src/mesa/drivers/dri/i965/brw_clip_util.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a 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, sublicense, 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
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32
  33 #include "main/glheader.h"
  34 #include "main/macros.h"
  35 #include "main/enums.h"
  36 #include "shader/program.h"
  37
  38 #include "intel_batchbuffer.h"
  39
  40 #include "brw_defines.h"
  41 #include "brw_context.h"
  42 #include "brw_eu.h"
  43 #include "brw_util.h"
  44 #include "brw_clip.h"
  45
  46
  47
  48
  49 struct brw_reg get_tmp( struct brw_clip_compile *c )
  50 {
  51    struct brw_reg tmp = brw_vec4_grf(c->last_tmp, 0);
  52
  53    if (++c->last_tmp > c->prog_data.total_grf)
  54       c->prog_data.total_grf = c->last_tmp;
  55
  56    return tmp;
  57 }
  58
  59 static void release_tmp( struct brw_clip_compile *c, struct brw_reg tmp )
  60 {
  61    if (tmp.nr == c->last_tmp-1)
  62       c->last_tmp--;
  63 }
  64
  65
  66 static struct brw_reg make_plane_ud(GLuint x, GLuint y, GLuint z, GLuint w)
  67 {
  68    return brw_imm_ud((w<<24) | (z<<16) | (y<<8) | x);
  69 }
  70
  71
  72 void brw_clip_init_planes( struct brw_clip_compile *c )
  73 {
  74    struct brw_compile *p = &c->func;
  75
  76    if (!c->key.nr_userclip) {
  77       brw_MOV(p, get_element_ud(c->reg.fixed_planes, 0), make_plane_ud( 0,    0, 0xff, 1));
  78       brw_MOV(p, get_element_ud(c->reg.fixed_planes, 1), make_plane_ud( 0,    0,    1, 1));
  79       brw_MOV(p, get_element_ud(c->reg.fixed_planes, 2), make_plane_ud( 0, 0xff,    0, 1));
  80       brw_MOV(p, get_element_ud(c->reg.fixed_planes, 3), make_plane_ud( 0,    1,    0, 1));
  81       brw_MOV(p, get_element_ud(c->reg.fixed_planes, 4), make_plane_ud(0xff,  0,    0, 1));
  82       brw_MOV(p, get_element_ud(c->reg.fixed_planes, 5), make_plane_ud( 1,    0,    0, 1));
  83    }
  84 }
  85
  86
  87
  88 #define W 3
  89
  90 /* Project 'pos' to screen space (or back again), overwrite with results:
  91  */
  92 void brw_clip_project_position(struct brw_clip_compile *c, struct brw_reg pos )
  93 {
  94    struct brw_compile *p = &c->func;
  95
  96    /* calc rhw
  97     */
  98    brw_math_invert(p, get_element(pos, W), get_element(pos, W));
  99
 100    /* value.xyz *= value.rhw
 101     */
 102    brw_set_access_mode(p, BRW_ALIGN_16);
 103    brw_MUL(p, brw_writemask(pos, WRITEMASK_XYZ), pos, brw_swizzle1(pos, W));
 104    brw_set_access_mode(p, BRW_ALIGN_1);
 105 }
 106
 107
 108 static void brw_clip_project_vertex( struct brw_clip_compile *c,
 109                                      struct brw_indirect vert_addr )
 110 {
 111    struct brw_compile *p = &c->func;
 112    struct brw_reg tmp = get_tmp(c);
 113
 114    /* Fixup position.  Extract from the original vertex and re-project
 115     * to screen space:
 116     */
 117    brw_MOV(p, tmp, deref_4f(vert_addr, c->offset[VERT_RESULT_HPOS]));
 118    brw_clip_project_position(c, tmp);
 119    brw_MOV(p, deref_4f(vert_addr, c->header_position_offset), tmp);
 120
 121    release_tmp(c, tmp);
 122 }
 123
 124
 125
 126
 127 /* Interpolate between two vertices and put the result into a0.0.
 128  * Increment a0.0 accordingly.
 129  */
 130 void brw_clip_interp_vertex( struct brw_clip_compile *c,
 131                              struct brw_indirect dest_ptr,
 132                              struct brw_indirect v0_ptr, /* from */
 133                              struct brw_indirect v1_ptr, /* to */
 134                              struct brw_reg t0,
 135                              GLboolean force_edgeflag)
 136 {
 137    struct brw_compile *p = &c->func;
 138    struct brw_reg tmp = get_tmp(c);
 139    GLuint i;
 140
 141    /* Just copy the vertex header:
 142     */
 143    /*
 144     * After CLIP stage, only first 256 bits of the VUE are read
 145     * back on IGDNG, so needn't change it
 146     */
 147    brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
 148
 149    /* Iterate over each attribute (could be done in pairs?)
 150     */
 151    for (i = 0; i < c->nr_attrs; i++) {
 152       GLuint delta = i*16 + 32;
 153
 154       if (BRW_IS_IGDNG(p->brw))
 155           delta = i * 16 + 32 * 3;
 156
 157       if (delta == c->offset[VERT_RESULT_EDGE]) {
 158          if (force_edgeflag)
 159             brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
 160          else
 161             brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
 162       }
 163       else {
 164          /* Interpolate:
 165           *
 166           *        New = attr0 + t*attr1 - t*attr0
 167           */
 168          brw_MUL(p,
 169                  vec4(brw_null_reg()),
 170                  deref_4f(v1_ptr, delta),
 171                  t0);
 172
 173          brw_MAC(p,
 174                  tmp,
 175                  negate(deref_4f(v0_ptr, delta)),
 176                  t0);
 177
 178          brw_ADD(p,
 179                  deref_4f(dest_ptr, delta),
 180                  deref_4f(v0_ptr, delta),
 181                  tmp);
 182       }
 183    }
 184
 185    if (i & 1) {
 186       GLuint delta = i*16 + 32;
 187
 188       if (BRW_IS_IGDNG(p->brw))
 189           delta = i * 16 + 32 * 3;
 190
 191       brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
 192    }
 193
 194    release_tmp(c, tmp);
 195
 196    /* Recreate the projected (NDC) coordinate in the new vertex
 197     * header:
 198     */
 199    brw_clip_project_vertex(c, dest_ptr );
 200 }
 201
 202
 203
 204
 205 #define MAX_MRF 16
 206
 207 void brw_clip_emit_vue(struct brw_clip_compile *c,
 208                        struct brw_indirect vert,
 209                        GLboolean allocate,
 210                        GLboolean eot,
 211                        GLuint header)
 212 {
 213    struct brw_compile *p = &c->func;
 214    GLuint start = c->last_mrf;
 215
 216    assert(!(allocate && eot));
 217
 218    /* Cycle through mrf regs - probably futile as we have to wait for
 219     * the allocation response anyway.  Also, the order this function
 220     * is invoked doesn't correspond to the order the instructions will
 221     * be executed, so it won't have any effect in many cases.
 222     */
 223 #if 0
 224    if (start + c->nr_regs + 1 >= MAX_MRF)
 225       start = 0;
 226
 227    c->last_mrf = start + c->nr_regs + 1;
 228 #endif
 229
 230    /* Copy the vertex from vertn into m1..mN+1:
 231     */
 232    brw_copy_from_indirect(p, brw_message_reg(start+1), vert, c->nr_regs);
 233
 234    /* Overwrite PrimType and PrimStart in the message header, for
 235     * each vertex in turn:
 236     */
 237    brw_MOV(p, get_element_ud(c->reg.R0, 2), brw_imm_ud(header));
 238
 239
 240    /* Send each vertex as a seperate write to the urb.  This
 241     * is different to the concept in brw_sf_emit.c, where
 242     * subsequent writes are used to build up a single urb
 243     * entry.  Each of these writes instantiates a seperate
 244     * urb entry - (I think... what about 'allocate'?)
 245     */
 246    brw_urb_WRITE(p,
 247                  allocate ? c->reg.R0 : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
 248                  start,
 249                  c->reg.R0,
 250                  allocate,
 251                  1,             /* used */
 252                  c->nr_regs + 1, /* msg length */
 253                  allocate ? 1 : 0, /* response_length */
 254                  eot,           /* eot */
 255                  1,             /* writes_complete */
 256                  0,             /* urb offset */
 257                  BRW_URB_SWIZZLE_NONE);
 258 }
 259
 260
 261
 262 void brw_clip_kill_thread(struct brw_clip_compile *c)
 263 {
 264    struct brw_compile *p = &c->func;
 265
 266    /* Send an empty message to kill the thread and release any
 267     * allocated urb entry:
 268     */
 269    brw_urb_WRITE(p,
 270                  retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
 271                  0,
 272                  c->reg.R0,
 273                  0,             /* allocate */
 274                  0,             /* used */
 275                  1,             /* msg len */
 276                  0,             /* response len */
 277                  1,             /* eot */
 278                  1,             /* writes complete */
 279                  0,
 280                  BRW_URB_SWIZZLE_NONE);
 281 }
 282
 283
 284
 285
 286 struct brw_reg brw_clip_plane0_address( struct brw_clip_compile *c )
 287 {
 288    return brw_address(c->reg.fixed_planes);
 289 }
 290
 291
 292 struct brw_reg brw_clip_plane_stride( struct brw_clip_compile *c )
 293 {
 294    if (c->key.nr_userclip) {
 295       return brw_imm_uw(16);
 296    }
 297    else {
 298       return brw_imm_uw(4);
 299    }
 300 }
 301
 302
 303 /* If flatshading, distribute color from provoking vertex prior to
 304  * clipping.
 305  */
 306 void brw_clip_copy_colors( struct brw_clip_compile *c,
 307                            GLuint to, GLuint from )
 308 {
 309    struct brw_compile *p = &c->func;
 310
 311    if (c->offset[VERT_RESULT_COL0])
 312       brw_MOV(p,
 313               byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_COL0]),
 314               byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_COL0]));
 315
 316    if (c->offset[VERT_RESULT_COL1])
 317       brw_MOV(p,
 318               byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_COL1]),
 319               byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_COL1]));
 320
 321    if (c->offset[VERT_RESULT_BFC0])
 322       brw_MOV(p,
 323               byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_BFC0]),
 324               byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_BFC0]));
 325
 326    if (c->offset[VERT_RESULT_BFC1])
 327       brw_MOV(p,
 328               byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_BFC1]),
 329               byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_BFC1]));
 330 }
 331
 332
 333
 334 void brw_clip_init_clipmask( struct brw_clip_compile *c )
 335 {
 336    struct brw_compile *p = &c->func;
 337    struct brw_reg incoming = get_element_ud(c->reg.R0, 2);
 338
 339    /* Shift so that lowest outcode bit is rightmost:
 340     */
 341    brw_SHR(p, c->reg.planemask, incoming, brw_imm_ud(26));
 342
 343    if (c->key.nr_userclip) {
 344       struct brw_reg tmp = retype(vec1(get_tmp(c)), BRW_REGISTER_TYPE_UD);
 345
 346       /* Rearrange userclip outcodes so that they come directly after
 347        * the fixed plane bits.
 348        */
 349       brw_AND(p, tmp, incoming, brw_imm_ud(0x3f<<14));
 350       brw_SHR(p, tmp, tmp, brw_imm_ud(8));
 351       brw_OR(p, c->reg.planemask, c->reg.planemask, tmp);
 352
 353       release_tmp(c, tmp);
 354    }
 355 }
 356
 357 void brw_clip_ff_sync(struct brw_clip_compile *c)
 358 {
 359         struct brw_compile *p = &c->func;
 360         brw_ff_sync(p,
 361                                 c->reg.R0,
 362                                 0,
 363                                 c->reg.R0,
 364                                 1,
 365                                 1,              /* used */
 366                                 1,      /* msg length */
 367                                 1,              /* response length */
 368                                 0,              /* eot */
 369                                 1,              /* write compelete */
 370                                 0,              /* urb offset */
 371                                 BRW_URB_SWIZZLE_NONE);
 372 }