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 #include "tgsi/util/tgsi_parse.h"
  42
  43 #ifdef MESA_LLVM
  44
  45 #include "gallivm/gallivm.h"
  46
  47 struct draw_llvm_vertex_shader {
  48    struct draw_vertex_shader base;
  49    struct gallivm_prog *llvm_prog;
  50 };
  51
  52
  53 static INLINE unsigned
  54 compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
  55 {
  56    unsigned mask = 0;
  57    unsigned i;
  58
  59    /* Do the hardwired planes first:
  60     */
  61    if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
  62    if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
  63    if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
  64    if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
  65    if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
  66    if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
  67
  68    /* Followed by any remaining ones:
  69     */
  70    for (i = 6; i < nr; i++) {
  71       if (dot4(clip, plane[i]) < 0)
  72          mask |= (1<<i);
  73    }
  74
  75    return mask;
  76 }
  77
  78
  79
  80 static void
  81 vs_llvm_prepare( struct draw_vertex_shader *base,
  82                  struct draw_context *draw )
  83 {
  84    draw_update_vertex_fetch( draw );
  85 }
  86
  87
  88
  89 /**
  90  * Transform vertices with the current vertex program/shader
  91  * Up to four vertices can be shaded at a time.
  92  * \param vbuffer  the input vertex data
  93  * \param elts  indexes of four input vertices
  94  * \param count  number of vertices to shade [1..4]
  95  * \param vOut  array of pointers to four output vertices
  96  */
  97 static void
  98 vs_llvm_run( struct draw_vertex_shader *base,
  99              struct draw_context *draw,
 100              const unsigned *elts,
 101              unsigned count,
 102              struct vertex_header *vOut[] )
 103 {
 104    struct draw_llvm_vertex_shader *shader =
 105       (struct draw_llvm_vertex_shader *)base;
 106
 107    struct tgsi_exec_machine *machine = &draw->machine;
 108    unsigned int j;
 109
 110    ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_MAX_ATTRIBS);
 111    ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_MAX_ATTRIBS);
 112    const float *scale = draw->viewport.scale;
 113    const float *trans = draw->viewport.translate;
 114
 115
 116    assert(count <= 4);
 117    assert(draw->vertex_shader->state->output_semantic_name[0]
 118           == TGSI_SEMANTIC_POSITION);
 119
 120    /* Consts does not require 16 byte alignment. */
 121    machine->Consts = (float (*)[4]) draw->user.constants;
 122
 123    machine->Inputs = ALIGN16_ASSIGN(inputs);
 124    if (draw->rasterizer->bypass_vs) {
 125       /* outputs are just the inputs */
 126       machine->Outputs = machine->Inputs;
 127    }
 128    else {
 129       machine->Outputs = ALIGN16_ASSIGN(outputs);
 130    }
 131
 132
 133    draw->vertex_fetch.fetch_func( draw, machine, elts, count );
 134
 135    if (!draw->rasterizer->bypass_vs) {
 136       /* run shader */
 137       gallivm_cpu_vs_exec(shader->llvm_prog,
 138                           machine->Inputs,
 139                           machine->Outputs,
 140                           machine->Consts,
 141                           machine->Temps);
 142    }
 143
 144    /* store machine results */
 145    for (j = 0; j < count; j++) {
 146       unsigned slot;
 147       float x, y, z, w;
 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       if (!draw->rasterizer->bypass_clipping) {
 155          vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
 156
 157          /* divide by w */
 158          w = 1.0f / w;
 159          x *= w;
 160          y *= w;
 161          z *= w;
 162       }
 163       else {
 164          vOut[j]->clipmask = 0;
 165       }
 166       vOut[j]->edgeflag = 1;
 167
 168       if (!draw->identity_viewport) {
 169          /* Viewport mapping */
 170          vOut[j]->data[0][0] = x * scale[0] + trans[0];
 171          vOut[j]->data[0][1] = y * scale[1] + trans[1];
 172          vOut[j]->data[0][2] = z * scale[2] + trans[2];
 173          vOut[j]->data[0][3] = w;
 174       }
 175       else {
 176          vOut[j]->data[0][0] = x;
 177          vOut[j]->data[0][1] = y;
 178          vOut[j]->data[0][2] = z;
 179          vOut[j]->data[0][3] = w;
 180       }
 181
 182       /* Remaining attributes are packed into sequential post-transform
 183        * vertex attrib slots.
 184        */
 185       for (slot = 1; slot < draw->num_vs_outputs; slot++) {
 186          vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 187          vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 188          vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 189          vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 190       }
 191    } /* loop over vertices */
 192 }
 193
 194 static void
 195 vs_llvm_delete( struct draw_vertex_shader *base )
 196 {
 197    struct draw_llvm_vertex_shader *shader =
 198       (struct draw_llvm_vertex_shader *)base;
 199
 200    /* Do something to free compiled shader:
 201     */
 202
 203    FREE( (void*) shader->base.state.tokens );
 204    FREE( shader );
 205 }
 206
 207
 208
 209
 210 struct draw_vertex_shader *
 211 draw_create_vs_llvm(struct draw_context *draw,
 212                     const struct pipe_shader_state *templ)
 213 {
 214    struct draw_llvm_vertex_shader *vs;
 215    uint nt = tgsi_num_tokens(templ->tokens);
 216
 217    vs = CALLOC_STRUCT( draw_llvm_vertex_shader );
 218    if (vs == NULL)
 219       return NULL;
 220
 221    /* we make a private copy of the tokens */
 222    vs->base.state.tokens = mem_dup(templ->tokens, nt * sizeof(templ->tokens[0]));
 223    vs->base.prepare = vs_llvm_prepare;
 224    vs->base.run = vs_llvm_run;
 225    vs->base.delete = vs_llvm_delete;
 226
 227    {
 228       struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);
 229       gallivm_ir_set_layout(ir, GALLIVM_SOA);
 230       gallivm_ir_set_components(ir, 4);
 231       gallivm_ir_fill_from_tgsi(ir, vs->base.state->tokens);
 232       vs->llvm_prog = gallivm_ir_compile(ir);
 233       gallivm_ir_delete(ir);
 234    }
 235
 236    draw->engine = gallivm_global_cpu_engine();
 237
 238    /* XXX: Why are there two versions of this?  Shouldn't creating the
 239     *      engine be a separate operation to compiling a shader?
 240     */
 241    if (!draw->engine) {
 242       draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
 243    }
 244    else {
 245       gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
 246    }
 247
 248    return &vs->base;
 249 }
 250
 251
 252
 253
 254
 255 #else
 256
 257 struct draw_vertex_shader *
 258 draw_create_vs_llvm(struct draw_context *draw,
 259                           const struct pipe_shader_state *shader)
 260 {
 261    return NULL;
 262 }
 263
 264 #endif