libm_tan.s

Glibc 2.3.2源代码(解压后有100多M)

{ .mfi
	nop.m 999
//
//    N even: Result = Result * r + c
//    N odd:  poly1  = 1.0 + S_hi * r        32 bits partial
//
(p12) fma.s1 S_hi = S_hi, poly1, S_hi
	nop.i 999 ;;
}

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

{ .mfi
	nop.m 999
//
//    N even: Result1 = Result + r  (Rounding mode S0)
//    N odd:  poly1  =  S_hi * r + 1.0       64 bits partial
//
(p12) fma.s1 S_hi = S_hi, poly1, S_hi
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N odd:  poly1  =  S_hi * poly + S_hi    64 bits
//
(p12) fma.s1 poly1 = S_hi, r, f1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N odd:  poly1  =  S_hi * r + 1.0
//
(p12) fma.s1 poly1 = S_hi, c, poly1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N odd:  poly1  =  S_hi * c + poly1
//
(p12) fmpy.s1 S_lo = S_hi, poly1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N odd:  S_lo  =  S_hi *  poly1
//
(p12) fma.s1 S_lo = P, r, S_lo
	nop.i 999 ;;
}

{ .mfb
	nop.m 999
//
//    N odd:  S_lo  =  S_lo + r * P
//
(p12) fadd.s0 Result = S_hi, S_lo
(p0)   br.ret.sptk b0 ;;
}


TAN_SMALL_R: 

{ .mii
	nop.m 999
(p0)  extr.u i_1 = N_fix_gr, 0, 1 ;;
(p0)  cmp.eq.unc p11, p12 = 0x0000, i_1
}

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

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

{ .mfi
(p0)  addl           table_ptr1   = @ltoff(TAN_BASE_CONSTANTS), gp
        nop.f 999
        nop.i 999
}
;;

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

// *****************************************************************
// *****************************************************************
// *****************************************************************

{ .mmi
(p0)  add table_ptr1 = 224, table_ptr1 ;;
(p0)  ldfe P1_1 = [table_ptr1], 16
	nop.i 999 ;;
}
//    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
//    |r| < 2**(-2)

{ .mfi
(p0)  ldfe P1_2 = [table_ptr1], 16
(p11) fmpy.s1 r_to_the_8 = rsq, rsq
	nop.i 999 ;;
}
//
//    Set table_ptr1 to beginning of constant table.
//    Get [i_1] - lsb of N_fix_gr.
//

{ .mfi
(p0)  ldfe P1_3 = [table_ptr1], 96
//
//    N even: rsq = r * r
//    N odd:  S_hi = frcpa(r)
//
(p12) fmerge.ns S_hi = S_hi, S_hi
	nop.i 999 ;;
}
//
//    Is i_1  even or odd?
//    if i_1 == 0, set PR_11.
//    if i_1 != 0, set PR_12.
//

{ .mfi
(p11) ldfe P1_9 = [table_ptr1], -16
//
//    N even: Poly2 = P1_7 + Poly2 * rsq
//    N odd:  poly2 = Q1_5 + poly2 * rsq
//
(p11) fadd.s1 CORR = rsq, f1
	nop.i 999 ;;
}

{ .mmi
(p11) ldfe P1_8 = [table_ptr1], -16 ;;
//
//    N even: Poly1 = P1_2 + P1_3 * rsq
//    N odd:  poly1 =  1.0 +  S_hi * r     
//    16 bits partial  account for necessary (-1)
//
(p11) ldfe P1_7 = [table_ptr1], -16
	nop.i 999 ;;
}
//
//    N even: Poly1 = P1_1 + Poly1 * rsq
//    N odd:  S_hi  =  S_hi + S_hi * poly1)     16 bits account for necessary
//

{ .mfi
(p11) ldfe P1_6 = [table_ptr1], -16
//
//    N even: Poly2 = P1_5 + Poly2 * rsq
//    N odd:  poly2 = Q1_3 + poly2 * rsq
//
(p11) fmpy.s1 r_to_the_8 = r_to_the_8, r_to_the_8
	nop.i 999 ;;
}
//
//    N even: Poly1 =  Poly1 * rsq
//    N odd:  poly1  = 1.0 + S_hi * r         32 bits partial
//

{ .mfi
(p11) ldfe P1_5 = [table_ptr1], -16
(p12) fma.s1 poly1 =  S_hi, r, f1
	nop.i 999 ;;
}
//
//    N even: CORR =  CORR * c
//    N odd:  S_hi  =  S_hi * poly1 + S_hi    32 bits
//

//
//    N even: Poly2 = P1_6 + Poly2 * rsq
//    N odd:  poly2 = Q1_4 + poly2 * rsq
//
{ .mmf
(p0)  addl           table_ptr2   = @ltoff(TAN_BASE_CONSTANTS), gp
(p11) ldfe P1_4 = [table_ptr1], -16
(p11) fmpy.s1 CORR =  CORR, c
}
;;


{ .mmi
(p0)  ld8 table_ptr2 = [table_ptr2]
      nop.m 999
      nop.i 999
}
;;


{ .mii
(p0)  add table_ptr2 = 464, table_ptr2
	nop.i 999 ;;
	nop.i 999
}

{ .mfi
	nop.m 999
(p11) fma.s1 Poly1 = P1_3, rsq, P1_2
	nop.i 999 ;;
}

{ .mfi
(p0)  ldfe Q1_7 = [table_ptr2], -16
(p12) fma.s1 S_hi = S_hi, poly1, S_hi
	nop.i 999 ;;
}

{ .mfi
(p0)  ldfe Q1_6 = [table_ptr2], -16
(p11) fma.s1 Poly2 = P1_9, rsq, P1_8
	nop.i 999 ;;
}

{ .mmi
(p0)  ldfe Q1_5 = [table_ptr2], -16 ;;
(p12) ldfe Q1_4 = [table_ptr2], -16
	nop.i 999 ;;
}

{ .mfi
(p12) ldfe Q1_3 = [table_ptr2], -16
//
//    N even: Poly2 = P1_8 + P1_9 * rsq
//    N odd:  poly2 = Q1_6 + Q1_7 * rsq
//
(p11) fma.s1 Poly1 = Poly1, rsq, P1_1
	nop.i 999 ;;
}

{ .mfi
(p12) ldfe Q1_2 = [table_ptr2], -16
(p12) fma.s1 poly1 = S_hi, r, f1
	nop.i 999 ;;
}

{ .mfi
(p12) ldfe Q1_1 = [table_ptr2], -16
(p11) fma.s1 Poly2 = Poly2, rsq, P1_7
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N even: CORR =  rsq + 1
//    N even: r_to_the_8 =  rsq * rsq
//
(p11) fmpy.s1 Poly1 = Poly1, rsq
	nop.i 999 ;;
}

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

{ .mfi
	nop.m 999
(p12) fma.s1 poly2 = Q1_7, rsq, Q1_6
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fma.s1 Poly2 = Poly2, rsq, P1_6
	nop.i 999 ;;
}

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

{ .mfi
	nop.m 999
(p12) fma.s1 poly2 = poly2, rsq, Q1_5
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
(p11) fma.s1 Poly2= Poly2, rsq, P1_5
	nop.i 999 ;;
}

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

{ .mfi
	nop.m 999
(p12) fma.s1 poly2 = poly2, rsq, Q1_4
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N even: r_to_the_8 = r_to_the_8 * r_to_the_8
//    N odd:  poly1  =  S_hi * r + 1.0       64 bits partial
//
(p11) fma.s1 Poly2 = Poly2, rsq, P1_4
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N even: Result = CORR + Poly * r
//    N odd:  P = Q1_1 + poly2 * rsq
//
(p12) fma.s1 poly1 = S_hi, r, f1
	nop.i 999
}

{ .mfi
	nop.m 999
(p12) fma.s1 poly2 = poly2, rsq, Q1_3
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N even: Poly2 = P1_4 + Poly2 * rsq
//    N odd:  poly2 = Q1_2 + poly2 * rsq
//
(p11) fma.s1 Poly = Poly2, r_to_the_8, Poly1
	nop.i 999 ;;
}

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

{ .mfi
	nop.m 999
(p12) fma.s1 poly2 = poly2, rsq, Q1_2
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N even: Poly = Poly1 + Poly2 * r_to_the_8
//    N odd:  S_hi =  S_hi * poly1 + S_hi    64 bits
//
(p11) fma.s1 Result = Poly, r, CORR
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N even: Result =  r + Result  (User supplied rounding mode)
//    N odd:  poly1  =  S_hi * c + poly1
//
(p12) fmpy.s1 S_lo = S_hi, poly1
	nop.i 999
}

{ .mfi
	nop.m 999
(p12) fma.s1 P = poly2, rsq, Q1_1
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N odd:  poly1  =  S_hi * r + 1.0
//
(p11) fadd.s0 Result = Result, r
	nop.i 999 ;;
}

{ .mfi
	nop.m 999
//
//    N odd:  S_lo  =  S_hi *  poly1
//
(p12) fma.s1 S_lo = Q1_1, c, S_lo
	nop.i 999
}

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

{ .mfi
	nop.m 999
//
//    N odd:  S_lo  =  Q1_1 * c + S_lo
//
(p12) fma.s1 Result = P, r, S_lo
	nop.i 999 ;;
}

{ .mfb
	nop.m 999
//
//    N odd:  Result =  S_lo + r * P
//
(p12) fadd.s0 Result = Result, S_hi
(p0)   br.ret.sptk b0 ;;
}


TAN_NORMAL_R: 

{ .mfi
(p0)  getf.sig sig_r = r
// *******************************************************************
// *******************************************************************
// *******************************************************************
//
//    r and c have been computed.
//    Make sure ftz mode is set - should be automatic when using wre
//
//
//    Get [i_1] -  lsb of N_fix_gr alone.
//
(p0)  fmerge.s  Pos_r = f1, r
(p0)  extr.u i_1 = N_fix_gr, 0, 1 ;;
}

{ .mfi
	nop.m 999
(p0)  fmerge.s  sgn_r =  r, f1
(p0)  cmp.eq.unc p11, p12 = 0x0000, i_1 ;;
}

{ .mfi
	nop.m 999
	nop.f 999
(p0)  extr.u lookup = sig_r, 58, 5
}

{ .mlx
	nop.m 999
(p0)  movl Create_B = 0x8200000000000000 ;;
}

{ .mfi
(p0)  addl           table_ptr1   = @ltoff(TAN_BASE_CONSTANTS), gp
	nop.f 999
(p0)  dep Create_B = lookup, Create_B, 58, 5
}
;;

//
//    Get [i_1] -  lsb of N_fix_gr alone.
//    Pos_r = abs (r)
//


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


{ .mmi
	nop.m 999
(p0)  setf.sig B = Create_B
//
//    Set table_ptr1 and table_ptr2 to base address of
//    constant table.
//
(p0)  add table_ptr1 = 480, table_ptr1 ;;
}

{ .mmb
	nop.m 999
//
//    Is i_1 or i_0  == 0 ?
//    Create the constant  1 00000 1000000000000000000000...
//
(p0)  ldfe P2_1 = [table_ptr1], 16
	nop.b 999
}

{ .mmi
	nop.m 999 ;;
(p0)  getf.exp exp_r = Pos_r
	nop.i 999
}
//
//    Get r's exponent
//    Get r's significand
//

{ .mmi
(p0)  ldfe P2_2 = [table_ptr1], 16 ;;
//
//    Get the 5 bits or r for the lookup.   1.xxxxx ....
//    from sig_r.
//    Grab  lsb of exp of B
//
(p0)  ldfe P2_3 = [table_ptr1], 16
	nop.i 999 ;;
}

{ .mii
	nop.m 999
(p0)  andcm table_offset = 0x0001, exp_r ;;
(p0)  shl table_offset = table_offset, 9 ;;
}

{ .mii
	nop.m 999
//
//    Deposit   0 00000 1000000000000000000000... on
//              1 xxxxx yyyyyyyyyyyyyyyyyyyyyy...,
//    getting rid of the ys.
//    Is  B = 2** -2 or  B=