brw_vs_tnl.c

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

static struct ureg register_param5( struct tnl_program *p, 
                                    GLint s0,
                                    GLint s1,
                                    GLint s2,
                                    GLint s3,
                                    GLint s4)
{
   gl_state_index tokens[STATE_LENGTH];
   GLint idx;
   tokens[0] = s0;
   tokens[1] = s1;
   tokens[2] = s2;
   tokens[3] = s3;
   tokens[4] = s4;
   idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
   return make_ureg(PROGRAM_STATE_VAR, idx);
}


#define register_param1(p,s0)          register_param5(p,s0,0,0,0,0)
#define register_param2(p,s0,s1)       register_param5(p,s0,s1,0,0,0)
#define register_param3(p,s0,s1,s2)    register_param5(p,s0,s1,s2,0,0)
#define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)


static void register_matrix_param5( struct tnl_program *p,
				    GLint s0, /* matrix name */
				    GLint s1, /* texture matrix number */
				    GLint s2, /* first row */
				    GLint s3, /* last row */
				    GLint s4, /* modifier */
				    struct ureg *matrix )
{
   GLint i;

   /* This is a bit sad as the support is there to pull the whole
    * matrix out in one go:
    */
   for (i = 0; i <= s3 - s2; i++) 
      matrix[i] = register_param5( p, s0, s1, i, i, s4 );
}


static void emit_arg( struct prog_src_register *src,
		      struct ureg reg )
{
   src->File = reg.file;
   src->Index = reg.idx;
   src->Swizzle = reg.swz;
   src->RelAddr = 0;
   src->NegateBase = reg.negate;
   src->Abs = 0;
   src->NegateAbs = 0;
}

static void emit_dst( struct prog_dst_register *dst,
		      struct ureg reg, GLuint mask )
{
   dst->File = reg.file;
   dst->Index = reg.idx;
   /* allow zero as a shorthand for xyzw */
   dst->WriteMask = mask ? mask : WRITEMASK_XYZW; 
   dst->CondMask = 0;
   dst->CondSwizzle = 0;
   dst->CondSrc = 0;
   dst->pad = 0;
}

static void debug_insn( struct prog_instruction *inst, const char *fn,
			GLuint line )
{
   if (DISASSEM) {
      static const char *last_fn;
   
      if (fn != last_fn) {
	 last_fn = fn;
	 _mesa_printf("%s:\n", fn);
      }
	 
      _mesa_printf("%d:\t", line);
      _mesa_print_instruction(inst);
   }
}


static void emit_op3fn(struct tnl_program *p,
		       GLuint op,
		       struct ureg dest,
		       GLuint mask,
		       struct ureg src0,
		       struct ureg src1,
		       struct ureg src2,
		       const char *fn,
		       GLuint line)
{
   GLuint nr = p->program->Base.NumInstructions++;
      
   if (nr >= p->nr_instructions) {
      int new_nr_instructions = p->nr_instructions * 2;

      p->program->Base.Instructions = 
	 _mesa_realloc(p->program->Base.Instructions,
		       sizeof(struct prog_instruction) * p->nr_instructions,
		       sizeof(struct prog_instruction) * new_nr_instructions);
      p->nr_instructions = new_nr_instructions;
   }

   {      
      struct prog_instruction *inst = &p->program->Base.Instructions[nr];
      memset(inst, 0, sizeof(*inst));
      inst->Opcode = op; 
      inst->StringPos = 0;
      inst->Data = 0;
   
      emit_arg( &inst->SrcReg[0], src0 );
      emit_arg( &inst->SrcReg[1], src1 );
      emit_arg( &inst->SrcReg[2], src2 );   

      emit_dst( &inst->DstReg, dest, mask );

      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 )
{
   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_W, src, src);
   emit_op1(p, OPCODE_RSQ, dest, WRITEMASK_W, swizzle1(dest,W));
   emit_op2(p, OPCODE_MUL, dest, WRITEMASK_XYZ, src, swizzle1(dest,W));
}

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;
}


#if 0
static struct ureg get_eye_z( struct tnl_program *p )
{
   if (!is_undef(p->eye_position)) {
      return swizzle1(p->eye_position, Z);
   }
   else if (!is_undef(p->eye_z)) {
      struct ureg pos = register_input( p, BRW_ATTRIB_POS ); 
      struct ureg modelview2;

      p->eye_z = reserve_temp(p);

      register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 2, 1, 
			      STATE_MATRIX, &modelview2 );

      emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
      emit_op2(p, OPCODE_DP4, p->eye_z, WRITEMASK_Z, pos, modelview2);
   }
   
   return swizzle1(p->eye_z, Z)
}
#endif



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);