s_cosl.s

Glibc 2.3.2源代码(解压后有100多M)

(p9)  	fma.s1 FR_poly_lo = FR_rsq, FR_poly_lo, FR_S_3
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10) 	fma.s1 FR_poly_hi = FR_rsq, FR_C_2, FR_C_1
	nop.i 999
}
{ .mfi
	nop.m 999
(p10)	fma.s0 FR_C_1 = FR_C_1, FR_C_1, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p9)  	fma.s1 FR_poly_hi = FR_poly_hi, FR_rsq, f0
	nop.i 999
}
{ .mfi
	nop.m 999
//
//      poly_lo = FR_rsq * poly_lo + S_3
//      poly_hi = FR_rsq * poly_hi
//
(p10) 	fma.s1 FR_poly_lo = FR_rsq, FR_poly_lo, FR_C_3
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10) 	fma.s1 FR_poly_hi = FR_poly_hi, FR_rsq, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
// 	if (i_1 == 0): dummy fmpy's to flag inexact
// 	r = 1
//
(p9)	fma.s1 FR_poly_hi = FR_r, FR_poly_hi, f0
	nop.i 999
}
{ .mfi
	nop.m 999
//
// 	poly_hi = r * poly_hi 
//
(p0)    fma.s1	FR_poly = FR_Z, FR_poly_lo, FR_c
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p12)	fms.s1	FR_r = f0, f1, FR_r
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      poly_hi = Z * poly_lo + c	
// 	if i_0 == 1: r = -r     
//
(p0) 	fma.s1	FR_poly = FR_poly, f1, FR_poly_hi
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p12)	fms.s0 FR_Input_X = FR_r, f1, FR_poly
	nop.i 999
}
{ .mfb
	nop.m 999
//
//      poly = poly + poly_hi	
//
(p11)	fma.s0 FR_Input_X = FR_r, f1, FR_poly
//
//      if (i_0 == 0) Result = r + poly
//      if (i_0 != 0) Result = r - poly
//
(p0)    br.ret.sptk   b0 ;;
}
L(SINCOSL_NORMAL_R): 
{ .mii
	nop.m 999
(p0)	extr.u	GR_i_1 = GR_N_Inc, 0, 1 ;;
//
//      Set table_ptr1 and table_ptr2 to base address of
//      constant table.
(p0)	cmp.eq.unc p9, p10 = 0x0, GR_i_1 ;;
}
{ .mfi
	nop.m 999
(p0)	fma.s1	FR_rsq = FR_r, FR_r, f0
(p0)	extr.u	GR_i_0 = GR_N_Inc, 1, 1 ;;
}
{ .mfi
	nop.m 999
(p0)	frcpa.s1 FR_r_hi, p6 = f1, FR_r
(p0)	cmp.eq.unc p11, p12 = 0x0, GR_i_0
}
;;

// ******************************************************************
// ******************************************************************
// ******************************************************************
//
//      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
//      Get [i_0,i_1] - two lsb of N_fix_gr alone.
//

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

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


{ .mfi
(p10)	add GR_Table_Base = 384, GR_Table_Base
(p12)	fms.s1 FR_Input_X = f0, f1, f1
(p9)	add GR_Table_Base = 224, GR_Table_Base ;;
}
{ .mfi
(p10)	ldfe FR_QQ_8 = [GR_Table_Base], 16
//
//      if (i_1==0) poly = poly * FR_rsq + PP_1_lo
//      else        poly = FR_rsq * poly
//
(p11)	fma.s1 FR_Input_X = f0, f1, f1
	nop.i 999 ;;
}
{ .mmb
(p10)	ldfe FR_QQ_7 = [GR_Table_Base], 16
//
// 	Adjust table pointers based on i_0 
//      Compute rsq = r * r
//
(p9)	ldfe FR_PP_8 = [GR_Table_Base], 16
	nop.b 999 ;;
}
{ .mfi
	nop.m 999
(p0)	fma.s1 FR_r_cubed = FR_r, FR_rsq, f0
	nop.i 999 ;;
}
{ .mmf
(p9)	ldfe FR_PP_7 = [GR_Table_Base], 16
(p10)	ldfe FR_QQ_6 = [GR_Table_Base], 16
//
//      Load PP_8 and QQ_8; PP_7 and QQ_7
//
(p0)	frcpa.s1 FR_r_hi, p6 = f1, FR_r_hi ;;
}
//
//      if (i_1==0) poly =   PP_7 + FR_rsq * PP_8.
//      else        poly =   QQ_7 + FR_rsq * QQ_8.
//
{ .mmb
(p9)	ldfe FR_PP_6 = [GR_Table_Base], 16
(p10)	ldfe FR_QQ_5 = [GR_Table_Base], 16
	nop.b 999 ;;
}
{ .mmb
(p9)	ldfe FR_PP_5 = [GR_Table_Base], 16
(p10)	ldfe FR_S_1 = [GR_Table_Base], 16
	nop.b 999 ;;
}
{ .mmb
(p10)	ldfe FR_QQ_1 = [GR_Table_Base], 16
(p9)	ldfe FR_C_1 = [GR_Table_Base], 16
	nop.b 999 ;;
}
{ .mmb
(p10)	ldfe FR_QQ_4 = [GR_Table_Base], 16
(p9)	ldfe FR_PP_1 = [GR_Table_Base], 16
	nop.b 999 ;;
}
{ .mmb
(p10)	ldfe FR_QQ_3 = [GR_Table_Base], 16
//
//      if (i_1=0) corr = corr + c*c
//      else       corr = corr * c 
//
(p9)	ldfe FR_PP_4 = [GR_Table_Base], 16
	nop.b 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_QQ_8, FR_QQ_7
	nop.i 999 ;;
}
//
//      if (i_1=0) poly = rsq * poly + PP_5 
//      else       poly = rsq * poly + QQ_5 
//      Load PP_4 or QQ_4
//
{ .mmi
(p9)	ldfe FR_PP_3 = [GR_Table_Base], 16 ;;
(p10)	ldfe FR_QQ_2 = [GR_Table_Base], 16
	nop.i 999
}
{ .mfi
	nop.m 999
//
//      r_hi =   frcpa(frcpa(r)).
//      r_cube = r * FR_rsq.
//
(p9)	fma.s1 FR_poly = FR_rsq, FR_PP_8, FR_PP_7
	nop.i 999 ;;
}
//
//      Do dummy multiplies so inexact is always set. 
//
{ .mfi
(p9)	ldfe FR_PP_2 = [GR_Table_Base], 16
//
//      r_lo = r - r_hi	
//
(p9)	fma.s1 FR_U_lo = FR_r_hi, FR_r_hi, f0
	nop.i 999 ;;
}
{ .mbb
(p9)	ldfe FR_PP_1_lo = [GR_Table_Base], 16
	nop.b 999
	nop.b 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_corr = FR_S_1, FR_r_cubed, FR_r
	nop.i 999
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_QQ_6
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1=0) U_lo = r_hi * r_hi
//      else       U_lo = r_hi + r
//
(p9)	fma.s1 FR_corr = FR_C_1, FR_rsq, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1=0) corr = C_1 * rsq
//      else       corr = S_1 * r_cubed + r
//
(p9)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_PP_6
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_U_lo = FR_r_hi, f1, FR_r
	nop.i 999
}
{ .mfi
	nop.m 999
//
//      if (i_1=0) U_hi = r_hi + U_hi 
//      else       U_hi = QQ_1 * U_hi + 1
//
(p9)	fma.s1 FR_U_lo = FR_r, FR_r_hi, FR_U_lo
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      U_hi = r_hi * r_hi	
//
(p0)	fms.s1 FR_r_lo = FR_r, f1, FR_r_hi
	nop.i 999
}
{ .mfi
	nop.m 999
//
//      Load PP_1, PP_6, PP_5, and C_1
//      Load QQ_1, QQ_6, QQ_5, and S_1
//
(p0)	fma.s1 FR_U_hi = FR_r_hi, FR_r_hi, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_QQ_5
	nop.i 999
}
{ .mfi
	nop.m 999
(p10)	fnma.s1	FR_corr = FR_corr, FR_c, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1=0) U_lo = r * r_hi + U_lo 
//      else       U_lo = r_lo * U_lo
//
(p9)	fma.s1 FR_corr = FR_corr, FR_c, FR_c
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p9)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_PP_5
	nop.i 999
}
{ .mfi
	nop.m 999
//
//      if (i_1 =0) U_hi = r + U_hi
//      if (i_1 =0) U_lo = r_lo * U_lo 
//      
//
(p9)	fma.s0 FR_PP_5 = FR_PP_5, FR_PP_4, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p9)	fma.s1 FR_U_lo = FR_r, FR_r, FR_U_lo
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_U_lo = FR_r_lo, FR_U_lo, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1=0) poly = poly * rsq + PP_6
//      else       poly = poly * rsq + QQ_6 
//
(p9)	fma.s1 FR_U_hi = FR_r_hi, FR_U_hi, f0
	nop.i 999
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_QQ_4
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_U_hi = FR_QQ_1, FR_U_hi, f1
	nop.i 999
}
{ .mfi
	nop.m 999
(p10)	fma.s0 FR_QQ_5 = FR_QQ_5, FR_QQ_5, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1!=0) U_hi = PP_1 * U_hi  
//      if (i_1!=0) U_lo = r * r  + U_lo  
//      Load PP_3 or QQ_3
//
(p9)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_PP_4
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p9)	fma.s1 FR_U_lo = FR_r_lo, FR_U_lo, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_U_lo = FR_QQ_1,FR_U_lo, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p9)	fma.s1 FR_U_hi = FR_PP_1, FR_U_hi, f0
	nop.i 999
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_QQ_3
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      Load PP_2, QQ_2
//
(p9)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_PP_3
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1==0) poly = FR_rsq * poly  + PP_3
//      else        poly = FR_rsq * poly  + QQ_3
//      Load PP_1_lo
//
(p9)	fma.s1 FR_U_lo = FR_PP_1, FR_U_lo, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1 =0) poly = poly * rsq + pp_r4
//      else        poly = poly * rsq + qq_r4
//
(p9)	fma.s1 FR_U_hi = FR_r, f1, FR_U_hi
	nop.i 999
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_QQ_2
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1==0) U_lo =  PP_1_hi * U_lo
//      else        U_lo =  QQ_1 * U_lo
//
(p9)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_PP_2
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_0==0)  Result = 1
//      else         Result = -1
//
(p0) 	fma.s1 FR_V = FR_U_lo, f1, FR_corr
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_poly, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1==0) poly =  FR_rsq * poly + PP_2
//      else poly =  FR_rsq * poly + QQ_2
// 
(p9)	fma.s1 FR_poly = FR_rsq, FR_poly, FR_PP_1_lo
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p10)	fma.s1 FR_poly = FR_rsq, FR_poly, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      V = U_lo + corr
//
(p9)	fma.s1 FR_poly = FR_r_cubed, FR_poly, f0
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
//
//      if (i_1==0) poly = r_cube * poly
//      else        poly = FR_rsq * poly
//
(p0)	fma.s1	FR_V = FR_poly, f1, FR_V
	nop.i 999 ;;
}
{ .mfi
	nop.m 999
(p12)	fms.s0 FR_Input_X = FR_Input_X, FR_U_hi, FR_V
	nop.i 999
}
{ .mfb
	nop.m 999
//
//      V = V + poly	
//
(p11)	fma.s0 FR_Input_X = FR_Input_X, FR_U_hi, FR_V
//
//      if (i_0==0) Result = Result * U_hi + V
//      else        Result = Result * U_hi - V
//
(p0)    br.ret.sptk   b0 
};;

//
//    	If cosine, FR_Input_X = 1
//    	If sine, FR_Input_X = +/-Zero (Input FR_Input_X)
//    	Results are exact, no exceptions
//

L(SINCOSL_ZERO):
{ .mbb
(p0)    cmp.eq.unc p6, p7 = 0x1, GR_Sin_or_Cos
        nop.b 999
        nop.b 999 ;;
}
{ .mfi
        nop.m 999
(p7)    fmerge.s FR_Input_X = FR_Input_X, FR_Input_X
        nop.i 999
}
{ .mfb
        nop.m 999
(p6)    fmerge.s FR_Input_X = f1, f1
(p0)    br.ret.sptk   b0 ;;
}
L(SINCOSL_SPECIAL):
{ .mfb
        nop.m 999
//
//      Path for Arg = +/- QNaN, SNaN, Inf
//      Invalid can be raised. SNaNs
//      become QNaNs
//
(p0)    fmpy.s0 FR_Input_X = FR_Input_X, f0
(p0)    br.ret.sptk   b0 ;;
}
.endp cosl#
ASM_SIZE_DIRECTIVE(cosl#)

//      Call int pi_by_2_reduce(double* x, double *y)
//      for |arguments| >= 2**63
//      Address to save r and c as double 
//
//             sp+32  -> f0
//      r45    sp+16  -> f0
//      r44 -> sp     -> InputX  
//      

.proc __libm_callout
__libm_callout:
L(SINCOSL_ARG_TOO_LARGE): 
.prologue
{ .mfi
        add   r45=-32,sp                        // Parameter: r address 
        nop.f 0
.save   ar.pfs,GR_SAVE_PFS
        mov  GR_SAVE_PFS=ar.pfs                 // Save ar.pfs
}
{ .mfi
.fframe 64
        add sp=-64,sp                           // Create new stack
        nop.f 0
        mov GR_SAVE_GP=gp                       // Save gp
};;
{ .mmi
        stfe [r45] = f0,16                      // Clear Parameter r on stack
        add  r44 = 16,sp                        // Parameter x address
.save   b0, GR_SAVE_B0
        mov GR_SAVE_B0=b0                       // Save b0
};;
.body
{ .mib
        stfe [r45] = f0,-16                     // Clear Parameter c on stack 
        nop.i 0
        nop.b 0
}
{ .mib
        stfe [r44] = FR_Input_X                 // Store Parameter x on stack
        nop.i 0
(p0)    br.call.sptk b0=__libm_pi_by_2_reduce# ;;
};;
{ .mii
(p0)    ldfe  FR_Input_X =[r44],16
//
//      Get r and c off stack
//
(p0)    adds  GR_Table_Base1 = -16, GR_Table_Base1
//
//      Get r and c off stack
//
(p0)    add   GR_N_Inc = GR_Sin_or_Cos,r8 ;;
}
{ .mmb
(p0)    ldfe  FR_r =[r45],16
//
//      Get X off the stack
//      Readjust Table ptr
//
(p0)    ldfs FR_Two_to_M3 = [GR_Table_Base1],4
	nop.b 999 ;;
}
{ .mmb
(p0)    ldfs FR_Neg_Two_to_M3 = [GR_Table_Base1],0
(p0)    ldfe  FR_c =[r45]
	nop.b 999 ;;
}
{ .mfi
.restore sp
        add   sp = 64,sp                       // Restore stack pointer
(p0)    fcmp.lt.unc.s1	p6, p0 = FR_r, FR_Two_to_M3
        mov   b0 = GR_SAVE_B0                  // Restore return address
};;
{ .mib
        mov   gp = GR_SAVE_GP                  // Restore gp
        mov   ar.pfs = GR_SAVE_PFS             // Restore ar.pfs
        nop.b 0
};;
{ .mfi
	nop.m 999
(p6)    fcmp.gt.unc.s1	p6, p0 = FR_r, FR_Neg_Two_to_M3
	nop.i 999 ;;
}
{ .mib
	nop.m 999
	nop.i 999
(p6)    br.cond.spnt L(SINCOSL_SMALL_R) ;;
}
{ .mib
	nop.m 999
	nop.i 999
(p0)    br.cond.sptk L(SINCOSL_NORMAL_R) ;;
}
.endp __libm_callout
ASM_SIZE_DIRECTIVE(__libm_callout)
.type   __libm_pi_by_2_reduce#,@function
.global __libm_pi_by_2_reduce#