src/gallium/auxiliary/draw/draw_vs_llvm.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   *   Zack Rusin
  31   *   Keith Whitwell <keith@tungstengraphics.com>
  32   *   Brian Paul
  33   */
  34
  35 #include "pipe/p_util.h"
  36 #include "pipe/p_shader_tokens.h"
  37 #include "draw_private.h"
  38 #include "draw_context.h"
  39 #include "draw_vs.h"
  40
  41 #ifdef MESA_LLVM
  42
  43 #include "llvm/gallivm.h"
  44
  45 struct draw_llvm_vertex_shader {
  46    struct draw_vertex_shader base;
  47    struct gallivm_prog *llvm_prog;
  48 };
  49
  50
  51 static INLINE unsigned
  52 compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
  53 {
  54    unsigned mask = 0;
  55    unsigned i;
  56
  57    /* Do the hardwired planes first:
  58     */
  59    if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
  60    if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
  61    if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
  62    if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
  63    if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
  64    if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
  65
  66    /* Followed by any remaining ones:
  67     */
  68    for (i = 6; i < nr; i++) {
  69       if (dot4(clip, plane[i]) < 0)
  70          mask |= (1<<i);
  71    }
  72
  73    return mask;
  74 }
  75
  76
  77
  78 static void
  79 vs_llvm_prepare( struct draw_vertex_shader *base,
  80                  struct draw_context *draw )
  81 {
  82    draw_update_vertex_fetch( draw );
  83 }
  84
  85
  86
  87 /**
  88  * Transform vertices with the current vertex program/shader
  89  * Up to four vertices can be shaded at a time.
  90  * \param vbuffer  the input vertex data
  91  * \param elts  indexes of four input vertices
  92  * \param count  number of vertices to shade [1..4]
  93  * \param vOut  array of pointers to four output vertices
  94  */
  95 static void
  96 vs_llvm_run( struct draw_vertex_shader *base,
  97              struct draw_context *draw,
  98              const unsigned *elts,
  99              unsigned count,
 100              struct vertex_header *vOut[] )
 101 {
 102    struct draw_llvm_vertex_shader *shader =
 103       (struct draw_llvm_vertex_shader *)base;
 104
 105    struct tgsi_exec_machine *machine = &draw->machine;
 106    unsigned int j;
 107
 108    ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
 109    ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
 110    const float *scale = draw->viewport.scale;
 111    const float *trans = draw->viewport.translate;
 112
 113
 114    assert(count <= 4);
 115    assert(draw->vertex_shader->state->output_semantic_name[0]
 116           == TGSI_SEMANTIC_POSITION);
 117
 118    /* Consts does not require 16 byte alignment. */
 119    machine->Consts = (float (*)[4]) draw->user.constants;
 120
 121    machine->Inputs = ALIGN16_ASSIGN(inputs);
 122    machine->Outputs = ALIGN16_ASSIGN(outputs);
 123
 124    draw->vertex_fetch.fetch_func( draw, machine, elts, count );
 125
 126    /* run shader */
 127    gallivm_cpu_vs_exec(shader->llvm_prog,
 128                        machine->Inputs,
 129                        machine->Outputs,
 130                        machine->Consts,
 131                        machine->Temps);
 132
 133    /* store machine results */
 134    for (j = 0; j < count; j++) {
 135       unsigned slot;
 136       float x, y, z, w;
 137
 138       x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
 139       y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
 140       z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
 141       w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 142
 143       vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
 144       vOut[j]->edgeflag = 1;
 145
 146       /* divide by w */
 147       w = 1.0f / w;
 148       x *= w;
 149       y *= w;
 150       z *= w;
 151
 152       /* Viewport mapping */
 153       vOut[j]->data[0][0] = x * scale[0] + trans[0];
 154       vOut[j]->data[0][1] = y * scale[1] + trans[1];
 155       vOut[j]->data[0][2] = z * scale[2] + trans[2];
 156       vOut[j]->data[0][3] = w;
 157
 158       /* Remaining attributes are packed into sequential post-transform
 159        * vertex attrib slots.
 160        */
 161       for (slot = 1; slot < draw->num_vs_outputs; slot++) {
 162          vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 163          vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 164          vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 165          vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 166       }
 167    } /* loop over vertices */
 168 }
 169
 170 static void
 171 vs_llvm_delete( struct draw_vertex_shader *base )
 172 {
 173    struct draw_llvm_vertex_shader *shader =
 174       (struct draw_llvm_vertex_shader *)base;
 175
 176    /* Do something to free compiled shader:
 177     */
 178
 179    FREE( shader );
 180 }
 181
 182
 183
 184
 185 struct draw_vertex_shader *
 186 draw_create_vs_llvm(struct draw_context *draw,
 187                     const struct pipe_shader_state *templ)
 188 {
 189    struct draw_llvm_vertex_shader *vs;
 190
 191    vs = CALLOC_STRUCT( draw_llvm_vertex_shader );
 192    if (vs == NULL)
 193       return NULL;
 194
 195    vs->base.state = templ;
 196    vs->base.prepare = vs_llvm_prepare;
 197    vs->base.run = vs_llvm_run;
 198    vs->base.delete = vs_llvm_delete;
 199
 200    {
 201       struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);
 202       gallivm_ir_set_layout(ir, GALLIVM_SOA);
 203       gallivm_ir_set_components(ir, 4);
 204       gallivm_ir_fill_from_tgsi(ir, vs->base.state->tokens);
 205       vs->llvm_prog = gallivm_ir_compile(ir);
 206       gallivm_ir_delete(ir);
 207    }
 208
 209    draw->engine = gallivm_global_cpu_engine();
 210
 211    /* XXX: Why are there two versions of this?  Shouldn't creating the
 212     *      engine be a separate operation to compiling a shader?
 213     */
 214    if (!draw->engine) {
 215       draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
 216    }
 217    else {
 218       gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
 219    }
 220
 221    return &vs->base;
 222 }
 223
 224
 225
 226
 227
 228 #else
 229
 230 struct draw_vertex_shader *
 231 draw_create_vs_llvm(struct draw_context *draw,
 232                           const struct pipe_shader_state *shader)
 233 {
 234    return NULL;
 235 }
 236
 237 #endif