summaryrefslogtreecommitdiffstats
path: root/video/out/gpu/utils.c
blob: f8dcbaac600ef732a6b30186bcb640932c508bb4 (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
#include "common/msg.h"
#include "video/out/vo.h"
#include "utils.h"

// Standard parallel 2D projection, except y1 < y0 means that the coordinate
// system is flipped, not the projection.
void gl_transform_ortho(struct gl_transform *t, float x0, float x1,
                        float y0, float y1)
{
    if (y1 < y0) {
        float tmp = y0;
        y0 = tmp - y1;
        y1 = tmp;
    }

    t->m[0][0] = 2.0f / (x1 - x0);
    t->m[0][1] = 0.0f;
    t->m[1][0] = 0.0f;
    t->m[1][1] = 2.0f / (y1 - y0);
    t->t[0] = -(x1 + x0) / (x1 - x0);
    t->t[1] = -(y1 + y0) / (y1 - y0);
}

// Apply the effects of one transformation to another, transforming it in the
// process. In other words: post-composes t onto x
void gl_transform_trans(struct gl_transform t, struct gl_transform *x)
{
    struct gl_transform xt = *x;
    x->m[0][0] = t.m[0][0] * xt.m[0][0] + t.m[0][1] * xt.m[1][0];
    x->m[1][0] = t.m[1][0] * xt.m[0][0] + t.m[1][1] * xt.m[1][0];
    x->m[0][1] = t.m[0][0] * xt.m[0][1] + t.m[0][1] * xt.m[1][1];
    x->m[1][1] = t.m[1][0] * xt.m[0][1] + t.m[1][1] * xt.m[1][1];
    gl_transform_vec(t, &x->t[0], &x->t[1]);
}

void gl_transform_ortho_fbodst(struct gl_transform *t, struct fbodst fbo)
{
    int y_dir = fbo.flip ? -1 : 1;
    gl_transform_ortho(t, 0, fbo.tex->params.w, 0, fbo.tex->params.h * y_dir);
}

void ra_buf_pool_uninit(struct ra *ra, struct ra_buf_pool *pool)
{
    for (int i = 0; i < pool->num_buffers; i++)
        ra_buf_free(ra, &pool->buffers[i]);

    talloc_free(pool->buffers);
    *pool = (struct ra_buf_pool){0};
}

static bool ra_buf_params_compatible(const struct ra_buf_params *new,
                                     const struct ra_buf_params *old)
{
    return new->type == old->type &&
           new->size <= old->size &&
           new->host_mapped  == old->host_mapped &&
           new->host_mutable == old->host_mutable;
}

static bool ra_buf_pool_grow(struct ra *ra, struct ra_buf_pool *pool)
{
    struct ra_buf *buf = ra_buf_create(ra, &pool->current_params);
    if (!buf)
        return false;

    MP_TARRAY_INSERT_AT(NULL, pool->buffers, pool->num_buffers, pool->index, buf);
    MP_VERBOSE(ra, "Resized buffer pool of type %u to size %d\n",
               pool->current_params.type, pool->num_buffers);
    return true;
}

struct ra_buf *ra_buf_pool_get(struct ra *ra, struct ra_buf_pool *pool,
                               const struct ra_buf_params *params)
{
    assert(!params->initial_data);

    if (!ra_buf_params_compatible(params, &pool->current_params)) {
        ra_buf_pool_uninit(ra, pool);
        pool->current_params = *params;
    }

    // Make sure we have at least one buffer available
    if (!pool->buffers && !ra_buf_pool_grow(ra, pool))
        return NULL;

    // Make sure the next buffer is available for use
    if (!ra->fns->buf_poll(ra, pool->buffers[pool->index]) &&
        !ra_buf_pool_grow(ra, pool))
    {
        return NULL;
    }

    struct ra_buf *buf = pool->buffers[pool->index++];
    pool->index %= pool->num_buffers;

    return buf;
}

bool ra_tex_upload_pbo(struct ra *ra, struct ra_buf_pool *pbo,
                       const struct ra_tex_upload_params *params)
{
    if (params->buf)
        return ra->fns->tex_upload(ra, params);

    struct ra_tex *tex = params->tex;
    size_t row_size = tex->params.dimensions == 2 ? params->stride :
                      tex->params.w * tex->params.format->pixel_size;

    struct ra_buf_params bufparams = {
        .type = RA_BUF_TYPE_TEX_UPLOAD,
        .size = row_size * tex->params.h * tex->params.d,
        .host_mutable = true,
    };

    struct ra_buf *buf = ra_buf_pool_get(ra, pbo, &bufparams);
    if (!buf)
        return false;

    ra->fns->buf_update(ra, buf, 0, params->src, bufparams.size);

    struct ra_tex_upload_params newparams = *params;
    newparams.buf = buf;
    newparams.src = NULL;

    return ra->fns->tex_upload(ra, &newparams);
}

struct ra_layout std140_layout(struct ra_renderpass_input *inp)
{
    size_t el_size = ra_vartype_size(inp->type);

    // 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
    size_t size = el_size * inp->dim_v;
    if (inp->dim_v == 3)
        size += el_size;
    if (inp->dim_m > 1)
        size = MP_ALIGN_UP(size, sizeof(float[4]));

    return (struct ra_layout) {
        .align  = size,
        .stride = size,
        .size   = size * inp->dim_m,
    };
}

struct ra_layout std430_layout(struct ra_renderpass_input *inp)
{
    size_t el_size = ra_vartype_size(inp->type);

    // std430 packing rules: like std140, except arrays/matrices are always
    // "tightly" packed, even arrays/matrices of vec3s
    size_t size = el_size * inp->dim_v;
    if (inp->dim_v == 3 && inp->dim_m == 1)
        size += el_size;

    return (struct ra_layout) {
        .align  = size,
        .stride = size,
        .size   = size * inp->dim_m,
    };
}

// Create a texture and a FBO using the texture as color attachments.
//  fmt: texture internal format
// If the parameters are the same as the previous call, do not touch it.
// flags can be 0, or a combination of FBOTEX_FUZZY_W and FBOTEX_FUZZY_H.
// Enabling FUZZY for W or H means the w or h does not need to be exact.
bool fbotex_change(struct fbotex *fbo, struct ra *ra, struct mp_log *log,
                   int w, int h, const struct ra_format *fmt, int flags)
{
    int lw = w, lh = h;

    if (fbo->tex) {
        int cw = w, ch = h;
        int rw = fbo->tex->params.w, rh = fbo->tex->params.h;

        if ((flags & FBOTEX_FUZZY_W) && cw < rw)
            cw = rw;
        if ((flags & FBOTEX_FUZZY_H) && ch < rh)
            ch = rh;

        if (rw == cw && rh == ch && fbo->tex->params.format == fmt)
            goto done;
    }

    if (flags & FBOTEX_FUZZY_W)
        w = MP_ALIGN_UP(w, 256);
    if (flags & FBOTEX_FUZZY_H)
        h = MP_ALIGN_UP(h, 256);

    mp_verbose(log, "Create FBO: %dx%d (%dx%d)\n", lw, lh, w, h);

    if (!fmt || !fmt->renderable || !fmt->linear_filter) {
        mp_err(log, "Format %s not supported.\n", fmt ? fmt->name : "(unset)");
        return false;
    }

    fbotex_uninit(fbo);

    *fbo = (struct fbotex) {
        .ra = ra,
    };

    struct ra_tex_params params = {
        .dimensions = 2,
        .w = w,
        .h = h,
        .d = 1,
        .format = fmt,
        .src_linear = true,
        .render_src = true,
        .render_dst = true,
        .storage_dst = true,
        .blit_src = true,
    };

    fbo->tex = ra_tex_create(fbo->ra, &params);

    if (!fbo->tex) {
        mp_err(log, "Error: framebuffer could not be created.\n");
        fbotex_uninit(fbo);
        return false;
    }

done:

    fbo->lw = lw;
    fbo->lh = lh;

    fbo->fbo = (struct fbodst){
        .tex = fbo->tex,
    };

    return true;
}

void fbotex_uninit(struct fbotex *fbo)
{
    if (fbo->ra) {
        ra_tex_free(fbo->ra, &fbo->tex);
        *fbo = (struct fbotex) {0};
    }
}

struct timer_pool {
    struct ra *ra;
    ra_timer *timer;
    bool running; // detect invalid usage

    uint64_t samples[VO_PERF_SAMPLE_COUNT];
    int sample_idx;
    int sample_count;

    uint64_t sum;
    uint64_t peak;
};

struct timer_pool *timer_pool_create(struct ra *ra)
{
    if (!ra->fns->timer_create)
        return NULL;

    ra_timer *timer = ra->fns->timer_create(ra);
    if (!timer)
        return NULL;

    struct timer_pool *pool = talloc(NULL, struct timer_pool);
    if (!pool) {
        ra->fns->timer_destroy(ra, timer);
        return NULL;
    }

    *pool = (struct timer_pool){ .ra = ra, .timer = timer };
    return pool;
}

void timer_pool_destroy(struct timer_pool *pool)
{
    if (!pool)
        return;

    pool->ra->fns->timer_destroy(pool->ra, pool->timer);
    talloc_free(pool);
}

void timer_pool_start(struct timer_pool *pool)
{
    if (!pool)
        return;

    assert(!pool->running);
    pool->ra->fns->timer_start(pool->ra, pool->timer);
    pool->running = true;
}

void timer_pool_stop(struct timer_pool *pool)
{
    if (!pool)
        return;

    assert(pool->running);
    uint64_t res = pool->ra->fns->timer_stop(pool->ra, pool->timer);
    pool->running = false;

    if (res) {
        // Input res into the buffer and grab the previous value
        uint64_t old = pool->samples[pool->sample_idx];
        pool->sample_count = MPMIN(pool->sample_count + 1, VO_PERF_SAMPLE_COUNT);
        pool->samples[pool->sample_idx++] = res;
        pool->sample_idx %= VO_PERF_SAMPLE_COUNT;
        pool->sum = pool->sum + res - old;

        // Update peak if necessary
        if (res >= pool->peak) {
            pool->peak = res;
        } else if (pool->peak == old) {
            // It's possible that the last peak was the value we just removed,
            // if so we need to scan for the new peak
            uint64_t peak = res;
            for (int i = 0; i < VO_PERF_SAMPLE_COUNT; i++)
                peak = MPMAX(peak, pool->samples[i]);
            pool->peak = peak;
        }
    }
}

struct mp_pass_perf timer_pool_measure(struct timer_pool *pool)
{
    if (!pool)
        return (struct mp_pass_perf){0};

    struct mp_pass_perf res = {
        .peak = pool->peak,
        .count = pool->sample_count,
    };

    int idx = pool->sample_idx - pool->sample_count + VO_PERF_SAMPLE_COUNT;
    for (int i = 0; i < res.count; i++) {
        idx %= VO_PERF_SAMPLE_COUNT;
        res.samples[i] = pool->samples[idx++];
    }

    if (res.count > 0) {
        res.last = res.samples[res.count - 1];
        res.avg = pool->sum / res.count;
    }

    return res;
}

void mp_log_source(struct mp_log *log, int lev, const char *src)
{
    int line = 1;
    if (!src)
        return;
    while (*src) {
        const char *end = strchr(src, '\n');
        const char *next = end + 1;
        if (!end)
            next = end = src + strlen(src);
        mp_msg(log, lev, "[%3d] %.*s\n", line, (int)(end - src), src);
        line++;
        src = next;
    }
}