brw_wm_emit.c

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

   brw_MOV(p, brw_message_reg(reg), aa);
   brw_pop_insn_state(p);
}


/* Post-fragment-program processing.  Send the results to the
 * framebuffer.
 */
static void emit_fb_write( struct brw_wm_compile *c,
			   struct brw_reg *arg0,
			   struct brw_reg *arg1,
			   struct brw_reg *arg2,
			   GLuint target,
			   GLuint eot)
{
   struct brw_compile *p = &c->func;
   GLuint nr = 2;
   GLuint channel;

   /* Reserve a space for AA - may not be needed:
    */
   if (c->key.aa_dest_stencil_reg)
      nr += 1;

   /* I don't really understand how this achieves the color interleave
    * (ie RGBARGBA) in the result:  [Do the saturation here]
    */
   {
      brw_push_insn_state(p);
      
      for (channel = 0; channel < 4; channel++) {
	 /*  mov (8) m2.0<1>:ud   r28.0<8;8,1>:ud  { Align1 } */
	 /*  mov (8) m6.0<1>:ud   r29.0<8;8,1>:ud  { Align1 SecHalf } */

	 brw_set_compression_control(p, BRW_COMPRESSION_NONE);
	 brw_MOV(p,
		 brw_message_reg(nr + channel),
		 arg0[channel]);
       
	 brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
	 brw_MOV(p,
		 brw_message_reg(nr + channel + 4),
		 sechalf(arg0[channel]));
      }

      /* skip over the regs populated above:
       */
      nr += 8;
   
      brw_pop_insn_state(p);
   }

   if (c->key.source_depth_to_render_target)
   {
      if (c->key.computes_depth) 
	 brw_MOV(p, brw_message_reg(nr), arg2[2]);
      else 
	 brw_MOV(p, brw_message_reg(nr), arg1[1]); /* ? */

      nr += 2;
   }

   if (c->key.dest_depth_reg)
   {
      GLuint comp = c->key.dest_depth_reg / 2;
      GLuint off = c->key.dest_depth_reg % 2;

      if (off != 0) {
         brw_push_insn_state(p);
         brw_set_compression_control(p, BRW_COMPRESSION_NONE);

         brw_MOV(p, brw_message_reg(nr), offset(arg1[comp],1));
         /* 2nd half? */
         brw_MOV(p, brw_message_reg(nr+1), arg1[comp+1]);
         brw_pop_insn_state(p);
      }
      else {
         brw_MOV(p, brw_message_reg(nr), arg1[comp]);
      }
      nr += 2;
   }


   if (!c->key.runtime_check_aads_emit) {
      if (c->key.aa_dest_stencil_reg)
	 emit_aa(c, arg1, 2);

      fire_fb_write(c, 0, nr, target, eot);
   }
   else {
      struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
      struct brw_reg ip = brw_ip_reg();
      struct brw_instruction *jmp;
      
      brw_set_compression_control(p, BRW_COMPRESSION_NONE);
      brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
      brw_AND(p, 
	      v1_null_ud, 
	      get_element_ud(brw_vec8_grf(1,0), 6), 
	      brw_imm_ud(1<<26)); 

      jmp = brw_JMPI(p, ip, ip, brw_imm_w(0));
      {
	 emit_aa(c, arg1, 2);
	 fire_fb_write(c, 0, nr, target, eot);
	 /* note - thread killed in subroutine */
      }
      brw_land_fwd_jump(p, jmp);

      /* ELSE: Shuffle up one register to fill in the hole left for AA:
       */
      fire_fb_write(c, 1, nr-1, target, eot);
   }
}




/* Post-fragment-program processing.  Send the results to the
 * framebuffer.
 */
static void emit_spill( struct brw_wm_compile *c,
			struct brw_reg reg,
			GLuint slot )
{
   struct brw_compile *p = &c->func;

   /*
     mov (16) m2.0<1>:ud   r2.0<8;8,1>:ud   { Align1 Compr }
   */
   brw_MOV(p, brw_message_reg(2), reg);

   /*
     mov (1) r0.2<1>:d    0x00000080:d     { Align1 NoMask }
     send (16) null.0<1>:uw m1               r0.0<8;8,1>:uw   0x053003ff:ud    { Align1 }
   */
   brw_dp_WRITE_16(p, 
		   retype(vec16(brw_vec8_grf(0, 0)), BRW_REGISTER_TYPE_UW),
		   1, 
		   slot);
}

static void emit_unspill( struct brw_wm_compile *c,
			  struct brw_reg reg,
			  GLuint slot )
{
   struct brw_compile *p = &c->func;

   /* Slot 0 is the undef value.
    */
   if (slot == 0) {
      brw_MOV(p, reg, brw_imm_f(0));
      return;
   }

   /*
     mov (1) r0.2<1>:d    0x000000c0:d     { Align1 NoMask }
     send (16) r110.0<1>:uw m1               r0.0<8;8,1>:uw   0x041243ff:ud    { Align1 }
   */

   brw_dp_READ_16(p,
		  retype(vec16(reg), BRW_REGISTER_TYPE_UW),
		  1, 
		  slot);
}



/**
 * Retrieve upto 4 GEN4 register pairs for the given wm reg:
 */
static void get_argument_regs( struct brw_wm_compile *c,
			       struct brw_wm_ref *arg[],
			       struct brw_reg *regs )
{
   GLuint i;

   for (i = 0; i < 4; i++) {
      if (arg[i]) {

	 if (arg[i]->unspill_reg) 
	    emit_unspill(c, 
			 brw_vec8_grf(arg[i]->unspill_reg, 0),
			 arg[i]->value->spill_slot);

	 regs[i] = arg[i]->hw_reg;	 
      }
      else {
	 regs[i] = brw_null_reg();
      }
   }
}

static void spill_values( struct brw_wm_compile *c,
			  struct brw_wm_value *values,
			  GLuint nr )
{
   GLuint i;

   for (i = 0; i < nr; i++)
      if (values[i].spill_slot) 
	 emit_spill(c, values[i].hw_reg, values[i].spill_slot);
}



/* Emit the fragment program instructions here.
 */
void brw_wm_emit( struct brw_wm_compile *c )
{
   struct brw_compile *p = &c->func;
   GLuint insn;

   brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED);

   /* Check if any of the payload regs need to be spilled:
    */
   spill_values(c, c->payload.depth, 4);
   spill_values(c, c->creg, c->nr_creg);
   spill_values(c, c->payload.input_interp, FRAG_ATTRIB_MAX);
   

   for (insn = 0; insn < c->nr_insns; insn++) {

      struct brw_wm_instruction *inst = &c->instruction[insn];
      struct brw_reg args[3][4], dst[4];
      GLuint i, dst_flags;
      
      /* Get argument regs:
       */
      for (i = 0; i < 3; i++) 
	 get_argument_regs(c, inst->src[i], args[i]);

      /* Get dest regs:
       */
      for (i = 0; i < 4; i++)
	 if (inst->dst[i])
	    dst[i] = inst->dst[i]->hw_reg;
	 else
	    dst[i] = brw_null_reg();
      
      /* Flags
       */
      dst_flags = inst->writemask;
      if (inst->saturate) 
	 dst_flags |= SATURATE;

      switch (inst->opcode) {
	 /* Generated instructions for calculating triangle interpolants:
	  */
      case WM_PIXELXY:
	 emit_pixel_xy(p, dst, dst_flags, args[0]);
	 break;

      case WM_DELTAXY:
	 emit_delta_xy(p, dst, dst_flags, args[0], args[1]);
	 break;

      case WM_WPOSXY:
	 emit_wpos_xy(c, dst, dst_flags, args[0]);
	 break;

      case WM_PIXELW:
	 emit_pixel_w(p, dst, dst_flags, args[0], args[1]);
	 break;

      case WM_LINTERP:
	 emit_linterp(p, dst, dst_flags, args[0], args[1]);
	 break;

      case WM_PINTERP:
	 emit_pinterp(p, dst, dst_flags, args[0], args[1], args[2]);
	 break;

      case WM_CINTERP:
	 emit_cinterp(p, dst, dst_flags, args[0]);
	 break;

      case WM_FB_WRITE:
	 emit_fb_write(c, args[0], args[1], args[2], inst->target, inst->eot);
	 break;

	 /* Straightforward arithmetic:
	  */
      case OPCODE_ADD:
	 emit_alu2(p, brw_ADD, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_FRC:
	 emit_alu1(p, brw_FRC, dst, dst_flags, args[0]);
	 break;

      case OPCODE_FLR:
	 emit_alu1(p, brw_RNDD, dst, dst_flags, args[0]);
	 break;

      case OPCODE_DP3:	/*  */
	 emit_dp3(p, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_DP4:
	 emit_dp4(p, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_DPH:
	 emit_dph(p, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_LRP:	/*  */
	 emit_lrp(p, dst, dst_flags, args[0], args[1], args[2]);
	 break;

      case OPCODE_MAD:	
	 emit_mad(p, dst, dst_flags, args[0], args[1], args[2]);
	 break;

      case OPCODE_MOV:
      case OPCODE_SWZ:
	 emit_alu1(p, brw_MOV, dst, dst_flags, args[0]);
	 break;

      case OPCODE_MUL:
	 emit_alu2(p, brw_MUL, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_XPD:
	 emit_xpd(p, dst, dst_flags, args[0], args[1]);
	 break;

	 /* Higher math functions:
	  */
      case OPCODE_RCP:
	 emit_math1(p, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]);
	 break;

      case OPCODE_RSQ:
	 emit_math1(p, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]);
	 break;

      case OPCODE_SIN:
	 emit_math1(p, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]);
	 break;

      case OPCODE_COS:
	 emit_math1(p, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]);
	 break;

      case OPCODE_EX2:
	 emit_math1(p, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]);
	 break;

      case OPCODE_LG2:
	 emit_math1(p, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]);
	 break;

      case OPCODE_SCS:
	 /* There is an scs math function, but it would need some
	  * fixup for 16-element execution.
	  */
	 if (dst_flags & WRITEMASK_X)
	    emit_math1(p, BRW_MATH_FUNCTION_COS, dst, (dst_flags&SATURATE)|WRITEMASK_X, args[0]);
	 if (dst_flags & WRITEMASK_Y)
	    emit_math1(p, BRW_MATH_FUNCTION_SIN, dst+1, (dst_flags&SATURATE)|WRITEMASK_X, args[0]);
	 break;

      case OPCODE_POW:
	 emit_math2(p, BRW_MATH_FUNCTION_POW, dst, dst_flags, args[0], args[1]);
	 break;

	 /* Comparisons:
	  */
      case OPCODE_CMP:
	 emit_cmp(p, dst, dst_flags, args[0], args[1], args[2]);
	 break;

      case OPCODE_MAX:
	 emit_max(p, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_MIN:
	 emit_min(p, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_SLT:
	 emit_slt(p, dst, dst_flags, args[0], args[1]);
	 break;

      case OPCODE_SLE:
	 emit_sle(p, dst, dst_flags, args[0], args[1]);
	break;
      case OPCODE_SGT:
	 emit_sgt(p, dst, dst_flags, args[0], args[1]);
	break;
      case OPCODE_SGE:
	 emit_sge(p, dst, dst_flags, args[0], args[1]);
	 break;
      case OPCODE_SEQ:
	 emit_seq(p, dst, dst_flags, args[0], args[1]);
	break;
      case OPCODE_SNE:
	 emit_sne(p, dst, dst_flags, args[0], args[1]);
	break;

      case OPCODE_LIT:
	 emit_lit(p, dst, dst_flags, args[0]);
	 break;

	 /* Texturing operations:
	  */
      case OPCODE_TEX:
	 emit_tex(c, inst, dst, dst_flags, args[0]);
	 break;

      case OPCODE_TXB:
	 emit_txb(c, inst, dst, dst_flags, args[0]);
	 break;

      case OPCODE_KIL:
	 emit_kil(c, args[0]);
	 break;

      default:
	_mesa_printf("unsupport opcode %d in fragment program\n", 
		inst->opcode);
      }
      
      for (i = 0; i < 4; i++)
	if (inst->dst[i] && inst->dst[i]->spill_slot) 
	   emit_spill(c, 
		      inst->dst[i]->hw_reg, 
		      inst->dst[i]->spill_slot);
   }
}