w_tgammaf.s

glibc 2.9,最新版的C语言库函数

(p10) fma.s1        FR_InvNormX2 = FR_InvNormX2,FR_InvAn,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      frcpa.s1      FR_Rcp0,p0 = f1,FR_GAMMA
      nop.i         0
};;
{ .mfi
      nop.m         0
      fms.s1        FR_Multplr = FR_NormX,f1,f1 // x - 1
      nop.i         0
};;
{ .mfi
      nop.m         0
      // NR-iteration
      fnma.s1       FR_Rcp1 = FR_Rcp0,FR_GAMMA,f1
      nop.i         0
};;
.pred.rel "mutex",p8,p9
{ .mfi
      nop.m         0
      // 1/x or 1/(An*x)
(p8)  fma.s1        FR_Multplr = FR_InvNormX2,FR_InvNormX1,FR_InvNormX2
      nop.i         0
}
{ .mfi
      nop.m         0
(p9)  fma.s1        FR_Multplr = f1,f1,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      // NR-iteration
      fma.s1        FR_Rcp1 = FR_Rcp0,FR_Rcp1,FR_Rcp0
      nop.i         0
};;
{ .mfi
      nop.m         0
      // NR-iteration
      fnma.s1       FR_Rcp2 = FR_Rcp1,FR_GAMMA,f1
      nop.i         0
}
{ .mfi
      nop.m         0
      // NR-iteration
      fma.s1        FR_Rcp1 = FR_Rcp1,FR_Multplr,f0
      nop.i         0
};;
{ .mfb
      nop.m         0
      fma.s.s0      f8 = FR_Rcp1,FR_Rcp2,FR_Rcp1
      br.ret.sptk   b0
};;

// here if 0 < x < 1
//--------------------------------------------------------------------
.align 32
tgammaf_from_0_to_1:
{ .mfi
      cmp.lt        p7,p0 = GR_Arg,GR_ExpOf05
      // NR-iteration
      fnma.s1       FR_Rcp1 = FR_Rcp0,FR_NormX,f1
      cmp.eq        p8,p0 = GR_Arg,GR_ExpOf05
}
{ .mfi
      cmp.gt        p9,p0 = GR_Arg,GR_ExpOf05
      fma.s1        FR_r = f0,f0,FR_NormX // reduced arg for (0;1)
      mov           GR_ExpOf025 = 0x7FA       
};;
{ .mfi
      getf.s        GR_ArgNz = f8
      fma.d.s0      FR_X = f0,f0,f8 // set deno flag
      shl           GR_OvfNzBound = GR_OvfNzBound,20
}
{ .mfi
(p8)  mov           GR_Tbl12Offs = 0x80 // 0.5 <= x < 0.75
      nop.f         0
(p7)  cmp.ge.unc    p6,p0 = GR_Arg,GR_ExpOf025
};;
.pred.rel "mutex",p6,p9
{ .mfi
(p9)  mov           GR_Tbl12Offs = 0xC0 // 0.75 <= x < 1
      nop.f         0
(p6)  mov           GR_Tbl12Offs = 0x40 // 0.25 <= x < 0.5
}
{ .mfi
      add           GR_ad_Ce = 0x2C0,GR_ad_Data      
      nop.f         0
      add           GR_ad_Co = 0x2A0,GR_ad_Data
};;
{ .mfi
      add           GR_ad_Co = GR_ad_Co,GR_Tbl12Offs
      nop.f         0
      cmp.lt        p12,p0 = GR_ArgNz,GR_OvfNzBound
}
{ .mib
      add           GR_ad_Ce = GR_ad_Ce,GR_Tbl12Offs
      cmp.eq        p7,p0 = GR_ArgNz,GR_OvfNzBound
      // jump if argument is 0x00200000
(p7)  br.cond.spnt  tgammaf_overflow_near0_bound
};;
{ .mmb
      ldfpd         FR_A7,FR_A6 = [GR_ad_Co],16
      ldfpd         FR_A5,FR_A4 = [GR_ad_Ce],16
      // jump if argument is close to 0 positive
(p12) br.cond.spnt  tgammaf_overflow      
};;
{ .mfi
      ldfpd         FR_A3,FR_A2 = [GR_ad_Co],16
      // NR-iteration
      fma.s1        FR_Rcp1 = FR_Rcp0,FR_Rcp1,FR_Rcp0
      nop.i         0
}
{ .mfb
      ldfpd         FR_A1,FR_A0 = [GR_ad_Ce],16
      nop.f         0
      br.cond.sptk  tgamma_from_0_to_2 
};;

// here if 1 < x < 2
//--------------------------------------------------------------------
.align 32
tgammaf_from_1_to_2:
{ .mfi
      add           GR_ad_Co = 0x2A0,GR_ad_Data
      fms.s1        FR_r = f0,f0,FR_1mX
      shr           GR_TblOffs = GR_Arg,47
}
{ .mfi
      add           GR_ad_Ce = 0x2C0,GR_ad_Data
      nop.f         0
      mov           GR_TblOffsMask = 0x18
};;
{ .mfi
      nop.m         0
      nop.f         0
      and           GR_TblOffs = GR_TblOffs,GR_TblOffsMask 
};;
{ .mfi
      shladd        GR_ad_Co = GR_TblOffs,3,GR_ad_Co
      nop.f         0
      nop.i         0
}
{ .mfi
      shladd        GR_ad_Ce = GR_TblOffs,3,GR_ad_Ce
      nop.f         0
      cmp.eq        p6,p7 = 8,GR_TblOffs
};;
{ .mmi
      ldfpd         FR_A7,FR_A6 = [GR_ad_Co],16
      ldfpd         FR_A5,FR_A4 = [GR_ad_Ce],16
      nop.i         0
};;
{ .mmi
      ldfpd         FR_A3,FR_A2 = [GR_ad_Co],16
      ldfpd         FR_A1,FR_A0 = [GR_ad_Ce],16
      nop.i         0
};;

.align 32
tgamma_from_0_to_2:
{ .mfi
      nop.m         0
(p6)  fms.s1        FR_r = FR_r,f1,FR_LocalMin
      nop.i         0
};;
{ .mfi
      nop.m         0
      // NR-iteration
(p10) fnma.s1       FR_Rcp2 = FR_Rcp1,FR_NormX,f1
      nop.i         0
};;
{ .mfi
      nop.m         0
      fms.s1        FR_r2 = FR_r,FR_r,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_A7 = FR_A7,FR_r,FR_A6
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_A5 = FR_A5,FR_r,FR_A4
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_A3 = FR_A3,FR_r,FR_A2
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_A1 = FR_A1,FR_r,FR_A0
      nop.i         0
};;
{ .mfi
      nop.m         0
      // NR-iteration
(p10) fma.s1        FR_Rcp2 = FR_Rcp1,FR_Rcp2,FR_Rcp1
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_A7 = FR_A7,FR_r2,FR_A5
      nop.i         0
}
{ .mfi
      nop.m         0
      fma.s1        FR_r4 = FR_r2,FR_r2,f0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fma.s1        FR_A3 = FR_A3,FR_r2,FR_A1
      nop.i         0
};;
{ .mfi
      nop.m         0 
(p10) fma.s1        FR_GAMMA = FR_A7,FR_r4,FR_A3
      nop.i         0
}
{ .mfi
      nop.m         0 
(p11) fma.s.s0      f8 = FR_A7,FR_r4,FR_A3
      nop.i         0
};;
{ .mfb
      nop.m         0 
(p10) fma.s.s0      f8 = FR_GAMMA,FR_Rcp2,f0
      br.ret.sptk   b0
};;


// overflow
//--------------------------------------------------------------------
.align 32
tgammaf_overflow_near0_bound:
.pred.rel "mutex",p14,p15
{ .mfi
	  mov           GR_fpsr = ar.fpsr
	  nop.f         0
(p15) mov           r8 = 0x7f8
}
{ .mfi
      nop.m         0
      nop.f         0
(p14) mov           r8 = 0xff8
};;
{ .mfi
	  nop.m         0
	  nop.f         0
	  shl           r8 = r8,20 
};;
{ .mfi
      sub           r8 = r8,r0,1
      nop.f         0
	  extr.u        GR_fpsr = GR_fpsr,10,2 // rounding mode
};;
.pred.rel "mutex",p14,p15
{ .mfi
      // set p8 to 0 in case of overflow and to 1 otherwise
	  // for negative arg: 
	  //    no overflow if rounding mode either Z or +Inf, i.e.
	  //    GR_fpsr > 1
(p14) cmp.lt        p8,p0 = 1,GR_fpsr
      nop.f         0
	  // for positive arg: 
	  //    no overflow if rounding mode either Z or -Inf, i.e.
	  //    (GR_fpsr & 1) == 0
(p15) tbit.z        p0,p8 = GR_fpsr,0
};;
{ .mib
(p8)  setf.s        f8 = r8 // set result to 0x7f7fffff without
                            // OVERFLOW flag raising
      nop.i         0
(p8)  br.ret.sptk   b0
};;

.align 32
tgammaf_overflow:
{ .mfi
      nop.m         0
      nop.f         0
      mov           r8 = 0x1FFFE
};;
{ .mfi
      setf.exp      f9 = r8
      fmerge.s      FR_X = f8,f8
      nop.i         0
};;
.pred.rel "mutex",p14,p15
{ .mfi
      nop.m         0
(p14) fnma.s.s0     f8 = f9,f9,f0 // set I,O and -INF result
      mov           GR_TAG = 261 // overflow
}
{ .mfb
      nop.m         0 
(p15) fma.s.s0      f8 = f9,f9,f0 // set I,O and +INF result
      br.cond.sptk  tgammaf_libm_err
};;

// x is negative integer or +/-0
//--------------------------------------------------------------------
.align 32
tgammaf_singularity:
{ .mfi
      nop.m         0
      fmerge.s      FR_X = f8,f8
      mov           GR_TAG = 262 // negative
}
{ .mfb
      nop.m         0
      frcpa.s0      f8,p0 = f0,f0
      br.cond.sptk  tgammaf_libm_err
};;
// x is negative noninteger with big absolute value
//--------------------------------------------------------------------
.align 32
tgammaf_underflow:
{ .mfi
      mov           r8 = 0x00001
      nop.f         0
      tbit.z        p6,p7 = GR_Sig,0
};;
{ .mfi
      setf.exp      f9 = r8
      nop.f         0
      nop.i         0
};;
.pred.rel "mutex",p6,p7
{ .mfi
      nop.m         0
(p6)  fms.s.s0      f8 = f9,f9,f9
      nop.i         0
}
{ .mfb
      nop.m         0
(p7)  fma.s.s0      f8 = f9,f9,f9
      br.ret.sptk   b0
};;

//  x for natval, nan, +/-inf or +/-0
//--------------------------------------------------------------------
.align 32
tgammaf_spec_args:
{ .mfi
      nop.m         0
      fclass.m      p6,p0 =  f8,0x1E1 // Test x for natval, nan, +inf
      nop.i         0
};;
{ .mfi
      nop.m         0
      fclass.m      p7,p8 =  f8,0x7 // +/-0
      nop.i         0
};;
{ .mfi
      nop.m         0
      fmerge.s      FR_X = f8,f8
      nop.i         0
}
{ .mfb
      nop.m         0
(p6)  fma.s.s0      f8 = f8,f1,f8
(p6)  br.ret.spnt   b0
};;
.pred.rel "mutex",p7,p8
{ .mfi
(p7)  mov           GR_TAG = 262 // negative
(p7)  frcpa.s0      f8,p0 = f1,f8
      nop.i         0 
}
{ .mib
      nop.m         0
      nop.i         0
(p8)  br.cond.spnt  tgammaf_singularity
};;

.align 32
tgammaf_libm_err:
{ .mfi
      alloc        r32 = ar.pfs,1,4,4,0
      nop.f        0
      mov          GR_Parameter_TAG = GR_TAG
};;

GLOBAL_LIBM_END(tgammaf)

LOCAL_LIBM_ENTRY(__libm_error_region)
.prologue
{ .mfi
        add   GR_Parameter_Y=-32,sp             // Parameter 2 value
        nop.f 0
.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
{ .mib
        stfs [GR_Parameter_X] = FR_X           // STORE Parameter 1 on stack 
        add   GR_Parameter_RESULT = 0,GR_Parameter_Y  // Parameter 3 address 
        nop.b 0                                      
}
{ .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
        nop.m 0
        nop.m 0
        add   GR_Parameter_RESULT = 48,sp
};;
{ .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#