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    static int new_vs = -1;
  58
  59    memset(&c, 0, sizeof(c));
  60    memcpy(&c.key, key, sizeof(*key));
  61
  62    mem_ctx = ralloc_context(NULL);
  63
  64    brw_init_compile(brw, &c.func, mem_ctx);
  65    c.vp = vp;
  66
  67    c.prog_data.outputs_written = vp->program.Base.OutputsWritten;
  68    c.prog_data.inputs_read = vp->program.Base.InputsRead;
  69
  70    if (c.key.copy_edgeflag) {
  71       c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_EDGE);
  72       c.prog_data.inputs_read |= 1<<VERT_ATTRIB_EDGEFLAG;
  73    }
  74
  75    /* Put dummy slots into the VUE for the SF to put the replaced
  76     * point sprite coords in.  We shouldn't need these dummy slots,
  77     * which take up precious URB space, but it would mean that the SF
  78     * doesn't get nice aligned pairs of input coords into output
  79     * coords, which would be a pain to handle.
  80     */
  81    for (i = 0; i < 8; i++) {
  82       if (c.key.point_coord_replace & (1 << i))
  83          c.prog_data.outputs_written |= BITFIELD64_BIT(VERT_RESULT_TEX0 + i);
  84    }
  85
  86    if (0) {
  87       _mesa_fprint_program_opt(stdout, &c.vp->program.Base, PROG_PRINT_DEBUG,
  88                                GL_TRUE);
  89    }
  90
  91    /* Emit GEN4 code.
  92     */
  93    if (new_vs == -1)
  94       new_vs = getenv("INTEL_NEW_VS") != NULL;
  95
  96    if (new_vs && prog) {
  97       if (!brw_vs_emit(prog, &c)) {
  98          ralloc_free(mem_ctx);
  99          return false;
 100       }
 101    } else {
 102       brw_old_vs_emit(&c);
 103    }
 104
 105    /* Scratch space is used for register spilling */
 106    if (c.last_scratch) {
 107       c.prog_data.total_scratch = brw_get_scratch_size(c.last_scratch);
 108
 109       brw_get_scratch_bo(intel, &brw->vs.scratch_bo,
 110                          c.prog_data.total_scratch * brw->vs_max_threads);
 111    }
 112
 113    /* get the program
 114     */
 115    program = brw_get_program(&c.func, &program_size);
 116
 117    /* We upload from &c.prog_data including the constant_map assuming
 118     * they're packed together.  It would be nice to have a
 119     * compile-time assert macro here.
 120     */
 121    assert(c.constant_map == (int8_t *)&c.prog_data +
 122           sizeof(c.prog_data));
 123    assert(ctx->Const.VertexProgram.MaxNativeParameters ==
 124           ARRAY_SIZE(c.constant_map));
 125    (void) ctx;
 126
 127    aux_size = sizeof(c.prog_data);
 128    /* constant_map */
 129    aux_size += c.vp->program.Base.Parameters->NumParameters;
 130
 131    brw_upload_cache(&brw->cache, BRW_VS_PROG,
 132                     &c.key, sizeof(c.key),
 133                     program, program_size,
 134                     &c.prog_data, aux_size,
 135                     &brw->vs.prog_offset, &brw->vs.prog_data);
 136    ralloc_free(mem_ctx);
 137
 138    return true;
 139 }
 140
 141
 142 static void brw_upload_vs_prog(struct brw_context *brw)
 143 {
 144    struct gl_context *ctx = &brw->intel.ctx;
 145    struct brw_vs_prog_key key;
 146    struct brw_vertex_program *vp =
 147       (struct brw_vertex_program *)brw->vertex_program;
 148    int i;
 149
 150    memset(&key, 0, sizeof(key));
 151
 152    /* Just upload the program verbatim for now.  Always send it all
 153     * the inputs it asks for, whether they are varying or not.
 154     */
 155    key.program_string_id = vp->id;
 156    key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
 157    key.copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 158                         ctx->Polygon.BackMode != GL_FILL);
 159    key.two_side_color = (ctx->Light.Enabled && ctx->Light.Model.TwoSide);
 160
 161    /* _NEW_LIGHT | _NEW_BUFFERS */
 162    key.clamp_vertex_color = ctx->Light._ClampVertexColor;
 163
 164    /* _NEW_POINT */
 165    if (ctx->Point.PointSprite) {
 166       for (i = 0; i < 8; i++) {
 167          if (ctx->Point.CoordReplace[i])
 168             key.point_coord_replace |= (1 << i);
 169       }
 170    }
 171
 172    /* BRW_NEW_VERTICES */
 173    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 174       if (vp->program.Base.InputsRead & (1 << i) &&
 175           brw->vb.inputs[i].glarray->Type == GL_FIXED) {
 176          key.gl_fixed_input_size[i] = brw->vb.inputs[i].glarray->Size;
 177       }
 178    }
 179
 180    if (!brw_search_cache(&brw->cache, BRW_VS_PROG,
 181                          &key, sizeof(key),
 182                          &brw->vs.prog_offset, &brw->vs.prog_data)) {
 183       bool success = do_vs_prog(brw, ctx->Shader.CurrentVertexProgram,
 184                                 vp, &key);
 185
 186       assert(success);
 187    }
 188    brw->vs.constant_map = ((int8_t *)brw->vs.prog_data +
 189                            sizeof(*brw->vs.prog_data));
 190 }
 191
 192 /* See brw_vs.c:
 193  */
 194 const struct brw_tracked_state brw_vs_prog = {
 195    .dirty = {
 196       .mesa  = (_NEW_TRANSFORM | _NEW_POLYGON | _NEW_POINT | _NEW_LIGHT |
 197                 _NEW_BUFFERS),
 198       .brw   = (BRW_NEW_VERTEX_PROGRAM |
 199                 BRW_NEW_VERTICES),
 200       .cache = 0
 201    },
 202    .prepare = brw_upload_vs_prog
 203 };
 204
 205 bool
 206 brw_vs_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
 207 {
 208    struct brw_context *brw = brw_context(ctx);
 209    struct brw_vs_prog_key key;
 210    struct gl_vertex_program *vp = prog->VertexProgram;
 211    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 212    uint32_t old_prog_offset = brw->vs.prog_offset;
 213    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 214    bool success;
 215
 216    if (!vp)
 217       return true;
 218
 219    memset(&key, 0, sizeof(key));
 220
 221    key.program_string_id = bvp->id;
 222    key.clamp_vertex_color = true;
 223
 224    success = do_vs_prog(brw, prog, bvp, &key);
 225
 226    brw->vs.prog_offset = old_prog_offset;
 227    brw->vs.prog_data = old_prog_data;
 228
 229    return success;
 230 }