IA-64 implementation of tanf.
authordrepper <drepper>
Mon, 19 Feb 2001 09:04:37 +0000 (09:04 +0000)
committerdrepper <drepper>
Mon, 19 Feb 2001 09:04:37 +0000 (09:04 +0000)
sysdeps/ia64/fpu/s_tanf.S [new file with mode: 0644]

diff --git a/sysdeps/ia64/fpu/s_tanf.S b/sysdeps/ia64/fpu/s_tanf.S
new file mode 100644 (file)
index 0000000..b4493c1
--- /dev/null
@@ -0,0 +1,757 @@
+.file "tanf.s"
+
+// Copyright (c) 2000, 2001, Intel Corporation
+// All rights reserved.
+// 
+// Contributed 2/2/2000 by John Harrison, Ted Kubaska, Bob Norin, Shane Story,
+// and Ping Tak Peter Tang of the Computational Software Lab, Intel Corporation.
+// 
+// WARRANTY DISCLAIMER
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS 
+// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 
+// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 
+// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY 
+// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 
+// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 
+// 
+// Intel Corporation is the author of this code, and requests that all
+// problem reports or change requests be submitted to it directly at 
+// http://developer.intel.com/opensource.
+//
+// History
+//==============================================================
+// 2/02/00: Initial version
+// 4/04/00  Unwind support added
+// 12/27/00 Improved speed
+//
+// API
+//==============================================================
+// float tan( float x);
+//
+// Overview of operation
+//==============================================================
+// If the input value in radians is |x| >= 1.xxxxx 2^10 call the
+// older slower version.
+//
+// The new algorithm is used when |x| <= 1.xxxxx 2^9.
+//
+// Represent the input X as Nfloat * pi/2 + r
+//    where r can be negative and |r| <= pi/4
+//
+//     tan_W  = x * 2/pi
+//     Nfloat = round_int(tan_W)
+//
+//     tan_r  = x - Nfloat * (pi/2)_hi
+//     tan_r  = tan_r - Nfloat * (pi/2)_lo
+//
+// We have two paths: p8, when Nfloat is even and p9. when Nfloat is odd.
+// p8: tan(X) =  tan(r)
+// p9: tan(X) = -cot(r)
+//
+// Each is evaluated as a series. The p9 path requires 1/r.
+//
+// The coefficients used in the series are stored in a table as
+// are the pi constants.
+//
+// Registers used
+//==============================================================
+//
+// predicate registers used:  
+// p6-10
+//
+// floating-point registers used:  
+// f10-15, f32-105
+// f8, input
+//
+// general registers used
+// r14-18, r32-43
+//
+
+#include "libm_support.h"
+
+// Assembly macros
+//==============================================================
+TAN_INV_PI_BY_2_2TO64        = f10
+TAN_RSHF_2TO64               = f11
+TAN_2TOM64                   = f12
+TAN_RSHF                     = f13
+TAN_W_2TO64_RSH              = f14
+TAN_NFLOAT                   = f15
+
+tan_Inv_Pi_by_2              = f32
+tan_Pi_by_2_hi               = f33
+tan_Pi_by_2_lo               = f34
+
+
+tan_P0                       = f35
+tan_P1                       = f36
+tan_P2                       = f37
+tan_P3                       = f38 
+tan_P4                       = f39 
+tan_P5                       = f40 
+tan_P6                       = f41
+tan_P7                       = f42
+tan_P8                       = f43 
+tan_P9                       = f44 
+tan_P10                      = f45 
+tan_P11                      = f46
+tan_P12                      = f47 
+tan_P13                      = f48
+tan_P14                      = f49
+tan_P15                      = f50
+
+tan_Q0                       = f51 
+tan_Q1                       = f52 
+tan_Q2                       = f53 
+tan_Q3                       = f54 
+tan_Q4                       = f55 
+tan_Q5                       = f56 
+tan_Q6                       = f57 
+tan_Q7                       = f58 
+tan_Q8                       = f59
+tan_Q9                       = f60
+tan_Q10                      = f61
+
+tan_r                        = f62
+tan_rsq                      = f63
+tan_rcube                    = f64
+
+tan_v18                      = f65
+tan_v16                      = f66
+tan_v17                      = f67
+tan_v12                      = f68
+tan_v13                      = f69
+tan_v7                       = f70
+tan_v8                       = f71
+tan_v4                       = f72
+tan_v5                       = f73
+tan_v15                      = f74
+tan_v11                      = f75
+tan_v14                      = f76
+tan_v3                       = f77
+tan_v6                       = f78
+tan_v10                      = f79
+tan_v2                       = f80
+tan_v9                       = f81
+tan_v1                       = f82
+tan_int_Nfloat               = f83 
+tan_Nfloat                   = f84 
+
+tan_NORM_f8                  = f85 
+tan_W                        = f86
+
+tan_y0                       = f87
+tan_d                        = f88 
+tan_y1                       = f89 
+tan_dsq                      = f90 
+tan_y2                       = f91 
+tan_d4                       = f92 
+tan_inv_r                    = f93 
+
+tan_z1                       = f94
+tan_z2                       = f95
+tan_z3                       = f96
+tan_z4                       = f97
+tan_z5                       = f98
+tan_z6                       = f99
+tan_z7                       = f100
+tan_z8                       = f101
+tan_z9                       = f102
+tan_z10                      = f103
+tan_z11                      = f104
+tan_z12                      = f105
+
+
+/////////////////////////////////////////////////////////////
+
+tan_GR_sig_inv_pi_by_2       = r14
+tan_GR_rshf_2to64            = r15
+tan_GR_exp_2tom64            = r16
+tan_GR_n                     = r17
+tan_GR_rshf                  = r18
+
+tan_AD                       = r33
+tan_GR_10009                 = r34 
+tan_GR_17_ones               = r35 
+tan_GR_N_odd_even            = r36 
+tan_GR_N                     = r37 
+tan_signexp                  = r38
+tan_exp                      = r39
+tan_ADQ                      = r40
+
+GR_SAVE_PFS                  = r41 
+GR_SAVE_B0                   = r42       
+GR_SAVE_GP                   = r43      
+
+
+#ifdef _LIBC
+.rodata
+#else
+.data
+#endif
+
+.align 16
+
+double_tan_constants:
+ASM_TYPE_DIRECTIVE(double_tan_constants,@object)
+//   data8 0xA2F9836E4E44152A, 0x00003FFE // 2/pi
+   data8 0xC90FDAA22168C234, 0x00003FFF // pi/2 hi
+
+   data8 0xBEEA54580DDEA0E1 // P14 
+   data8 0x3ED3021ACE749A59 // P15
+   data8 0xBEF312BD91DC8DA1 // P12 
+   data8 0x3EFAE9AFC14C5119 // P13
+   data8 0x3F2F342BF411E769 // P8
+   data8 0x3F1A60FC9F3B0227 // P9
+   data8 0x3EFF246E78E5E45B // P10
+   data8 0x3F01D9D2E782875C // P11
+   data8 0x3F8226E34C4499B6 // P4
+   data8 0x3F6D6D3F12C236AC // P5
+   data8 0x3F57DA1146DCFD8B // P6
+   data8 0x3F43576410FE3D75 // P7
+   data8 0x3FD5555555555555 // P0
+   data8 0x3FC11111111111C2 // P1
+   data8 0x3FABA1BA1BA0E850 // P2
+   data8 0x3F9664F4886725A7 // P3
+ASM_SIZE_DIRECTIVE(double_tan_constants)
+
+double_Q_tan_constants:
+ASM_TYPE_DIRECTIVE(double_Q_tan_constants,@object)
+   data8 0xC4C6628B80DC1CD1, 0x00003FBF // pi/2 lo
+   data8 0x3E223A73BA576E48 // Q8
+   data8 0x3DF54AD8D1F2CA43 // Q9
+   data8 0x3EF66A8EE529A6AA // Q4
+   data8 0x3EC2281050410EE6 // Q5
+   data8 0x3E8D6BB992CC3CF5 // Q6
+   data8 0x3E57F88DE34832E4 // Q7
+   data8 0x3FD5555555555555 // Q0
+   data8 0x3F96C16C16C16DB8 // Q1
+   data8 0x3F61566ABBFFB489 // Q2
+   data8 0x3F2BBD77945C1733 // Q3
+   data8 0x3D927FB33E2B0E04 // Q10
+ASM_SIZE_DIRECTIVE(double_Q_tan_constants)
+
+
+   
+.align 32
+.global tanf#
+#ifdef _LIBC
+.global __tanf#
+#endif
+
+////////////////////////////////////////////////////////
+
+
+
+.section .text
+.proc  tanf#
+#ifdef _LIBC
+.proc  __tanf#
+#endif
+.align 32
+tanf:
+#ifdef _LIBC
+__tanf:
+#endif
+// The initial fnorm will take any unmasked faults and
+// normalize any single/double unorms
+
+{ .mlx
+      alloc          r32=ar.pfs,1,11,0,0               
+      movl tan_GR_sig_inv_pi_by_2 = 0xA2F9836E4E44152A // significand of 2/pi
+}
+{ .mlx
+      addl           tan_AD   = @ltoff(double_tan_constants), gp
+      movl tan_GR_rshf_2to64 = 0x47e8000000000000 // 1.1000 2^(63+63+1)
+}
+;;
+
+{ .mfi
+      ld8 tan_AD = [tan_AD]
+      fnorm     tan_NORM_f8  = f8                      
+      mov tan_GR_exp_2tom64 = 0xffff-64 // exponent of scaling factor 2^-64
+}
+{ .mlx
+      nop.m 999
+      movl tan_GR_rshf = 0x43e8000000000000 // 1.1000 2^63 for right shift
+}
+;;
+
+
+// Form two constants we need
+//   2/pi * 2^1 * 2^63, scaled by 2^64 since we just loaded the significand
+//   1.1000...000 * 2^(63+63+1) to right shift int(W) into the significand
+{ .mmi
+      setf.sig TAN_INV_PI_BY_2_2TO64 = tan_GR_sig_inv_pi_by_2
+      setf.d TAN_RSHF_2TO64 = tan_GR_rshf_2to64
+      mov       tan_GR_17_ones     = 0x1ffff             ;;
+}
+
+
+// Form another constant
+//   2^-64 for scaling Nfloat
+//   1.1000...000 * 2^63, the right shift constant
+{ .mmf
+      setf.exp TAN_2TOM64 = tan_GR_exp_2tom64
+      adds tan_ADQ = double_Q_tan_constants - double_tan_constants, tan_AD
+      fclass.m.unc  p6,p0 = f8, 0x07  // Test for x=0
+}
+;;
+
+
+// Form another constant
+//   2^-64 for scaling Nfloat
+//   1.1000...000 * 2^63, the right shift constant
+{ .mmf
+      setf.d TAN_RSHF = tan_GR_rshf
+      ldfe      tan_Pi_by_2_hi = [tan_AD],16 
+      fclass.m.unc  p7,p0 = f8, 0x23  // Test for x=inf
+}
+;;
+
+{ .mfb
+      ldfe      tan_Pi_by_2_lo = [tan_ADQ],16           
+      fclass.m.unc  p8,p0 = f8, 0xc3  // Test for x=nan
+(p6)  br.ret.spnt    b0    ;;         // Exit for x=0
+}
+
+{ .mfi
+      ldfpd     tan_P14,tan_P15 = [tan_AD],16                         
+(p7)  frcpa.s0  f8,p9=f0,f0           // Set qnan indef if x=inf
+      mov       tan_GR_10009 = 0x10009
+}
+{ .mib
+      ldfpd      tan_Q8,tan_Q9  = [tan_ADQ],16                        
+      nop.i 999
+(p7)  br.ret.spnt    b0    ;;         // Exit for x=inf
+}
+
+{ .mfi
+      ldfpd      tan_P12,tan_P13 = [tan_AD],16                         
+(p8)  fma.s f8=f8,f1,f8               // Set qnan if x=nan
+      nop.i 999
+}
+{ .mib
+      ldfpd      tan_Q4,tan_Q5  = [tan_ADQ],16                        
+      nop.i 999
+(p8)  br.ret.spnt    b0    ;;         // Exit for x=nan
+}
+
+{ .mmi
+      getf.exp  tan_signexp    = tan_NORM_f8                 
+      ldfpd      tan_P8,tan_P9  = [tan_AD],16                         
+      nop.i 999 ;;
+}
+
+// Multiply x by scaled 2/pi and add large const to shift integer part of W to 
+//   rightmost bits of significand
+{ .mfi
+      ldfpd      tan_Q6,tan_Q7  = [tan_ADQ],16
+      fma.s1 TAN_W_2TO64_RSH = tan_NORM_f8,TAN_INV_PI_BY_2_2TO64,TAN_RSHF_2TO64
+      nop.i 999 ;;
+}
+
+{ .mmi
+      ldfpd      tan_P10,tan_P11 = [tan_AD],16                         
+      nop.m 999
+      and       tan_exp = tan_GR_17_ones, tan_signexp         ;;
+}
+
+
+// p7 is true if we must call DBX TAN
+// p7 is true if f8 exp is > 0x10009 (which includes all ones
+//    NAN or inf)
+{ .mmi
+      ldfpd      tan_Q0,tan_Q1  = [tan_ADQ],16                         
+      cmp.ge.unc  p7,p0 = tan_exp,tan_GR_10009               
+      nop.i 999 ;;
+}
+
+
+{ .mmb
+      ldfpd      tan_P4,tan_P5  = [tan_AD],16                         
+      nop.m 999
+(p7)  br.cond.spnt   L(TAN_DBX) ;;                                  
+}
+
+
+{ .mmi
+      ldfpd      tan_Q2,tan_Q3  = [tan_ADQ],16                         
+      nop.m 999
+      nop.i 999 ;;
+}
+
+
+
+// TAN_NFLOAT = Round_Int_Nearest(tan_W)
+{ .mfi
+      ldfpd      tan_P6,tan_P7  = [tan_AD],16                         
+      fms.s1 TAN_NFLOAT = TAN_W_2TO64_RSH,TAN_2TOM64,TAN_RSHF      
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      ldfd      tan_Q10 = [tan_ADQ]
+      nop.f 999
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      ldfpd      tan_P0,tan_P1  = [tan_AD],16                         
+      nop.f 999
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      getf.sig    tan_GR_n = TAN_W_2TO64_RSH
+      nop.f 999
+      nop.i 999 ;;
+}
+
+// tan_r          = -tan_Nfloat * tan_Pi_by_2_hi + x
+{ .mfi
+      ldfpd      tan_P2,tan_P3  = [tan_AD]
+      fnma.s1  tan_r      = TAN_NFLOAT, tan_Pi_by_2_hi,  tan_NORM_f8         
+      nop.i 999 ;;
+}
+
+
+// p8 ==> even
+// p9 ==> odd
+{ .mmi
+      and         tan_GR_N_odd_even = 0x1, tan_GR_n ;;          
+      nop.m 999
+      cmp.eq.unc  p8,p9          = tan_GR_N_odd_even, r0      ;;
+}
+
+
+// tan_r          = tan_r -tan_Nfloat * tan_Pi_by_2_lo 
+{ .mfi
+      nop.m 999
+      fnma.s1  tan_r      = TAN_NFLOAT, tan_Pi_by_2_lo,  tan_r      
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      nop.m 999
+      fma.s1   tan_rsq    = tan_r, tan_r,   f0                      
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p9)  frcpa.s1   tan_y0, p10 = f1,tan_r                  
+      nop.i 999  ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v18 = tan_rsq, tan_P15, tan_P14        
+      nop.i 999
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v4  = tan_rsq, tan_P1, tan_P0          
+      nop.i 999  ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v16 = tan_rsq, tan_P13, tan_P12        
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v17 = tan_rsq, tan_rsq, f0             
+      nop.i 999 ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v12 = tan_rsq, tan_P9, tan_P8          
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v13 = tan_rsq, tan_P11, tan_P10        
+      nop.i 999 ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v7  = tan_rsq, tan_P5, tan_P4          
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v8  = tan_rsq, tan_P7, tan_P6          
+      nop.i 999 ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p9)  fnma.s1    tan_d   = tan_r, tan_y0, f1   
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v5  = tan_rsq, tan_P3, tan_P2          
+      nop.i 999 ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z11 = tan_rsq, tan_Q9, tan_Q8         
+      nop.i 999
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z12 = tan_rsq, tan_rsq, f0            
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v15 = tan_v17, tan_v18, tan_v16        
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z7 = tan_rsq, tan_Q5, tan_Q4          
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v11 = tan_v17, tan_v13, tan_v12        
+      nop.i 999
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z8 = tan_rsq, tan_Q7, tan_Q6          
+      nop.i 999 ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v14 = tan_v17, tan_v17, f0             
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z3 = tan_rsq, tan_Q1, tan_Q0          
+      nop.i 999 ;; 
+}
+
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v3 = tan_v17, tan_v5, tan_v4           
+      nop.i 999
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v6 = tan_v17, tan_v8, tan_v7           
+      nop.i 999 ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p9)  fma.s1     tan_y1  = tan_y0, tan_d, tan_y0    
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1     tan_dsq = tan_d, tan_d, f0        
+      nop.i 999 ;; 
+}
+
+
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z10 = tan_z12, tan_Q10, tan_z11       
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z9  = tan_z12, tan_z12,f0             
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z4 = tan_rsq, tan_Q3, tan_Q2          
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z6  = tan_z12, tan_z8, tan_z7         
+      nop.i 999 ;; 
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v10 = tan_v14, tan_v15, tan_v11        
+      nop.i 999 ;; 
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p9)  fma.s1     tan_y2  = tan_y1, tan_d, tan_y0         
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1     tan_d4  = tan_dsq, tan_dsq, tan_d       
+      nop.i 999  ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v2 = tan_v14, tan_v6, tan_v3           
+      nop.i 999
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v9 = tan_v14, tan_v14, f0              
+      nop.i 999 ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z2  = tan_z12, tan_z4, tan_z3         
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z5  = tan_z9, tan_z10, tan_z6         
+      nop.i 999  ;;
+}
+
+
+{ .mfi
+      nop.m 999
+(p9)  fma.s1     tan_inv_r = tan_d4, tan_y2, tan_y0      
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p8)  fma.s1   tan_rcube  = tan_rsq, tan_r,   f0
+      nop.i 999  ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s1  tan_v1 = tan_v9, tan_v10, tan_v2           
+      nop.i 999 
+}
+{ .mfi
+      nop.m 999
+(p9)  fma.s1  tan_z1  = tan_z9, tan_z5, tan_z2          
+      nop.i 999   ;;
+}
+
+
+
+{ .mfi
+      nop.m 999
+(p8)  fma.s.s0  f8  = tan_v1, tan_rcube, tan_r             
+      nop.i 999  
+}
+{ .mfb
+      nop.m 999
+(p9)  fms.s.s0  f8  = tan_r, tan_z1, tan_inv_r        
+      br.ret.sptk    b0 ;;    
+}
+.endp tanf#
+ASM_SIZE_DIRECTIVE(tanf#)
+
+
+.proc __libm_callout
+__libm_callout:
+L(TAN_DBX): 
+.prologue
+
+{ .mfi
+        nop.m 0
+     fmerge.s f9 = f0,f0 
+.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
+      nop.f 999
+       br.call.sptk.many  b0=__libm_tan# ;;
+}
+
+
+{ .mfi
+       mov gp        = GR_SAVE_GP
+      fnorm.s     f8 = f8
+       mov b0        = GR_SAVE_B0 
+}
+;;
+
+
+{ .mib
+         nop.m 999
+      mov ar.pfs    = GR_SAVE_PFS
+      br.ret.sptk     b0
+;;
+}
+
+
+.endp  __libm_callout
+ASM_SIZE_DIRECTIVE(__libm_callout)
+
+.type __libm_tan#,@function
+.global __libm_tan#