w_tgammaf.s

glibc 2.9,最新版的C语言库函数

//
// 1/An for [40; 48)
data8 0xCAA7576DE621FCD5, 0x3F68
LOCAL_OBJECT_END(tgammaf_data)

//==============================================================
// Code
//==============================================================

.section .text
GLOBAL_LIBM_ENTRY(tgammaf)
{ .mfi
      getf.exp      GR_SignExp = f8
      fma.s1        FR_NormX = f8,f1,f0
      addl          GR_ad_Data = @ltoff(tgammaf_data), gp
}
{ .mfi
      mov           GR_ExpOf05 = 0xFFFE
      fcvt.fx.trunc.s1 FR_iXt = f8 // [x]
      mov           GR_Offs = 0 // 2 <= x < 8
};;
{ .mfi
      getf.d        GR_Arg = f8
      fcmp.lt.s1    p14,p15 = f8,f0
      mov           GR_Tbl12Offs = 0
}
{ .mfi
      setf.exp      FR_05 = GR_ExpOf05
      fma.s1        FR_2 = f1,f1,f1 // 2
      mov           GR_Correction = 0
};;
{ .mfi
      ld8           GR_ad_Data = [GR_ad_Data]
      fclass.m      p10,p0 = f8,0x1E7 // is x  NaTVal, NaN, +/-0 or +/-INF?
      tbit.z        p12,p13 = GR_SignExp,16 // p13 if |x| >= 2
}
{ .mfi
      mov           GR_ExpOf1 = 0xFFFF
      fcvt.fx.s1    FR_rs = f8 // round(x)
      and           GR_Exp2Ind = 7,GR_SignExp
};;
.pred.rel "mutex",p14,p15
{ .mfi
(p15) cmp.eq.unc    p11,p0 = GR_ExpOf1,GR_SignExp // p11 if 1 <= x < 2
(p14) fma.s1        FR_1mX = f1,f1,f8 // 1 - |x|
      mov           GR_Sig = 0 // if |x| < 2
}
{ .mfi
(p13) cmp.eq.unc    p7,p0 = 2,GR_Exp2Ind
(p15) fms.s1        FR_1mX = f1,f1,f8 // 1 - |x|
(p13) cmp.eq.unc    p8,p0 = 3,GR_Exp2Ind
};;
.pred.rel "mutex",p7,p8
{ .mfi
(p7)  mov           GR_Offs = 0x7    // 8 <= |x| < 16
      nop.f         0
(p8)  tbit.z.unc    p0,p6 = GR_Arg,51
}
{ .mib
(p13) cmp.lt.unc    p9,p0 = 3,GR_Exp2Ind
(p8)  mov           GR_Offs = 0xE // 16 <= |x| < 32
      // jump if x is NaTVal, NaN, +/-0 or +/-INF?
(p10) br.cond.spnt  tgammaf_spec_args
};;
.pred.rel "mutex",p14,p15
.pred.rel "mutex",p6,p9
{ .mfi
(p9)  mov           GR_Offs = 0x1C // 32 <= |x|
(p14) fma.s1        FR_X2mX = FR_NormX,FR_NormX,FR_NormX // x^2-|x|
(p9)  tbit.z.unc    p0,p8 = GR_Arg,50
}
{ .mfi
      ldfpd         FR_LocalMin,FR_10 = [GR_ad_Data],16
(p15) fms.s1        FR_X2mX = FR_NormX,FR_NormX,FR_NormX // x^2-|x|
(p6)  add           GR_Offs = 0x7,GR_Offs // 24 <= x < 32
};;
.pred.rel "mutex",p8,p12
{ .mfi
      add           GR_ad_Ce = 0x50,GR_ad_Data
(p15) fcmp.lt.unc.s1 p10,p0 = f8,f1 // p10 if 0 <= x < 1
      mov           GR_OvfNzBound = 2
}
{ .mib
      ldfpd         FR_S32,FR_S31 = [GR_ad_Data],16
(p8)  add           GR_Offs = 0x7,GR_Offs // 40 <= |x|
      // jump if 1 <= x < 2
(p11) br.cond.spnt  tgammaf_from_1_to_2
};;
{ .mfi
      shladd        GR_ad_Ce = GR_Offs,4,GR_ad_Ce
      fcvt.xf       FR_Xt = FR_iXt // [x]
(p13) cmp.eq.unc    p7,p0 = r0,GR_Offs // p7 if 2 <= |x| < 8
}
{ .mfi
      shladd        GR_ad_Co = GR_Offs,4,GR_ad_Data
      fma.s1        FR_6 = FR_2,FR_2,FR_2
      mov           GR_ExpOf05 = 0x7FC
};;
{ .mfi
(p13) getf.sig      GR_Sig = FR_iXt // if |x| >= 2
      frcpa.s1      FR_Rcp0,p0 = f1,FR_NormX
(p10) shr           GR_Arg = GR_Arg,51
}
{ .mib
      ldfpd         FR_C01,FR_C11 = [GR_ad_Co],16
(p7)  mov           GR_Correction = 2
      // jump if 0 < x < 1
(p10) br.cond.spnt  tgammaf_from_0_to_1
};;
{ .mfi
      ldfpd         FR_C21,FR_C31 = [GR_ad_Ce],16
      fma.s1        FR_Rq2 = f1,f1,FR_1mX // 2 - |x|
(p14) sub           GR_Correction = r0,GR_Correction
}
{ .mfi
      ldfpd         FR_C41,FR_C51 = [GR_ad_Co],16
(p14) fcvt.xf       FR_rs = FR_rs
(p14) add           GR_ad_SinO = 0x3A0,GR_ad_Data
};;
.pred.rel "mutex",p14,p15
{ .mfi
      ldfpd         FR_C00,FR_C10 = [GR_ad_Ce],16
      nop.f         0
(p14) sub           GR_Sig = GR_Correction,GR_Sig
}
{ .mfi
      ldfpd         FR_C20,FR_C30 = [GR_ad_Co],16
      fma.s1        FR_Rq1 = FR_1mX,FR_2,FR_X2mX // (x-1)*(x-2)
(p15) sub           GR_Sig = GR_Sig,GR_Correction
};;
{ .mfi
(p14) ldfpd         FR_S01,FR_S11 = [GR_ad_SinO],16
      fma.s1        FR_Rq3 = FR_2,f1,FR_1mX // 3 - |x|
      and           GR_RqDeg = 0x6,GR_Sig
}
{ .mfi
      ldfpd         FR_C40,FR_C50 = [GR_ad_Ce],16
(p14) fma.d.s0      FR_X = f0,f0,f8 // set deno flag
      mov           GR_NanBound = 0x30016 // -2^23
};;
.pred.rel "mutex",p14,p15
{ .mfi
(p14) add           GR_ad_SinE = 0x3C0,GR_ad_Data
(p15) fms.s1        FR_r = FR_NormX,f1,FR_Xt // r = x - [x]
      cmp.eq        p8,p0 = 2,GR_RqDeg
}
{ .mfi
      ldfpd         FR_An,FR_OvfBound = [GR_ad_Co]
(p14) fms.s1        FR_r = FR_Xt,f1,FR_NormX // r = |x - [x]|
      cmp.eq        p9,p0 = 4,GR_RqDeg
};;
.pred.rel "mutex",p8,p9
{ .mfi
(p14) ldfpd         FR_S21,FR_S00 = [GR_ad_SinE],16
(p8)  fma.s1        FR_Rq0 = FR_2,f1,FR_1mX // (3-x)
      tbit.z        p0,p6 = GR_Sig,0
}
{ .mfi
(p14) ldfpd         FR_S10,FR_S20 = [GR_ad_SinO],16
(p9)  fma.s1        FR_Rq0 = FR_2,FR_2,FR_1mX // (5-x)
      cmp.eq        p10,p0 = 6,GR_RqDeg
};;
{ .mfi
(p14) getf.s        GR_Arg = f8
(p14) fcmp.eq.unc.s1 p13,p0 = FR_NormX,FR_Xt
(p14) mov           GR_ZeroResBound = 0xC22C // -43
}
{ .mfi
(p14) ldfe          FR_InvAn = [GR_ad_SinE]
(p10) fma.s1        FR_Rq0 = FR_6,f1,FR_1mX // (7-x)
      cmp.eq        p7,p0 = r0,GR_RqDeg
};;
{ .mfi
(p14) cmp.ge.unc    p11,p0 = GR_SignExp,GR_NanBound
      fma.s1        FR_Rq2 = FR_Rq2,FR_6,FR_X2mX // (x-3)*(x-4)
(p14) shl           GR_ZeroResBound = GR_ZeroResBound,16
}
{ .mfb
(p14) mov           GR_OvfNzBound = 0x802
(p14) fms.s1        FR_rs = FR_rs,f1,FR_NormX // rs = round(x) - x
      // jump if  x < -2^23 i.e. x is negative integer
(p11) br.cond.spnt  tgammaf_singularity
};;
{ .mfi
      nop.m         0
(p7)  fma.s1        FR_Rq1 = f0,f0,f1
(p14) shl           GR_OvfNzBound = GR_OvfNzBound,20
}
{ .mfb
      nop.m         0
      fma.s1        FR_Rq3 = FR_Rq3,FR_10,FR_X2mX // (x-5)*(x-6)
      // jump if x is negative integer such that -2^23 < x < 0
(p13) br.cond.spnt  tgammaf_singularity
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C01 = FR_C01,f1,FR_r
(p14) mov           GR_ExpOf05 = 0xFFFE
}
{ .mfi
(p14) cmp.eq.unc    p7,p0 = GR_Arg,GR_OvfNzBound
      fma.s1        FR_C11 = FR_C11,f1,FR_r
(p14) cmp.ltu.unc   p11,p0 = GR_Arg,GR_OvfNzBound
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C21 = FR_C21,f1,FR_r
(p14) cmp.ltu.unc   p9,p0 = GR_ZeroResBound,GR_Arg
}
{ .mfb
      nop.m         0
      fma.s1        FR_C31 = FR_C31,f1,FR_r
      // jump if argument is close to 0 negative
(p11) br.cond.spnt  tgammaf_overflow
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C41 = FR_C41,f1,FR_r
      nop.i         0
}
{ .mfb
      nop.m         0
      fma.s1        FR_C51 = FR_C51,f1,FR_r
      // jump if x is negative noninteger such that -2^23 < x < -43
(p9)  br.cond.spnt  tgammaf_underflow
};;
{ .mfi
      nop.m         0
(p14) fma.s1        FR_rs2 = FR_rs,FR_rs,f0
      nop.i         0 
}
{ .mfb
      nop.m         0
(p14) fma.s1        FR_S01 = FR_rs,FR_rs,FR_S01
      // jump if argument is 0x80200000
(p7)  br.cond.spnt  tgammaf_overflow_near0_bound
};;
{ .mfi
      nop.m         0 
(p6)  fnma.s1       FR_Rq1 = FR_Rq1,FR_Rq0,f0
      nop.i         0 
}
{ .mfi
      nop.m         0 
(p10) fma.s1        FR_Rq2 = FR_Rq2,FR_Rq3,f0
      and           GR_Sig = 0x7,GR_Sig
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C01 = FR_C01,FR_r,FR_C00
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_C11 = FR_C11,FR_r,FR_C10
      cmp.eq        p6,p7 = r0,GR_Sig // p6 if |x| from one of base intervals
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C21 = FR_C21,FR_r,FR_C20
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_C31 = FR_C31,FR_r,FR_C30
(p7)  cmp.lt.unc    p9,p0 = 2,GR_RqDeg
};;
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S11 = FR_rs,FR_rs,FR_S11
      nop.i         0
}
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S21 = FR_rs,FR_rs,FR_S21
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C41 = FR_C41,FR_r,FR_C40
      nop.i         0
}
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S32 = FR_rs2,FR_S32,FR_S31
      nop.i         0
};;
{ .mfi
      nop.m         0 
(p9)  fma.s1        FR_Rq1 = FR_Rq1,FR_Rq2,f0
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_C51 = FR_C51,FR_r,FR_C50
      nop.i         0 
};;
{ .mfi
(p14) getf.exp      GR_SignExp = FR_rs
      fma.s1        FR_C01 = FR_C01,FR_C11,f0
      nop.i         0 
}
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S01 = FR_S01,FR_rs2,FR_S00
      nop.i         0 
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C21 = FR_C21,FR_C31,f0
      nop.i         0
}
{ .mfi
      nop.m         0
      // NR-iteration
(p14) fnma.s1       FR_InvNormX1 = FR_Rcp0,FR_NormX,f1
      nop.i         0
};;
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S11 = FR_S11,FR_rs2,FR_S10
(p14) tbit.z.unc    p11,p12 = GR_SignExp,17 
}
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S21 = FR_S21,FR_rs2,FR_S20
      nop.i         0
};;
{ .mfi
      nop.m         0
(p15) fcmp.lt.unc.s1 p0,p13 = FR_NormX,FR_OvfBound
      nop.i         0
}
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S32 = FR_rs2,FR_S32,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_C41 = FR_C41,FR_C51,f0
      nop.i         0
}
{ .mfi
      nop.m         0
(p7)  fma.s1        FR_An = FR_Rq1,FR_An,f0
      nop.i         0 
};;
{ .mfb
      nop.m         0
      nop.f         0
      // jump if x > 35.04010009765625
(p13) br.cond.spnt  tgammaf_overflow
};;
{ .mfi
      nop.m         0
      // NR-iteration
(p14) fma.s1        FR_InvNormX1 = FR_Rcp0,FR_InvNormX1,FR_Rcp0
      nop.i         0
};;
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S01 = FR_S01,FR_S11,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
(p14) fma.s1        FR_S21 = FR_S21,FR_S32,f0
      nop.i         0
};;
{ .mfi
(p14) getf.exp      GR_SignExp = FR_NormX
      fma.s1        FR_C01 = FR_C01,FR_C21,f0
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_C41 = FR_C41,FR_An,f0
(p14) mov           GR_ExpOf1 = 0x2FFFF
};;
{ .mfi
      nop.m         0
      // NR-iteration
(p14) fnma.s1       FR_InvNormX2 = FR_InvNormX1,FR_NormX,f1
      nop.i         0
};;
.pred.rel "mutex",p11,p12
{ .mfi
      nop.m         0
(p12) fnma.s1       FR_S01 = FR_S01,FR_S21,f0
      nop.i         0
}
{ .mfi
      nop.m         0
(p11) fma.s1        FR_S01 = FR_S01,FR_S21,f0
      nop.i         0
};;

{ .mfi
      nop.m         0 
(p14) fma.s1        FR_GAMMA = FR_C01,FR_C41,f0
(p14) tbit.z.unc    p6,p7 = GR_Sig,0
}
{ .mfb
      nop.m         0
(p15) fma.s.s0      f8 = FR_C01,FR_C41,f0
(p15) br.ret.spnt   b0 // exit for positives
};;
.pred.rel "mutex",p11,p12
{ .mfi
      nop.m         0
(p12) fms.s1        FR_S01 = FR_rs,FR_S01,FR_rs
      nop.i         0
}
{ .mfi
      nop.m         0
(p11) fma.s1        FR_S01 = FR_rs,FR_S01,FR_rs
      nop.i         0
};;
{ .mfi
      nop.m         0
      // NR-iteration
      fma.s1        FR_InvNormX2 = FR_InvNormX1,FR_InvNormX2,FR_InvNormX1
      cmp.eq        p10,p0 = 0x23,GR_Offs
};;
.pred.rel "mutex",p6,p7
{ .mfi
      nop.m         0
(p6)  fma.s1        FR_GAMMA = FR_S01,FR_GAMMA,f0
      cmp.gtu       p8,p0 = GR_SignExp,GR_ExpOf1
}
{ .mfi
      nop.m         0
(p7)  fnma.s1       FR_GAMMA = FR_S01,FR_GAMMA,f0
      cmp.eq        p9,p0 = GR_SignExp,GR_ExpOf1
};;
{ .mfi
      nop.m         0
      // NR-iteration
      fnma.s1       FR_InvNormX1 = FR_InvNormX2,FR_NormX,f1
      nop.i         0
}
{ .mfi
      nop.m         0