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

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics 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 <keithw@vmware.com>
  30   */
  31
  32 #include "main/glheader.h"
  33 #include "main/macros.h"
  34 #include "main/enums.h"
  35 #include "main/transformfeedback.h"
  36
  37 #include "intel_batchbuffer.h"
  38
  39 #include "brw_defines.h"
  40 #include "brw_context.h"
  41 #include "brw_eu.h"
  42 #include "brw_util.h"
  43 #include "brw_state.h"
  44 #include "brw_gs.h"
  45
  46 #include "glsl/ralloc.h"
  47
  48 static void compile_ff_gs_prog(struct brw_context *brw,
  49                                struct brw_ff_gs_prog_key *key)
  50 {
  51    struct brw_ff_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    c.vue_map = brw->vs.prog_data->base.vue_map;
  60    c.nr_regs = (c.vue_map.num_slots + 1)/2;
  61
  62    mem_ctx = ralloc_context(NULL);
  63
  64    /* Begin the compilation:
  65     */
  66    brw_init_compile(brw, &c.func, mem_ctx);
  67
  68    c.func.single_program_flow = 1;
  69
  70    /* For some reason the thread is spawned with only 4 channels
  71     * unmasked.
  72     */
  73    brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
  74
  75    if (brw->gen >= 6) {
  76       unsigned num_verts;
  77       bool check_edge_flag;
  78       /* On Sandybridge, we use the GS for implementing transform feedback
  79        * (called "Stream Out" in the PRM).
  80        */
  81       switch (key->primitive) {
  82       case _3DPRIM_POINTLIST:
  83          num_verts = 1;
  84          check_edge_flag = false;
  85          break;
  86       case _3DPRIM_LINELIST:
  87       case _3DPRIM_LINESTRIP:
  88       case _3DPRIM_LINELOOP:
  89          num_verts = 2;
  90          check_edge_flag = false;
  91          break;
  92       case _3DPRIM_TRILIST:
  93       case _3DPRIM_TRIFAN:
  94       case _3DPRIM_TRISTRIP:
  95       case _3DPRIM_RECTLIST:
  96          num_verts = 3;
  97          check_edge_flag = false;
  98          break;
  99       case _3DPRIM_QUADLIST:
 100       case _3DPRIM_QUADSTRIP:
 101       case _3DPRIM_POLYGON:
 102          num_verts = 3;
 103          check_edge_flag = true;
 104          break;
 105       default:
 106          assert(!"Unexpected primitive type in Gen6 SOL program.");
 107          return;
 108       }
 109       gen6_sol_program(&c, key, num_verts, check_edge_flag);
 110    } else {
 111       /* On Gen4-5, we use the GS to decompose certain types of primitives.
 112        * Note that primitives which don't require a GS program have already
 113        * been weeded out by now.
 114        */
 115       switch (key->primitive) {
 116       case _3DPRIM_QUADLIST:
 117          brw_ff_gs_quads( &c, key );
 118          break;
 119       case _3DPRIM_QUADSTRIP:
 120          brw_ff_gs_quad_strip( &c, key );
 121          break;
 122       case _3DPRIM_LINELOOP:
 123          brw_ff_gs_lines( &c );
 124          break;
 125       default:
 126          ralloc_free(mem_ctx);
 127          return;
 128       }
 129    }
 130
 131    brw_compact_instructions(&c.func);
 132
 133    /* get the program
 134     */
 135    program = brw_get_program(&c.func, &program_size);
 136
 137    if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
 138       fprintf(stderr, "gs:\n");
 139       brw_disassemble(brw, c.func.store, 0, program_size, stderr);
 140       fprintf(stderr, "\n");
 141     }
 142
 143    brw_upload_cache(&brw->cache, BRW_FF_GS_PROG,
 144                     &c.key, sizeof(c.key),
 145                     program, program_size,
 146                     &c.prog_data, sizeof(c.prog_data),
 147                     &brw->ff_gs.prog_offset, &brw->ff_gs.prog_data);
 148    ralloc_free(mem_ctx);
 149 }
 150
 151 static void populate_key(struct brw_context *brw,
 152                          struct brw_ff_gs_prog_key *key)
 153 {
 154    static const unsigned swizzle_for_offset[4] = {
 155       BRW_SWIZZLE4(0, 1, 2, 3),
 156       BRW_SWIZZLE4(1, 2, 3, 3),
 157       BRW_SWIZZLE4(2, 3, 3, 3),
 158       BRW_SWIZZLE4(3, 3, 3, 3)
 159    };
 160
 161    struct gl_context *ctx = &brw->ctx;
 162
 163    memset(key, 0, sizeof(*key));
 164
 165    /* CACHE_NEW_VS_PROG (part of VUE map) */
 166    key->attrs = brw->vs.prog_data->base.vue_map.slots_valid;
 167
 168    /* BRW_NEW_PRIMITIVE */
 169    key->primitive = brw->primitive;
 170
 171    /* _NEW_LIGHT */
 172    key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
 173    if (key->primitive == _3DPRIM_QUADLIST && ctx->Light.ShadeModel != GL_FLAT) {
 174       /* Provide consistent primitive order with brw_set_prim's
 175        * optimization of single quads to trifans.
 176        */
 177       key->pv_first = true;
 178    }
 179
 180    if (brw->gen >= 7) {
 181       /* On Gen7 and later, we don't use GS (yet). */
 182       key->need_gs_prog = false;
 183    } else if (brw->gen == 6) {
 184       /* On Gen6, GS is used for transform feedback. */
 185       /* BRW_NEW_TRANSFORM_FEEDBACK */
 186       if (_mesa_is_xfb_active_and_unpaused(ctx)) {
 187          const struct gl_shader_program *shaderprog =
 188             ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
 189          const struct gl_transform_feedback_info *linked_xfb_info =
 190             &shaderprog->LinkedTransformFeedback;
 191          int i;
 192
 193          /* Make sure that the VUE slots won't overflow the unsigned chars in
 194           * key->transform_feedback_bindings[].
 195           */
 196          STATIC_ASSERT(BRW_VARYING_SLOT_COUNT <= 256);
 197
 198          /* Make sure that we don't need more binding table entries than we've
 199           * set aside for use in transform feedback.  (We shouldn't, since we
 200           * set aside enough binding table entries to have one per component).
 201           */
 202          assert(linked_xfb_info->NumOutputs <= BRW_MAX_SOL_BINDINGS);
 203
 204          key->need_gs_prog = true;
 205          key->num_transform_feedback_bindings = linked_xfb_info->NumOutputs;
 206          for (i = 0; i < key->num_transform_feedback_bindings; ++i) {
 207             key->transform_feedback_bindings[i] =
 208                linked_xfb_info->Outputs[i].OutputRegister;
 209             key->transform_feedback_swizzles[i] =
 210                swizzle_for_offset[linked_xfb_info->Outputs[i].ComponentOffset];
 211          }
 212       }
 213    } else {
 214       /* Pre-gen6, GS is used to transform QUADLIST, QUADSTRIP, and LINELOOP
 215        * into simpler primitives.
 216        */
 217       key->need_gs_prog = (brw->primitive == _3DPRIM_QUADLIST ||
 218                            brw->primitive == _3DPRIM_QUADSTRIP ||
 219                            brw->primitive == _3DPRIM_LINELOOP);
 220    }
 221 }
 222
 223 /* Calculate interpolants for triangle and line rasterization.
 224  */
 225 static void
 226 brw_upload_ff_gs_prog(struct brw_context *brw)
 227 {
 228    struct brw_ff_gs_prog_key key;
 229    /* Populate the key:
 230     */
 231    populate_key(brw, &key);
 232
 233    if (brw->ff_gs.prog_active != key.need_gs_prog) {
 234       brw->state.dirty.cache |= CACHE_NEW_FF_GS_PROG;
 235       brw->ff_gs.prog_active = key.need_gs_prog;
 236    }
 237
 238    if (brw->ff_gs.prog_active) {
 239       if (!brw_search_cache(&brw->cache, BRW_FF_GS_PROG,
 240                             &key, sizeof(key),
 241                             &brw->ff_gs.prog_offset, &brw->ff_gs.prog_data)) {
 242          compile_ff_gs_prog( brw, &key );
 243       }
 244    }
 245 }
 246
 247
 248 const struct brw_tracked_state brw_ff_gs_prog = {
 249    .dirty = {
 250       .mesa  = (_NEW_LIGHT),
 251       .brw   = (BRW_NEW_PRIMITIVE |
 252                 BRW_NEW_TRANSFORM_FEEDBACK),
 253       .cache = CACHE_NEW_VS_PROG
 254    },
 255    .emit = brw_upload_ff_gs_prog
 256 };