summaryrefslogtreecommitdiffstats
path: root/video/out/opengl/shader_cache.c
blob: e702d7e4a116099fa3ccd1fc3467206544bf2c97 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <stdarg.h>
#include <assert.h>

#include <libavutil/sha.h>
#include <libavutil/mem.h>

#include "osdep/io.h"

#include "common/common.h"
#include "options/path.h"
#include "stream/stream.h"
#include "shader_cache.h"
#include "formats.h"
#include "utils.h"

// Force cache flush if more than this number of shaders is created.
#define SC_MAX_ENTRIES 48

union uniform_val {
    float f[9];         // RA_VARTYPE_FLOAT
    int i[4];           // RA_VARTYPE_INT
    struct ra_tex *tex; // RA_VARTYPE_TEX, RA_VARTYPE_IMG_*
    struct ra_buf *buf; // RA_VARTYPE_BUF_*
};

struct sc_uniform {
    struct ra_renderpass_input input;
    const char *glsl_type;
    union uniform_val v;
    char *buffer_format;
    // for UBO entries: these are all assumed to be arrays as far as updating
    // is concerned. ("regular" values are treated like arrays of length 1)
    size_t ubo_length;  // number of array elements (or 0 if not using UBO)
    size_t ubo_rowsize; // size of data in each array row
    size_t ubo_stride;  // stride of each array row
    size_t ubo_offset;  // byte offset within the uniform buffer
};

struct sc_cached_uniform {
    union uniform_val v;
    int index; // for ra_renderpass_input_val
};

struct sc_entry {
    struct ra_renderpass *pass;
    struct sc_cached_uniform *cached_uniforms;
    int num_cached_uniforms;
    bstr total;
    struct timer_pool *timer;
    struct ra_buf *ubo;
    int ubo_index; // for ra_renderpass_input_val.index
};

struct gl_shader_cache {
    struct ra *ra;
    struct mp_log *log;

    // permanent
    char **exts;
    int num_exts;

    // this is modified during use (gl_sc_add() etc.) and reset for each shader
    bstr prelude_text;
    bstr header_text;
    bstr text;

    // Next binding point (texture unit, image unit, buffer binding, etc.)
    // In OpenGL these are separate for each input type
    int next_binding[RA_VARTYPE_COUNT];

    struct ra_renderpass_params params;

    struct sc_entry **entries;
    int num_entries;

    struct sc_entry *current_shader; // set by gl_sc_generate()

    struct sc_uniform *uniforms;
    int num_uniforms;

    int ubo_binding;
    size_t ubo_size;

    struct ra_renderpass_input_val *values;
    int num_values;

    // For checking that the user is calling gl_sc_reset() properly.
    bool needs_reset;

    bool error_state; // true if an error occurred

    // temporary buffers (avoids frequent reallocations)
    bstr tmp[6];

    // For the disk-cache.
    char *cache_dir;
    struct mpv_global *global; // can be NULL
};

static void gl_sc_reset(struct gl_shader_cache *sc);

struct gl_shader_cache *gl_sc_create(struct ra *ra, struct mpv_global *global,
                                     struct mp_log *log)
{
    struct gl_shader_cache *sc = talloc_ptrtype(NULL, sc);
    *sc = (struct gl_shader_cache){
        .ra = ra,
        .global = global,
        .log = log,
    };
    gl_sc_reset(sc);
    return sc;
}

// Reset the previous pass. This must be called after gl_sc_generate and before
// starting a new shader.
static void gl_sc_reset(struct gl_shader_cache *sc)
{
    sc->prelude_text.len = 0;
    sc->header_text.len = 0;
    sc->text.len = 0;
    for (int n = 0; n < sc->num_uniforms; n++)
        talloc_free((void *)sc->uniforms[n].input.name);
    sc->num_uniforms = 0;
    sc->ubo_binding = 0;
    sc->ubo_size = 0;
    for (int i = 0; i < RA_VARTYPE_COUNT; i++)
        sc->next_binding[i] = 0;
    sc->current_shader = NULL;
    sc->params = (struct ra_renderpass_params){0};
    sc->needs_reset = false;
}

static void sc_flush_cache(struct gl_shader_cache *sc)
{
    MP_VERBOSE(sc, "flushing shader cache\n");

    for (int n = 0; n < sc->num_entries; n++) {
        struct sc_entry *e = sc->entries[n];
        ra_buf_free(sc->ra, &e->ubo);
        if (e->pass)
            sc->ra->fns->renderpass_destroy(sc->ra, e->pass);
        timer_pool_destroy(e->timer);
        talloc_free(e);
    }
    sc->num_entries = 0;
}

void gl_sc_destroy(struct gl_shader_cache *sc)
{
    if (!sc)
        return;
    gl_sc_reset(sc);
    sc_flush_cache(sc);
    talloc_free(sc);
}

bool gl_sc_error_state(struct gl_shader_cache *sc)
{
    return sc->error_state;
}

void gl_sc_reset_error(struct gl_shader_cache *sc)
{
    sc->error_state = false;
}

void gl_sc_enable_extension(struct gl_shader_cache *sc, char *name)
{
    for (int n = 0; n < sc->num_exts; n++) {
        if (strcmp(sc->exts[n], name) == 0)
            return;
    }
    MP_TARRAY_APPEND(sc, sc->exts, sc->num_exts, talloc_strdup(sc, name));
}

#define bstr_xappend0(sc, b, s) bstr_xappend(sc, b, bstr0(s))

void gl_sc_add(struct gl_shader_cache *sc, const char *text)
{
    bstr_xappend0(sc, &sc->text, text);
}

void gl_sc_addf(struct gl_shader_cache *sc, const char *textf, ...)
{
    va_list ap;
    va_start(ap, textf);
    bstr_xappend_vasprintf(sc, &sc->text, textf, ap);
    va_end(ap);
}

void gl_sc_hadd(struct gl_shader_cache *sc, const char *text)
{
    bstr_xappend0(sc, &sc->header_text, text);
}

void gl_sc_haddf(struct gl_shader_cache *sc, const char *textf, ...)
{
    va_list ap;
    va_start(ap, textf);
    bstr_xappend_vasprintf(sc, &sc->header_text, textf, ap);
    va_end(ap);
}

void gl_sc_hadd_bstr(struct gl_shader_cache *sc, struct bstr text)
{
    bstr_xappend(sc, &sc->header_text, text);
}

void gl_sc_paddf(struct gl_shader_cache *sc, const char *textf, ...)
{
    va_list ap;
    va_start(ap, textf);
    bstr_xappend_vasprintf(sc, &sc->prelude_text, textf, ap);
    va_end(ap);
}

static struct sc_uniform *find_uniform(struct gl_shader_cache *sc,
                                       const char *name)
{
    struct sc_uniform new = {
        .input = {
            .dim_v = 1,
            .dim_m = 1,
        },
    };

    for (int n = 0; n < sc->num_uniforms; n++) {
        struct sc_uniform *u = &sc->uniforms[n];
        if (strcmp(u->input.name, name) == 0) {
            const char *allocname = u->input.name;
            *u = new;
            u->input.name = allocname;
            return u;
        }
    }

    // not found -> add it
    new.input.name = talloc_strdup(NULL, name);
    MP_TARRAY_APPEND(sc, sc->uniforms, sc->num_uniforms, new);
    return &sc->uniforms[sc->num_uniforms - 1];
}

static int gl_sc_next_binding(struct gl_shader_cache *sc, enum ra_vartype type)
{
    if (sc->ra->caps & RA_CAP_SHARED_BINDING) {
        return sc->next_binding[type]++;
    } else {
        return sc->next_binding[0]++;
    }
}

// Updates the UBO metadata for the given sc_uniform. Assumes type, dim_v and
// dim_m are already set. Computes the correct alignment, size and array stride
// as per the std140 specification.
static void update_ubo_params(struct gl_shader_cache *sc, struct sc_uniform *u)
{
    if (!(sc->ra->caps & RA_CAP_BUF_RO))
        return;

    // Using UBOs with explicit layout(offset) like we do requires GLSL version
    // 440 or higher. In theory the UBO code can also use older versions, but
    // just try and avoid potential headaches. This also ensures they're only
    // used on drivers that are probably modern enough to actually support them
    // correctly.
    if (sc->ra->glsl_version < 440)
        return;

    size_t el_size;
    switch (u->input.type) {
    case RA_VARTYPE_INT:   el_size = sizeof(int);   break;
    case RA_VARTYPE_FLOAT: el_size = sizeof(float); break;
    default: abort();
    }

    u->ubo_rowsize = el_size * u->input.dim_v;

    // std140 packing rules:
    // 1. The alignment of generic values is their size in bytes
    // 2. The alignment of vectors is the vector length * the base count, with
    // the exception of vec3 which is always aligned like vec4
    // 3. The alignment of arrays is that of the element size rounded up to
    // the nearest multiple of vec4
    // 4. Matrices are treated like arrays of vectors
    // 5. Arrays/matrices are laid out with a stride equal to the alignment
    u->ubo_stride = u->ubo_rowsize;
    if (u->input.dim_v == 3)
        u->ubo_stride += el_size;
    if (u->input.dim_m > 1)
        u->ubo_stride = MP_ALIGN_UP(u->ubo_stride, sizeof(float[4]));

    u->ubo_offset = MP_ALIGN_UP(sc->ubo_size, u->ubo_stride);
    u->ubo_length = u->input.dim_m;
    sc->ubo_size = u->ubo_offset + u->ubo_stride * u->ubo_length;
}

void gl_sc_uniform_texture(struct gl_shader_cache *sc, char *name,
                           struct ra_tex *tex)
{
    const char *glsl_type = "sampler2D";
    if (tex->params.dimensions == 1) {
        glsl_type = "sampler1D";
    } else if (tex->params.dimensions == 3) {
        glsl_type = "sampler3D";
    } else if (tex->params.non_normalized) {
        glsl_type = "sampler2DRect";
    } else if (tex->params.external_oes) {
        glsl_type = "samplerExternalOES";
    } else if (tex->params.format->ctype == RA_CTYPE_UINT) {
        glsl_type = sc->ra->glsl_es ? "highp usampler2D" : "usampler2D";
    }

    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_TEX;
    u->glsl_type = glsl_type;
    u->input.binding = gl_sc_next_binding(sc, u->input.type);
    u->v.tex = tex;
}

void gl_sc_uniform_image2D_wo(struct gl_shader_cache *sc, const char *name,
                              struct ra_tex *tex)
{
    gl_sc_enable_extension(sc, "GL_ARB_shader_image_load_store");

    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_IMG_W;
    u->glsl_type = "writeonly image2D";
    u->input.binding = gl_sc_next_binding(sc, u->input.type);
    u->v.tex = tex;
}

void gl_sc_ssbo(struct gl_shader_cache *sc, char *name, struct ra_buf *buf,
                char *format, ...)
{
    assert(sc->ra->caps & RA_CAP_BUF_RW);
    gl_sc_enable_extension(sc, "GL_ARB_shader_storage_buffer_object");

    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_BUF_RW;
    u->glsl_type = "";
    u->input.binding = gl_sc_next_binding(sc, u->input.type);
    u->v.buf = buf;

    va_list ap;
    va_start(ap, format);
    u->buffer_format = ta_vasprintf(sc, format, ap);
    va_end(ap);
}

void gl_sc_uniform_f(struct gl_shader_cache *sc, char *name, float f)
{
    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_FLOAT;
    u->glsl_type = "float";
    update_ubo_params(sc, u);
    u->v.f[0] = f;
}

void gl_sc_uniform_i(struct gl_shader_cache *sc, char *name, int i)
{
    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_INT;
    u->glsl_type = "int";
    update_ubo_params(sc, u);
    u->v.i[0] = i;
}

void gl_sc_uniform_vec2(struct gl_shader_cache *sc, char *name, float f[2])
{
    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_FLOAT;
    u->input.dim_v = 2;
    u->glsl_type = "vec2";
    update_ubo_params(sc, u);
    u->v.f[0] = f[0];
    u->v.f[1] = f[1];
}

void gl_sc_uniform_vec3(struct gl_shader_cache *sc, char *name, GLfloat f[3])
{
    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_FLOAT;
    u->input.dim_v = 3;
    u->glsl_type = "vec3";
    update_ubo_params(sc, u);
    u->v.f[0] = f[0];
    u->v.f[1] = f[1];
    u->v.f[2] = f[2];
}

static void transpose2x2(float r[2 * 2])
{
    MPSWAP(float, r[0+2*1], r[1+2*0]);
}

void gl_sc_uniform_mat2(struct gl_shader_cache *sc, char *name,
                        bool transpose, GLfloat *v)
{
    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_FLOAT;
    u->input.dim_v = 2;
    u->input.dim_m = 2;
    u->glsl_type = "mat2";
    update_ubo_params(sc, u);
    for (int n = 0; n < 4; n++)
        u->v.f[n] = v[n];
    if (transpose)
        transpose2x2(&u->v.f[0]);
}

static void transpose3x3(float r[3 * 3])
{
    MPSWAP(float, r[0+3*1], r[1+3*0]);
    MPSWAP(float, r[0+3*2], r[2+3*0]);
    MPSWAP(float, r[1+3*2], r[2+3*1]);
}

void gl_sc_uniform_mat3(struct gl_shader_cache *sc, char *name,
                        bool transpose, GLfloat *v)
{
    struct sc_uniform *u = find_uniform(sc, name);
    u->input.type = RA_VARTYPE_FLOAT;
    u->input.dim_v = 3;
    u->input.dim_m = 3;
    u->glsl_type = "mat3";
    update_ubo_params(sc, u);
    for (int n = 0; n < 9; n++)
        u->v.f[n] = v[n];
    if (transpose)
        transpose3x3(&u->v.f[0]);
}

// Tell the shader generator (and later gl_sc_draw_data()) about the vertex
// data layout and attribute names. The entries array is terminated with a {0}
// entry. The array memory must remain valid indefinitely (for now).
void gl_sc_set_vertex_format(struct gl_shader_cache *sc,
                             const struct ra_renderpass_input *entries,
                             int vertex_stride)
{
    sc->params.vertex_attribs = (struct ra_renderpass_input *)entries;
    sc->params.num_vertex_attribs = 0;
    while (entries[sc->params.num_vertex_attribs].name)
        sc->params.num_vertex_attribs++;
    sc->params.vertex_stride = vertex_stride;
}

void gl_sc_blend(struct gl_shader_cache *sc,
                 enum ra_blend blend_src_rgb,
                 enum ra_blend blend_dst_rgb,
                 enum ra_blend blend_src_alpha,
                 enum ra_blend blend_dst_alpha)
{
    sc->params.enable_blend = true;
    sc->params.blend_src_rgb = blend_src_rgb;
    sc->params.blend_dst_rgb = blend_dst_rgb;
    sc->params.blend_src_alpha = blend_src_alpha;
    sc->params.blend_dst_alpha = blend_dst_alpha;
}

static const char *vao_glsl_type(const struct ra_renderpass_input *e)
{
    // pretty dumb... too dumb, but works for us
    switch (e->dim_v) {
    case 1: return "float";
    case 2: return "vec2";
    case 3: return "vec3";
    case 4: return "vec4";
    default: abort();
    }
}

static void update_ubo(struct ra *ra, struct ra_buf *ubo, struct sc_uniform *u)
{
    uintptr_t src = (uintptr_t) &u->v;
    size_t dst = u->ubo_offset;
    size_t src_stride = u->ubo_rowsize;
    size_t dst_stride = u->ubo_stride;

    for (int i = 0; i < u->ubo_length; i++) {
        ra->fns->buf_update(ra, ubo, dst, (void *)src, src_stride);
        src += src_stride;
        dst += dst_stride;
    }
}

static void update_uniform(struct gl_shader_cache *sc, struct sc_entry *e,
                           struct sc_uniform *u, int n)
{
    struct sc_cached_uniform *un = &e->cached_uniforms[n];
    size_t size = ra_render_pass_input_data_size(&u->input);
    bool changed = true;
    if (size > 0)
        changed = memcmp(&un->v, &u->v, size) != 0;

    if (changed) {
        un->v = u->v;

        if (u->ubo_length) {
            assert(e->ubo);
            update_ubo(sc->ra, e->ubo, u);
        } else {
            struct ra_renderpass_input_val value = {
                .index = un->index,
                .data = &un->v,
            };
            MP_TARRAY_APPEND(sc, sc->values, sc->num_values, value);
        }
    }
}

void gl_sc_set_cache_dir(struct gl_shader_cache *sc, const char *dir)
{
    talloc_free(sc->cache_dir);
    sc->cache_dir = talloc_strdup(sc, dir);
}

static bool create_pass(struct gl_shader_cache *sc, struct sc_entry *entry)
{
    bool ret = false;

    void *tmp = talloc_new(NULL);
    struct ra_renderpass_params params = sc->params;

    MP_VERBOSE(sc, "new shader program:\n");
    if (sc->header_text.len) {
        MP_VERBOSE(sc, "header:\n");
        mp_log_source(sc->log, MSGL_V, sc->header_text.start);
        MP_VERBOSE(sc, "body:\n");
    }
    if (sc->text.len)
        mp_log_source(sc->log, MSGL_V, sc->text.start);

    // The vertex shader uses mangled names for the vertex attributes, so that
    // the fragment shader can use the "real" names. But the shader is expecting
    // the vertex attribute names (at least with older GLSL targets for GL).
    params.vertex_attribs = talloc_memdup(tmp, params.vertex_attribs,
                params.num_vertex_attribs * sizeof(params.vertex_attribs[0]));
    for (int n = 0; n < params.num_vertex_attribs; n++) {
        struct ra_renderpass_input *attrib = &params.vertex_attribs[n];
        attrib->name = talloc_asprintf(tmp, "vertex_%s", attrib->name);
    }

    const char *cache_header = "mpv shader cache v1\n";
    char *cache_filename = NULL;
    char *cache_dir = NULL;

    if (sc->cache_dir && sc->cache_dir[0]) {
        // Try to load it from a disk cache.
        cache_dir = mp_get_user_path(tmp, sc->global, sc->cache_dir);

        struct AVSHA *sha = av_sha_alloc();
        if (!sha)
            abort();
        av_sha_init(sha, 256);
        av_sha_update(sha, entry->total.start, entry->total.len);

        uint8_t hash[256 / 8];
        av_sha_final(sha, hash);
        av_free(sha);

        char hashstr[256 / 8 * 2 + 1];
        for (int n = 0; n < 256 / 8; n++)
            snprintf(hashstr + n * 2, sizeof(hashstr) - n * 2, "%02X", hash[n]);

        cache_filename = mp_path_join(tmp, cache_dir, hashstr);
        if (stat(cache_filename, &(struct stat){0}) == 0) {
            MP_VERBOSE(sc, "Trying to load shader from disk...\n");
            struct bstr cachedata =
                stream_read_file(cache_filename, tmp, sc->global, 1000000000);
            if (bstr_eatstart0(&cachedata, cache_header))
                params.cached_program = cachedata;
        }
    }

    // If using a UBO, also make sure to add it as an input value so the RA
    // can see it
    if (sc->ubo_size) {
        entry->ubo_index = sc->params.num_inputs;
        struct ra_renderpass_input ubo_input = {
            .name = "UBO",
            .type = RA_VARTYPE_BUF_RO,
            .dim_v = 1,
            .dim_m = 1,
            .binding = sc->ubo_binding,
        };
        MP_TARRAY_APPEND(sc, params.inputs, params.num_inputs, ubo_input);
    }

    entry->pass = sc->ra->fns->renderpass_create(sc->ra, &params);
    if (!entry->pass)
        goto error;

    if (sc->ubo_size) {
        struct ra_buf_params ubo_params = {
            .type = RA_BUF_TYPE_UNIFORM,
            .size = sc->ubo_size,
            .host_mutable = true,
        };

        entry->ubo = ra_buf_create(sc->ra, &ubo_params);
        if (!entry->ubo) {
            MP_ERR(sc, "Failed creating uniform buffer!\n");
            goto error;
        }
    }

    if (entry->pass && cache_filename) {
        bstr nc = entry->pass->params.cached_program;
        if (nc.len && !bstr_equals(params.cached_program, nc)) {
            mp_mkdirp(cache_dir);

            MP_VERBOSE(sc, "Writing shader cache file: %s\n", cache_filename);
            FILE *out = fopen(cache_filename, "wb");
            if (out) {
                fwrite(cache_header, strlen(cache_header), 1, out);
                fwrite(nc.start, nc.len, 1, out);
                fclose(out);
            }
        }
    }

    ret = true;

error:
    talloc_free(tmp);
    return ret;
}

#define ADD(x, ...) bstr_xappend_asprintf(sc, (x), __VA_ARGS__)
#define ADD_BSTR(x, s) bstr_xappend(sc, (x), (s))

static void add_uniforms(struct gl_shader_cache *sc, bstr *dst)
{
    // Add all of the UBO entries separately as members of their own buffer
    if (sc->ubo_size > 0) {
        ADD(dst, "layout(std140, binding=%d) uniform UBO {\n", sc->ubo_binding);
        for (int n = 0; n < sc->num_uniforms; n++) {
            struct sc_uniform *u = &sc->uniforms[n];
            if (!u->ubo_length)
                continue;
            ADD(dst, "layout(offset=%zu) %s %s;\n", u->ubo_offset,
                u->glsl_type, u->input.name);
        }
        ADD(dst, "};\n");
    }

    for (int n = 0; n < sc->num_uniforms; n++) {
        struct sc_uniform *u = &sc->uniforms[n];
        if (u->ubo_length)
            continue;
        switch (u->input.type) {
        case RA_VARTYPE_INT:
        case RA_VARTYPE_FLOAT:
            assert(sc->ra->caps & RA_CAP_GLOBAL_UNIFORM);
            // fall through
        case RA_VARTYPE_TEX:
        case RA_VARTYPE_IMG_W:
            // Vulkan requires explicitly assigning the bindings in the shader
            // source. For OpenGL it's optional, but requires higher GL version
            // so we don't do it (and instead have ra_gl update the bindings
            // after program creation).
            if (sc->ra->glsl_vulkan)
                ADD(dst, "layout(binding=%d) ", u->input.binding);
            ADD(dst, "uniform %s %s;\n", u->glsl_type, u->input.name);
            break;
        case RA_VARTYPE_BUF_RO:
            ADD(dst, "layout(std140, binding=%d) uniform %s { %s };\n",
                u->input.binding, u->input.name, u->buffer_format);
            break;
        case RA_VARTYPE_BUF_RW:
            ADD(dst, "layout(std430, binding=%d) buffer %s { %s };\n",
                u->input.binding, u->input.name, u->buffer_format);
            break;
        }
    }
}

// 1. Generate vertex and fragment shaders from the fragment shader text added
//    with gl_sc_add(). The generated shader program is cached (based on the
//    text), so actual compilation happens only the first time.
// 2. Update the uniforms and textures set with gl_sc_uniform_*.
// 3. Make the new shader program current (glUseProgram()).
// After that, you render, and then you call gc_sc_reset(), which does:
// 1. Unbind the program and all textures.
// 2. Reset the sc state and prepare for a new shader program. (All uniforms
//    and fragment operations needed for the next program have to be re-added.)
static void gl_sc_generate(struct gl_shader_cache *sc,
                           enum ra_renderpass_type type,
                           const struct ra_format *target_format)
{
    int glsl_version = sc->ra->glsl_version;
    int glsl_es = sc->ra->glsl_es ? glsl_version : 0;

    sc->params.type = type;

    // gl_sc_reset() must be called after ending the previous render process,
    // and before starting a new one.
    assert(!sc->needs_reset);
    sc->needs_reset = true;

    // gl_sc_set_vertex_format() must always be called
    assert(sc->params.vertex_attribs);

    // If using a UBO, pick a binding (needed for shader generation)
    if (sc->ubo_size)
        sc->ubo_binding = gl_sc_next_binding(sc, RA_VARTYPE_BUF_RO);

    for (int n = 0; n < MP_ARRAY_SIZE(sc->tmp); n++)
        sc->tmp[n].len = 0;

    // set up shader text (header + uniforms + body)
    bstr *header = &sc->tmp[0];
    ADD(header, "#version %d%s\n", glsl_version, glsl_es >= 300 ? " es" : "");
    if (type == RA_RENDERPASS_TYPE_COMPUTE) {
        // This extension cannot be enabled in fragment shader. Enable it as
        // an exception for compute shader.
        ADD(header, "#extension GL_ARB_compute_shader : enable\n");
    }
    for (int n = 0; n < sc->num_exts; n++)
        ADD(header, "#extension %s : enable\n", sc->exts[n]);
    if (glsl_es) {
        ADD(header, "precision mediump float;\n");
        ADD(header, "precision mediump sampler2D;\n");
        if (sc->ra->caps & RA_CAP_TEX_3D)
            ADD(header, "precision mediump sampler3D;\n");
    }

    if (glsl_version >= 130) {
        ADD(header, "#define tex1D texture\n");
        ADD(header, "#define tex3D texture\n");
    } else {
        ADD(header, "#define tex1D texture1D\n");
        ADD(header, "#define tex3D texture3D\n");
        ADD(header, "#define texture texture2D\n");
    }

    if (sc->ra->glsl_vulkan && type == RA_RENDERPASS_TYPE_COMPUTE) {
        ADD(header, "#define gl_GlobalInvocationIndex "
                    "(gl_WorkGroupID * gl_WorkGroupSize + gl_LocalInvocationID)\n");
    }

    // Additional helpers.
    ADD(header, "#define LUT_POS(x, lut_size)"
                " mix(0.5 / (lut_size), 1.0 - 0.5 / (lut_size), (x))\n");

    char *vert_in = glsl_version >= 130 ? "in" : "attribute";
    char *vert_out = glsl_version >= 130 ? "out" : "varying";
    char *frag_in = glsl_version >= 130 ? "in" : "varying";

    struct bstr *vert = NULL, *frag = NULL, *comp = NULL;

    if (type == RA_RENDERPASS_TYPE_RASTER) {
        // vertex shader: we don't use the vertex shader, so just setup a
        // dummy, which passes through the vertex array attributes.
        bstr *vert_head = &sc->tmp[1];
        ADD_BSTR(vert_head, *header);
        bstr *vert_body = &sc->tmp[2];
        ADD(vert_body, "void main() {\n");
        bstr *frag_vaos = &sc->tmp[3];
        for (int n = 0; n < sc->params.num_vertex_attribs; n++) {
            const struct ra_renderpass_input *e = &sc->params.vertex_attribs[n];
            const char *glsl_type = vao_glsl_type(e);
            char loc[32] = {0};
            if (sc->ra->glsl_vulkan)
                snprintf(loc, sizeof(loc), "layout(location=%d) ", n);
            if (strcmp(e->name, "position") == 0) {
                // setting raster pos. requires setting gl_Position magic variable
                assert(e->dim_v == 2 && e->type == RA_VARTYPE_FLOAT);
                ADD(vert_head, "%s%s vec2 vertex_position;\n", loc, vert_in);
                ADD(vert_body, "gl_Position = vec4(vertex_position, 1.0, 1.0);\n");
            } else {
                ADD(vert_head, "%s%s %s vertex_%s;\n", loc, vert_in, glsl_type, e->name);
                ADD(vert_head, "%s%s %s %s;\n", loc, vert_out, glsl_type, e->name);
                ADD(vert_body, "%s = vertex_%s;\n", e->name, e->name);
                ADD(frag_vaos, "%s%s %s %s;\n", loc, frag_in, glsl_type, e->name);
            }
        }
        ADD(vert_body, "}\n");
        vert = vert_head;
        ADD_BSTR(vert, *vert_body);

        // fragment shader; still requires adding used uniforms and VAO elements
        frag = &sc->tmp[4];
        ADD_BSTR(frag, *header);
        if (glsl_version >= 130) {
            ADD(frag, "%sout vec4 out_color;\n",
                sc->ra->glsl_vulkan ? "layout(location=0) " : "");
        }
        ADD_BSTR(frag, *frag_vaos);
        add_uniforms(sc, frag);

        ADD_BSTR(frag, sc->prelude_text);
        ADD_BSTR(frag, sc->header_text);

        ADD(frag, "void main() {\n");
        // we require _all_ frag shaders to write to a "vec4 color"
        ADD(frag, "vec4 color = vec4(0.0, 0.0, 0.0, 1.0);\n");
        ADD_BSTR(frag, sc->text);
        if (glsl_version >= 130) {
            ADD(frag, "out_color = color;\n");
        } else {
            ADD(frag, "gl_FragColor = color;\n");
        }
        ADD(frag, "}\n");

        // We need to fix the format of the render dst at renderpass creation
        // time
        assert(target_format);
        sc->params.target_format = target_format;
    }

    if (type == RA_RENDERPASS_TYPE_COMPUTE) {
        comp = &sc->tmp[4];
        ADD_BSTR(comp, *header);

        add_uniforms(sc, comp);

        ADD_BSTR(comp, sc->prelude_text);
        ADD_BSTR(comp, sc->header_text);

        ADD(comp, "void main() {\n");
        ADD(comp, "vec4 color = vec4(0.0, 0.0, 0.0, 1.0);\n"); // convenience
        ADD_BSTR(comp, sc->text);
        ADD(comp, "}\n");
    }

    bstr *hash_total = &sc->tmp[5];

    ADD(hash_total, "type %d\n", sc->params.type);

    if (frag) {
        ADD_BSTR(hash_total, *frag);
        sc->params.frag_shader = frag->start;
    }
    ADD(hash_total, "\n");
    if (vert) {
        ADD_BSTR(hash_total, *vert);
        sc->params.vertex_shader = vert->start;
    }
    ADD(hash_total, "\n");
    if (comp) {
        ADD_BSTR(hash_total, *comp);
        sc->params.compute_shader = comp->start;
    }
    ADD(hash_total, "\n");

    if (sc->params.enable_blend) {
        ADD(hash_total, "blend %d %d %d %d\n",
            sc->params.blend_src_rgb, sc->params.blend_dst_rgb,
            sc->params.blend_src_alpha, sc->params.blend_dst_alpha);
    }

    if (sc->params.target_format)
        ADD(hash_total, "format %s\n", sc->params.target_format->name);

    struct sc_entry *entry = NULL;
    for (int n = 0; n < sc->num_entries; n++) {
        struct sc_entry *cur = sc->entries[n];
        if (bstr_equals(cur->total, *hash_total)) {
            entry = cur;
            break;
        }
    }
    if (!entry) {
        if (sc->num_entries == SC_MAX_ENTRIES)
            sc_flush_cache(sc);
        entry = talloc_ptrtype(NULL, entry);
        *entry = (struct sc_entry){
            .total = bstrdup(entry, *hash_total),
            .timer = timer_pool_create(sc->ra),
        };
        for (int n = 0; n < sc->num_uniforms; n++) {
            struct sc_cached_uniform u = {0};
            if (!sc->uniforms[n].ubo_length) {
                u.index = sc->params.num_inputs;
                MP_TARRAY_APPEND(sc, sc->params.inputs, sc->params.num_inputs,
                                 sc->uniforms[n].input);
            }
            MP_TARRAY_APPEND(entry, entry->cached_uniforms,
                             entry->num_cached_uniforms, u);
        }
        if (!create_pass(sc, entry))
            sc->error_state = true;
        MP_TARRAY_APPEND(sc, sc->entries, sc->num_entries, entry);
    }
    if (sc->error_state)
        return;

    assert(sc->num_uniforms == entry->num_cached_uniforms);

    sc->num_values = 0;
    for (int n = 0; n < sc->num_uniforms; n++)
        update_uniform(sc, entry, &sc->uniforms[n], n);

    // If we're using a UBO, make sure to bind it as well
    if (sc->ubo_size) {
        struct ra_renderpass_input_val ubo_val = {
            .index = entry->ubo_index,
            .data = &entry->ubo,
        };
        MP_TARRAY_APPEND(sc, sc->values, sc->num_values, ubo_val);
    }

    sc->current_shader = entry;
}

struct mp_pass_perf gl_sc_dispatch_draw(struct gl_shader_cache *sc,
                                        struct ra_tex *target,
                                        void *ptr, size_t num)
{
    struct timer_pool *timer = NULL;

    gl_sc_generate(sc, RA_RENDERPASS_TYPE_RASTER, target->params.format);
    if (!sc->current_shader)
        goto error;

    timer = sc->current_shader->timer;

    struct mp_rect full_rc = {0, 0, target->params.w, target->params.h};

    struct ra_renderpass_run_params run = {
        .pass = sc->current_shader->pass,
        .values = sc->values,
        .num_values = sc->num_values,
        .target = target,
        .vertex_data = ptr,
        .vertex_count = num,
        .viewport = full_rc,
        .scissors = full_rc,
    };

    timer_pool_start(timer);
    sc->ra->fns->renderpass_run(sc->ra, &run);
    timer_pool_stop(timer);

error:
    gl_sc_reset(sc);
    return timer_pool_measure(timer);
}

struct mp_pass_perf gl_sc_dispatch_compute(struct gl_shader_cache *sc,
                                           int w, int h, int d)
{
    struct timer_pool *timer = NULL;

    gl_sc_generate(sc, RA_RENDERPASS_TYPE_COMPUTE, NULL);
    if (!sc->current_shader)
        goto error;

    timer = sc->current_shader->timer;

    struct ra_renderpass_run_params run = {
        .pass = sc->current_shader->pass,
        .values = sc->values,
        .num_values = sc->num_values,
        .compute_groups = {w, h, d},
    };

    timer_pool_start(timer);
    sc->ra->fns->renderpass_run(sc->ra, &run);
    timer_pool_stop(timer);

error:
    gl_sc_reset(sc);
    return timer_pool_measure(timer);
}