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_resource.h"
  34
  35 #include "pipe/p_context.h"
  36 #include "pipe/p_defines.h"
  37
  38 #include "util/u_math.h"
  39
  40 /**
  41  * Examine framebuffer state to determine width, height.
  42  */
  43 static boolean
  44 framebuffer_size(const struct pipe_framebuffer_state *fb,
  45                  uint *width, uint *height)
  46 {
  47    if (fb->cbufs[0]) {
  48       *width = fb->cbufs[0]->width;
  49       *height = fb->cbufs[0]->height;
  50       return TRUE;
  51    }
  52    else if (fb->zsbuf) {
  53       *width = fb->zsbuf->width;
  54       *height = fb->zsbuf->height;
  55       return TRUE;
  56    }
  57    else {
  58       *width = *height = 0;
  59       return FALSE;
  60    }
  61 }
  62
  63
  64 /* Push the state into the sarea and/or texture memory.
  65  */
  66 void
  67 i915_emit_hardware_state(struct i915_context *i915 )
  68 {
  69    /* XXX: there must be an easier way */
  70    const unsigned dwords = ( 14 +
  71                              7 +
  72                              I915_MAX_DYNAMIC +
  73                              8 +
  74                              2 + I915_TEX_UNITS*3 +
  75                              2 + I915_TEX_UNITS*3 +
  76                              2 + I915_MAX_CONSTANT*4 +
  77 #if 0
  78                              i915->current.program_len +
  79 #else
  80                              i915->fs->program_len +
  81 #endif
  82                              6
  83                            ) * 3/2; /* plus 50% margin */
  84    const unsigned relocs = ( I915_TEX_UNITS +
  85                              3
  86                            ) * 3/2; /* plus 50% margin */
  87
  88    uintptr_t save_ptr;
  89    size_t save_relocs;
  90
  91    if (I915_DBG_ON(DBG_ATOMS))
  92       i915_dump_hardware_dirty(i915, __FUNCTION__);
  93
  94    if(!BEGIN_BATCH(dwords, relocs)) {
  95       FLUSH_BATCH(NULL);
  96       assert(BEGIN_BATCH(dwords, relocs));
  97    }
  98
  99    save_ptr = (uintptr_t)i915->batch->ptr;
 100    save_relocs = i915->batch->relocs;
 101
 102    /* 14 dwords, 0 relocs */
 103    if (i915->hardware_dirty & I915_HW_INVARIANT)
 104    {
 105       OUT_BATCH(_3DSTATE_AA_CMD |
 106                 AA_LINE_ECAAR_WIDTH_ENABLE |
 107                 AA_LINE_ECAAR_WIDTH_1_0 |
 108                 AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0);
 109
 110       OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
 111       OUT_BATCH(0);
 112
 113       OUT_BATCH(_3DSTATE_DFLT_SPEC_CMD);
 114       OUT_BATCH(0);
 115
 116       OUT_BATCH(_3DSTATE_DFLT_Z_CMD);
 117       OUT_BATCH(0);
 118
 119       OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS |
 120                 CSB_TCB(0, 0) |
 121                 CSB_TCB(1, 1) |
 122                 CSB_TCB(2, 2) |
 123                 CSB_TCB(3, 3) |
 124                 CSB_TCB(4, 4) |
 125                 CSB_TCB(5, 5) |
 126                 CSB_TCB(6, 6) |
 127                 CSB_TCB(7, 7));
 128
 129       OUT_BATCH(_3DSTATE_RASTER_RULES_CMD |
 130                 ENABLE_POINT_RASTER_RULE |
 131                 OGL_POINT_RASTER_RULE |
 132                 ENABLE_LINE_STRIP_PROVOKE_VRTX |
 133                 ENABLE_TRI_FAN_PROVOKE_VRTX |
 134                 LINE_STRIP_PROVOKE_VRTX(1) |
 135                 TRI_FAN_PROVOKE_VRTX(2) |
 136                 ENABLE_TEXKILL_3D_4D |
 137                 TEXKILL_4D);
 138
 139       OUT_BATCH(_3DSTATE_DEPTH_SUBRECT_DISABLE);
 140
 141       /* disable indirect state for now
 142        */
 143       OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0);
 144       OUT_BATCH(0);
 145    }
 146
 147    /* 7 dwords, 1 relocs */
 148    if (i915->hardware_dirty & I915_HW_IMMEDIATE)
 149    {
 150       /* remove unwatned bits and S7 */
 151       unsigned dirty = (1 << I915_IMMEDIATE_S0 | 1 << I915_IMMEDIATE_S1 |
 152                         1 << I915_IMMEDIATE_S2 | 1 << I915_IMMEDIATE_S3 |
 153                         1 << I915_IMMEDIATE_S3 | 1 << I915_IMMEDIATE_S4 |
 154                         1 << I915_IMMEDIATE_S5 | 1 << I915_IMMEDIATE_S6) &
 155                        i915->immediate_dirty;
 156       int i, num = util_bitcount(dirty);
 157       assert(num && num <= I915_MAX_IMMEDIATE);
 158
 159       OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 |
 160                 dirty << 4 | (num - 1));
 161
 162       if (i915->immediate_dirty & (1 << I915_IMMEDIATE_S0)) {
 163          if (i915->vbo)
 164             OUT_RELOC(i915->vbo, I915_USAGE_VERTEX,
 165                       i915->current.immediate[I915_IMMEDIATE_S0]);
 166          else
 167             OUT_BATCH(0);
 168       }
 169
 170       for (i = 1; i < I915_MAX_IMMEDIATE; i++) {
 171          if (dirty & (1 << i))
 172             OUT_BATCH(i915->current.immediate[i]);
 173       }
 174    }
 175
 176 #if 01
 177    /* I915_MAX_DYNAMIC dwords, 0 relocs */
 178    if (i915->hardware_dirty & I915_HW_DYNAMIC)
 179    {
 180       int i;
 181       for (i = 0; i < I915_MAX_DYNAMIC; i++) {
 182          if (i915->dynamic_dirty & (1 << i))
 183             OUT_BATCH(i915->current.dynamic[i]);
 184       }
 185    }
 186 #endif
 187
 188 #if 01
 189    /* 8 dwords, 2 relocs */
 190    if (i915->hardware_dirty & I915_HW_STATIC)
 191    {
 192       if (i915->current.cbuf_bo) {
 193          OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
 194          OUT_BATCH(i915->current.cbuf_flags);
 195          OUT_RELOC(i915->current.cbuf_bo,
 196                    I915_USAGE_RENDER,
 197                    0);
 198       }
 199
 200       /* What happens if no zbuf??
 201        */
 202       if (i915->current.depth_bo) {
 203          OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
 204          OUT_BATCH(i915->current.depth_flags);
 205          OUT_RELOC(i915->current.depth_bo,
 206                    I915_USAGE_RENDER,
 207                    0);
 208       }
 209
 210       {
 211          OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD);
 212          OUT_BATCH(i915->current.dst_buf_vars);
 213       }
 214    }
 215 #endif
 216
 217 #if 01
 218       /* texture images */
 219       /* 2 + I915_TEX_UNITS*3 dwords, I915_TEX_UNITS relocs */
 220       if (i915->hardware_dirty & (I915_HW_MAP | I915_HW_SAMPLER))
 221       {
 222          const uint nr = i915->current.sampler_enable_nr;
 223          if (nr) {
 224             const uint enabled = i915->current.sampler_enable_flags;
 225             uint unit;
 226             uint count = 0;
 227             OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
 228             OUT_BATCH(enabled);
 229             for (unit = 0; unit < I915_TEX_UNITS; unit++) {
 230                if (enabled & (1 << unit)) {
 231                   struct i915_texture *texture = i915_texture(i915->fragment_sampler_views[unit]->texture);
 232                   struct i915_winsys_buffer *buf = texture->buffer;
 233                   assert(buf);
 234
 235                   count++;
 236
 237                   OUT_RELOC(buf, I915_USAGE_SAMPLER, 0);
 238                   OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */
 239                   OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */
 240                }
 241             }
 242             assert(count == nr);
 243          }
 244       }
 245 #endif
 246
 247 #if 01
 248    /* samplers */
 249    /* 2 + I915_TEX_UNITS*3 dwords, 0 relocs */
 250    if (i915->hardware_dirty & I915_HW_SAMPLER)
 251    {
 252       if (i915->current.sampler_enable_nr) {
 253          int i;
 254
 255          OUT_BATCH( _3DSTATE_SAMPLER_STATE |
 256                     (3 * i915->current.sampler_enable_nr) );
 257
 258          OUT_BATCH( i915->current.sampler_enable_flags );
 259
 260          for (i = 0; i < I915_TEX_UNITS; i++) {
 261             if (i915->current.sampler_enable_flags & (1<<i)) {
 262                OUT_BATCH( i915->current.sampler[i][0] );
 263                OUT_BATCH( i915->current.sampler[i][1] );
 264                OUT_BATCH( i915->current.sampler[i][2] );
 265             }
 266          }
 267       }
 268    }
 269 #endif
 270
 271 #if 01
 272    /* constants */
 273    /* 2 + I915_MAX_CONSTANT*4 dwords, 0 relocs */
 274    if (i915->hardware_dirty & I915_HW_CONSTANTS)
 275    {
 276       /* Collate the user-defined constants with the fragment shader's
 277        * immediates according to the constant_flags[] array.
 278        */
 279       const uint nr = i915->fs->num_constants;
 280       if (nr) {
 281          uint i;
 282
 283          OUT_BATCH( _3DSTATE_PIXEL_SHADER_CONSTANTS | (nr * 4) );
 284          OUT_BATCH((1 << nr) - 1);
 285
 286          for (i = 0; i < nr; i++) {
 287             const uint *c;
 288             if (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER) {
 289                /* grab user-defined constant */
 290                c = (uint *) i915_buffer(i915->constants[PIPE_SHADER_FRAGMENT])->data;
 291                c += 4 * i;
 292             }
 293             else {
 294                /* emit program constant */
 295                c = (uint *) i915->fs->constants[i];
 296             }
 297 #if 0 /* debug */
 298             {
 299                float *f = (float *) c;
 300                printf("Const %2d: %f %f %f %f %s\n", i, f[0], f[1], f[2], f[3],
 301                       (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER
 302                        ? "user" : "immediate"));
 303             }
 304 #endif
 305             OUT_BATCH(*c++);
 306             OUT_BATCH(*c++);
 307             OUT_BATCH(*c++);
 308             OUT_BATCH(*c++);
 309          }
 310       }
 311    }
 312 #endif
 313
 314 #if 01
 315    /* Fragment program */
 316    /* i915->current.program_len dwords, 0 relocs */
 317    if (i915->hardware_dirty & I915_HW_PROGRAM)
 318    {
 319       uint i;
 320       /* we should always have, at least, a pass-through program */
 321       assert(i915->fs->program_len > 0);
 322       for (i = 0; i < i915->fs->program_len; i++) {
 323          OUT_BATCH(i915->fs->program[i]);
 324       }
 325    }
 326 #endif
 327
 328 #if 01
 329    /* drawing surface size */
 330    /* 6 dwords, 0 relocs */
 331    if (i915->hardware_dirty & I915_HW_STATIC)
 332    {
 333       uint w, h;
 334       struct pipe_surface *cbuf_surface = i915->framebuffer.cbufs[0];
 335       unsigned x, y;
 336       int layer;
 337       uint32_t draw_offset;
 338       boolean ret;
 339
 340       ret = framebuffer_size(&i915->framebuffer, &w, &h);
 341       assert(ret);
 342
 343       if (cbuf_surface) {
 344          struct i915_texture *tex = i915_texture(cbuf_surface->texture);
 345          layer = cbuf_surface->u.tex.first_layer;
 346
 347          x = tex->image_offset[cbuf_surface->u.tex.level][layer].nblocksx;
 348          y = tex->image_offset[cbuf_surface->u.tex.level][layer].nblocksy;
 349
 350       } else
 351          x = y = 0;
 352
 353       draw_offset = x | (y << 16);
 354
 355       /* XXX flush only required when the draw_offset changes! */
 356       OUT_BATCH(MI_FLUSH | INHIBIT_FLUSH_RENDER_CACHE);
 357       OUT_BATCH(_3DSTATE_DRAW_RECT_CMD);
 358       OUT_BATCH(DRAW_RECT_DIS_DEPTH_OFS);
 359       OUT_BATCH(draw_offset);
 360       OUT_BATCH((w - 1 + x) | ((h - 1 + y) << 16));
 361       OUT_BATCH(draw_offset);
 362    }
 363 #endif
 364
 365    I915_DBG(DBG_EMIT, "%s: used %d dwords, %d relocs\n", __FUNCTION__,
 366             ((uintptr_t)i915->batch->ptr - save_ptr) / 4,
 367             i915->batch->relocs - save_relocs);
 368
 369    i915->hardware_dirty = 0;
 370    i915->immediate_dirty = 0;
 371    i915->dynamic_dirty = 0;
 372 }