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 bool
  44 do_vs_prog(struct brw_context *brw,
  45            struct gl_shader_program *prog,
  46            struct brw_vertex_program *vp,
  47            struct brw_vs_prog_key *key)
  48 {
  49    struct gl_context *ctx = &brw->intel.ctx;
  50    struct intel_context *intel = &brw->intel;
  51    GLuint program_size;
  52    const GLuint *program;
  53    struct brw_vs_compile c;
  54    void *mem_ctx;
  55    int aux_size;
  56    int i;
  57
  58    memset(&c, 0, sizeof(c));
  59    memcpy(&c.key, key, sizeof(*key));
  60
  61    mem_ctx = ralloc_context(NULL);
  62
  63    brw_init_compile(brw, &c.func, mem_ctx);
  64    c.vp = vp;
  65
  66    c.prog_data.outputs_written = vp->program.Base.OutputsWritten;
  67    c.prog_data.inputs_read = vp->program.Base.InputsRead;
  68
  69    if (c.key.copy_edgeflag) {
  70       c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_EDGE);
  71       c.prog_data.inputs_read |= 1<<VERT_ATTRIB_EDGEFLAG;
  72    }
  73
  74    /* Put dummy slots into the VUE for the SF to put the replaced
  75     * point sprite coords in.  We shouldn't need these dummy slots,
  76     * which take up precious URB space, but it would mean that the SF
  77     * doesn't get nice aligned pairs of input coords into output
  78     * coords, which would be a pain to handle.
  79     */
  80    for (i = 0; i < 8; i++) {
  81       if (c.key.point_coord_replace & (1 << i))
  82          c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_TEX0 + i);
  83    }
  84
  85    if (0) {
  86       _mesa_fprint_program_opt(stdout, &c.vp->program.Base, PROG_PRINT_DEBUG,
  87                                GL_TRUE);
  88    }
  89
  90    /* Emit GEN4 code.
  91     */
  92    if (brw->new_vs_backend && prog) {
  93       if (!brw_vs_emit(prog, &c)) {
  94          ralloc_free(mem_ctx);
  95          return false;
  96       }
  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    return true;
 135 }
 136
 137
 138 static void brw_upload_vs_prog(struct brw_context *brw)
 139 {
 140    struct gl_context *ctx = &brw->intel.ctx;
 141    struct brw_vs_prog_key key;
 142    struct brw_vertex_program *vp =
 143       (struct brw_vertex_program *)brw->vertex_program;
 144    int i;
 145
 146    memset(&key, 0, sizeof(key));
 147
 148    /* Just upload the program verbatim for now.  Always send it all
 149     * the inputs it asks for, whether they are varying or not.
 150     */
 151    key.program_string_id = vp->id;
 152    key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
 153    key.copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 154                         ctx->Polygon.BackMode != GL_FILL);
 155    key.two_side_color = (ctx->Light.Enabled && ctx->Light.Model.TwoSide);
 156
 157    /* _NEW_LIGHT | _NEW_BUFFERS */
 158    key.clamp_vertex_color = ctx->Light._ClampVertexColor;
 159
 160    /* _NEW_POINT */
 161    if (ctx->Point.PointSprite) {
 162       for (i = 0; i < 8; i++) {
 163          if (ctx->Point.CoordReplace[i])
 164             key.point_coord_replace |= (1 << i);
 165       }
 166    }
 167
 168    /* BRW_NEW_VERTICES */
 169    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 170       if (vp->program.Base.InputsRead & (1 << i) &&
 171           brw->vb.inputs[i].glarray->Type == GL_FIXED) {
 172          key.gl_fixed_input_size[i] = brw->vb.inputs[i].glarray->Size;
 173       }
 174    }
 175
 176    if (!brw_search_cache(&brw->cache, BRW_VS_PROG,
 177                          &key, sizeof(key),
 178                          &brw->vs.prog_offset, &brw->vs.prog_data)) {
 179       bool success = do_vs_prog(brw, ctx->Shader.CurrentVertexProgram,
 180                                 vp, &key);
 181
 182       assert(success);
 183    }
 184    brw->vs.constant_map = ((int8_t *)brw->vs.prog_data +
 185                            sizeof(*brw->vs.prog_data));
 186 }
 187
 188 /* See brw_vs.c:
 189  */
 190 const struct brw_tracked_state brw_vs_prog = {
 191    .dirty = {
 192       .mesa  = (_NEW_TRANSFORM | _NEW_POLYGON | _NEW_POINT | _NEW_LIGHT |
 193                 _NEW_BUFFERS),
 194       .brw   = (BRW_NEW_VERTEX_PROGRAM |
 195                 BRW_NEW_VERTICES),
 196       .cache = 0
 197    },
 198    .prepare = brw_upload_vs_prog
 199 };
 200
 201 bool
 202 brw_vs_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
 203 {
 204    struct brw_context *brw = brw_context(ctx);
 205    struct brw_vs_prog_key key;
 206    struct gl_vertex_program *vp = prog->VertexProgram;
 207    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 208    uint32_t old_prog_offset = brw->vs.prog_offset;
 209    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 210    bool success;
 211
 212    if (!vp)
 213       return true;
 214
 215    memset(&key, 0, sizeof(key));
 216
 217    key.program_string_id = bvp->id;
 218    key.clamp_vertex_color = true;
 219
 220    success = do_vs_prog(brw, prog, bvp, &key);
 221
 222    brw->vs.prog_offset = old_prog_offset;
 223    brw->vs.prog_data = old_prog_data;
 224
 225    return success;
 226 }