s_expm1f.s

glibc 2.9,最新版的C语言库函数

(p14) fms.s.s0        f8 = f0, f0, f1       // result if x = -inf
(p14) br.ret.spnt     b0                    // exit here if x = -inf
}
;;

{ .mfb
      ldfpd           fA4, fA3 = [rTblAddr], 16
      fclass.m        p15, p0 = f8 , 0x1e1  // test for NaT,NaN,+Inf
(p13) br.ret.spnt     b0                    // exit here if x =0.0, result is x
}
;;

{ .mfi
      // overflow thresholds
      ldfps           fMIN_SGL_OFLOW_ARG, fMAX_SGL_NORM_ARG = [rTblAddr], 8
      fma.s1          fXsq = fNormX, fNormX, f0      // x^2 for small path
      and             rExp_x = rExp_mask, rSignexp_x // Biased exponent of x
}
{ .mlx
      nop.m           0
      movl            rM1_lim = 0xc1c00000  // Minus -1 limit (-24.0), SP
}
;;

{ .mfi
      setf.exp        fA2 = rExp_half
      // x*(64/ln(2)) + Right Shifter
      fma.s1          fNint = fNormX, f64DivLn2, fRightShifter
      sub             rExp_x = rExp_x, rExp_bias     // True exponent of x
}
{ .mfb
      nop.m           0
(p15) fma.s.s0        f8 = f8, f1, f0       // result if x = NaT,NaN,+Inf
(p15) br.ret.spnt     b0                    // exit here if x = NaT,NaN,+Inf
}
;;

{ .mfi
      setf.s          fMAX_SGL_MINUS_1_ARG = rM1_lim // -1 threshold, -24.0
      nop.f           0
      cmp.gt          p7, p8 = -2, rExp_x      // Test |x| < 2^(-2)
}
;;

{ .mfi
(p7)  cmp.gt.unc      p6, p7 = -40, rExp_x     // Test |x| < 2^(-40)
      fma.s1          fA87 = fA8, fNormX, fA7  // Small path, A8*x+A7
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fA65 = fA6, fNormX, fA5  // Small path, A6*x+A5
      nop.i           0
}
;;

{ .mfb
      nop.m           0
(p6)  fma.s.s0        f8 = f8, f8, f8          // If x < 2^-40, result=x+x*x
(p6)  br.ret.spnt     b0                       // Exit if x < 2^-40
}
;;

{ .mfi
      nop.m           0
      // check for overflow
      fcmp.gt.s1      p15, p14 = fNormX, fMIN_SGL_OFLOW_ARG
      nop.i           0
}
{ .mfi
      nop.m           0
      fms.s1          fN = fNint, f1, fRightShifter // n in FP register
      nop.i           0
}
;;

{ .mfi
      nop.m           0
(p7)  fma.s1          fA43 = fA4, fNormX, fA3   // Small path, A4*x+A3
      nop.i           0
}
;;

{ .mfi
      getf.sig        rNJ = fNint               // bits of n, j
(p7)  fma.s1          fA8765 = fA87, fXsq, fA65 // Small path, A87*xsq+A65
      nop.i           0
}
{ .mfb
      nop.m           0
(p7)  fma.s1          fX3 = fXsq, fNormX, f0    // Small path, x^3
      // branch out if overflow
(p15) br.cond.spnt    EXPM1_CERTAIN_OVERFLOW
}
;;

{ .mfi
      addl            rN = 0xffff-63, rNJ    // biased and shifted n
      fnma.s1         fR = fLn2Div64, fN, fNormX // R = x - N*ln(2)/64
      extr.u          rJ = rNJ , 0 , 6       // bits of j
}
;;

{ .mfi
      shladd          rJ = rJ, 3, rTblAddr   // address in the 2^(j/64) table
      // check for certain -1
      fcmp.le.s1      p13, p0 = fNormX, fMAX_SGL_MINUS_1_ARG
      shr             rN = rN, 6             // biased n
}
{ .mfi
      nop.m           0
(p7)  fma.s1          fA432 = fA43, fNormX, fA2 // Small path, A43*x+A2
      nop.i           0
}
;;

{ .mfi
      ld8             rJ = [rJ]
      nop.f           0
      shl             rN = rN , 52           // 2^n bits in DP format
}
;;

{ .mmi
      or              rN = rN, rJ        // bits of 2^n * 2^(j/64) in DP format
(p13) mov             rTmp = 1           // Make small value for -1 path
      nop.i           0
}
;;

{ .mfi
      setf.d          fT = rN            // 2^n
      // check for possible overflow (only happens if input higher precision)
(p14) fcmp.gt.s1      p14, p0 = fNormX, fMAX_SGL_NORM_ARG
      nop.i           0
}
{ .mfi
      nop.m           0
(p7)  fma.s1          fA8765432 = fA8765, fX3, fA432 // A8765*x^3+A432
      nop.i           0
}
;;

{ .mfi
(p13) setf.exp        fTmp = rTmp        // Make small value for -1 path
      fma.s1          fP = fA3, fR, fA2  // A3*R + A2
      nop.i           0
}
{ .mfb
      nop.m           0
      fma.s1          fRSqr = fR, fR, f0 // R^2
(p13) br.cond.spnt    EXPM1_CERTAIN_MINUS_ONE // Branch if x < -24.0
}
;;

{ .mfb
      nop.m           0
(p7)  fma.s.s0        f8 = fA8765432, fXsq, fNormX // Small path, 
                                         // result=xsq*A8765432+x
(p7)  br.ret.spnt     b0                 // Exit if 2^-40 <= |x| < 2^-2
}
;;

{ .mfi
      nop.m           0
      fma.s1          fP = fP, fRSqr, fR // P = (A3*R + A2)*Rsqr + R
      nop.i           0
}
;;

{ .mfb
      nop.m           0
      fms.s1          fTm1 = fT, f1, f1  // T - 1.0
(p14) br.cond.spnt    EXPM1_POSSIBLE_OVERFLOW
}
;;

{ .mfb
      nop.m           0
      fma.s.s0        f8 = fP, fT, fTm1
      br.ret.sptk     b0                 // Result for main path
                                         // minus_one_limit < x < -2^-2
                                         // and +2^-2 <= x < overflow_limit
}
;;

// Here if x unorm
EXPM1_UNORM:
{ .mfb
      getf.exp        rSignexp_x = fNormX // Must recompute if x unorm
      fcmp.eq.s0      p6, p0 = f8, f0     // Set D flag
      br.cond.sptk    EXPM1_COMMON
}
;;

// here if result will be -1 and inexact, x <= -24.0
EXPM1_CERTAIN_MINUS_ONE:
{ .mfb
      nop.m           0
      fms.s.s0        f8 = fTmp, fTmp, f1  // Result -1, and Inexact set
      br.ret.sptk     b0
}
;;

EXPM1_POSSIBLE_OVERFLOW:

// Here if fMAX_SGL_NORM_ARG < x < fMIN_SGL_OFLOW_ARG
// This cannot happen if input is a single, only if input higher precision.
// Overflow is a possibility, not a certainty.

// Recompute result using status field 2 with user's rounding mode,
// and wre set.  If result is larger than largest single, then we have
// overflow

{ .mfi
      mov             rGt_ln  = 0x1007f // Exponent for largest sgl + 1 ulp
      fsetc.s2        0x7F,0x42         // Get user's round mode, set wre
      nop.i           0
}
;;

{ .mfi
      setf.exp        fGt_pln = rGt_ln  // Create largest single + 1 ulp
      fma.s.s2        fWre_urm_f8 = fP, fT, fTm1  // Result with wre set
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fsetc.s2        0x7F,0x40                   // Turn off wre in sf2
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fcmp.ge.s1      p6, p0 =  fWre_urm_f8, fGt_pln // Test for overflow
      nop.i           0
}
;;

{ .mfb
      nop.m           0
      nop.f           0
(p6)  br.cond.spnt    EXPM1_CERTAIN_OVERFLOW // Branch if overflow
}
;;

{ .mfb
      nop.m           0
      fma.s.s0        f8 = fP, fT, fTm1
      br.ret.sptk     b0                     // Exit if really no overflow
}
;;

// here if overflow
EXPM1_CERTAIN_OVERFLOW:
{ .mmi
      addl            rTmp = 0x1FFFE, r0;;
      setf.exp        fTmp = rTmp
      nop.i 999
}
;;

{ .mfi
      alloc           r32 = ar.pfs, 0, 3, 4, 0 // get some registers
      fmerge.s        FR_X = fNormX,fNormX
      nop.i           0
}
{ .mfb
      mov             GR_Parameter_TAG = 43
      fma.s.s0        FR_RESULT = fTmp, fTmp, f0 // Set I,O and +INF result
      br.cond.sptk    __libm_error_region
}
;;

GLOBAL_IEEE754_END(expm1f)


LOCAL_LIBM_ENTRY(__libm_error_region)
.prologue
{ .mfi
      add   GR_Parameter_Y=-32,sp             // Parameter 2 value
      nop.f 999
.save   ar.pfs,GR_SAVE_PFS
      mov  GR_SAVE_PFS=ar.pfs                 // Save ar.pfs
}
{ .mfi
.fframe 64
      add sp=-64,sp                           // Create new stack
      nop.f 0
      mov GR_SAVE_GP=gp                       // Save gp
};;
{ .mmi
      stfs [GR_Parameter_Y] = FR_Y,16         // Store Parameter 2 on stack
      add GR_Parameter_X = 16,sp              // Parameter 1 address
.save   b0, GR_SAVE_B0
      mov GR_SAVE_B0=b0                       // Save b0
};;
.body
{ .mfi
      stfs [GR_Parameter_X] = FR_X            // Store Parameter 1 on stack
      nop.f 0
      add   GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address
}
{ .mib
      stfs [GR_Parameter_Y] = FR_RESULT       // Store Parameter 3 on stack
      add   GR_Parameter_Y = -16,GR_Parameter_Y
      br.call.sptk b0=__libm_error_support#   // Call error handling function
};;

{ .mmi
      add   GR_Parameter_RESULT = 48,sp
      nop.m 0
      nop.i 0
};;

{ .mmi
      ldfs  f8 = [GR_Parameter_RESULT]       // Get return result off stack
.restore sp
      add   sp = 64,sp                       // Restore stack pointer
      mov   b0 = GR_SAVE_B0                  // Restore return address
};;
{ .mib
      mov   gp = GR_SAVE_GP                  // Restore gp
      mov   ar.pfs = GR_SAVE_PFS             // Restore ar.pfs
      br.ret.sptk     b0                     // Return
};;

LOCAL_LIBM_END(__libm_error_region)


.type   __libm_error_support#,@function
.global __libm_error_support#