t_vp_build.c

Mesa is an open-source implementation of the OpenGL specification - a system for rendering interacti

	    emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z)));

	    if (!is_undef(att)) 
	       emit_op2(p, OPCODE_MUL, lit, 0, lit, att);

	    if (count == nr_lights) {
	       if (separate) {
		  mask0 = WRITEMASK_XYZ;
		  mask1 = WRITEMASK_XYZ;
		  res0 = register_output( p, VERT_RESULT_BFC0 );
		  res1 = register_output( p, VERT_RESULT_BFC1 );
	       }
	       else {
		  mask0 = 0;
		  mask1 = WRITEMASK_XYZ;
		  res0 = _bfc0;
		  res1 = register_output( p, VERT_RESULT_BFC0 );
	       }
	    } else {
	       res0 = _bfc0;
	       res1 = _bfc1;
	       mask0 = 0;
	       mask1 = 0;
	    }

	    emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);

	    release_temp(p, ambient);
	    release_temp(p, diffuse);
	    release_temp(p, specular);
	 }

	 release_temp(p, att);
      }
   }

   release_temps( p );
}


static void build_fog( struct tnl_program *p )
{
   struct ureg fog = register_output(p, VERT_RESULT_FOGC);
   struct ureg input;

   if (p->state->fog_source_is_depth) {
      input = swizzle1(get_eye_position(p), Z);
   }
   else {
      input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
   }

   if (p->state->fog_mode && p->state->tnl_do_vertex_fog) {
      struct ureg params = register_param2(p, STATE_INTERNAL,
					   STATE_FOG_PARAMS_OPTIMIZED);
      struct ureg tmp = get_temp(p);
      GLboolean useabs = (p->state->fog_mode != FOG_EXP2);

      if (useabs) {
	 emit_op1(p, OPCODE_ABS, tmp, 0, input);
      }

      switch (p->state->fog_mode) {
      case FOG_LINEAR: {
	 struct ureg id = get_identity_param(p);
	 emit_op3(p, OPCODE_MAD, tmp, 0, useabs ? tmp : input,
			swizzle1(params,X), swizzle1(params,Y));
	 emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
	 emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
	 break;
      }
      case FOG_EXP:
	 emit_op2(p, OPCODE_MUL, tmp, 0, useabs ? tmp : input,
			swizzle1(params,Z));
	 emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
	 break;
      case FOG_EXP2:
	 emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
	 emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
	 emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
	 break;
      }

      release_temp(p, tmp);
   }
   else {
      /* results = incoming fog coords (compute fog per-fragment later) 
       *
       * KW:  Is it really necessary to do anything in this case?
       * BP: Yes, we always need to compute the absolute value, unless
       * we want to push that down into the fragment program...
       */
      GLboolean useabs = GL_TRUE;
      emit_op1(p, useabs ? OPCODE_ABS : OPCODE_MOV, fog, WRITEMASK_X, input);
   }
}
 
static void build_reflect_texgen( struct tnl_program *p,
				  struct ureg dest,
				  GLuint writemask )
{
   struct ureg normal = get_eye_normal(p);
   struct ureg eye_hat = get_eye_position_normalized(p);
   struct ureg tmp = get_temp(p);

   /* n.u */
   emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat); 
   /* 2n.u */
   emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp); 
   /* (-2n.u)n + u */
   emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);

   release_temp(p, tmp);
}

static void build_sphere_texgen( struct tnl_program *p,
				 struct ureg dest,
				 GLuint writemask )
{
   struct ureg normal = get_eye_normal(p);
   struct ureg eye_hat = get_eye_position_normalized(p);
   struct ureg tmp = get_temp(p);
   struct ureg half = register_scalar_const(p, .5);
   struct ureg r = get_temp(p);
   struct ureg inv_m = get_temp(p);
   struct ureg id = get_identity_param(p);

   /* Could share the above calculations, but it would be
    * a fairly odd state for someone to set (both sphere and
    * reflection active for different texture coordinate
    * components.  Of course - if two texture units enable
    * reflect and/or sphere, things start to tilt in favour
    * of seperating this out:
    */

   /* n.u */
   emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat); 
   /* 2n.u */
   emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp); 
   /* (-2n.u)n + u */
   emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat); 
   /* r + 0,0,1 */
   emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z)); 
   /* rx^2 + ry^2 + (rz+1)^2 */
   emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp); 
   /* 2/m */
   emit_op1(p, OPCODE_RSQ, tmp, 0, tmp); 
   /* 1/m */
   emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half); 
   /* r/m + 1/2 */
   emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half); 
	       
   release_temp(p, tmp);
   release_temp(p, r);
   release_temp(p, inv_m);
}


static void build_texture_transform( struct tnl_program *p )
{
   GLuint i, j;

   for (i = 0; i < MAX_TEXTURE_UNITS; i++) {

      if (!(p->state->fragprog_inputs_read & FRAG_BIT_TEX(i)))
	 continue;
							     
      if (p->state->unit[i].texgen_enabled || 
	  p->state->unit[i].texmat_enabled) {
	 
	 GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
	 struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
	 struct ureg out_texgen = undef;

	 if (p->state->unit[i].texgen_enabled) {
	    GLuint copy_mask = 0;
	    GLuint sphere_mask = 0;
	    GLuint reflect_mask = 0;
	    GLuint normal_mask = 0;
	    GLuint modes[4];
	 
	    if (texmat_enabled) 
	       out_texgen = get_temp(p);
	    else
	       out_texgen = out;

	    modes[0] = p->state->unit[i].texgen_mode0;
	    modes[1] = p->state->unit[i].texgen_mode1;
	    modes[2] = p->state->unit[i].texgen_mode2;
	    modes[3] = p->state->unit[i].texgen_mode3;

	    for (j = 0; j < 4; j++) {
	       switch (modes[j]) {
	       case TXG_OBJ_LINEAR: {
		  struct ureg obj = register_input(p, VERT_ATTRIB_POS);
		  struct ureg plane = 
		     register_param3(p, STATE_TEXGEN, i,
				     STATE_TEXGEN_OBJECT_S + j);

		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j, 
			   obj, plane );
		  break;
	       }
	       case TXG_EYE_LINEAR: {
		  struct ureg eye = get_eye_position(p);
		  struct ureg plane = 
		     register_param3(p, STATE_TEXGEN, i, 
				     STATE_TEXGEN_EYE_S + j);

		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j, 
			   eye, plane );
		  break;
	       }
	       case TXG_SPHERE_MAP: 
		  sphere_mask |= WRITEMASK_X << j;
		  break;
	       case TXG_REFLECTION_MAP:
		  reflect_mask |= WRITEMASK_X << j;
		  break;
	       case TXG_NORMAL_MAP: 
		  normal_mask |= WRITEMASK_X << j;
		  break;
	       case TXG_NONE:
		  copy_mask |= WRITEMASK_X << j;
	       }

	    }

	 
	    if (sphere_mask) {
	       build_sphere_texgen(p, out_texgen, sphere_mask);
	    }

	    if (reflect_mask) {
	       build_reflect_texgen(p, out_texgen, reflect_mask);
	    }

	    if (normal_mask) {
	       struct ureg normal = get_eye_normal(p);
	       emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
	    }

	    if (copy_mask) {
	       struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
	       emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
	    }
	 }

	 if (texmat_enabled) {
	    struct ureg texmat[4];
	    struct ureg in = (!is_undef(out_texgen) ? 
			      out_texgen : 
			      register_input(p, VERT_ATTRIB_TEX0+i));
	    if (PREFER_DP4) {
	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
				       0, texmat );
	       emit_matrix_transform_vec4( p, out, texmat, in );
	    }
	    else {
	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
				       STATE_MATRIX_TRANSPOSE, texmat );
	       emit_transpose_matrix_transform_vec4( p, out, texmat, in );
	    }
	 }

	 release_temps(p);
      } 
      else {
	 emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
      }
   }
}


static void build_pointsize( struct tnl_program *p )
{
   struct ureg eye = get_eye_position(p);
   struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
   struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
   struct ureg out = register_output(p, VERT_RESULT_PSIZ);
   struct ureg ut = get_temp(p);

   /* dist = |eyez| */
   emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
   /* p1 + dist * (p2 + dist * p3); */
   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
		swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
		ut, swizzle1(state_attenuation, X));

   /* 1 / sqrt(factor) */
   emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );

#if 1
   /* out = pointSize / sqrt(factor) */
   emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
#else
   /* not sure, might make sense to do clamping here,
      but it's not done in t_vb_points neither */
   emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
   emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
   emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
#endif

   release_temp(p, ut);
}

static void build_tnl_program( struct tnl_program *p )
{   /* Emit the program, starting with modelviewproject:
    */
   build_hpos(p);

   /* Lighting calculations:
    */
   if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) {
      if (p->state->light_global_enabled)
	 build_lighting(p);
      else {
	 if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
	    emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0);

	 if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
	    emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1);
      }
   }

   if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) ||
       p->state->fog_mode != FOG_NONE)
      build_fog(p);

   if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY)
      build_texture_transform(p);

   if (p->state->point_attenuated)
      build_pointsize(p);

   /* Finish up:
    */
   emit_op1(p, OPCODE_END, undef, 0, undef);

   /* Disassemble:
    */
   if (DISASSEM) {
      _mesa_printf ("\n");
   }
}


static void
create_new_program( const struct state_key *key,
                    struct gl_vertex_program *program,
                    GLuint max_temps)
{
   struct tnl_program p;

   _mesa_memset(&p, 0, sizeof(p));
   p.state = key;
   p.program = program;
   p.eye_position = undef;
   p.eye_position_normalized = undef;
   p.eye_normal = undef;
   p.identity = undef;
   p.temp_in_use = 0;
   
   if (max_temps >= sizeof(int) * 8)
      p.temp_reserved = 0;
   else
      p.temp_reserved = ~((1<<max_temps)-1);

   p.program->Base.Instructions = _mesa_alloc_instructions(MAX_INSN);
   p.program->Base.String = NULL;
   p.program->Base.NumInstructions =
   p.program->Base.NumTemporaries =
   p.program->Base.NumParameters =
   p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
   p.program->Base.Parameters = _mesa_new_parameter_list();
   p.program->Base.InputsRead = 0;
   p.program->Base.OutputsWritten = 0;

   build_tnl_program( &p );
}


static struct gl_vertex_program *
search_cache(struct tnl_cache *cache, GLuint hash,
             const void *key, GLuint keysize)
{
   struct tnl_cache_item *c;

   for (c = cache->items[hash % cache->size]; c; c = c->next) {
      if (c->hash == hash && _mesa_memcmp(c->key, key, keysize) == 0)
	 return c->prog;
   }

   return NULL;
}


static void rehash( struct tnl_cache *cache )
{
   struct tnl_cache_item **items;
   struct tnl_cache_item *c, *next;
   GLuint size, i;

   size = cache->size * 3;
   items = (struct tnl_cache_item**) _mesa_malloc(size * sizeof(*items));
   _mesa_memset(items, 0, size * sizeof(*items));

   for (i = 0; i < cache->size; i++)
      for (c = cache->items[i]; c; c = next) {
	 next = c->next;
	 c->next = items[c->hash % size];
	 items[c->hash % size] = c;
      }

   FREE(cache->items);
   cache->items = items;
   cache->size = size;
}

static void cache_item( GLcontext *ctx,
                        struct tnl_cache *cache,
			GLuint hash,
			void *key,
			struct gl_vertex_program *prog )
{
   struct tnl_cache_item *c = CALLOC_STRUCT(tnl_cache_item);
   c->hash = hash;
   c->key = key;

   c->prog = prog;

   if (++cache->n_items > cache->size * 1.5)
      rehash(cache);

   c->next = cache->items[hash % cache->size];
   cache->items[hash % cache->size] = c;
}

static GLuint hash_key( struct state_key *key )
{
   GLuint *ikey = (GLuint *)key;
   GLuint hash = 0, i;

   /* I'm sure this can be improved on, but speed is important:
    */
   for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
      hash ^= ikey[i];

   return hash;
}

void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
{
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct state_key *key;
   GLuint hash;
   const struct gl_vertex_program *prev = ctx->VertexProgram._Current;

   if (!ctx->VertexProgram._Current ||
       ctx->VertexProgram._Current == ctx->VertexProgram._TnlProgram) {
      struct gl_vertex_program *newProg;

      /* Grab all the relevent state and put it in a single structure:
       */
      key = make_state_key(ctx);
      hash = hash_key(key);

      /* Look for an already-prepared program for this state:
       */
      newProg = search_cache( tnl->vp_cache, hash, key, sizeof(*key));
   
      /* OK, we'll have to build a new one:
       */
      if (!newProg) {

	 if (0)
	    _mesa_printf("Build new TNL program\n");
	 
	 newProg = (struct gl_vertex_program *)
	    ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0); 

	 create_new_program( key, newProg, ctx->Const.VertexProgram.MaxTemps );

	 if (ctx->Driver.ProgramStringNotify)
	    ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB, 
                                             &newProg->Base );

         /* Our ownership of newProg is transferred to the cache */
	 cache_item(ctx, tnl->vp_cache, hash, key, newProg);
      }
      else {
	 FREE(key);
      }

      _mesa_reference_vertprog(ctx, &ctx->VertexProgram._TnlProgram, newProg);
      _mesa_reference_vertprog(ctx, &ctx->VertexProgram._Current, newProg);
   }

   /* Tell the driver about the change.  Could define a new target for
    * this?
    */
   if (ctx->VertexProgram._Current != prev && ctx->Driver.BindProgram) {
      ctx->Driver.BindProgram(ctx, GL_VERTEX_PROGRAM_ARB,
                            (struct gl_program *) ctx->VertexProgram._Current);
   }
}

void _tnl_ProgramCacheInit( GLcontext *ctx )
{
   TNLcontext *tnl = TNL_CONTEXT(ctx);

   tnl->vp_cache = (struct tnl_cache *) MALLOC(sizeof(*tnl->vp_cache));
   tnl->vp_cache->size = 17;
   tnl->vp_cache->n_items = 0;
   tnl->vp_cache->items = (struct tnl_cache_item**)
      _mesa_calloc(tnl->vp_cache->size * sizeof(*tnl->vp_cache->items));
}

void _tnl_ProgramCacheDestroy( GLcontext *ctx )
{
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct tnl_cache_item *c, *next;
   GLuint i;

   for (i = 0; i < tnl->vp_cache->size; i++)
      for (c = tnl->vp_cache->items[i]; c; c = next) {
	 next = c->next;
	 FREE(c->key);
	 _mesa_reference_vertprog(ctx, &c->prog, NULL);
	 FREE(c);
      }

   FREE(tnl->vp_cache->items);
   FREE(tnl->vp_cache);
}