libm_sincos_large.s

glibc 2.9,最新版的C语言库函数

      addl           GR_Table_Base   = @ltoff(FSINCOS_CONSTANTS#), gp
      nop.i 999
}
;;

{ .mmi
      ld8 GR_Table_Base = [GR_Table_Base]
      nop.m 999
      nop.i 999
}
;;


//
//     Point to  2*-14
//     N_0 = Arg * Inv_P_0
//

{ .mmi
       add GR_Table_Base = 688, GR_Table_Base ;;
       ldfs FR_Two_to_M14 = [GR_Table_Base], 4
      nop.i 999 ;;
}

{ .mfi
       ldfs FR_Neg_Two_to_M14 = [GR_Table_Base], 0
      nop.f 999
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
//
//     Load values 2**(-14) and -2**(-14)
//
       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
      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 ;;
}

{ .mfi
      nop.m 999
(p9)   fms.s1 FR_U_lo = FR_N_0, FR_d_1, FR_U_hi
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p9)   fmerge.s FR_V_hiabs = f0, FR_V_hi
      nop.i 999
}

{ .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 SINCOS_SMALL_R ;;
}

{ .mib
      nop.m 999
      nop.i 999
(p13)  br.cond.sptk SINCOS_NORMAL_R ;;
}

{ .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.
//
(p9)   fcmp.ge.unc.s1 p10, p11 = FR_U_hiabs, FR_V_hiabs
       extr.u   GR_i_1 = GR_N_Inc, 0, 1 ;;
}

{ .mfi
      nop.m 999
//
//     C_hi = S + A
//
(p9)   fma.s1 FR_t = FR_U_lo, f1, FR_V_lo
       extr.u   GR_i_0 = GR_N_Inc, 1, 1 ;;
}

{ .mfi
      nop.m 999
//
//     t = U_lo + V_lo
//
//
(p10)  fms.s1 FR_a = FR_U_hi, f1, FR_A
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p11)  fma.s1 FR_a = FR_V_hi, f1, FR_A
      nop.i 999
}
;;

{ .mmi
      nop.m 999
      addl           GR_Table_Base   = @ltoff(FSINCOS_CONSTANTS#), gp
      nop.i 999
}
;;

{ .mmi
      ld8 GR_Table_Base = [GR_Table_Base]
      nop.m 999
      nop.i 999
}
;;


{ .mfi
       add GR_Table_Base = 528, GR_Table_Base
//
//     Is U_hiabs >= V_hiabs?
//
(p9)   fma.s1 FR_C_hi = FR_s, f1, FR_A
      nop.i 999 ;;
}

{ .mmi
       ldfe FR_C_1 = [GR_Table_Base], 16 ;;
       ldfe FR_C_2 = [GR_Table_Base], 64
      nop.i 999 ;;
}

{ .mmf
      nop.m 999
//
//     c = c + C_lo  finished.
//     Load  C_2
//
       ldfe FR_S_1 = [GR_Table_Base], 16
//
//     C_lo = S - C_hi
//
       fma.s1 FR_t = FR_t, f1, FR_w ;;
}
//
//     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_Table_Base], 64
//
//     t = t + w
//
(p10)  fms.s1 FR_a = FR_a, f1, FR_V_hi
       cmp.eq.unc p9, p10 = 0x0, GR_i_0
}

{ .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
(p11)  fms.s1 FR_a = FR_U_hi, f1, FR_a
       cmp.eq.unc p11, p12 = 0x0, GR_i_1
}

{ .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
//     Adjust Table_Base to beginning of table
//
       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
//
//     Table_Base points to C_1
//     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
//
//(p11)  fma.s1 FR_Input_X = f0, f1, FR_r
(p11)  fma.s1 FR_prelim = f0, f1, FR_r
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
//(p12)  fma.s1 FR_Input_X = f0, f1, f1
(p12)  fma.s1 FR_prelim = f0, f1, f1
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
//
//     Compute r_cube = FR_rsq * r
//
(p11)  fma.s1 FR_poly = FR_rsq, FR_S_2, FR_S_1
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p12)  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
//
//(p10)  fms.s1 FR_Input_X = f0, f1, FR_Input_X
(p10)  fms.s1 FR_prelim = f0, f1, FR_prelim
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
//
//     if i_1 ==0: Result = r
//     else        Result = 1.0
//
(p11)  fma.s1 FR_poly = FR_r_cubed, FR_poly, FR_c
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p12)  fma.s1 FR_poly = FR_rsq, FR_poly, f0
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
//
//     if i_0 !=0: Result = -Result
//
(p9)   fma.s1 FR_Input_X = FR_prelim, f1, FR_poly
      nop.i 999 ;;
}

{ .mfb
      nop.m 999
(p10)  fms.s1 FR_Input_X = FR_prelim, f1, FR_poly
//
//     if i_0 == 0: Result = Result + poly
//     else         Result = Result - poly
//
       br.ret.sptk   b0 ;;
}
SINCOS_SMALL_R:

{ .mii
      nop.m 999
        extr.u  GR_i_1 = GR_N_Inc, 0, 1 ;;
//
//
//      Compare both i_1 and i_0 with 0.
//      if i_1 == 0, set p9.
//      if i_0 == 0, set p11.
//
        cmp.eq.unc p9, p10 = 0x0, GR_i_1 ;;
}

{ .mfi
      nop.m 999
        fma.s1 FR_rsq = FR_r, FR_r, f0
        extr.u  GR_i_0 = GR_N_Inc, 1, 1 ;;
}

{ .mfi
      nop.m 999
//
//  Z = Z * FR_rsq
//
(p10)   fnma.s1 FR_c = FR_c, FR_r, f0
        cmp.eq.unc p11, p12 = 0x0, GR_i_0
}
;;

// ******************************************************************
// ******************************************************************
// ******************************************************************
//      r and c have been computed.
//      We know whether this is the sine or cosine routine.
//      Make sure ftz mode is set - should be automatic when using wre
//      |r| < 2**(-3)
//
//      Set table_ptr1 to beginning of constant table.
//      Get [i_0,i_1] - two lsb of N_fix_gr.
//

{ .mmi
      nop.m 999
      addl           GR_Table_Base   = @ltoff(FSINCOS_CONSTANTS#), gp
      nop.i 999
}
;;

{ .mmi
      ld8 GR_Table_Base = [GR_Table_Base]
      nop.m 999
      nop.i 999
}
;;


//
//      Set table_ptr1 to point to S_5.
//      Set table_ptr1 to point to C_5.
//      Compute FR_rsq = r * r
//

{ .mfi
(p9)    add GR_Table_Base = 672, GR_Table_Base
(p10)   fmerge.s FR_r = f1, f1
(p10)   add GR_Table_Base = 592, GR_Table_Base ;;
}
//
//      Set table_ptr1 to point to S_5.
//      Set table_ptr1 to point to C_5.
//

{ .mmi
(p9)    ldfe FR_S_5 = [GR_Table_Base], -16 ;;
//
//      if (i_1 == 0) load S_5
//      if (i_1 != 0) load C_5
//
(p9)    ldfe FR_S_4 = [GR_Table_Base], -16
      nop.i 999 ;;
}

{ .mmf
(p10)   ldfe FR_C_5 = [GR_Table_Base], -16
//
//      Z = FR_rsq * FR_rsq
//
(p9)    ldfe FR_S_3 = [GR_Table_Base], -16
//
//      Compute FR_rsq = r * r
//      if (i_1 == 0) load S_4
//      if (i_1 != 0) load C_4
//
        fma.s1 FR_Z = FR_rsq, FR_rsq, f0 ;;
}
//
//      if (i_1 == 0) load S_3
//      if (i_1 != 0) load C_3
//

{ .mmi
(p9)    ldfe FR_S_2 = [GR_Table_Base], -16 ;;
//
//      if (i_1 == 0) load S_2
//      if (i_1 != 0) load C_2
//
(p9)    ldfe FR_S_1 = [GR_Table_Base], -16
      nop.i 999
}

{ .mmi
(p10)   ldfe FR_C_4 = [GR_Table_Base], -16 ;;
(p10)   ldfe FR_C_3 = [GR_Table_Base], -16
      nop.i 999 ;;
}

{ .mmi
(p10)   ldfe FR_C_2 = [GR_Table_Base], -16 ;;
(p10)   ldfe FR_C_1 = [GR_Table_Base], -16
      nop.i 999
}

{ .mfi
      nop.m 999
//
//      if (i_1 != 0):
//      poly_lo = FR_rsq * C_5 + C_4
//      poly_hi = FR_rsq * C_2 + C_1
//
(p9)    fma.s1 FR_Z = FR_Z, FR_r, f0
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
//
//      if (i_1 == 0) load S_1
//      if (i_1 != 0) load C_1
//
(p9)    fma.s1 FR_poly_lo = FR_rsq, FR_S_5, FR_S_4
      nop.i 999
}

{ .mfi
      nop.m 999
//
//      c = -c * r
//      dummy fmpy's to flag inexact.
//
(p9)    fma.d.s1 FR_S_4 = FR_S_4, FR_S_4, f0
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
//
//      poly_lo = FR_rsq * poly_lo + C_3
//      poly_hi = FR_rsq * poly_hi
//
        fma.s1  FR_Z = FR_Z, FR_rsq, f0
      nop.i 999 ;;