src/mesa/drivers/dri/i965/brw_vs.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/compiler.h"
  34 #include "brw_context.h"
  35 #include "brw_vs.h"
  36 #include "brw_util.h"
  37 #include "brw_state.h"
  38 #include "program/prog_print.h"
  39 #include "program/prog_parameter.h"
  40
  41 #include "../glsl/ralloc.h"
  42
  43 static void do_vs_prog( struct brw_context *brw,
  44                         struct brw_vertex_program *vp,
  45                         struct brw_vs_prog_key *key )
  46 {
  47    struct gl_context *ctx = &brw->intel.ctx;
  48    struct intel_context *intel = &brw->intel;
  49    GLuint program_size;
  50    const GLuint *program;
  51    struct brw_vs_compile c;
  52    void *mem_ctx;
  53    int aux_size;
  54    int i;
  55    static int new_vs = -1;
  56
  57    memset(&c, 0, sizeof(c));
  58    memcpy(&c.key, key, sizeof(*key));
  59
  60    mem_ctx = ralloc_context(NULL);
  61
  62    brw_init_compile(brw, &c.func, mem_ctx);
  63    c.vp = vp;
  64
  65    c.prog_data.outputs_written = vp->program.Base.OutputsWritten;
  66    c.prog_data.inputs_read = vp->program.Base.InputsRead;
  67
  68    if (c.key.copy_edgeflag) {
  69       c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_EDGE);
  70       c.prog_data.inputs_read |= 1<<VERT_ATTRIB_EDGEFLAG;
  71    }
  72
  73    /* Put dummy slots into the VUE for the SF to put the replaced
  74     * point sprite coords in.  We shouldn't need these dummy slots,
  75     * which take up precious URB space, but it would mean that the SF
  76     * doesn't get nice aligned pairs of input coords into output
  77     * coords, which would be a pain to handle.
  78     */
  79    for (i = 0; i < 8; i++) {
  80       if (c.key.point_coord_replace & (1 << i))
  81          c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_TEX0 + i);
  82    }
  83
  84    if (0) {
  85       _mesa_fprint_program_opt(stdout, &c.vp->program.Base, PROG_PRINT_DEBUG,
  86                                GL_TRUE);
  87    }
  88
  89    /* Emit GEN4 code.
  90     */
  91    if (new_vs == -1)
  92       new_vs = getenv("INTEL_NEW_VS") != NULL;
  93
  94    if (new_vs) {
  95       if (!brw_vs_emit(&c))
  96          brw_old_vs_emit(&c);
  97    } else {
  98       brw_old_vs_emit(&c);
  99    }
 100
 101    /* Scratch space is used for register spilling */
 102    if (c.last_scratch) {
 103       c.prog_data.total_scratch = brw_get_scratch_size(c.last_scratch);
 104
 105       brw_get_scratch_bo(intel, &brw->vs.scratch_bo,
 106                          c.prog_data.total_scratch * brw->vs_max_threads);
 107    }
 108
 109    /* get the program
 110     */
 111    program = brw_get_program(&c.func, &program_size);
 112
 113    /* We upload from &c.prog_data including the constant_map assuming
 114     * they're packed together.  It would be nice to have a
 115     * compile-time assert macro here.
 116     */
 117    assert(c.constant_map == (int8_t *)&c.prog_data +
 118           sizeof(c.prog_data));
 119    assert(ctx->Const.VertexProgram.MaxNativeParameters ==
 120           ARRAY_SIZE(c.constant_map));
 121    (void) ctx;
 122
 123    aux_size = sizeof(c.prog_data);
 124    /* constant_map */
 125    aux_size += c.vp->program.Base.Parameters->NumParameters;
 126
 127    brw_upload_cache(&brw->cache, BRW_VS_PROG,
 128                     &c.key, sizeof(c.key),
 129                     program, program_size,
 130                     &c.prog_data, aux_size,
 131                     &brw->vs.prog_offset, &brw->vs.prog_data);
 132    ralloc_free(mem_ctx);
 133 }
 134
 135
 136 static void brw_upload_vs_prog(struct brw_context *brw)
 137 {
 138    struct gl_context *ctx = &brw->intel.ctx;
 139    struct brw_vs_prog_key key;
 140    struct brw_vertex_program *vp =
 141       (struct brw_vertex_program *)brw->vertex_program;
 142    int i;
 143
 144    memset(&key, 0, sizeof(key));
 145
 146    /* Just upload the program verbatim for now.  Always send it all
 147     * the inputs it asks for, whether they are varying or not.
 148     */
 149    key.program_string_id = vp->id;
 150    key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
 151    key.copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 152                         ctx->Polygon.BackMode != GL_FILL);
 153    key.two_side_color = (ctx->Light.Enabled && ctx->Light.Model.TwoSide);
 154
 155    /* _NEW_LIGHT | _NEW_BUFFERS */
 156    key.clamp_vertex_color = ctx->Light._ClampVertexColor;
 157
 158    /* _NEW_POINT */
 159    if (ctx->Point.PointSprite) {
 160       for (i = 0; i < 8; i++) {
 161          if (ctx->Point.CoordReplace[i])
 162             key.point_coord_replace |= (1 << i);
 163       }
 164    }
 165
 166    /* BRW_NEW_VERTICES */
 167    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 168       if (vp->program.Base.InputsRead & (1 << i) &&
 169           brw->vb.inputs[i].glarray->Type == GL_FIXED) {
 170          key.gl_fixed_input_size[i] = brw->vb.inputs[i].glarray->Size;
 171       }
 172    }
 173
 174    if (!brw_search_cache(&brw->cache, BRW_VS_PROG,
 175                          &key, sizeof(key),
 176                          &brw->vs.prog_offset, &brw->vs.prog_data)) {
 177       do_vs_prog(brw, vp, &key);
 178    }
 179    brw->vs.constant_map = ((int8_t *)brw->vs.prog_data +
 180                            sizeof(*brw->vs.prog_data));
 181 }
 182
 183
 184 /* See brw_vs.c:
 185  */
 186 const struct brw_tracked_state brw_vs_prog = {
 187    .dirty = {
 188       .mesa  = (_NEW_TRANSFORM | _NEW_POLYGON | _NEW_POINT | _NEW_LIGHT |
 189                 _NEW_BUFFERS),
 190       .brw   = (BRW_NEW_VERTEX_PROGRAM |
 191                 BRW_NEW_VERTICES),
 192       .cache = 0
 193    },
 194    .prepare = brw_upload_vs_prog
 195 };