w_tgamma.s

glibc 2.9,最新版的C语言库函数

      fcmp.lt.s1    p11,p0 = FR_05,FR_r
      mov           GR_185 = 185
}
{ .mfi
      nop.m         0
      fma.s1        FR_Rq6 = FR_Rq6,FR_12,FR_X2pX // (x-11)*(x-12)
      nop.i         0
};;
{ .mfi
      ldfe          FR_CA0 = [GR_ad_Ce7],32
      fma.s1        FR_C31 = FR_C31,f1,FR_r
(p12) mov           GR_iSig = 0
}
{ .mfi
      ldfe          FR_An = [GR_ad_Co7],0x80
      fma.s1        FR_C41 = FR_C41,f1,FR_r
      nop.i         0
};;
{ .mfi
(p14) getf.sig      GR_Sig = FR_r
      fma.s1        FR_C51 = FR_C51,f1,FR_r
(p14) sub           GR_iSig = r0,GR_iSig
}
{ .mfi
      ldfe          FR_S21 = [GR_ad_Co],32
      fma.s1        FR_C61 = FR_C61,f1,FR_r
      nop.i         0
};;
{ .mfi
      ldfe          FR_S19 = [GR_ad_Ce],32
      fma.s1        FR_C71 = FR_C71,f1,FR_r
      and           GR_SigRqLin = 0xF,GR_iSig
}
{ .mfi
      ldfe          FR_S17 = [GR_ad_Co],32
      fma.s1        FR_C81 = FR_C81,f1,FR_r
      mov           GR_2 = 2
};;
{ .mfi
(p14) ldfe          FR_InvAn = [GR_ad_Co7]
      fma.s1        FR_C91 = FR_C91,f1,FR_r
      // if significand of r is 0 tnan argument is negative integer
(p14) cmp.eq.unc    p12,p0 = r0,GR_Sig
}
{ .mfb
(p8)  sub           GR_SigRqLin = GR_SigRqLin,GR_2 // subtract 2 if 2 <= x < 16
      fma.s1        FR_CA1 = FR_CA1,f1,FR_r
      // jump if x is negative integer such that -2^52 < x < -185
(p12) br.cond.spnt  tgamma_singularity
};;
{ .mfi
      setf.sig      FR_Xt = GR_SigRqLin
(p11) fms.s1        FR_rs = f1,f1,FR_r
(p14) cmp.ltu.unc   p7,p0 = GR_185,GR_iSig
}
{ .mfb
      ldfe          FR_S15 = [GR_ad_Ce],32
      fma.s1        FR_Rq7 = FR_Rq7,FR_14,FR_X2pX // (x-13)*(x-14)
      // jump if x is noninteger such that -2^52 < x < -185
(p7)  br.cond.spnt  tgamma_underflow
};;
{ .mfi
      ldfe          FR_S13 = [GR_ad_Co],48
      fma.s1        FR_C01 = FR_C01,FR_r,FR_C00
      and           GR_Sig2 = 0xE,GR_SigRqLin
}
{ .mfi
      ldfe          FR_S11 = [GR_ad_Ce],48
      fma.s1        FR_C11 = FR_C11,FR_r,FR_C10
      nop.i         0
};;
{ .mfi
      ldfe          FR_S9 = [GR_ad_Co],32
      fma.s1        FR_C21 = FR_C21,FR_r,FR_C20
      // should we mul by polynomial of recursion?
      cmp.eq        p13,p12 = r0,GR_SigRqLin
}
{ .mfi
      ldfe          FR_S7 = [GR_ad_Ce],32
      fma.s1        FR_C31 = FR_C31,FR_r,FR_C30
      nop.i         0
};;
{ .mfi
      ldfe          FR_S5 = [GR_ad_Co],32
      fma.s1        FR_C41 = FR_C41,FR_r,FR_C40
      nop.i         0
}
{ .mfi
      ldfe          FR_S3 = [GR_ad_Ce],32
      fma.s1        FR_C51 = FR_C51,FR_r,FR_C50
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C61 = FR_C61,FR_r,FR_C60
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_C71 = FR_C71,FR_r,FR_C70
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C81 = FR_C81,FR_r,FR_C80
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_C91 = FR_C91,FR_r,FR_C90
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_CA1 = FR_CA1,FR_r,FR_CA0
      nop.i         0
}
{ .mfi
      nop.m         0 
      fma.s1        FR_C01 = FR_C01,FR_C11,f0
      nop.i         0
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C21 = FR_C21,FR_C31,f0
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_rs2 = FR_rs,FR_rs,f0
(p12) cmp.lt.unc    p7,p0 = 2,GR_Sig2 // should mul by FR_Rq2?
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C41 = FR_C41,FR_C51,f0
      nop.i         0 
}
{ .mfi
      nop.m         0
(p7)  fma.s1        FR_Rq1 = FR_Rq1,FR_Rq2,f0
(p12) cmp.lt.unc    p9,p0 = 6,GR_Sig2 // should mul by FR_Rq4?
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C61 = FR_C61,FR_C71,f0
(p15) cmp.eq        p11,p0 = r0,r0
}
{ .mfi
      nop.m         0
(p9)  fma.s1        FR_Rq3 = FR_Rq3,FR_Rq4,f0
(p12) cmp.lt.unc    p8,p0 = 10,GR_Sig2 // should mul by FR_Rq6?
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C81 = FR_C81,FR_C91,f0
      nop.i         0
}
{ .mfi
      nop.m         0
(p8)  fma.s1        FR_Rq5 = FR_Rq5,FR_Rq6,f0
(p14) cmp.ltu       p0,p11 = 0x9,GR_Tbl_Ind
};;
{ .mfi
      nop.m         0 
      fcvt.xf       FR_RqLin = FR_Xt  
      nop.i         0
}
{ .mfi
      nop.m         0
(p11) fma.s1        FR_CA1 = FR_CA1,FR_An,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_S21 = FR_S21,FR_rs2,FR_S19
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_S17 = FR_S17,FR_rs2,FR_S15
      nop.i         0
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C01 = FR_C01,FR_C21,f0
      nop.i         0
}
{ .mfi
      nop.m         0 
      fma.s1        FR_rs4 = FR_rs2,FR_rs2,f0
(p12) cmp.lt.unc    p8,p0 = 4,GR_Sig2 // should mul by FR_Rq3?
};;
{ .mfi
      nop.m         0
(p8)  fma.s1        FR_Rq1 = FR_Rq1,FR_Rq3,f0
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_S13 = FR_S13,FR_rs2,FR_S11
(p12) cmp.lt.unc    p9,p0 = 12,GR_Sig2 // should mul by FR_Rq7?
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C41 = FR_C41,FR_C61,f0
      nop.i         0
}
{ .mfi
      nop.m         0 
(p9)  fma.s1        FR_Rq5 = FR_Rq5,FR_Rq7,f0
      nop.i         0 
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C81 = FR_C81,FR_CA1,f0
      nop.i         0 
}
{ .mfi
      nop.m         0
      fma.s1        FR_S9 = FR_S9,FR_rs2,FR_S7
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_S5 = FR_S5,FR_rs2,FR_S3
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_rs3 = FR_rs2,FR_rs,f0
(p12) tbit.nz.unc   p6,p0 = GR_SigRqLin,0
}
{ .mfi
      nop.m         0
      fma.s1        FR_rs8 = FR_rs4,FR_rs4,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_S21 = FR_S21,FR_rs4,FR_S17
      mov           GR_ExpOf1 = 0x2FFFF
}
{ .mfi
      nop.m         0 
(p6)  fms.s1        FR_RqLin = FR_AbsX,f1,FR_RqLin
(p12) cmp.lt.unc    p8,p0 = 8,GR_Sig2 // should mul by FR_Rq5?
};;
{ .mfi
      nop.m         0 
      fma.s1        FR_C01 = FR_C01,FR_C41,f0
      nop.i         0
}
{ .mfi
      nop.m         0
(p8)  fma.s1        FR_Rq1 = FR_Rq1,FR_Rq5,f0
(p14) cmp.gtu.unc   p7,p0 = GR_Sign_Exp,GR_ExpOf1
};;
{ .mfi
      nop.m         0
      fma.s1        FR_S13 = FR_S13,FR_rs4,FR_S9
      nop.i         0
}
{ .mfi
      nop.m         0
(p7)  fma.s1        FR_C81 = FR_C81,FR_AbsX,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
(p14) fma.s1        FR_AbsXp1 = f1,f1,FR_AbsX // |x|+1
      nop.i         0
}
{ .mfi
      nop.m         0
(p15) fcmp.lt.unc.s1 p0,p10 = FR_AbsX,FR_OvfBound // x >= overflow_boundary 
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_rs7 = FR_rs4,FR_rs3,f0
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_S5 = FR_S5,FR_rs3,FR_rs
      nop.i         0
};;
{ .mib
(p14) cmp.lt        p13,p0 = r0,r0 // set p13 to 0 if x < 0
(p12) cmp.eq.unc    p8,p9 = 1,GR_SigRqLin
(p10) br.cond.spnt  tgamma_spec_res
};;
{ .mfi
      getf.sig      GR_Sig = FR_iXt
(p6)  fma.s1        FR_Rq1 = FR_Rq1,FR_RqLin,f0
      // should we mul by polynomial of recursion?
(p15) cmp.eq.unc    p0,p11 = r0,GR_SigRqLin
}
{ .mfb
      nop.m         0 
      fma.s1        FR_GAMMA = FR_C01,FR_C81,f0
(p11) br.cond.spnt  tgamma_positives
};;
{ .mfi
      nop.m         0
      fma.s1        FR_S21 = FR_S21,FR_rs8,FR_S13
      nop.i         0
}
{ .mfb
      nop.m         0
(p13) fma.d.s0      f8 = FR_C01,FR_C81,f0
(p13) br.ret.spnt   b0
};;
.pred.rel "mutex",p8,p9
{ .mfi
      nop.m         0 
(p9)  fma.s1        FR_GAMMA = FR_GAMMA,FR_Rq1,f0
      tbit.z        p6,p7 = GR_Sig,0 // p6 if sin<0, p7 if sin>0
}
{ .mfi
      nop.m         0 
(p8)  fma.s1        FR_GAMMA = FR_GAMMA,FR_RqLin,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_S21 = FR_S21,FR_rs7,FR_S5
      nop.i         0
};;
.pred.rel "mutex",p6,p7
{ .mfi
      nop.m         0 
(p6)  fnma.s1       FR_GAMMA = FR_GAMMA,FR_S21,f0 
      nop.i         0 
}
{ .mfi
      nop.m         0 
(p7)  fma.s1        FR_GAMMA = FR_GAMMA,FR_S21,f0
      mov           GR_Sig2 = 1
};;
{ .mfi
      nop.m         0
      frcpa.s1      FR_Rcp0,p0 = f1,FR_GAMMA
      cmp.ltu       p13,p0 = GR_Sign_Exp,GR_ExpOf1
};;
// NR method: ineration #1
{ .mfi
(p13) getf.exp      GR_Sign_Exp = FR_AbsX
      fnma.s1       FR_Rcp1 = FR_Rcp0,FR_GAMMA,f1 // t = 1 - r0*x
(p13) shl           GR_Sig2 = GR_Sig2,63
};;
{ .mfi
(p13) getf.sig      GR_Sig = FR_AbsX
      nop.f         0
(p13) mov           GR_NzOvfBound = 0xFBFF
};;
{ .mfi
(p13) cmp.ltu.unc   p8,p0 = GR_Sign_Exp,GR_NzOvfBound // p8 <- overflow
      nop.f         0
(p13) cmp.eq.unc    p9,p0 = GR_Sign_Exp,GR_NzOvfBound
};;
{ .mfb
      nop.m         0
(p13) fma.d.s0      FR_X = f1,f1,f8 // set deno & inexact flags
(p8)  br.cond.spnt  tgamma_ovf_near_0 //tgamma_neg_overflow
};;
{ .mib
      nop.m         0
(p9)  cmp.eq.unc    p8,p0 = GR_Sig,GR_Sig2
(p8)  br.cond.spnt  tgamma_ovf_near_0_boundary //tgamma_neg_overflow
};;
{ .mfi
      nop.m         0
      fma.s1        FR_Rcp1 = FR_Rcp0,FR_Rcp1,FR_Rcp0
      nop.i         0 
};;
// NR method: ineration #2
{ .mfi
      nop.m         0
      fnma.s1       FR_Rcp2 = FR_Rcp1,FR_GAMMA,f1 // t = 1 - r1*x
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_Rcp2 = FR_Rcp1,FR_Rcp2,FR_Rcp1
      nop.i         0
};;
// NR method: ineration #3
{ .mfi
      nop.m         0
      fnma.s1       FR_Rcp3 = FR_Rcp2,FR_GAMMA,f1 // t = 1 - r2*x
      nop.i         0
}
{ .mfi
      nop.m         0
(p13) fma.s1        FR_Rcp2 = FR_Rcp2,FR_AbsXp1,f0
(p14) cmp.ltu       p10,p11 = 0x9,GR_Tbl_Ind
};;
.pred.rel "mutex",p10,p11
{ .mfi
      nop.m         0
(p10) fma.s1        FR_GAMMA = FR_Rcp2,FR_Rcp3,FR_Rcp2
      nop.i         0
}
{ .mfi
      nop.m         0
(p11) fma.d.s0      f8 = FR_Rcp2,FR_Rcp3,FR_Rcp2
      nop.i         0
};;
{ .mfb
      nop.m         0
(p10) fma.d.s0      f8 = FR_GAMMA,FR_InvAn,f0
      br.ret.sptk   b0
};;


// here if x >= 3
//--------------------------------------------------------------------
.align 32
tgamma_positives:
.pred.rel "mutex",p8,p9
{ .mfi
      nop.m         0 
(p9)  fma.d.s0      f8 = FR_GAMMA,FR_Rq1,f0
      nop.i         0
}
{ .mfb
      nop.m         0 
(p8)  fma.d.s0      f8 = FR_GAMMA,FR_RqLin,f0
      br.ret.sptk   b0
};;

// here if 0 < x < 1
//--------------------------------------------------------------------
.align 32
tgamma_from_0_to_2:
{ .mfi
      getf.exp      GR_Sign_Exp = FR_r02
      fms.s1        FR_r = FR_r02,f1,FR_Xmin
      mov           GR_ExpOf025 = 0xFFFD
}
{ .mfi
      add           GR_ad_Co = 0x1200,GR_ad_Data
(p6)  fnma.s1       FR_Rcp1 = FR_Rcp0,FR_NormX,f1  // t = 1 - r0*x
(p6)  mov           GR_Sig2 = 1
};;
{ .mfi
(p6)  getf.sig      GR_Sig = FR_NormX
      nop.f         0 
(p6)  shl           GR_Sig2 = GR_Sig2,63
}
{ .mfi
      add           GR_ad_Ce = 0x1210,GR_ad_Data
      nop.f         0
(p6)  mov           GR_NzOvfBound = 0xFBFF
};;
{ .mfi
      cmp.eq        p8,p0 = GR_Sign_Exp,GR_ExpOf05 // r02 >= 1/2 
      nop.f         0
      cmp.eq        p9,p10 = GR_Sign_Exp,GR_ExpOf025 // r02 >= 1/4