s_expm1l.s

Glibc 2.3.2源代码(解压后有100多M)

.global expm1l#
.align 64 
expm1l: 
#ifdef _LIBC
.global __expm1l#
__expm1l:
#endif
{ .mii
alloc r32 = ar.pfs,0,30,4,0
(p0)  add r33 = 1, r0  
(p0)  cmp.eq.unc  p7, p0 =  r0, r0 
}
{ .mbb
	nop.m 999
(p0)  br.cond.sptk exp_continue 
	nop.b 999 ;;
}

//
//    Set p7 true for expm1
//    Set Flag = r33 = 1 for expm1
//    

.endp expm1l
ASM_SIZE_DIRECTIVE(expm1l)

#ifdef _LIBC
libm_hidden_def (__expm1l)
#endif

.section .text
.proc expl#
.global expl#
.align 64 
expl: 
#ifdef _LIBC
.global __ieee754_expl#
__ieee754_expl:
#endif
{ .mii
alloc r32 = ar.pfs,0,30,4,0
(p0)  add r33 = r0, r0  
(p0)  cmp.eq.unc  p0, p7 =  r0, r0 ;; 
}
exp_continue: 
{ .mfi
(p0)  add r32 = 2,r0  
(p0)  fnorm.s1 f9 = f8 
      nop.i 0
}
{ .mfi
(p0)  nop.m 0 
//
//    Set p7 false for exp
//    Set Flag = r33 = 0 for exp
//    
(p0)  fclass.m.unc p6, p8 =  f8, 0x1E7 
      nop.i 0;;
}
{ .mfi
	nop.m 999
(p0)  fclass.nm.unc p9, p0 =  f8, 0x1FF 
      nop.i 0
}
{ .mfi
	nop.m 999
(p0)  mov f36 = f1 
	nop.i 999 ;;
}
{ .mfb
	nop.m 999
//     
//    Identify NatVals, NaNs, Infs, and Zeros. 
//    Identify EM unsupporteds. 
//    Save special input registers 
(p0)  mov f32 = f0 
//
//    Create FR_X_cor      = 0.0 
//           GR_Flag       = 0 
//           GR_Expo_Range = 2 (r32) for double-extended precision 
//           FR_Scale      = 1.0
//
(p6)  br.cond.spnt EXPL_64_SPECIAL ;; 
}
{ .mib
	nop.m 999
	nop.i 999
(p9)  br.cond.spnt EXPL_64_UNSUPPORTED ;; 
}
{ .mfi
(p0)  cmp.ne.unc p12, p13 = 0x01, r33
//     
//    Branch out for special input values 
//     
(p0)  fcmp.lt.unc.s0 p9,p0 =  f8, f0 
(p0)  cmp.eq.unc  p15, p0 =  r0, r0 
}
{ .mmi
	nop.m 999
//     
//    Raise possible denormal operand exception 
//    Normalize x 
//     
//    This function computes expl( x  + x_cor) 
//    Input  FR 1: FR_X            
//    Input  FR 2: FR_X_cor  
//    Input  GR 1: GR_Flag  
//    Input  GR 2: GR_Expo_Range  
//    Output FR 3: FR_Y_hi  
//    Output FR 4: FR_Y_lo  
//    Output FR 5: FR_Scale  
//    Output PR 1: PR_Safe  
(p0)  addl r34 = @ltoff(Constants_exp_64_Arg#),gp  
(p0)  addl r40 = @ltoff(Constants_exp_64_W1#),gp 
};;
//
//    Prepare to load constants
//    Set Safe = True
//

{ .mmi
      ld8  r34 = [r34]
      ld8  r40 = [r40]
(p0)  addl r41 = @ltoff(Constants_exp_64_W2#),gp  
};;

{ .mmi
(p0)  ldfe f37 = [r34],16 
(p0)  ld8 r41 = [r41] ;; 
}

//
//    N = fcvt.fx(float_N)
//    Set p14 if -6 > expo_X 
//
//
//    Bias = 0x0FFFF
//    expo_X = expo_X and Mask  
//

{ .mmi
(p0)  ldfe f40 = [r34],16 
      nop.m 999
//
//    Load L_lo
//    Set p10 if 14 < expo_X 
//
(p0)  addl r50 = @ltoff(Constants_exp_64_T1#),gp 
}
{ .mmi
	nop.m 999
	nop.m 999
(p0)  addl r51 = @ltoff(Constants_exp_64_T2#),gp ;; 
}
//
//    Load W2_ptr
//    Branch to SMALL is expo_X < -6
//

{.mmi
(p0)  ld8 r50 = [r50]  
(p0)  ld8 r51 = [r51]  
};;

{ .mlx
(p0)  ldfe f41 = [r34],16 
//
//    float_N = X * L_Inv
//    expo_X = exponent of X
//    Mask = 0x1FFFF
//
(p0)  movl r58 = 0x0FFFF 
}
{ .mlx
	nop.m 999
(p0)  movl r39 = 0x1FFFF ;; 
}
{ .mmi
(p0)  getf.exp r37 = f9 
	nop.m 999
(p0)  addl r34 = @ltoff(Constants_exp_64_Exponents#),gp ;; 
}
{ .mii
(p0)  ld8 r34 = [r34]  
      nop.i 999 
(p0)  and  r37 = r37, r39 ;;  
}
{ .mmi
(p0)  sub r37 = r37, r58 ;;  
(p0)  cmp.gt.unc  p14, p0 =  -6, r37 
(p0)  cmp.lt.unc  p10, p0 =  14, r37 ;; 
}
{ .mfi
(p0)  nop.m 0  
//
//    Load L_inv 
//    Set p12 true for Flag = 0 (exp)
//    Set p13 true for Flag = 1 (expm1)
//
(p0)  fmpy.s1 f38 = f9, f37 
	nop.i 999 ;;
}
{ .mfb
	nop.m 999
//
//    Load L_hi
//    expo_X = expo_X - Bias
//    get W1_ptr      
//
(p0)  fcvt.fx.s1 f39 = f38
(p14) br.cond.spnt EXPL_SMALL ;; 
}
{ .mib
	nop.m 999
	nop.i 999
(p10) br.cond.spnt EXPL_HUGE ;; 
}
{ .mmi
(p0)  shladd r34 = r32,4,r34 
      nop.m 999
(p0)  addl r35 = @ltoff(Constants_exp_64_A#),gp ;; 
}
//
//    Load T_1,T_2
//
{ .mmi
   nop.m 999
   ld8   r35 =[r35]
   nop.i 99
};;
{ .mmb
(p0)  ldfe f51 = [r35],16 
(p0)  ld8 r45 = [r34],8
	nop.b 999 ;;
}
//    
//    Set Safe = True  if k >= big_expo_neg  
//    Set Safe = False if k < big_expo_neg  
//    
{ .mmb
(p0)  ldfe f49 = [r35],16 
(p0)  ld8 r48 = [r34],0
	nop.b 999 ;;
}
{ .mfi
	nop.m 999
//
//    Branch to HUGE is expo_X > 14 
//
(p0)  fcvt.xf f38 = f39 
	nop.i 999 ;;
}
{ .mfi
(p0)  getf.sig r52 = f39 
	nop.f 999
	nop.i 999 ;;
}
{ .mii
	nop.m 999
(p0)  extr.u r43 = r52, 6, 6 ;;  
//
//    r = r - float_N * L_lo
//    K = extr(N_fix,12,52)
//
(p0)  shladd r40 = r43,3,r40 ;; 
}
{ .mfi
(p0)  shladd r50 = r43,2,r50 
(p0)  fnma.s1 f42 = f40, f38, f9 
//
//    float_N = float(N)
//    N_fix = signficand N 
//
(p0)  extr.u r42 = r52, 0, 6  
}
{ .mmi
(p0)  ldfd  f43 = [r40],0 ;; 
(p0)  shladd r41 = r42,3,r41 
(p0)  shladd r51 = r42,2,r51 
}
//
//    W_1_p1 = 1 + W_1
//
{ .mmi
(p0)  ldfs  f44 = [r50],0 ;; 
(p0)  ldfd  f45 = [r41],0 
//
//    M_2 = extr(N_fix,0,6)
//    M_1 = extr(N_fix,6,6)
//    r = X - float_N * L_hi
//
(p0)  extr r44 = r52, 12, 52  
}
{ .mmi
(p0)  ldfs  f46 = [r51],0 ;; 
(p0)  sub r46 = r58, r44  
(p0)  cmp.gt.unc  p8, p15 =  r44, r45 
}
//    
//    W = W_1 + W_1_p1*W_2 
//    Load  A_2 
//    Bias_m_K = Bias - K
//
{ .mii
(p0)  ldfe f40 = [r35],16 
//
//    load A_1
//    poly = A_2 + r*A_3 
//    rsq = r * r  
//    neg_2_mK = exponent of Bias_m_k
//
(p0)  add r47 = r58, r44 ;;  
//    
//    Set Safe = True  if k <= big_expo_pos  
//    Set Safe = False  if k >  big_expo_pos  
//    Load A_3
//    
(p15) cmp.lt p8,p15 = r44,r48 ;;
}
{ .mmf
(p0)  setf.exp f61 = r46 
//    
//    Bias_p + K = Bias + K
//    T = T_1 * T_2
//    
(p0)  setf.exp f36 = r47 
(p0)  fnma.s1 f42 = f41, f38, f42 ;; 
}
{ .mfi
	nop.m 999
//
//    Load W_1,W_2
//    Load big_exp_pos, load big_exp_neg
//
(p0)  fadd.s1 f47 = f43, f1 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)  fma.s1 f52 = f42, f51, f49 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)  fmpy.s1 f48 = f42, f42 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)  fmpy.s1 f53 = f44, f46 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)  fma.s1 f54 = f45, f47, f43 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)  fneg f61 =  f61 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)  fma.s1 f52 = f42, f52, f40 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)  fadd.s1 f55 = f54, f1 
	nop.i 999
}
{ .mfi
	nop.m 999
//
//    W + Wp1 * poly     
// 
(p0)  mov f34 = f53 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//    A_1 + r * poly 
//    Scale = setf_expl(Bias_p_k) 
//
(p0)  fma.s1 f52 = f48, f52, f42 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//    poly = r + rsq(A_1 + r*poly) 
//    Wp1 = 1 + W
//    neg_2_mK = -neg_2_mK
//
(p0)  fma.s1 f35 = f55, f52, f54
	nop.i 999 ;;
}
{ .mfb
	nop.m 999
(p0)  fmpy.s1 f35 = f35, f53 
//   
//    Y_hi = T
//    Y_lo = T * (W + Wp1*poly)
//
(p12) br.cond.sptk EXPL_MAIN ;; 
}
//
//    Branch if expl(x)  
//    Continue for expl(x-1)
//
{ .mii
(p0)  cmp.lt.unc  p12, p13 =  10, r44 
	nop.i 999 ;;
//
//    Set p12 if 10 < K, Else p13 
//
(p13) cmp.gt.unc  p13, p14 =  -10, r44 ;; 
}
//
//    K > 10:  Y_lo = Y_lo + neg_2_mK
//    K <=10:  Set p13 if -10 > K, Else set p14 
//
{ .mfi
(p13) cmp.eq  p15, p0 =  r0, r0 
(p14) fadd.s1 f34 = f61, f34 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p12) fadd.s1 f35 = f35, f61 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p13) fadd.s1 f35 = f35, f34 
	nop.i 999
}
{ .mfb
	nop.m 999
//
//    K <= 10 and K < -10, Set Safe = True
//    K <= 10 and K < 10,   Y_lo = Y_hi + Y_lo 
//    K <= 10 and K > =-10, Y_hi = Y_hi + neg_2_mk 
// 
(p13) mov f34 = f61 
(p0)  br.cond.sptk EXPL_MAIN ;; 
}
EXPL_SMALL: 
{ .mmi
      nop.m 999
(p0)  addl r34 = @ltoff(Constants_exp_64_Exponents#),gp  
(p12) addl r35 = @ltoff(Constants_exp_64_P#),gp ;; 
}
.pred.rel "mutex",p12,p13
{ .mmi
(p12) ld8  r35=[r35]      
nop.m 999
(p13) addl r35 = @ltoff(Constants_exp_64_Q#),gp 
};;
{ .mmi
(p13) ld8  r35=[r35]      
(p0) ld8  r34=[r34]      
nop.i 999
};;
{ .mfi
(p0)  add r34 = 0x48,r34  
// 
//    Return
//    K <= 10 and K < 10,   Y_hi = neg_2_mk 
// 
//    /*******************************************************/
//    /*********** Branch EXPL_SMALL  ************************/
//    /*******************************************************/
(p0)  mov f42 = f9 
	nop.i 999 ;;
}
//
//    Flag = 0
//    r4 = rsq * rsq
//
{ .mfi
(p0)  ld8 r49 =[r34],0
	nop.f 999
	nop.i 999 ;;
}
{ .mii
	nop.m 999
	nop.i 999 ;;
//
//    Flag = 1
//
(p0)  cmp.lt.unc  p14, p0 =  r37, r49 ;; 
}
{ .mfi
	nop.m 999
//
//    r = X
//
(p0)  fmpy.s1 f48 = f42, f42 
	nop.i 999 ;;
}
{ .mfb
	nop.m 999
//
//    rsq = r * r
//
(p0)  fmpy.s1 f50 = f48, f48 
//
//    Is input very small?
//
(p14) br.cond.spnt EXPL_VERY_SMALL ;; 
}
//
//    Flag_not1: Y_hi = 1.0
//    Flag is 1: r6 = rsq * r4
//
{ .mfi
(p12) ldfe f52 = [r35],16 
(p12) mov f34 = f1 
(p0)  add r53 = 0x1,r0 ;;  
}
{ .mfi
(p13) ldfe f51 = [r35],16 
//
//    Flag_not_1: Y_lo = poly_hi + r4 * poly_lo
//
(p13) mov f34 = f9 
	nop.i 999 ;;
}
{ .mmf
(p12) ldfe f53 = [r35],16 
//
//    For Flag_not_1, Y_hi = X
//    Scale = 1
//    Create 0x000...01
//
(p0)  setf.sig f37 = r53 
(p0)  mov f36 = f1 ;; 
}