t_vp_build.c

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

   debug_insn(inst, fn, line);
}


#define emit_op3(p, op, dst, mask, src0, src1, src2) \
   emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)

#define emit_op2(p, op, dst, mask, src0, src1) \
    emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)

#define emit_op1(p, op, dst, mask, src0) \
    emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)


static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
{
   if (reg.file == PROGRAM_TEMPORARY && 
       !(p->temp_reserved & (1<<reg.idx)))
      return reg;
   else {
      struct ureg temp = get_temp(p);
      emit_op1(p, OPCODE_MOV, temp, 0, reg);
      return temp;
   }
}


/* Currently no tracking performed of input/output/register size or
 * active elements.  Could be used to reduce these operations, as
 * could the matrix type.
 */
static void emit_matrix_transform_vec4( struct tnl_program *p,
					struct ureg dest,
					const struct ureg *mat,
					struct ureg src)
{
   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
}

/* This version is much easier to implement if writemasks are not
 * supported natively on the target or (like SSE), the target doesn't
 * have a clean/obvious dotproduct implementation.
 */
static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
						  struct ureg dest,
						  const struct ureg *mat,
						  struct ureg src)
{
   struct ureg tmp;

   if (dest.file != PROGRAM_TEMPORARY)
      tmp = get_temp(p);
   else
      tmp = dest;

   emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
   emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
   emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
   emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);

   if (dest.file != PROGRAM_TEMPORARY)
      release_temp(p, tmp);
}

static void emit_matrix_transform_vec3( struct tnl_program *p,
					struct ureg dest,
					const struct ureg *mat,
					struct ureg src)
{
   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
}


static void emit_normalize_vec3( struct tnl_program *p,
				 struct ureg dest,
				 struct ureg src )
{
   struct ureg tmp = get_temp(p);
   emit_op2(p, OPCODE_DP3, tmp, 0, src, src);
   emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
   emit_op2(p, OPCODE_MUL, dest, 0, src, tmp);
   release_temp(p, tmp);
}

static void emit_passthrough( struct tnl_program *p, 
			      GLuint input,
			      GLuint output )
{
   struct ureg out = register_output(p, output);
   emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input)); 
}

static struct ureg get_eye_position( struct tnl_program *p )
{
   if (is_undef(p->eye_position)) {
      struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 
      struct ureg modelview[4];

      p->eye_position = reserve_temp(p);

      if (PREFER_DP4) {
	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
                                 0, modelview );

	 emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
      }
      else {
	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
				 STATE_MATRIX_TRANSPOSE, modelview );

	 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
      }
   }
   
   return p->eye_position;
}


static struct ureg get_eye_position_normalized( struct tnl_program *p )
{
   if (is_undef(p->eye_position_normalized)) {
      struct ureg eye = get_eye_position(p);
      p->eye_position_normalized = reserve_temp(p);
      emit_normalize_vec3(p, p->eye_position_normalized, eye);
   }
   
   return p->eye_position_normalized;
}


static struct ureg get_eye_normal( struct tnl_program *p )
{
   if (is_undef(p->eye_normal)) {
      struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
      struct ureg mvinv[3];

      register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
			      STATE_MATRIX_INVTRANS, mvinv );

      p->eye_normal = reserve_temp(p);

      /* Transform to eye space:
       */
      emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );

      /* Normalize/Rescale:
       */
      if (p->state->normalize) {
	 emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
      }
      else if (p->state->rescale_normals) {
	 struct ureg rescale = register_param2(p, STATE_INTERNAL,
					       STATE_NORMAL_SCALE);

	 emit_op2( p, OPCODE_MUL, p->eye_normal, 0, p->eye_normal,
		   swizzle1(rescale, X));
      }
   }

   return p->eye_normal;
}



static void build_hpos( struct tnl_program *p )
{
   struct ureg pos = register_input( p, VERT_ATTRIB_POS ); 
   struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
   struct ureg mvp[4];

   if (PREFER_DP4) {
      register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3, 
			      0, mvp );
      emit_matrix_transform_vec4( p, hpos, mvp, pos );
   }
   else {
      register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3, 
			      STATE_MATRIX_TRANSPOSE, mvp );
      emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
   }
}


static GLuint material_attrib( GLuint side, GLuint property )
{
   return ((property - STATE_AMBIENT) * 2 + 
	   side);
}

/* Get a bitmask of which material values vary on a per-vertex basis.
 */
static void set_material_flags( struct tnl_program *p )
{
   p->color_materials = 0;
   p->materials = 0;

   if (p->state->light_color_material) {
      p->materials = 
	 p->color_materials = p->state->light_color_material_mask;
   }

   p->materials |= p->state->light_material_mask;
}


static struct ureg get_material( struct tnl_program *p, GLuint side, 
				 GLuint property )
{
   GLuint attrib = material_attrib(side, property);

   if (p->color_materials & (1<<attrib))
      return register_input(p, VERT_ATTRIB_COLOR0);
   else if (p->materials & (1<<attrib)) 
      return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT );
   else
      return register_param3( p, STATE_MATERIAL, side, property );
}

#define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
				   MAT_BIT_FRONT_AMBIENT | \
				   MAT_BIT_FRONT_DIFFUSE) << (side))

/* Either return a precalculated constant value or emit code to
 * calculate these values dynamically in the case where material calls
 * are present between begin/end pairs.
 *
 * Probably want to shift this to the program compilation phase - if
 * we always emitted the calculation here, a smart compiler could
 * detect that it was constant (given a certain set of inputs), and
 * lift it out of the main loop.  That way the programs created here
 * would be independent of the vertex_buffer details.
 */
static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
{
   if (p->materials & SCENE_COLOR_BITS(side)) {
      struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
      struct ureg material_emission = get_material(p, side, STATE_EMISSION);
      struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
      struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
      struct ureg tmp = make_temp(p, material_diffuse);
      emit_op3(p, OPCODE_MAD, tmp,  WRITEMASK_XYZ, lm_ambient, 
	       material_ambient, material_emission);
      return tmp;
   }
   else
      return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
}


static struct ureg get_lightprod( struct tnl_program *p, GLuint light, 
				  GLuint side, GLuint property )
{
   GLuint attrib = material_attrib(side, property);
   if (p->materials & (1<<attrib)) {
      struct ureg light_value = 
	 register_param3(p, STATE_LIGHT, light, property);
      struct ureg material_value = get_material(p, side, property);
      struct ureg tmp = get_temp(p);
      emit_op2(p, OPCODE_MUL, tmp,  0, light_value, material_value);
      return tmp;
   }
   else
      return register_param4(p, STATE_LIGHTPROD, light, side, property);
}

static struct ureg calculate_light_attenuation( struct tnl_program *p,
						GLuint i, 
						struct ureg VPpli,
						struct ureg dist )
{
   struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
					     STATE_ATTENUATION);
   struct ureg att = get_temp(p);

   /* Calculate spot attenuation:
    */
   if (!p->state->unit[i].light_spotcutoff_is_180) {
      struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
						  STATE_SPOT_DIR_NORMALIZED, i);
      struct ureg spot = get_temp(p);
      struct ureg slt = get_temp(p);

      emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
      emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
      emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
      emit_op2(p, OPCODE_MUL, att, 0, slt, spot);

      release_temp(p, spot);
      release_temp(p, slt);
   }

   /* Calculate distance attenuation:
    */
   if (p->state->unit[i].light_attenuated) {

      /* 1/d,d,d,1/d */
      emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist); 
      /* 1,d,d*d,1/d */
      emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y)); 
      /* 1/dist-atten */
      emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist); 

      if (!p->state->unit[i].light_spotcutoff_is_180) {
	 /* dist-atten */
	 emit_op1(p, OPCODE_RCP, dist, 0, dist); 
	 /* spot-atten * dist-atten */
	 emit_op2(p, OPCODE_MUL, att, 0, dist, att);	
      } else {
	 /* dist-atten */
	 emit_op1(p, OPCODE_RCP, att, 0, dist); 
      }
   }

   return att;
}
						




/* Need to add some addtional parameters to allow lighting in object
 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
 * space lighting.
 */
static void build_lighting( struct tnl_program *p )
{
   const GLboolean twoside = p->state->light_twoside;
   const GLboolean separate = p->state->separate_specular;
   GLuint nr_lights = 0, count = 0;
   struct ureg normal = get_eye_normal(p);
   struct ureg lit = get_temp(p);
   struct ureg dots = get_temp(p);
   struct ureg _col0 = undef, _col1 = undef;
   struct ureg _bfc0 = undef, _bfc1 = undef;
   GLuint i;

   for (i = 0; i < MAX_LIGHTS; i++) 
      if (p->state->unit[i].light_enabled)
	 nr_lights++;
   
   set_material_flags(p);

   {
      struct ureg shininess = get_material(p, 0, STATE_SHININESS);
      emit_op1(p, OPCODE_MOV, dots,  WRITEMASK_W, swizzle1(shininess,X));

      _col0 = make_temp(p, get_scenecolor(p, 0));
      if (separate)
	 _col1 = make_temp(p, get_identity_param(p));
      else
	 _col1 = _col0;

   }

   if (twoside) {
      struct ureg shininess = get_material(p, 1, STATE_SHININESS);
      emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z, 
	       negate(swizzle1(shininess,X)));

      _bfc0 = make_temp(p, get_scenecolor(p, 1));
      if (separate)
	 _bfc1 = make_temp(p, get_identity_param(p));
      else
	 _bfc1 = _bfc0;
   }


   /* If no lights, still need to emit the scenecolor.
    */
      {
	 struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
	 emit_op1(p, OPCODE_MOV, res0, 0, _col0);
      }

      if (separate) {
	 struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
	 emit_op1(p, OPCODE_MOV, res1, 0, _col1);
      }

      if (twoside) {
	 struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
	 emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
      }
      
      if (twoside && separate) {
	 struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
	 emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
      }
      
   if (nr_lights == 0) {
      release_temps(p);
      return;
   }


   for (i = 0; i < MAX_LIGHTS; i++) {
      if (p->state->unit[i].light_enabled) {
	 struct ureg half = undef;
	 struct ureg att = undef, VPpli = undef;
	  
	 count++;

	 if (p->state->unit[i].light_eyepos3_is_zero) {
	    /* Can used precomputed constants in this case.
	     * Attenuation never applies to infinite lights.
	     */
	    VPpli = register_param3(p, STATE_LIGHT, i, 
				    STATE_POSITION_NORMALIZED); 
            if (p->state->light_local_viewer) {
                struct ureg eye_hat = get_eye_position_normalized(p);
                half = get_temp(p);
                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
                emit_normalize_vec3(p, half, half);
            } else {
                half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR);
            }
	 } 
	 else {
	    struct ureg Ppli = register_param3(p, STATE_LIGHT, i, 
					       STATE_POSITION); 
	    struct ureg V = get_eye_position(p);
	    struct ureg dist = get_temp(p);
	    struct ureg tmpPpli = get_temp(p);

	    VPpli = get_temp(p); 
 
            /* In homogeneous object coordinates
             */
            emit_op1(p, OPCODE_RCP, dist, 0, swizzle1(Ppli, W));
            emit_op2(p, OPCODE_MUL, tmpPpli, 0, Ppli, dist);

	    /* Calculate VPpli vector
	     */
	    emit_op2(p, OPCODE_SUB, VPpli, 0, tmpPpli, V); 

            /* we're done with tmpPpli now */
	    release_temp(p, tmpPpli);

	    /* Normalize VPpli.  The dist value also used in
	     * attenuation below.
	     */
	    emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
	    emit_op1(p, OPCODE_RSQ, dist, 0, dist);
	    emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);


	    /* Calculate  attenuation:
	     */ 
	    if (!p->state->unit[i].light_spotcutoff_is_180 ||
		p->state->unit[i].light_attenuated) {
	       att = calculate_light_attenuation(p, i, VPpli, dist);
	    }
	    
	    /* We're done with dist now */
	    release_temp(p, dist);
	 
      
	    /* Calculate viewer direction, or use infinite viewer:
	     */
	    half = get_temp(p);
	    if (p->state->light_local_viewer) {
	       struct ureg eye_hat = get_eye_position_normalized(p);
	       emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
	    }
	    else {
	       struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z); 
	       emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
	    }

	    emit_normalize_vec3(p, half, half);
	 }

	 /* Calculate dot products:
	  */
	 emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
	 emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);

	 /* we're done with VPpli and half now, so free them as to not drive up
	    our temp usage unnecessary */
	 release_temp(p, VPpli);
	 release_temp(p, half);
	
	 /* Front face lighting:
	  */
	 {
	    struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
	    struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
	    struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
	    struct ureg res0, res1;
	    GLuint mask0, mask1;

	    emit_op1(p, OPCODE_LIT, lit, 0, dots);
   
	    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_COL0 );
		  res1 = register_output( p, VERT_RESULT_COL1 );
	       }
	       else {
		  mask0 = 0;
		  mask1 = WRITEMASK_XYZ;
		  res0 = _col0;
		  res1 = register_output( p, VERT_RESULT_COL0 );
	       }
	    } else {
	       mask0 = 0;
	       mask1 = 0;
	       res0 = _col0;
	       res1 = _col1;
	    }

	    emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
      
	    release_temp(p, ambient);
	    release_temp(p, diffuse);
	    release_temp(p, specular);
	 }

	 /* Back face lighting:
	  */
	 if (twoside) {
	    struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
	    struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
	    struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
	    struct ureg res0, res1;
	    GLuint mask0, mask1;