s_log1pf.s

Glibc 2.3.2源代码(解压后有100多M)

FR_AA      = f47 
FR_log2_hi = f47  
// Shared with AA  
FR_BB          = f48
FR_log2_lo     = f48  
// Shared with BB  
FR_S_lo        = f49 
FR_two_negN    = f50  
FR_float_N     = f51 
FR_Q4          = f52 
FR_dummy       = f52  
// Shared with Q4
FR_P4          = f52  
// Shared with Q4
FR_Threshold    = f52
// Shared with Q4
FR_Q3          = f53  
FR_P3          = f53  
// Shared with Q3
FR_Tiny        = f53  
// Shared with Q3
FR_Q2          = f54 
FR_P2          = f54  
// Shared with Q2
FR_1LN10_hi     = f54 
// Shared with Q2
FR_Q1           = f55 
FR_P1           = f55 
// Shared with Q1 
FR_1LN10_lo     = f55 
// Shared with Q1 
FR_P5           = f98 
FR_SCALE        = f98 
FR_Output_X_tmp = f99 

GR_Expo_Range   = r32
GR_Table_Base   = r34
GR_Table_Base1  = r35
GR_Table_ptr    = r36 
GR_Index2       = r37 
GR_signif       = r38 
GR_X_0          = r39 
GR_X_1          = r40 
GR_X_2          = r41 
GR_Z_1          = r42 
GR_Z_2          = r43 
GR_N            = r44 
GR_Bias         = r45 
GR_M            = r46 
GR_ScaleN       = r47  
GR_Index3       = r48 
GR_Perturb      = r49 
GR_Table_Scale  = r50 


GR_SAVE_PFS     = r51
GR_SAVE_B0      = r52
GR_SAVE_GP      = r53

GR_Parameter_X       = r54
GR_Parameter_Y       = r55
GR_Parameter_RESULT  = r56

GR_Parameter_TAG = r57 


.section .text
.proc log1pf#
.global log1pf#
.align 64 
log1pf:
#ifdef _LIBC
.global __log1pf
__log1pf:
#endif

{ .mfi
alloc r32 = ar.pfs,0,22,4,0
(p0)  fsub.s1 FR_Neg_One = f0,f1 
(p0)  cmp.eq.unc  p7, p0 = r0, r0 
}

{ .mfi
(p0)  cmp.ne.unc  p14, p0 = r0, r0 
(p0)  fnorm.s1 FR_X_Prime = FR_Input_X 
(p0)  cmp.eq.unc  p15, p0 = r0, r0 ;; 
}

{ .mfi
      nop.m 999
(p0)  fclass.m.unc p6, p0 =  FR_Input_X, 0x1E3 
      nop.i 999
}
;;

{ .mfi
	nop.m 999
(p0)  fclass.nm.unc p10, p0 =  FR_Input_X, 0x1FF 
      nop.i 999
}
;;

{ .mfi
	nop.m 999
(p0)  fcmp.eq.unc.s1 p9, p0 =  FR_Input_X, f0 
      nop.i 999
}

{ .mfi
	nop.m 999
(p0)  fadd FR_Em1 = f0,f0 
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p0)  fadd FR_E = f0,f1 
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p0)  fcmp.eq.unc.s1 p8, p0 =  FR_Input_X, FR_Neg_One 
	nop.i 999
}

{ .mfi
	nop.m 999
(p0)  fcmp.lt.unc.s1 p13, p0 =  FR_Input_X, FR_Neg_One 
	nop.i 999
}


L(LOG_BEGIN): 

{ .mfi
	nop.m 999
(p0)  fadd.s1 FR_Z = FR_X_Prime, FR_E 
	nop.i 999
}

{ .mlx
	nop.m 999
(p0)  movl GR_Table_Scale = 0x0000000000000018 ;; 
}

{ .mmi
	nop.m 999
//     
//    Create E = 1 and Em1 = 0 
//    Check for X == 0, meaning log(1+0)
//    Check for X < -1, meaning log(negative)
//    Check for X == -1, meaning log(0)
//    Normalize x 
//    Identify NatVals, NaNs, Infs. 
//    Identify EM unsupporteds. 
//    Identify Negative values - us S1 so as
//    not to raise denormal operand exception 
//    Set p15 to true for log1pf
//    Set p14 to false for log1pf
//    Set p7 true for log and log1pf
//    
(p0)  addl GR_Table_Base = @ltoff(Constants_Z_G_H_h1#),gp
      nop.i  999
}

{ .mfi
	nop.m 999
(p0)  fmax.s1 FR_AA = FR_X_Prime, FR_E 
	nop.i 999 ;;
}

{ .mfi
      ld8    GR_Table_Base = [GR_Table_Base]
(p0)  fmin.s1 FR_BB = FR_X_Prime, FR_E 
	nop.i 999
}

{ .mfb
	nop.m 999
(p0)  fadd.s1 FR_W = FR_X_Prime, FR_Em1 
//     
//    Begin load of constants base
//    FR_Z = Z = |x| + E 
//    FR_W = W = |x| + Em1
//    AA = fmax(|x|,E)
//    BB = fmin(|x|,E)
//
(p6)  br.cond.spnt L(LOG_64_special) ;; 
}

{ .mib
	nop.m 999
	nop.i 999
(p10) br.cond.spnt L(LOG_64_unsupported) ;; 
}

{ .mib
	nop.m 999
	nop.i 999
(p13) br.cond.spnt L(LOG_64_negative) ;; 
}

{ .mib
(p0)  getf.sig GR_signif = FR_Z 
	nop.i 999
(p9)  br.cond.spnt L(LOG_64_one) ;; 
}

{ .mib
	nop.m 999
	nop.i 999
(p8)  br.cond.spnt L(LOG_64_zero) ;; 
}

{ .mfi
(p0)  getf.exp GR_N =  FR_Z 
//   
//    Raise possible denormal operand exception 
//    Create Bias
// 
//    This function computes ln( x + e ) 
//    Input  FR 1: FR_X   = FR_Input_X          
//    Input  FR 2: FR_E   = FR_E
//    Input  FR 3: FR_Em1 = FR_Em1 
//    Input  GR 1: GR_Expo_Range = GR_Expo_Range = 1
//    Output FR 4: FR_Y_hi  
//    Output FR 5: FR_Y_lo  
//    Output FR 6: FR_Scale  
//    Output PR 7: PR_Safe  
//
(p0)  fsub.s1 FR_S_lo = FR_AA, FR_Z 
//
//    signif = getf.sig(Z)
//    abs_W = fabs(w)
//
(p0)  extr.u GR_Table_ptr = GR_signif, 59, 4 ;; 
}

{ .mfi
	nop.m 999
(p0)  fmerge.se FR_S_hi =  f1,FR_Z 
(p0)  extr.u GR_X_0 = GR_signif, 49, 15  
}

{ .mmi
      nop.m 999
(p0)  addl GR_Table_Base1 = @ltoff(Constants_Z_G_H_h2#),gp  
      nop.i 999
}
;;

{ .mlx
      ld8    GR_Table_Base1 = [GR_Table_Base1]
(p0)  movl GR_Bias = 0x000000000000FFFF ;; 
}

{ .mfi
	nop.m 999
(p0)  fabs FR_abs_W =  FR_W 
(p0)  pmpyshr2.u GR_Table_ptr = GR_Table_ptr,GR_Table_Scale,0 
}

{ .mfi
	nop.m 999
//    
//    Branch out for special input values 
//    
(p0)  fcmp.lt.unc.s0 p8, p0 =  FR_Input_X, f0 
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    X_0 = extr.u(signif,49,15)
//    Index1 = extr.u(signif,59,4)
//
(p0)  fadd.s1 FR_S_lo = FR_S_lo, FR_BB 
	nop.i 999 ;;
}

{ .mii
	nop.m 999
	nop.i 999 ;;
//
//    Offset_to_Z1 = 24 * Index1
//    For performance, don't use result
//    for 3 or 4 cycles.
//
(p0)  add GR_Table_ptr = GR_Table_ptr, GR_Table_Base ;; 
}
//
//    Add Base to Offset for Z1
//    Create Bias

{ .mmi
(p0)  ld4 GR_Z_1 = [GR_Table_ptr],4 ;; 
(p0)  ldfs  FR_G = [GR_Table_ptr],4 
	nop.i 999 ;;
}

{ .mmi
(p0)  ldfs  FR_H = [GR_Table_ptr],8 ;; 
(p0)  ldfd  FR_h = [GR_Table_ptr],0 
(p0)  pmpyshr2.u GR_X_1 = GR_X_0,GR_Z_1,15 
}
//
//    Load Z_1 
//    Get Base of Table2 
//

{ .mfi
(p0)  getf.exp GR_M = FR_abs_W 
	nop.f 999
	nop.i 999 ;;
}

{ .mii
	nop.m 999
	nop.i 999 ;;
//
//    M = getf.exp(abs_W)
//    S_lo = AA - Z
//    X_1 = pmpyshr2(X_0,Z_1,15)
//
(p0)  sub GR_M = GR_M, GR_Bias ;; 
}
//     
//    M = M - Bias
//    Load G1
//    N = getf.exp(Z)
//

{ .mii
(p0)  cmp.gt.unc  p11, p0 =  -80, GR_M 
(p0)  cmp.gt.unc  p12, p0 =  -7, GR_M ;; 
(p0)  extr.u GR_Index2 = GR_X_1, 6, 4 ;; 
}

{ .mib
	nop.m 999
//
//    if -80 > M, set p11
//    Index2 = extr.u(X_1,6,4)
//    if -7  > M, set p12
//    Load H1
//
(p0)  pmpyshr2.u GR_Index2 = GR_Index2,GR_Table_Scale,0 
(p11) br.cond.spnt L(log1pf_small) ;; 
}

{ .mib
      nop.m 999
	nop.i 999
(p12) br.cond.spnt L(log1pf_near) ;; 
}

{ .mii
(p0)  sub GR_N = GR_N, GR_Bias 
//
//    poly_lo = r * poly_lo 
//
(p0)  add GR_Perturb = 0x1, r0 ;; 
(p0)  sub GR_ScaleN = GR_Bias, GR_N  
}

{ .mii
(p0)  setf.sig FR_float_N = GR_N 
	nop.i 999 ;;
//
//    Prepare Index2 - pmpyshr2.u(X_1,Z_2,15)
//    Load h1
//    S_lo = S_lo + BB 
//    Branch for -80 > M
//   
(p0)  add GR_Index2 = GR_Index2, GR_Table_Base1
}

{ .mmi
(p0)  setf.exp FR_two_negN = GR_ScaleN 
      nop.m 999
(p0)  addl GR_Table_Base = @ltoff(Constants_Z_G_H_h3#),gp  
};;

//
//    Index2 points to Z2
//    Branch for -7 > M
//

{ .mmb
(p0)  ld4 GR_Z_2 = [GR_Index2],4 
      ld8 GR_Table_Base = [GR_Table_Base]
      nop.b 999 ;;
}
(p0)  nop.i 999
//
//    Load Z_2
//    N = N - Bias
//    Tablebase points to Table3
//

{ .mmi
(p0)  ldfs  FR_G_tmp = [GR_Index2],4 ;; 
//
//    Load G_2
//    pmpyshr2  X_2= (X_1,Z_2,15)
//    float_N = setf.sig(N)
//    ScaleN = Bias - N
//
(p0)  ldfs  FR_H_tmp = [GR_Index2],8 
	nop.i 999 ;;
}
//
//    Load H_2
//    two_negN = setf.exp(scaleN)
//    G = G_1 * G_2
//

{ .mfi
(p0)  ldfd  FR_h_tmp = [GR_Index2],0 
	nop.f 999
(p0)  pmpyshr2.u GR_X_2 = GR_X_1,GR_Z_2,15 ;; 
}

{ .mii
	nop.m 999
(p0)  extr.u GR_Index3 = GR_X_2, 1, 5 ;; 
//
//    Load h_2
//    H = H_1 + H_2 
//    h = h_1 + h_2 
//    Index3 = extr.u(X_2,1,5)
//
(p0)  shladd GR_Index3 = GR_Index3,4,GR_Table_Base 
}

{ .mmi
	nop.m 999
	nop.m 999
//
//    float_N = fcvt.xf(float_N)
//    load G3
//
(p0)  addl GR_Table_Base = @ltoff(Constants_Q#),gp ;; 
}

{ .mfi
ld8    GR_Table_Base = [GR_Table_Base]
nop.f 999
nop.i 999
} ;;

{ .mfi
(p0)  ldfe FR_log2_hi = [GR_Table_Base],16 
(p0)  fmpy.s1 FR_S_lo = FR_S_lo, FR_two_negN 
	nop.i 999 ;;
}

{ .mmf
	nop.m 999
//
//    G = G3 * G
//    Load h3
//    Load log2_hi
//    H = H + H3
//
(p0)  ldfe FR_log2_lo = [GR_Table_Base],16 
(p0)  fmpy.s1 FR_G = FR_G, FR_G_tmp ;; 
}

{ .mmf
(p0)  ldfs  FR_G_tmp = [GR_Index3],4 
//
//    h = h + h3
//    r = G * S_hi + 1 
//    Load log2_lo
//
(p0)  ldfe FR_Q4 = [GR_Table_Base],16 
(p0)  fadd.s1 FR_h = FR_h, FR_h_tmp ;; 
}

{ .mfi
(p0)  ldfe FR_Q3 = [GR_Table_Base],16 
(p0)  fadd.s1 FR_H = FR_H, FR_H_tmp 
	nop.i 999 ;;
}

{ .mmf
(p0)  ldfs  FR_H_tmp = [GR_Index3],4 
(p0)  ldfe FR_Q2 = [GR_Table_Base],16 
//
//    Comput Index for Table3
//    S_lo = S_lo * two_negN
//
(p0)  fcvt.xf FR_float_N = FR_float_N ;; 
}
//
//    If S_lo == 0, set p8 false
//    Load H3
//    Load ptr to table of polynomial coeff.
//

{ .mmf
(p0)  ldfd  FR_h_tmp = [GR_Index3],0 
(p0)  ldfe FR_Q1 = [GR_Table_Base],0 
(p0)  fcmp.eq.unc.s1 p0, p8 =  FR_S_lo, f0 ;; 
}

{ .mfi
	nop.m 999
(p0)  fmpy.s1 FR_G = FR_G, FR_G_tmp 
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p0)  fadd.s1 FR_H = FR_H, FR_H_tmp 
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p0)  fms.s1 FR_r = FR_G, FR_S_hi, f1 
	nop.i 999
}

{ .mfi
	nop.m 999
(p0)  fadd.s1 FR_h = FR_h, FR_h_tmp 
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p0)  fma.s1 FR_Y_hi = FR_float_N, FR_log2_hi, FR_H 
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Load Q4 
//    Load Q3