libm_lgammaf.s

glibc 2.9,最新版的C语言库函数

};;
{ .mfi
(p13) ldfpd         FR_A2,FR_A1 = [GR_ad_Co],16
      fms.s1        FR_xm2 = FR_xm2,f1,f1
(p14) extr.u        GR_Arg = GR_Sig,60,4
}
{ .mfi
      mov           GR_SignOfGamma = 1 // set sign of gamma(x) to 1
      fcvt.xf       FR_Ntrunc = FR_int_Ntrunc
      nop.i         0
};;
{ .mfi
      ldfd          FR_T = [GR_ad_T]
      fma.s1        FR_r2 = FR_r,FR_r,f0
      shl           GR_ReqBound = GR_ReqBound,3
}
{ .mfi
      add           GR_ad_Co = 0xCA0,GR_ad_Data
      fnma.s1       FR_Req = FR_Xp1,FR_NormX,f0 // -x*(x+1)
(p14) shladd        GR_Arg = GR_Exp,4,GR_Arg
};;
{ .mfi
(p13) ldfd          FR_A0 = [GR_ad_C650]
      fma.s1        FR_Xp3 = FR_2,f1,FR_Xp1 // (x+3)
(p14) cmp.le.unc    p9,p0 = GR_Arg,GR_ReqBound
}
{ .mfi
(p14) add           GR_ad_Ce = 0x20,GR_ad_Co
      fma.s1        FR_Xp4 = FR_2,FR_2,FR_NormX // (x+4)
(p15) add           GR_ad_OvfBound = 0xBB8,GR_ad_Data
};;
{ .mfi
      // load coefficients of polynomial approximation
      // of ln(sin(Pi*xf)/(Pi*xf)), |xf| <= 0.5
(p14) ldfpd         FR_S16,FR_S14 = [GR_ad_Co],16
(p14) fms.s1        FR_Xf = FR_NormX,f1,FR_N  // xf = x - [x]
(p14) sub           GR_SignOfGamma = r0,GR_SignOfGamma // set sign of
                                                       // gamma(x) to -1
}
{ .mfb
(p14) ldfpd         FR_S12,FR_S10 = [GR_ad_Ce],16
      fma.s1        FR_Xp5 = FR_2,FR_2,FR_Xp1 // (x+5)
      // jump if x is from the interval (-9; 0)
(p9)  br.cond.spnt  lgammaf_negrecursion
};;
{ .mfi
(p14) ldfpd         FR_S8,FR_S6 = [GR_ad_Co],16
      fma.s1        FR_P32 = FR_P3,FR_r,FR_P2
      nop.i         0
}
{ .mfb
(p14) ldfpd         FR_S4,FR_S2 = [GR_ad_Ce],16
      fma.s1        FR_x2 = FR_x,FR_x,f0
      // jump if x is from the interval (-2^13; -9)
(p14) br.cond.spnt  lgammaf_negpoly
};;
{ .mfi
      ldfd          FR_OverflowBound = [GR_ad_OvfBound]
(p12) fcvt.xf       FR_N = FR_int_Ln
      // set p9  if signgum is 32-bit int
      // set p10 if signgum is 64-bit int
      cmp.eq        p10,p9 = 8,r34
}
{ .mfi
      nop.m         0
(p12) fma.s1        FR_P10 = FR_P1,FR_r,f1
      nop.i         0
};;
.pred.rel "mutex",p6,p7
.pred.rel "mutex",p9,p10
{ .mfi
      // store sign of gamma(x) as 32-bit int
(p9)  st4           [r33] = GR_SignOfGamma
(p6)  fma.s1        FR_xx = FR_x,FR_xm2,f0
      nop.i         0
}
{ .mfi
      // store sign of gamma(x) as 64-bit int
(p10) st8           [r33] = GR_SignOfGamma
(p7)  fma.s1        FR_xx = f0,f0,FR_x
      nop.i         0
};;
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A9 = FR_A10,FR_x,FR_A9
      nop.i         0
}
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A7 = FR_A8,FR_x,FR_A7
      nop.i         0
};;
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A5 = FR_A6,FR_x,FR_A5
      nop.i         0
}
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A3 = FR_A4,FR_x,FR_A3
      nop.i         0
};;
{ .mfi
      nop.m         0
(p15) fcmp.eq.unc.s1 p8,p0 = FR_NormX,FR_2 // is input argument 2.0?
      nop.i         0
}
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A1 = FR_A2,FR_x,FR_A1
      nop.i         0
};;
{ .mfi
      nop.m         0
(p12) fma.s1        FR_T = FR_N,FR_Ln2,FR_T
      nop.i         0
}
{ .mfi
      nop.m         0
(p12) fma.s1        FR_P32 = FR_P32,FR_r2,FR_P10
      nop.i         0
};;
{ .mfi
      nop.m         0
(p13) fma.s1        FR_x4 = FR_x2,FR_x2,f0
      nop.i         0
}
{ .mfi
      nop.m         0
(p13) fma.s1        FR_x3 = FR_x2,FR_xx,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A7 = FR_A9,FR_x2,FR_A7
      nop.i         0
}
{ .mfb
      nop.m         0
(p8)  fma.s.s0      f8 = f0,f0,f0
(p8)  br.ret.spnt   b0 // fast exit for 2.0
};;
{ .mfi
      nop.m         0
(p6)  fma.s1        FR_A0 = FR_A0,FR_xm2,f0
      nop.i         0
}
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A3 = FR_A5,FR_x2,FR_A3
      nop.i         0
};;
{ .mfi
      nop.m         0
(p15) fcmp.le.unc.s1 p8,p0 = FR_OverflowBound,FR_NormX // overflow test
      nop.i         0
}
{ .mfi
      nop.m         0
(p12) fms.s1        FR_xm05 = FR_NormX,f1,FR_05
      nop.i         0
};;
{ .mfi
      nop.m         0
(p12) fma.s1        FR_Ln = FR_P32,FR_r,FR_T
      nop.i         0
}
{ .mfi
      nop.m         0
(p12) fms.s1        FR_LnSqrt2Pi = FR_LnSqrt2Pi,f1,FR_NormX
      nop.i         0
};;
{ .mfi
      nop.m         0
(p13) fma.s1        FR_A0 = FR_A1,FR_xx,FR_A0
      nop.i         0
}
{ .mfb
      nop.m         0
(p13) fma.s1        FR_A3 = FR_A7,FR_x4,FR_A3
      // jump if result overflows
(p8)  br.cond.spnt  lgammaf_overflow
};;
.pred.rel "mutex",p12,p13
{ .mfi
      nop.m         0
(p12) fma.s.s0      f8 = FR_Ln,FR_xm05,FR_LnSqrt2Pi
      nop.i         0
}
{ .mfb
      nop.m         0
(p13) fma.s.s0      f8 = FR_A3,FR_x3,FR_A0
      br.ret.sptk   b0
};;
// branch for calculating of ln(GAMMA(x)) for 0 < x < 1
//---------------------------------------------------------------------
.align 32
lgammaf_0_1:
{ .mfi
      getf.sig      GR_Ind = FR_Xp1
      fma.s1        FR_r2 = FR_r,FR_r,f0
      mov           GR_fff7 = 0xFFF7
}
{ .mfi
      ldfpd         FR_Ln2,FR_05 = [GR_ad_Data],16
      fma.s1        FR_P32 = FR_P3,FR_r,FR_P2
      // input argument cann't be equal to 1.0
      cmp.eq        p0,p14 = r0,r0
};;
{ .mfi
      getf.exp      GR_Exp = FR_w
      fcvt.xf       FR_N = FR_int_Ln
      add           GR_ad_Co = 0xCE0,GR_ad_Data
}
{ .mfi
      shladd        GR_ad_T = GR_Ind4T,3,GR_ad_Data
      fma.s1        FR_P10 = FR_P1,FR_r,f1
      add           GR_ad_Ce = 0xD00,GR_ad_Data
};;
{ .mfi
      ldfd          FR_T = [GR_ad_T]
      fma.s1        FR_w2 = FR_w,FR_w,f0
      extr.u        GR_Ind = GR_Ind,61,2
}
{ .mfi
      nop.m         0
      fma.s1        FR_Q32 = FR_P3,FR_w,FR_P2
////      add           GR_ad_C0 = 0xB30,GR_ad_Data
      add           GR_ad_C0 = 0xB38,GR_ad_Data
};;
{ .mfi
      and           GR_Exp = GR_Exp,GR_ExpMask
      nop.f         0
      shladd        GR_IndX8 = GR_Ind,3,r0
}
{ .mfi
      shladd        GR_IndX2 = GR_Ind,1,r0
      fma.s1        FR_Q10 = FR_P1,FR_w,f1
      cmp.eq        p6,p15 = 0,GR_Ind
};;
{ .mfi
      shladd        GR_ad_Co = GR_IndX8,3,GR_ad_Co
(p6)  fma.s1        FR_x = f0,f0,FR_NormX
      shladd        GR_ad_C0 = GR_IndX2,4,GR_ad_C0
}
{ .mfi
      shladd        GR_ad_Ce = GR_IndX8,3,GR_ad_Ce
      nop.f         0
(p15) cmp.eq.unc    p7,p8 = 1,GR_Ind
};;
.pred.rel "mutex",p7,p8
{ .mfi
      ldfpd         FR_A8,FR_A7 = [GR_ad_Co],16
(p7)  fms.s1        FR_x = FR_NormX,f1,FR_LocalMin
      cmp.ge        p10,p11 = GR_Exp,GR_fff7
}
{ .mfb
      ldfpd         FR_A6,FR_A5 = [GR_ad_Ce],16
(p8)  fma.s1        FR_x = f1,f1,FR_NormX
      br.cond.sptk  lgamma_0_2_core
};;
// branch for calculating of ln(GAMMA(x)) for 1 <= x < 2
//---------------------------------------------------------------------
.align 32
lgammaf_1_2:
{ .mfi
      add           GR_ad_Co = 0xCF0,GR_ad_Data
      fcmp.eq.s1    p14,p0 = f1,FR_NormX // is input argument 1.0?
      extr.u        GR_Ind = GR_Sig,61,2
}
{ .mfi
      add           GR_ad_Ce = 0xD10,GR_ad_Data
      nop.f         0
////      add           GR_ad_C0 = 0xB40,GR_ad_Data
      add           GR_ad_C0 = 0xB48,GR_ad_Data
};;
{ .mfi
      shladd        GR_IndX8 = GR_Ind,3,r0
      nop.f         0
      shladd        GR_IndX2 = GR_Ind,1,r0
}
{ .mfi
      cmp.eq        p6,p15 = 0,GR_Ind // p6 <- x from [1;1.25)
      nop.f         0
      cmp.ne        p9,p0 = r0,r0
};;
{ .mfi
      shladd        GR_ad_Co = GR_IndX8,3,GR_ad_Co
(p6)  fms.s1        FR_x = FR_NormX,f1,f1 // reduced x for [1;1.25)
      shladd        GR_ad_C0 = GR_IndX2,4,GR_ad_C0
}
{ .mfi
      shladd        GR_ad_Ce = GR_IndX8,3,GR_ad_Ce
(p14) fma.s.s0      f8 = f0,f0,f0
(p15) cmp.eq.unc    p7,p8 = 1,GR_Ind // p7 <- x from [1.25;1.5)
};;
.pred.rel "mutex",p7,p8
{ .mfi
      ldfpd         FR_A8,FR_A7 = [GR_ad_Co],16
(p7)  fms.s1        FR_x = FR_xm2,f1,FR_LocalMin
      nop.i         0
}
{ .mfi
      ldfpd         FR_A6,FR_A5 = [GR_ad_Ce],16
(p8)  fma.s1        FR_x = f0,f0,FR_NormX
(p9)  cmp.eq.unc    p10,p11 = r0,r0
};;
lgamma_0_2_core:
{ .mmi
      ldfpd         FR_A4,FR_A3 = [GR_ad_Co],16
      ldfpd         FR_A2,FR_A1 = [GR_ad_Ce],16
      mov           GR_SignOfGamma = 1 // set sign of gamma(x) to 1
};;
{ .mfi
//      add           GR_ad_C0 = 8,GR_ad_C0
      ldfd          FR_A0 = [GR_ad_C0]
      nop.f         0
      // set p13 if signgum is 32-bit int
      // set p15 if signgum is 64-bit int
      cmp.eq        p15,p13 = 8,r34
};;
.pred.rel "mutex",p13,p15
{ .mmf
      // store sign of gamma(x)
(p13) st4           [r33] = GR_SignOfGamma // as 32-bit int
(p15) st8           [r33] = GR_SignOfGamma // as 64-bit int
(p11) fma.s1        FR_Q32 = FR_Q32,FR_w2,FR_Q10
};;
{ .mfb
      nop.m         0
(p10) fma.s1        FR_P32 = FR_P32,FR_r2,FR_P10
(p14) br.ret.spnt   b0 // fast exit for 1.0
};;
{ .mfi
      nop.m         0
(p10) fma.s1        FR_T = FR_N,FR_Ln2,FR_T
      cmp.eq        p6,p7 = 0,GR_Ind // p6 <- x from [1;1.25)
}
{ .mfi
      nop.m         0
      fma.s1        FR_x2 = FR_x,FR_x,f0
      cmp.eq        p8,p0 = r0,r0 // set p8 to 1 that means we on [1;2]
};;
{ .mfi
      nop.m         0
(p11) fma.s1        FR_Ln = FR_Q32,FR_w,f0
      nop.i         0
}
{ .mfi
      nop.m         0
      nop.f         0
      nop.i         0
};;
.pred.rel "mutex",p6,p7
{ .mfi
      nop.m         0
(p6)  fma.s1        FR_xx = f0,f0,FR_x
      nop.i         0
}
{ .mfi
      nop.m         0
(p7)  fma.s1        FR_xx = f0,f0,f1
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_A7 = FR_A8,FR_x,FR_A7
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_A5 = FR_A6,FR_x,FR_A5
(p9)  cmp.ne        p8,p0 = r0,r0 // set p8 to 0 that means we on [0;1]
};;
{ .mfi
      nop.m         0
      fma.s1        FR_A3 = FR_A4,FR_x,FR_A3
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_A1 = FR_A2,FR_x,FR_A1
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_x4 = FR_x2,FR_x2,f0
      nop.i         0
}
{ .mfi
      nop.m         0
(p10) fma.s1        FR_Ln = FR_P32,FR_r,FR_T
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_A5 = FR_A7,FR_x2,FR_A5
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_A1 = FR_A3,FR_x2,FR_A1
      nop.i         0
};;
.pred.rel "mutex",p9,p8
{ .mfi
      nop.m         0
(p9)  fms.d.s1      FR_A0 = FR_A0,FR_xx,FR_Ln
      nop.i         0
}
{ .mfi
      nop.m         0
(p8)  fms.s1        FR_A0 = FR_A0,FR_xx,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.d.s1      FR_A1 = FR_A5,FR_x4,FR_A1
      nop.i         0
}
{ .mfi
      nop.m         0
      nop.f         0
      nop.i         0
};;
.pred.rel "mutex",p6,p7
{ .mfi
      nop.m         0
(p6)  fma.s.s0      f8 = FR_A1,FR_x2,FR_A0
      nop.i         0