src/mesa/drivers/dri/i965/brw_gs.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 #include "main/glheader.h"
  33 #include "main/macros.h"
  34 #include "main/enums.h"
  35
  36 #include "intel_batchbuffer.h"
  37
  38 #include "brw_defines.h"
  39 #include "brw_context.h"
  40 #include "brw_eu.h"
  41 #include "brw_util.h"
  42 #include "brw_state.h"
  43 #include "brw_gs.h"
  44
  45 #include "glsl/ralloc.h"
  46
  47 static void compile_gs_prog( struct brw_context *brw,
  48                              struct brw_gs_prog_key *key )
  49 {
  50    struct intel_context *intel = &brw->intel;
  51    struct brw_gs_compile c;
  52    const GLuint *program;
  53    void *mem_ctx;
  54    GLuint program_size;
  55
  56    memset(&c, 0, sizeof(c));
  57
  58    c.key = *key;
  59    /* The geometry shader needs to access the entire VUE. */
  60    struct brw_vue_map vue_map;
  61    brw_compute_vue_map(&vue_map, intel, c.key.userclip_active, c.key.attrs);
  62    c.nr_regs = (vue_map.num_slots + 1)/2;
  63
  64    mem_ctx = NULL;
  65
  66    /* Begin the compilation:
  67     */
  68    brw_init_compile(brw, &c.func, mem_ctx);
  69
  70    c.func.single_program_flow = 1;
  71
  72    /* For some reason the thread is spawned with only 4 channels
  73     * unmasked.
  74     */
  75    brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
  76
  77    if (intel->gen >= 6) {
  78       unsigned num_verts;
  79       bool check_edge_flag;
  80       /* On Sandybridge, we use the GS for implementing transform feedback
  81        * (called "Stream Out" in the PRM).
  82        */
  83       switch (key->primitive) {
  84       case _3DPRIM_POINTLIST:
  85          num_verts = 1;
  86          check_edge_flag = false;
  87          break;
  88       case _3DPRIM_LINELIST:
  89       case _3DPRIM_LINESTRIP:
  90       case _3DPRIM_LINELOOP:
  91          num_verts = 2;
  92          check_edge_flag = false;
  93          break;
  94       case _3DPRIM_TRILIST:
  95       case _3DPRIM_TRIFAN:
  96       case _3DPRIM_TRISTRIP:
  97       case _3DPRIM_RECTLIST:
  98          num_verts = 3;
  99          check_edge_flag = false;
 100          break;
 101       case _3DPRIM_QUADLIST:
 102       case _3DPRIM_QUADSTRIP:
 103       case _3DPRIM_POLYGON:
 104          num_verts = 3;
 105          check_edge_flag = true;
 106          break;
 107       default:
 108          assert(!"Unexpected primitive type in Gen6 SOL program.");
 109          return;
 110       }
 111       gen6_sol_program(&c, key, num_verts, check_edge_flag);
 112    } else {
 113       /* On Gen4-5, we use the GS to decompose certain types of primitives.
 114        * Note that primitives which don't require a GS program have already
 115        * been weeded out by now.
 116        */
 117       switch (key->primitive) {
 118       case _3DPRIM_QUADLIST:
 119          brw_gs_quads( &c, key );
 120          break;
 121       case _3DPRIM_QUADSTRIP:
 122          brw_gs_quad_strip( &c, key );
 123          break;
 124       case _3DPRIM_LINELOOP:
 125          brw_gs_lines( &c );
 126          break;
 127       default:
 128          ralloc_free(mem_ctx);
 129          return;
 130       }
 131    }
 132
 133    /* get the program
 134     */
 135    program = brw_get_program(&c.func, &program_size);
 136
 137    if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
 138       int i;
 139
 140       printf("gs:\n");
 141       for (i = 0; i < program_size / sizeof(struct brw_instruction); i++)
 142          brw_disasm(stdout, &((struct brw_instruction *)program)[i],
 143                     intel->gen);
 144       printf("\n");
 145     }
 146
 147    brw_upload_cache(&brw->cache, BRW_GS_PROG,
 148                     &c.key, sizeof(c.key),
 149                     program, program_size,
 150                     &c.prog_data, sizeof(c.prog_data),
 151                     &brw->gs.prog_offset, &brw->gs.prog_data);
 152    ralloc_free(mem_ctx);
 153 }
 154
 155 static void populate_key( struct brw_context *brw,
 156                           struct brw_gs_prog_key *key )
 157 {
 158    struct gl_context *ctx = &brw->intel.ctx;
 159    struct intel_context *intel = &brw->intel;
 160
 161    memset(key, 0, sizeof(*key));
 162
 163    /* CACHE_NEW_VS_PROG */
 164    key->attrs = brw->vs.prog_data->outputs_written;
 165
 166    /* BRW_NEW_PRIMITIVE */
 167    key->primitive = brw->primitive;
 168
 169    /* _NEW_LIGHT */
 170    key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
 171    if (key->primitive == _3DPRIM_QUADLIST && ctx->Light.ShadeModel != GL_FLAT) {
 172       /* Provide consistent primitive order with brw_set_prim's
 173        * optimization of single quads to trifans.
 174        */
 175       key->pv_first = true;
 176    }
 177
 178    /* _NEW_TRANSFORM */
 179    key->userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
 180
 181    if (intel->gen >= 7) {
 182       /* On Gen7 and later, we don't use GS (yet). */
 183       key->need_gs_prog = false;
 184    } else if (intel->gen == 6) {
 185       /* On Gen6, GS is used for transform feedback. */
 186       /* _NEW_TRANSFORM_FEEDBACK */
 187       key->need_gs_prog = ctx->TransformFeedback.CurrentObject->Active;
 188    } else {
 189       /* Pre-gen6, GS is used to transform QUADLIST, QUADSTRIP, and LINELOOP
 190        * into simpler primitives.
 191        */
 192       key->need_gs_prog = (brw->primitive == _3DPRIM_QUADLIST ||
 193                            brw->primitive == _3DPRIM_QUADSTRIP ||
 194                            brw->primitive == _3DPRIM_LINELOOP);
 195    }
 196    /* For testing, the environment variable INTEL_FORCE_GS can be used to
 197     * force a GS program to be used, even if it's not necessary.
 198     */
 199    if (getenv("INTEL_FORCE_GS"))
 200       key->need_gs_prog = true;
 201 }
 202
 203 /* Calculate interpolants for triangle and line rasterization.
 204  */
 205 static void
 206 brw_upload_gs_prog(struct brw_context *brw)
 207 {
 208    struct brw_gs_prog_key key;
 209    /* Populate the key:
 210     */
 211    populate_key(brw, &key);
 212
 213    if (brw->gs.prog_active != key.need_gs_prog) {
 214       brw->state.dirty.cache |= CACHE_NEW_GS_PROG;
 215       brw->gs.prog_active = key.need_gs_prog;
 216    }
 217
 218    if (brw->gs.prog_active) {
 219       if (!brw_search_cache(&brw->cache, BRW_GS_PROG,
 220                             &key, sizeof(key),
 221                             &brw->gs.prog_offset, &brw->gs.prog_data)) {
 222          compile_gs_prog( brw, &key );
 223       }
 224    }
 225 }
 226
 227
 228 const struct brw_tracked_state brw_gs_prog = {
 229    .dirty = {
 230       .mesa  = (_NEW_LIGHT |
 231                 _NEW_TRANSFORM |
 232                 _NEW_TRANSFORM_FEEDBACK),
 233       .brw   = BRW_NEW_PRIMITIVE,
 234       .cache = CACHE_NEW_VS_PROG
 235    },
 236    .emit = brw_upload_gs_prog
 237 };