src/gallium/drivers/i915/i915_state_emit.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28
  29 #include "i915_reg.h"
  30 #include "i915_context.h"
  31 #include "i915_batch.h"
  32 #include "i915_debug.h"
  33 #include "i915_reg.h"
  34 #include "i915_resource.h"
  35
  36 #include "pipe/p_context.h"
  37 #include "pipe/p_defines.h"
  38
  39 static unsigned translate_format( enum pipe_format format )
  40 {
  41    switch (format) {
  42    case PIPE_FORMAT_B8G8R8A8_UNORM:
  43       return COLOR_BUF_ARGB8888;
  44    case PIPE_FORMAT_B5G6R5_UNORM:
  45       return COLOR_BUF_RGB565;
  46    default:
  47       assert(0);
  48       return 0;
  49    }
  50 }
  51
  52 static unsigned translate_depth_format( enum pipe_format zformat )
  53 {
  54    switch (zformat) {
  55    case PIPE_FORMAT_Z24X8_UNORM:
  56    case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
  57       return DEPTH_FRMT_24_FIXED_8_OTHER;
  58    case PIPE_FORMAT_Z16_UNORM:
  59       return DEPTH_FRMT_16_FIXED;
  60    default:
  61       assert(0);
  62       return 0;
  63    }
  64 }
  65
  66
  67 /**
  68  * Examine framebuffer state to determine width, height.
  69  */
  70 static boolean
  71 framebuffer_size(const struct pipe_framebuffer_state *fb,
  72                  uint *width, uint *height)
  73 {
  74    if (fb->cbufs[0]) {
  75       *width = fb->cbufs[0]->width;
  76       *height = fb->cbufs[0]->height;
  77       return TRUE;
  78    }
  79    else if (fb->zsbuf) {
  80       *width = fb->zsbuf->width;
  81       *height = fb->zsbuf->height;
  82       return TRUE;
  83    }
  84    else {
  85       *width = *height = 0;
  86       return FALSE;
  87    }
  88 }
  89
  90
  91 /* Push the state into the sarea and/or texture memory.
  92  */
  93 void
  94 i915_emit_hardware_state(struct i915_context *i915 )
  95 {
  96    /* XXX: there must be an easier way */
  97    const unsigned dwords = ( 14 +
  98                              7 +
  99                              I915_MAX_DYNAMIC +
 100                              8 +
 101                              2 + I915_TEX_UNITS*3 +
 102                              2 + I915_TEX_UNITS*3 +
 103                              2 + I915_MAX_CONSTANT*4 +
 104 #if 0
 105                              i915->current.program_len +
 106 #else
 107                              i915->fs->program_len +
 108 #endif
 109                              6
 110                            ) * 3/2; /* plus 50% margin */
 111    const unsigned relocs = ( I915_TEX_UNITS +
 112                              3
 113                            ) * 3/2; /* plus 50% margin */
 114
 115    uintptr_t save_ptr;
 116    size_t save_relocs;
 117
 118    if (I915_DBG_ON(DBG_ATOMS))
 119       i915_dump_hardware_dirty(i915, __FUNCTION__);
 120
 121    if(!BEGIN_BATCH(dwords, relocs)) {
 122       FLUSH_BATCH(NULL);
 123       assert(BEGIN_BATCH(dwords, relocs));
 124    }
 125
 126    save_ptr = (uintptr_t)i915->batch->ptr;
 127    save_relocs = i915->batch->relocs;
 128
 129    /* 14 dwords, 0 relocs */
 130    if (i915->hardware_dirty & I915_HW_INVARIENT)
 131    {
 132       OUT_BATCH(_3DSTATE_AA_CMD |
 133                 AA_LINE_ECAAR_WIDTH_ENABLE |
 134                 AA_LINE_ECAAR_WIDTH_1_0 |
 135                 AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0);
 136
 137       OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
 138       OUT_BATCH(0);
 139
 140       OUT_BATCH(_3DSTATE_DFLT_SPEC_CMD);
 141       OUT_BATCH(0);
 142
 143       OUT_BATCH(_3DSTATE_DFLT_Z_CMD);
 144       OUT_BATCH(0);
 145
 146       OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS |
 147                 CSB_TCB(0, 0) |
 148                 CSB_TCB(1, 1) |
 149                 CSB_TCB(2, 2) |
 150                 CSB_TCB(3, 3) |
 151                 CSB_TCB(4, 4) |
 152                 CSB_TCB(5, 5) |
 153                 CSB_TCB(6, 6) |
 154                 CSB_TCB(7, 7));
 155
 156       OUT_BATCH(_3DSTATE_RASTER_RULES_CMD |
 157                 ENABLE_POINT_RASTER_RULE |
 158                 OGL_POINT_RASTER_RULE |
 159                 ENABLE_LINE_STRIP_PROVOKE_VRTX |
 160                 ENABLE_TRI_FAN_PROVOKE_VRTX |
 161                 LINE_STRIP_PROVOKE_VRTX(1) |
 162                 TRI_FAN_PROVOKE_VRTX(2) |
 163                 ENABLE_TEXKILL_3D_4D |
 164                 TEXKILL_4D);
 165
 166       /* Need to initialize this to zero.
 167        */
 168       OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(3) | (0));
 169       OUT_BATCH(0);
 170
 171       OUT_BATCH(_3DSTATE_DEPTH_SUBRECT_DISABLE);
 172
 173       /* disable indirect state for now
 174        */
 175       OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0);
 176       OUT_BATCH(0);
 177    }
 178
 179    /* 7 dwords, 1 relocs */
 180    if (i915->hardware_dirty & I915_HW_IMMEDIATE)
 181    {
 182       OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 |
 183                 I1_LOAD_S(0) |
 184                 I1_LOAD_S(1) |
 185                 I1_LOAD_S(2) |
 186                 I1_LOAD_S(4) |
 187                 I1_LOAD_S(5) |
 188                 I1_LOAD_S(6) |
 189                 (5));
 190
 191       if(i915->vbo)
 192          OUT_RELOC(i915->vbo,
 193                    I915_USAGE_VERTEX,
 194                    i915->current.immediate[I915_IMMEDIATE_S0]);
 195       else
 196          /* FIXME: we should not do this */
 197          OUT_BATCH(0);
 198       OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S1]);
 199       OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S2]);
 200       OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S4]);
 201       OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S5]);
 202       OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S6]);
 203    }
 204
 205    /* I915_MAX_DYNAMIC dwords, 0 relocs */
 206    if (i915->hardware_dirty & I915_HW_DYNAMIC)
 207    {
 208       int i;
 209       for (i = 0; i < I915_MAX_DYNAMIC; i++) {
 210          OUT_BATCH(i915->current.dynamic[i]);
 211       }
 212    }
 213
 214    /* 8 dwords, 2 relocs */
 215    if (i915->hardware_dirty & I915_HW_STATIC)
 216    {
 217       struct pipe_surface *cbuf_surface = i915->framebuffer.cbufs[0];
 218       struct pipe_surface *depth_surface = i915->framebuffer.zsbuf;
 219
 220       if (cbuf_surface) {
 221          unsigned ctile = BUF_3D_USE_FENCE;
 222          struct i915_texture *tex = i915_texture(cbuf_surface->texture);
 223          assert(tex);
 224
 225          if (tex && tex->sw_tiled) {
 226             ctile = BUF_3D_TILED_SURFACE;
 227          }
 228
 229          OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
 230
 231          OUT_BATCH(BUF_3D_ID_COLOR_BACK |
 232                    BUF_3D_PITCH(tex->stride) |  /* pitch in bytes */
 233                    ctile);
 234
 235          OUT_RELOC(tex->buffer,
 236                    I915_USAGE_RENDER,
 237                    cbuf_surface->offset);
 238       }
 239
 240       /* What happens if no zbuf??
 241        */
 242       if (depth_surface) {
 243          unsigned ztile = BUF_3D_USE_FENCE;
 244          struct i915_texture *tex = i915_texture(depth_surface->texture);
 245          assert(tex);
 246
 247          if (tex && tex->sw_tiled) {
 248             ztile = BUF_3D_TILED_SURFACE;
 249          }
 250
 251          OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
 252
 253          assert(tex);
 254          OUT_BATCH(BUF_3D_ID_DEPTH |
 255                    BUF_3D_PITCH(tex->stride) |  /* pitch in bytes */
 256                    ztile);
 257
 258          OUT_RELOC(tex->buffer,
 259                    I915_USAGE_RENDER,
 260                    depth_surface->offset);
 261       }
 262
 263       {
 264          unsigned cformat, zformat = 0;
 265
 266          if (cbuf_surface)
 267             cformat = cbuf_surface->format;
 268          else
 269             cformat = PIPE_FORMAT_B8G8R8A8_UNORM; /* arbitrary */
 270          cformat = translate_format(cformat);
 271
 272          if (depth_surface)
 273             zformat = translate_depth_format( i915->framebuffer.zsbuf->format );
 274
 275          OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD);
 276          OUT_BATCH(DSTORG_HORT_BIAS(0x8) | /* .5 */
 277                    DSTORG_VERT_BIAS(0x8) | /* .5 */
 278                    LOD_PRECLAMP_OGL |
 279                    TEX_DEFAULT_COLOR_OGL |
 280                    cformat |
 281                    zformat );
 282       }
 283    }
 284
 285 #if 01
 286       /* texture images */
 287       /* 2 + I915_TEX_UNITS*3 dwords, I915_TEX_UNITS relocs */
 288       if (i915->hardware_dirty & (I915_HW_MAP | I915_HW_SAMPLER))
 289       {
 290          const uint nr = i915->current.sampler_enable_nr;
 291          if (nr) {
 292             const uint enabled = i915->current.sampler_enable_flags;
 293             uint unit;
 294             uint count = 0;
 295             OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
 296             OUT_BATCH(enabled);
 297             for (unit = 0; unit < I915_TEX_UNITS; unit++) {
 298                if (enabled & (1 << unit)) {
 299                   struct i915_texture *texture = i915_texture(i915->fragment_sampler_views[unit]->texture);
 300                   struct i915_winsys_buffer *buf = texture->buffer;
 301                   uint offset = 0;
 302                   assert(buf);
 303
 304                   count++;
 305
 306                   OUT_RELOC(buf, I915_USAGE_SAMPLER, offset);
 307                   OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */
 308                   OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */
 309                }
 310             }
 311             assert(count == nr);
 312          }
 313       }
 314 #endif
 315
 316 #if 01
 317    /* samplers */
 318    /* 2 + I915_TEX_UNITS*3 dwords, 0 relocs */
 319    if (i915->hardware_dirty & I915_HW_SAMPLER)
 320    {
 321       if (i915->current.sampler_enable_nr) {
 322          int i;
 323
 324          OUT_BATCH( _3DSTATE_SAMPLER_STATE |
 325                     (3 * i915->current.sampler_enable_nr) );
 326
 327          OUT_BATCH( i915->current.sampler_enable_flags );
 328
 329          for (i = 0; i < I915_TEX_UNITS; i++) {
 330             if (i915->current.sampler_enable_flags & (1<<i)) {
 331                OUT_BATCH( i915->current.sampler[i][0] );
 332                OUT_BATCH( i915->current.sampler[i][1] );
 333                OUT_BATCH( i915->current.sampler[i][2] );
 334             }
 335          }
 336       }
 337    }
 338 #endif
 339
 340    /* constants */
 341    /* 2 + I915_MAX_CONSTANT*4 dwords, 0 relocs */
 342    if (i915->hardware_dirty & I915_HW_PROGRAM)
 343    {
 344       /* Collate the user-defined constants with the fragment shader's
 345        * immediates according to the constant_flags[] array.
 346        */
 347       const uint nr = i915->fs->num_constants;
 348       if (nr) {
 349          uint i;
 350
 351          OUT_BATCH( _3DSTATE_PIXEL_SHADER_CONSTANTS | (nr * 4) );
 352          OUT_BATCH( (1 << (nr - 1)) | ((1 << (nr - 1)) - 1) );
 353
 354          for (i = 0; i < nr; i++) {
 355             const uint *c;
 356             if (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER) {
 357                /* grab user-defined constant */
 358                c = (uint *) i915->current.constants[PIPE_SHADER_FRAGMENT][i];
 359             }
 360             else {
 361                /* emit program constant */
 362                c = (uint *) i915->fs->constants[i];
 363             }
 364 #if 0 /* debug */
 365             {
 366                float *f = (float *) c;
 367                printf("Const %2d: %f %f %f %f %s\n", i, f[0], f[1], f[2], f[3],
 368                       (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER
 369                        ? "user" : "immediate"));
 370             }
 371 #endif
 372             OUT_BATCH(*c++);
 373             OUT_BATCH(*c++);
 374             OUT_BATCH(*c++);
 375             OUT_BATCH(*c++);
 376          }
 377       }
 378    }
 379
 380    /* Fragment program */
 381    /* i915->current.program_len dwords, 0 relocs */
 382    if (i915->hardware_dirty & I915_HW_PROGRAM)
 383    {
 384       uint i;
 385       /* we should always have, at least, a pass-through program */
 386       assert(i915->fs->program_len > 0);
 387       for (i = 0; i < i915->fs->program_len; i++) {
 388          OUT_BATCH(i915->fs->program[i]);
 389       }
 390    }
 391
 392    /* drawing surface size */
 393    /* 6 dwords, 0 relocs */
 394    {
 395       uint w, h;
 396       boolean k = framebuffer_size(&i915->framebuffer, &w, &h);
 397       (void)k;
 398       assert(k);
 399
 400       OUT_BATCH(_3DSTATE_DRAW_RECT_CMD);
 401       OUT_BATCH(0);
 402       OUT_BATCH(0);
 403       OUT_BATCH(((w - 1) & 0xffff) | ((h - 1) << 16));
 404       OUT_BATCH(0);
 405       OUT_BATCH(0);
 406    }
 407
 408    I915_DBG(DBG_EMIT, "%s: used %d dwords, %d relocs\n", __FUNCTION__,
 409             ((uintptr_t)i915->batch->ptr - save_ptr) / 4,
 410             i915->batch->relocs - save_relocs);
 411
 412    i915->hardware_dirty = 0;
 413 }