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
  42 #include "x86/rtasm/x86sse.h"
  43 #include "pipe/llvm/gallivm.h"
  44
  45
  46 #define DBG_VS 0
  47
  48
  49 static INLINE unsigned
  50 compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
  51 {
  52    unsigned mask = 0;
  53    unsigned i;
  54
  55    /* Do the hardwired planes first:
  56     */
  57    if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
  58    if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
  59    if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
  60    if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
  61    if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
  62    if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
  63
  64    /* Followed by any remaining ones:
  65     */
  66    for (i = 6; i < nr; i++) {
  67       if (dot4(clip, plane[i]) < 0)
  68          mask |= (1<<i);
  69    }
  70
  71    return mask;
  72 }
  73
  74
  75 typedef void (XSTDCALL *codegen_function) (
  76    const struct tgsi_exec_vector *input,
  77    struct tgsi_exec_vector *output,
  78    float (*constant)[4],
  79    struct tgsi_exec_vector *temporary );
  80
  81
  82 /**
  83  * Transform vertices with the current vertex program/shader
  84  * Up to four vertices can be shaded at a time.
  85  * \param vbuffer  the input vertex data
  86  * \param elts  indexes of four input vertices
  87  * \param count  number of vertices to shade [1..4]
  88  * \param vOut  array of pointers to four output vertices
  89  */
  90 static void
  91 run_vertex_program(struct draw_context *draw,
  92                    unsigned elts[4], unsigned count,
  93                    struct vertex_header *vOut[])
  94 {
  95    struct tgsi_exec_machine *machine = &draw->machine;
  96    unsigned int j;
  97
  98    ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
  99    ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
 100    const float *scale = draw->viewport.scale;
 101    const float *trans = draw->viewport.translate;
 102
 103    assert(count <= 4);
 104    assert(draw->vertex_shader->state->output_semantic_name[0]
 105           == TGSI_SEMANTIC_POSITION);
 106
 107    /* Consts does not require 16 byte alignment. */
 108    machine->Consts = (float (*)[4]) draw->user.constants;
 109
 110    machine->Inputs = ALIGN16_ASSIGN(inputs);
 111    machine->Outputs = ALIGN16_ASSIGN(outputs);
 112
 113    draw->vertex_fetch.fetch_func( draw, machine, elts, count );
 114
 115    /* run shader */
 116 #if defined(__i386__) || defined(__386__)
 117    if (draw->use_sse) {
 118       /* SSE */
 119       /* cast away const */
 120       struct draw_vertex_shader *shader
 121          = (struct draw_vertex_shader *)draw->vertex_shader;
 122       codegen_function func
 123          = (codegen_function) x86_get_func( &shader->sse2_program );
 124       func(
 125          machine->Inputs,
 126          machine->Outputs,
 127          machine->Consts,
 128          machine->Temps );
 129    }
 130    else
 131 #endif
 132    {
 133       /* interpreter */
 134       tgsi_exec_machine_run( machine );
 135    }
 136
 137    /* store machine results */
 138    for (j = 0; j < count; j++) {
 139       unsigned slot;
 140       float x, y, z, w;
 141
 142       /* Handle attr[0] (position) specially:
 143        *
 144        * XXX: Computing the clipmask should be done in the vertex
 145        * program as a set of DP4 instructions appended to the
 146        * user-provided code.
 147        */
 148       x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
 149       y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
 150       z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
 151       w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 152
 153       vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
 154       vOut[j]->edgeflag = 1;
 155
 156       /* divide by w */
 157       w = 1.0f / w;
 158       x *= w;
 159       y *= w;
 160       z *= w;
 161
 162       /* Viewport mapping */
 163       vOut[j]->data[0][0] = x * scale[0] + trans[0];
 164       vOut[j]->data[0][1] = y * scale[1] + trans[1];
 165       vOut[j]->data[0][2] = z * scale[2] + trans[2];
 166       vOut[j]->data[0][3] = w;
 167
 168 #if DBG_VS
 169       printf("output[%d]win: %f %f %f %f\n", j,
 170              vOut[j]->data[0][0],
 171              vOut[j]->data[0][1],
 172              vOut[j]->data[0][2],
 173              vOut[j]->data[0][3]);
 174 #endif
 175       /* Remaining attributes are packed into sequential post-transform
 176        * vertex attrib slots.
 177        */
 178       for (slot = 1; slot < draw->num_vs_outputs; slot++) {
 179          vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 180          vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 181          vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 182          vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 183 #if DBG_VS
 184          printf("output[%d][%d]: %f %f %f %f\n", j, slot,
 185                 vOut[j]->data[slot][0],
 186                 vOut[j]->data[slot][1],
 187                 vOut[j]->data[slot][2],
 188                 vOut[j]->data[slot][3]);
 189 #endif
 190       }
 191    } /* loop over vertices */
 192 }
 193
 194
 195 /**
 196  * Run the vertex shader on all vertices in the vertex queue.
 197  * Called by the draw module when the vertx cache needs to be flushed.
 198  */
 199 void
 200 draw_vertex_shader_queue_flush(struct draw_context *draw)
 201 {
 202    unsigned i;
 203
 204    assert(draw->vs.queue_nr != 0);
 205
 206    /* XXX: do this on statechange:
 207     */
 208    draw_update_vertex_fetch( draw );
 209
 210 //   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
 211 #ifdef MESA_LLVM
 212    if (draw->vertex_shader->llvm_prog) {
 213       draw_vertex_shader_queue_flush_llvm(draw);
 214       return;
 215    }
 216 #endif
 217
 218    /* run vertex shader on vertex cache entries, four per invokation */
 219    for (i = 0; i < draw->vs.queue_nr; i += 4) {
 220       struct vertex_header *dests[4];
 221       unsigned elts[4];
 222       int j, n = MIN2(4, draw->vs.queue_nr - i);
 223
 224       for (j = 0; j < n; j++) {
 225          elts[j] = draw->vs.queue[i + j].elt;
 226          dests[j] = draw->vs.queue[i + j].dest;
 227       }
 228
 229       for ( ; j < 4; j++) {
 230          elts[j] = elts[0];
 231          dests[j] = dests[0];
 232       }
 233
 234       assert(n > 0);
 235       assert(n <= 4);
 236
 237       run_vertex_program(draw, elts, n, dests);
 238    }
 239
 240    draw->vs.queue_nr = 0;
 241 }
 242
 243
 244 struct draw_vertex_shader *
 245 draw_create_vertex_shader(struct draw_context *draw,
 246                           const struct pipe_shader_state *shader)
 247 {
 248    struct draw_vertex_shader *vs;
 249
 250    vs = CALLOC_STRUCT( draw_vertex_shader );
 251    if (vs == NULL) {
 252       return NULL;
 253    }
 254
 255    vs->state = shader;
 256
 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 #elif defined(__i386__) || defined(__386__)
 267    if (draw->use_sse) {
 268       /* cast-away const */
 269       struct pipe_shader_state *sh = (struct pipe_shader_state *) shader;
 270
 271       x86_init_func( &vs->sse2_program );
 272       tgsi_emit_sse2( (struct tgsi_token *) sh->tokens, &vs->sse2_program );
 273    }
 274 #endif
 275
 276    return vs;
 277 }
 278
 279
 280 void
 281 draw_bind_vertex_shader(struct draw_context *draw,
 282                         struct draw_vertex_shader *dvs)
 283 {
 284    draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
 285
 286    draw->vertex_shader = dvs;
 287    draw->num_vs_outputs = dvs->state->num_outputs;
 288
 289    /* specify the fragment program to interpret/execute */
 290    tgsi_exec_machine_init(&draw->machine,
 291                           draw->vertex_shader->state->tokens,
 292                           PIPE_MAX_SAMPLERS,
 293                           NULL /*samplers*/ );
 294 }
 295
 296
 297 void
 298 draw_delete_vertex_shader(struct draw_context *draw,
 299                           struct draw_vertex_shader *dvs)
 300 {
 301 #if defined(__i386__) || defined(__386__)
 302    x86_release_func( (struct x86_function *) &dvs->sse2_program );
 303 #endif
 304
 305    FREE( dvs );
 306 }