1/*
2 * Copyright (C) Intel Corp.  2006.  All Rights Reserved.
3 * Intel funded Tungsten Graphics to
4 * develop this 3D driver.
5 *
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:
13 *
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.
17 *
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.
25 */
26#include "brw_context.h"
27#include "brw_wm.h"
28#include "brw_state.h"
29#include "main/enums.h"
30#include "main/formats.h"
31#include "main/fbobject.h"
32#include "main/samplerobj.h"
33#include "main/framebuffer.h"
34#include "program/prog_parameter.h"
35#include "program/program.h"
36#include "brw_mipmap_tree.h"
37#include "brw_image.h"
38#include "brw_fbo.h"
39#include "compiler/brw_nir.h"
40#include "brw_program.h"
41
42#include "util/ralloc.h"
43#include "util/u_math.h"
44
45static void
46assign_fs_binding_table_offsets(const struct intel_device_info *devinfo,
47                                const struct gl_program *prog,
48                                const struct brw_wm_prog_key *key,
49                                struct brw_wm_prog_data *prog_data)
50{
51   /* Render targets implicitly start at surface index 0.  Even if there are
52    * no color regions, we still perform an FB write to a null render target,
53    * which will be surface 0.
54    */
55   uint32_t next_binding_table_offset = MAX2(key->nr_color_regions, 1);
56
57   next_binding_table_offset =
58      brw_assign_common_binding_table_offsets(devinfo, prog, &prog_data->base,
59                                              next_binding_table_offset);
60
61   if (prog->nir->info.outputs_read && !key->coherent_fb_fetch) {
62      prog_data->binding_table.render_target_read_start =
63         next_binding_table_offset;
64      next_binding_table_offset += key->nr_color_regions;
65   }
66
67   /* Update the binding table size */
68   prog_data->base.binding_table.size_bytes = next_binding_table_offset * 4;
69}
70
71static bool
72brw_codegen_wm_prog(struct brw_context *brw,
73                    struct brw_program *fp,
74                    struct brw_wm_prog_key *key,
75                    struct brw_vue_map *vue_map)
76{
77   const struct intel_device_info *devinfo = &brw->screen->devinfo;
78   void *mem_ctx = ralloc_context(NULL);
79   struct brw_wm_prog_data prog_data;
80   const GLuint *program;
81   bool start_busy = false;
82   double start_time = 0;
83
84   nir_shader *nir = nir_shader_clone(mem_ctx, fp->program.nir);
85
86   memset(&prog_data, 0, sizeof(prog_data));
87
88   /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
89   if (fp->program.info.is_arb_asm)
90      prog_data.base.use_alt_mode = true;
91
92   assign_fs_binding_table_offsets(devinfo, &fp->program, key, &prog_data);
93
94   if (!fp->program.info.is_arb_asm) {
95      brw_nir_setup_glsl_uniforms(mem_ctx, nir, &fp->program,
96                                  &prog_data.base, true);
97      if (brw->can_push_ubos) {
98         brw_nir_analyze_ubo_ranges(brw->screen->compiler, nir,
99                                    NULL, prog_data.base.ubo_ranges);
100      }
101   } else {
102      brw_nir_setup_arb_uniforms(mem_ctx, nir, &fp->program, &prog_data.base);
103
104      if (INTEL_DEBUG(DEBUG_WM))
105         brw_dump_arb_asm("fragment", &fp->program);
106   }
107
108   if (unlikely(brw->perf_debug)) {
109      start_busy = (brw->batch.last_bo &&
110                    brw_bo_busy(brw->batch.last_bo));
111      start_time = get_time();
112   }
113
114   struct brw_compile_fs_params params = {
115      .nir = nir,
116      .key = key,
117      .prog_data = &prog_data,
118
119      .allow_spilling = true,
120      .vue_map = vue_map,
121
122      .log_data = brw,
123   };
124
125   if (INTEL_DEBUG(DEBUG_SHADER_TIME)) {
126      params.shader_time = true;
127      params.shader_time_index8 =
128         brw_get_shader_time_index(brw, &fp->program, ST_FS8,
129                                   !fp->program.info.is_arb_asm);
130      params.shader_time_index16 =
131         brw_get_shader_time_index(brw, &fp->program, ST_FS16,
132                                   !fp->program.info.is_arb_asm);
133      params.shader_time_index32 =
134         brw_get_shader_time_index(brw, &fp->program, ST_FS32,
135                                   !fp->program.info.is_arb_asm);
136   }
137
138   program = brw_compile_fs(brw->screen->compiler, mem_ctx, &params);
139
140   if (program == NULL) {
141      if (!fp->program.info.is_arb_asm) {
142         fp->program.sh.data->LinkStatus = LINKING_FAILURE;
143         ralloc_strcat(&fp->program.sh.data->InfoLog, params.error_str);
144      }
145
146      _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", params.error_str);
147
148      ralloc_free(mem_ctx);
149      return false;
150   }
151
152   if (unlikely(brw->perf_debug)) {
153      if (fp->compiled_once) {
154         brw_debug_recompile(brw, MESA_SHADER_FRAGMENT, fp->program.Id,
155                             &key->base);
156      }
157      fp->compiled_once = true;
158
159      if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
160         perf_debug("FS compile took %.03f ms and stalled the GPU\n",
161                    (get_time() - start_time) * 1000);
162      }
163   }
164
165   brw_alloc_stage_scratch(brw, &brw->wm.base, prog_data.base.total_scratch);
166
167   if (INTEL_DEBUG(DEBUG_WM) && fp->program.info.is_arb_asm)
168      fprintf(stderr, "\n");
169
170   /* The param and pull_param arrays will be freed by the shader cache. */
171   ralloc_steal(NULL, prog_data.base.param);
172   ralloc_steal(NULL, prog_data.base.pull_param);
173   brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
174                    key, sizeof(struct brw_wm_prog_key),
175                    program, prog_data.base.program_size,
176                    &prog_data, sizeof(prog_data),
177                    &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
178
179   ralloc_free(mem_ctx);
180
181   return true;
182}
183
184static uint8_t
185gfx6_gather_workaround(GLenum internalformat)
186{
187   switch (internalformat) {
188   case GL_R8I: return WA_SIGN | WA_8BIT;
189   case GL_R8UI: return WA_8BIT;
190   case GL_R16I: return WA_SIGN | WA_16BIT;
191   case GL_R16UI: return WA_16BIT;
192   default:
193      /* Note that even though GL_R32I and GL_R32UI have format overrides in
194       * the surface state, there is no shader w/a required.
195       */
196      return 0;
197   }
198}
199
200static void
201brw_populate_sampler_prog_key_data(struct gl_context *ctx,
202                                   const struct gl_program *prog,
203                                   struct brw_sampler_prog_key_data *key)
204{
205   struct brw_context *brw = brw_context(ctx);
206   const struct intel_device_info *devinfo = &brw->screen->devinfo;
207   GLbitfield mask = prog->SamplersUsed;
208
209   while (mask) {
210      const int s = u_bit_scan(&mask);
211
212      key->swizzles[s] = SWIZZLE_NOOP;
213      key->scale_factors[s] = 0.0f;
214
215      int unit_id = prog->SamplerUnits[s];
216      const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
217
218      if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
219         const struct gl_texture_object *t = unit->_Current;
220         const struct gl_texture_image *img = t->Image[0][t->Attrib.BaseLevel];
221         struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
222
223         const bool alpha_depth = t->Attrib.DepthMode == GL_ALPHA &&
224            (img->_BaseFormat == GL_DEPTH_COMPONENT ||
225             img->_BaseFormat == GL_DEPTH_STENCIL);
226
227         /* Haswell handles texture swizzling as surface format overrides
228          * (except for GL_ALPHA); all other platforms need MOVs in the shader.
229          */
230         if (alpha_depth || (devinfo->verx10 <= 70))
231            key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
232
233         if (devinfo->ver < 8 &&
234             sampler->Attrib.MinFilter != GL_NEAREST &&
235             sampler->Attrib.MagFilter != GL_NEAREST) {
236            if (sampler->Attrib.WrapS == GL_CLAMP)
237               key->gl_clamp_mask[0] |= 1 << s;
238            if (sampler->Attrib.WrapT == GL_CLAMP)
239               key->gl_clamp_mask[1] |= 1 << s;
240            if (sampler->Attrib.WrapR == GL_CLAMP)
241               key->gl_clamp_mask[2] |= 1 << s;
242         }
243
244         /* gather4 for RG32* is broken in multiple ways on Gfx7. */
245         if (devinfo->ver == 7 && prog->info.uses_texture_gather) {
246            switch (img->InternalFormat) {
247            case GL_RG32I:
248            case GL_RG32UI: {
249               /* We have to override the format to R32G32_FLOAT_LD.
250                * This means that SCS_ALPHA and SCS_ONE will return 0x3f8
251                * (1.0) rather than integer 1.  This needs shader hacks.
252                *
253                * On Ivybridge, we whack W (alpha) to ONE in our key's
254                * swizzle.  On Haswell, we look at the original texture
255                * swizzle, and use XYZW with channels overridden to ONE,
256                * leaving normal texture swizzling to SCS.
257                */
258               unsigned src_swizzle =
259                  devinfo->is_haswell ? t->Attrib._Swizzle : key->swizzles[s];
260               for (int i = 0; i < 4; i++) {
261                  unsigned src_comp = GET_SWZ(src_swizzle, i);
262                  if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) {
263                     key->swizzles[i] &= ~(0x7 << (3 * i));
264                     key->swizzles[i] |= SWIZZLE_ONE << (3 * i);
265                  }
266               }
267            }
268            FALLTHROUGH;
269            case GL_RG32F:
270               /* The channel select for green doesn't work - we have to
271                * request blue.  Haswell can use SCS for this, but Ivybridge
272                * needs a shader workaround.
273                */
274               if (!devinfo->is_haswell)
275                  key->gather_channel_quirk_mask |= 1 << s;
276               break;
277            }
278         }
279
280         /* Gfx6's gather4 is broken for UINT/SINT; we treat them as
281          * UNORM/FLOAT instead and fix it in the shader.
282          */
283         if (devinfo->ver == 6 && prog->info.uses_texture_gather) {
284            key->gfx6_gather_wa[s] = gfx6_gather_workaround(img->InternalFormat);
285         }
286
287         /* If this is a multisample sampler, and uses the CMS MSAA layout,
288          * then we need to emit slightly different code to first sample the
289          * MCS surface.
290          */
291         struct brw_texture_object *intel_tex =
292            brw_texture_object((struct gl_texture_object *)t);
293
294         /* From gfx9 onwards some single sampled buffers can also be
295          * compressed. These don't need ld2dms sampling along with mcs fetch.
296          */
297         if (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) {
298            assert(devinfo->ver >= 7);
299            assert(intel_tex->mt->surf.samples > 1);
300            assert(intel_tex->mt->aux_buf);
301            assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
302            key->compressed_multisample_layout_mask |= 1 << s;
303
304            if (intel_tex->mt->surf.samples >= 16) {
305               assert(devinfo->ver >= 9);
306               key->msaa_16 |= 1 << s;
307            }
308         }
309
310         if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
311
312            /* Setup possible scaling factor. */
313            key->scale_factors[s] = intel_tex->planar_format->scaling_factor;
314
315            switch (intel_tex->planar_format->components) {
316            case __DRI_IMAGE_COMPONENTS_Y_UV:
317               key->y_uv_image_mask |= 1 << s;
318               break;
319            case __DRI_IMAGE_COMPONENTS_Y_U_V:
320               key->y_u_v_image_mask |= 1 << s;
321               break;
322            case __DRI_IMAGE_COMPONENTS_Y_XUXV:
323               key->yx_xuxv_image_mask |= 1 << s;
324               break;
325            case __DRI_IMAGE_COMPONENTS_Y_UXVX:
326               key->xy_uxvx_image_mask |= 1 << s;
327               break;
328            case __DRI_IMAGE_COMPONENTS_AYUV:
329               key->ayuv_image_mask |= 1 << s;
330               break;
331            case __DRI_IMAGE_COMPONENTS_XYUV:
332               key->xyuv_image_mask |= 1 << s;
333               break;
334            default:
335               break;
336            }
337
338            switch (intel_tex->yuv_color_space) {
339            case __DRI_YUV_COLOR_SPACE_ITU_REC709:
340              key->bt709_mask |= 1 << s;
341              break;
342            case __DRI_YUV_COLOR_SPACE_ITU_REC2020:
343              key->bt2020_mask |= 1 << s;
344              break;
345            default:
346              break;
347            }
348         }
349
350      }
351   }
352}
353
354void
355brw_populate_base_prog_key(struct gl_context *ctx,
356                           const struct brw_program *prog,
357                           struct brw_base_prog_key *key)
358{
359   key->program_string_id = prog->id;
360   key->subgroup_size_type = BRW_SUBGROUP_SIZE_UNIFORM;
361   brw_populate_sampler_prog_key_data(ctx, &prog->program, &key->tex);
362}
363
364void
365brw_populate_default_base_prog_key(const struct intel_device_info *devinfo,
366                                   const struct brw_program *prog,
367                                   struct brw_base_prog_key *key)
368{
369   key->program_string_id = prog->id;
370   key->subgroup_size_type = BRW_SUBGROUP_SIZE_UNIFORM;
371   brw_setup_tex_for_precompile(devinfo, &key->tex, &prog->program);
372}
373
374static bool
375brw_wm_state_dirty(const struct brw_context *brw)
376{
377   return brw_state_dirty(brw,
378                          _NEW_BUFFERS |
379                          _NEW_COLOR |
380                          _NEW_DEPTH |
381                          _NEW_FRAG_CLAMP |
382                          _NEW_HINT |
383                          _NEW_LIGHT |
384                          _NEW_LINE |
385                          _NEW_MULTISAMPLE |
386                          _NEW_POLYGON |
387                          _NEW_STENCIL |
388                          _NEW_TEXTURE,
389                          BRW_NEW_FRAGMENT_PROGRAM |
390                          BRW_NEW_REDUCED_PRIMITIVE |
391                          BRW_NEW_STATS_WM |
392                          BRW_NEW_VUE_MAP_GEOM_OUT);
393}
394
395void
396brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
397{
398   const struct intel_device_info *devinfo = &brw->screen->devinfo;
399   struct gl_context *ctx = &brw->ctx;
400   /* BRW_NEW_FRAGMENT_PROGRAM */
401   const struct gl_program *prog = brw->programs[MESA_SHADER_FRAGMENT];
402   const struct brw_program *fp = brw_program_const(prog);
403   GLuint lookup = 0;
404   GLuint line_aa;
405
406   memset(key, 0, sizeof(*key));
407
408   /* Build the index for table lookup
409    */
410   if (devinfo->ver < 6) {
411      struct brw_renderbuffer *depth_irb =
412         brw_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH);
413
414      /* _NEW_COLOR */
415      if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
416         lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
417      }
418
419      if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
420         lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
421      }
422
423      /* _NEW_DEPTH */
424      if (depth_irb && ctx->Depth.Test) {
425         lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
426
427         if (brw_depth_writes_enabled(brw))
428            lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
429      }
430
431      /* _NEW_STENCIL | _NEW_BUFFERS */
432      if (brw->stencil_enabled) {
433         lookup |= BRW_WM_IZ_STENCIL_TEST_ENABLE_BIT;
434
435         if (ctx->Stencil.WriteMask[0] ||
436             ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
437            lookup |= BRW_WM_IZ_STENCIL_WRITE_ENABLE_BIT;
438      }
439      key->iz_lookup = lookup;
440   }
441
442   line_aa = BRW_WM_AA_NEVER;
443
444   /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
445   if (ctx->Line.SmoothFlag) {
446      if (brw->reduced_primitive == GL_LINES) {
447         line_aa = BRW_WM_AA_ALWAYS;
448      }
449      else if (brw->reduced_primitive == GL_TRIANGLES) {
450         if (ctx->Polygon.FrontMode == GL_LINE) {
451            line_aa = BRW_WM_AA_SOMETIMES;
452
453            if (ctx->Polygon.BackMode == GL_LINE ||
454                (ctx->Polygon.CullFlag &&
455                 ctx->Polygon.CullFaceMode == GL_BACK))
456               line_aa = BRW_WM_AA_ALWAYS;
457         }
458         else if (ctx->Polygon.BackMode == GL_LINE) {
459            line_aa = BRW_WM_AA_SOMETIMES;
460
461            if ((ctx->Polygon.CullFlag &&
462                 ctx->Polygon.CullFaceMode == GL_FRONT))
463               line_aa = BRW_WM_AA_ALWAYS;
464         }
465      }
466   }
467
468   key->line_aa = line_aa;
469
470   /* _NEW_HINT */
471   key->high_quality_derivatives =
472      prog->info.uses_fddx_fddy &&
473      ctx->Hint.FragmentShaderDerivative == GL_NICEST;
474
475   if (devinfo->ver < 6)
476      key->stats_wm = brw->stats_wm;
477
478   /* _NEW_LIGHT */
479   key->flat_shade =
480      (prog->info.inputs_read & (VARYING_BIT_COL0 | VARYING_BIT_COL1)) &&
481      (ctx->Light.ShadeModel == GL_FLAT);
482
483   /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
484   key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
485
486   /* _NEW_TEXTURE */
487   brw_populate_base_prog_key(ctx, fp, &key->base);
488
489   /* _NEW_BUFFERS */
490   key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
491
492   /* _NEW_COLOR */
493   key->force_dual_color_blend = brw->dual_color_blend_by_location &&
494      (ctx->Color.BlendEnabled & 1) && ctx->Color._BlendUsesDualSrc & 0x1;
495
496   /* _NEW_MULTISAMPLE, _NEW_BUFFERS */
497   key->alpha_to_coverage =  _mesa_is_alpha_to_coverage_enabled(ctx);
498
499   /* _NEW_COLOR, _NEW_BUFFERS */
500   key->alpha_test_replicate_alpha =
501      ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
502      _mesa_is_alpha_test_enabled(ctx);
503
504   /* _NEW_BUFFERS _NEW_MULTISAMPLE */
505   /* Ignore sample qualifier while computing this flag. */
506   if (ctx->Multisample.Enabled) {
507      key->persample_interp =
508         ctx->Multisample.SampleShading &&
509         (ctx->Multisample.MinSampleShadingValue *
510          _mesa_geometric_samples(ctx->DrawBuffer) > 1);
511
512      key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
513   }
514
515   key->ignore_sample_mask_out = !key->multisample_fbo;
516
517   /* BRW_NEW_VUE_MAP_GEOM_OUT */
518   if (devinfo->ver < 6 || util_bitcount64(prog->info.inputs_read &
519                                             BRW_FS_VARYING_INPUT_MASK) > 16) {
520      key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
521   }
522
523   /* _NEW_COLOR | _NEW_BUFFERS */
524   /* Pre-gfx6, the hardware alpha test always used each render
525    * target's alpha to do alpha test, as opposed to render target 0's alpha
526    * like GL requires.  Fix that by building the alpha test into the
527    * shader, and we'll skip enabling the fixed function alpha test.
528    */
529   if (devinfo->ver < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
530       ctx->Color.AlphaEnabled) {
531      key->alpha_test_func = ctx->Color.AlphaFunc;
532      key->alpha_test_ref = ctx->Color.AlphaRef;
533   }
534
535   /* Whether reads from the framebuffer should behave coherently. */
536   key->coherent_fb_fetch = ctx->Extensions.EXT_shader_framebuffer_fetch;
537}
538
539void
540brw_upload_wm_prog(struct brw_context *brw)
541{
542   struct brw_wm_prog_key key;
543   struct brw_program *fp =
544      (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
545
546   if (!brw_wm_state_dirty(brw))
547      return;
548
549   brw_wm_populate_key(brw, &key);
550
551   if (brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG, &key, sizeof(key),
552                        &brw->wm.base.prog_offset, &brw->wm.base.prog_data,
553                        true))
554      return;
555
556   if (brw_disk_cache_upload_program(brw, MESA_SHADER_FRAGMENT))
557      return;
558
559   fp = (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
560   fp->id = key.base.program_string_id;
561
562   ASSERTED bool success = brw_codegen_wm_prog(brw, fp, &key,
563                                                   &brw->vue_map_geom_out);
564   assert(success);
565}
566
567void
568brw_wm_populate_default_key(const struct brw_compiler *compiler,
569                            struct brw_wm_prog_key *key,
570                            struct gl_program *prog)
571{
572   const struct intel_device_info *devinfo = compiler->devinfo;
573
574   memset(key, 0, sizeof(*key));
575
576   brw_populate_default_base_prog_key(devinfo, brw_program(prog),
577                                      &key->base);
578
579   uint64_t outputs_written = prog->info.outputs_written;
580
581   if (devinfo->ver < 6) {
582      if (prog->info.fs.uses_discard)
583         key->iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
584
585      if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
586         key->iz_lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
587
588      /* Just assume depth testing. */
589      key->iz_lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
590      key->iz_lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
591   }
592
593   if (devinfo->ver < 6 || util_bitcount64(prog->info.inputs_read &
594                                             BRW_FS_VARYING_INPUT_MASK) > 16) {
595      key->input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
596   }
597
598   key->nr_color_regions = util_bitcount64(outputs_written &
599         ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
600           BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
601           BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
602
603   /* Whether reads from the framebuffer should behave coherently. */
604   key->coherent_fb_fetch = devinfo->ver >= 9;
605}
606
607bool
608brw_fs_precompile(struct gl_context *ctx, struct gl_program *prog)
609{
610   struct brw_context *brw = brw_context(ctx);
611   const struct intel_device_info *devinfo = &brw->screen->devinfo;
612   struct brw_wm_prog_key key;
613
614   struct brw_program *bfp = brw_program(prog);
615
616   brw_wm_populate_default_key(brw->screen->compiler, &key, prog);
617
618   /* check brw_wm_populate_default_key coherent_fb_fetch setting */
619   assert(key.coherent_fb_fetch ==
620          ctx->Extensions.EXT_shader_framebuffer_fetch);
621
622   uint32_t old_prog_offset = brw->wm.base.prog_offset;
623   struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
624
625   struct brw_vue_map vue_map;
626   if (devinfo->ver < 6) {
627      brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
628                          prog->info.inputs_read | VARYING_BIT_POS,
629                          false, 1);
630   }
631
632   bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map);
633
634   brw->wm.base.prog_offset = old_prog_offset;
635   brw->wm.base.prog_data = old_prog_data;
636
637   return success;
638}
639