libm_tan.s

Glibc 2.3.2源代码(解压后有100多M)

;;

//
// ARGUMENT REDUCTION CODE - CASE 3 and 4
//
//
//    Adjust table_ptr1 to beginning of table.
//    N_0 = Arg * Inv_P_0
//


{ .mmi
(p0)  ld8 table_ptr1 = [table_ptr1]
      nop.m 999
      nop.i 999
}
;;


{ .mmi
(p0)  add table_ptr1 = 8, table_ptr1 ;;
//
//    Point to  2*-14
//
(p0)  ldfs TWO_TO_NEG14 = [table_ptr1], 4
	nop.i 999 ;;
}
//
//    Load 2**(-14).
//

{ .mmi
(p0)  ldfs NEGTWO_TO_NEG14 = [table_ptr1], 180 ;;
//
//    N_0_fix  = integer part of N_0 .
//    Adjust table_ptr1 to beginning of table.
//
(p0)  ldfs TWO_TO_NEG2 = [table_ptr1], 4
	nop.i 999 ;;
}
//
//    Make N_0 the integer part.
//

{ .mfi
(p0)  ldfs NEGTWO_TO_NEG2 = [table_ptr1]
//
//    Load -2**(-14).
//
(p0)  fcvt.fx.s1 N_0_fix = N_0
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p0)  fcvt.xf N_0 = N_0_fix
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p0)  fnma.s1 ArgPrime = N_0, P_0, Arg
	nop.i 999
}

{ .mfi
	nop.m 999
(p0)  fmpy.s1 w = N_0, d_1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    ArgPrime = -N_0 * P_0 + Arg
//    w  = N_0 * d_1
//
(p0)  fmpy.s1 N = ArgPrime, two_by_PI
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N = ArgPrime * 2/pi
//
(p0)  fcvt.fx.s1 N_fix = N
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N_fix is the integer part.
//
(p0)  fcvt.xf N = N_fix
	nop.i 999 ;;
}

{ .mfi
(p0)  getf.sig N_fix_gr = N_fix
	nop.f 999
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N is the integer part of the reduced-reduced argument.
//    Put the integer in a GP register.
//
(p0)  fnma.s1 s_val = N, P_1, ArgPrime
	nop.i 999
}

{ .mfi
	nop.m 999
(p0)  fnma.s1 w = N, P_2, w
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    s_val = -N*P_1 + ArgPrime
//    w = -N*P_2 + w
//
(p0)  fcmp.lt.unc.s1 p11, p10 = s_val, TWO_TO_NEG14
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fcmp.gt.s1 p11, p10 = s_val, NEGTWO_TO_NEG14
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 3: r = s_val + w (Z complete)
//    Case 4: U_hi = N_0 * d_1
//
(p10) fmpy.s1 V_hi = N, P_2
	nop.i 999
}

{ .mfi
	nop.m 999
(p11) fmpy.s1 U_hi = N_0, d_1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 3: r = s_val + w (Z complete)
//    Case 4: U_hi = N_0 * d_1
//
(p11) fmpy.s1 V_hi = N, P_2
	nop.i 999
}

{ .mfi
	nop.m 999
(p11) fmpy.s1 U_hi = N_0, d_1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Decide between case 3 and 4:
//    Case 3:  s <= -2**(-14) or s >= 2**(-14)
//    Case 4: -2**(-14) < s < 2**(-14)
//
(p10) fadd.s1 r = s_val, w
	nop.i 999
}

{ .mfi
	nop.m 999
(p11) fmpy.s1 w = N, P_3
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 4: We need abs of both U_hi and V_hi - dont
//    worry about switched sign of V_hi .
//
(p11) fsub.s1 A = U_hi, V_hi
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    Case 4: A =  U_hi + V_hi
//    Note: Worry about switched sign of V_hi, so subtract instead of add.
//
(p11) fnma.s1 V_lo = N, P_2, V_hi
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fms.s1 U_lo = N_0, d_1, U_hi
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fabs V_hiabs = V_hi
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    Case 4: V_hi = N * P_2
//            w = N * P_3
//    Note the product does not include the (-) as in the writeup
//    so (-) missing for V_hi and w .
(p10) fadd.s1 r = s_val, w
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 3: c = s_val - r
//    Case 4: U_lo = N_0 * d_1 - U_hi
//
(p11) fabs U_hiabs = U_hi
	nop.i 999
}

{ .mfi
	nop.m 999
(p11) fmpy.s1 w = N, P_3
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 4: Set P_12 if U_hiabs >= V_hiabs
//
(p11) fadd.s1 C_hi = s_val, A
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 4: C_hi = s_val + A
//
(p11) fadd.s1 t = U_lo, V_lo
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 3: Is |r| < 2**(-2), if so set PR_7
//    else set PR_8.
//    Case 3: If PR_7 is set, prepare to branch to Small_R.
//    Case 3: If PR_8 is set, prepare to branch to Normal_R.
//
(p10) fsub.s1 c = s_val, r
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 3: c = (s - r) + w (c complete)
//
(p11) fcmp.ge.unc.s1 p12, p13 = U_hiabs, V_hiabs
	nop.i 999
}

{ .mfi
	nop.m 999
(p11) fms.s1 w = N_0, d_2, w
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 4: V_hi = N * P_2
//            w = N * P_3
//    Note the product does not include the (-) as in the writeup
//    so (-) missing for V_hi and w .
//
(p10) fcmp.lt.unc.s1 p14, p15 = r, TWO_TO_NEG2
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p14) fcmp.gt.s1 p14, p15 = r, NEGTWO_TO_NEG2
	nop.i 999 ;;
}

{ .mfb
	nop.m 999
//
//    Case 4: V_lo = -N * P_2 - V_hi (U_hi is in place of V_hi in writeup)
//    Note: the (-) is still missing for V_hi .
//    Case 4: w = w + N_0 * d_2
//    Note: the (-) is now incorporated in w .
//
(p10) fadd.s1 c = c, w
//
//    Case 4: t = U_lo + V_lo
//    Note: remember V_lo should be (-), subtract instead of add. NO
//
(p14) br.cond.spnt TAN_SMALL_R ;;
}

{ .mib
	nop.m 999
	nop.i 999
(p15) br.cond.spnt TAN_NORMAL_R ;;
}

{ .mfi
	nop.m 999
//
//    Case 3: Vector off when |r| < 2**(-2).  Recall that PR_3 will be true.
//    The remaining stuff is for Case 4.
//
(p12) fsub.s1 a = U_hi, A
(p11) extr.u i_1 = N_fix_gr, 0, 1 ;;
}

{ .mfi
	nop.m 999
//
//    Case 4: C_lo = s_val - C_hi
//
(p11) fadd.s1 t = t, w
	nop.i 999
}

{ .mfi
	nop.m 999
(p13) fadd.s1 a = V_hi, A
	nop.i 999 ;;
}

//
//    Case 4: a = U_hi - A
//            a = V_hi - A (do an add to account for missing (-) on V_hi
//

{ .mfi
(p11)  addl           table_ptr1   = @ltoff(TAN_BASE_CONSTANTS), gp
(p11) fsub.s1 C_lo = s_val, C_hi
	nop.i 999
}
;;

{ .mmi
(p11) ld8 table_ptr1 = [table_ptr1]
      nop.m 999
      nop.i 999
}
;;

//
//    Case 4: a = (U_hi - A)  + V_hi
//            a = (V_hi - A)  + U_hi
//    In each case account for negative missing form V_hi .
//
//
//    Case 4: C_lo = (s_val - C_hi) + A
//

{ .mmi
(p11) add table_ptr1 = 224, table_ptr1 ;;
(p11) ldfe P1_1 = [table_ptr1], 16
	nop.i 999 ;;
}

{ .mfi
(p11) ldfe P1_2 = [table_ptr1], 128
//
//    Case 4: w = U_lo + V_lo  + w
//
(p12) fsub.s1 a = a, V_hi
	nop.i 999 ;;
}
//
//    Case 4: r = C_hi + C_lo
//

{ .mfi
(p11) ldfe Q1_1 = [table_ptr1], 16
(p11) fadd.s1 C_lo = C_lo, A
	nop.i 999 ;;
}
//
//    Case 4: c = C_hi - r
//    Get [i_1] - lsb of N_fix_gr.
//

{ .mfi
(p11) ldfe Q1_2 = [table_ptr1], 16
	nop.f 999
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p13) fsub.s1 a = U_hi, a
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fadd.s1 t = t, a
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 4: t = t + a
//
(p11) fadd.s1 C_lo = C_lo, t
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    Case 4: C_lo = C_lo + t
//
(p11) fadd.s1 r = C_hi, C_lo
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fsub.s1 c = C_hi, r
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    Case 4: c = c + C_lo  finished.
//    Is i_1  even or odd?
//    if i_1 == 0, set PR_4, else set PR_5.
//
// r and c have been computed.
// We known whether this is the sine or cosine routine.
// Make sure ftz mode is set - should be automatic when using wre
(p0)  fmpy.s1 rsq = r, r
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fadd.s1 c = c , C_lo
(p11) cmp.eq.unc p11, p12 =  0x0000, i_1 ;;
}

{ .mfi
	nop.m 999
(p12) frcpa.s1 S_hi, p0 = f1, r
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    N odd: Change sign of S_hi
//
(p11) fma.s1 Result = rsq, P1_2, P1_1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p12) fma.s1 P = rsq, Q1_2, Q1_1
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    N odd:  Result  =  S_hi + S_lo      (User supplied rounding mode for C1)
//
(p0)   fmpy.s0 fp_tmp = fp_tmp, fp_tmp  // Dummy mult to set inexact
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N even: rsq = r * r
//    N odd:  S_hi = frcpa(r)
//
(p12) fmerge.ns S_hi = S_hi, S_hi
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    N even: rsq = rsq * P1_2 + P1_1
//    N odd:  poly1 =  1.0 +  S_hi * r    16 bits partial  account for necessary
//
(p11) fmpy.s1 Result = rsq, Result
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p12) fma.s1 poly1 = S_hi, r,f1
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    N even: Result =  Result * rsq
//    N odd:  S_hi  = S_hi + S_hi*poly1  16 bits account for necessary
//
(p11) fma.s1 Result = r, Result, c
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p12) fma.s1 S_hi = S_hi, poly1, S_hi
	nop.i 999
}

{ .mfi
	nop.m 999
//
//    N odd:   S_hi  = S_hi * poly1 + S_hi   32 bits
//
(p11) fadd.s0 Result= r, Result
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p12) fma.s1 poly1 =  S_hi, r, f1
	nop.i 999 ;;
}