C++ brw_CMP函数代码示例

/* Kill pixel - set execution mask to zero for those pixels which
 * fail.
 */
static void emit_kil( struct brw_wm_compile *c,
		      struct brw_reg *arg0)
{
   struct brw_compile *p = &c->func;
   struct intel_context *intel = &p->brw->intel;
   struct brw_reg pixelmask;
   GLuint i, j;

   if (intel->gen >= 6)
      pixelmask = retype(brw_vec1_grf(1, 7), BRW_REGISTER_TYPE_UW);
   else
      pixelmask = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);

   for (i = 0; i < 4; i++) {
      /* Check if we've already done the comparison for this reg
       * -- common when someone does KIL TEMP.wwww.
       */
      for (j = 0; j < i; j++) {
	 if (memcmp(&arg0[j], &arg0[i], sizeof(arg0[0])) == 0)
	    break;
      }
      if (j != i)
	 continue;

      brw_push_insn_state(p);
      brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_GE, arg0[i], brw_imm_f(0));   
      brw_set_predicate_control_flag_value(p, 0xff);
      brw_set_compression_control(p, BRW_COMPRESSION_NONE);
      brw_AND(p, pixelmask, brw_flag_reg(), pixelmask);
      brw_pop_insn_state(p);
   }
}

static void emit_min(struct brw_wm_compile *c,
		struct prog_instruction *inst)
{
    struct brw_compile *p = &c->func;
    GLuint mask = inst->DstReg.WriteMask;
    struct brw_reg src0, src1, dst;
    int i;
    brw_push_insn_state(p);
    for (i = 0; i < 4; i++) {
	if (mask & (1<<i)) {
	    dst = get_dst_reg(c, inst, i, 1);
	    src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
	    src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
	    brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
	    brw_MOV(p, dst, src0);
	    brw_set_saturate(p, 0);

	    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, src1, src0);
	    brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
	    brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
	    brw_MOV(p, dst, src1);
	    brw_set_saturate(p, 0);
	    brw_set_predicate_control_flag_value(p, 0xff);
	}
    }
    brw_pop_insn_state(p);
}

static void merge_edgeflags( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp0 = get_element_ud(c->reg.tmp0, 0);

   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK)); 
   brw_CMP(p, 
	   vec1(brw_null_reg()), 
	   BRW_CONDITIONAL_EQ, 
	   tmp0,
	   brw_imm_ud(_3DPRIM_POLYGON));

   /* Get away with using reg.vertex because we know that this is not
    * a _3DPRIM_TRISTRIP_REVERSE:
    */
   brw_IF(p, BRW_EXECUTE_1);
   {   
      brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
      brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<8));
      brw_MOV(p, byte_offset(c->reg.vertex[0],
                             brw_varying_to_offset(&c->vue_map,
                                                   VARYING_SLOT_EDGE)),
              brw_imm_f(0));
      brw_set_predicate_control(p, BRW_PREDICATE_NONE);

      brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
      brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<9));
      brw_MOV(p, byte_offset(c->reg.vertex[2],
                             brw_varying_to_offset(&c->vue_map,
                                                   VARYING_SLOT_EDGE)),
              brw_imm_f(0));
      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
   }
   brw_ENDIF(p);
}

static void emit_unfilled_primitives( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;

   /* Direction culling has already been done.
    */
   if (c->key.fill_ccw != c->key.fill_cw &&
       c->key.fill_ccw != CLIP_CULL &&
       c->key.fill_cw != CLIP_CULL)
   {
      brw_CMP(p,
	      vec1(brw_null_reg()),
	      BRW_CONDITIONAL_GE,
	      get_element(c->reg.dir, 2),
	      brw_imm_f(0));

      brw_IF(p, BRW_EXECUTE_1);
      {
	 emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw);
      }
      brw_ELSE(p);
      {
	 emit_primitives(c, c->key.fill_cw, c->key.offset_cw);
      }
      brw_ENDIF(p);
   }
   else if (c->key.fill_cw != CLIP_CULL) {
      emit_primitives(c, c->key.fill_cw, c->key.offset_cw);
   }
   else if (c->key.fill_ccw != CLIP_CULL) {
      emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw);
   }
}

static void cull_direction( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;
   GLuint conditional;

   assert (!(c->key.fill_ccw == CLIP_CULL &&
	     c->key.fill_cw == CLIP_CULL));

   if (c->key.fill_ccw == CLIP_CULL)
      conditional = BRW_CONDITIONAL_GE;
   else
      conditional = BRW_CONDITIONAL_L;

   brw_CMP(p,
	   vec1(brw_null_reg()),
	   conditional,
	   get_element(c->reg.dir, 2),
	   brw_imm_f(0));

   brw_IF(p, BRW_EXECUTE_1);
   {
      brw_clip_kill_thread(c);
   }
   brw_ENDIF(p);
}

void brw_emit_unfilled_clip( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;

   c->need_direction = ((c->key.offset_ccw || c->key.offset_cw) ||
			(c->key.fill_ccw != c->key.fill_cw) ||
			c->key.fill_ccw == CLIP_CULL ||
			c->key.fill_cw == CLIP_CULL ||
			c->key.copy_bfc_cw ||
			c->key.copy_bfc_ccw);

   brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6);
   brw_clip_tri_init_vertices(c);
   brw_clip_init_ff_sync(c);

   assert(brw_clip_have_varying(c, VARYING_SLOT_EDGE));

   if (c->key.fill_ccw == CLIP_CULL &&
       c->key.fill_cw == CLIP_CULL) {
      brw_clip_kill_thread(c);
      return;
   }

   merge_edgeflags(c);

   /* Need to use the inlist indirection here:
    */
   if (c->need_direction)
      compute_tri_direction(c);

   if (c->key.fill_ccw == CLIP_CULL ||
       c->key.fill_cw == CLIP_CULL)
      cull_direction(c);

   if (c->key.offset_ccw ||
       c->key.offset_cw)
      compute_offset(c);

   if (c->key.copy_bfc_ccw ||
       c->key.copy_bfc_cw)
      copy_bfc(c);

   /* Need to do this whether we clip or not:
    */
   if (c->has_flat_shading)
      brw_clip_tri_flat_shade(c);

   brw_clip_init_clipmask(c);
   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0));
   brw_IF(p, BRW_EXECUTE_1);
   {
      brw_clip_init_planes(c);
      brw_clip_tri(c);
      check_nr_verts(c);
   }
   brw_ENDIF(p);

   emit_unfilled_primitives(c);
   brw_clip_kill_thread(c);
}

void brw_clip_tri_flat_shade( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;
   struct brw_instruction *is_poly;
   struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */

   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK)); 
   brw_CMP(p, 
	   vec1(brw_null_reg()), 
	   BRW_CONDITIONAL_EQ, 
	   tmp0,
	   brw_imm_ud(_3DPRIM_POLYGON));

   is_poly = brw_IF(p, BRW_EXECUTE_1);
   {   
      brw_clip_copy_colors(c, 1, 0);
      brw_clip_copy_colors(c, 2, 0);
   }
   is_poly = brw_ELSE(p, is_poly);
   {
      brw_clip_copy_colors(c, 0, 2);
      brw_clip_copy_colors(c, 1, 2);
   }
   brw_ENDIF(p, is_poly);
}

static void do_twoside_color( struct brw_sf_compile *c )
{
   struct brw_compile *p = &c->func;
   GLuint backface_conditional = c->key.frontface_ccw ? BRW_CONDITIONAL_G : BRW_CONDITIONAL_L;

   /* Already done in clip program:
    */
   if (c->key.primitive == SF_UNFILLED_TRIS)
      return;

   /* If the vertex shader provides backface color, do the selection. The VS
    * promises to set up the front color if the backface color is provided, but
    * it may contain junk if never written to.
    */
   if (!(have_attr(c, VARYING_SLOT_COL0) && have_attr(c, VARYING_SLOT_BFC0)) &&
       !(have_attr(c, VARYING_SLOT_COL1) && have_attr(c, VARYING_SLOT_BFC1)))
      return;

   /* Need to use BRW_EXECUTE_4 and also do an 4-wide compare in order
    * to get all channels active inside the IF.  In the clipping code
    * we run with NoMask, so it's not an option and we can use
    * BRW_EXECUTE_1 for all comparisions.
    */
   brw_CMP(p, vec4(brw_null_reg()), backface_conditional, c->det, brw_imm_f(0));
   brw_IF(p, BRW_EXECUTE_4);
   {
      switch (c->nr_verts) {
      case 3: copy_bfc(c, c->vert[2]);
      case 2: copy_bfc(c, c->vert[1]);
      case 1: copy_bfc(c, c->vert[0]);
      }
   }
   brw_ENDIF(p);
}

static void emit_lit( struct brw_compile *p, 
		      const struct brw_reg *dst,
		      GLuint mask,
		      const struct brw_reg *arg0 )
{
   assert((mask & WRITEMASK_XW) == 0);

   if (mask & WRITEMASK_Y) {
      brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
      brw_MOV(p, dst[1], arg0[0]);
      brw_set_saturate(p, 0);
   }

   if (mask & WRITEMASK_Z) {
      emit_math2(p, BRW_MATH_FUNCTION_POW,
		 &dst[2],
		 WRITEMASK_X | (mask & SATURATE),
		 &arg0[1],
		 &arg0[3]);
   }

   /* Ordinarily you'd use an iff statement to skip or shortcircuit
    * some of the POW calculations above, but 16-wide iff statements
    * seem to lock c1 hardware, so this is a nasty workaround:
    */
   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_LE, arg0[0], brw_imm_f(0));
   {
      if (mask & WRITEMASK_Y) 
	 brw_MOV(p, dst[1], brw_imm_f(0));

      if (mask & WRITEMASK_Z) 
	 brw_MOV(p, dst[2], brw_imm_f(0)); 
   }
   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}

/* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */
void emit_frontfacing(struct brw_compile *p,
		      const struct brw_reg *dst,
		      GLuint mask)
{
   struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD);
   GLuint i;

   if (!(mask & WRITEMASK_XYZW))
      return;

   for (i = 0; i < 4; i++) {
      if (mask & (1<<i)) {
	 brw_MOV(p, dst[i], brw_imm_f(0.0));
      }
   }

   /* bit 31 is "primitive is back face", so checking < (1 << 31) gives
    * us front face
    */
   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, r1_6ud, brw_imm_ud(1 << 31));
   for (i = 0; i < 4; i++) {
      if (mask & (1<<i)) {
	 brw_MOV(p, dst[i], brw_imm_f(1.0));
      }
   }
   brw_set_predicate_control_flag_value(p, 0xff);
}

void brw_clip_tri_flat_shade( struct brw_clip_compile *c )
{
   struct brw_codegen *p = &c->func;
   struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */

   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
   brw_CMP(p,
	   vec1(brw_null_reg()),
	   BRW_CONDITIONAL_EQ,
	   tmp0,
	   brw_imm_ud(_3DPRIM_POLYGON));

   brw_IF(p, BRW_EXECUTE_1);
   {
      brw_clip_copy_flatshaded_attributes(c, 1, 0);
      brw_clip_copy_flatshaded_attributes(c, 2, 0);
   }
   brw_ELSE(p);
   {
      if (c->key.pv_first) {
	 brw_CMP(p,
		 vec1(brw_null_reg()),
		 BRW_CONDITIONAL_EQ,
		 tmp0,
		 brw_imm_ud(_3DPRIM_TRIFAN));
	 brw_IF(p, BRW_EXECUTE_1);
	 {
	    brw_clip_copy_flatshaded_attributes(c, 0, 1);
	    brw_clip_copy_flatshaded_attributes(c, 2, 1);
	 }
	 brw_ELSE(p);
	 {
	    brw_clip_copy_flatshaded_attributes(c, 1, 0);
	    brw_clip_copy_flatshaded_attributes(c, 2, 0);
	 }
	 brw_ENDIF(p);
      }
      else {
         brw_clip_copy_flatshaded_attributes(c, 0, 2);
         brw_clip_copy_flatshaded_attributes(c, 1, 2);
      }
   }
   brw_ENDIF(p);
}

static void copy_bfc( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;
   GLuint conditional;

   /* Do we have any colors to copy?
    */
   if (!(brw_clip_have_varying(c, VARYING_SLOT_COL0) &&
         brw_clip_have_varying(c, VARYING_SLOT_BFC0)) &&
       !(brw_clip_have_varying(c, VARYING_SLOT_COL1) &&
         brw_clip_have_varying(c, VARYING_SLOT_BFC1)))
      return;

   /* In some wierd degnerate cases we can end up testing the
    * direction twice, once for culling and once for bfc copying.  Oh
    * well, that's what you get for setting wierd GL state.
    */
   if (c->key.copy_bfc_ccw)
      conditional = BRW_CONDITIONAL_GE;
   else
      conditional = BRW_CONDITIONAL_L;

   brw_CMP(p,
	   vec1(brw_null_reg()),
	   conditional,
	   get_element(c->reg.dir, 2),
	   brw_imm_f(0));

   brw_IF(p, BRW_EXECUTE_1);
   {
      GLuint i;

      for (i = 0; i < 3; i++) {
	 if (brw_clip_have_varying(c, VARYING_SLOT_COL0) &&
             brw_clip_have_varying(c, VARYING_SLOT_BFC0))
	    brw_MOV(p,
		    byte_offset(c->reg.vertex[i],
                                brw_varying_to_offset(&c->vue_map,
                                                      VARYING_SLOT_COL0)),
		    byte_offset(c->reg.vertex[i],
                                brw_varying_to_offset(&c->vue_map,
                                                      VARYING_SLOT_BFC0)));

	 if (brw_clip_have_varying(c, VARYING_SLOT_COL1) &&
             brw_clip_have_varying(c, VARYING_SLOT_BFC1))
	    brw_MOV(p,
		    byte_offset(c->reg.vertex[i],
                                brw_varying_to_offset(&c->vue_map,
                                                      VARYING_SLOT_COL1)),
		    byte_offset(c->reg.vertex[i],
                                brw_varying_to_offset(&c->vue_map,
                                                      VARYING_SLOT_BFC1)));
      }
   }
   brw_ENDIF(p);
}

static void maybe_do_clip_tri( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;

   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0));
   brw_IF(p, BRW_EXECUTE_1);
   {
      do_clip_tri(c);
   }
   brw_ENDIF(p);
}

static void check_nr_verts( struct brw_clip_compile *c )
{
   struct brw_compile *p = &c->func;

   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.nr_verts, brw_imm_d(3));
   brw_IF(p, BRW_EXECUTE_1);
   {
      brw_clip_kill_thread(c);
   }
   brw_ENDIF(p);
}

void emit_sign(struct brw_compile *p,
	       const struct brw_reg *dst,
	       GLuint mask,
	       const struct brw_reg *arg0)
{
   GLuint i;

   for (i = 0; i < 4; i++) {
      if (mask & (1<<i)) {
	 brw_MOV(p, dst[i], brw_imm_f(0.0));

	 brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0[i], brw_imm_f(0));
	 brw_MOV(p, dst[i], brw_imm_f(-1.0));
	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);

	 brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, arg0[i], brw_imm_f(0));
	 brw_MOV(p, dst[i], brw_imm_f(1.0));
	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
      }
   }
}