e_powl.s

Glibc 2.3.2源代码(解压后有100多M)

//     load G_2 
//     X_2 = X_1 * Z_2 
//     Add offset to Table 2 ptr.
//     float_N = significand of N
//
{ .mmi
(p0)   ldfs  FR_H_2 = [GR_Index2],8 ;; 
//
//     load H_2 
//     G = G * G_2
//
(p0)   ldfe FR_h_2 = [GR_Index2],0 
(p0)   pmpyshr2.u GR_X_2 = GR_X_1,GR_Z_2,15 ;; 
}
{ .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
        nop.i 999;;
}
{ .mii
	nop.m 999
	nop.i 999 ;;
(p0)   extr.u GR_Index3 = GR_X_2, 1, 5 ;;  
}
{ .mfi
(p0)   shladd GR_Table_Ptr1 = GR_Index3,2,GR_Table_Ptr1 
	nop.f 999
//
//     h = h_1 + h_2  
//     Adjust Index3 
//
(p0)   shladd GR_Index3 = GR_Index3,4,GR_Table_Ptr ;; 
}
{ .mmb
	nop.m 999
(p0)   ldfe FR_h_3 = [GR_Index3],12 
	nop.b 999 ;;
}
{ .mmf
(p0)   ldfs FR_H_3 = [GR_Table_Ptr1],0 
//
//     float_N = Make N a fp number
//     Load h_3
//     Get pointer to Q table.     
//
(p0)   ldfs  FR_G_3 = [GR_Index3],0 
(p0)   fmpy.s1 FR_G = FR_G_1, FR_G_2 
}
;;

{ .mmi
      nop.m 999
(p0)  addl           GR_Table_Ptr   = @ltoff(Constants_log_80_Q#), gp
      nop.i 999
}
;;

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



{ .mfi
(p0)   ldfe FR_log2_hi = [GR_Table_Ptr],16
(p0)   fadd.s1 FR_H = FR_H_1, FR_H_2 
	nop.i 999 ;;
}
{ .mmf
	nop.m 999
//
//     G = G_1 * G_2 * G_3 
//
(p0)   ldfe FR_log2_lo = [GR_Table_Ptr],16 
//
//     load h_2 
//     H = H_1 + H_2 
//     Get Index3
//
(p0)   fadd.s1 FR_h = FR_h_1, FR_h_2 ;; 
}
//
//     Load log2_lo part
//     r = G*S -1
//
{ .mfi
(p0)   ldfe FR_Q_6 = [GR_Table_Ptr],16 
//
//     Load H_3
//
(p0)   fcvt.xf FR_float_N = FR_float_N 
	nop.i 999 ;;
}
//
//     Load Q_6
//
{ .mmi
(p0)   ldfe FR_Q_5 = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_Q_4 = [GR_Table_Ptr],16 
	nop.i 999 ;;
}
{ .mmi
(p0)   ldfe FR_Q_3 = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_Q_2 = [GR_Table_Ptr],16 
	nop.i 999 ;;
}
{ .mmf
	nop.m 999
//
//     poly_lo = Q_5 + r * Q_6
//     Load Q_2
//     rsq = r * r 
//
(p0)   ldfe FR_Q_1 = [GR_Table_Ptr],16 
//
//     h = h_1 + h_2 + h_3   
//     H = H_1 + H_2 + H_3 
//     Load G_3.
//     Begin Loading Q's -  load log2_hi part
//
(p0)   fmpy.s1 FR_G = FR_G, FR_G_3 
}
{ .mfi
	nop.m 999
(p0)   fadd.s1 FR_H = FR_H, FR_H_3 
	nop.i 999 
}
;;

//
//     Y_lo = poly + Y_lo 
//

{ .mmi
      nop.m 999
(p0)  addl           GR_Table_Ptr   = @ltoff(Constants_exp_64_Arg#), gp
      nop.i 999
}
;;

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


{ .mfi
	nop.m 999
(p0)   fadd.s1 FR_h = FR_h, FR_h_3 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     Load Q_5
//
(p0)   fmpy.s1 FR_GS_hi = FR_G, FR_S 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fms.s1 FR_r = FR_G, FR_S, f1 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fma.s1 FR_poly_lo = FR_r, FR_Q_6, FR_Q_5 
	nop.i 999
}
{ .mfi
	nop.m 999
//
//     GS_hi = G*S
//     Load Q_4
//
(p0)   fsub.s1 FR_r_cor = FR_GS_hi, f1 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fms.s1 FR_GS_lo = FR_G, FR_S, FR_GS_hi 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fma.s1 FR_poly = FR_r, FR_Q_2, FR_Q_1 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     Load Q_3
//     r_cor = GS_hi -1
//     GS_lo  = G*S - GS_hi
//
(p0)   fmpy.s1 FR_rsq = FR_r, FR_r 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fma.s1 FR_G = FR_float_N, FR_log2_hi, FR_H 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     poly = poly_hi + rsq * poly_lo 
//     Tbl = float_N*log2_hi + H
//
(p0)   fma.s1 FR_Y_lo = FR_float_N, FR_log2_lo, FR_h 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     r_cor = r_cor - r
//     poly_hi = r * Q_2 + Q_1
//
(p0)   fma.s1 FR_poly_lo = FR_r, FR_poly_lo, FR_Q_4 
	nop.i 999
}
{ .mfi
	nop.m 999
//
//     Load Q_1
//
(p0)   fsub.s1 FR_r_cor = FR_r_cor, FR_r 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
// 
//     Y_lo = float_N*log2_lo + h
// 
(p0)   fadd.s1 FR_Y_hi = FR_G, FR_r 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     poly_lo = Q_4 + r * poly_lo;;
//     r_cor = r_cor + GS_lo;;
//
(p0)   fma.s1 FR_poly_lo = FR_r, FR_poly_lo, FR_Q_3 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fadd.s1 FR_r_cor = FR_r_cor, FR_GS_lo 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fadd.s1 FR_r_cor = FR_r_cor, FR_Y_lo 
	nop.i 999
}
{ .mfi
	nop.m 999
//
//     poly_lo = Q_3 + r * poly_lo;;
//
(p0)   fma.s1 FR_poly = FR_rsq, FR_poly_lo, FR_poly 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fsub.s1 FR_Y_lo = FR_G, FR_Y_hi 
	nop.i 999
}
{ .mmi
(p0)   ldfe FR_L_Inv = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_L_hi = [GR_Table_Ptr],16 
	nop.i 999 ;;
}
{ .mfi
(p0)   ldfe FR_L_lo = [GR_Table_Ptr],16 
	nop.f 999
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     Y_hi = Tbl + r 
//     r_cor = r_cor + Y_lo 
//
(p0)   fma.s1 FR_poly = FR_rsq, FR_poly, FR_r_cor 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//     Y_lo = Tbl - Y_hi 
//     poly = rsq * poly + r_cor
//
(p0)   fadd.s1 FR_Y_lo = FR_Y_lo, FR_r 
	nop.i 999 ;;
}
{ .mfb
	nop.m 999
//
//     Y_lo =  Y_lo + r  
//
(p0)   fadd.s1 FR_Y_lo = FR_Y_lo, FR_poly 
//
//     Load L_Inv
//     Load L_hi
//     Load L_lo
//     all long before they are needed.
//     They are used in LOGL_RETURN PATH
//
br.cond.sptk L(LOGL_RETURN) ;; 
}
L(LOGL80_NEAR): 
//
//     Branch LOGL80_NEAR
//

{ .mmi
      nop.m 999
(p0)  addl           GR_Table_Ptr   = @ltoff(Constants_log_80_P#), gp
      nop.i 999
}
;;

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

{ .mfi
	nop.m 999
(p0)   fmpy.s1 FR_Wsq = FR_W, FR_W 
(p0)   add GR_Table_Ptr1 = 0x50,GR_Table_Ptr  
}
//
//     Adjust ptr to 1/2 
//     Adjust Ptr1 to P_4
//
{ .mmi
(p0)   ldfe FR_Half = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_P_4 = [GR_Table_Ptr1],16 
	nop.i 999
}
//
//     Load 1/2 
//
{ .mmi
(p0)   ldfe FR_P_8 = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_P_3 = [GR_Table_Ptr1],16 
	nop.i 999
}
{ .mmi
(p0)   ldfe FR_P_7 = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_P_2 = [GR_Table_Ptr1],16 
	nop.i 999
}
//
//     Load P_7
//     half_W = .5 * W
//     Load P_3
//
{ .mmi
(p0)   ldfe FR_P_6 = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_P_1 = [GR_Table_Ptr1],16 
	nop.i 999 ;;
}
//
//     Load P_6
//     Wsq = w * w
//     poly = w*P_4 + P_3 
//     Load P_2
//
{ .mfi
(p0)   ldfe FR_P_5 = [GR_Table_Ptr],16 
//
//     Load P_5
//     poly_lo =  w * P_8 + P_7 
//     Y_hi = w - (1/2)w*w
//     Load P_1
//
(p0)   fmpy.s1 FR_W4 = FR_Wsq, FR_Wsq 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fmpy.s1 FR_W3 = FR_Wsq, FR_W 
	nop.i 999 
}
;;

//
//     Y_lo = W3 * poly + Y_lo
//

{ .mmi
      nop.m 999
(p0)  addl           GR_Table_Ptr   = @ltoff(Constants_exp_64_Arg#), gp
      nop.i 999
}
;;

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


{ .mmi
(p0)   ldfe FR_L_Inv = [GR_Table_Ptr],16 ;; 
(p0)   ldfe FR_L_hi = [GR_Table_Ptr],16 
	nop.i 999 ;;
}
{ .mfi
(p0)   ldfe FR_L_lo = [GR_Table_Ptr],16 
//
//     Load P_8
//     Load P_4
//
(p0)   fmpy.s1 FR_half_W = FR_Half, FR_W 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fma.s1 FR_poly_lo = FR_W, FR_P_8,FR_P_7 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fma.s1 FR_poly = FR_W, FR_P_4, FR_P_3 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fnma.s1 FR_Y_hi = FR_W, FR_half_W, FR_W 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     W4 = Wsq * Wsq
//     poly = w *poly + P_2
//
(p0)   fma.s1 FR_poly_lo = FR_W, FR_poly_lo, FR_P_6 
	nop.i 999
}
{ .mfi
	nop.m 999
(p0)   fma.s1 FR_poly = FR_W, FR_poly, FR_P_2 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fsub.s1 FR_Y_lo = FR_W, FR_Y_hi 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     poly = w * poly + P_1
//     w3 = wsq * w
//
(p0)   fma.s1 FR_poly_lo = FR_W, FR_poly_lo, FR_P_5 
	nop.i 999
}
{ .mfi
	nop.m 999
//
//     poly_lo = w * poly_lo + P_6
//     Y_lo = W - Y_hi
//
(p0)   fma.s1 FR_poly = FR_W, FR_poly, FR_P_1 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p0)   fnma.s1 FR_Y_lo = FR_W, FR_half_W, FR_Y_lo 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     poly_lo = w * poly_lo + 
//     Y_lo = Y_lo - w * (1/2)w
//
(p0)   fma.s1 FR_poly = FR_poly_lo, FR_W4, FR_poly 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     Y_lo = (W-Y_hi) - w * (1/2)w
//     poly =  W4* poly_lo + poly 
//
(p0)   fma.s1 FR_Y_lo = FR_poly, FR_W3, FR_Y_lo 
	nop.i 999 ;;
}
L(LOGL_RETURN): 
{ .mfi
(p0)   add GR_Expo_Range = 0x2,r0  
//
//     Load L_Inv
//     Load L_hi
//     Load L_lo
//     all long before they are needed.
//
//
//     kernel_log_80 computed ln(X)
//     and return logX_hi and logX_lo as results.
//     PR_pow_Safe set as well. 
//
(p0)   fmpy.s1 FR_X_lo = FR_Input_Y, FR_logx_lo 
//
//     Compute Y * (logX_hi + logX_lo) 
//     P_hi -> X 
//     P_lo -> X_cor 
//     (Manipulate names so that inputs are in
//     the place kernel_exp expects them)
//     Set GR_Flag to 2 
//     Set GR_Expo_Range to Double
// 
//     This function computes exp( 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)   cmp.eq.unc  p15, p0 =  r0, r0 
}
;;

{ .mmi
(p0)  addl           GR_W1_ptr   = @ltoff(Constants_exp_64_W1#), gp
(p0)  addl           GR_W2_ptr   = @ltoff(Constants_exp_64_W2#), gp
(p0)  add GR_Flag = 0x2,r0  
}
;;

{ .mmi
      ld8 GR_W1_ptr = [GR_W1_ptr]
      ld8 GR_W2_ptr = [GR_W2_ptr]
(p0)   cmp.ne.unc  p7, p0 =  0x1, GR_Flag 
}
;;

{ .mlx
	nop.m 999
(p0)   movl GR_Mask = 0x1FFFF ;; 
}


{ .mlx
	nop.m 999
(p0)   movl GR_BIAS = 0x0FFFF ;; 
}
{ .mfi
	nop.m 999
//
//     X_lo =  Y * logX_lo
//
(p0)   fma.s1 FR_P_hi = FR_Input_Y, FR_logx_hi,FR_X_lo 
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//     Set Safe=True 
//     Flag is always 2 for this routine
//
(p0)   fmpy.s1 FR_float_N = FR_X, FR_L_Inv 
	nop.i 999
}
{ .mfi
	nop.m 999
//
//     X_hi  = Y * logX_hi + X_lo
//     Set GR_Flag = 2 for exp(x + xcor)
//
(p0)   fms.s1 FR_P_lo= FR_Input_Y, FR_logx_hi, FR_P_hi 
	nop.i 999 ;;
}
{ .mmi
	nop.m 999 ;;
(p0)   getf.exp GR_Expo_X = FR_X