libm_sincosl.s

glibc 2.9,最新版的C语言库函数

//
{ .mfi
      nop.m 0
(p8)  fcmp.lt.unc.s1 p12, p13 = FR_r, FR_Two_to_M3
      nop.i 0
};;

{ .mfi
      nop.m 0
(p12) fcmp.gt.s1 p12, p13 = FR_r, FR_Neg_Two_to_M3
      nop.i 0
};;

{ .mfi
      nop.m 0
(p8)  fma.s1 FR_c = FR_c, f1, FR_w
      nop.i 0
}
{ .mfb
      nop.m 0
(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 0
      nop.i 0
(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
//
//    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.
//
{ .mfi
      and GR_N_SinCos = 0x1, GR_N_Inc
      fcmp.ge.unc.s1 p6, p7 = FR_U_hiabs, FR_V_hiabs
      tbit.z p8,p12       = GR_N_Inc, 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_t = FR_U_lo, f1, FR_V_lo //     C_hi = S + A
      nop.i 0
};;

{ .mfi
      sub GR_N_SignS =  GR_N_Inc, GR_N_SinCos
(p6)  fms.s1 FR_a = FR_U_hi, f1, FR_A
      add GR_N_SignC =  GR_N_Inc, GR_N_SinCos
}
{ .mfi
      nop.m 0
(p7)  fma.s1 FR_a = FR_V_hi, f1, FR_A
      nop.i 0
};;

{ .mmf
      ldfe FR_C_1 = [GR_ad_c], 16
      ldfe  FR_S_1 = [GR_ad_s], 16
      fma.s1 FR_C_hi = FR_s, f1, FR_A
};;

{ .mmi
      ldfe FR_C_2 = [GR_ad_c], 64
      ldfe FR_S_2 = [GR_ad_s], 64
(p8)  tbit.z.unc p10,p11   = GR_N_SignC, 1
};;

//
//    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.
//
//    For larger u than v: a = U_hi - A
//    Else a = V_hi - A (do an add to account for missing (-) on V_hi
//
{ .mfi
      nop.m 0
      fma.s1 FR_t = FR_t, f1, FR_w //     t = t + w
(p8)  tbit.z.unc p8,p9    = GR_N_SignS, 1
}
{ .mfi
      nop.m 0
(p6)  fms.s1 FR_a = FR_a, f1, FR_V_hi
      nop.i 0
};;

//
//     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.
//
{ .mfi
      nop.m 0
      fms.s1 FR_C_lo = FR_s, f1, FR_C_hi
(p12) tbit.z.unc p14,p15  = GR_N_SignC, 1
}
{ .mfi
      nop.m 0
(p7)  fms.s1 FR_a = FR_U_hi, f1, FR_a
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_C_lo = FR_C_lo, f1, FR_A //     C_lo = (S - C_hi) + A
(p12) tbit.z.unc p12,p13  = GR_N_SignS, 1
}
{ .mfi
      nop.m 0
      fma.s1 FR_t = FR_t, f1, FR_a //     t = t + a
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_r = FR_C_hi, f1, FR_C_lo
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_C_lo = FR_C_lo, f1, FR_t //     C_lo = C_lo + t
      nop.i 0
};;


{ .mfi
      nop.m 0
      fma.s1 FR_rsq = FR_r, FR_r, f0
      nop.i 0
}
{ .mfi
      nop.m 0
      fms.s1 FR_c = FR_C_hi, f1, FR_r
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_FirstS = f0, f1, FR_r
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_FirstC = f0, f1, f1
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_polyS = FR_rsq, FR_S_2, FR_S_1
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_polyC = FR_rsq, FR_C_2, FR_C_1
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_r_cubed = FR_rsq, FR_r, f0
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_c = FR_c, f1, FR_C_lo
      nop.i 0
};;

.pred.rel "mutex",p9,p15
{ .mfi
      nop.m 0
(p9)  fms.s0 FR_FirstS   = f1, f0, FR_FirstS
      nop.i 0
}
{ .mfi
      nop.m 0
(p15) fms.s0 FR_FirstS   = f1, f0, FR_FirstS
      nop.i 0
};;

.pred.rel "mutex",p11,p13
{ .mfi
      nop.m 0
(p11) fms.s0 FR_FirstC   = f1, f0, FR_FirstC
      nop.i 0
}
{ .mfi
      nop.m 0
(p13) fms.s0 FR_FirstC   = f1, f0, FR_FirstC
      nop.i 0
};;


{ .mfi
      nop.m 0
      fma.s1 FR_polyS = FR_r_cubed, FR_polyS, FR_c
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_polyC = FR_rsq, FR_polyC, f0
      nop.i 0
};;



.pred.rel "mutex",p8,p9
{ .mfi
      nop.m 0
(p8)  fma.s0 FR_ResultS = FR_FirstS, f1, FR_polyS
      nop.i 0
}
{ .mfi
      nop.m 0
(p9)  fms.s0 FR_ResultS = FR_FirstS, f1, FR_polyS
      nop.i 0
};;

.pred.rel "mutex",p10,p11
{ .mfi
      nop.m 0
(p10) fma.s0 FR_ResultC = FR_FirstC, f1, FR_polyC
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fms.s0 FR_ResultC = FR_FirstC, f1, FR_polyC
      nop.i 0
};;



.pred.rel "mutex",p12,p13
{ .mfi
      nop.m 0
(p12) fma.s0 FR_ResultS = FR_FirstC, f1, FR_polyC
      nop.i 0
}
{ .mfi
      nop.m 0
(p13) fms.s0 FR_ResultS = FR_FirstC, f1, FR_polyC
      nop.i 0
};;

.pred.rel "mutex",p14,p15
{ .mfi
      nop.m 0
(p14) fma.s0 FR_ResultC = FR_FirstS, f1, FR_polyS
      nop.i 0
}
{ .mfb
      cmp.eq  p10, p0 = 0x1, GR_Cis
(p15) fms.s0 FR_ResultC = FR_FirstS, f1, FR_polyS
(p10) br.ret.sptk               b0
};;


{ .mmb       // exit for sincosl
      stfe  [sincos_pResSin] =  FR_ResultS
      stfe  [sincos_pResCos] =  FR_ResultC
      br.ret.sptk               b0
};;



SINCOSL_SMALL_R:
//
// Here if |r| < 2^-3
//
// Enter with r, c, and N_Inc computed
//
{ .mfi
      nop.m 0
      fma.s1 FR_rsq = FR_r, FR_r, f0   // rsq = r * r
      nop.i 0
};;

{ .mmi
      ldfe FR_S_5 = [GR_ad_se], -16    // Load S_5
      ldfe FR_C_5 = [GR_ad_ce], -16    // Load C_5
      nop.i 0
};;

{ .mmi
      ldfe FR_S_4 = [GR_ad_se], -16    // Load S_4
      ldfe FR_C_4 = [GR_ad_ce], -16    // Load C_4
      nop.i 0
};;

SINCOSL_SMALL_R_0:
// Entry point for 2^-3 < |x| < pi/4
SINCOSL_SMALL_R_1:
// Entry point for pi/4 < |x| < 2^24 and |r| < 2^-3
{ .mfi
      ldfe   FR_S_3 = [GR_ad_se], -16    // Load S_3
      fma.s1 FR_r6  = FR_rsq, FR_rsq, f0 // Z = rsq * rsq
      tbit.z p7,p11       = GR_N_Inc, 0
}
{ .mfi
      ldfe    FR_C_3 = [GR_ad_ce], -16   // Load C_3
      nop.f 0
      and GR_N_SinCos = 0x1, GR_N_Inc
};;

{ .mfi
      ldfe   FR_S_2 = [GR_ad_se], -16    // Load S_2
      fnma.s1 FR_cC = FR_c, FR_r, f0     // c = -c * r
      sub GR_N_SignS =  GR_N_Inc, GR_N_SinCos
}
{ .mfi
      ldfe   FR_C_2 = [GR_ad_ce], -16    // Load C_2
      nop.f 0
      add GR_N_SignC =  GR_N_Inc, GR_N_SinCos
};;

{ .mmi
      ldfe FR_S_1 = [GR_ad_se], -16    // Load S_1
      ldfe FR_C_1 = [GR_ad_ce], -16    // Load C_1
(p7)  tbit.z.unc p9,p10   = GR_N_SignC, 1
};;

{ .mfi
      nop.m 0
      fma.s1 FR_r7 = FR_r6, FR_r, f0     // Z = Z * r
(p7)  tbit.z.unc p7,p8    = GR_N_SignS, 1
};;

{ .mfi
      nop.m 0
      fma.s1 FR_poly_loS = FR_rsq, FR_S_5, FR_S_4 // poly_lo=rsq*S_5+S_4
(p11) tbit.z.unc p13,p14  = GR_N_SignC, 1
}
{ .mfi
      nop.m 0
      fma.s1 FR_poly_loC = FR_rsq, FR_C_5, FR_C_4 // poly_lo=rsq*C_5+C_4
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_poly_hiS = FR_rsq, FR_S_2, FR_S_1 // poly_hi=rsq*S_2+S_1
(p11) tbit.z.unc p11,p12  = GR_N_SignS, 1
}
{ .mfi
      nop.m 0
      fma.s1 FR_poly_hiC = FR_rsq, FR_C_2, FR_C_1 // poly_hi=rsq*C_2+C_1
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s0 FR_FirstS = FR_r, f1, f0
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s0 FR_FirstC = f1, f1, f0
      nop.i 0
};;


{ .mfi
      nop.m 0
      fma.s1 FR_r6 = FR_r6, FR_rsq, f0
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_r7 = FR_r7, FR_rsq, f0
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_poly_loS = FR_rsq, FR_poly_loS, FR_S_3 // p_lo=p_lo*rsq+S_3
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_poly_loC = FR_rsq, FR_poly_loC, FR_C_3 // p_lo=p_lo*rsq+C_3
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s0 FR_inexact = FR_S_4, FR_S_4, f0     // Dummy op to set inexact
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_poly_hiS = FR_poly_hiS, FR_rsq, f0     // p_hi=p_hi*rsq
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_poly_hiC = FR_poly_hiC, FR_rsq, f0     // p_hi=p_hi*rsq
      nop.i 0
};;

.pred.rel "mutex",p8,p14
{ .mfi
      nop.m 0
(p8)  fms.s0 FR_FirstS   = f1, f0, FR_FirstS
      nop.i 0
}
{ .mfi
      nop.m 0
(p14) fms.s0 FR_FirstS   = f1, f0, FR_FirstS
      nop.i 0
};;

.pred.rel "mutex",p10,p12
{ .mfi
      nop.m 0
(p10) fms.s0 FR_FirstC   = f1, f0, FR_FirstC
      nop.i 0
}
{ .mfi
      nop.m 0
(p12) fms.s0 FR_FirstC   = f1, f0, FR_FirstC
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_polyS = FR_r7, FR_poly_loS, FR_cS        // poly=Z*poly_lo+c
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_polyC = FR_r6, FR_poly_loC, FR_cC        // poly=Z*poly_lo+c
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1 FR_poly_hiS = FR_r, FR_poly_hiS, f0       // p_hi=r*p_hi
      nop.i 0
};;


{ .mfi
      nop.m 0
      fma.s1 FR_polyS = FR_polyS, f1, FR_poly_hiS
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1 FR_polyC = FR_polyC, f1, FR_poly_hiC
      nop.i 0
};;

.pred.rel "mutex",p7,p8
{ .mfi
      nop.m 0
(p7)  fma.s0 FR_ResultS = FR_FirstS, f1, FR_polyS
      nop.i 0
}
{ .mfi
      nop.m 0
(p8)  fms.s0 FR_ResultS = FR_FirstS, f1, FR_polyS
      nop.i 0
};;

.pred.rel "mutex",p9,p10
{ .mfi
      nop.m 0
(p9)  fma.s0 FR_ResultC = FR_FirstC, f1, FR_polyC
      nop.i 0
}
{ .mfi
      nop.m 0
(p10) fms.s0 FR_ResultC = FR_FirstC, f1, FR_polyC
      nop.i 0
};;

.pred.rel "mutex",p11,p12
{ .mfi
      nop.m 0
(p11) fma.s0 FR_ResultS = FR_FirstC, f1, FR_polyC
      nop.i 0
}
{ .mfi
      nop.m 0
(p12) fms.s0 FR_ResultS = FR_FirstC, f1, FR_polyC
      nop.i 0
};;

.pred.rel "mutex",p13,p14
{ .mfi
      nop.m 0
(p13) fma.s0 FR_ResultC = FR_FirstS, f1, FR_polyS
      nop.i 0