s_asinhl.s

glibc 2.9,最新版的C语言库函数

      nop.i 0
};;

{ .mfi
      nop.m 0
      fadd.s1   FR_h          = FR_h, FR_h2              // h = h_1 + h_2
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1    FR_S_lo       = FR_XLog_Lo, FR_2_to_minus_N, f0 //S_lo=S_lo*2^-N
      nop.i 0
};;

{ .mfi
      nop.m 0
      fmpy.s1   FR_G          = FR_G, FR_G3             // G = (G_1 * G_2) * G_3
      nop.i 0
}
{ .mfi
      nop.m 0
      fadd.s1   FR_H          = FR_H, FR_H3             // H = (H_1 + H_2) + H_3
      nop.i 0
};;

{ .mfi
      nop.m 0
      fadd.s1   FR_h          = FR_h, FR_h3             // h = (h_1 + h_2) + h_3
      nop.i 0
};;

{ .mfi
      nop.m 0
      fms.s1    FR_r          = FR_G, FR_S_hi, f1           // r = G * S_hi - 1
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1    FR_Y_hi       = FR_float_N, FR_log2_hi, FR_H // Y_hi=N*log2_hi+H
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_h          = FR_float_N, FR_log2_lo, FR_h  // h=N*log2_lo+h
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1    FR_r          = FR_G, FR_S_lo, FR_r  // r=G*S_lo+(G*S_hi-1)
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_poly_lo    = FR_r, FR_Q4, FR_Q3      // poly_lo = r * Q4 + Q3
      nop.i 0
}
{ .mfi
      nop.m 0
      fmpy.s1   FR_rsq        = FR_r, FR_r              // rsq = r * r
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_poly_lo    = FR_poly_lo, FR_r, FR_Q2 // poly_lo=poly_lo*r+Q2
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1    FR_rcub       = FR_rsq, FR_r, f0        // rcub = r^3
      nop.i 0
};;

.pred.rel "mutex",p12,p11
{ .mfi
      nop.m 0
(p12) fma.s1    FR_poly_hi    = FR_Q1, FR_rsq, FR_r // poly_hi = Q1*rsq + r
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fms.s1    FR_poly_hi    = FR_Q1, FR_rsq, FR_r     // poly_hi = Q1*rsq + r
      nop.i 0
};;


.pred.rel "mutex",p12,p11
{ .mfi
      nop.m 0
(p12) fma.s1    FR_poly_lo    = FR_poly_lo, FR_rcub, FR_h//poly_lo=poly_lo*r^3+h
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fms.s1    FR_poly_lo    = FR_poly_lo, FR_rcub, FR_h//poly_lo=poly_lo*r^3+h
      nop.i 0
}
;;

{ .mfi
      nop.m 0
      fadd.s0   FR_Y_lo       = FR_poly_hi, FR_poly_lo  
	                                                     // Y_lo=poly_hi+poly_lo
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fma.s0    FR_Y_hi       = FR_Y_hi, FR_Neg_One, f0 // FR_Y_hi sign for neg
      nop.i 0
};;

{ .mfb
      nop.m 0
      fadd.s0   FR_Res        = FR_Y_lo,FR_Y_hi    // Result=Y_lo+Y_hi
      br.ret.sptk   b0                         // Common exit for 2^-7 < x < inf
};;

// * SPECIAL VERSION OF LOGL FOR HUGE ARGUMENTS *

huges_logl:
{ .mfi
      getf.sig  GR_signif     = FR_XLog_Hi     // Get significand of x+1
      fmerge.ns FR_Neg_One    = f1, f1         // Form -1.0
      mov       GR_exp_2tom7  = 0x0fff8        // Exponent of 2^-7
};;

{ .mfi
      add       GR_ad_tbl_1   = 0x040, GR_ad_z_1    // Point to Constants_G_H_h1
      nop.f 0
      add       GR_ad_q       = -0x60, GR_ad_z_1    // Point to Constants_P
}
{ .mfi
      add       GR_ad_z_2     = 0x140, GR_ad_z_1    // Point to Constants_Z_2
      nop.f 0
      add       GR_ad_tbl_2   = 0x180, GR_ad_z_1    // Point to Constants_G_H_h2
};;

{ .mfi
      nop.m 0
      nop.f 0
      extr.u    GR_Index1     = GR_signif, 59, 4    // Get high 4 bits of signif
}
{ .mfi
      add       GR_ad_tbl_3   = 0x280, GR_ad_z_1    // Point to Constants_G_H_h3
      nop.f 0
      nop.i 0
};;

{ .mfi
      shladd    GR_ad_z_1     = GR_Index1, 2, GR_ad_z_1  // Point to Z_1
      nop.f 0
      extr.u    GR_X_0        = GR_signif, 49, 15 // Get high 15 bits of signif.
};;

{ .mfi
      ld4       GR_Z_1        = [GR_ad_z_1]    // Load Z_1
      nop.f 0
      mov       GR_exp_mask   = 0x1FFFF        // Create exponent mask
}
{ .mfi
      shladd    GR_ad_tbl_1   = GR_Index1, 4, GR_ad_tbl_1  // Point to G_1
      nop.f 0
      mov       GR_Bias       = 0x0FFFF            // Create exponent bias
};;

{ .mfi
      ldfps     FR_G, FR_H    = [GR_ad_tbl_1],8     // Load G_1, H_1
      fmerge.se FR_S_hi       =  f1,FR_XLog_Hi            // Form |x+1|
      nop.i 0
};;

{ .mmi
      getf.exp  GR_N          =  FR_XLog_Hi          // Get N = exponent of x+1
      ldfd      FR_h          = [GR_ad_tbl_1]        // Load h_1
      nop.i 0
};;

{ .mfi
      ldfe      FR_log2_hi    = [GR_ad_q],16      // Load log2_hi
      nop.f 0
      pmpyshr2.u GR_X_1       = GR_X_0,GR_Z_1,15  // Get bits 30-15 of X_0 * Z_1
};;

// WE CANNOT USE GR_X_1 IN NEXT 3 CYCLES BECAUSE OF POSSIBLE 10 CLOCKS STALL!
// "DEAD" ZONE!

{ .mmi
      ldfe      FR_log2_lo    = [GR_ad_q],16      // Load log2_lo
      sub       GR_N          = GR_N, GR_Bias 
      mov       GR_exp_2tom80 = 0x0ffaf           // Exponent of 2^-80
};;

{ .mfi
      ldfe      FR_Q4         = [GR_ad_q],16          // Load Q4
      nop.f 0
      sub       GR_minus_N    = GR_Bias, GR_N         // Form exponent of 2^(-N)
};;

{ .mmf
      ldfe      FR_Q3         = [GR_ad_q],16   // Load Q3
      setf.sig  FR_float_N    = GR_N        // Put integer N into rightmost sign
      nop.f 0
};;

{ .mmi
      nop.m 0
      ldfe      FR_Q2         = [GR_ad_q],16      // Load Q2
      extr.u    GR_Index2     = GR_X_1, 6, 4      // Extract bits 6-9 of X_1 
};;

{ .mmi
      ldfe      FR_Q1         = [GR_ad_q]                // Load Q1
      shladd    GR_ad_z_2     = GR_Index2, 2, GR_ad_z_2  // Point to Z_2
      nop.i 0
};;

{ .mmi
      ld4       GR_Z_2        = [GR_ad_z_2]                // Load Z_2
      shladd    GR_ad_tbl_2   = GR_Index2, 4, GR_ad_tbl_2  // Point to G_2
      nop.i 0
};;

{ .mmi
      ldfps     FR_G2, FR_H2  = [GR_ad_tbl_2],8       // Load G_2, H_2
      nop.m 0
      nop.i 0
};;

{ .mfi
      ldfd      FR_h2         = [GR_ad_tbl_2]             // Load h_2
      nop.f 0
      nop.i 0
}
{ .mfi
      setf.exp  FR_2_to_minus_N = GR_minus_N   // Form 2^(-N)
      nop.f 0
      nop.i 0
};;

{ .mfi
      nop.m 0
      nop.f 0
      pmpyshr2.u GR_X_2       = GR_X_1,GR_Z_2,15 // Get bits 30-15 of X_1 * Z_2
};;

// WE CANNOT USE GR_X_2 IN NEXT 3 CYCLES BECAUSE OF POSSIBLE 10 CLOCKS STALL!
// "DEAD" ZONE!
// JUST HAVE TO INSERT 3 NOP CYCLES (nothing to do here)

{ .mfi
      nop.m 0
      nop.f 0
      nop.i 0
};;

{ .mfi
      nop.m 0
      nop.f 0
      nop.i 0
};;

{ .mfi
      nop.m 0
      nop.f 0
      nop.i 0
};;

{ .mfi
      nop.m 0
(p11) fma.s1    FR_Q4         = FR_Q4, FR_Neg_One, f0 // Negate Q4
      extr.u    GR_Index3     = GR_X_2, 1, 5          // Extract bits 1-5 of X_2
 };;

{ .mfi
      shladd    GR_ad_tbl_3   = GR_Index3, 4, GR_ad_tbl_3  // Point to G_3
	  fcvt.xf   FR_float_N    = FR_float_N
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fma.s1    FR_Q3         = FR_Q3, FR_Neg_One, f0 // Negate Q3
      nop.i 0
};;

{ .mfi
      ldfps     FR_G3, FR_H3  = [GR_ad_tbl_3],8   // Load G_3, H_3
(p11) fma.s1    FR_Q2         = FR_Q2, FR_Neg_One, f0 // Negate Q2
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fma.s1    FR_Q1         = FR_Q1, FR_Neg_One, f0 // Negate Q1
      nop.i 0
};;

{ .mfi
      ldfd      FR_h3         = [GR_ad_tbl_3]            // Load h_3
      fmpy.s1   FR_G          = FR_G, FR_G2              // G = G_1 * G_2
      nop.i 0
}
{ .mfi
      nop.m 0
      fadd.s1   FR_H          = FR_H, FR_H2              // H = H_1 + H_2
      nop.i 0
};;

{ .mmf
      nop.m 0
      nop.m 0
      fadd.s1   FR_h          = FR_h, FR_h2              // h = h_1 + h_2
};;

{ .mfi
      nop.m 0
      fmpy.s1   FR_G          = FR_G, FR_G3             // G = (G_1 * G_2) * G_3
      nop.i 0
}
{ .mfi
      nop.m 0
      fadd.s1   FR_H          = FR_H, FR_H3             // H = (H_1 + H_2) + H_3
      nop.i 0
};;

{ .mfi
      nop.m 0
      fadd.s1   FR_h          = FR_h, FR_h3             // h = (h_1 + h_2) + h_3
      nop.i 0
};;

{ .mfi
      nop.m 0
      fms.s1    FR_r          = FR_G, FR_S_hi, f1           // r = G * S_hi - 1
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1    FR_Y_hi       = FR_float_N, FR_log2_hi, FR_H // Y_hi=N*log2_hi+H
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_h          = FR_float_N, FR_log2_lo, FR_h  // h=N*log2_lo+h
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_poly_lo    = FR_r, FR_Q4, FR_Q3      // poly_lo = r * Q4 + Q3
      nop.i 0
}
{ .mfi
      nop.m 0
      fmpy.s1   FR_rsq        = FR_r, FR_r              // rsq = r * r
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_poly_lo    = FR_poly_lo, FR_r, FR_Q2 // poly_lo=poly_lo*r+Q2
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1    FR_rcub       = FR_rsq, FR_r, f0        // rcub = r^3
      nop.i 0
};;

.pred.rel "mutex",p12,p11
{ .mfi
      nop.m 0
(p12) fma.s1    FR_poly_hi    = FR_Q1, FR_rsq, FR_r // poly_hi = Q1*rsq + r
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fms.s1    FR_poly_hi    = FR_Q1, FR_rsq, FR_r // poly_hi = Q1*rsq + r
      nop.i 0
};;


.pred.rel "mutex",p12,p11
{ .mfi
      nop.m 0
(p12) fma.s1    FR_poly_lo    = FR_poly_lo, FR_rcub, FR_h//poly_lo=poly_lo*r^3+h
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fms.s1    FR_poly_lo    = FR_poly_lo, FR_rcub, FR_h//poly_lo=poly_lo*r^3+h
      nop.i 0
};;

{ .mfi
      nop.m 0
      fadd.s0   FR_Y_lo       = FR_poly_hi, FR_poly_lo  // Y_lo=poly_hi+poly_lo
      nop.i 0
}
{ .mfi
      nop.m 0
(p11) fma.s0    FR_Y_hi       = FR_Y_hi, FR_Neg_One, f0 // FR_Y_hi sign for neg
      nop.i 0
};;

{ .mfb
      nop.m 0
      fadd.s0   FR_Res        = FR_Y_lo,FR_Y_hi    // Result=Y_lo+Y_hi
      br.ret.sptk   b0                         // Common exit for 2^-7 < x < inf
};;

// NEAR ZERO POLYNOMIAL INTERVAL
near_0:
{ .mfi
      nop.m 0
      fma.s1    FR_X4         = FR_X2, FR_X2, f0 // x^4 = x^2 * x^2
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_P9         = FR_C9,FR_X2,FR_C7  // p9 = C9*x^2 + C7
      nop.i 0
}
{ .mfi
      nop.m 0
      fma.s1    FR_P5         = FR_C5,FR_X2,FR_C3  // p5 = C5*x^2 + C3
      nop.i 0
};;

{ .mfi
      nop.m 0
      fma.s1    FR_P3         = FR_P9,FR_X4,FR_P5  // p3 = p9*x^4 + p5
      nop.i 0
};;

{ .mfb
      nop.m 0
      fma.s0    FR_Res        = FR_P3,FR_X3,FR_Arg // res = p3*C3 + x
      br.ret.sptk   b0                          // Near 0 path return
};;

GLOBAL_LIBM_END(asinhl)