• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2018 Collabora Ltd.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * on the rights to use, copy, modify, merge, publish, distribute, sub
8  * license, and/or sell copies of the Software, and to permit persons to whom
9  * the Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
22  */
23 
24 #include "zink_program.h"
25 
26 #include "zink_compiler.h"
27 #include "zink_context.h"
28 #include "zink_descriptors.h"
29 #include "zink_helpers.h"
30 #include "zink_render_pass.h"
31 #include "zink_resource.h"
32 #include "zink_screen.h"
33 #include "zink_state.h"
34 #include "zink_inlines.h"
35 
36 #include "util/hash_table.h"
37 #include "util/set.h"
38 #include "util/u_debug.h"
39 #include "util/u_memory.h"
40 #include "util/u_prim.h"
41 #include "tgsi/tgsi_from_mesa.h"
42 
43 /* for pipeline cache */
44 #define XXH_INLINE_ALL
45 #include "util/xxhash.h"
46 
47 struct gfx_pipeline_cache_entry {
48    struct zink_gfx_pipeline_state state;
49    VkPipeline pipeline;
50 };
51 
52 struct compute_pipeline_cache_entry {
53    struct zink_compute_pipeline_state state;
54    VkPipeline pipeline;
55 };
56 
57 void
debug_describe_zink_gfx_program(char * buf,const struct zink_gfx_program * ptr)58 debug_describe_zink_gfx_program(char *buf, const struct zink_gfx_program *ptr)
59 {
60    sprintf(buf, "zink_gfx_program");
61 }
62 
63 void
debug_describe_zink_compute_program(char * buf,const struct zink_compute_program * ptr)64 debug_describe_zink_compute_program(char *buf, const struct zink_compute_program *ptr)
65 {
66    sprintf(buf, "zink_compute_program");
67 }
68 
69 static bool
shader_key_matches(const struct zink_shader_module * zm,bool ignore_size,const struct zink_shader_key * key,unsigned num_uniforms)70 shader_key_matches(const struct zink_shader_module *zm, bool ignore_size,
71                    const struct zink_shader_key *key, unsigned num_uniforms)
72 {
73    bool key_size_differs = ignore_size ? false : zm->key_size != key->size;
74    if (key_size_differs || zm->num_uniforms != num_uniforms || zm->has_nonseamless != !!key->base.nonseamless_cube_mask)
75       return false;
76    const uint32_t nonseamless_size = zm->has_nonseamless ? sizeof(uint32_t) : 0;
77    return !memcmp(zm->key, key, zm->key_size) &&
78           (!nonseamless_size || !memcmp(zm->key + zm->key_size, &key->base.nonseamless_cube_mask, nonseamless_size)) &&
79           (!num_uniforms || !memcmp(zm->key + zm->key_size + nonseamless_size,
80                                     key->base.inlined_uniform_values, zm->num_uniforms * sizeof(uint32_t)));
81 }
82 
83 static uint32_t
shader_module_hash(const struct zink_shader_module * zm)84 shader_module_hash(const struct zink_shader_module *zm)
85 {
86    const uint32_t nonseamless_size = zm->has_nonseamless ? sizeof(uint32_t) : 0;
87    unsigned key_size = zm->key_size + nonseamless_size + zm->num_uniforms * sizeof(uint32_t);
88    return _mesa_hash_data(zm->key, key_size);
89 }
90 
91 static struct zink_shader_module *
get_shader_module_for_stage(struct zink_context * ctx,struct zink_screen * screen,struct zink_shader * zs,struct zink_gfx_program * prog,struct zink_gfx_pipeline_state * state)92 get_shader_module_for_stage(struct zink_context *ctx, struct zink_screen *screen,
93                             struct zink_shader *zs, struct zink_gfx_program *prog,
94                             struct zink_gfx_pipeline_state *state)
95 {
96    gl_shader_stage stage = zs->nir->info.stage;
97    enum pipe_shader_type pstage = pipe_shader_type_from_mesa(stage);
98    VkShaderModule mod;
99    struct zink_shader_module *zm = NULL;
100    unsigned inline_size = 0, nonseamless_size = 0;
101    struct zink_shader_key *key = &state->shader_keys.key[pstage];
102    bool ignore_key_size = false;
103    if (pstage == PIPE_SHADER_TESS_CTRL && !zs->is_generated) {
104       /* non-generated tcs won't use the shader key */
105       ignore_key_size = true;
106    }
107    if (ctx && zs->nir->info.num_inlinable_uniforms &&
108        ctx->inlinable_uniforms_valid_mask & BITFIELD64_BIT(pstage)) {
109       if (zs->can_inline && (screen->is_cpu || prog->inlined_variant_count[pstage] < ZINK_MAX_INLINED_VARIANTS))
110          inline_size = zs->nir->info.num_inlinable_uniforms;
111       else
112          key->inline_uniforms = false;
113    }
114    if (key->base.nonseamless_cube_mask)
115       nonseamless_size = sizeof(uint32_t);
116 
117    struct zink_shader_module *iter, *next;
118    LIST_FOR_EACH_ENTRY_SAFE(iter, next, &prog->shader_cache[pstage][!!nonseamless_size][!!inline_size], list) {
119       if (!shader_key_matches(iter, ignore_key_size, key, inline_size))
120          continue;
121       list_delinit(&iter->list);
122       zm = iter;
123       break;
124    }
125 
126    if (!zm) {
127       zm = malloc(sizeof(struct zink_shader_module) + key->size + nonseamless_size + inline_size * sizeof(uint32_t));
128       if (!zm) {
129          return NULL;
130       }
131       unsigned patch_vertices = state->shader_keys.key[PIPE_SHADER_TESS_CTRL ].key.tcs.patch_vertices;
132       if (pstage == PIPE_SHADER_TESS_CTRL && zs->is_generated && zs->spirv) {
133          assert(ctx); //TODO async
134          mod = zink_shader_tcs_compile(screen, zs, patch_vertices);
135       } else {
136          mod = zink_shader_compile(screen, zs, prog->nir[stage], key);
137       }
138       if (!mod) {
139          FREE(zm);
140          return NULL;
141       }
142       zm->shader = mod;
143       list_inithead(&zm->list);
144       zm->num_uniforms = inline_size;
145       if (!ignore_key_size) {
146          zm->key_size = key->size;
147          memcpy(zm->key, key, key->size);
148       } else {
149          zm->key_size = 0;
150          memset(zm->key, 0, key->size);
151       }
152       if (nonseamless_size) {
153          /* nonseamless mask gets added to base key if it exists */
154          memcpy(zm->key + key->size, &key->base.nonseamless_cube_mask, nonseamless_size);
155       }
156       zm->has_nonseamless = !!nonseamless_size;
157       if (inline_size)
158          memcpy(zm->key + key->size + nonseamless_size, key->base.inlined_uniform_values, inline_size * sizeof(uint32_t));
159       if (pstage == PIPE_SHADER_TESS_CTRL && zs->is_generated)
160          zm->hash = patch_vertices;
161       else
162          zm->hash = shader_module_hash(zm);
163       zm->default_variant = !inline_size && list_is_empty(&prog->shader_cache[pstage][0][0]);
164       if (inline_size)
165          prog->inlined_variant_count[pstage]++;
166    }
167    list_add(&zm->list, &prog->shader_cache[pstage][!!nonseamless_size][!!inline_size]);
168    return zm;
169 }
170 
171 static void
zink_destroy_shader_module(struct zink_screen * screen,struct zink_shader_module * zm)172 zink_destroy_shader_module(struct zink_screen *screen, struct zink_shader_module *zm)
173 {
174    VKSCR(DestroyShaderModule)(screen->dev, zm->shader, NULL);
175    free(zm);
176 }
177 
178 static void
destroy_shader_cache(struct zink_screen * screen,struct list_head * sc)179 destroy_shader_cache(struct zink_screen *screen, struct list_head *sc)
180 {
181    struct zink_shader_module *zm, *next;
182    LIST_FOR_EACH_ENTRY_SAFE(zm, next, sc, list) {
183       list_delinit(&zm->list);
184       zink_destroy_shader_module(screen, zm);
185    }
186 }
187 
188 static void
update_gfx_shader_modules(struct zink_context * ctx,struct zink_screen * screen,struct zink_gfx_program * prog,uint32_t mask,struct zink_gfx_pipeline_state * state)189 update_gfx_shader_modules(struct zink_context *ctx,
190                       struct zink_screen *screen,
191                       struct zink_gfx_program *prog, uint32_t mask,
192                       struct zink_gfx_pipeline_state *state)
193 {
194    bool hash_changed = false;
195    bool default_variants = true;
196    bool first = !prog->modules[PIPE_SHADER_VERTEX];
197    uint32_t variant_hash = prog->last_variant_hash;
198    u_foreach_bit(pstage, mask) {
199       assert(prog->shaders[pstage]);
200       struct zink_shader_module *zm = get_shader_module_for_stage(ctx, screen, prog->shaders[pstage], prog, state);
201       state->modules[pstage] = zm->shader;
202       if (prog->modules[pstage] == zm)
203          continue;
204       if (prog->modules[pstage])
205          variant_hash ^= prog->modules[pstage]->hash;
206       hash_changed = true;
207       default_variants &= zm->default_variant;
208       prog->modules[pstage] = zm;
209       variant_hash ^= prog->modules[pstage]->hash;
210    }
211 
212    if (hash_changed && state) {
213       if (default_variants && !first)
214          prog->last_variant_hash = prog->default_variant_hash;
215       else {
216          prog->last_variant_hash = variant_hash;
217          if (first) {
218             p_atomic_dec(&prog->base.reference.count);
219             prog->default_variant_hash = prog->last_variant_hash;
220          }
221       }
222 
223       state->modules_changed = true;
224    }
225 }
226 
227 static uint32_t
hash_gfx_pipeline_state(const void * key)228 hash_gfx_pipeline_state(const void *key)
229 {
230    const struct zink_gfx_pipeline_state *state = key;
231    uint32_t hash = _mesa_hash_data(key, offsetof(struct zink_gfx_pipeline_state, hash));
232    if (!state->have_EXT_extended_dynamic_state2)
233       hash = XXH32(&state->dyn_state2, sizeof(state->dyn_state2), hash);
234    if (state->have_EXT_extended_dynamic_state)
235       return hash;
236    return XXH32(&state->dyn_state1, sizeof(state->dyn_state1), hash);
237 }
238 
239 static bool
equals_gfx_pipeline_state(const void * a,const void * b)240 equals_gfx_pipeline_state(const void *a, const void *b)
241 {
242    const struct zink_gfx_pipeline_state *sa = a;
243    const struct zink_gfx_pipeline_state *sb = b;
244    if (sa->uses_dynamic_stride != sb->uses_dynamic_stride)
245       return false;
246    /* dynamic vs rp */
247    if (!!sa->render_pass != !!sb->render_pass)
248       return false;
249    if (!sa->have_EXT_extended_dynamic_state || !sa->uses_dynamic_stride) {
250       if (sa->vertex_buffers_enabled_mask != sb->vertex_buffers_enabled_mask)
251          return false;
252       /* if we don't have dynamic states, we have to hash the enabled vertex buffer bindings */
253       uint32_t mask_a = sa->vertex_buffers_enabled_mask;
254       uint32_t mask_b = sb->vertex_buffers_enabled_mask;
255       while (mask_a || mask_b) {
256          unsigned idx_a = u_bit_scan(&mask_a);
257          unsigned idx_b = u_bit_scan(&mask_b);
258          if (sa->vertex_strides[idx_a] != sb->vertex_strides[idx_b])
259             return false;
260       }
261    }
262    if (!sa->have_EXT_extended_dynamic_state) {
263       if (memcmp(&sa->dyn_state1, &sb->dyn_state1, offsetof(struct zink_pipeline_dynamic_state1, depth_stencil_alpha_state)))
264          return false;
265       if (!!sa->dyn_state1.depth_stencil_alpha_state != !!sb->dyn_state1.depth_stencil_alpha_state ||
266           (sa->dyn_state1.depth_stencil_alpha_state &&
267            memcmp(sa->dyn_state1.depth_stencil_alpha_state, sb->dyn_state1.depth_stencil_alpha_state,
268                   sizeof(struct zink_depth_stencil_alpha_hw_state))))
269          return false;
270    }
271    if (!sa->have_EXT_extended_dynamic_state2) {
272       if (memcmp(&sa->dyn_state2, &sb->dyn_state2, sizeof(sa->dyn_state2)))
273          return false;
274    } else if (!sa->extendedDynamicState2PatchControlPoints) {
275       if (sa->dyn_state2.vertices_per_patch != sb->dyn_state2.vertices_per_patch)
276          return false;
277    }
278    return !memcmp(sa->modules, sb->modules, sizeof(sa->modules)) &&
279           !memcmp(a, b, offsetof(struct zink_gfx_pipeline_state, hash));
280 }
281 
282 void
zink_update_gfx_program(struct zink_context * ctx,struct zink_gfx_program * prog)283 zink_update_gfx_program(struct zink_context *ctx, struct zink_gfx_program *prog)
284 {
285    update_gfx_shader_modules(ctx, zink_screen(ctx->base.screen), prog, ctx->dirty_shader_stages & prog->stages_present, &ctx->gfx_pipeline_state);
286 }
287 
288 static void
update_cs_shader_module(struct zink_context * ctx,struct zink_compute_program * comp)289 update_cs_shader_module(struct zink_context *ctx, struct zink_compute_program *comp)
290 {
291    struct zink_screen *screen = zink_screen(ctx->base.screen);
292    struct zink_shader *zs = comp->shader;
293    VkShaderModule mod;
294    struct zink_shader_module *zm = NULL;
295    unsigned inline_size = 0, nonseamless_size = 0;
296    struct zink_shader_key *key = &ctx->compute_pipeline_state.key;
297 
298    if (ctx && zs->nir->info.num_inlinable_uniforms &&
299        ctx->inlinable_uniforms_valid_mask & BITFIELD64_BIT(PIPE_SHADER_COMPUTE)) {
300       if (screen->is_cpu || comp->inlined_variant_count < ZINK_MAX_INLINED_VARIANTS)
301          inline_size = zs->nir->info.num_inlinable_uniforms;
302       else
303          key->inline_uniforms = false;
304    }
305    if (key->base.nonseamless_cube_mask)
306       nonseamless_size = sizeof(uint32_t);
307 
308    if (inline_size || nonseamless_size) {
309       struct zink_shader_module *iter, *next;
310       LIST_FOR_EACH_ENTRY_SAFE(iter, next, &comp->shader_cache[!!nonseamless_size], list) {
311          if (!shader_key_matches(iter, false, key, inline_size))
312             continue;
313          list_delinit(&iter->list);
314          zm = iter;
315          break;
316       }
317    } else {
318       zm = comp->module;
319    }
320 
321    if (!zm) {
322       zm = malloc(sizeof(struct zink_shader_module) + nonseamless_size + inline_size * sizeof(uint32_t));
323       if (!zm) {
324          return;
325       }
326       mod = zink_shader_compile(screen, zs, comp->shader->nir, key);
327       if (!mod) {
328          FREE(zm);
329          return;
330       }
331       zm->shader = mod;
332       list_inithead(&zm->list);
333       zm->num_uniforms = inline_size;
334       zm->key_size = 0;
335       zm->has_nonseamless = !!nonseamless_size;
336       assert(nonseamless_size || inline_size);
337       if (nonseamless_size)
338          memcpy(zm->key, &key->base.nonseamless_cube_mask, nonseamless_size);
339       if (inline_size)
340          memcpy(zm->key + nonseamless_size, key->base.inlined_uniform_values, inline_size * sizeof(uint32_t));
341       zm->hash = shader_module_hash(zm);
342       zm->default_variant = false;
343       if (inline_size)
344          comp->inlined_variant_count++;
345    }
346    if (zm->num_uniforms || nonseamless_size)
347       list_add(&zm->list, &comp->shader_cache[!!nonseamless_size]);
348    if (comp->curr == zm)
349       return;
350    ctx->compute_pipeline_state.final_hash ^= ctx->compute_pipeline_state.module_hash;
351    comp->curr = zm;
352    ctx->compute_pipeline_state.module_hash = zm->hash;
353    ctx->compute_pipeline_state.final_hash ^= ctx->compute_pipeline_state.module_hash;
354    ctx->compute_pipeline_state.module_changed = true;
355 }
356 
357 void
zink_update_compute_program(struct zink_context * ctx)358 zink_update_compute_program(struct zink_context *ctx)
359 {
360    update_cs_shader_module(ctx, ctx->curr_compute);
361 }
362 
363 VkPipelineLayout
zink_pipeline_layout_create(struct zink_screen * screen,struct zink_program * pg,uint32_t * compat)364 zink_pipeline_layout_create(struct zink_screen *screen, struct zink_program *pg, uint32_t *compat)
365 {
366    VkPipelineLayoutCreateInfo plci = {0};
367    plci.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
368 
369    plci.pSetLayouts = pg->dsl;
370    plci.setLayoutCount = pg->num_dsl;
371 
372    VkPushConstantRange pcr[2] = {0};
373    if (pg->is_compute) {
374       if (((struct zink_compute_program*)pg)->shader->nir->info.stage == MESA_SHADER_KERNEL) {
375          pcr[0].stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
376          pcr[0].offset = 0;
377          pcr[0].size = sizeof(struct zink_cs_push_constant);
378          plci.pushConstantRangeCount = 1;
379       }
380    } else {
381       pcr[0].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
382       pcr[0].offset = offsetof(struct zink_gfx_push_constant, draw_mode_is_indexed);
383       pcr[0].size = 2 * sizeof(unsigned);
384       pcr[1].stageFlags = VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT;
385       pcr[1].offset = offsetof(struct zink_gfx_push_constant, default_inner_level);
386       pcr[1].size = sizeof(float) * 6;
387       plci.pushConstantRangeCount = 2;
388    }
389    plci.pPushConstantRanges = &pcr[0];
390 
391    VkPipelineLayout layout;
392    VkResult result = VKSCR(CreatePipelineLayout)(screen->dev, &plci, NULL, &layout);
393    if (result != VK_SUCCESS) {
394       mesa_loge("vkCreatePipelineLayout failed (%s)", vk_Result_to_str(result));
395       return VK_NULL_HANDLE;
396    }
397 
398    *compat = _mesa_hash_data(pg->dsl, pg->num_dsl * sizeof(pg->dsl[0]));
399 
400    return layout;
401 }
402 
403 static void
assign_io(struct zink_gfx_program * prog,struct zink_shader * stages[ZINK_SHADER_COUNT])404 assign_io(struct zink_gfx_program *prog, struct zink_shader *stages[ZINK_SHADER_COUNT])
405 {
406    struct zink_shader *shaders[PIPE_SHADER_TYPES];
407 
408    /* build array in pipeline order */
409    for (unsigned i = 0; i < ZINK_SHADER_COUNT; i++)
410       shaders[tgsi_processor_to_shader_stage(i)] = stages[i];
411 
412    for (unsigned i = 0; i < MESA_SHADER_FRAGMENT;) {
413       nir_shader *producer = shaders[i]->nir;
414       for (unsigned j = i + 1; j < ZINK_SHADER_COUNT; i++, j++) {
415          struct zink_shader *consumer = shaders[j];
416          if (!consumer)
417             continue;
418          if (!prog->nir[producer->info.stage])
419             prog->nir[producer->info.stage] = nir_shader_clone(prog, producer);
420          if (!prog->nir[j])
421             prog->nir[j] = nir_shader_clone(prog, consumer->nir);
422          zink_compiler_assign_io(prog->nir[producer->info.stage], prog->nir[j]);
423          i = j;
424          break;
425       }
426    }
427 }
428 
429 struct zink_gfx_program *
zink_create_gfx_program(struct zink_context * ctx,struct zink_shader * stages[ZINK_SHADER_COUNT],unsigned vertices_per_patch)430 zink_create_gfx_program(struct zink_context *ctx,
431                         struct zink_shader *stages[ZINK_SHADER_COUNT],
432                         unsigned vertices_per_patch)
433 {
434    struct zink_screen *screen = zink_screen(ctx->base.screen);
435    struct zink_gfx_program *prog = rzalloc(NULL, struct zink_gfx_program);
436    if (!prog)
437       goto fail;
438 
439    pipe_reference_init(&prog->base.reference, 1);
440    util_queue_fence_init(&prog->base.cache_fence);
441 
442    for (int i = 0; i < ZINK_SHADER_COUNT; ++i) {
443       list_inithead(&prog->shader_cache[i][0][0]);
444       list_inithead(&prog->shader_cache[i][0][1]);
445       list_inithead(&prog->shader_cache[i][1][0]);
446       list_inithead(&prog->shader_cache[i][1][1]);
447       if (stages[i]) {
448          prog->shaders[i] = stages[i];
449          prog->stages_present |= BITFIELD_BIT(i);
450       }
451    }
452    if (stages[PIPE_SHADER_TESS_EVAL] && !stages[PIPE_SHADER_TESS_CTRL]) {
453       prog->shaders[PIPE_SHADER_TESS_EVAL]->generated =
454       prog->shaders[PIPE_SHADER_TESS_CTRL] =
455         zink_shader_tcs_create(screen, stages[PIPE_SHADER_VERTEX], vertices_per_patch);
456       prog->stages_present |= BITFIELD_BIT(PIPE_SHADER_TESS_CTRL);
457    }
458 
459    assign_io(prog, prog->shaders);
460 
461    if (stages[PIPE_SHADER_GEOMETRY])
462       prog->last_vertex_stage = stages[PIPE_SHADER_GEOMETRY];
463    else if (stages[PIPE_SHADER_TESS_EVAL])
464       prog->last_vertex_stage = stages[PIPE_SHADER_TESS_EVAL];
465    else
466       prog->last_vertex_stage = stages[PIPE_SHADER_VERTEX];
467 
468    for (int i = 0; i < ARRAY_SIZE(prog->pipelines); ++i) {
469       _mesa_hash_table_init(&prog->pipelines[i], prog, NULL, equals_gfx_pipeline_state);
470       /* only need first 3/4 for point/line/tri/patch */
471       if (screen->info.have_EXT_extended_dynamic_state &&
472           i == (prog->last_vertex_stage->nir->info.stage == MESA_SHADER_TESS_EVAL ? 4 : 3))
473          break;
474    }
475 
476    struct mesa_sha1 sctx;
477    _mesa_sha1_init(&sctx);
478    for (int i = 0; i < ZINK_SHADER_COUNT; ++i) {
479       if (prog->shaders[i]) {
480          simple_mtx_lock(&prog->shaders[i]->lock);
481          _mesa_set_add(prog->shaders[i]->programs, prog);
482          simple_mtx_unlock(&prog->shaders[i]->lock);
483          zink_gfx_program_reference(ctx, NULL, prog);
484          _mesa_sha1_update(&sctx, prog->shaders[i]->base.sha1, sizeof(prog->shaders[i]->base.sha1));
485       }
486    }
487    _mesa_sha1_final(&sctx, prog->base.sha1);
488 
489    if (!screen->descriptor_program_init(ctx, &prog->base))
490       goto fail;
491 
492    zink_screen_get_pipeline_cache(screen, &prog->base);
493    return prog;
494 
495 fail:
496    if (prog)
497       zink_destroy_gfx_program(ctx, prog);
498    return NULL;
499 }
500 
501 static uint32_t
hash_compute_pipeline_state(const void * key)502 hash_compute_pipeline_state(const void *key)
503 {
504    const struct zink_compute_pipeline_state *state = key;
505    uint32_t hash = _mesa_hash_data(state, offsetof(struct zink_compute_pipeline_state, hash));
506    if (state->use_local_size)
507       hash = XXH32(&state->local_size[0], sizeof(state->local_size), hash);
508    return hash;
509 }
510 
511 void
zink_program_update_compute_pipeline_state(struct zink_context * ctx,struct zink_compute_program * comp,const uint block[3])512 zink_program_update_compute_pipeline_state(struct zink_context *ctx, struct zink_compute_program *comp, const uint block[3])
513 {
514    struct zink_shader *zs = comp->shader;
515    bool use_local_size = !(zs->nir->info.workgroup_size[0] ||
516                            zs->nir->info.workgroup_size[1] ||
517                            zs->nir->info.workgroup_size[2]);
518    if (ctx->compute_pipeline_state.use_local_size != use_local_size)
519       ctx->compute_pipeline_state.dirty = true;
520    ctx->compute_pipeline_state.use_local_size = use_local_size;
521 
522    if (ctx->compute_pipeline_state.use_local_size) {
523       for (int i = 0; i < ARRAY_SIZE(ctx->compute_pipeline_state.local_size); i++) {
524          if (ctx->compute_pipeline_state.local_size[i] != block[i])
525             ctx->compute_pipeline_state.dirty = true;
526          ctx->compute_pipeline_state.local_size[i] = block[i];
527       }
528    } else
529       ctx->compute_pipeline_state.local_size[0] =
530       ctx->compute_pipeline_state.local_size[1] =
531       ctx->compute_pipeline_state.local_size[2] = 0;
532 }
533 
534 static bool
equals_compute_pipeline_state(const void * a,const void * b)535 equals_compute_pipeline_state(const void *a, const void *b)
536 {
537    const struct zink_compute_pipeline_state *sa = a;
538    const struct zink_compute_pipeline_state *sb = b;
539    return !memcmp(a, b, offsetof(struct zink_compute_pipeline_state, hash)) &&
540           sa->module == sb->module;
541 }
542 
543 struct zink_compute_program *
zink_create_compute_program(struct zink_context * ctx,struct zink_shader * shader)544 zink_create_compute_program(struct zink_context *ctx, struct zink_shader *shader)
545 {
546    struct zink_screen *screen = zink_screen(ctx->base.screen);
547    struct zink_compute_program *comp = rzalloc(NULL, struct zink_compute_program);
548    if (!comp)
549       goto fail;
550 
551    pipe_reference_init(&comp->base.reference, 1);
552    util_queue_fence_init(&comp->base.cache_fence);
553    comp->base.is_compute = true;
554 
555    comp->curr = comp->module = CALLOC_STRUCT(zink_shader_module);
556    assert(comp->module);
557    comp->module->shader = zink_shader_compile(screen, shader, shader->nir, NULL);
558    assert(comp->module->shader);
559    list_inithead(&comp->shader_cache[0]);
560    list_inithead(&comp->shader_cache[1]);
561 
562    comp->pipelines = _mesa_hash_table_create(NULL, NULL,
563                                              equals_compute_pipeline_state);
564 
565    _mesa_set_add(shader->programs, comp);
566    comp->shader = shader;
567    memcpy(comp->base.sha1, shader->base.sha1, sizeof(shader->base.sha1));
568 
569    if (!screen->descriptor_program_init(ctx, &comp->base))
570       goto fail;
571 
572    zink_screen_get_pipeline_cache(screen, &comp->base);
573    return comp;
574 
575 fail:
576    if (comp)
577       zink_destroy_compute_program(ctx, comp);
578    return NULL;
579 }
580 
581 uint32_t
zink_program_get_descriptor_usage(struct zink_context * ctx,enum pipe_shader_type stage,enum zink_descriptor_type type)582 zink_program_get_descriptor_usage(struct zink_context *ctx, enum pipe_shader_type stage, enum zink_descriptor_type type)
583 {
584    struct zink_shader *zs = NULL;
585    switch (stage) {
586    case PIPE_SHADER_VERTEX:
587    case PIPE_SHADER_TESS_CTRL:
588    case PIPE_SHADER_TESS_EVAL:
589    case PIPE_SHADER_GEOMETRY:
590    case PIPE_SHADER_FRAGMENT:
591       zs = ctx->gfx_stages[stage];
592       break;
593    case PIPE_SHADER_COMPUTE: {
594       zs = ctx->compute_stage;
595       break;
596    }
597    default:
598       unreachable("unknown shader type");
599    }
600    if (!zs)
601       return 0;
602    switch (type) {
603    case ZINK_DESCRIPTOR_TYPE_UBO:
604       return zs->ubos_used;
605    case ZINK_DESCRIPTOR_TYPE_SSBO:
606       return zs->ssbos_used;
607    case ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW:
608       return BITSET_TEST_RANGE(zs->nir->info.textures_used, 0, PIPE_MAX_SAMPLERS - 1);
609    case ZINK_DESCRIPTOR_TYPE_IMAGE:
610       return BITSET_TEST_RANGE(zs->nir->info.images_used, 0, PIPE_MAX_SAMPLERS - 1);
611    default:
612       unreachable("unknown descriptor type!");
613    }
614    return 0;
615 }
616 
617 bool
zink_program_descriptor_is_buffer(struct zink_context * ctx,enum pipe_shader_type stage,enum zink_descriptor_type type,unsigned i)618 zink_program_descriptor_is_buffer(struct zink_context *ctx, enum pipe_shader_type stage, enum zink_descriptor_type type, unsigned i)
619 {
620    struct zink_shader *zs = NULL;
621    switch (stage) {
622    case PIPE_SHADER_VERTEX:
623    case PIPE_SHADER_TESS_CTRL:
624    case PIPE_SHADER_TESS_EVAL:
625    case PIPE_SHADER_GEOMETRY:
626    case PIPE_SHADER_FRAGMENT:
627       zs = ctx->gfx_stages[stage];
628       break;
629    case PIPE_SHADER_COMPUTE: {
630       zs = ctx->compute_stage;
631       break;
632    }
633    default:
634       unreachable("unknown shader type");
635    }
636    if (!zs)
637       return false;
638    return zink_shader_descriptor_is_buffer(zs, type, i);
639 }
640 
641 static unsigned
get_num_bindings(struct zink_shader * zs,enum zink_descriptor_type type)642 get_num_bindings(struct zink_shader *zs, enum zink_descriptor_type type)
643 {
644    switch (type) {
645    case ZINK_DESCRIPTOR_TYPE_UBO:
646    case ZINK_DESCRIPTOR_TYPE_SSBO:
647       return zs->num_bindings[type];
648    default:
649       break;
650    }
651    unsigned num_bindings = 0;
652    for (int i = 0; i < zs->num_bindings[type]; i++)
653       num_bindings += zs->bindings[type][i].size;
654    return num_bindings;
655 }
656 
657 unsigned
zink_program_num_bindings_typed(const struct zink_program * pg,enum zink_descriptor_type type,bool is_compute)658 zink_program_num_bindings_typed(const struct zink_program *pg, enum zink_descriptor_type type, bool is_compute)
659 {
660    unsigned num_bindings = 0;
661    if (is_compute) {
662       struct zink_compute_program *comp = (void*)pg;
663       return get_num_bindings(comp->shader, type);
664    }
665    struct zink_gfx_program *prog = (void*)pg;
666    for (unsigned i = 0; i < ZINK_SHADER_COUNT; i++) {
667       if (prog->shaders[i])
668          num_bindings += get_num_bindings(prog->shaders[i], type);
669    }
670    return num_bindings;
671 }
672 
673 unsigned
zink_program_num_bindings(const struct zink_program * pg,bool is_compute)674 zink_program_num_bindings(const struct zink_program *pg, bool is_compute)
675 {
676    unsigned num_bindings = 0;
677    for (unsigned i = 0; i < ZINK_DESCRIPTOR_TYPES; i++)
678       num_bindings += zink_program_num_bindings_typed(pg, i, is_compute);
679    return num_bindings;
680 }
681 
682 void
zink_destroy_gfx_program(struct zink_context * ctx,struct zink_gfx_program * prog)683 zink_destroy_gfx_program(struct zink_context *ctx,
684                          struct zink_gfx_program *prog)
685 {
686    struct zink_screen *screen = zink_screen(ctx->base.screen);
687    util_queue_fence_wait(&prog->base.cache_fence);
688    if (prog->base.layout)
689       VKSCR(DestroyPipelineLayout)(screen->dev, prog->base.layout, NULL);
690 
691    for (int i = 0; i < ZINK_SHADER_COUNT; ++i) {
692       if (prog->shaders[i]) {
693          _mesa_set_remove_key(prog->shaders[i]->programs, prog);
694          prog->shaders[i] = NULL;
695       }
696       destroy_shader_cache(screen, &prog->shader_cache[i][0][0]);
697       destroy_shader_cache(screen, &prog->shader_cache[i][0][1]);
698       destroy_shader_cache(screen, &prog->shader_cache[i][1][0]);
699       destroy_shader_cache(screen, &prog->shader_cache[i][1][1]);
700       ralloc_free(prog->nir[i]);
701    }
702 
703    unsigned max_idx = ARRAY_SIZE(prog->pipelines);
704    if (screen->info.have_EXT_extended_dynamic_state) {
705       /* only need first 3/4 for point/line/tri/patch */
706       if ((prog->stages_present &
707           (BITFIELD_BIT(PIPE_SHADER_TESS_EVAL) | BITFIELD_BIT(PIPE_SHADER_GEOMETRY))) ==
708           BITFIELD_BIT(PIPE_SHADER_TESS_EVAL))
709          max_idx = 4;
710       else
711          max_idx = 3;
712       max_idx++;
713    }
714 
715    for (int i = 0; i < max_idx; ++i) {
716       hash_table_foreach(&prog->pipelines[i], entry) {
717          struct gfx_pipeline_cache_entry *pc_entry = entry->data;
718 
719          VKSCR(DestroyPipeline)(screen->dev, pc_entry->pipeline, NULL);
720          free(pc_entry);
721       }
722    }
723    if (prog->base.pipeline_cache)
724       VKSCR(DestroyPipelineCache)(screen->dev, prog->base.pipeline_cache, NULL);
725    screen->descriptor_program_deinit(ctx, &prog->base);
726 
727    ralloc_free(prog);
728 }
729 
730 void
zink_destroy_compute_program(struct zink_context * ctx,struct zink_compute_program * comp)731 zink_destroy_compute_program(struct zink_context *ctx,
732                              struct zink_compute_program *comp)
733 {
734    struct zink_screen *screen = zink_screen(ctx->base.screen);
735    util_queue_fence_wait(&comp->base.cache_fence);
736    if (comp->base.layout)
737       VKSCR(DestroyPipelineLayout)(screen->dev, comp->base.layout, NULL);
738 
739    if (comp->shader)
740       _mesa_set_remove_key(comp->shader->programs, comp);
741    destroy_shader_cache(screen, &comp->shader_cache[0]);
742    destroy_shader_cache(screen, &comp->shader_cache[1]);
743 
744    hash_table_foreach(comp->pipelines, entry) {
745       struct compute_pipeline_cache_entry *pc_entry = entry->data;
746 
747       VKSCR(DestroyPipeline)(screen->dev, pc_entry->pipeline, NULL);
748       free(pc_entry);
749    }
750    _mesa_hash_table_destroy(comp->pipelines, NULL);
751    VKSCR(DestroyShaderModule)(screen->dev, comp->module->shader, NULL);
752    free(comp->module);
753    if (comp->base.pipeline_cache)
754       VKSCR(DestroyPipelineCache)(screen->dev, comp->base.pipeline_cache, NULL);
755    screen->descriptor_program_deinit(ctx, &comp->base);
756 
757    ralloc_free(comp);
758 }
759 
760 static unsigned
get_pipeline_idx(bool have_EXT_extended_dynamic_state,enum pipe_prim_type mode,VkPrimitiveTopology vkmode)761 get_pipeline_idx(bool have_EXT_extended_dynamic_state, enum pipe_prim_type mode, VkPrimitiveTopology vkmode)
762 {
763    /* VK_DYNAMIC_STATE_PRIMITIVE_TOPOLOGY specifies that the topology state in
764     * VkPipelineInputAssemblyStateCreateInfo only specifies the topology class,
765     * and the specific topology order and adjacency must be set dynamically
766     * with vkCmdSetPrimitiveTopology before any drawing commands.
767     */
768    if (have_EXT_extended_dynamic_state) {
769       if (mode == PIPE_PRIM_PATCHES)
770          return 3;
771       switch (u_reduced_prim(mode)) {
772       case PIPE_PRIM_POINTS:
773          return 0;
774       case PIPE_PRIM_LINES:
775          return 1;
776       default:
777          return 2;
778       }
779    }
780    return vkmode;
781 }
782 
783 /*
784    VUID-vkCmdBindVertexBuffers2-pStrides-06209
785    If pStrides is not NULL each element of pStrides must be either 0 or greater than or equal
786    to the maximum extent of all vertex input attributes fetched from the corresponding
787    binding, where the extent is calculated as the VkVertexInputAttributeDescription::offset
788    plus VkVertexInputAttributeDescription::format size
789 
790    * thus, if the stride doesn't meet the minimum requirement for a binding,
791    * disable the dynamic state here and use a fully-baked pipeline
792  */
793 static bool
check_vertex_strides(struct zink_context * ctx)794 check_vertex_strides(struct zink_context *ctx)
795 {
796    const struct zink_vertex_elements_state *ves = ctx->element_state;
797    for (unsigned i = 0; i < ves->hw_state.num_bindings; i++) {
798       const struct pipe_vertex_buffer *vb = ctx->vertex_buffers + ves->binding_map[i];
799       unsigned stride = vb->buffer.resource ? vb->stride : 0;
800       if (stride && stride < ves->min_stride[i])
801          return false;
802    }
803    return true;
804 }
805 
806 VkPipeline
zink_get_gfx_pipeline(struct zink_context * ctx,struct zink_gfx_program * prog,struct zink_gfx_pipeline_state * state,enum pipe_prim_type mode)807 zink_get_gfx_pipeline(struct zink_context *ctx,
808                       struct zink_gfx_program *prog,
809                       struct zink_gfx_pipeline_state *state,
810                       enum pipe_prim_type mode)
811 {
812    struct zink_screen *screen = zink_screen(ctx->base.screen);
813    const bool have_EXT_vertex_input_dynamic_state = screen->info.have_EXT_vertex_input_dynamic_state;
814    const bool have_EXT_extended_dynamic_state = screen->info.have_EXT_extended_dynamic_state;
815    bool uses_dynamic_stride = state->uses_dynamic_stride;
816 
817    VkPrimitiveTopology vkmode = zink_primitive_topology(mode);
818    const unsigned idx = get_pipeline_idx(screen->info.have_EXT_extended_dynamic_state, mode, vkmode);
819    assert(idx <= ARRAY_SIZE(prog->pipelines));
820    if (!state->dirty && !state->modules_changed &&
821        (have_EXT_vertex_input_dynamic_state || !ctx->vertex_state_changed) &&
822        idx == state->idx)
823       return state->pipeline;
824 
825    struct hash_entry *entry = NULL;
826 
827    if (state->dirty) {
828       if (state->pipeline) //avoid on first hash
829          state->final_hash ^= state->hash;
830       state->hash = hash_gfx_pipeline_state(state);
831       state->final_hash ^= state->hash;
832       state->dirty = false;
833    }
834    if (!have_EXT_vertex_input_dynamic_state && ctx->vertex_state_changed) {
835       if (state->pipeline)
836          state->final_hash ^= state->vertex_hash;
837       if (have_EXT_extended_dynamic_state)
838          uses_dynamic_stride = check_vertex_strides(ctx);
839       if (!uses_dynamic_stride) {
840          uint32_t hash = 0;
841          /* if we don't have dynamic states, we have to hash the enabled vertex buffer bindings */
842          uint32_t vertex_buffers_enabled_mask = state->vertex_buffers_enabled_mask;
843          hash = XXH32(&vertex_buffers_enabled_mask, sizeof(uint32_t), hash);
844 
845          for (unsigned i = 0; i < state->element_state->num_bindings; i++) {
846             const unsigned buffer_id = ctx->element_state->binding_map[i];
847             struct pipe_vertex_buffer *vb = ctx->vertex_buffers + buffer_id;
848             state->vertex_strides[buffer_id] = vb->buffer.resource ? vb->stride : 0;
849             hash = XXH32(&state->vertex_strides[buffer_id], sizeof(uint32_t), hash);
850          }
851          state->vertex_hash = hash ^ state->element_state->hash;
852       } else
853          state->vertex_hash = state->element_state->hash;
854       state->final_hash ^= state->vertex_hash;
855    }
856    state->modules_changed = false;
857    state->uses_dynamic_stride = uses_dynamic_stride;
858    ctx->vertex_state_changed = false;
859 
860    entry = _mesa_hash_table_search_pre_hashed(&prog->pipelines[idx], state->final_hash, state);
861 
862    if (!entry) {
863       util_queue_fence_wait(&prog->base.cache_fence);
864       VkPipeline pipeline = zink_create_gfx_pipeline(screen, prog, state,
865                                                      ctx->element_state->binding_map,
866                                                      vkmode);
867       if (pipeline == VK_NULL_HANDLE)
868          return VK_NULL_HANDLE;
869 
870       zink_screen_update_pipeline_cache(screen, &prog->base);
871       struct gfx_pipeline_cache_entry *pc_entry = CALLOC_STRUCT(gfx_pipeline_cache_entry);
872       if (!pc_entry)
873          return VK_NULL_HANDLE;
874 
875       memcpy(&pc_entry->state, state, sizeof(*state));
876       pc_entry->pipeline = pipeline;
877 
878       entry = _mesa_hash_table_insert_pre_hashed(&prog->pipelines[idx], state->final_hash, pc_entry, pc_entry);
879       assert(entry);
880    }
881 
882    struct gfx_pipeline_cache_entry *cache_entry = entry->data;
883    state->pipeline = cache_entry->pipeline;
884    state->idx = idx;
885    return state->pipeline;
886 }
887 
888 VkPipeline
zink_get_compute_pipeline(struct zink_screen * screen,struct zink_compute_program * comp,struct zink_compute_pipeline_state * state)889 zink_get_compute_pipeline(struct zink_screen *screen,
890                       struct zink_compute_program *comp,
891                       struct zink_compute_pipeline_state *state)
892 {
893    struct hash_entry *entry = NULL;
894 
895    if (!state->dirty && !state->module_changed)
896       return state->pipeline;
897    if (state->dirty) {
898       if (state->pipeline) //avoid on first hash
899          state->final_hash ^= state->hash;
900       state->hash = hash_compute_pipeline_state(state);
901       state->dirty = false;
902       state->final_hash ^= state->hash;
903    }
904    entry = _mesa_hash_table_search_pre_hashed(comp->pipelines, state->final_hash, state);
905 
906    if (!entry) {
907       util_queue_fence_wait(&comp->base.cache_fence);
908       VkPipeline pipeline = zink_create_compute_pipeline(screen, comp, state);
909 
910       if (pipeline == VK_NULL_HANDLE)
911          return VK_NULL_HANDLE;
912 
913       struct compute_pipeline_cache_entry *pc_entry = CALLOC_STRUCT(compute_pipeline_cache_entry);
914       if (!pc_entry)
915          return VK_NULL_HANDLE;
916 
917       memcpy(&pc_entry->state, state, sizeof(*state));
918       pc_entry->pipeline = pipeline;
919 
920       entry = _mesa_hash_table_insert_pre_hashed(comp->pipelines, state->final_hash, pc_entry, pc_entry);
921       assert(entry);
922    }
923 
924    struct compute_pipeline_cache_entry *cache_entry = entry->data;
925    state->pipeline = cache_entry->pipeline;
926    return state->pipeline;
927 }
928 
929 static inline void
bind_stage(struct zink_context * ctx,enum pipe_shader_type stage,struct zink_shader * shader)930 bind_stage(struct zink_context *ctx, enum pipe_shader_type stage,
931            struct zink_shader *shader)
932 {
933    if (shader && shader->nir->info.num_inlinable_uniforms)
934       ctx->shader_has_inlinable_uniforms_mask |= 1 << stage;
935    else
936       ctx->shader_has_inlinable_uniforms_mask &= ~(1 << stage);
937 
938    if (stage == PIPE_SHADER_COMPUTE) {
939       if (ctx->compute_stage) {
940          ctx->compute_pipeline_state.final_hash ^= ctx->compute_pipeline_state.module_hash;
941          ctx->compute_pipeline_state.module = VK_NULL_HANDLE;
942          ctx->compute_pipeline_state.module_hash = 0;
943       }
944       if (shader && shader != ctx->compute_stage) {
945          struct hash_entry *entry = _mesa_hash_table_search(&ctx->compute_program_cache, shader);
946          if (entry) {
947             ctx->compute_pipeline_state.dirty = true;
948             ctx->curr_compute = entry->data;
949          } else {
950             struct zink_compute_program *comp = zink_create_compute_program(ctx, shader);
951             _mesa_hash_table_insert(&ctx->compute_program_cache, comp->shader, comp);
952             ctx->compute_pipeline_state.dirty = true;
953             ctx->curr_compute = comp;
954             zink_batch_reference_program(&ctx->batch, &ctx->curr_compute->base);
955          }
956          ctx->compute_pipeline_state.module_hash = ctx->curr_compute->curr->hash;
957          ctx->compute_pipeline_state.module = ctx->curr_compute->curr->shader;
958          ctx->compute_pipeline_state.final_hash ^= ctx->compute_pipeline_state.module_hash;
959          if (ctx->compute_pipeline_state.key.base.nonseamless_cube_mask)
960             ctx->dirty_shader_stages |= BITFIELD_BIT(PIPE_SHADER_COMPUTE);
961       } else if (!shader)
962          ctx->curr_compute = NULL;
963       ctx->compute_stage = shader;
964       zink_select_launch_grid(ctx);
965    } else {
966       if (ctx->gfx_stages[stage])
967          ctx->gfx_hash ^= ctx->gfx_stages[stage]->hash;
968       ctx->gfx_stages[stage] = shader;
969       ctx->gfx_dirty = ctx->gfx_stages[PIPE_SHADER_FRAGMENT] && ctx->gfx_stages[PIPE_SHADER_VERTEX];
970       ctx->gfx_pipeline_state.modules_changed = true;
971       if (shader) {
972          ctx->shader_stages |= BITFIELD_BIT(stage);
973          ctx->gfx_hash ^= ctx->gfx_stages[stage]->hash;
974       } else {
975          ctx->gfx_pipeline_state.modules[stage] = VK_NULL_HANDLE;
976          if (ctx->curr_program)
977             ctx->gfx_pipeline_state.final_hash ^= ctx->curr_program->last_variant_hash;
978          ctx->curr_program = NULL;
979          ctx->shader_stages &= ~BITFIELD_BIT(stage);
980       }
981    }
982 }
983 
984 static void
bind_last_vertex_stage(struct zink_context * ctx)985 bind_last_vertex_stage(struct zink_context *ctx)
986 {
987    enum pipe_shader_type old = ctx->last_vertex_stage ? pipe_shader_type_from_mesa(ctx->last_vertex_stage->nir->info.stage) : PIPE_SHADER_TYPES;
988    if (ctx->gfx_stages[PIPE_SHADER_GEOMETRY])
989       ctx->last_vertex_stage = ctx->gfx_stages[PIPE_SHADER_GEOMETRY];
990    else if (ctx->gfx_stages[PIPE_SHADER_TESS_EVAL])
991       ctx->last_vertex_stage = ctx->gfx_stages[PIPE_SHADER_TESS_EVAL];
992    else
993       ctx->last_vertex_stage = ctx->gfx_stages[PIPE_SHADER_VERTEX];
994    enum pipe_shader_type current = ctx->last_vertex_stage ? pipe_shader_type_from_mesa(ctx->last_vertex_stage->nir->info.stage) : PIPE_SHADER_VERTEX;
995    if (old != current) {
996       if (old != PIPE_SHADER_TYPES) {
997          memset(&ctx->gfx_pipeline_state.shader_keys.key[old].key.vs_base, 0, sizeof(struct zink_vs_key_base));
998          ctx->dirty_shader_stages |= BITFIELD_BIT(old);
999       } else {
1000          /* always unset vertex shader values when changing to a non-vs last stage */
1001          memset(&ctx->gfx_pipeline_state.shader_keys.key[PIPE_SHADER_VERTEX].key.vs_base, 0, sizeof(struct zink_vs_key_base));
1002       }
1003 
1004       unsigned num_viewports = ctx->vp_state.num_viewports;
1005       struct zink_screen *screen = zink_screen(ctx->base.screen);
1006       /* number of enabled viewports is based on whether last vertex stage writes viewport index */
1007       if (ctx->last_vertex_stage) {
1008          if (ctx->last_vertex_stage->nir->info.outputs_written & (VARYING_BIT_VIEWPORT | VARYING_BIT_VIEWPORT_MASK))
1009             ctx->vp_state.num_viewports = MIN2(screen->info.props.limits.maxViewports, PIPE_MAX_VIEWPORTS);
1010          else
1011             ctx->vp_state.num_viewports = 1;
1012       } else {
1013          ctx->vp_state.num_viewports = 1;
1014       }
1015       ctx->vp_state_changed |= num_viewports != ctx->vp_state.num_viewports;
1016       if (!screen->info.have_EXT_extended_dynamic_state) {
1017          if (ctx->gfx_pipeline_state.dyn_state1.num_viewports != ctx->vp_state.num_viewports)
1018             ctx->gfx_pipeline_state.dirty = true;
1019          ctx->gfx_pipeline_state.dyn_state1.num_viewports = ctx->vp_state.num_viewports;
1020       }
1021       ctx->last_vertex_stage_dirty = true;
1022    }
1023 }
1024 
1025 static void
zink_bind_vs_state(struct pipe_context * pctx,void * cso)1026 zink_bind_vs_state(struct pipe_context *pctx,
1027                    void *cso)
1028 {
1029    struct zink_context *ctx = zink_context(pctx);
1030    if (!cso && !ctx->gfx_stages[PIPE_SHADER_VERTEX])
1031       return;
1032    bind_stage(ctx, PIPE_SHADER_VERTEX, cso);
1033    bind_last_vertex_stage(ctx);
1034    if (cso) {
1035       struct zink_shader *zs = cso;
1036       ctx->shader_reads_drawid = BITSET_TEST(zs->nir->info.system_values_read, SYSTEM_VALUE_DRAW_ID);
1037       ctx->shader_reads_basevertex = BITSET_TEST(zs->nir->info.system_values_read, SYSTEM_VALUE_BASE_VERTEX);
1038    } else {
1039       ctx->shader_reads_drawid = false;
1040       ctx->shader_reads_basevertex = false;
1041    }
1042 }
1043 
1044 /* if gl_SampleMask[] is written to, we have to ensure that we get a shader with the same sample count:
1045  * in GL, samples==1 means ignore gl_SampleMask[]
1046  * in VK, gl_SampleMask[] is never ignored
1047  */
1048 void
zink_update_fs_key_samples(struct zink_context * ctx)1049 zink_update_fs_key_samples(struct zink_context *ctx)
1050 {
1051    if (!ctx->gfx_stages[PIPE_SHADER_FRAGMENT])
1052       return;
1053    nir_shader *nir = ctx->gfx_stages[PIPE_SHADER_FRAGMENT]->nir;
1054    if (nir->info.outputs_written & (1 << FRAG_RESULT_SAMPLE_MASK)) {
1055       bool samples = zink_get_fs_key(ctx)->samples;
1056       if (samples != (ctx->fb_state.samples > 1))
1057          zink_set_fs_key(ctx)->samples = ctx->fb_state.samples > 1;
1058    }
1059 }
1060 
1061 /* if gl_SampleMask[] is written to, we have to ensure that we get a shader with the same sample count:
1062  * in GL, samples==1 means ignore gl_SampleMask[]
1063  * in VK, gl_SampleMask[] is never ignored
1064  */
1065 void
zink_update_fs_key_samples(struct zink_context * ctx)1066 zink_update_fs_key_samples(struct zink_context *ctx)
1067 {
1068    if (!ctx->gfx_stages[PIPE_SHADER_FRAGMENT])
1069       return;
1070    nir_shader *nir = ctx->gfx_stages[PIPE_SHADER_FRAGMENT]->nir;
1071    if (nir->info.outputs_written & (1 << FRAG_RESULT_SAMPLE_MASK)) {
1072       bool samples = zink_get_fs_key(ctx)->samples;
1073       if (samples != (ctx->fb_state.samples > 1))
1074          zink_set_fs_key(ctx)->samples = ctx->fb_state.samples > 1;
1075    }
1076 }
1077 
1078 static void
zink_bind_fs_state(struct pipe_context * pctx,void * cso)1079 zink_bind_fs_state(struct pipe_context *pctx,
1080                    void *cso)
1081 {
1082    struct zink_context *ctx = zink_context(pctx);
1083    if (!cso && !ctx->gfx_stages[PIPE_SHADER_FRAGMENT])
1084       return;
1085    bind_stage(ctx, PIPE_SHADER_FRAGMENT, cso);
1086    ctx->fbfetch_outputs = 0;
1087    if (cso) {
1088       nir_shader *nir = ctx->gfx_stages[PIPE_SHADER_FRAGMENT]->nir;
1089       if (nir->info.fs.uses_fbfetch_output) {
1090          nir_foreach_shader_out_variable(var, ctx->gfx_stages[PIPE_SHADER_FRAGMENT]->nir) {
1091             if (var->data.fb_fetch_output)
1092                ctx->fbfetch_outputs |= BITFIELD_BIT(var->data.location - FRAG_RESULT_DATA0);
1093          }
1094       }
1095       zink_update_fs_key_samples(ctx);
1096    }
1097    zink_update_fbfetch(ctx);
1098 }
1099 
1100 static void
zink_bind_gs_state(struct pipe_context * pctx,void * cso)1101 zink_bind_gs_state(struct pipe_context *pctx,
1102                    void *cso)
1103 {
1104    struct zink_context *ctx = zink_context(pctx);
1105    if (!cso && !ctx->gfx_stages[PIPE_SHADER_GEOMETRY])
1106       return;
1107    bool had_points = ctx->gfx_stages[PIPE_SHADER_GEOMETRY] ? ctx->gfx_stages[PIPE_SHADER_GEOMETRY]->nir->info.gs.output_primitive == SHADER_PRIM_POINTS : false;
1108    bind_stage(ctx, PIPE_SHADER_GEOMETRY, cso);
1109    bind_last_vertex_stage(ctx);
1110    if (cso) {
1111       if (!had_points && ctx->last_vertex_stage->nir->info.gs.output_primitive == SHADER_PRIM_POINTS)
1112          ctx->gfx_pipeline_state.has_points++;
1113    } else {
1114       if (had_points)
1115          ctx->gfx_pipeline_state.has_points--;
1116    }
1117 }
1118 
1119 static void
zink_bind_tcs_state(struct pipe_context * pctx,void * cso)1120 zink_bind_tcs_state(struct pipe_context *pctx,
1121                    void *cso)
1122 {
1123    bind_stage(zink_context(pctx), PIPE_SHADER_TESS_CTRL, cso);
1124 }
1125 
1126 static void
zink_bind_tes_state(struct pipe_context * pctx,void * cso)1127 zink_bind_tes_state(struct pipe_context *pctx,
1128                    void *cso)
1129 {
1130    struct zink_context *ctx = zink_context(pctx);
1131    if (!cso && !ctx->gfx_stages[PIPE_SHADER_TESS_EVAL])
1132       return;
1133    if (!!ctx->gfx_stages[PIPE_SHADER_TESS_EVAL] != !!cso) {
1134       if (!cso) {
1135          /* if unsetting a TESS that uses a generated TCS, ensure the TCS is unset */
1136          if (ctx->gfx_stages[PIPE_SHADER_TESS_EVAL]->generated)
1137             ctx->gfx_stages[PIPE_SHADER_TESS_CTRL] = NULL;
1138       }
1139    }
1140    bind_stage(ctx, PIPE_SHADER_TESS_EVAL, cso);
1141    bind_last_vertex_stage(ctx);
1142 }
1143 
1144 static void *
zink_create_cs_state(struct pipe_context * pctx,const struct pipe_compute_state * shader)1145 zink_create_cs_state(struct pipe_context *pctx,
1146                      const struct pipe_compute_state *shader)
1147 {
1148    struct nir_shader *nir;
1149    if (shader->ir_type != PIPE_SHADER_IR_NIR)
1150       nir = zink_tgsi_to_nir(pctx->screen, shader->prog);
1151    else
1152       nir = (struct nir_shader *)shader->prog;
1153 
1154    return zink_shader_create(zink_screen(pctx->screen), nir, NULL);
1155 }
1156 
1157 static void
zink_bind_cs_state(struct pipe_context * pctx,void * cso)1158 zink_bind_cs_state(struct pipe_context *pctx,
1159                    void *cso)
1160 {
1161    bind_stage(zink_context(pctx), PIPE_SHADER_COMPUTE, cso);
1162 }
1163 
1164 void
zink_delete_shader_state(struct pipe_context * pctx,void * cso)1165 zink_delete_shader_state(struct pipe_context *pctx, void *cso)
1166 {
1167    zink_shader_free(zink_context(pctx), cso);
1168 }
1169 
1170 void *
zink_create_gfx_shader_state(struct pipe_context * pctx,const struct pipe_shader_state * shader)1171 zink_create_gfx_shader_state(struct pipe_context *pctx, const struct pipe_shader_state *shader)
1172 {
1173    nir_shader *nir;
1174    if (shader->type != PIPE_SHADER_IR_NIR)
1175       nir = zink_tgsi_to_nir(pctx->screen, shader->tokens);
1176    else
1177       nir = (struct nir_shader *)shader->ir.nir;
1178 
1179    return zink_shader_create(zink_screen(pctx->screen), nir, &shader->stream_output);
1180 }
1181 
1182 static void
zink_delete_cached_shader_state(struct pipe_context * pctx,void * cso)1183 zink_delete_cached_shader_state(struct pipe_context *pctx, void *cso)
1184 {
1185    struct zink_screen *screen = zink_screen(pctx->screen);
1186    util_shader_reference(pctx, &screen->shaders, &cso, NULL);
1187 }
1188 
1189 static void *
zink_create_cached_shader_state(struct pipe_context * pctx,const struct pipe_shader_state * shader)1190 zink_create_cached_shader_state(struct pipe_context *pctx, const struct pipe_shader_state *shader)
1191 {
1192    bool cache_hit;
1193    struct zink_screen *screen = zink_screen(pctx->screen);
1194    return util_live_shader_cache_get(pctx, &screen->shaders, shader, &cache_hit);
1195 }
1196 
1197 void
zink_program_init(struct zink_context * ctx)1198 zink_program_init(struct zink_context *ctx)
1199 {
1200    ctx->base.create_vs_state = zink_create_cached_shader_state;
1201    ctx->base.bind_vs_state = zink_bind_vs_state;
1202    ctx->base.delete_vs_state = zink_delete_cached_shader_state;
1203 
1204    ctx->base.create_fs_state = zink_create_cached_shader_state;
1205    ctx->base.bind_fs_state = zink_bind_fs_state;
1206    ctx->base.delete_fs_state = zink_delete_cached_shader_state;
1207 
1208    ctx->base.create_gs_state = zink_create_cached_shader_state;
1209    ctx->base.bind_gs_state = zink_bind_gs_state;
1210    ctx->base.delete_gs_state = zink_delete_cached_shader_state;
1211 
1212    ctx->base.create_tcs_state = zink_create_cached_shader_state;
1213    ctx->base.bind_tcs_state = zink_bind_tcs_state;
1214    ctx->base.delete_tcs_state = zink_delete_cached_shader_state;
1215 
1216    ctx->base.create_tes_state = zink_create_cached_shader_state;
1217    ctx->base.bind_tes_state = zink_bind_tes_state;
1218    ctx->base.delete_tes_state = zink_delete_cached_shader_state;
1219 
1220    ctx->base.create_compute_state = zink_create_cs_state;
1221    ctx->base.bind_compute_state = zink_bind_cs_state;
1222    ctx->base.delete_compute_state = zink_delete_shader_state;
1223 }
1224 
1225 bool
zink_set_rasterizer_discard(struct zink_context * ctx,bool disable)1226 zink_set_rasterizer_discard(struct zink_context *ctx, bool disable)
1227 {
1228    bool value = disable ? false : (ctx->rast_state ? ctx->rast_state->base.rasterizer_discard : false);
1229    bool changed = ctx->gfx_pipeline_state.dyn_state2.rasterizer_discard != value;
1230    ctx->gfx_pipeline_state.dyn_state2.rasterizer_discard = value;
1231    if (!changed)
1232       return false;
1233    if (!zink_screen(ctx->base.screen)->info.have_EXT_extended_dynamic_state2)
1234       ctx->gfx_pipeline_state.dirty |= true;
1235    ctx->rasterizer_discard_changed = true;
1236    return true;
1237 }
1238