src/mesa/pipe/draw/draw_vertex_shader.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 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   * Authors:
  30   *   Keith Whitwell <keith@tungstengraphics.com>
  31   *   Brian Paul
  32   */
  33
  34 #include "pipe/p_util.h"
  35 #include "draw_private.h"
  36 #include "draw_context.h"
  37 #include "draw_vertex.h"
  38
  39 #include "x86/rtasm/x86sse.h"
  40
  41 #include "pipe/tgsi/exec/tgsi_core.h"
  42
  43
  44 #define DBG 0
  45
  46
  47 static INLINE unsigned
  48 compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
  49 {
  50    unsigned mask = 0;
  51    unsigned i;
  52
  53    for (i = 0; i < nr; i++) {
  54       if (dot4(clip, plane[i]) < 0)
  55          mask |= (1<<i);
  56    }
  57
  58    return mask;
  59 }
  60
  61
  62 typedef void (XSTDCALL *codegen_function) (
  63    const struct tgsi_exec_vector *input,
  64    struct tgsi_exec_vector *output,
  65    float (*constant)[4],
  66    struct tgsi_exec_vector *temporary );
  67
  68
  69 /**
  70  * Transform vertices with the current vertex program/shader
  71  * Up to four vertices can be shaded at a time.
  72  * \param vbuffer  the input vertex data
  73  * \param elts  indexes of four input vertices
  74  * \param count  number of vertices to shade [1..4]
  75  * \param vOut  array of pointers to four output vertices
  76  */
  77 static void
  78 run_vertex_program(struct draw_context *draw,
  79                    unsigned elts[4], unsigned count,
  80                    struct vertex_header *vOut[])
  81 {
  82    struct tgsi_exec_machine *machine = &draw->machine;
  83    unsigned int j;
  84
  85    ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
  86    ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
  87    const float *scale = draw->viewport.scale;
  88    const float *trans = draw->viewport.translate;
  89
  90    assert(count <= 4);
  91    assert(draw->vertex_shader->state->output_semantic_name[0]
  92           == TGSI_SEMANTIC_POSITION);
  93
  94    /* Consts does not require 16 byte alignment. */
  95    machine->Consts = (float (*)[4]) draw->mapped_constants;
  96
  97    machine->Inputs = ALIGN16_ASSIGN(inputs);
  98    machine->Outputs = ALIGN16_ASSIGN(outputs);
  99
 100    draw_vertex_fetch( draw, machine, elts, count );
 101
 102    /* run shader */
 103 #if defined(__i386__) || defined(__386__)
 104    if (draw->use_sse) {
 105       /* SSE */
 106       codegen_function func = (codegen_function) x86_get_func( &draw->vertex_shader->sse2_program );
 107       func(
 108          machine->Inputs,
 109          machine->Outputs,
 110          machine->Consts,
 111          machine->Temps );
 112    }
 113    else
 114 #else
 115    {
 116       /* interpreter */
 117       tgsi_exec_machine_run( machine );
 118    }
 119 #endif
 120
 121    /* store machine results */
 122    for (j = 0; j < count; j++) {
 123       unsigned slot;
 124       float x, y, z, w;
 125
 126       /* Handle attr[0] (position) specially:
 127        *
 128        * XXX: Computing the clipmask should be done in the vertex
 129        * program as a set of DP4 instructions appended to the
 130        * user-provided code.
 131        */
 132       x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
 133       y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
 134       z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
 135       w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 136
 137       vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
 138       vOut[j]->edgeflag = 1;
 139
 140       /* divide by w */
 141       w = 1.0f / w;
 142       x *= w;
 143       y *= w;
 144       z *= w;
 145
 146       /* Viewport mapping */
 147       vOut[j]->data[0][0] = x * scale[0] + trans[0];
 148       vOut[j]->data[0][1] = y * scale[1] + trans[1];
 149       vOut[j]->data[0][2] = z * scale[2] + trans[2];
 150       vOut[j]->data[0][3] = w;
 151
 152 #if DBG
 153       printf("output[%d]win: %f %f %f %f\n", j,
 154              vOut[j]->data[0][0],
 155              vOut[j]->data[0][1],
 156              vOut[j]->data[0][2],
 157              vOut[j]->data[0][3]);
 158 #endif
 159       /* Remaining attributes are packed into sequential post-transform
 160        * vertex attrib slots.
 161        * Skip 0 since we just did it above.
 162        * Subtract two because of the VERTEX_HEADER, CLIP_POS attribs.
 163        */
 164       for (slot = 1; slot < draw->vertex_info.num_attribs - 2; slot++) {
 165          vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 166          vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 167          vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 168          vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 169 #if DBG
 170          printf("output[%d][%d]: %f %f %f %f\n", j, slot,
 171                 vOut[j]->data[slot][0],
 172                 vOut[j]->data[slot][1],
 173                 vOut[j]->data[slot][2],
 174                 vOut[j]->data[slot][3]);
 175 #endif
 176       }
 177    } /* loop over vertices */
 178 }
 179
 180
 181 /**
 182  * Called by the draw module when the vertx cache needs to be flushed.
 183  * This involves running the vertex shader.
 184  */
 185 void draw_vertex_shader_queue_flush( struct draw_context *draw )
 186 {
 187    unsigned i, j;
 188
 189 //   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
 190 #ifdef MESA_LLVM
 191    if (draw->vertex_shader->state->llvm_prog) {
 192       draw_vertex_shader_queue_flush_llvm(draw);
 193       return;
 194    }
 195 #endif
 196
 197    /* run vertex shader on vertex cache entries, four per invokation */
 198    for (i = 0; i < draw->vs.queue_nr; i += 4) {
 199       struct vertex_header *dests[4];
 200       unsigned elts[4];
 201       int n;
 202
 203       for (j = 0; j < 4; j++) {
 204          elts[j] = draw->vs.queue[i + j].elt;
 205          dests[j] = draw->vs.queue[i + j].dest;
 206       }
 207
 208       n = MIN2(4, draw->vs.queue_nr - i);
 209       assert(n > 0);
 210       assert(n <= 4);
 211
 212       run_vertex_program(draw, elts, n, dests);
 213    }
 214
 215    draw->vs.queue_nr = 0;
 216 }
 217
 218
 219 void *
 220 draw_create_vertex_shader(struct draw_context *draw,
 221                           const struct pipe_shader_state *shader)
 222 {
 223    struct draw_vertex_shader *vs;
 224
 225    vs = calloc( 1, sizeof( struct draw_vertex_shader ) );
 226    vs->state = shader;
 227
 228 #if defined(__i386__) || defined(__386__)
 229    if (draw->use_sse) {
 230       /* cast-away const */
 231       struct pipe_shader_state *sh = (struct pipe_shader_state *) shader;
 232
 233       x86_init_func( &vs->sse2_program );
 234       tgsi_emit_sse2( sh->tokens, &vs->sse2_program );
 235    }
 236 #endif
 237
 238    return vs;
 239 }
 240
 241 void draw_bind_vertex_shader(struct draw_context *draw,
 242                              void *vcso)
 243 {
 244    draw_flush(draw);
 245    draw->vertex_shader = (struct draw_vertex_shader*)(vcso);
 246
 247    /* specify the fragment program to interpret/execute */
 248    tgsi_exec_machine_init(&draw->machine,
 249                           draw->vertex_shader->state->tokens,
 250                           PIPE_MAX_SAMPLERS,
 251                           NULL /*samplers*/ );
 252 }
 253
 254 void draw_delete_vertex_shader(struct draw_context *draw,
 255                                void *vcso)
 256 {
 257    struct draw_vertex_shader *vs;
 258
 259    vs = (struct draw_vertex_shader *) vcso;
 260
 261 #if defined(__i386__) || defined(__386__)
 262    x86_release_func( (struct x86_function *) &vs->sse2_program );
 263 #endif
 264
 265    free( vs->state );
 266    free( vs );
 267 }