/**
 * Generate texel lookup instruction.
 */
static void
emit_tex(struct i915_fp_compile *p,
         const struct i915_full_instruction *inst,
         uint opcode,
         struct i915_fragment_shader* fs)
{
   uint texture = inst->Texture.Texture;
   uint unit = inst->Src[1].Register.Index;
   uint tex = translate_tex_src_target( p, texture );
   uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
   uint coord = src_vector( p, &inst->Src[0], fs);

   i915_emit_texld( p,
                    get_result_vector( p, &inst->Dst[0] ),
                    get_result_flags( inst ),
                    sampler,
                    coord,
                    opcode,
                    i915_num_coords(texture) );
}
/**
 * Generate a simple arithmetic instruction
 * \param opcode  the i915 opcode
 * \param numArgs  the number of input/src arguments
 */
static void
emit_simple_arith(struct i915_fp_compile *p,
                  const struct i915_full_instruction *inst,
                  uint opcode, uint numArgs,
                  struct i915_fragment_shader *fs)
{
   uint arg1, arg2, arg3;

   assert(numArgs <= 3);

   arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->Src[0], fs );
   arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->Src[1], fs );
   arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->Src[2], fs );

   i915_emit_arith( p,
                    opcode,
                    get_result_vector( p, &inst->Dst[0]),
                    get_result_flags( inst ), 0,
                    arg1,
                    arg2,
                    arg3 );
}
/* Possible concerns:
 *
 * SIN, COS -- could use another taylor step?
 * LIT      -- results seem a little different to sw mesa
 * LOG      -- different to mesa on negative numbers, but this is conformant.
 * 
 * Parse failures -- Mesa doesn't currently give a good indication
 * internally whether a particular program string parsed or not.  This
 * can lead to confusion -- hopefully we cope with it ok now.
 *
 */
static void
upload_program(struct i915_fragment_program *p)
{
   const struct gl_fragment_program *program =
      p->ctx->FragmentProgram._Current;
   const struct prog_instruction *inst = program->Base.Instructions;

/*    _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */

   /* Is this a parse-failed program?  Ensure a valid program is
    * loaded, as the flagging of an error isn't sufficient to stop
    * this being uploaded to hardware.
    */
   if (inst[0].Opcode == OPCODE_END) {
      GLuint tmp = i915_get_utemp(p);
      i915_emit_arith(p,
                      A0_MOV,
                      UREG(REG_TYPE_OC, 0),
                      A0_DEST_CHANNEL_ALL, 0,
                      swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
      return;
   }

   if (program->Base.NumInstructions > I915_MAX_INSN) {
       i915_program_error( p, "Exceeded max instructions" );
       return;
    }

   /* Not always needed:
    */
   calc_live_regs(p);

   while (1) {
      GLuint src0, src1, src2, flags;
      GLuint tmp = 0, consts0 = 0, consts1 = 0;

      switch (inst->Opcode) {
      case OPCODE_ABS:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         i915_emit_arith(p,
                         A0_MAX,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         src0, negate(src0, 1, 1, 1, 1), 0);
         break;

      case OPCODE_ADD:
         EMIT_2ARG_ARITH(A0_ADD);
         break;

      case OPCODE_CMP:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         src2 = src_vector(p, &inst->SrcReg[2], program);
         i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
         break;

      case OPCODE_COS:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);
	 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
	 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);

	 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
         i915_emit_arith(p,
                         A0_MAD,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         src0,
			 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
			 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */

         i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);

	 i915_emit_arith(p,
			 A0_MAD,
			 tmp, A0_DEST_CHANNEL_X, 0,
			 tmp,
			 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
			 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */

	 /* Compute COS with the same calculation used for SIN, but a
	  * different source range has been mapped to [-1,1] this time.
	  */

	 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
	 i915_emit_arith(p,
                         A0_MAX,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
			 0);

	 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
	 i915_emit_arith(p,
			 A0_MUL,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 tmp,
			 0);

	 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
         i915_emit_arith(p,
                         A0_DP3,
                         tmp, A0_DEST_CHANNEL_X, 0,
			 tmp,
                         swizzle(consts1, X, Y, ZERO, ZERO),
			 0);

	 /* tmp.x now contains a first approximation (y).  Now, weight it
	  * against tmp.y**2 to get closer.
	  */
	 i915_emit_arith(p,
                         A0_MAX,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
			 0);

	 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
	 i915_emit_arith(p,
			 A0_MAD,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 swizzle(tmp, ZERO, Y, ZERO, ZERO),
			 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));

	 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
	 i915_emit_arith(p,
			 A0_MAD,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
			 swizzle(consts1, W, W, W, W),
			 swizzle(tmp, Y, Y, Y, Y),
			 swizzle(tmp, X, X, X, X));
         break;

      case OPCODE_DP3:
         EMIT_2ARG_ARITH(A0_DP3);
         break;

      case OPCODE_DP4:
         EMIT_2ARG_ARITH(A0_DP4);
         break;

      case OPCODE_DPH:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);

         i915_emit_arith(p,
                         A0_DP4,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, Y, Z, ONE), src1, 0);
         break;

      case OPCODE_DST:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);

         /* result[0] = 1    * 1;
          * result[1] = a[1] * b[1];
          * result[2] = a[2] * 1;
          * result[3] = 1    * b[3];
          */
         i915_emit_arith(p,
                         A0_MUL,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, ONE, Y, Z, ONE),
                         swizzle(src1, ONE, Y, ONE, W), 0);
         break;

      case OPCODE_EX2:
         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_EXP,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_FLR:
         EMIT_1ARG_ARITH(A0_FLR);
         break;

      case OPCODE_FRC:
         EMIT_1ARG_ARITH(A0_FRC);
         break;

      case OPCODE_KIL:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         i915_emit_texld(p, get_live_regs(p, inst),
                         tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
                         0, src0, T0_TEXKILL);
         break;

      case OPCODE_LG2:
         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_LOG,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_LIT:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         /* tmp = max( a.xyzw, a.00zw )
          * XXX: Clamp tmp.w to -128..128
          * tmp.y = log(tmp.y)
          * tmp.y = tmp.w * tmp.y
          * tmp.y = exp(tmp.y)
          * result = cmp (a.11-x1, a.1x01, a.1xy1 )
          */
         i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
                         src0, swizzle(src0, ZERO, ZERO, Z, W), 0);

         i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
                         swizzle(tmp, Y, Y, Y, Y), 0, 0);

         i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
                         swizzle(tmp, ZERO, Y, ZERO, ZERO),
                         swizzle(tmp, ZERO, W, ZERO, ZERO), 0);

         i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
                         swizzle(tmp, Y, Y, Y, Y), 0, 0);

         i915_emit_arith(p, A0_CMP,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
                         swizzle(tmp, ONE, X, ZERO, ONE),
                         swizzle(tmp, ONE, X, Y, ONE));

         break;

      case OPCODE_LRP:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         src2 = src_vector(p, &inst->SrcReg[2], program);
         flags = get_result_flags(inst);
         tmp = i915_get_utemp(p);

         /* b*a + c*(1-a)
          *
          * b*a + c - ca 
          *
          * tmp = b*a + c, 
          * result = (-c)*a + tmp 
          */
         i915_emit_arith(p, A0_MAD, tmp,
                         flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);

         i915_emit_arith(p, A0_MAD,
                         get_result_vector(p, inst),
                         flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
         break;

      case OPCODE_MAD:
         EMIT_3ARG_ARITH(A0_MAD);
         break;

      case OPCODE_MAX:
         EMIT_2ARG_ARITH(A0_MAX);
         break;

      case OPCODE_MIN:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         tmp = i915_get_utemp(p);
         flags = get_result_flags(inst);

         i915_emit_arith(p,
                         A0_MAX,
                         tmp, flags & A0_DEST_CHANNEL_ALL, 0,
                         negate(src0, 1, 1, 1, 1),
                         negate(src1, 1, 1, 1, 1), 0);

         i915_emit_arith(p,
                         A0_MOV,
                         get_result_vector(p, inst),
                         flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
         break;

      case OPCODE_MOV:
         EMIT_1ARG_ARITH(A0_MOV);
         break;

      case OPCODE_MUL:
         EMIT_2ARG_ARITH(A0_MUL);
         break;

      case OPCODE_POW:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         tmp = i915_get_utemp(p);
         flags = get_result_flags(inst);

         /* XXX: masking on intermediate values, here and elsewhere.
          */
         i915_emit_arith(p,
                         A0_LOG,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         swizzle(src0, X, X, X, X), 0, 0);

         i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);


         i915_emit_arith(p,
                         A0_EXP,
                         get_result_vector(p, inst),
                         flags, 0, swizzle(tmp, X, X, X, X), 0, 0);

         break;

      case OPCODE_RCP:
         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_RCP,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_RSQ:

         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_RSQ,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_SCS:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         /* 
          * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
          * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
          * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
          * scs.x = DP4 t1, sin_constants
          * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
          * scs.y = DP4 t1, cos_constants
          */
         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_XY, 0,
                         swizzle(src0, X, X, ONE, ONE),
                         swizzle(src0, X, ONE, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_ALL, 0,
                         swizzle(tmp, X, Y, X, Y),
                         swizzle(tmp, X, X, ONE, ONE), 0);

         if (inst->DstReg.WriteMask & WRITEMASK_Y) {
            GLuint tmp1;

            if (inst->DstReg.WriteMask & WRITEMASK_X)
               tmp1 = i915_get_utemp(p);
            else
               tmp1 = tmp;

            i915_emit_arith(p,
                            A0_MUL,
                            tmp1, A0_DEST_CHANNEL_ALL, 0,
                            swizzle(tmp, X, Y, Y, W),
                            swizzle(tmp, X, Z, ONE, ONE), 0);

            i915_emit_arith(p,
                            A0_DP4,
                            get_result_vector(p, inst),
                            A0_DEST_CHANNEL_Y, 0,
                            swizzle(tmp1, W, Z, Y, X),
                            i915_emit_const4fv(p, sin_constants), 0);
         }

         if (inst->DstReg.WriteMask & WRITEMASK_X) {
            i915_emit_arith(p,
                            A0_MUL,
                            tmp, A0_DEST_CHANNEL_XYZ, 0,
                            swizzle(tmp, X, X, Z, ONE),
                            swizzle(tmp, Z, ONE, ONE, ONE), 0);

            i915_emit_arith(p,
                            A0_DP4,
                            get_result_vector(p, inst),
                            A0_DEST_CHANNEL_X, 0,
                            swizzle(tmp, ONE, Z, Y, X),
                            i915_emit_const4fv(p, cos_constants), 0);
         }
         break;

      case OPCODE_SGE:
         EMIT_2ARG_ARITH(A0_SGE);
         break;

      case OPCODE_SIN:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);
	 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
	 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);

	 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
         i915_emit_arith(p,
                         A0_MAD,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         src0,
			 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
			 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */

         i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);

	 i915_emit_arith(p,
			 A0_MAD,
			 tmp, A0_DEST_CHANNEL_X, 0,
			 tmp,
			 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
			 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */

	 /* Compute sin using a quadratic and quartic.  It gives continuity
	  * that repeating the Taylor series lacks every 2*pi, and has
	  * reduced error.
	  *
	  * The idea was described at:
	  * http://www.devmaster.net/forums/showthread.php?t=5784
	  */

	 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
	 i915_emit_arith(p,
                         A0_MAX,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
			 0);

	 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
	 i915_emit_arith(p,
			 A0_MUL,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 tmp,
			 0);

	 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
         i915_emit_arith(p,
                         A0_DP3,
                         tmp, A0_DEST_CHANNEL_X, 0,
			 tmp,
                         swizzle(consts1, X, Y, ZERO, ZERO),
			 0);

	 /* tmp.x now contains a first approximation (y).  Now, weight it
	  * against tmp.y**2 to get closer.
	  */
	 i915_emit_arith(p,
                         A0_MAX,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
			 0);

	 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
	 i915_emit_arith(p,
			 A0_MAD,
			 tmp, A0_DEST_CHANNEL_Y, 0,
			 swizzle(tmp, ZERO, X, ZERO, ZERO),
			 swizzle(tmp, ZERO, Y, ZERO, ZERO),
			 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));

	 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
	 i915_emit_arith(p,
			 A0_MAD,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
			 swizzle(consts1, W, W, W, W),
			 swizzle(tmp, Y, Y, Y, Y),
			 swizzle(tmp, X, X, X, X));

         break;

      case OPCODE_SLT:
         EMIT_2ARG_ARITH(A0_SLT);
         break;

      case OPCODE_SUB:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);

         i915_emit_arith(p,
                         A0_ADD,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         src0, negate(src1, 1, 1, 1, 1), 0);
         break;

      case OPCODE_SWZ:
         EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
         break;

      case OPCODE_TEX:
         EMIT_TEX(T0_TEXLD);
         break;

      case OPCODE_TXB:
         EMIT_TEX(T0_TEXLDB);
         break;

      case OPCODE_TXP:
         EMIT_TEX(T0_TEXLDP);
         break;

      case OPCODE_XPD:
         /* Cross product:
          *      result.x = src0.y * src1.z - src0.z * src1.y;
          *      result.y = src0.z * src1.x - src0.x * src1.z;
          *      result.z = src0.x * src1.y - src0.y * src1.x;
          *      result.w = undef;
          */
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         tmp = i915_get_utemp(p);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_ALL, 0,
                         swizzle(src0, Z, X, Y, ONE),
                         swizzle(src1, Y, Z, X, ONE), 0);

         i915_emit_arith(p,
                         A0_MAD,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, Y, Z, X, ONE),
                         swizzle(src1, Z, X, Y, ONE),
                         negate(tmp, 1, 1, 1, 0));
         break;

      case OPCODE_END:
         return;

      default:
         i915_program_error(p, "bad opcode");
         return;
      }

      inst++;
      i915_release_utemps(p);
   }
}
/* Possible concerns:
 *
 * SIN, COS -- could use another taylor step?
 * LIT      -- results seem a little different to sw mesa
 * LOG      -- different to mesa on negative numbers, but this is conformant.
 * 
 * Parse failures -- Mesa doesn't currently give a good indication
 * internally whether a particular program string parsed or not.  This
 * can lead to confusion -- hopefully we cope with it ok now.
 *
 */
static void
upload_program(struct i915_fragment_program *p)
{
   const struct gl_fragment_program *program =
      p->ctx->FragmentProgram._Current;
   const struct prog_instruction *inst = program->Base.Instructions;

/*    _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */

   /* Is this a parse-failed program?  Ensure a valid program is
    * loaded, as the flagging of an error isn't sufficient to stop
    * this being uploaded to hardware.
    */
   if (inst[0].Opcode == OPCODE_END) {
      GLuint tmp = i915_get_utemp(p);
      i915_emit_arith(p,
                      A0_MOV,
                      UREG(REG_TYPE_OC, 0),
                      A0_DEST_CHANNEL_ALL, 0,
                      swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
      return;
   }

   while (1) {
      GLuint src0, src1, src2, flags;
      GLuint tmp = 0;

      switch (inst->Opcode) {
      case OPCODE_ABS:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         i915_emit_arith(p,
                         A0_MAX,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         src0, negate(src0, 1, 1, 1, 1), 0);
         break;

      case OPCODE_ADD:
         EMIT_2ARG_ARITH(A0_ADD);
         break;

      case OPCODE_CMP:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         src2 = src_vector(p, &inst->SrcReg[2], program);
         i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
         break;

      case OPCODE_COS:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         src0, i915_emit_const1f(p, 1.0 / (M_PI)), 0);

         i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);

         /* By choosing different taylor constants, could get rid of this mul:
          */
         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         tmp, i915_emit_const1f(p, (M_PI)), 0);

         /* 
          * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
          * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
          * t0 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
          * result = DP4 t0, cos_constants
          */
         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_XY, 0,
                         swizzle(tmp, X, X, ONE, ONE),
                         swizzle(tmp, X, ONE, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_XYZ, 0,
                         swizzle(tmp, X, Y, X, ONE),
                         swizzle(tmp, X, X, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_XYZ, 0,
                         swizzle(tmp, X, X, Z, ONE),
                         swizzle(tmp, Z, ONE, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_DP4,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(tmp, ONE, Z, Y, X),
                         i915_emit_const4fv(p, cos_constants), 0);

         break;

      case OPCODE_DP3:
         EMIT_2ARG_ARITH(A0_DP3);
         break;

      case OPCODE_DP4:
         EMIT_2ARG_ARITH(A0_DP4);
         break;

      case OPCODE_DPH:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);

         i915_emit_arith(p,
                         A0_DP4,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, Y, Z, ONE), src1, 0);
         break;

      case OPCODE_DST:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);

         /* result[0] = 1    * 1;
          * result[1] = a[1] * b[1];
          * result[2] = a[2] * 1;
          * result[3] = 1    * b[3];
          */
         i915_emit_arith(p,
                         A0_MUL,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, ONE, Y, Z, ONE),
                         swizzle(src1, ONE, Y, ONE, W), 0);
         break;

      case OPCODE_EX2:
         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_EXP,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_FLR:
         EMIT_1ARG_ARITH(A0_FLR);
         break;

      case OPCODE_FRC:
         EMIT_1ARG_ARITH(A0_FRC);
         break;

      case OPCODE_KIL:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
                         0, src0, T0_TEXKILL);
         break;

      case OPCODE_LG2:
         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_LOG,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_LIT:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         /* tmp = max( a.xyzw, a.00zw )
          * XXX: Clamp tmp.w to -128..128
          * tmp.y = log(tmp.y)
          * tmp.y = tmp.w * tmp.y
          * tmp.y = exp(tmp.y)
          * result = cmp (a.11-x1, a.1x01, a.1xy1 )
          */
         i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
                         src0, swizzle(src0, ZERO, ZERO, Z, W), 0);

         i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
                         swizzle(tmp, Y, Y, Y, Y), 0, 0);

         i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
                         swizzle(tmp, ZERO, Y, ZERO, ZERO),
                         swizzle(tmp, ZERO, W, ZERO, ZERO), 0);

         i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
                         swizzle(tmp, Y, Y, Y, Y), 0, 0);

         i915_emit_arith(p, A0_CMP,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
                         swizzle(tmp, ONE, X, ZERO, ONE),
                         swizzle(tmp, ONE, X, Y, ONE));

         break;

      case OPCODE_LRP:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         src2 = src_vector(p, &inst->SrcReg[2], program);
         flags = get_result_flags(inst);
         tmp = i915_get_utemp(p);

         /* b*a + c*(1-a)
          *
          * b*a + c - ca 
          *
          * tmp = b*a + c, 
          * result = (-c)*a + tmp 
          */
         i915_emit_arith(p, A0_MAD, tmp,
                         flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);

         i915_emit_arith(p, A0_MAD,
                         get_result_vector(p, inst),
                         flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
         break;

      case OPCODE_MAD:
         EMIT_3ARG_ARITH(A0_MAD);
         break;

      case OPCODE_MAX:
         EMIT_2ARG_ARITH(A0_MAX);
         break;

      case OPCODE_MIN:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         tmp = i915_get_utemp(p);
         flags = get_result_flags(inst);

         i915_emit_arith(p,
                         A0_MAX,
                         tmp, flags & A0_DEST_CHANNEL_ALL, 0,
                         negate(src0, 1, 1, 1, 1),
                         negate(src1, 1, 1, 1, 1), 0);

         i915_emit_arith(p,
                         A0_MOV,
                         get_result_vector(p, inst),
                         flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
         break;

      case OPCODE_MOV:
         EMIT_1ARG_ARITH(A0_MOV);
         break;

      case OPCODE_MUL:
         EMIT_2ARG_ARITH(A0_MUL);
         break;

      case OPCODE_POW:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         tmp = i915_get_utemp(p);
         flags = get_result_flags(inst);

         /* XXX: masking on intermediate values, here and elsewhere.
          */
         i915_emit_arith(p,
                         A0_LOG,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         swizzle(src0, X, X, X, X), 0, 0);

         i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);


         i915_emit_arith(p,
                         A0_EXP,
                         get_result_vector(p, inst),
                         flags, 0, swizzle(tmp, X, X, X, X), 0, 0);

         break;

      case OPCODE_RCP:
         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_RCP,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_RSQ:

         src0 = src_vector(p, &inst->SrcReg[0], program);

         i915_emit_arith(p,
                         A0_RSQ,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, X, X, X, X), 0, 0);
         break;

      case OPCODE_SCS:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         /* 
          * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
          * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
          * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
          * scs.x = DP4 t1, sin_constants
          * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
          * scs.y = DP4 t1, cos_constants
          */
         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_XY, 0,
                         swizzle(src0, X, X, ONE, ONE),
                         swizzle(src0, X, ONE, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_ALL, 0,
                         swizzle(tmp, X, Y, X, Y),
                         swizzle(tmp, X, X, ONE, ONE), 0);

         if (inst->DstReg.WriteMask & WRITEMASK_Y) {
            GLuint tmp1;

            if (inst->DstReg.WriteMask & WRITEMASK_X)
               tmp1 = i915_get_utemp(p);
            else
               tmp1 = tmp;

            i915_emit_arith(p,
                            A0_MUL,
                            tmp1, A0_DEST_CHANNEL_ALL, 0,
                            swizzle(tmp, X, Y, Y, W),
                            swizzle(tmp, X, Z, ONE, ONE), 0);

            i915_emit_arith(p,
                            A0_DP4,
                            get_result_vector(p, inst),
                            A0_DEST_CHANNEL_Y, 0,
                            swizzle(tmp1, W, Z, Y, X),
                            i915_emit_const4fv(p, sin_constants), 0);
         }

         if (inst->DstReg.WriteMask & WRITEMASK_X) {
            i915_emit_arith(p,
                            A0_MUL,
                            tmp, A0_DEST_CHANNEL_XYZ, 0,
                            swizzle(tmp, X, X, Z, ONE),
                            swizzle(tmp, Z, ONE, ONE, ONE), 0);

            i915_emit_arith(p,
                            A0_DP4,
                            get_result_vector(p, inst),
                            A0_DEST_CHANNEL_X, 0,
                            swizzle(tmp, ONE, Z, Y, X),
                            i915_emit_const4fv(p, cos_constants), 0);
         }
         break;

      case OPCODE_SGE:
         EMIT_2ARG_ARITH(A0_SGE);
         break;

      case OPCODE_SIN:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         tmp = i915_get_utemp(p);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         src0, i915_emit_const1f(p, 1.0 / (M_PI)), 0);

         i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);

         /* By choosing different taylor constants, could get rid of this mul:
          */
         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_X, 0,
                         tmp, i915_emit_const1f(p, (M_PI)), 0);

         /* 
          * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
          * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
          * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
          * result = DP4 t1.wzyx, sin_constants
          */
         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_XY, 0,
                         swizzle(tmp, X, X, ONE, ONE),
                         swizzle(tmp, X, ONE, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_ALL, 0,
                         swizzle(tmp, X, Y, X, Y),
                         swizzle(tmp, X, X, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_ALL, 0,
                         swizzle(tmp, X, Y, Y, W),
                         swizzle(tmp, X, Z, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_DP4,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(tmp, W, Z, Y, X),
                         i915_emit_const4fv(p, sin_constants), 0);
         break;

      case OPCODE_SLT:
         EMIT_2ARG_ARITH(A0_SLT);
         break;

      case OPCODE_SUB:
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);

         i915_emit_arith(p,
                         A0_ADD,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         src0, negate(src1, 1, 1, 1, 1), 0);
         break;

      case OPCODE_SWZ:
         EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
         break;

      case OPCODE_TEX:
         EMIT_TEX(T0_TEXLD);
         break;

      case OPCODE_TXB:
         EMIT_TEX(T0_TEXLDB);
         break;

      case OPCODE_TXP:
         EMIT_TEX(T0_TEXLDP);
         break;

      case OPCODE_XPD:
         /* Cross product:
          *      result.x = src0.y * src1.z - src0.z * src1.y;
          *      result.y = src0.z * src1.x - src0.x * src1.z;
          *      result.z = src0.x * src1.y - src0.y * src1.x;
          *      result.w = undef;
          */
         src0 = src_vector(p, &inst->SrcReg[0], program);
         src1 = src_vector(p, &inst->SrcReg[1], program);
         tmp = i915_get_utemp(p);

         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_ALL, 0,
                         swizzle(src0, Z, X, Y, ONE),
                         swizzle(src1, Y, Z, X, ONE), 0);

         i915_emit_arith(p,
                         A0_MAD,
                         get_result_vector(p, inst),
                         get_result_flags(inst), 0,
                         swizzle(src0, Y, Z, X, ONE),
                         swizzle(src1, Z, X, Y, ONE),
                         negate(tmp, 1, 1, 1, 0));
         break;

      case OPCODE_END:
         return;

      default:
         i915_program_error(p, "bad opcode");
         return;
      }

      inst++;
      i915_release_utemps(p);
   }
}
/*
 * Translate TGSI instruction to i915 instruction.
 *
 * Possible concerns:
 *
 * DDX, DDY -- return 0
 * SIN, COS -- could use another taylor step?
 * LIT      -- results seem a little different to sw mesa
 * LOG      -- different to mesa on negative numbers, but this is conformant.
 */
static void
i915_translate_instruction(struct i915_fp_compile *p,
                           const struct i915_full_instruction *inst,
                           struct i915_fragment_shader *fs)
{
   uint writemask;
   uint src0, src1, src2, flags;
   uint tmp = 0;

   switch (inst->Instruction.Opcode) {
   case TGSI_OPCODE_ABS:
      src0 = src_vector(p, &inst->Src[0], fs);
      i915_emit_arith(p,
                      A0_MAX,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      src0, negate(src0, 1, 1, 1, 1), 0);
      break;

   case TGSI_OPCODE_ADD:
      emit_simple_arith(p, inst, A0_ADD, 2, fs);
      break;

   case TGSI_OPCODE_CEIL:
      src0 = src_vector(p, &inst->Src[0], fs);
      tmp = i915_get_utemp(p);
      flags = get_result_flags(inst);
      i915_emit_arith(p,
                      A0_FLR,
                      tmp,
                      flags & A0_DEST_CHANNEL_ALL, 0,
                      negate(src0, 1, 1, 1, 1), 0, 0);
      i915_emit_arith(p,
                      A0_MOV,
                      get_result_vector(p, &inst->Dst[0]),
                      flags, 0,
                      negate(tmp, 1, 1, 1, 1), 0, 0);
      break;

   case TGSI_OPCODE_CMP:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);
      src2 = src_vector(p, &inst->Src[2], fs);
      i915_emit_arith(p, A0_CMP,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst),
                      0, src0, src2, src1);   /* NOTE: order of src2, src1 */
      break;

   case TGSI_OPCODE_COS:
      src0 = src_vector(p, &inst->Src[0], fs);
      tmp = i915_get_utemp(p);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_X, 0,
                      src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);

      i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);

      /* 
       * t0.xy = MUL x.xx11, x.x111  ; x^2, x, 1, 1
       * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
       * t0 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
       * result = DP4 t0, cos_constants
       */
      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_XY, 0,
                      swizzle(tmp, X, X, ONE, ONE),
                      swizzle(tmp, X, ONE, ONE, ONE), 0);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_XYZ, 0,
                      swizzle(tmp, X, Y, X, ONE),
                      swizzle(tmp, X, X, ONE, ONE), 0);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_XYZ, 0,
                      swizzle(tmp, X, X, Z, ONE),
                      swizzle(tmp, Z, ONE, ONE, ONE), 0);

      i915_emit_arith(p,
                      A0_DP4,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(tmp, ONE, Z, Y, X),
                      i915_emit_const4fv(p, cos_constants), 0);
      break;

  case TGSI_OPCODE_DDX:
  case TGSI_OPCODE_DDY:
      /* XXX We just output 0 here */
      debug_printf("Punting DDX/DDX\n");
      src0 = get_result_vector(p, &inst->Dst[0]);
      i915_emit_arith(p,
                      A0_MOV,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0, 0);
      break;

  case TGSI_OPCODE_DP2:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);

      i915_emit_arith(p,
                      A0_DP3,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, X, Y, ZERO, ZERO), src1, 0);
      break;

   case TGSI_OPCODE_DP3:
      emit_simple_arith(p, inst, A0_DP3, 2, fs);
      break;

   case TGSI_OPCODE_DP4:
      emit_simple_arith(p, inst, A0_DP4, 2, fs);
      break;

   case TGSI_OPCODE_DPH:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);

      i915_emit_arith(p,
                      A0_DP4,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, X, Y, Z, ONE), src1, 0);
      break;

   case TGSI_OPCODE_DST:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);

      /* result[0] = 1    * 1;
       * result[1] = a[1] * b[1];
       * result[2] = a[2] * 1;
       * result[3] = 1    * b[3];
       */
      i915_emit_arith(p,
                      A0_MUL,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, ONE, Y, Z, ONE),
                      swizzle(src1, ONE, Y, ONE, W), 0);
      break;

   case TGSI_OPCODE_END:
      /* no-op */
      break;

   case TGSI_OPCODE_EX2:
      src0 = src_vector(p, &inst->Src[0], fs);

      i915_emit_arith(p,
                      A0_EXP,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, X, X, X, X), 0, 0);
      break;

   case TGSI_OPCODE_FLR:
      emit_simple_arith(p, inst, A0_FLR, 1, fs);
      break;

   case TGSI_OPCODE_FRC:
      emit_simple_arith(p, inst, A0_FRC, 1, fs);
      break;

   case TGSI_OPCODE_KILL_IF:
      /* kill if src[0].x < 0 || src[0].y < 0 ... */
      src0 = src_vector(p, &inst->Src[0], fs);
      tmp = i915_get_utemp(p);

      i915_emit_texld(p,
                      tmp,                   /* dest reg: a dummy reg */
                      A0_DEST_CHANNEL_ALL,   /* dest writemask */
                      0,                     /* sampler */
                      src0,                  /* coord*/
                      T0_TEXKILL,            /* opcode */
                      1);                    /* num_coord */
      break;

   case TGSI_OPCODE_KILL:
      /* unconditional kill */
      tmp = i915_get_utemp(p);

      i915_emit_texld(p,
                      tmp,                                   /* dest reg: a dummy reg */
                      A0_DEST_CHANNEL_ALL,                   /* dest writemask */
                      0,                                     /* sampler */
                      negate(swizzle(0, ONE, ONE, ONE, ONE), 1, 1, 1, 1), /* coord */
                      T0_TEXKILL,                            /* opcode */
                      1);                                    /* num_coord */
      break;

   case TGSI_OPCODE_LG2:
      src0 = src_vector(p, &inst->Src[0], fs);

      i915_emit_arith(p,
                      A0_LOG,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, X, X, X, X), 0, 0);
      break;

   case TGSI_OPCODE_LIT:
      src0 = src_vector(p, &inst->Src[0], fs);
      tmp = i915_get_utemp(p);

      /* tmp = max( a.xyzw, a.00zw )
       * XXX: Clamp tmp.w to -128..128
       * tmp.y = log(tmp.y)
       * tmp.y = tmp.w * tmp.y
       * tmp.y = exp(tmp.y)
       * result = cmp (a.11-x1, a.1x01, a.1xy1 )
       */
      i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
                      src0, swizzle(src0, ZERO, ZERO, Z, W), 0);

      i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
                      swizzle(tmp, Y, Y, Y, Y), 0, 0);

      i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
                      swizzle(tmp, ZERO, Y, ZERO, ZERO),
                      swizzle(tmp, ZERO, W, ZERO, ZERO), 0);

      i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
                      swizzle(tmp, Y, Y, Y, Y), 0, 0);

      i915_emit_arith(p, A0_CMP,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
                      swizzle(tmp, ONE, X, ZERO, ONE),
                      swizzle(tmp, ONE, X, Y, ONE));

      break;

   case TGSI_OPCODE_LRP:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);
      src2 = src_vector(p, &inst->Src[2], fs);
      flags = get_result_flags(inst);
      tmp = i915_get_utemp(p);

      /* b*a + c*(1-a)
       *
       * b*a + c - ca 
       *
       * tmp = b*a + c, 
       * result = (-c)*a + tmp 
       */
      i915_emit_arith(p, A0_MAD, tmp,
                      flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);

      i915_emit_arith(p, A0_MAD,
                      get_result_vector(p, &inst->Dst[0]),
                      flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
      break;

   case TGSI_OPCODE_MAD:
      emit_simple_arith(p, inst, A0_MAD, 3, fs);
      break;

   case TGSI_OPCODE_MAX:
      emit_simple_arith(p, inst, A0_MAX, 2, fs);
      break;

   case TGSI_OPCODE_MIN:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);
      tmp = i915_get_utemp(p);
      flags = get_result_flags(inst);

      i915_emit_arith(p,
                      A0_MAX,
                      tmp, flags & A0_DEST_CHANNEL_ALL, 0,
                      negate(src0, 1, 1, 1, 1),
                      negate(src1, 1, 1, 1, 1), 0);

      i915_emit_arith(p,
                      A0_MOV,
                      get_result_vector(p, &inst->Dst[0]),
                      flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
      break;

   case TGSI_OPCODE_MOV:
      emit_simple_arith(p, inst, A0_MOV, 1, fs);
      break;

   case TGSI_OPCODE_MUL:
      emit_simple_arith(p, inst, A0_MUL, 2, fs);
      break;

   case TGSI_OPCODE_NOP:
      break;

   case TGSI_OPCODE_POW:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);
      tmp = i915_get_utemp(p);
      flags = get_result_flags(inst);

      /* XXX: masking on intermediate values, here and elsewhere.
       */
      i915_emit_arith(p,
                      A0_LOG,
                      tmp, A0_DEST_CHANNEL_X, 0,
                      swizzle(src0, X, X, X, X), 0, 0);

      i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);

      i915_emit_arith(p,
                      A0_EXP,
                      get_result_vector(p, &inst->Dst[0]),
                      flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
      break;

   case TGSI_OPCODE_RET:
      /* XXX: no-op? */
      break;

   case TGSI_OPCODE_RCP:
      src0 = src_vector(p, &inst->Src[0], fs);

      i915_emit_arith(p,
                      A0_RCP,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, X, X, X, X), 0, 0);
      break;

   case TGSI_OPCODE_RSQ:
      src0 = src_vector(p, &inst->Src[0], fs);

      i915_emit_arith(p,
                      A0_RSQ,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, X, X, X, X), 0, 0);
      break;

   case TGSI_OPCODE_SCS:
      src0 = src_vector(p, &inst->Src[0], fs);
      tmp = i915_get_utemp(p);

      /* 
       * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
       * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
       * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
       * scs.x = DP4 t1, scs_sin_constants
       * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
       * scs.y = DP4 t1, scs_cos_constants
       */
      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_XY, 0,
                      swizzle(src0, X, X, ONE, ONE),
                      swizzle(src0, X, ONE, ONE, ONE), 0);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_ALL, 0,
                      swizzle(tmp, X, Y, X, Y),
                      swizzle(tmp, X, X, ONE, ONE), 0);

      writemask = inst->Dst[0].Register.WriteMask;

      if (writemask & TGSI_WRITEMASK_Y) {
         uint tmp1;

         if (writemask & TGSI_WRITEMASK_X)
            tmp1 = i915_get_utemp(p);
         else
            tmp1 = tmp;

         i915_emit_arith(p,
                         A0_MUL,
                         tmp1, A0_DEST_CHANNEL_ALL, 0,
                         swizzle(tmp, X, Y, Y, W),
                         swizzle(tmp, X, Z, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_DP4,
                         get_result_vector(p, &inst->Dst[0]),
                         A0_DEST_CHANNEL_Y, 0,
                         swizzle(tmp1, W, Z, Y, X),
                         i915_emit_const4fv(p, scs_sin_constants), 0);
      }

      if (writemask & TGSI_WRITEMASK_X) {
         i915_emit_arith(p,
                         A0_MUL,
                         tmp, A0_DEST_CHANNEL_XYZ, 0,
                         swizzle(tmp, X, X, Z, ONE),
                         swizzle(tmp, Z, ONE, ONE, ONE), 0);

         i915_emit_arith(p,
                         A0_DP4,
                         get_result_vector(p, &inst->Dst[0]),
                         A0_DEST_CHANNEL_X, 0,
                         swizzle(tmp, ONE, Z, Y, X),
                         i915_emit_const4fv(p, scs_cos_constants), 0);
      }
      break;

   case TGSI_OPCODE_SEQ:
      /* if we're both >= and <= then we're == */
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);
      tmp = i915_get_utemp(p);

      i915_emit_arith(p,
                      A0_SGE,
                      tmp, A0_DEST_CHANNEL_ALL, 0,
                      src0,
                      src1, 0);

      i915_emit_arith(p,
                      A0_SGE,
                      get_result_vector(p, &inst->Dst[0]),
                      A0_DEST_CHANNEL_ALL, 0,
                      src1,
                      src0, 0);

      i915_emit_arith(p,
                      A0_MUL,
                      get_result_vector(p, &inst->Dst[0]),
                      A0_DEST_CHANNEL_ALL, 0,
                      get_result_vector(p, &inst->Dst[0]),
                      tmp, 0);

      break;

   case TGSI_OPCODE_SGE:
      emit_simple_arith(p, inst, A0_SGE, 2, fs);
      break;

   case TGSI_OPCODE_SIN:
      src0 = src_vector(p, &inst->Src[0], fs);
      tmp = i915_get_utemp(p);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_X, 0,
                      src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);

      i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);

      /* 
       * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
       * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
       * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
       * result = DP4 t1.wzyx, sin_constants
       */
      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_XY, 0,
                      swizzle(tmp, X, X, ONE, ONE),
                      swizzle(tmp, X, ONE, ONE, ONE), 0);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_ALL, 0,
                      swizzle(tmp, X, Y, X, Y),
                      swizzle(tmp, X, X, ONE, ONE), 0);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_ALL, 0,
                      swizzle(tmp, X, Y, Y, W),
                      swizzle(tmp, X, Z, ONE, ONE), 0);

      i915_emit_arith(p,
                      A0_DP4,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(tmp, W, Z, Y, X),
                      i915_emit_const4fv(p, sin_constants), 0);
      break;

   case TGSI_OPCODE_SLE:
      /* like SGE, but swap reg0, reg1 */
      emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
      break;

   case TGSI_OPCODE_SLT:
      emit_simple_arith(p, inst, A0_SLT, 2, fs);
      break;

   case TGSI_OPCODE_SGT:
      /* like SLT, but swap reg0, reg1 */
      emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
      break;

   case TGSI_OPCODE_SNE:
      /* if we're < or > then we're != */
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);
      tmp = i915_get_utemp(p);

      i915_emit_arith(p,
                      A0_SLT,
                      tmp,
                      A0_DEST_CHANNEL_ALL, 0,
                      src0,
                      src1, 0);

      i915_emit_arith(p,
                      A0_SLT,
                      get_result_vector(p, &inst->Dst[0]),
                      A0_DEST_CHANNEL_ALL, 0,
                      src1,
                      src0, 0);

      i915_emit_arith(p,
                      A0_ADD,
                      get_result_vector(p, &inst->Dst[0]),
                      A0_DEST_CHANNEL_ALL, 0,
                      get_result_vector(p, &inst->Dst[0]),
                      tmp, 0);
      break;

   case TGSI_OPCODE_SSG:
      /* compute (src>0) - (src<0) */
      src0 = src_vector(p, &inst->Src[0], fs);
      tmp = i915_get_utemp(p);

      i915_emit_arith(p,
                      A0_SLT,
                      tmp,
                      A0_DEST_CHANNEL_ALL, 0,
                      src0,
                      swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);

      i915_emit_arith(p,
                      A0_SLT,
                      get_result_vector(p, &inst->Dst[0]),
                      A0_DEST_CHANNEL_ALL, 0,
                      swizzle(src0, ZERO, ZERO, ZERO, ZERO),
                      src0, 0);

      i915_emit_arith(p,
                      A0_ADD,
                      get_result_vector(p, &inst->Dst[0]),
                      A0_DEST_CHANNEL_ALL, 0,
                      get_result_vector(p, &inst->Dst[0]),
                      negate(tmp, 1, 1, 1, 1), 0);
      break;

   case TGSI_OPCODE_SUB:
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);

      i915_emit_arith(p,
                      A0_ADD,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      src0, negate(src1, 1, 1, 1, 1), 0);
      break;

   case TGSI_OPCODE_TEX:
      emit_tex(p, inst, T0_TEXLD, fs);
      break;

   case TGSI_OPCODE_TRUNC:
      emit_simple_arith(p, inst, A0_TRC, 1, fs);
      break;

   case TGSI_OPCODE_TXB:
      emit_tex(p, inst, T0_TEXLDB, fs);
      break;

   case TGSI_OPCODE_TXP:
      emit_tex(p, inst, T0_TEXLDP, fs);
      break;

   case TGSI_OPCODE_XPD:
      /* Cross product:
       *      result.x = src0.y * src1.z - src0.z * src1.y;
       *      result.y = src0.z * src1.x - src0.x * src1.z;
       *      result.z = src0.x * src1.y - src0.y * src1.x;
       *      result.w = undef;
       */
      src0 = src_vector(p, &inst->Src[0], fs);
      src1 = src_vector(p, &inst->Src[1], fs);
      tmp = i915_get_utemp(p);

      i915_emit_arith(p,
                      A0_MUL,
                      tmp, A0_DEST_CHANNEL_ALL, 0,
                      swizzle(src0, Z, X, Y, ONE),
                      swizzle(src1, Y, Z, X, ONE), 0);

      i915_emit_arith(p,
                      A0_MAD,
                      get_result_vector(p, &inst->Dst[0]),
                      get_result_flags(inst), 0,
                      swizzle(src0, Y, Z, X, ONE),
                      swizzle(src1, Z, X, Y, ONE),
                      negate(tmp, 1, 1, 1, 0));
      break;

   default:
      i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
      p->error = 1;
      return;
   }

   i915_release_utemps(p);
}