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"
40 #include "util/ralloc.h"
43 assign_fs_binding_table_offsets(const struct brw_device_info
*devinfo
,
44 const struct gl_shader_program
*shader_prog
,
45 const struct gl_program
*prog
,
46 const struct brw_wm_prog_key
*key
,
47 struct brw_wm_prog_data
*prog_data
)
49 uint32_t next_binding_table_offset
= 0;
51 /* If there are no color regions, we still perform an FB write to a null
52 * renderbuffer, which we place at surface index 0.
54 prog_data
->binding_table
.render_target_start
= next_binding_table_offset
;
55 next_binding_table_offset
+= MAX2(key
->nr_color_regions
, 1);
57 brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT
, devinfo
,
58 shader_prog
, prog
, &prog_data
->base
,
59 next_binding_table_offset
);
63 * All Mesa program -> GPU code generation goes through this function.
64 * Depending on the instructions used (i.e. flow control instructions)
65 * we'll use one of two code generators.
68 brw_codegen_wm_prog(struct brw_context
*brw
,
69 struct gl_shader_program
*prog
,
70 struct brw_fragment_program
*fp
,
71 struct brw_wm_prog_key
*key
)
73 struct gl_context
*ctx
= &brw
->ctx
;
74 void *mem_ctx
= ralloc_context(NULL
);
75 struct brw_wm_prog_data prog_data
;
76 const GLuint
*program
;
77 struct brw_shader
*fs
= NULL
;
79 bool start_busy
= false;
80 double start_time
= 0;
83 fs
= (struct brw_shader
*)prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
85 memset(&prog_data
, 0, sizeof(prog_data
));
87 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
89 prog_data
.base
.use_alt_mode
= true;
91 assign_fs_binding_table_offsets(brw
->intelScreen
->devinfo
, prog
,
92 &fp
->program
.Base
, key
, &prog_data
);
94 /* Allocate the references to the uniforms that will end up in the
95 * prog_data associated with the compiled program, and which will be freed
98 int param_count
= fp
->program
.Base
.nir
->num_uniforms
;
100 prog_data
.base
.nr_image_params
= fs
->base
.NumImages
;
101 /* The backend also sometimes adds params for texture size. */
102 param_count
+= 2 * ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTextureImageUnits
;
103 prog_data
.base
.param
=
104 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
105 prog_data
.base
.pull_param
=
106 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
107 prog_data
.base
.image_param
=
108 rzalloc_array(NULL
, struct brw_image_param
,
109 prog_data
.base
.nr_image_params
);
110 prog_data
.base
.nr_params
= param_count
;
113 brw_nir_setup_glsl_uniforms(fp
->program
.Base
.nir
, prog
, &fp
->program
.Base
,
114 &prog_data
.base
, true);
116 brw_nir_setup_arb_uniforms(fp
->program
.Base
.nir
, &fp
->program
.Base
,
120 if (unlikely(brw
->perf_debug
)) {
121 start_busy
= (brw
->batch
.last_bo
&&
122 drm_intel_bo_busy(brw
->batch
.last_bo
));
123 start_time
= get_time();
126 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
127 brw_dump_ir("fragment", prog
, fs
? &fs
->base
: NULL
, &fp
->program
.Base
);
129 int st_index8
= -1, st_index16
= -1;
130 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
131 st_index8
= brw_get_shader_time_index(brw
, prog
, &fp
->program
.Base
, ST_FS8
);
132 st_index16
= brw_get_shader_time_index(brw
, prog
, &fp
->program
.Base
, ST_FS16
);
135 char *error_str
= NULL
;
136 program
= brw_compile_fs(brw
->intelScreen
->compiler
, brw
, mem_ctx
,
137 key
, &prog_data
, fp
->program
.Base
.nir
,
138 &fp
->program
.Base
, st_index8
, st_index16
,
139 brw
->use_rep_send
, &program_size
, &error_str
);
140 if (program
== NULL
) {
142 prog
->LinkStatus
= false;
143 ralloc_strcat(&prog
->InfoLog
, error_str
);
146 _mesa_problem(NULL
, "Failed to compile fragment shader: %s\n", error_str
);
148 ralloc_free(mem_ctx
);
152 if (unlikely(brw
->perf_debug
) && fs
) {
153 if (fs
->compiled_once
)
154 brw_wm_debug_recompile(brw
, prog
, key
);
155 fs
->compiled_once
= true;
157 if (start_busy
&& !drm_intel_bo_busy(brw
->batch
.last_bo
)) {
158 perf_debug("FS compile took %.03f ms and stalled the GPU\n",
159 (get_time() - start_time
) * 1000);
163 if (prog_data
.base
.total_scratch
) {
164 brw_get_scratch_bo(brw
, &brw
->wm
.base
.scratch_bo
,
165 prog_data
.base
.total_scratch
* brw
->max_wm_threads
);
168 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
169 fprintf(stderr
, "\n");
171 brw_upload_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
172 key
, sizeof(struct brw_wm_prog_key
),
173 program
, program_size
,
174 &prog_data
, sizeof(prog_data
),
175 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
);
177 ralloc_free(mem_ctx
);
183 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
186 perf_debug(" %s %d->%d\n", name
, a
, b
);
194 brw_debug_recompile_sampler_key(struct brw_context
*brw
,
195 const struct brw_sampler_prog_key_data
*old_key
,
196 const struct brw_sampler_prog_key_data
*key
)
200 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
201 found
|= key_debug(brw
, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
202 old_key
->swizzles
[i
], key
->swizzles
[i
]);
204 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 1st coordinate",
205 old_key
->gl_clamp_mask
[0], key
->gl_clamp_mask
[0]);
206 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
207 old_key
->gl_clamp_mask
[1], key
->gl_clamp_mask
[1]);
208 found
|= key_debug(brw
, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
209 old_key
->gl_clamp_mask
[2], key
->gl_clamp_mask
[2]);
210 found
|= key_debug(brw
, "gather channel quirk on any texture unit",
211 old_key
->gather_channel_quirk_mask
, key
->gather_channel_quirk_mask
);
212 found
|= key_debug(brw
, "compressed multisample layout",
213 old_key
->compressed_multisample_layout_mask
,
214 key
->compressed_multisample_layout_mask
);
215 found
|= key_debug(brw
, "16x msaa",
219 for (unsigned int i
= 0; i
< MAX_SAMPLERS
; i
++) {
220 found
|= key_debug(brw
, "textureGather workarounds",
221 old_key
->gen6_gather_wa
[i
], key
->gen6_gather_wa
[i
]);
228 brw_wm_debug_recompile(struct brw_context
*brw
,
229 struct gl_shader_program
*prog
,
230 const struct brw_wm_prog_key
*key
)
232 struct brw_cache_item
*c
= NULL
;
233 const struct brw_wm_prog_key
*old_key
= NULL
;
236 perf_debug("Recompiling fragment shader for program %d\n", prog
->Name
);
238 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
239 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
240 if (c
->cache_id
== BRW_CACHE_FS_PROG
) {
243 if (old_key
->program_string_id
== key
->program_string_id
)
252 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
256 found
|= key_debug(brw
, "alphatest, computed depth, depth test, or "
258 old_key
->iz_lookup
, key
->iz_lookup
);
259 found
|= key_debug(brw
, "depth statistics",
260 old_key
->stats_wm
, key
->stats_wm
);
261 found
|= key_debug(brw
, "flat shading",
262 old_key
->flat_shade
, key
->flat_shade
);
263 found
|= key_debug(brw
, "per-sample shading",
264 old_key
->persample_shading
, key
->persample_shading
);
265 found
|= key_debug(brw
, "per-sample shading and 2x MSAA",
266 old_key
->persample_2x
, key
->persample_2x
);
267 found
|= key_debug(brw
, "number of color buffers",
268 old_key
->nr_color_regions
, key
->nr_color_regions
);
269 found
|= key_debug(brw
, "MRT alpha test or alpha-to-coverage",
270 old_key
->replicate_alpha
, key
->replicate_alpha
);
271 found
|= key_debug(brw
, "rendering to FBO",
272 old_key
->render_to_fbo
, key
->render_to_fbo
);
273 found
|= key_debug(brw
, "fragment color clamping",
274 old_key
->clamp_fragment_color
, key
->clamp_fragment_color
);
275 found
|= key_debug(brw
, "line smoothing",
276 old_key
->line_aa
, key
->line_aa
);
277 found
|= key_debug(brw
, "renderbuffer height",
278 old_key
->drawable_height
, key
->drawable_height
);
279 found
|= key_debug(brw
, "input slots valid",
280 old_key
->input_slots_valid
, key
->input_slots_valid
);
281 found
|= key_debug(brw
, "mrt alpha test function",
282 old_key
->alpha_test_func
, key
->alpha_test_func
);
283 found
|= key_debug(brw
, "mrt alpha test reference value",
284 old_key
->alpha_test_ref
, key
->alpha_test_ref
);
286 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
289 perf_debug(" Something else\n");
294 gen6_gather_workaround(GLenum internalformat
)
296 switch (internalformat
) {
297 case GL_R8I
: return WA_SIGN
| WA_8BIT
;
298 case GL_R8UI
: return WA_8BIT
;
299 case GL_R16I
: return WA_SIGN
| WA_16BIT
;
300 case GL_R16UI
: return WA_16BIT
;
302 /* Note that even though GL_R32I and GL_R32UI have format overrides in
303 * the surface state, there is no shader w/a required.
310 brw_populate_sampler_prog_key_data(struct gl_context
*ctx
,
311 const struct gl_program
*prog
,
312 unsigned sampler_count
,
313 struct brw_sampler_prog_key_data
*key
)
315 struct brw_context
*brw
= brw_context(ctx
);
317 for (int s
= 0; s
< sampler_count
; s
++) {
318 key
->swizzles
[s
] = SWIZZLE_NOOP
;
320 if (!(prog
->SamplersUsed
& (1 << s
)))
323 int unit_id
= prog
->SamplerUnits
[s
];
324 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[unit_id
];
326 if (unit
->_Current
&& unit
->_Current
->Target
!= GL_TEXTURE_BUFFER
) {
327 const struct gl_texture_object
*t
= unit
->_Current
;
328 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
329 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, unit_id
);
331 const bool alpha_depth
= t
->DepthMode
== GL_ALPHA
&&
332 (img
->_BaseFormat
== GL_DEPTH_COMPONENT
||
333 img
->_BaseFormat
== GL_DEPTH_STENCIL
);
335 /* Haswell handles texture swizzling as surface format overrides
336 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
338 if (alpha_depth
|| (brw
->gen
< 8 && !brw
->is_haswell
))
339 key
->swizzles
[s
] = brw_get_texture_swizzle(ctx
, t
);
342 sampler
->MinFilter
!= GL_NEAREST
&&
343 sampler
->MagFilter
!= GL_NEAREST
) {
344 if (sampler
->WrapS
== GL_CLAMP
)
345 key
->gl_clamp_mask
[0] |= 1 << s
;
346 if (sampler
->WrapT
== GL_CLAMP
)
347 key
->gl_clamp_mask
[1] |= 1 << s
;
348 if (sampler
->WrapR
== GL_CLAMP
)
349 key
->gl_clamp_mask
[2] |= 1 << s
;
352 /* gather4's channel select for green from RG32F is broken; requires
353 * a shader w/a on IVB; fixable with just SCS on HSW.
355 if (brw
->gen
== 7 && !brw
->is_haswell
&& prog
->UsesGather
) {
356 if (img
->InternalFormat
== GL_RG32F
)
357 key
->gather_channel_quirk_mask
|= 1 << s
;
360 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
361 * UNORM/FLOAT instead and fix it in the shader.
363 if (brw
->gen
== 6 && prog
->UsesGather
) {
364 key
->gen6_gather_wa
[s
] = gen6_gather_workaround(img
->InternalFormat
);
367 /* If this is a multisample sampler, and uses the CMS MSAA layout,
368 * then we need to emit slightly different code to first sample the
371 struct intel_texture_object
*intel_tex
=
372 intel_texture_object((struct gl_texture_object
*)t
);
375 intel_tex
->mt
->msaa_layout
== INTEL_MSAA_LAYOUT_CMS
) {
376 key
->compressed_multisample_layout_mask
|= 1 << s
;
378 if (intel_tex
->mt
->num_samples
>= 16) {
379 assert(brw
->gen
>= 9);
380 key
->msaa_16
|= 1 << s
;
388 brw_wm_state_dirty (struct brw_context
*brw
)
390 return brw_state_dirty(brw
,
402 BRW_NEW_FRAGMENT_PROGRAM
|
403 BRW_NEW_REDUCED_PRIMITIVE
|
405 BRW_NEW_VUE_MAP_GEOM_OUT
);
409 brw_wm_populate_key(struct brw_context
*brw
, struct brw_wm_prog_key
*key
)
411 struct gl_context
*ctx
= &brw
->ctx
;
412 /* BRW_NEW_FRAGMENT_PROGRAM */
413 const struct brw_fragment_program
*fp
=
414 (struct brw_fragment_program
*) brw
->fragment_program
;
415 const struct gl_program
*prog
= (struct gl_program
*) brw
->fragment_program
;
418 bool program_uses_dfdy
= fp
->program
.UsesDFdy
;
419 const bool multisample_fbo
= _mesa_geometric_samples(ctx
->DrawBuffer
) > 1;
421 memset(key
, 0, sizeof(*key
));
423 /* Build the index for table lookup
427 if (fp
->program
.UsesKill
|| ctx
->Color
.AlphaEnabled
)
428 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
430 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
431 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
435 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
437 if (ctx
->Depth
.Test
&& ctx
->Depth
.Mask
) /* ?? */
438 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
440 /* _NEW_STENCIL | _NEW_BUFFERS */
441 if (ctx
->Stencil
._Enabled
) {
442 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
444 if (ctx
->Stencil
.WriteMask
[0] ||
445 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
446 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
448 key
->iz_lookup
= lookup
;
453 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
454 if (ctx
->Line
.SmoothFlag
) {
455 if (brw
->reduced_primitive
== GL_LINES
) {
458 else if (brw
->reduced_primitive
== GL_TRIANGLES
) {
459 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
460 line_aa
= AA_SOMETIMES
;
462 if (ctx
->Polygon
.BackMode
== GL_LINE
||
463 (ctx
->Polygon
.CullFlag
&&
464 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
467 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
468 line_aa
= AA_SOMETIMES
;
470 if ((ctx
->Polygon
.CullFlag
&&
471 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
477 key
->line_aa
= line_aa
;
480 key
->high_quality_derivatives
=
481 ctx
->Hint
.FragmentShaderDerivative
== GL_NICEST
;
484 key
->stats_wm
= brw
->stats_wm
;
487 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
489 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
490 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
493 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->wm
.base
.sampler_count
,
498 * Include the draw buffer origin and height so that we can calculate
499 * fragment position values relative to the bottom left of the drawable,
500 * from the incoming screen origin relative position we get as part of our
503 * This is only needed for the WM_WPOSXY opcode when the fragment program
504 * uses the gl_FragCoord input.
506 * We could avoid recompiling by including this as a constant referenced by
507 * our program, but if we were to do that it would also be nice to handle
508 * getting that constant updated at batchbuffer submit time (when we
509 * hold the lock and know where the buffer really is) rather than at emit
510 * time when we don't hold the lock and are just guessing. We could also
511 * just avoid using this as key data if the program doesn't use
514 * For DRI2 the origin_x/y will always be (0,0) but we still need the
515 * drawable height in order to invert the Y axis.
517 if (fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) {
518 key
->drawable_height
= _mesa_geometric_height(ctx
->DrawBuffer
);
521 if ((fp
->program
.Base
.InputsRead
& VARYING_BIT_POS
) || program_uses_dfdy
) {
522 key
->render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
);
526 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
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 key
->persample_shading
=
535 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, true) > 1;
536 if (key
->persample_shading
)
537 key
->persample_2x
= _mesa_geometric_samples(ctx
->DrawBuffer
) == 2;
539 key
->compute_pos_offset
=
540 _mesa_get_min_invocations_per_fragment(ctx
, &fp
->program
, false) > 1 &&
541 fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_POS
;
543 key
->compute_sample_id
=
545 ctx
->Multisample
.Enabled
&&
546 (fp
->program
.Base
.SystemValuesRead
& SYSTEM_BIT_SAMPLE_ID
);
548 /* BRW_NEW_VUE_MAP_GEOM_OUT */
549 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->program
.Base
.InputsRead
&
550 BRW_FS_VARYING_INPUT_MASK
) > 16)
551 key
->input_slots_valid
= brw
->vue_map_geom_out
.slots_valid
;
554 /* _NEW_COLOR | _NEW_BUFFERS */
555 /* Pre-gen6, the hardware alpha test always used each render
556 * target's alpha to do alpha test, as opposed to render target 0's alpha
557 * like GL requires. Fix that by building the alpha test into the
558 * shader, and we'll skip enabling the fixed function alpha test.
560 if (brw
->gen
< 6 && ctx
->DrawBuffer
->_NumColorDrawBuffers
> 1 &&
561 ctx
->Color
.AlphaEnabled
) {
562 key
->alpha_test_func
= ctx
->Color
.AlphaFunc
;
563 key
->alpha_test_ref
= ctx
->Color
.AlphaRef
;
566 /* The unique fragment program ID */
567 key
->program_string_id
= fp
->id
;
571 brw_upload_wm_prog(struct brw_context
*brw
)
573 struct gl_context
*ctx
= &brw
->ctx
;
574 struct gl_shader_program
*current
= ctx
->_Shader
->_CurrentFragmentProgram
;
575 struct brw_wm_prog_key key
;
576 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
577 brw
->fragment_program
;
579 if (!brw_wm_state_dirty(brw
))
582 brw_wm_populate_key(brw
, &key
);
584 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_FS_PROG
,
586 &brw
->wm
.base
.prog_offset
, &brw
->wm
.prog_data
)) {
587 bool success
= brw_codegen_wm_prog(brw
, current
, fp
, &key
);
591 brw
->wm
.base
.prog_data
= &brw
->wm
.prog_data
->base
;
595 brw_fs_precompile(struct gl_context
*ctx
,
596 struct gl_shader_program
*shader_prog
,
597 struct gl_program
*prog
)
599 struct brw_context
*brw
= brw_context(ctx
);
600 struct brw_wm_prog_key key
;
602 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*) prog
;
603 struct brw_fragment_program
*bfp
= brw_fragment_program(fp
);
604 bool program_uses_dfdy
= fp
->UsesDFdy
;
606 memset(&key
, 0, sizeof(key
));
610 key
.iz_lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
612 if (fp
->Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
613 key
.iz_lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
615 /* Just assume depth testing. */
616 key
.iz_lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
617 key
.iz_lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
620 if (brw
->gen
< 6 || _mesa_bitcount_64(fp
->Base
.InputsRead
&
621 BRW_FS_VARYING_INPUT_MASK
) > 16)
622 key
.input_slots_valid
= fp
->Base
.InputsRead
| VARYING_BIT_POS
;
624 brw_setup_tex_for_precompile(brw
, &key
.tex
, &fp
->Base
);
626 if (fp
->Base
.InputsRead
& VARYING_BIT_POS
) {
627 key
.drawable_height
= ctx
->DrawBuffer
->Height
;
630 key
.nr_color_regions
= _mesa_bitcount_64(fp
->Base
.OutputsWritten
&
631 ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH
) |
632 BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)));
634 if ((fp
->Base
.InputsRead
& VARYING_BIT_POS
) || program_uses_dfdy
) {
635 key
.render_to_fbo
= _mesa_is_user_fbo(ctx
->DrawBuffer
) ||
636 key
.nr_color_regions
> 1;
639 key
.program_string_id
= bfp
->id
;
641 uint32_t old_prog_offset
= brw
->wm
.base
.prog_offset
;
642 struct brw_wm_prog_data
*old_prog_data
= brw
->wm
.prog_data
;
644 bool success
= brw_codegen_wm_prog(brw
, shader_prog
, bfp
, &key
);
646 brw
->wm
.base
.prog_offset
= old_prog_offset
;
647 brw
->wm
.prog_data
= old_prog_data
;