e_acosl.s

Glibc 2.3.2源代码(解压后有100多M)

      nop.m 999
(p0)  frsqrta.s1     acos_y0,p8  = acos_B                                
      nop.i 999
}

{ .mfi
      nop.m 999
(p0)  fms.s1 acos_1mA = f1,f1, acos_A                                       
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1 acos_Bb  = acos_1mB,f1, acos_ABS_NORM_f8                       
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_Hh     = acos_HALF, acos_B, f0                 
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_t1     = acos_y0, acos_y0, f0                  
      nop.i 999
}

{ .mfi
      nop.m 999
(p0)  fms.s1 acos_Aa  = acos_1mA,f1, acos_ABS_NORM_f8                       
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_t2     = acos_t1, acos_Hh, acos_HALF           
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_y1     = acos_t2, acos_y0, acos_y0             
      nop.i 999
}


// Step 1.2:
/////////////////////////
// Get V = sqrt(A)
/////////////////////////
{ .mfi
      nop.m 999
(p0)  frsqrta.s1     acos_y0,p8  = acos_A                                
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_t3     = acos_y1, acos_Hh, f0                  
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_t1     = acos_y0, acos_y0, f0                  
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_t4     = acos_t3, acos_y1, acos_HALF           
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_y2     = acos_t4, acos_y1, acos_y1             
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_S      = acos_B, acos_y2, f0                   
      nop.i 999
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_H      = acos_y2, acos_HALF, f0                
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_t5     = acos_Hh, acos_y2, f0                  
      nop.i 999
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_Hh     = acos_HALF, acos_A, f0                 
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_Dd     = acos_S, acos_S, acos_B                
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_t2     = acos_t1, acos_Hh, acos_HALF           
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_U      = acos_Dd, acos_H, acos_S               
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_y1     = acos_t2, acos_y0, acos_y0             
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_2U       = acos_U, f1, acos_U                  
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_t3     = acos_y1, acos_Hh, f0                  
      nop.i 999
}


// Step 1.3: 
// sqrt(A + a) = V + v
// sqrt(B + b) = U + u

/////////////////////////
// Get u
/////////////////////////

// acos_BmUU   = B - UU
// acos_BmUUpb = (B - UU) + b

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_BmUU     = acos_U, acos_U, acos_B              
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)   fmerge.se f9 = acos_U, acos_U                           
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_t4     = acos_t3, acos_y1, acos_HALF           
      nop.i 999 ;;
}

// acos_1d2U = frcpa(2U)
{ .mfi
      nop.m 999
(p0)  frcpa.s1       acos_1d2U,p9  = f1, acos_2U                         
      nop.i 999
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_BmUUpb   = acos_BmUU, f1, acos_Bb              
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_y2     = acos_t4, acos_y1, acos_y1             
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
// acos_Uu = ((B - UU) + b) * frcpa(2U)
(p0)  fma.s1         acos_Uu       = acos_BmUUpb, acos_1d2U, f0          
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_S      = acos_A, acos_y2, f0                   
      nop.i 999
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_H      = acos_y2, acos_HALF, f0                
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_t5     = acos_Hh, acos_y2, f0                  
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_Dd     = acos_S, acos_S, acos_A                
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_V      = acos_Dd, acos_H, acos_S               
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_2V       = acos_V, f1, acos_V                  
      nop.i 999
}

// Step 3
/////////////////////////
// Calculate the correction, acos_corr
/////////////////////////
// acos_corr = U*v - (V*u)

{ .mfi
      nop.m 999
(p0)  fma.s1  acos_Vu   = acos_V,acos_Uu, f0                  
      nop.i 999 ;;
}

/////////////////////////
// Get v
/////////////////////////
// acos_AmVV   = A - VV
// acos_AmVVpa = (A - VV) + a

{ .mfi
      nop.m 999
(p0)  fnma.s1        acos_AmVV     = acos_V, acos_V, acos_A              
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)   fmerge.se f8 = acos_V, acos_V                           
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)  fma.s1         acos_AmVVpa   = acos_AmVV, f1, acos_Aa              
      nop.i 999 ;;
}

// acos_1d2V = frcpa(2V)
{ .mfi
      nop.m 999
(p0)  frcpa.s1       acos_1d2V,p9  = f1, acos_2V                         
      nop.i 999 ;;
}

// acos_Vv = ((A - VV) + a) * frcpa(2V)
{ .mfi
      nop.m 999
(p0)  fma.s1         acos_Vv       = acos_AmVVpa, acos_1d2V, f0          
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p0)   fma.s1  acos_Uv   = acos_U,acos_Vv, f0                  
      nop.i 999 ;;
}


.endp acosl#
ASM_SIZE_DIRECTIVE(acosl#)


.proc __libm_callout
__libm_callout:
.prologue
{ .mfi
        nop.m 0
        nop.f 0
.save   ar.pfs,GR_SAVE_PFS
        mov  GR_SAVE_PFS=ar.pfs
}
;;

{ .mfi
        mov GR_SAVE_GP=gp
        nop.f 0
.save   b0, GR_SAVE_B0
        mov GR_SAVE_B0=b0
}

.body
{ .mfb
      nop.m 999
(p0)   fms.s1  acos_corr = acos_Uv,f1, acos_Vu                 
(p0)   br.call.sptk.many  b0=__libm_atan2_reg# ;;                        
}


// p6 ==> X is negative
// p7 ==> x is positive
// We know that |X| >= 1/4

{ .mfi
(p0)   mov gp              = GR_SAVE_GP                           
(p0)   fcmp.lt.unc   p6,p7 = acos_X , f0                       
(p0)   mov b0              = GR_SAVE_B0 ;;                           
}

// acos_2_Z_hi    = 2 * acos_Z_hi
// acos_s_lo_Z_lo = s_lo * Z_lo

{ .mfi
       nop.m 999
(p0)   fma.s1  acos_2_Z_hi      = acos_Z_hi, f1, acos_Z_hi               
(p0)   mov ar.pfs               = GR_SAVE_PFS                                     
}

{ .mfi
      nop.m 999
(p0)   fma.s1  acos_s_lo_Z_lo   = acos_s_lo, acos_Z_lo, f0               
      nop.i 999 ;;
}

// 2 is a constant needed later
{ .mfi
      nop.m 999
(p0)  fma.s1     acos_2 = f1,f1,f1                             
      nop.i 999 ;;
}

// X >= 1/4
// acos_result_lo = 2(s_lo * Z_lo) - corr
// f8             = (2*Z_hi) + (2(s_lo * Z_lo) - corr)

{ .mfi
      nop.m 999
(p7)   fma.s1  acos_result_lo     = acos_s_lo_Z_lo, acos_2, acos_corr      
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p7)  fma.s0   f8                 = acos_2_Z_hi, f1, acos_result_lo        
      nop.i 999
}

// acos_result_lo = (pi_lo - corr)
// acos_result_lo = (pi_lo - corr) + acos_Ww 
{ .mfi
      nop.m 999
(p6)  fms.s1  acos_result_lo     = acos_pi_lo, f1, acos_corr              
      nop.i 999 ;;
}

// X <= -1/4
// acos_W = pi_hi - 2 * Z_hi
{ .mfi
      nop.m 999
(p6)  fnma.s1 acos_W             = acos_2, acos_Z_hi, acos_pi_hi          
      nop.i 999 ;;
}

// acos_Ww = pi_hi - W
// acos_Ww = (pi_hi - W) + (2 * Z_hi)
{ .mfi
      nop.m 999
(p6)  fms.s1  acos_Ww            = acos_pi_hi, f1, acos_W                 
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p6)   fms.s1  acos_Ww            = acos_Ww, f1, acos_2_Z_hi               
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p6)   fma.s1  acos_result_lo     = acos_result_lo, f1, acos_Ww            
      nop.i 999 ;;
}

// acos_Z_lo = ((pi_lo - corr) + acos_Ww) - 2 * (s_lo * Z_lo)
{ .mfi
      nop.m 999
(p6)  fnma.s1  acos_Z_lo          = acos_s_lo_Z_lo, acos_2, acos_result_lo 
      nop.i 999 ;;
}

{ .mfb
      nop.m 999
(p6)  fma.s0   f8                  = acos_W, f1, acos_Z_lo                
(p0)  br.ret.sptk   b0 ;;                          
}
.endp __libm_callout
ASM_SIZE_DIRECTIVE(__libm_callout)

.proc SPECIAL
SPECIAL:
L(ACOS_NAN): 
{ .mfb
      nop.m 999
(p0)  fma.s0 f8 = f8,f1,f0                       
(p0)  br.ret.sptk   b0 ;;                          
}

L(ACOS_ERROR_RETURN): 
// Save ar.pfs, b0, and gp; restore on exit

// qnan snan inf norm     unorm 0 -+
// 1    1    0   0        0     0 11 = 0xc3

// Coming in as X = +- 1
// What should we return?

// If X is 1, return (sign of X)pi/2


{ .mfi
      nop.m 999
(p0)  fcmp.eq.unc p6,p7 = acos_ABS_NORM_f8,f1              
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p6)  fcmp.lt.unc p8,p9 = f8,f0                            
      nop.i 999 ;;
}

{ .mfi
      nop.m 999
(p8)  fma.s0 f8          = acos_pi_hi, f1, acos_pi_lo       
      nop.i 999
}

{ .mfb
      nop.m 999
(p9)  fmerge.s    f8 = f8,f0                               
(p6)  br.ret.spnt   b0 ;;                                     
}

// If X is a NAN, leave
{ .mfi
      nop.m 999
(p0)  fclass.m.unc p12,p0 = f8, 0xc3            
      nop.i 999 ;;
}

{ .mfb
      nop.m 999
(p12) fma.s0 f8 = f8,f1,f0                       
(p12) br.ret.spnt   b0 ;;                          
}

{ .mfi
(p0)   mov   GR_Parameter_TAG = 57 
(p0)   frcpa f10, p6 = f0, f0
nop.i 999
};;

.endp SPECIAL
ASM_SIZE_DIRECTIVE(SPECIAL)

.proc __libm_error_region
__libm_error_region:
.prologue
// (1)
{ .mfi
        add   GR_Parameter_Y=-32,sp             // Parameter 2 value
        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
};;


// (2)
{ .mmi
        stfe [GR_Parameter_Y] = f1,16         // Store Parameter 2 on stack
        add GR_Parameter_X = 16,sp            // Parameter 1 address
.save   b0, GR_SAVE_B0
        mov GR_SAVE_B0=b0                     // Save b0
};;

.body
// (3)
{ .mib
        stfe [GR_Parameter_X] = f8              // Store Parameter 1 on stack
        add   GR_Parameter_RESULT = 0,GR_Parameter_Y
        nop.b 0                                 // Parameter 3 address
}
{ .mib
        stfe [GR_Parameter_Y] = f10             // Store Parameter 3 on stack
        add   GR_Parameter_Y = -16,GR_Parameter_Y
        br.call.sptk b0=__libm_error_support#   // Call error handling function
};;
{ .mmi
        nop.m 0
        nop.m 0
        add   GR_Parameter_RESULT = 48,sp
};;

// (4)
{ .mmi
        ldfe  f8 = [GR_Parameter_RESULT]       // Get return result off stack
.restore sp
        add   sp = 64,sp                       // Restore stack pointer
        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
        br.ret.sptk     b0                     // Return
};;

.endp __libm_error_region
ASM_SIZE_DIRECTIVE(__libm_error_region)

.type   __libm_error_support#,@function
.global __libm_error_support#

.type   __libm_atan2_reg#,@function
.global __libm_atan2_reg#