s_cosl.s

glibc 2.9,最新版的C语言库函数

                                       // p12 if i_0=1, N mod 4 = 1,3
}
;;

{ .mfi
        nop.m 999
       fms.s1        FR_s = FR_s, f1, FR_r
        nop.i 999
}
{ .mfi
        nop.m 999
//
//     S = S - r
//     U_2 = U_2 + w
//     load S_1
//
       fma.s1        FR_rsq = FR_r, FR_r, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
       fma.s1        FR_U_2 = FR_U_2, f1, FR_w
        nop.i 999
}
{ .mfi
        nop.m 999
       fmerge.se FR_tmp_result = FR_r, FR_r
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p10)  fma.s1 FR_tmp_result = f0, f1, f1
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     FR_rsq = r * r
//     Save r as the result.
//
       fms.s1        FR_c = FR_s, f1, FR_U_1
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     if ( i_1 ==0) poly = c + S_1*r*r*r
//     else Result = 1
//
(p12)  fnma.s1 FR_tmp_result = FR_tmp_result, f1, f0
        nop.i 999
}
{ .mfi
        nop.m 999
       fma.s1        FR_r = FR_S_1, FR_r, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
       fma.s0        FR_S_1 = FR_S_1, FR_S_1, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     If i_1 != 0, poly = 2**(-67)
//
       fms.s1 FR_c = FR_c, f1, FR_U_2
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     c = c - U_2
//
(p9)   fma.s1 FR_poly = FR_r, FR_rsq, FR_c
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     i_0 != 0, so Result = -Result
//
(p11)  fma.s0 FR_Result = FR_tmp_result, f1, FR_poly
        nop.i 999 ;;
}
{ .mfb
        nop.m 999
(p12)  fms.s0 FR_Result = FR_tmp_result, f1, FR_poly
//
//     if (i_0 == 0),  Result = Result + poly
//     else            Result = Result - poly
//
        br.ret.sptk   b0         // Exit if |s| < 2^-33, and pi/4 <= |x| < 2^24
}
;;

SINCOSL_LARGER_ARG:
//
// Here if 2^24 <= |x| < 2^63
//
{ .mfi
      ldfe FR_d_1 = [GR_ad_p], 16          // Load d_1 for |x| >= 2^24 path
       fma.s1 FR_N_0 = FR_Input_X, FR_Inv_P_0, f0
        nop.i 999
}
;;

//
//     N_0 = Arg * Inv_P_0
//
//     Load values 2**(-14) and -2**(-14)
{ .mmi
       ldfps FR_Two_to_M14, FR_Neg_Two_to_M14 = [GR_ad_m14]
        nop.i 999 ;;
}
{ .mfi
      ldfe FR_d_2 = [GR_ad_p], 16          // Load d_2 for |x| >= 2^24 path
        nop.f 999
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//
       fcvt.fx.s1 FR_N_0_fix = FR_N_0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     N_0_fix  = integer part of N_0
//
       fcvt.xf FR_N_0 = FR_N_0_fix
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     Make N_0 the integer part
//
       fnma.s1 FR_ArgPrime = FR_N_0, FR_P_0, FR_Input_X
        nop.i 999
}
{ .mfi
        nop.m 999
       fma.s1 FR_w = FR_N_0, FR_d_1, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     Arg' = -N_0 * P_0 + Arg
//     w  = N_0 * d_1
//
       fma.s1 FR_N_float = FR_ArgPrime, FR_Inv_pi_by_2, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     N = A' * 2/pi
//
       fcvt.fx.s1 FR_N_fix = FR_N_float
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     N_fix is the integer part
//
       fcvt.xf FR_N_float = FR_N_fix
        nop.i 999 ;;
}
{ .mfi
       getf.sig GR_N_Inc = FR_N_fix
        nop.f 999
        nop.i 999 ;;
}
{ .mii
        nop.m 999
        nop.i 999 ;;
       add GR_N_Inc = GR_N_Inc, GR_Sin_or_Cos ;;
}
{ .mfi
        nop.m 999
//
//     N is the integer part of the reduced-reduced argument.
//     Put the integer in a GP register
//
       fnma.s1 FR_s = FR_N_float, FR_P_1, FR_ArgPrime
        nop.i 999
}
{ .mfi
        nop.m 999
       fnma.s1 FR_w = FR_N_float, FR_P_2, FR_w
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     s = -N*P_1 + Arg'
//     w = -N*P_2 + w
//     N_fix_gr = N_fix_gr + N_inc
//
       fcmp.lt.unc.s1 p9, p8 = FR_s, FR_Two_to_M14
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p9)   fcmp.gt.s1 p9, p8 = FR_s, FR_Neg_Two_to_M14  // p9 if |s| < 2^-14
        nop.i 999 ;;
}

{ .mfi
        nop.m 999
//
//     For |s|  > 2**(-14) r = S + w (r complete)
//     Else       U_hi = N_0 * d_1
//
(p9)   fma.s1 FR_V_hi = FR_N_float, FR_P_2, f0
        nop.i 999
}
{ .mfi
        nop.m 999
(p9)   fma.s1 FR_U_hi = FR_N_0, FR_d_1, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     Either S <= -2**(-14) or S >= 2**(-14)
//     or -2**(-14) < s < 2**(-14)
//
(p8)   fma.s1 FR_r = FR_s, f1, FR_w
        nop.i 999
}
{ .mfi
        nop.m 999
(p9)   fma.s1 FR_w = FR_N_float, FR_P_3, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     We need abs of both U_hi and V_hi - don't
//     worry about switched sign of V_hi.
//
(p9)   fms.s1 FR_A = FR_U_hi, f1, FR_V_hi
        nop.i 999
}
{ .mfi
        nop.m 999
//
//     Big s: finish up c = (S - r) + w (c complete)
//     Case 4: A =  U_hi + V_hi
//     Note: Worry about switched sign of V_hi, so subtract instead of add.
//
(p9)   fnma.s1 FR_V_lo = FR_N_float, FR_P_2, FR_V_hi
        nop.i 999 ;;
}
{ .mmf
        nop.m 999
        nop.m 999
(p9)   fms.s1 FR_U_lo = FR_N_0, FR_d_1, FR_U_hi
}
{ .mfi
        nop.m 999
(p9)   fmerge.s FR_V_hiabs = f0, FR_V_hi
        nop.i 999 ;;
}
//{ .mfb
//(p9)   fmerge.s f8= FR_V_lo,FR_V_lo
//(p9)   br.ret.sptk b0
//}
//;;
{ .mfi
        nop.m 999
//     For big s: c = S - r
//     For small s do more work: U_lo = N_0 * d_1 - U_hi
//
(p9)   fmerge.s FR_U_hiabs = f0, FR_U_hi
        nop.i 999
}
{ .mfi
        nop.m 999
//
//     For big s: Is |r| < 2**(-3)
//     For big s: if p12 set, prepare to branch to Small_R.
//     For big s: If p13 set, prepare to branch to Normal_R.
//
(p8)   fms.s1 FR_c = FR_s, f1, FR_r
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     For small S: V_hi = N * P_2
//                  w = N * P_3
//     Note the product does not include the (-) as in the writeup
//     so (-) missing for V_hi and w.
//
(p8)   fcmp.lt.unc.s1 p12, p13 = FR_r, FR_Two_to_M3
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p12)  fcmp.gt.s1 p12, p13 = FR_r, FR_Neg_Two_to_M3
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p8)   fma.s1 FR_c = FR_c, f1, FR_w
        nop.i 999
}
{ .mfb
        nop.m 999
(p9)   fms.s1 FR_w = FR_N_0, FR_d_2, FR_w
(p12)  br.cond.spnt SINCOSL_SMALL_R      // Branch if |r| < 2^-3
                                         // and 2^24 <= |x| < 2^63
}
;;

{ .mib
        nop.m 999
        nop.i 999
(p13)  br.cond.sptk SINCOSL_NORMAL_R     // Branch if |r| >= 2^-3
                                         // and 2^24 <= |x| < 2^63
}
;;

SINCOSL_LARGER_S_TINY:
//
// Here if |s| < 2^-14, and 2^24 <= |x| < 2^63
//
{ .mfi
        nop.m 999
//
//     Big s: Vector off when |r| < 2**(-3).  Recall that p8 will be true.
//     The remaining stuff is for Case 4.
//     Small s: V_lo = N * P_2 + U_hi (U_hi is in place of V_hi in writeup)
//     Note: the (-) is still missing for V_lo.
//     Small s: w = w + N_0 * d_2
//     Note: the (-) is now incorporated in w.
//
       fcmp.ge.unc.s1 p7, p8 = FR_U_hiabs, FR_V_hiabs
}
{ .mfi
        nop.m 999
//
//     C_hi = S + A
//
       fma.s1 FR_t = FR_U_lo, f1, FR_V_lo
}
;;

{ .mfi
        nop.m 999
//
//     t = U_lo + V_lo
//
//
(p7)  fms.s1 FR_a = FR_U_hi, f1, FR_A
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p8)  fma.s1 FR_a = FR_V_hi, f1, FR_A
        nop.i 999
}
;;

{ .mfi
//
//     Is U_hiabs >= V_hiabs?
//
        nop.m 999
       fma.s1 FR_C_hi = FR_s, f1, FR_A
        nop.i 999 ;;
}
{ .mmi
       ldfe FR_C_1 = [GR_ad_c], 16 ;;
       ldfe FR_C_2 = [GR_ad_c], 64
        nop.i 999 ;;
}
//
//     c = c + C_lo  finished.
//     Load  C_2
//
{ .mfi
       ldfe        FR_S_1 = [GR_ad_s], 16
//
//     C_lo = S - C_hi
//
       fma.s1 FR_t = FR_t, f1, FR_w
        nop.i 999 ;;
}
//
//     r and c have been computed.
//     Make sure ftz mode is set - should be automatic when using wre
//     |r| < 2**(-3)
//     Get [i_0,i_1] - two lsb of N_fix.
//     Load S_1
//
{ .mfi
       ldfe FR_S_2 = [GR_ad_s], 64
//
//     t = t + w
//
(p7)  fms.s1 FR_a = FR_a, f1, FR_V_hi
      tbit.z p9,p10 = GR_N_Inc, 0      // p9  if i_1=0, N mod 4 = 0,1
                                       // p10 if i_1=1, N mod 4 = 2,3
}
;;
{ .mfi
        nop.m 999
//
//     For larger u than v: a = U_hi - A
//     Else a = V_hi - A (do an add to account for missing (-) on V_hi
//
       fms.s1 FR_C_lo = FR_s, f1, FR_C_hi
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p8)  fms.s1 FR_a = FR_U_hi, f1, FR_a
      tbit.z p11,p12 = GR_N_Inc, 1     // p11 if i_0=0, N mod 4 = 0,2
                                       // p12 if i_0=1, N mod 4 = 1,3
}
;;

{ .mfi
        nop.m 999
//
//     If u > v: a = (U_hi - A)  + V_hi
//     Else      a = (V_hi - A)  + U_hi
//     In each case account for negative missing from V_hi.
//
       fma.s1 FR_C_lo = FR_C_lo, f1, FR_A
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     C_lo = (S - C_hi) + A
//
       fma.s1 FR_t = FR_t, f1, FR_a
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     t = t + a
//
       fma.s1 FR_C_lo = FR_C_lo, f1, FR_t
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     C_lo = C_lo + t
//
       fma.s1 FR_r = FR_C_hi, f1, FR_C_lo
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     Load S_2
//
       fma.s1 FR_rsq = FR_r, FR_r, f0
        nop.i 999
}
{ .mfi
        nop.m 999
//
//     r = C_hi + C_lo
//
       fms.s1 FR_c = FR_C_hi, f1, FR_r
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     if i_1 ==0: poly = S_2 * FR_rsq + S_1
//     else        poly = C_2 * FR_rsq + C_1
//
(p9)  fma.s1 FR_tmp_result = f0, f1, FR_r
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p10)  fma.s1 FR_tmp_result = f0, f1, f1
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     Compute r_cube = FR_rsq * r
//
(p9)  fma.s1 FR_poly = FR_rsq, FR_S_2, FR_S_1
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p10)  fma.s1 FR_poly = FR_rsq, FR_C_2, FR_C_1
        nop.i 999
}
{ .mfi
        nop.m 999
//
//     Compute FR_rsq = r * r
//     Is i_1 == 0 ?
//
       fma.s1 FR_r_cubed = FR_rsq, FR_r, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     c = C_hi - r
//     Load  C_1
//
       fma.s1 FR_c = FR_c, f1, FR_C_lo
        nop.i 999
}
{ .mfi
        nop.m 999
//
//     if i_1 ==0: poly = r_cube * poly + c
//     else        poly = FR_rsq * poly
//
(p12)  fms.s1 FR_tmp_result = f0, f1, FR_tmp_result
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     if i_1 ==0: Result = r
//     else        Result = 1.0
//
(p9)  fma.s1 FR_poly = FR_r_cubed, FR_poly, FR_c
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
(p10)  fma.s1 FR_poly = FR_rsq, FR_poly, f0
        nop.i 999 ;;
}
{ .mfi
        nop.m 999
//
//     if i_0 !=0: Result = -Result
//
(p11)   fma.s0 FR_Result = FR_tmp_result, f1, FR_poly
        nop.i 999 ;;
}
{ .mfb
        nop.m 999
(p12)  fms.s0 FR_Result = FR_tmp_result, f1, FR_poly
//
//     if i_0 == 0: Result = Result + poly
//     else         Result = Result - poly
//
      br.ret.sptk   b0         // Exit for |s| < 2^-14, and 2^24 <= |x| < 2^63
}
;;


SINCOSL_SMALL_R:
//
// Here if |r| < 2^-3
//
// Enter with r, c, and N_Inc computed
//
//      Compare both i_1 and i_0 with 0.
//      if i_1 == 0, set p9.
//      if i_0 == 0, set p11.
//

{ .mfi
      nop.m 999
      fma.s1 FR_rsq = FR_r, FR_r, f0   // rsq = r * r
      tbit.z p9,p10 = GR_N_Inc, 0      // p9  if i_1=0, N mod 4 = 0,1
                                       // p10 if i_1=1, N mod 4 = 2,3
}
;;

{ .mmi
(p9)  ldfe FR_S_5 = [GR_ad_se], -16    // Load S_5 if i_1=0
(p10) ldfe FR_C_5 = [GR_ad_ce], -16    // Load C_5 if i_1=1
      nop.i 999