src/mesa/state_tracker/st_atom_shader.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * 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, sub license, 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 portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 /**
  29  * State validation for vertex/fragment shaders.
  30  * Note that we have to delay most vertex/fragment shader translation
  31  * until rendering time since the linkage between the vertex outputs and
  32  * fragment inputs can vary depending on the pairing of shaders.
  33  *
  34  * Authors:
  35  *   Brian Paul
  36  */
  37
  38
  39
  40 #include "main/imports.h"
  41 #include "main/mtypes.h"
  42
  43 #include "pipe/p_context.h"
  44 #include "pipe/p_shader_tokens.h"
  45
  46 #include "pipe/cso_cache/cso_cache.h"
  47
  48 #include "st_context.h"
  49 #include "st_cache.h"
  50 #include "st_atom.h"
  51 #include "st_program.h"
  52 #include "st_atom_shader.h"
  53 #include "st_mesa_to_tgsi.h"
  54
  55
  56 /**
  57  * This represents a vertex program, especially translated to match
  58  * the inputs of a particular fragment shader.
  59  */
  60 struct translated_vertex_program
  61 {
  62    struct st_vertex_program *master;
  63
  64    /** The fragment shader "signature" this vertex shader is meant for: */
  65    GLbitfield frag_inputs;
  66
  67    /** Compared against master vertex program's serialNo: */
  68    GLuint serialNo;
  69
  70    /** Maps VERT_RESULT_x to slot */
  71    GLuint output_to_slot[VERT_RESULT_MAX];
  72
  73    /** The program in TGSI format */
  74    struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
  75
  76    /** Pointer to the translated vertex program */
  77    struct st_vertex_program *vp;
  78
  79    struct translated_vertex_program *next;  /**< next in linked list */
  80 };
  81
  82
  83
  84 /**
  85  * Free data hanging off the st vert prog.
  86  */
  87 void
  88 st_remove_vertex_program(struct st_context *st, struct st_vertex_program *stvp)
  89 {
  90    /* no-op, for now? */
  91 }
  92
  93
  94 /**
  95  * Free data hanging off the st frag prog.
  96  */
  97 void
  98 st_remove_fragment_program(struct st_context *st,
  99                           struct st_fragment_program *stfp)
 100 {
 101    struct translated_vertex_program *xvp, *next;
 102
 103    for (xvp = stfp->vertex_programs; xvp; xvp = next) {
 104       next = xvp->next;
 105       /* XXX free xvp->vs */
 106       free(xvp);
 107    }
 108 }
 109
 110
 111
 112 /**
 113  * Given a vertex program output attribute, return the corresponding
 114  * fragment program input attribute.
 115  * \return -1 for vertex outputs that have no corresponding fragment input
 116  */
 117 static GLint
 118 vp_out_to_fp_in(GLuint vertResult)
 119 {
 120    if (vertResult >= VERT_RESULT_TEX0 &&
 121        vertResult < VERT_RESULT_TEX0 + MAX_TEXTURE_COORD_UNITS)
 122       return FRAG_ATTRIB_TEX0 + (vertResult - VERT_RESULT_TEX0);
 123
 124    if (vertResult >= VERT_RESULT_VAR0 &&
 125        vertResult < VERT_RESULT_VAR0 + MAX_VARYING)
 126       return FRAG_ATTRIB_VAR0 + (vertResult - VERT_RESULT_VAR0);
 127
 128    switch (vertResult) {
 129    case VERT_RESULT_HPOS:
 130       return FRAG_ATTRIB_WPOS;
 131    case VERT_RESULT_COL0:
 132       return FRAG_ATTRIB_COL0;
 133    case VERT_RESULT_COL1:
 134       return FRAG_ATTRIB_COL1;
 135    case VERT_RESULT_FOGC:
 136       return FRAG_ATTRIB_FOGC;
 137    default:
 138       /* Back-face colors, edge flags, etc */
 139       return -1;
 140    }
 141 }
 142
 143
 144 /**
 145  * Find a translated vertex program that corresponds to stvp and
 146  * has outputs matched to stfp's inputs.
 147  * This performs vertex and fragment translation (to TGSI) when needed.
 148  */
 149 static struct translated_vertex_program *
 150 find_translated_vp(struct st_context *st,
 151                    struct st_vertex_program *stvp,
 152                    struct st_fragment_program *stfp)
 153 {
 154    static const GLuint UNUSED = ~0;
 155    struct translated_vertex_program *xvp;
 156    const GLbitfield fragInputsRead = stfp->Base.Base.InputsRead;
 157
 158    /*
 159     * Translate fragment program if needed.
 160     */
 161    if (!stfp->fs) {
 162       GLuint inAttr, numIn = 0;
 163
 164       for (inAttr = 0; inAttr < FRAG_ATTRIB_MAX; inAttr++) {
 165          if (fragInputsRead & (1 << inAttr)) {
 166             stfp->input_to_slot[inAttr] = numIn;
 167             numIn++;
 168          }
 169          else {
 170             stfp->input_to_slot[inAttr] = UNUSED;
 171          }
 172       }
 173
 174       stfp->num_input_slots = numIn;
 175
 176       assert(stfp->Base.Base.NumInstructions > 1);
 177
 178       (void) st_translate_fragment_program(st, stfp,
 179                                            stfp->input_to_slot,
 180                                            stfp->tokens,
 181                                            ST_MAX_SHADER_TOKENS);
 182       assert(stfp->fs);
 183    }
 184
 185
 186    /* See if we've got a translated vertex program whose outputs match
 187     * the fragment program's inputs.
 188     * XXX This could be a hash lookup, using InputsRead as the key.
 189     */
 190    for (xvp = stfp->vertex_programs; xvp; xvp = xvp->next) {
 191       if (xvp->master == stvp && xvp->frag_inputs == fragInputsRead) {
 192          break;
 193       }
 194    }
 195
 196    /* No?  Allocate translated vp object now */
 197    if (!xvp) {
 198       xvp = CALLOC_STRUCT(translated_vertex_program);
 199       xvp->frag_inputs = fragInputsRead;
 200       xvp->master = stvp;
 201
 202       xvp->next = stfp->vertex_programs;
 203       stfp->vertex_programs = xvp;
 204    }
 205
 206    /* See if we need to translate vertex program to TGSI form */
 207    if (xvp->serialNo != stvp->serialNo) {
 208       GLuint outAttr, dummySlot;
 209       const GLbitfield outputsWritten = stvp->Base.Base.OutputsWritten;
 210       GLuint numVpOuts = 0;
 211
 212       /* Compute mapping of vertex program outputs to slots, which depends
 213        * on the fragment program's input->slot mapping.
 214        */
 215       for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
 216          /* set default: */
 217          xvp->output_to_slot[outAttr] = UNUSED;
 218
 219          if (outAttr == VERT_RESULT_HPOS) {
 220             /* always put xformed position into slot zero */
 221             xvp->output_to_slot[VERT_RESULT_HPOS] = 0;
 222             numVpOuts++;
 223          }
 224          else if (outputsWritten & (1 << outAttr)) {
 225             /* see if the frag prog wants this vert output */
 226             GLint fpInAttrib = vp_out_to_fp_in(outAttr);
 227             if (fpInAttrib >= 0) {
 228                GLuint fpInSlot = stfp->input_to_slot[fpInAttrib];
 229                GLuint vpOutSlot = stfp->fs->state.input_map[fpInSlot];
 230                xvp->output_to_slot[outAttr] = vpOutSlot;
 231                numVpOuts++;
 232             }
 233             else if (outAttr == VERT_RESULT_PSIZ ||
 234                      outAttr == VERT_RESULT_BFC0 ||
 235                      outAttr == VERT_RESULT_BFC1) {
 236                /* backface colors go into last slots */
 237                xvp->output_to_slot[outAttr] = numVpOuts++;
 238             }
 239          }
 240          /*
 241          printf("output_to_slot[%d] = %d\n", outAttr,
 242                 xvp->output_to_slot[outAttr]);
 243          */
 244       }
 245
 246       /* Unneeded vertex program outputs will go to this slot.
 247        * We could use this info to do dead code elimination in the
 248        * vertex program.
 249        */
 250       dummySlot = stfp->num_input_slots;
 251
 252       /* Map vert program outputs that aren't used to the dummy slot */
 253       for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
 254          if (outputsWritten & (1 << outAttr)) {
 255             if (xvp->output_to_slot[outAttr] == UNUSED)
 256                xvp->output_to_slot[outAttr] = dummySlot;
 257          }
 258       }
 259
 260       assert(stvp->Base.Base.NumInstructions > 1);
 261
 262       st_translate_vertex_program(st, stvp,
 263                                   xvp->output_to_slot,
 264                                   xvp->tokens,
 265                                   ST_MAX_SHADER_TOKENS);
 266
 267       assert(stvp->cso);
 268       xvp->vp = stvp;
 269
 270       /* translated VP is up to date now */
 271       xvp->serialNo = stvp->serialNo;
 272    }
 273
 274    return xvp;
 275 }
 276
 277
 278 static void
 279 update_linkage( struct st_context *st )
 280 {
 281    struct st_vertex_program *stvp;
 282    struct st_fragment_program *stfp;
 283    struct translated_vertex_program *xvp;
 284
 285    /* find active shader and params -- Should be covered by
 286     * ST_NEW_VERTEX_PROGRAM
 287     */
 288    assert(st->ctx->VertexProgram._Current);
 289    stvp = st_vertex_program(st->ctx->VertexProgram._Current);
 290
 291    assert(st->ctx->FragmentProgram._Current);
 292    stfp = st_fragment_program(st->ctx->FragmentProgram._Current);
 293
 294    xvp = find_translated_vp(st, stvp, stfp);
 295
 296    st->vp = stvp;
 297    st->state.vs = xvp->vp;
 298    st->pipe->bind_vs_state(st->pipe, st->state.vs->cso->data);
 299
 300    st->fp = stfp;
 301    st->state.fs = stfp->fs;
 302    st->pipe->bind_fs_state(st->pipe, st->state.fs->data);
 303
 304    st->vertex_result_to_slot = xvp->output_to_slot;
 305 }
 306
 307
 308 const struct st_tracked_state st_update_shader = {
 309    .name = "st_update_shader",
 310    .dirty = {
 311       .mesa  = 0,
 312       .st   = ST_NEW_VERTEX_PROGRAM | ST_NEW_FRAGMENT_PROGRAM
 313    },
 314    .update = update_linkage
 315 };