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 "pipe/p_shader_tokens.h"
  36 #if defined(__i386__) || defined(__386__)
  37 #include "pipe/tgsi/exec/tgsi_sse2.h"
  38 #endif
  39 #include "draw_private.h"
  40 #include "draw_context.h"
  41 #include "draw_vertex.h"
  42
  43 #include "x86/rtasm/x86sse.h"
  44 #include "pipe/llvm/gallivm.h"
  45
  46
  47 #define DBG_VS 0
  48
  49
  50 static INLINE unsigned
  51 compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
  52 {
  53    unsigned mask = 0;
  54    unsigned i;
  55
  56    /* Do the hardwired planes first:
  57     */
  58    if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
  59    if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
  60    if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
  61    if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
  62    if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
  63    if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
  64
  65    /* Followed by any remaining ones:
  66     */
  67    for (i = 6; i < nr; i++) {
  68       if (dot4(clip, plane[i]) < 0)
  69          mask |= (1<<i);
  70    }
  71
  72    return mask;
  73 }
  74
  75
  76 typedef void (XSTDCALL *codegen_function) (
  77    const struct tgsi_exec_vector *input,
  78    struct tgsi_exec_vector *output,
  79    float (*constant)[4],
  80    struct tgsi_exec_vector *temporary );
  81
  82
  83 /**
  84  * Transform vertices with the current vertex program/shader
  85  * Up to four vertices can be shaded at a time.
  86  * \param vbuffer  the input vertex data
  87  * \param elts  indexes of four input vertices
  88  * \param count  number of vertices to shade [1..4]
  89  * \param vOut  array of pointers to four output vertices
  90  */
  91 static void
  92 run_vertex_program(struct draw_context *draw,
  93                    unsigned elts[4], unsigned count,
  94                    struct vertex_header *vOut[])
  95 {
  96    struct tgsi_exec_machine *machine = &draw->machine;
  97    unsigned int j;
  98
  99    ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
 100    ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
 101    const float *scale = draw->viewport.scale;
 102    const float *trans = draw->viewport.translate;
 103
 104    assert(count <= 4);
 105    assert(draw->vertex_shader->state->output_semantic_name[0]
 106           == TGSI_SEMANTIC_POSITION);
 107
 108    /* Consts does not require 16 byte alignment. */
 109    machine->Consts = (float (*)[4]) draw->user.constants;
 110
 111    machine->Inputs = ALIGN16_ASSIGN(inputs);
 112    machine->Outputs = ALIGN16_ASSIGN(outputs);
 113
 114    draw_vertex_fetch( draw, machine, elts, count );
 115
 116    /* run shader */
 117 #if defined(__i386__) || defined(__386__)
 118    if (draw->use_sse) {
 119       /* SSE */
 120       /* cast away const */
 121       struct draw_vertex_shader *shader
 122          = (struct draw_vertex_shader *)draw->vertex_shader;
 123       codegen_function func
 124          = (codegen_function) x86_get_func( &shader->sse2_program );
 125       func(
 126          machine->Inputs,
 127          machine->Outputs,
 128          machine->Consts,
 129          machine->Temps );
 130    }
 131    else
 132 #endif
 133    {
 134       /* interpreter */
 135       tgsi_exec_machine_run( machine );
 136    }
 137
 138    /* store machine results */
 139    for (j = 0; j < count; j++) {
 140       unsigned slot;
 141       float x, y, z, w;
 142
 143       /* Handle attr[0] (position) specially:
 144        *
 145        * XXX: Computing the clipmask should be done in the vertex
 146        * program as a set of DP4 instructions appended to the
 147        * user-provided code.
 148        */
 149       x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
 150       y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
 151       z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
 152       w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 153
 154       vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
 155       vOut[j]->edgeflag = 1;
 156
 157       /* divide by w */
 158       w = 1.0f / w;
 159       x *= w;
 160       y *= w;
 161       z *= w;
 162
 163       /* Viewport mapping */
 164       vOut[j]->data[0][0] = x * scale[0] + trans[0];
 165       vOut[j]->data[0][1] = y * scale[1] + trans[1];
 166       vOut[j]->data[0][2] = z * scale[2] + trans[2];
 167       vOut[j]->data[0][3] = w;
 168
 169 #if DBG_VS
 170       printf("output[%d]win: %f %f %f %f\n", j,
 171              vOut[j]->data[0][0],
 172              vOut[j]->data[0][1],
 173              vOut[j]->data[0][2],
 174              vOut[j]->data[0][3]);
 175 #endif
 176       /* Remaining attributes are packed into sequential post-transform
 177        * vertex attrib slots.
 178        */
 179       for (slot = 1; slot < draw->vertex_info.num_attribs; slot++) {
 180          vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 181          vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 182          vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 183          vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 184 #if DBG_VS
 185          printf("output[%d][%d]: %f %f %f %f\n", j, slot,
 186                 vOut[j]->data[slot][0],
 187                 vOut[j]->data[slot][1],
 188                 vOut[j]->data[slot][2],
 189                 vOut[j]->data[slot][3]);
 190 #endif
 191       }
 192    } /* loop over vertices */
 193 }
 194
 195
 196 /**
 197  * Run the vertex shader on all vertices in the vertex queue.
 198  * Called by the draw module when the vertx cache needs to be flushed.
 199  */
 200 void
 201 draw_vertex_shader_queue_flush(struct draw_context *draw)
 202 {
 203    unsigned i, j;
 204
 205 //   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
 206 #ifdef MESA_LLVM
 207    if (draw->vertex_shader->llvm_prog) {
 208       draw_vertex_shader_queue_flush_llvm(draw);
 209       return;
 210    }
 211 #endif
 212
 213    /* run vertex shader on vertex cache entries, four per invokation */
 214    for (i = 0; i < draw->vs.queue_nr; i += 4) {
 215       struct vertex_header *dests[4];
 216       unsigned elts[4];
 217       int n;
 218
 219       for (j = 0; j < 4; j++) {
 220          elts[j] = draw->vs.queue[i + j].elt;
 221          dests[j] = draw->vs.queue[i + j].dest;
 222       }
 223
 224       n = MIN2(4, draw->vs.queue_nr - i);
 225       assert(n > 0);
 226       assert(n <= 4);
 227
 228       run_vertex_program(draw, elts, n, dests);
 229    }
 230
 231    draw->vs.queue_nr = 0;
 232 }
 233
 234
 235 struct draw_vertex_shader *
 236 draw_create_vertex_shader(struct draw_context *draw,
 237                           const struct pipe_shader_state *shader)
 238 {
 239    struct draw_vertex_shader *vs;
 240
 241    vs = CALLOC_STRUCT( draw_vertex_shader );
 242    if (vs == NULL) {
 243       return NULL;
 244    }
 245
 246    vs->state = shader;
 247
 248 #if defined(__i386__) || defined(__386__)
 249    if (draw->use_sse) {
 250       /* cast-away const */
 251       struct pipe_shader_state *sh = (struct pipe_shader_state *) shader;
 252
 253       x86_init_func( &vs->sse2_program );
 254       tgsi_emit_sse2( (struct tgsi_token *) sh->tokens, &vs->sse2_program );
 255    }
 256 #endif
 257 #ifdef MESA_LLVM
 258    vs->llvm_prog = gallivm_from_tgsi(shader->tokens, GALLIVM_VS);
 259    draw->engine = gallivm_global_cpu_engine();
 260    if (!draw->engine) {
 261       draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
 262    }
 263    else {
 264       gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
 265    }
 266 #endif
 267
 268    return vs;
 269 }
 270
 271
 272 void
 273 draw_bind_vertex_shader(struct draw_context *draw,
 274                         struct draw_vertex_shader *dvs)
 275 {
 276    draw_flush(draw);
 277    draw->vertex_shader = dvs;
 278
 279    /* specify the fragment program to interpret/execute */
 280    tgsi_exec_machine_init(&draw->machine,
 281                           draw->vertex_shader->state->tokens,
 282                           PIPE_MAX_SAMPLERS,
 283                           NULL /*samplers*/ );
 284 }
 285
 286
 287 void
 288 draw_delete_vertex_shader(struct draw_context *draw,
 289                           struct draw_vertex_shader *dvs)
 290 {
 291 #if defined(__i386__) || defined(__386__)
 292    x86_release_func( (struct x86_function *) &dvs->sse2_program );
 293 #endif
 294
 295    FREE( dvs );
 296 }