src/gallium/drivers/nouveau/nv50/nv50_context.c

   1 /*
   2  * Copyright 2010 Christoph Bumiller
   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  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20  * OTHER DEALINGS IN THE SOFTWARE.
  21  */
  22
  23 #include "pipe/p_defines.h"
  24 #include "util/u_framebuffer.h"
  25 #include "util/u_upload_mgr.h"
  26
  27 #include "nv50/nv50_context.h"
  28 #include "nv50/nv50_screen.h"
  29 #include "nv50/nv50_resource.h"
  30
  31 static void
  32 nv50_flush(struct pipe_context *pipe,
  33            struct pipe_fence_handle **fence,
  34            unsigned flags)
  35 {
  36    struct nouveau_screen *screen = nouveau_screen(pipe->screen);
  37
  38    if (fence)
  39       nouveau_fence_ref(screen->fence.current, (struct nouveau_fence **)fence);
  40
  41    PUSH_KICK(screen->pushbuf);
  42
  43    nouveau_context_update_frame_stats(nouveau_context(pipe));
  44 }
  45
  46 static void
  47 nv50_texture_barrier(struct pipe_context *pipe, unsigned flags)
  48 {
  49    struct nouveau_pushbuf *push = nv50_context(pipe)->base.pushbuf;
  50
  51    BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
  52    PUSH_DATA (push, 0);
  53    BEGIN_NV04(push, NV50_3D(TEX_CACHE_CTL), 1);
  54    PUSH_DATA (push, 0x20);
  55 }
  56
  57 static void
  58 nv50_memory_barrier(struct pipe_context *pipe, unsigned flags)
  59 {
  60    struct nv50_context *nv50 = nv50_context(pipe);
  61    int i, s;
  62
  63    if (flags & PIPE_BARRIER_MAPPED_BUFFER) {
  64       for (i = 0; i < nv50->num_vtxbufs; ++i) {
  65          if (!nv50->vtxbuf[i].buffer)
  66             continue;
  67          if (nv50->vtxbuf[i].buffer->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
  68             nv50->base.vbo_dirty = true;
  69       }
  70
  71       if (nv50->idxbuf.buffer &&
  72           nv50->idxbuf.buffer->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
  73          nv50->base.vbo_dirty = true;
  74
  75       for (s = 0; s < 3 && !nv50->cb_dirty; ++s) {
  76          uint32_t valid = nv50->constbuf_valid[s];
  77
  78          while (valid && !nv50->cb_dirty) {
  79             const unsigned i = ffs(valid) - 1;
  80             struct pipe_resource *res;
  81
  82             valid &= ~(1 << i);
  83             if (nv50->constbuf[s][i].user)
  84                continue;
  85
  86             res = nv50->constbuf[s][i].u.buf;
  87             if (!res)
  88                continue;
  89
  90             if (res->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
  91                nv50->cb_dirty = true;
  92          }
  93       }
  94    }
  95 }
  96
  97 static void
  98 nv50_emit_string_marker(struct pipe_context *pipe, const char *str, int len)
  99 {
 100    struct nouveau_pushbuf *push = nv50_context(pipe)->base.pushbuf;
 101    int string_words = len / 4;
 102    int data_words;
 103
 104    if (len <= 0)
 105       return;
 106    string_words = MIN2(string_words, NV04_PFIFO_MAX_PACKET_LEN);
 107    if (string_words == NV04_PFIFO_MAX_PACKET_LEN)
 108       data_words = string_words;
 109    else
 110       data_words = string_words + !!(len & 3);
 111    BEGIN_NI04(push, SUBC_3D(NV04_GRAPH_NOP), data_words);
 112    if (string_words)
 113       PUSH_DATAp(push, str, string_words);
 114    if (string_words != data_words) {
 115       int data = 0;
 116       memcpy(&data, &str[string_words * 4], len & 3);
 117       PUSH_DATA (push, data);
 118    }
 119 }
 120
 121 void
 122 nv50_default_kick_notify(struct nouveau_pushbuf *push)
 123 {
 124    struct nv50_screen *screen = push->user_priv;
 125
 126    if (screen) {
 127       nouveau_fence_next(&screen->base);
 128       nouveau_fence_update(&screen->base, true);
 129       if (screen->cur_ctx)
 130          screen->cur_ctx->state.flushed = true;
 131    }
 132 }
 133
 134 static void
 135 nv50_context_unreference_resources(struct nv50_context *nv50)
 136 {
 137    unsigned s, i;
 138
 139    nouveau_bufctx_del(&nv50->bufctx_3d);
 140    nouveau_bufctx_del(&nv50->bufctx);
 141    nouveau_bufctx_del(&nv50->bufctx_cp);
 142
 143    util_unreference_framebuffer_state(&nv50->framebuffer);
 144
 145    assert(nv50->num_vtxbufs <= PIPE_MAX_ATTRIBS);
 146    for (i = 0; i < nv50->num_vtxbufs; ++i)
 147       pipe_resource_reference(&nv50->vtxbuf[i].buffer, NULL);
 148
 149    pipe_resource_reference(&nv50->idxbuf.buffer, NULL);
 150
 151    for (s = 0; s < 3; ++s) {
 152       assert(nv50->num_textures[s] <= PIPE_MAX_SAMPLERS);
 153       for (i = 0; i < nv50->num_textures[s]; ++i)
 154          pipe_sampler_view_reference(&nv50->textures[s][i], NULL);
 155
 156       for (i = 0; i < NV50_MAX_PIPE_CONSTBUFS; ++i)
 157          if (!nv50->constbuf[s][i].user)
 158             pipe_resource_reference(&nv50->constbuf[s][i].u.buf, NULL);
 159    }
 160
 161    for (i = 0; i < nv50->global_residents.size / sizeof(struct pipe_resource *);
 162         ++i) {
 163       struct pipe_resource **res = util_dynarray_element(
 164          &nv50->global_residents, struct pipe_resource *, i);
 165       pipe_resource_reference(res, NULL);
 166    }
 167    util_dynarray_fini(&nv50->global_residents);
 168 }
 169
 170 static void
 171 nv50_destroy(struct pipe_context *pipe)
 172 {
 173    struct nv50_context *nv50 = nv50_context(pipe);
 174
 175    if (nv50->screen->cur_ctx == nv50) {
 176       nv50->screen->cur_ctx = NULL;
 177       /* Save off the state in case another context gets created */
 178       nv50->screen->save_state = nv50->state;
 179    }
 180
 181    if (nv50->base.pipe.stream_uploader)
 182       u_upload_destroy(nv50->base.pipe.stream_uploader);
 183
 184    nouveau_pushbuf_bufctx(nv50->base.pushbuf, NULL);
 185    nouveau_pushbuf_kick(nv50->base.pushbuf, nv50->base.pushbuf->channel);
 186
 187    nv50_context_unreference_resources(nv50);
 188
 189    FREE(nv50->blit);
 190
 191    nouveau_context_destroy(&nv50->base);
 192 }
 193
 194 static int
 195 nv50_invalidate_resource_storage(struct nouveau_context *ctx,
 196                                  struct pipe_resource *res,
 197                                  int ref)
 198 {
 199    struct nv50_context *nv50 = nv50_context(&ctx->pipe);
 200    unsigned bind = res->bind ? res->bind : PIPE_BIND_VERTEX_BUFFER;
 201    unsigned s, i;
 202
 203    if (bind & PIPE_BIND_RENDER_TARGET) {
 204       assert(nv50->framebuffer.nr_cbufs <= PIPE_MAX_COLOR_BUFS);
 205       for (i = 0; i < nv50->framebuffer.nr_cbufs; ++i) {
 206          if (nv50->framebuffer.cbufs[i] &&
 207              nv50->framebuffer.cbufs[i]->texture == res) {
 208             nv50->dirty_3d |= NV50_NEW_3D_FRAMEBUFFER;
 209             nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_FB);
 210             if (!--ref)
 211                return ref;
 212          }
 213       }
 214    }
 215    if (bind & PIPE_BIND_DEPTH_STENCIL) {
 216       if (nv50->framebuffer.zsbuf &&
 217           nv50->framebuffer.zsbuf->texture == res) {
 218          nv50->dirty_3d |= NV50_NEW_3D_FRAMEBUFFER;
 219          nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_FB);
 220          if (!--ref)
 221             return ref;
 222       }
 223    }
 224
 225    if (bind & (PIPE_BIND_VERTEX_BUFFER |
 226                PIPE_BIND_INDEX_BUFFER |
 227                PIPE_BIND_CONSTANT_BUFFER |
 228                PIPE_BIND_STREAM_OUTPUT |
 229                PIPE_BIND_SAMPLER_VIEW)) {
 230
 231       assert(nv50->num_vtxbufs <= PIPE_MAX_ATTRIBS);
 232       for (i = 0; i < nv50->num_vtxbufs; ++i) {
 233          if (nv50->vtxbuf[i].buffer == res) {
 234             nv50->dirty_3d |= NV50_NEW_3D_ARRAYS;
 235             nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_VERTEX);
 236             if (!--ref)
 237                return ref;
 238          }
 239       }
 240
 241       if (nv50->idxbuf.buffer == res) {
 242          /* Just rebind to the bufctx as there is no separate dirty bit */
 243          nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_INDEX);
 244          BCTX_REFN(nv50->bufctx_3d, 3D_INDEX, nv04_resource(res), RD);
 245          if (!--ref)
 246             return ref;
 247       }
 248
 249       for (s = 0; s < 3; ++s) {
 250       assert(nv50->num_textures[s] <= PIPE_MAX_SAMPLERS);
 251       for (i = 0; i < nv50->num_textures[s]; ++i) {
 252          if (nv50->textures[s][i] &&
 253              nv50->textures[s][i]->texture == res) {
 254             nv50->dirty_3d |= NV50_NEW_3D_TEXTURES;
 255             nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_TEXTURES);
 256             if (!--ref)
 257                return ref;
 258          }
 259       }
 260       }
 261
 262       for (s = 0; s < 3; ++s) {
 263       for (i = 0; i < NV50_MAX_PIPE_CONSTBUFS; ++i) {
 264          if (!(nv50->constbuf_valid[s] & (1 << i)))
 265             continue;
 266          if (!nv50->constbuf[s][i].user &&
 267              nv50->constbuf[s][i].u.buf == res) {
 268             nv50->dirty_3d |= NV50_NEW_3D_CONSTBUF;
 269             nv50->constbuf_dirty[s] |= 1 << i;
 270             nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_3D_CB(s, i));
 271             if (!--ref)
 272                return ref;
 273          }
 274       }
 275       }
 276    }
 277
 278    return ref;
 279 }
 280
 281 static void
 282 nv50_context_get_sample_position(struct pipe_context *, unsigned, unsigned,
 283                                  float *);
 284
 285 struct pipe_context *
 286 nv50_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags)
 287 {
 288    struct nv50_screen *screen = nv50_screen(pscreen);
 289    struct nv50_context *nv50;
 290    struct pipe_context *pipe;
 291    int ret;
 292    uint32_t flags;
 293
 294    nv50 = CALLOC_STRUCT(nv50_context);
 295    if (!nv50)
 296       return NULL;
 297    pipe = &nv50->base.pipe;
 298
 299    if (!nv50_blitctx_create(nv50))
 300       goto out_err;
 301
 302    nv50->base.pushbuf = screen->base.pushbuf;
 303    nv50->base.client = screen->base.client;
 304
 305    ret = nouveau_bufctx_new(screen->base.client, 2, &nv50->bufctx);
 306    if (!ret)
 307       ret = nouveau_bufctx_new(screen->base.client, NV50_BIND_3D_COUNT,
 308                                &nv50->bufctx_3d);
 309    if (!ret)
 310       ret = nouveau_bufctx_new(screen->base.client, NV50_BIND_CP_COUNT,
 311                                &nv50->bufctx_cp);
 312    if (ret)
 313       goto out_err;
 314
 315    nv50->base.screen    = &screen->base;
 316    nv50->base.copy_data = nv50_m2mf_copy_linear;
 317    nv50->base.push_data = nv50_sifc_linear_u8;
 318    nv50->base.push_cb   = nv50_cb_push;
 319
 320    nv50->screen = screen;
 321    pipe->screen = pscreen;
 322    pipe->priv = priv;
 323    pipe->stream_uploader = u_upload_create_default(pipe);
 324    if (!pipe->stream_uploader)
 325       goto out_err;
 326    pipe->const_uploader = pipe->stream_uploader;
 327
 328    pipe->destroy = nv50_destroy;
 329
 330    pipe->draw_vbo = nv50_draw_vbo;
 331    pipe->clear = nv50_clear;
 332    pipe->launch_grid = nv50_launch_grid;
 333
 334    pipe->flush = nv50_flush;
 335    pipe->texture_barrier = nv50_texture_barrier;
 336    pipe->memory_barrier = nv50_memory_barrier;
 337    pipe->get_sample_position = nv50_context_get_sample_position;
 338    pipe->emit_string_marker = nv50_emit_string_marker;
 339
 340    if (!screen->cur_ctx) {
 341       /* Restore the last context's state here, normally handled during
 342        * context switch
 343        */
 344       nv50->state = screen->save_state;
 345       screen->cur_ctx = nv50;
 346       nouveau_pushbuf_bufctx(screen->base.pushbuf, nv50->bufctx);
 347    }
 348    nv50->base.pushbuf->kick_notify = nv50_default_kick_notify;
 349
 350    nouveau_context_init(&nv50->base);
 351    nv50_init_query_functions(nv50);
 352    nv50_init_surface_functions(nv50);
 353    nv50_init_state_functions(nv50);
 354    nv50_init_resource_functions(pipe);
 355
 356    nv50->base.invalidate_resource_storage = nv50_invalidate_resource_storage;
 357
 358    if (screen->base.device->chipset < 0x84 ||
 359        debug_get_bool_option("NOUVEAU_PMPEG", false)) {
 360       /* PMPEG */
 361       nouveau_context_init_vdec(&nv50->base);
 362    } else if (screen->base.device->chipset < 0x98 ||
 363               screen->base.device->chipset == 0xa0) {
 364       /* VP2 */
 365       pipe->create_video_codec = nv84_create_decoder;
 366       pipe->create_video_buffer = nv84_video_buffer_create;
 367    } else {
 368       /* VP3/4 */
 369       pipe->create_video_codec = nv98_create_decoder;
 370       pipe->create_video_buffer = nv98_video_buffer_create;
 371    }
 372
 373    flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD;
 374
 375    BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->code);
 376    BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->uniforms);
 377    BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->txc);
 378    BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->stack_bo);
 379    if (screen->compute) {
 380       BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->code);
 381       BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->txc);
 382       BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->stack_bo);
 383    }
 384
 385    flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
 386
 387    BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->fence.bo);
 388    BCTX_REFN_bo(nv50->bufctx, FENCE, flags, screen->fence.bo);
 389    if (screen->compute)
 390       BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
 391
 392    nv50->base.scratch.bo_size = 2 << 20;
 393
 394    util_dynarray_init(&nv50->global_residents);
 395
 396    return pipe;
 397
 398 out_err:
 399    if (pipe->stream_uploader)
 400       u_upload_destroy(pipe->stream_uploader);
 401    if (nv50->bufctx_3d)
 402       nouveau_bufctx_del(&nv50->bufctx_3d);
 403    if (nv50->bufctx_cp)
 404       nouveau_bufctx_del(&nv50->bufctx_cp);
 405    if (nv50->bufctx)
 406       nouveau_bufctx_del(&nv50->bufctx);
 407    FREE(nv50->blit);
 408    FREE(nv50);
 409    return NULL;
 410 }
 411
 412 void
 413 nv50_bufctx_fence(struct nouveau_bufctx *bufctx, bool on_flush)
 414 {
 415    struct nouveau_list *list = on_flush ? &bufctx->current : &bufctx->pending;
 416    struct nouveau_list *it;
 417
 418    for (it = list->next; it != list; it = it->next) {
 419       struct nouveau_bufref *ref = (struct nouveau_bufref *)it;
 420       struct nv04_resource *res = ref->priv;
 421       if (res)
 422          nv50_resource_validate(res, (unsigned)ref->priv_data);
 423    }
 424 }
 425
 426 static void
 427 nv50_context_get_sample_position(struct pipe_context *pipe,
 428                                  unsigned sample_count, unsigned sample_index,
 429                                  float *xy)
 430 {
 431    static const uint8_t ms1[1][2] = { { 0x8, 0x8 } };
 432    static const uint8_t ms2[2][2] = {
 433       { 0x4, 0x4 }, { 0xc, 0xc } }; /* surface coords (0,0), (1,0) */
 434    static const uint8_t ms4[4][2] = {
 435       { 0x6, 0x2 }, { 0xe, 0x6 },   /* (0,0), (1,0) */
 436       { 0x2, 0xa }, { 0xa, 0xe } }; /* (0,1), (1,1) */
 437    static const uint8_t ms8[8][2] = {
 438       { 0x1, 0x7 }, { 0x5, 0x3 },   /* (0,0), (1,0) */
 439       { 0x3, 0xd }, { 0x7, 0xb },   /* (0,1), (1,1) */
 440       { 0x9, 0x5 }, { 0xf, 0x1 },   /* (2,0), (3,0) */
 441       { 0xb, 0xf }, { 0xd, 0x9 } }; /* (2,1), (3,1) */
 442 #if 0
 443    /* NOTE: there are alternative modes for MS2 and MS8, currently not used */
 444    static const uint8_t ms8_alt[8][2] = {
 445       { 0x9, 0x5 }, { 0x7, 0xb },   /* (2,0), (1,1) */
 446       { 0xd, 0x9 }, { 0x5, 0x3 },   /* (3,1), (1,0) */
 447       { 0x3, 0xd }, { 0x1, 0x7 },   /* (0,1), (0,0) */
 448       { 0xb, 0xf }, { 0xf, 0x1 } }; /* (2,1), (3,0) */
 449 #endif
 450
 451    const uint8_t (*ptr)[2];
 452
 453    switch (sample_count) {
 454    case 0:
 455    case 1: ptr = ms1; break;
 456    case 2: ptr = ms2; break;
 457    case 4: ptr = ms4; break;
 458    case 8: ptr = ms8; break;
 459    default:
 460       assert(0);
 461       return; /* bad sample count -> undefined locations */
 462    }
 463    xy[0] = ptr[sample_index][0] * 0.0625f;
 464    xy[1] = ptr[sample_index][1] * 0.0625f;
 465 }