src/gallium/drivers/nouveau/nv30/nv30_draw.c

   1 /*
   2  * Copyright 2012 Red Hat Inc.
   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  * Authors: Ben Skeggs
  23  *
  24  */
  25
  26 #include "draw/draw_context.h"
  27 #include "draw/draw_vertex.h"
  28 #include "draw/draw_pipe.h"
  29 #include "draw/draw_vbuf.h"
  30 #include "draw/draw_private.h"
  31
  32 #include "nv_object.xml.h"
  33 #include "nv30/nv30-40_3d.xml.h"
  34 #include "nv30/nv30_context.h"
  35 #include "nv30/nv30_format.h"
  36
  37 struct nv30_render {
  38    struct vbuf_render base;
  39    struct nv30_context *nv30;
  40
  41    struct pipe_transfer *transfer;
  42    struct pipe_resource *buffer;
  43    unsigned offset;
  44    unsigned length;
  45
  46    struct vertex_info vertex_info;
  47
  48    struct nouveau_heap *vertprog;
  49    uint32_t vtxprog[16][4];
  50    uint32_t vtxfmt[16];
  51    uint32_t vtxptr[16];
  52    uint32_t prim;
  53 };
  54
  55 static inline struct nv30_render *
  56 nv30_render(struct vbuf_render *render)
  57 {
  58    return (struct nv30_render *)render;
  59 }
  60
  61 static const struct vertex_info *
  62 nv30_render_get_vertex_info(struct vbuf_render *render)
  63 {
  64    return &nv30_render(render)->vertex_info;
  65 }
  66
  67 static boolean
  68 nv30_render_allocate_vertices(struct vbuf_render *render,
  69                               ushort vertex_size, ushort nr_vertices)
  70 {
  71    struct nv30_render *r = nv30_render(render);
  72    struct nv30_context *nv30 = r->nv30;
  73
  74    r->length = (uint32_t)vertex_size * (uint32_t)nr_vertices;
  75
  76    if (r->offset + r->length >= render->max_vertex_buffer_bytes) {
  77       pipe_resource_reference(&r->buffer, NULL);
  78       r->buffer = pipe_buffer_create(&nv30->screen->base.base,
  79                                      PIPE_BIND_VERTEX_BUFFER, PIPE_USAGE_STREAM,
  80                                      render->max_vertex_buffer_bytes);
  81       if (!r->buffer)
  82          return false;
  83
  84       r->offset = 0;
  85    }
  86
  87    return true;
  88 }
  89
  90 static void *
  91 nv30_render_map_vertices(struct vbuf_render *render)
  92 {
  93    struct nv30_render *r = nv30_render(render);
  94    char *map = pipe_buffer_map_range(
  95          &r->nv30->base.pipe, r->buffer,
  96          r->offset, r->length,
  97          PIPE_TRANSFER_WRITE |
  98          PIPE_TRANSFER_DISCARD_RANGE,
  99          &r->transfer);
 100    assert(map);
 101    return map;
 102 }
 103
 104 static void
 105 nv30_render_unmap_vertices(struct vbuf_render *render,
 106                            ushort min_index, ushort max_index)
 107 {
 108    struct nv30_render *r = nv30_render(render);
 109    pipe_buffer_unmap(&r->nv30->base.pipe, r->transfer);
 110    r->transfer = NULL;
 111 }
 112
 113 static void
 114 nv30_render_set_primitive(struct vbuf_render *render, unsigned prim)
 115 {
 116    struct nv30_render *r = nv30_render(render);
 117
 118    r->prim = nv30_prim_gl(prim);
 119 }
 120
 121 static void
 122 nv30_render_draw_elements(struct vbuf_render *render,
 123                           const ushort *indices, uint count)
 124 {
 125    struct nv30_render *r = nv30_render(render);
 126    struct nv30_context *nv30 = r->nv30;
 127    struct nouveau_pushbuf *push = nv30->screen->base.pushbuf;
 128    unsigned i;
 129
 130    BEGIN_NV04(push, NV30_3D(VTXBUF(0)), r->vertex_info.num_attribs);
 131    for (i = 0; i < r->vertex_info.num_attribs; i++) {
 132       PUSH_RESRC(push, NV30_3D(VTXBUF(i)), BUFCTX_VTXTMP,
 133                        nv04_resource(r->buffer), r->offset + r->vtxptr[i],
 134                        NOUVEAU_BO_LOW | NOUVEAU_BO_RD, 0, NV30_3D_VTXBUF_DMA1);
 135    }
 136
 137    if (!nv30_state_validate(nv30, ~0, false))
 138       return;
 139
 140    BEGIN_NV04(push, NV30_3D(VERTEX_BEGIN_END), 1);
 141    PUSH_DATA (push, r->prim);
 142
 143    if (count & 1) {
 144       BEGIN_NV04(push, NV30_3D(VB_ELEMENT_U32), 1);
 145       PUSH_DATA (push, *indices++);
 146    }
 147
 148    count >>= 1;
 149    while (count) {
 150       unsigned npush = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN);
 151       count -= npush;
 152
 153       BEGIN_NI04(push, NV30_3D(VB_ELEMENT_U16), npush);
 154       while (npush--) {
 155          PUSH_DATA(push, (indices[1] << 16) | indices[0]);
 156          indices += 2;
 157       }
 158    }
 159
 160    BEGIN_NV04(push, NV30_3D(VERTEX_BEGIN_END), 1);
 161    PUSH_DATA (push, NV30_3D_VERTEX_BEGIN_END_STOP);
 162    PUSH_RESET(push, BUFCTX_VTXTMP);
 163 }
 164
 165 static void
 166 nv30_render_draw_arrays(struct vbuf_render *render, unsigned start, uint nr)
 167 {
 168    struct nv30_render *r = nv30_render(render);
 169    struct nv30_context *nv30 = r->nv30;
 170    struct nouveau_pushbuf *push = nv30->base.pushbuf;
 171    unsigned fn = nr >> 8, pn = nr & 0xff;
 172    unsigned ps = fn + (pn ? 1 : 0);
 173    unsigned i;
 174
 175    BEGIN_NV04(push, NV30_3D(VTXBUF(0)), r->vertex_info.num_attribs);
 176    for (i = 0; i < r->vertex_info.num_attribs; i++) {
 177       PUSH_RESRC(push, NV30_3D(VTXBUF(i)), BUFCTX_VTXTMP,
 178                        nv04_resource(r->buffer), r->offset + r->vtxptr[i],
 179                        NOUVEAU_BO_LOW | NOUVEAU_BO_RD, 0, NV30_3D_VTXBUF_DMA1);
 180    }
 181
 182    if (!nv30_state_validate(nv30, ~0, false))
 183       return;
 184
 185    BEGIN_NV04(push, NV30_3D(VERTEX_BEGIN_END), 1);
 186    PUSH_DATA (push, r->prim);
 187
 188    BEGIN_NI04(push, NV30_3D(VB_VERTEX_BATCH), ps);
 189    while (fn--) {
 190       PUSH_DATA (push, 0xff000000 | start);
 191       start += 256;
 192    }
 193
 194    if (pn)
 195       PUSH_DATA (push, ((pn - 1) << 24) | start);
 196
 197    BEGIN_NV04(push, NV30_3D(VERTEX_BEGIN_END), 1);
 198    PUSH_DATA (push, NV30_3D_VERTEX_BEGIN_END_STOP);
 199    PUSH_RESET(push, BUFCTX_VTXTMP);
 200 }
 201
 202 static void
 203 nv30_render_release_vertices(struct vbuf_render *render)
 204 {
 205    struct nv30_render *r = nv30_render(render);
 206    r->offset += r->length;
 207 }
 208
 209 static const struct {
 210    unsigned emit;
 211    unsigned vp30;
 212    unsigned vp40;
 213    unsigned ow40;
 214 } vroute [] = {
 215    [TGSI_SEMANTIC_POSITION] = { EMIT_4F, 0, 0, 0x00000000 },
 216    [TGSI_SEMANTIC_COLOR   ] = { EMIT_4F, 3, 1, 0x00000001 },
 217    [TGSI_SEMANTIC_BCOLOR  ] = { EMIT_4F, 1, 3, 0x00000004 },
 218    [TGSI_SEMANTIC_FOG     ] = { EMIT_4F, 5, 5, 0x00000010 },
 219    [TGSI_SEMANTIC_PSIZE   ] = { EMIT_1F_PSIZE, 6, 6, 0x00000020 },
 220    [TGSI_SEMANTIC_TEXCOORD] = { EMIT_4F, 8, 7, 0x00004000 },
 221 };
 222
 223 static bool
 224 vroute_add(struct nv30_render *r, uint attrib, uint sem, uint *idx)
 225 {
 226    struct nv30_screen *screen = r->nv30->screen;
 227    struct nv30_fragprog *fp = r->nv30->fragprog.program;
 228    struct vertex_info *vinfo = &r->vertex_info;
 229    enum pipe_format format;
 230    uint emit = EMIT_OMIT;
 231    uint result = *idx;
 232
 233    if (sem == TGSI_SEMANTIC_GENERIC) {
 234       uint num_texcoords = (screen->eng3d->oclass < NV40_3D_CLASS) ? 8 : 10;
 235       for (result = 0; result < num_texcoords; result++) {
 236          if (fp->texcoord[result] == *idx + 8) {
 237             sem = TGSI_SEMANTIC_TEXCOORD;
 238             emit = vroute[sem].emit;
 239             break;
 240          }
 241       }
 242    } else {
 243       emit = vroute[sem].emit;
 244    }
 245
 246    if (emit == EMIT_OMIT)
 247       return false;
 248
 249    draw_emit_vertex_attr(vinfo, emit, attrib);
 250    format = draw_translate_vinfo_format(emit);
 251
 252    r->vtxfmt[attrib] = nv30_vtxfmt(&screen->base.base, format)->hw;
 253    r->vtxptr[attrib] = vinfo->size;
 254    vinfo->size += draw_translate_vinfo_size(emit);
 255
 256    if (screen->eng3d->oclass < NV40_3D_CLASS) {
 257       r->vtxprog[attrib][0] = 0x001f38d8;
 258       r->vtxprog[attrib][1] = 0x0080001b | (attrib << 9);
 259       r->vtxprog[attrib][2] = 0x0836106c;
 260       r->vtxprog[attrib][3] = 0x2000f800 | (result + vroute[sem].vp30) << 2;
 261    } else {
 262       r->vtxprog[attrib][0] = 0x401f9c6c;
 263       r->vtxprog[attrib][1] = 0x0040000d | (attrib << 8);
 264       r->vtxprog[attrib][2] = 0x8106c083;
 265       r->vtxprog[attrib][3] = 0x6041ff80 | (result + vroute[sem].vp40) << 2;
 266    }
 267
 268    if (result < 8)
 269       *idx = vroute[sem].ow40 << result;
 270    else {
 271       assert(sem == TGSI_SEMANTIC_TEXCOORD);
 272       *idx = 0x00001000 << (result - 8);
 273    }
 274    return true;
 275 }
 276
 277 static bool
 278 nv30_render_validate(struct nv30_context *nv30)
 279 {
 280    struct nv30_render *r = nv30_render(nv30->draw->render);
 281    struct nv30_rasterizer_stateobj *rast = nv30->rast;
 282    struct pipe_screen *pscreen = &nv30->screen->base.base;
 283    struct nouveau_pushbuf *push = nv30->screen->base.pushbuf;
 284    struct nouveau_object *eng3d = nv30->screen->eng3d;
 285    struct nv30_vertprog *vp = nv30->vertprog.program;
 286    struct vertex_info *vinfo = &r->vertex_info;
 287    unsigned vp_attribs = 0;
 288    unsigned vp_results = 0;
 289    unsigned attrib = 0;
 290    unsigned pntc;
 291    int i;
 292
 293    if (!r->vertprog) {
 294       struct nouveau_heap *heap = nv30_screen(pscreen)->vp_exec_heap;
 295       if (nouveau_heap_alloc(heap, 16, &r->vertprog, &r->vertprog)) {
 296          while (heap->next && heap->size < 16) {
 297             struct nouveau_heap **evict = heap->next->priv;
 298             nouveau_heap_free(evict);
 299          }
 300
 301          if (nouveau_heap_alloc(heap, 16, &r->vertprog, &r->vertprog))
 302             return false;
 303       }
 304    }
 305
 306    vinfo->num_attribs = 0;
 307    vinfo->size = 0;
 308
 309    /* setup routing for all necessary vp outputs */
 310    for (i = 0; i < vp->info.num_outputs && attrib < 16; i++) {
 311       uint semantic = vp->info.output_semantic_name[i];
 312       uint index = vp->info.output_semantic_index[i];
 313       if (vroute_add(r, attrib, semantic, &index)) {
 314          vp_attribs |= (1 << attrib++);
 315          vp_results |= index;
 316       }
 317    }
 318
 319    /* setup routing for replaced point coords not written by vp */
 320    if (rast && rast->pipe.point_quad_rasterization)
 321       pntc = rast->pipe.sprite_coord_enable & 0x000002ff;
 322    else
 323       pntc = 0;
 324
 325    while (pntc && attrib < 16) {
 326       uint index = ffs(pntc) - 1; pntc &= ~(1 << index);
 327       if (vroute_add(r, attrib, TGSI_SEMANTIC_TEXCOORD, &index)) {
 328          vp_attribs |= (1 << attrib++);
 329          vp_results |= index;
 330       }
 331    }
 332
 333    /* modify vertex format for correct stride, and stub out unused ones */
 334    BEGIN_NV04(push, NV30_3D(VP_UPLOAD_FROM_ID), 1);
 335    PUSH_DATA (push, r->vertprog->start);
 336    r->vtxprog[attrib - 1][3] |= 1;
 337    for (i = 0; i < attrib; i++) {
 338       BEGIN_NV04(push, NV30_3D(VP_UPLOAD_INST(0)), 4);
 339       PUSH_DATAp(push, r->vtxprog[i], 4);
 340       r->vtxfmt[i] |= vinfo->size << 8;
 341    }
 342    for (; i < 16; i++)
 343       r->vtxfmt[i]  = NV30_3D_VTXFMT_TYPE_V32_FLOAT;
 344
 345    BEGIN_NV04(push, NV30_3D(VIEWPORT_TRANSLATE_X), 8);
 346    PUSH_DATAf(push, 0.0);
 347    PUSH_DATAf(push, 0.0);
 348    PUSH_DATAf(push, 0.0);
 349    PUSH_DATAf(push, 0.0);
 350    PUSH_DATAf(push, 1.0);
 351    PUSH_DATAf(push, 1.0);
 352    PUSH_DATAf(push, 1.0);
 353    PUSH_DATAf(push, 1.0);
 354    BEGIN_NV04(push, NV30_3D(DEPTH_RANGE_NEAR), 2);
 355    PUSH_DATAf(push, 0.0);
 356    PUSH_DATAf(push, 1.0);
 357
 358    BEGIN_NV04(push, NV30_3D(VTXFMT(0)), 16);
 359    PUSH_DATAp(push, r->vtxfmt, 16);
 360
 361    BEGIN_NV04(push, NV30_3D(VP_START_FROM_ID), 1);
 362    PUSH_DATA (push, r->vertprog->start);
 363    BEGIN_NV04(push, NV30_3D(ENGINE), 1);
 364    PUSH_DATA (push, 0x00000103);
 365    if (eng3d->oclass >= NV40_3D_CLASS) {
 366       BEGIN_NV04(push, NV40_3D(VP_ATTRIB_EN), 2);
 367       PUSH_DATA (push, vp_attribs);
 368       PUSH_DATA (push, vp_results);
 369    }
 370
 371    vinfo->size /= 4;
 372    return true;
 373 }
 374
 375 void
 376 nv30_render_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 377 {
 378    struct nv30_context *nv30 = nv30_context(pipe);
 379    struct draw_context *draw = nv30->draw;
 380    struct pipe_transfer *transfer[PIPE_MAX_ATTRIBS] = {NULL};
 381    struct pipe_transfer *transferi = NULL;
 382    int i;
 383
 384    nv30_render_validate(nv30);
 385
 386    if (nv30->draw_dirty & NV30_NEW_VIEWPORT)
 387       draw_set_viewport_states(draw, 0, 1, &nv30->viewport);
 388    if (nv30->draw_dirty & NV30_NEW_RASTERIZER)
 389       draw_set_rasterizer_state(draw, &nv30->rast->pipe, NULL);
 390    if (nv30->draw_dirty & NV30_NEW_CLIP)
 391       draw_set_clip_state(draw, &nv30->clip);
 392    if (nv30->draw_dirty & NV30_NEW_ARRAYS) {
 393       draw_set_vertex_buffers(draw, 0, nv30->num_vtxbufs, nv30->vtxbuf);
 394       draw_set_vertex_elements(draw, nv30->vertex->num_elements, nv30->vertex->pipe);
 395    }
 396    if (nv30->draw_dirty & NV30_NEW_FRAGPROG) {
 397       struct nv30_fragprog *fp = nv30->fragprog.program;
 398       if (!fp->draw)
 399          fp->draw = draw_create_fragment_shader(draw, &fp->pipe);
 400       draw_bind_fragment_shader(draw, fp->draw);
 401    }
 402    if (nv30->draw_dirty & NV30_NEW_VERTPROG) {
 403       struct nv30_vertprog *vp = nv30->vertprog.program;
 404       if (!vp->draw)
 405          vp->draw = draw_create_vertex_shader(draw, &vp->pipe);
 406       draw_bind_vertex_shader(draw, vp->draw);
 407    }
 408    if (nv30->draw_dirty & NV30_NEW_VERTCONST) {
 409       if (nv30->vertprog.constbuf) {
 410          void *map = nv04_resource(nv30->vertprog.constbuf)->data;
 411          draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX, 0,
 412                                          map, nv30->vertprog.constbuf_nr * 16);
 413       } else {
 414          draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX, 0, NULL, 0);
 415       }
 416    }
 417
 418    for (i = 0; i < nv30->num_vtxbufs; i++) {
 419       const void *map = nv30->vtxbuf[i].user_buffer;
 420       if (!map) {
 421          if (nv30->vtxbuf[i].buffer)
 422             map = pipe_buffer_map(pipe, nv30->vtxbuf[i].buffer,
 423                                   PIPE_TRANSFER_UNSYNCHRONIZED |
 424                                   PIPE_TRANSFER_READ, &transfer[i]);
 425       }
 426       draw_set_mapped_vertex_buffer(draw, i, map, ~0);
 427    }
 428
 429    if (info->indexed) {
 430       const void *map = nv30->idxbuf.user_buffer;
 431       if (!map)
 432          map = pipe_buffer_map(pipe, nv30->idxbuf.buffer,
 433                                PIPE_TRANSFER_UNSYNCHRONIZED |
 434                                PIPE_TRANSFER_READ, &transferi);
 435       draw_set_indexes(draw,
 436                        (ubyte *) map + nv30->idxbuf.offset,
 437                        nv30->idxbuf.index_size, ~0);
 438    } else {
 439       draw_set_indexes(draw, NULL, 0, 0);
 440    }
 441
 442    draw_vbo(draw, info);
 443    draw_flush(draw);
 444
 445    if (info->indexed && transferi)
 446       pipe_buffer_unmap(pipe, transferi);
 447    for (i = 0; i < nv30->num_vtxbufs; i++)
 448       if (transfer[i])
 449          pipe_buffer_unmap(pipe, transfer[i]);
 450
 451    nv30->draw_dirty = 0;
 452    nv30_state_release(nv30);
 453 }
 454
 455 static void
 456 nv30_render_destroy(struct vbuf_render *render)
 457 {
 458    struct nv30_render *r = nv30_render(render);
 459
 460    if (r->transfer)
 461       pipe_buffer_unmap(&r->nv30->base.pipe, r->transfer);
 462    pipe_resource_reference(&r->buffer, NULL);
 463    nouveau_heap_free(&r->vertprog);
 464    FREE(render);
 465 }
 466
 467 static struct vbuf_render *
 468 nv30_render_create(struct nv30_context *nv30)
 469 {
 470    struct nv30_render *r = CALLOC_STRUCT(nv30_render);
 471    if (!r)
 472       return NULL;
 473
 474    r->nv30 = nv30;
 475    r->offset = 1 * 1024 * 1024;
 476
 477    r->base.max_indices = 16 * 1024;
 478    r->base.max_vertex_buffer_bytes = r->offset;
 479
 480    r->base.get_vertex_info = nv30_render_get_vertex_info;
 481    r->base.allocate_vertices = nv30_render_allocate_vertices;
 482    r->base.map_vertices = nv30_render_map_vertices;
 483    r->base.unmap_vertices = nv30_render_unmap_vertices;
 484    r->base.set_primitive = nv30_render_set_primitive;
 485    r->base.draw_elements = nv30_render_draw_elements;
 486    r->base.draw_arrays = nv30_render_draw_arrays;
 487    r->base.release_vertices = nv30_render_release_vertices;
 488    r->base.destroy = nv30_render_destroy;
 489    return &r->base;
 490 }
 491
 492 void
 493 nv30_draw_init(struct pipe_context *pipe)
 494 {
 495    struct nv30_context *nv30 = nv30_context(pipe);
 496    struct vbuf_render *render;
 497    struct draw_context *draw;
 498    struct draw_stage *stage;
 499
 500    draw = draw_create(pipe);
 501    if (!draw)
 502       return;
 503
 504    render = nv30_render_create(nv30);
 505    if (!render) {
 506       draw_destroy(draw);
 507       return;
 508    }
 509
 510    stage = draw_vbuf_stage(draw, render);
 511    if (!stage) {
 512       render->destroy(render);
 513       draw_destroy(draw);
 514       return;
 515    }
 516
 517    draw_set_render(draw, render);
 518    draw_set_rasterize_stage(draw, stage);
 519    draw_wide_line_threshold(draw, 10000000.f);
 520    draw_wide_point_threshold(draw, 10000000.f);
 521    draw_wide_point_sprites(draw, true);
 522    nv30->draw = draw;
 523 }