src/gallium/drivers/nv50/nv50_state_validate.c

   1 /*
   2  * Copyright 2008 Ben Skeggs
   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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
  18  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
  19  * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  20  * SOFTWARE.
  21  */
  22
  23 #include "nv50_context.h"
  24 #include "nouveau/nouveau_stateobj.h"
  25
  26 static void
  27 nv50_state_validate_fb(struct nv50_context *nv50)
  28 {
  29         struct nouveau_grobj *tesla = nv50->screen->tesla;
  30         struct nouveau_stateobj *so = so_new(128, 18);
  31         struct pipe_framebuffer_state *fb = &nv50->framebuffer;
  32         unsigned i, w, h, gw = 0;
  33
  34         for (i = 0; i < fb->nr_cbufs; i++) {
  35                 struct pipe_texture *pt = fb->cbufs[i]->texture;
  36                 struct nouveau_bo *bo = nv50_miptree(pt)->base.bo;
  37
  38                 if (!gw) {
  39                         w = fb->cbufs[i]->width;
  40                         h = fb->cbufs[i]->height;
  41                         gw = 1;
  42                 } else {
  43                         assert(w == fb->cbufs[i]->width);
  44                         assert(h == fb->cbufs[i]->height);
  45                 }
  46
  47                 so_method(so, tesla, NV50TCL_RT_HORIZ(i), 2);
  48                 so_data  (so, fb->cbufs[i]->width);
  49                 so_data  (so, fb->cbufs[i]->height);
  50
  51                 so_method(so, tesla, NV50TCL_RT_ADDRESS_HIGH(i), 5);
  52                 so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM |
  53                               NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0);
  54                 so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM |
  55                               NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0);
  56                 switch (fb->cbufs[i]->format) {
  57                 case PIPE_FORMAT_A8R8G8B8_UNORM:
  58                         so_data(so, NV50TCL_RT_FORMAT_A8R8G8B8_UNORM);
  59                         break;
  60                 case PIPE_FORMAT_X8R8G8B8_UNORM:
  61                         so_data(so, NV50TCL_RT_FORMAT_X8R8G8B8_UNORM);
  62                         break;
  63                 case PIPE_FORMAT_R5G6B5_UNORM:
  64                         so_data(so, NV50TCL_RT_FORMAT_R5G6B5_UNORM);
  65                         break;
  66                 default:
  67                         NOUVEAU_ERR("AIIII unknown format %s\n",
  68                                     pf_name(fb->cbufs[i]->format));
  69                         so_data(so, NV50TCL_RT_FORMAT_X8R8G8B8_UNORM);
  70                         break;
  71                 }
  72                 so_data(so, nv50_miptree(pt)->
  73                                 level[fb->cbufs[i]->level].tile_mode << 4);
  74                 so_data(so, 0x00000000);
  75
  76                 so_method(so, tesla, 0x1224, 1);
  77                 so_data  (so, 1);
  78         }
  79
  80         if (fb->zsbuf) {
  81                 struct pipe_texture *pt = fb->zsbuf->texture;
  82                 struct nouveau_bo *bo = nv50_miptree(pt)->base.bo;
  83
  84                 if (!gw) {
  85                         w = fb->zsbuf->width;
  86                         h = fb->zsbuf->height;
  87                         gw = 1;
  88                 } else {
  89                         assert(w == fb->zsbuf->width);
  90                         assert(h == fb->zsbuf->height);
  91                 }
  92
  93                 so_method(so, tesla, NV50TCL_ZETA_ADDRESS_HIGH, 5);
  94                 so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM |
  95                               NOUVEAU_BO_HIGH | NOUVEAU_BO_RDWR, 0, 0);
  96                 so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM |
  97                               NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0);
  98                 switch (fb->zsbuf->format) {
  99                 case PIPE_FORMAT_Z32_FLOAT:
 100                         so_data(so, NV50TCL_ZETA_FORMAT_Z32_FLOAT);
 101                         break;
 102                 case PIPE_FORMAT_Z24S8_UNORM:
 103                         so_data(so, NV50TCL_ZETA_FORMAT_Z24S8_UNORM);
 104                         break;
 105                 case PIPE_FORMAT_X8Z24_UNORM:
 106                         so_data(so, NV50TCL_ZETA_FORMAT_X8Z24_UNORM);
 107                         break;
 108                 case PIPE_FORMAT_S8Z24_UNORM:
 109                         so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM);
 110                         break;
 111                 default:
 112                         NOUVEAU_ERR("AIIII unknown format %s\n",
 113                                     pf_name(fb->zsbuf->format));
 114                         so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM);
 115                         break;
 116                 }
 117                 so_data(so, nv50_miptree(pt)->
 118                                 level[fb->zsbuf->level].tile_mode << 4);
 119                 so_data(so, 0x00000000);
 120
 121                 so_method(so, tesla, 0x1538, 1);
 122                 so_data  (so, 1);
 123                 so_method(so, tesla, NV50TCL_ZETA_HORIZ, 3);
 124                 so_data  (so, fb->zsbuf->width);
 125                 so_data  (so, fb->zsbuf->height);
 126                 so_data  (so, 0x00010001);
 127         }
 128
 129         so_method(so, tesla, NV50TCL_VIEWPORT_HORIZ, 2);
 130         so_data  (so, w << 16);
 131         so_data  (so, h << 16);
 132         /* set window lower left corner */
 133         so_method(so, tesla, NV50TCL_WINDOW_LEFT, 2);
 134         so_data  (so, 0);
 135         so_data  (so, 0);
 136         /* set screen scissor rectangle */
 137         so_method(so, tesla, NV50TCL_SCREEN_SCISSOR_HORIZ, 2);
 138         so_data  (so, w << 16);
 139         so_data  (so, h << 16);
 140
 141         /* we set scissors to framebuffer size when they're 'turned off' */
 142         nv50->dirty |= NV50_NEW_SCISSOR;
 143         so_ref(NULL, &nv50->state.scissor);
 144
 145         so_ref(so, &nv50->state.fb);
 146         so_ref(NULL, &so);
 147 }
 148
 149 static void
 150 nv50_state_emit(struct nv50_context *nv50)
 151 {
 152         struct nv50_screen *screen = nv50->screen;
 153         struct nouveau_channel *chan = screen->base.channel;
 154
 155         if (nv50->pctx_id != screen->cur_pctx) {
 156                 if (nv50->state.fb)
 157                         nv50->state.dirty |= NV50_NEW_FRAMEBUFFER;
 158                 if (nv50->state.blend)
 159                         nv50->state.dirty |= NV50_NEW_BLEND;
 160                 if (nv50->state.zsa)
 161                         nv50->state.dirty |= NV50_NEW_ZSA;
 162                 if (nv50->state.vertprog)
 163                         nv50->state.dirty |= NV50_NEW_VERTPROG;
 164                 if (nv50->state.fragprog)
 165                         nv50->state.dirty |= NV50_NEW_FRAGPROG;
 166                 if (nv50->state.rast)
 167                         nv50->state.dirty |= NV50_NEW_RASTERIZER;
 168                 if (nv50->state.blend_colour)
 169                         nv50->state.dirty |= NV50_NEW_BLEND_COLOUR;
 170                 if (nv50->state.stipple)
 171                         nv50->state.dirty |= NV50_NEW_STIPPLE;
 172                 if (nv50->state.scissor)
 173                         nv50->state.dirty |= NV50_NEW_SCISSOR;
 174                 if (nv50->state.viewport)
 175                         nv50->state.dirty |= NV50_NEW_VIEWPORT;
 176                 if (nv50->state.tsc_upload)
 177                         nv50->state.dirty |= NV50_NEW_SAMPLER;
 178                 if (nv50->state.tic_upload)
 179                         nv50->state.dirty |= NV50_NEW_TEXTURE;
 180                 if (nv50->state.vtxfmt && nv50->state.vtxbuf)
 181                         nv50->state.dirty |= NV50_NEW_ARRAYS;
 182                 screen->cur_pctx = nv50->pctx_id;
 183         }
 184
 185         if (nv50->state.dirty & NV50_NEW_FRAMEBUFFER)
 186                 so_emit(chan, nv50->state.fb);
 187         if (nv50->state.dirty & NV50_NEW_BLEND)
 188                 so_emit(chan, nv50->state.blend);
 189         if (nv50->state.dirty & NV50_NEW_ZSA)
 190                 so_emit(chan, nv50->state.zsa);
 191         if (nv50->state.dirty & NV50_NEW_VERTPROG)
 192                 so_emit(chan, nv50->state.vertprog);
 193         if (nv50->state.dirty & NV50_NEW_FRAGPROG)
 194                 so_emit(chan, nv50->state.fragprog);
 195         if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG))
 196                 so_emit(chan, nv50->state.programs);
 197         if (nv50->state.dirty & NV50_NEW_RASTERIZER)
 198                 so_emit(chan, nv50->state.rast);
 199         if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR)
 200                 so_emit(chan, nv50->state.blend_colour);
 201         if (nv50->state.dirty & NV50_NEW_STIPPLE)
 202                 so_emit(chan, nv50->state.stipple);
 203         if (nv50->state.dirty & NV50_NEW_SCISSOR)
 204                 so_emit(chan, nv50->state.scissor);
 205         if (nv50->state.dirty & NV50_NEW_VIEWPORT)
 206                 so_emit(chan, nv50->state.viewport);
 207         if (nv50->state.dirty & NV50_NEW_SAMPLER)
 208                 so_emit(chan, nv50->state.tsc_upload);
 209         if (nv50->state.dirty & NV50_NEW_TEXTURE)
 210                 so_emit(chan, nv50->state.tic_upload);
 211         if (nv50->state.dirty & NV50_NEW_ARRAYS) {
 212                 so_emit(chan, nv50->state.vtxfmt);
 213                 so_emit(chan, nv50->state.vtxbuf);
 214                 if (nv50->state.vtxattr)
 215                         so_emit(chan, nv50->state.vtxattr);
 216         }
 217         nv50->state.dirty = 0;
 218 }
 219
 220 void
 221 nv50_state_flush_notify(struct nouveau_channel *chan)
 222 {
 223         struct nv50_context *nv50 = chan->user_private;
 224
 225         if (nv50->state.tic_upload && !(nv50->dirty & NV50_NEW_TEXTURE))
 226                 so_emit(chan, nv50->state.tic_upload);
 227
 228         so_emit_reloc_markers(chan, nv50->state.fb);
 229         so_emit_reloc_markers(chan, nv50->state.vertprog);
 230         so_emit_reloc_markers(chan, nv50->state.fragprog);
 231         so_emit_reloc_markers(chan, nv50->state.vtxbuf);
 232         so_emit_reloc_markers(chan, nv50->screen->static_init);
 233 }
 234
 235 boolean
 236 nv50_state_validate(struct nv50_context *nv50)
 237 {
 238         struct nouveau_grobj *tesla = nv50->screen->tesla;
 239         struct nouveau_grobj *eng2d = nv50->screen->eng2d;
 240         struct nouveau_stateobj *so;
 241         unsigned i;
 242
 243         if (nv50->dirty & NV50_NEW_FRAMEBUFFER)
 244                 nv50_state_validate_fb(nv50);
 245
 246         if (nv50->dirty & NV50_NEW_BLEND)
 247                 so_ref(nv50->blend->so, &nv50->state.blend);
 248
 249         if (nv50->dirty & NV50_NEW_ZSA)
 250                 so_ref(nv50->zsa->so, &nv50->state.zsa);
 251
 252         if (nv50->dirty & (NV50_NEW_VERTPROG | NV50_NEW_VERTPROG_CB))
 253                 nv50_vertprog_validate(nv50);
 254
 255         if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB))
 256                 nv50_fragprog_validate(nv50);
 257
 258         if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG))
 259                 nv50_linkage_validate(nv50);
 260
 261         if (nv50->dirty & NV50_NEW_RASTERIZER)
 262                 so_ref(nv50->rasterizer->so, &nv50->state.rast);
 263
 264         if (nv50->dirty & NV50_NEW_BLEND_COLOUR) {
 265                 so = so_new(5, 0);
 266                 so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4);
 267                 so_data  (so, fui(nv50->blend_colour.color[0]));
 268                 so_data  (so, fui(nv50->blend_colour.color[1]));
 269                 so_data  (so, fui(nv50->blend_colour.color[2]));
 270                 so_data  (so, fui(nv50->blend_colour.color[3]));
 271                 so_ref(so, &nv50->state.blend_colour);
 272                 so_ref(NULL, &so);
 273         }
 274
 275         if (nv50->dirty & NV50_NEW_STIPPLE) {
 276                 so = so_new(33, 0);
 277                 so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32);
 278                 for (i = 0; i < 32; i++)
 279                         so_data(so, nv50->stipple.stipple[i]);
 280                 so_ref(so, &nv50->state.stipple);
 281                 so_ref(NULL, &so);
 282         }
 283
 284         if (nv50->dirty & (NV50_NEW_SCISSOR | NV50_NEW_RASTERIZER)) {
 285                 struct pipe_rasterizer_state *rast = &nv50->rasterizer->pipe;
 286                 struct pipe_scissor_state *s = &nv50->scissor;
 287
 288                 if (nv50->state.scissor &&
 289                     (rast->scissor == 0 && nv50->state.scissor_enabled == 0))
 290                         goto scissor_uptodate;
 291                 nv50->state.scissor_enabled = rast->scissor;
 292
 293                 so = so_new(3, 0);
 294                 so_method(so, tesla, NV50TCL_SCISSOR_HORIZ, 2);
 295                 if (nv50->state.scissor_enabled) {
 296                         so_data(so, (s->maxx << 16) | s->minx);
 297                         so_data(so, (s->maxy << 16) | s->miny);
 298                 } else {
 299                         so_data(so, (nv50->framebuffer.width << 16));
 300                         so_data(so, (nv50->framebuffer.height << 16));
 301                 }
 302                 so_ref(so, &nv50->state.scissor);
 303                 so_ref(NULL, &so);
 304                 nv50->state.dirty |= NV50_NEW_SCISSOR;
 305         }
 306 scissor_uptodate:
 307
 308         if (nv50->dirty & (NV50_NEW_VIEWPORT | NV50_NEW_RASTERIZER)) {
 309                 unsigned bypass;
 310
 311                 if (!nv50->rasterizer->pipe.bypass_vs_clip_and_viewport)
 312                         bypass = 0;
 313                 else
 314                         bypass = 1;
 315
 316                 if (nv50->state.viewport &&
 317                     (bypass || !(nv50->dirty & NV50_NEW_VIEWPORT)) &&
 318                     nv50->state.viewport_bypass == bypass)
 319                         goto viewport_uptodate;
 320                 nv50->state.viewport_bypass = bypass;
 321
 322                 so = so_new(14, 0);
 323                 if (!bypass) {
 324                         so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE(0), 3);
 325                         so_data  (so, fui(nv50->viewport.translate[0]));
 326                         so_data  (so, fui(nv50->viewport.translate[1]));
 327                         so_data  (so, fui(nv50->viewport.translate[2]));
 328                         so_method(so, tesla, NV50TCL_VIEWPORT_SCALE(0), 3);
 329                         so_data  (so, fui(nv50->viewport.scale[0]));
 330                         so_data  (so, fui(nv50->viewport.scale[1]));
 331                         so_data  (so, fui(nv50->viewport.scale[2]));
 332
 333                         so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
 334                         so_data  (so, 1);
 335                         /* 0x0000 = remove whole primitive only (xyz)
 336                          * 0x1018 = remove whole primitive only (xy), clamp z
 337                          * 0x1080 = clip primitive (xyz)
 338                          * 0x1098 = clip primitive (xy), clamp z
 339                          */
 340                         so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
 341                         so_data  (so, 0x1080);
 342                         /* no idea what 0f90 does */
 343                         so_method(so, tesla, 0x0f90, 1);
 344                         so_data  (so, 0);
 345                 } else {
 346                         so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
 347                         so_data  (so, 0);
 348                         so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
 349                         so_data  (so, 0x0000);
 350                         so_method(so, tesla, 0x0f90, 1);
 351                         so_data  (so, 1);
 352                 }
 353
 354                 so_ref(so, &nv50->state.viewport);
 355                 so_ref(NULL, &so);
 356                 nv50->state.dirty |= NV50_NEW_VIEWPORT;
 357         }
 358 viewport_uptodate:
 359
 360         if (nv50->dirty & NV50_NEW_SAMPLER) {
 361                 unsigned i;
 362
 363                 so = so_new(nv50->sampler_nr * 9 + 23 + 4, 2);
 364
 365                 nv50_so_init_sifc(nv50, so, nv50->screen->tsc, NOUVEAU_BO_VRAM,
 366                                   nv50->sampler_nr * 8 * 4);
 367
 368                 for (i = 0; i < nv50->sampler_nr; i++) {
 369                         if (!nv50->sampler[i])
 370                                 continue;
 371                         so_method(so, eng2d, NV50_2D_SIFC_DATA | (2 << 29), 8);
 372                         so_datap (so, nv50->sampler[i]->tsc, 8);
 373                 }
 374
 375                 so_method(so, tesla, 0x1440, 1); /* sync SIFC */
 376                 so_data  (so, 0);
 377                 so_method(so, tesla, 0x1334, 1); /* flush TSC */
 378                 so_data  (so, 0);
 379
 380                 so_ref(so, &nv50->state.tsc_upload);
 381                 so_ref(NULL, &so);
 382         }
 383
 384         if (nv50->dirty & (NV50_NEW_TEXTURE | NV50_NEW_SAMPLER))
 385                 nv50_tex_validate(nv50);
 386
 387         if (nv50->dirty & NV50_NEW_ARRAYS)
 388                 nv50_vbo_validate(nv50);
 389
 390         nv50->state.dirty |= nv50->dirty;
 391         nv50->dirty = 0;
 392         nv50_state_emit(nv50);
 393
 394         return TRUE;
 395 }
 396
 397 void nv50_so_init_sifc(struct nv50_context *nv50,
 398                        struct nouveau_stateobj *so,
 399                        struct nouveau_bo *bo, unsigned reloc, unsigned size)
 400 {
 401         struct nouveau_grobj *eng2d = nv50->screen->eng2d;
 402
 403         so_method(so, eng2d, NV50_2D_DST_FORMAT, 2);
 404         so_data  (so, NV50_2D_DST_FORMAT_R8_UNORM);
 405         so_data  (so, 1);
 406         so_method(so, eng2d, NV50_2D_DST_PITCH, 5);
 407         so_data  (so, 262144);
 408         so_data  (so, 65536);
 409         so_data  (so, 1);
 410         so_reloc (so, bo, 0, reloc | NOUVEAU_BO_WR | NOUVEAU_BO_HIGH, 0, 0);
 411         so_reloc (so, bo, 0, reloc | NOUVEAU_BO_WR | NOUVEAU_BO_LOW, 0, 0);
 412         so_method(so, eng2d, NV50_2D_SIFC_UNK0800, 2);
 413         so_data  (so, 0);
 414         so_data  (so, NV50_2D_SIFC_FORMAT_R8_UNORM);
 415         so_method(so, eng2d, NV50_2D_SIFC_WIDTH, 10);
 416         so_data  (so, size);
 417         so_data  (so, 1);
 418         so_data  (so, 0);
 419         so_data  (so, 1);
 420         so_data  (so, 0);
 421         so_data  (so, 1);
 422         so_data  (so, 0);
 423         so_data  (so, 0);
 424         so_data  (so, 0);
 425         so_data  (so, 0);
 426 }