src/mesa/pipe/softpipe/sp_quad_fs.c

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  6.5
   4  *
   5  * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  11  * and/or sell copies of the Software, and to permit persons to whom the
  12  * Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included
  15  * in all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  21  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  22  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25 /* Vertices are just an array of floats, with all the attributes
  26  * packed.  We currently assume a layout like:
  27  *
  28  * attr[0][0..3] - window position
  29  * attr[1..n][0..3] - remaining attributes.
  30  *
  31  * Attributes are assumed to be 4 floats wide but are packed so that
  32  * all the enabled attributes run contiguously.
  33  */
  34
  35 #include "glheader.h"
  36 #include "imports.h"
  37 #include "sp_context.h"
  38 #include "sp_headers.h"
  39 #include "sp_quad.h"
  40 #include "core/tgsi_core.h"
  41
  42 #if defined __GNUC__
  43 #define ALIGNED_ATTRIBS 1
  44 #else
  45 #define ALIGNED_ATTRIBS 0
  46 #endif
  47
  48 struct exec_machine {
  49    const struct setup_coefficient *coef; /**< will point to quad->coef */
  50
  51 #if ALIGNED_ATTRIBS
  52    GLfloat attr[FRAG_ATTRIB_MAX][NUM_CHANNELS][QUAD_SIZE] __attribute__(( aligned( 16 ) ));
  53 #else
  54    GLfloat attr[FRAG_ATTRIB_MAX][NUM_CHANNELS][QUAD_SIZE];
  55 #endif
  56 };
  57
  58
  59 /**
  60  * Compute quad's attributes values, as constants (GL_FLAT shading).
  61  */
  62 static INLINE void cinterp( struct exec_machine *exec,
  63                             GLuint attrib,
  64                             GLuint i )
  65 {
  66    GLuint j;
  67
  68    for (j = 0; j < QUAD_SIZE; j++) {
  69       exec->attr[attrib][i][j] = exec->coef[attrib].a0[i];
  70    }
  71 }
  72
  73
  74 /**
  75  * Compute quad's attribute values by linear interpolation.
  76  *
  77  * Push into the fp:
  78  *
  79  *   INPUT[attr] = MAD COEF_A0[attr], COEF_DADX[attr], INPUT_WPOS.xxxx
  80  *   INPUT[attr] = MAD INPUT[attr],   COEF_DADY[attr], INPUT_WPOS.yyyy
  81  */
  82 static INLINE void linterp( struct exec_machine *exec,
  83                             GLuint attrib,
  84                             GLuint i )
  85 {
  86    GLuint j;
  87
  88    for (j = 0; j < QUAD_SIZE; j++) {
  89       const GLfloat x = exec->attr[FRAG_ATTRIB_WPOS][0][j];
  90       const GLfloat y = exec->attr[FRAG_ATTRIB_WPOS][1][j];
  91       exec->attr[attrib][i][j] = (exec->coef[attrib].a0[i] +
  92                                   exec->coef[attrib].dadx[i] * x +
  93                                   exec->coef[attrib].dady[i] * y);
  94    }
  95 }
  96
  97
  98 /**
  99  * Compute quad's attribute values by linear interpolation with
 100  * perspective correction.
 101  *
 102  * Push into the fp:
 103  *
 104  *   INPUT[attr] = MAD COEF_DADX[attr], INPUT_WPOS.xxxx, COEF_A0[attr]
 105  *   INPUT[attr] = MAD COEF_DADY[attr], INPUT_WPOS.yyyy, INPUT[attr]
 106  *   TMP         = RCP INPUT_WPOS.w
 107  *   INPUT[attr] = MUL INPUT[attr], TMP.xxxx
 108  *
 109  */
 110 static INLINE void pinterp( struct exec_machine *exec,
 111                             GLuint attrib,
 112                             GLuint i )
 113 {
 114    GLuint j;
 115
 116    for (j = 0; j < QUAD_SIZE; j++) {
 117       const GLfloat x = exec->attr[FRAG_ATTRIB_WPOS][0][j];
 118       const GLfloat y = exec->attr[FRAG_ATTRIB_WPOS][1][j];
 119       /* FRAG_ATTRIB_WPOS.w here is really 1/w */
 120       const GLfloat w = 1.0 / exec->attr[FRAG_ATTRIB_WPOS][3][j];
 121       exec->attr[attrib][i][j] = ((exec->coef[attrib].a0[i] +
 122                                    exec->coef[attrib].dadx[i] * x +
 123                                    exec->coef[attrib].dady[i] * y) * w);
 124    }
 125 }
 126
 127
 128
 129 /* This should be done by the fragment shader execution unit (code
 130  * generated from the decl instructions).  Do it here for now.
 131  */
 132 static void
 133 shade_quad( struct quad_stage *qs, struct quad_header *quad )
 134 {
 135    const struct softpipe_context *softpipe = qs->softpipe;
 136    struct exec_machine exec;
 137    const GLfloat fx = quad->x0;
 138    const GLfloat fy = quad->y0;
 139    GLuint attr, i;
 140
 141    exec.coef = quad->coef;
 142
 143    /* Position:
 144     */
 145    exec.attr[FRAG_ATTRIB_WPOS][0][0] = fx;
 146    exec.attr[FRAG_ATTRIB_WPOS][0][1] = fx + 1.0;
 147    exec.attr[FRAG_ATTRIB_WPOS][0][2] = fx;
 148    exec.attr[FRAG_ATTRIB_WPOS][0][3] = fx + 1.0;
 149
 150    exec.attr[FRAG_ATTRIB_WPOS][1][0] = fy;
 151    exec.attr[FRAG_ATTRIB_WPOS][1][1] = fy;
 152    exec.attr[FRAG_ATTRIB_WPOS][1][2] = fy + 1.0;
 153    exec.attr[FRAG_ATTRIB_WPOS][1][3] = fy + 1.0;
 154
 155    /* Z and W are done by linear interpolation */
 156    if (softpipe->need_z) {
 157       linterp(&exec, 0, 2);   /* attr[0].z */
 158    }
 159
 160    if (softpipe->need_w) {
 161       linterp(&exec, 0, 3);  /* attr[0].w */
 162       /*invert(&exec, 0, 3);*/
 163    }
 164
 165    /* Interpolate all the remaining attributes.  This will get pushed
 166     * into the fragment program's responsibilities at some point.
 167     * Start at 1 to skip fragment position attribute (computed above).
 168     */
 169    for (attr = 1; attr < quad->nr_attrs; attr++) {
 170       switch (softpipe->interp[attr]) {
 171       case INTERP_CONSTANT:
 172          for (i = 0; i < NUM_CHANNELS; i++)
 173             cinterp(&exec, attr, i);
 174          break;
 175
 176       case INTERP_LINEAR:
 177          for (i = 0; i < NUM_CHANNELS; i++)
 178             linterp(&exec, attr, i);
 179          break;
 180
 181       case INTERP_PERSPECTIVE:
 182          for (i = 0; i < NUM_CHANNELS; i++)
 183             pinterp(&exec, attr, i);
 184          break;
 185       }
 186    }
 187
 188 #if 1
 189    /*softpipe->run_fs( tri->fp, quad, &tri->outputs );*/
 190
 191    {
 192       struct tgsi_exec_machine machine;
 193       struct tgsi_exec_vector outputs[FRAG_ATTRIB_MAX + 1];
 194       struct tgsi_exec_vector *aoutputs;
 195       GLuint i;
 196
 197 #if !ALIGNED_ATTRIBS
 198       struct tgsi_exec_vector inputs[FRAG_ATTRIB_MAX + 1];
 199       struct tgsi_exec_vector *ainputs;
 200 #endif
 201
 202 #ifdef DEBUG
 203       memset(&machine, 0, sizeof(machine));
 204 #endif
 205
 206       /* init machine state */
 207       tgsi_exec_machine_init(
 208          &machine,
 209          softpipe->fs.tokens );
 210
 211       /* Consts does not require 16 byte alignment. */
 212       machine.Consts = softpipe->fs.constants->constant;
 213
 214       aoutputs = (struct tgsi_exec_vector *) tgsi_align_128bit( outputs );
 215       machine.Outputs = aoutputs;
 216
 217       assert( sizeof( struct tgsi_exec_vector ) == sizeof( exec.attr[0] ) );
 218
 219 #if ALIGNED_ATTRIBS
 220       machine.Inputs = (struct tgsi_exec_vector *) exec.attr;
 221
 222       for (i = 0; i < softpipe->nr_attrs; i++) {
 223          /* Make sure fp_attr_to_slot[] is an identity transform. */
 224          assert( softpipe->fp_attr_to_slot[i] == i );
 225       }
 226 #else
 227       ainputs = (struct tgsi_exec_vector *) tgsi_align_128bit( inputs );
 228       machine.Inputs = ainputs;
 229
 230       /* load input registers */
 231       for (i = 0; i < softpipe->nr_attrs; i++) {
 232          /* Make sure fp_attr_to_slot[] is an identity transform. */
 233          assert( softpipe->fp_attr_to_slot[i] == i );
 234
 235          memcpy(
 236             &ainputs[i],
 237             exec.attr[i],
 238             sizeof( ainputs[0] ) );
 239       }
 240 #endif
 241
 242       /* run shader */
 243       tgsi_exec_machine_run( &machine );
 244
 245       /* store result color */
 246       memcpy(quad->outputs.color,
 247              &aoutputs[FRAG_ATTRIB_COL0].xyzw[0].f[0],
 248              sizeof(quad->outputs.color));
 249       if (softpipe->need_z) {
 250          /* XXX temporary */
 251          quad->outputs.depth[0] = exec.attr[0][2][0];
 252          quad->outputs.depth[1] = exec.attr[0][2][1];
 253          quad->outputs.depth[2] = exec.attr[0][2][2];
 254          quad->outputs.depth[3] = exec.attr[0][2][3];
 255       }
 256    }
 257 #else
 258    {
 259       GLuint attr = softpipe->fp_attr_to_slot[FRAG_ATTRIB_COL0];
 260       assert(attr);
 261
 262       memcpy(quad->outputs.color,
 263              exec.attr[attr],
 264              sizeof(quad->outputs.color));
 265
 266       if (softpipe->need_z) {
 267          quad->outputs.depth[0] = exec.attr[0][2][0];
 268          quad->outputs.depth[1] = exec.attr[0][2][1];
 269          quad->outputs.depth[2] = exec.attr[0][2][2];
 270          quad->outputs.depth[3] = exec.attr[0][2][3];
 271       }
 272    }
 273 #endif
 274
 275    /* shader may cull fragments */
 276    if (quad->mask)
 277       qs->next->run(qs->next, quad);
 278 }
 279
 280
 281
 282 struct quad_stage *sp_quad_shade_stage( struct softpipe_context *softpipe )
 283 {
 284    struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
 285
 286    stage->softpipe = softpipe;
 287    stage->run = shade_quad;
 288
 289    return stage;
 290 }