e_powl.s

Glibc 2.3.2源代码(解压后有100多M)

GR_X_1              = r40
GR_W1_ptr           = r40

GR_W2_ptr           = r41
GR_X_2              = r41

GR_Z_1              = r42
GR_M2               = r42

GR_M1               = r43
GR_Z_2              = r43

GR_N                = r44
GR_k                = r44

GR_Big_Pos_Exp      = r45


GR_BIAS_p_k         = r47
GR_BIASed_exp_y     = r47

GR_Big_Neg_Exp      = r48
GR_Index3           = r48
GR_temp             = r48

GR_vsm_expo         = r49
GR_y_sign           = r49

GR_T1_ptr           = r50
GR_T2_ptr           = r51
GR_N_fix            = r52
GR_exp_y            = r53
GR_signif_y         = r54
GR_exp_and_sign_y   = r55
GR_low_order_bit    = r56
GR_get_exp_mask     = r57
GR_exponent_zero    = r58

// ** Registers for unwind support

GR_SAVE_PFS         = r59
GR_SAVE_B0          = r60
GR_SAVE_GP          = r61
GR_Parameter_X      = r62 
GR_Parameter_Y      = r63 
GR_Parameter_RESULT = r64 
GR_Parameter_TAG    = r65 

FR_X      = f8
FR_Y      = f9
FR_RESULT = f99

// **

FR_Input_X          = f8
FR_Output           = f8
FR_Input_Y          = f9

FR_Neg              = f10
FR_P_hi             = f10
FR_X                = f10

FR_Half             = f11
FR_h_3              = f11
FR_poly_hi          = f11

FR_Sgn              = f12

FR_Neg_X            = f13
FR_half_W           = f13

FR_X_cor            = f14
FR_P_lo             = f14

FR_W                = f15

FR_X_lo             = f32

FR_S                = f33
FR_W3               = f33

FR_Y_hi             = f34
FR_logx_hi          = f34

FR_Z                = f35
FR_logx_lo          = f35
FR_GS_hi            = f35
FR_Y_lo             = f35

FR_r_cor            = f36
FR_Scale            = f36

FR_G_1              = f37
FR_G                = f37
FR_Wsq              = f37
FR_L_Inv            = f37
FR_temp             = f37

FR_H_1              = f38
FR_H                = f38
FR_W4               = f38
FR_float_N          = f38

FR_h                = f39
FR_h_1              = f39
FR_N                = f39
FR_P_7              = f39

FR_G_2              = f40
FR_P_8              = f40
FR_L_hi             = f40

FR_H_2              = f41
FR_L_lo             = f41
FR_A_1              = f41

FR_h_2              = f42
FR_P_6              = f42

FR_abs_W            = f43
FR_W1               = f43

FR_G_3              = f44
FR_P_8              = f44
FR_T1               = f44

FR_log2_hi          = f45
FR_W2               = f45

FR_GS_lo            = f46
FR_T2               = f46

FR_W_1_p1           = f47
FR_H_3              = f47

FR_float_N          = f48

FR_P_4              = f49
FR_A_2              = f49

FR_Q_4              = f50
FR_r4               = f50

FR_Q_3              = f51
FR_A_3              = f51

FR_Q_2              = f52
FR_P_2              = f52

FR_Q_1              = f53
FR_P_1              = f53
FR_T                = f53

FR_Wp1              = f54
FR_Q_5              = f54
FR_P_3              = f54

FR_Q_6              = f55

FR_log2_lo          = f56
FR_Two              = f56

FR_Big              = f57

FR_neg_2_mK         = f58
FR_NBig             = f58

FR_r                = f59

FR_poly_lo          = f60

FR_poly             = f61

FR_P_5              = f62

FR_rsq              = f63

FR_Result           = f99
FR_Result_small     = f100
FR_Result_big       = f101

.section .text
.proc powl#
.global powl#
.align 64 

powl: 
{ .mfi
alloc  GR_Expo_Range = ar.pfs,0,30,4,0
(p0)   fclass.m.unc p7, p13 =  FR_Input_Y, 0x1E7 
nop.i 0
}
{ .mfi
(p0)   getf.exp GR_exp_and_sign_y = FR_Input_Y 
//
//     Save State
//
(p0)   fclass.m.unc p6, p12 =  FR_Input_X, 0x1E7 
nop.i 0
};;
{ .mfi
(p0)   getf.sig GR_signif_y = FR_Input_Y 
(p0)   fcmp.eq.unc.s1 p12, p13 =  FR_Input_X, f1 
nop.i 0
}
{ .mfi
	nop.m 999
//
//     Check for y = 1 
//     Identify EM unsupporteds.
//     Load FR_half = .5
//
(p0)   fadd.s1 FR_Two = f1, f1 
//
//     Load 1/2 in GP register
//
nop.i 0 
}
;;

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

{ .mmi
      ld8 GR_Table_Ptr = [GR_Table_Ptr]
      nop.m 999
      nop.i 999
}
;;

{ .mlx
(p0)   ldfe FR_Half =[GR_Table_Ptr],0
(p0)   movl GR_get_exp_mask = 0x1FFFF ;; 
}

{ .mfi
	nop.m 999
(p0)   fclass.nm.unc p9, p15 =  FR_Input_Y, 0x1FF 
//
//     Create FR_Two = 2
//     Get exp and significand of Y
//     Crate Masks
//     sgn = 1
//
(p0)   and GR_exp_y = GR_get_exp_mask,GR_exp_and_sign_y
}
{ .mlx
	nop.m 999
(p0)   movl GR_exponent_zero = 0xFFFF ;; 
}
{ .mfi
	nop.m 999
(p0)   mov FR_Sgn = f1 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fcmp.eq.unc.s1 p10, p11 =  FR_Input_Y, f1 
	nop.i 999 ;;
}
{ .mfb
	nop.m 999
//
//     Identify NatVals, NaNs, Infs, and Zeros.
//     Load Half
//
(p0)   fclass.nm.unc p8, p14 =  FR_Input_X, 0x1FF 
//      
//     Remove sign bit from exponent of y.
//     Check for x = 1 
//
(p6)   br.cond.spnt L(POWL_64_SPECIAL) ;; 
}
{ .mib
	nop.m 999
	nop.i 999
(p7)   br.cond.spnt L(POWL_64_SPECIAL) ;; 
}
{ .mib
	nop.m 999
	nop.i 999
(p8)   br.cond.spnt L(POWL_64_UNSUPPORT) ;; 
}
{ .mib
	nop.m 999
	nop.i 999
(p9)   br.cond.spnt L(POWL_64_UNSUPPORT) ;; 
}
{ .mfi
(p0)   cmp.lt.unc  p9, p0 = GR_exp_y,GR_exponent_zero 
(p0)   fcmp.lt.unc.s1 p6, p13  =  FR_Input_X, f0 
//
//     Branch on Infs, Nans, Zeros, and Natvals
//     Check to see that exponent < 0
//
(p0)   sub GR_exp_y = GR_exp_y,GR_exponent_zero
}
//     x not zero, is y ==2? 
{ .mfi
	nop.m 999
(p11)  fcmp.eq.unc.s1 p7, p14 =  FR_Input_Y, FR_Two 
	nop.i 999 ;;
}
{ .mfb
	nop.m 999
(p9)   fcmp.lt.unc.s1 p9, p0   =  FR_Input_X, f0 
(p7)   br.cond.spnt L(POWL_64_SQUARE) ;;   // Branch if x not zero and y=2
}
{ .mfi
	nop.m 999
(p6)   fmerge.ns FR_Neg_X = FR_Input_X, FR_Input_X 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)  fmpy.s0 FR_Result = FR_Input_X, f1 
//
//     For y = 1, compute result = x 
//     For x = 1, compute 1 
//     When Y is one return X and possible raise 
//     denormal operand exception.
//     Remove exponent BIAS
//
(p6)   shl GR_exp_and_sign_y=  GR_signif_y,GR_exp_y ;; 
}
{ .mfi
(p9)   or  GR_exp_and_sign_y = 0xF,GR_signif_y 
(p12)  fma.s0 FR_Result = FR_Input_Y, f0, f1 
	nop.i 999 ;;
}
{ .mii
	nop.m 999
(p6)   extr.u GR_exp_y = GR_exp_and_sign_y,63,1 ;; 
(p6)   cmp.ne.unc  p9, p0 =  GR_exp_y, r0 
}
{ .mii
	nop.m 999
//
//     Both predicates can be set. 
//     Don't consider  y's < 1.
//
(p6)   shl GR_signif_y=  GR_exp_and_sign_y,1 ;; 
//
//     Is shift off integer part of y.
//     Get y's  even or odd bit.
//
(p6)   cmp.ne.unc  p8, p0  =  GR_signif_y, r0 
}
{ .mib
	nop.m 999
	nop.i 999
//
//     Is the fractional part of the y = 0?
//     Is the integer even or odd. 
//
(p10)  br.cond.spnt L(POWL_64_RETURN) ;; 
}
{ .mib
	nop.m 999
	nop.i 999
(p12)  br.cond.spnt L(POWL_64_RETURN) ;; 
}
{ .mib
	nop.m 999
	nop.i 999
(p8)   br.cond.spnt L(POWL_64_XNEG) ;;
}
{ .mfi
	nop.m 999
(p9)   fmerge.ns FR_Sgn = FR_Sgn, FR_Sgn 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fcmp.eq.unc.s0 p11, p0 =  FR_Input_Y, FR_Half 
	nop.i 999 ;;
}
//
//     Raise possible denormal operand exception for both
//     X and Y.
//
{ .mfb
	nop.m 999
//
//     Branch for (x < 0) and Y not an integer.
//
(p0)   fcmp.eq.unc.s0 p12, p0 =  FR_Input_X, f1 
//
//     For x < 0 and y integer, make x positive 
//     For x < 0 and y odd integer,, set sign = -1.
//
(p11)  br.cond.spnt L(POWL_64_SQRT) ;; 
}
{ .mmf
(p0)   cmp.eq.unc  p15, p14 =  r0, r0 
	nop.m 999
(p13)  fnorm.s1 FR_Z = FR_Input_X ;; 
}
{ .mfi
	nop.m 999
(p6)   fnorm.s1 FR_Z = FR_Neg_X 
	nop.i 999
}
;;

//
//     Branch to embedded sqrt(x)
//
//
//     Computes ln( x ) to extra precision 
//     Input  FR 1: FR_X            
//     Output FR 2: FR_Y_hi  
//     Output FR 3: FR_Y_lo  
//     Output PR 1: PR_Safe  
//

{ .mmi
      nop.m 999
(p0)  addl           GR_Table_Ptr   = @ltoff(Constants_log_80_Z_G_H_h1#), gp
      nop.i 999
}
;;

{ .mmi
      ld8 GR_Table_Ptr = [GR_Table_Ptr]
      nop.m 999
      nop.i 999
}
;;


{ .mlx
	nop.m 999
(p0)   movl GR_BIAS = 0x000000000000FFFF ;; 
}
{ .mfi
	nop.m 999
(p0)   fsub.s1 FR_W = FR_Z, f1 
	nop.i 999 ;;
}
//     
//     Z = Norm(X) - both + and - case 
//     Set Safe = True
//     
{ .mmb
(p0)   getf.sig GR_signif_Z = FR_Z 
(p0)   getf.exp GR_N =  FR_Z 
	nop.b 999 ;;
}
{ .mii
	nop.m 999
//     
//     Get significand of Z 
//     W = Z - 1
//     
(p0)   extr.u GR_Index1 = GR_signif_Z, 59, 4 ;;  
//     
//     Index1 = High order 4 bits of Z
//     X_0  = High order 15 bit of Z 
//
(p0)   shl GR_Index1 = GR_Index1,5 ;; 
}
{ .mfi
	nop.m 999
//     
//     Add offset to Index1 ptr.
//     
(p0)   fabs FR_abs_W =  FR_W 
//     
//     BIAS = 0x000...FFFF
//     Adjust Index1 ptr ( x 32) .
//     
(p0)   add GR_Index1 = GR_Index1,GR_Table_Ptr  
}
{ .mmi
	nop.m 999 ;;
(p0)   ld2 GR_Z_1 =[GR_Index1],4
(p0)   extr.u GR_X_0 = GR_signif_Z, 49, 15  
}
;;

{ .mmi
      nop.m 999
(p0)  addl           GR_Table_Ptr   = @ltoff(Constants_log_80_Z_G_H_h2#), gp
      nop.i 999
}
;;

{ .mmi
      ld8 GR_Table_Ptr = [GR_Table_Ptr]
      nop.m 999
      nop.i 999
}
;;


{ .mmi
(p0)   ldfs  FR_G_1 = [GR_Index1],4 ;; 
(p0)   ldfs  FR_H_1 = [GR_Index1],8 
	nop.i 999 ;;
}
//
//     Adjust Index2 (x 32). 
//
{ .mfi
(p0)   ldfe FR_h_1 = [GR_Index1],0 
	nop.f 999
(p0)   pmpyshr2.u GR_X_1 = GR_X_0,GR_Z_1,15 ;; 
}
{ .mmi
	nop.m 999 ;;
//
//     load Z_1 from Index1
//     abs_W = |W|
//     Point to Table2 
//
(p0)   getf.exp GR_M = FR_abs_W 
//
//     M = M - BIAS 
//     Load G_1 
//     N = exponent of Z 
//
       nop.i 999;;
}
{ .mmi
       nop.m 999
       nop.m 999
       nop.i 999;;
}
{ .mmi
       nop.m 999
       nop.m 999
       nop.i 999;;
}
{ .mmi
        nop.m 999
        nop.m 999
(p0)   extr.u GR_Index2 = GR_X_1, 6, 4 ;;  
}
{ .mii
	nop.m 999
//     
//     Extract Index2 
//     Load H_1
//     Is -8 > M ?
//
(p0)   shl GR_Index2=GR_Index2,5 ;; 
(p0)   add GR_Index2 = GR_Index2, GR_Table_Ptr  
}
//
//     M = exponent of abs_W
//     X_1 = X_0 * Z_1 
//     
{ .mii
(p0)   sub GR_M = GR_M, GR_BIAS  
	nop.i 999 ;;
(p0)   cmp.gt.unc  p7, p14 =  -8, GR_M 
}
{ .mib
	nop.m 999
	nop.i 999
(p7)   br.cond.spnt L(LOGL80_NEAR) ;; 
}
//
//     Load h_1
//     Possible branch out.  
//     Add offset of table to Index2 
//
{ .mfi
(p0)   ld2 GR_Z_2 =[GR_Index2],4
(p0)   fmerge.se FR_S =  f1,FR_Z
(p0)   sub GR_N = GR_N, GR_BIAS  
}
;;

{ .mmi
      nop.m 999
(p0)  addl           GR_Table_Ptr   = @ltoff(Constants_log_80_h3_G_H#), gp
      nop.i 999
}
;;

{ .mmi
      ld8 GR_Table_Ptr = [GR_Table_Ptr]
      nop.m 999
      nop.i 999
}
;;

//     
//     load Z_2 
//     N - BIAS 
//     Point to Table 3.
//     S = merging of Z and 1.0
//     
{ .mmi
(p0)   ldfs  FR_G_2 = [GR_Index2],4 
(p0)   setf.sig FR_float_N = GR_N 
(p0)   add GR_Table_Ptr1 = 0x200,GR_Table_Ptr ;;  
}
//