha test.
558 if (brw
->gen
< 6 && ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
559 ctx
->Color
.AlphaEnabled
) {
560 key
->alpha_test_func
= ctx
->Color
.AlphaFunc
;
561 key
->alpha_test_ref
= ctx
->Color
.AlphaRef
;
564 /* The unique fragment program ID */
565 key
->program_string_id
= fp
->id
;
567 /* Whether reads from the framebuffer should behave coherently. */
568 key
->coherent_fb_fetch
= ctx
->Extensions
.MESA_shader_framebuffer_fetch
;
572 brw_upload_wm_prog(struct brw_context
*brw
)
574 struct gl_context
*ctx
= &brw
->ctx
;
575 struct gl_shader_program
*current
= ctx
->_Shader
->_CurrentFragmentProgram
;
576 struct brw_wm_prog_key key
;
577 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
578 brw
->fragment_program
;
580 if (!brw_wm_state_dirty(brw
))
583 brw_wm_populate_key(brw
, &key
);
585 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
587 &brw
->wm
.base
.prog_offset
,
588 &brw
->wm
.base
.prog_data
)) {
589 bool success
= brw_codegen_wm_prog(brw
, current
, fp
, &key
,
590 &brw
->vue_map_geom_out
);
597 brw_fs_precompile(struct gl_context
*ctx
,
598 struct gl_shader_program
*shader_prog
,
599 struct gl_program
*prog
)
601 struct brw_context
*brw
= brw_context(ctx
);
602 struct brw_wm_prog_key key
;
604 struct brw_fragment_program
*bfp
= brw_fragment_program(prog
);
606 memset(&key
, 0, sizeof(key
));
608 uint64_t outputs_written
= prog
->info
.outputs_written
;
611 if (prog
->info
.fs
.uses_discard
)
612 key
.iz_lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
614 if (outputs_written
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
615 key
.iz_lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
617 /* Just assume depth testing. */
618 key
.iz_lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
619 key
.iz_lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
622 if (brw
->gen
< 6 || _mesa_bitcount_64(prog
->info
.inputs_read
&
623 BRW_FS_VARYING_INPUT_MASK
) > 16) {
624 key
.input_slots_valid
= prog
->info
.inputs_read
| VARYING_BIT_POS
;
627 brw_setup_tex_for_precompile(brw
, &key
.tex
, prog
);
629 key
.nr_color_regions
= _mesa_bitcount_64(outputs_written
&
630 ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH
) |
631 BITFIELD64_BIT(FRAG_RESULT_STENCIL
) |
632 BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)));
634 key
.program_string_id
= bfp
->id
;
636 /* Whether reads from the framebuffer should behave coherently. */
637 key
.coherent_fb_fetch
= ctx
->Extensions
.MESA_shader_framebuffer_fetch
;
639 uint32_t old_prog_offset
= brw
->wm
.base
.prog_offset
;
640 struct brw_stage_prog_data
*old_prog_data
= brw
->wm
.base
.prog_data
;
642 struct brw_vue_map vue_map
;
644 brw_compute_vue_map(&brw
->screen
->devinfo
, &vue_map
,
645 prog
->info
.inputs_read
| VARYING_BIT_POS
,
649 bool success
= brw_codegen_wm_prog(brw
, shader_prog
, bfp
, &key
, &vue_map
);
651 brw
->wm
.base
.prog_offset
= old_prog_offset
;
652 brw
->wm
.base
.prog_data
= old_prog_data
;