diff mbox series

[5/5] ia64: Use generic exp10f

Message ID 20200409195945.10393-5-adhemerval.zanella@linaro.org
State New
Headers show
Series [1/5] benchtests: Add exp10f benchmark | expand

Commit Message

Adhemerval Zanella April 9, 2020, 7:59 p.m. UTC
The generic implementation is slight worse (Itanium(R) Processor 9020):

Before new code:
  "exp10f": {
   "": {
    "duration": 3.52876e+08,
    "iterations": 2.1456e+07,
    "max": 232.907,
    "min": 7.766,
    "mean": 16.4465
   }
  }

With new code:
  "exp10f": {
   "": {
    "duration": 3.67279e+08,
    "iterations": 2.1456e+07,
    "max": 220.718,
    "min": 15.503,
    "mean": 17.1178
   }
  }

However it fixes all the issues on both:

  math/test-float-exp10
  math/test-float32-exp10

(all the issues wrong results for non default rounding modes).

Checked on ia64-linux-gnu.
---
 sysdeps/ia64/fpu/e_exp10f.S        | 568 -----------------------------
 sysdeps/ia64/fpu/e_exp2f_data.c    |   1 -
 sysdeps/ia64/fpu/libm-test-ulps    |   3 +
 sysdeps/ia64/fpu/math_errf.c       |   1 -
 sysdeps/ia64/fpu/w_exp10f.c        |   1 -
 sysdeps/ia64/fpu/w_exp10f_compat.c |   1 -
 6 files changed, 3 insertions(+), 572 deletions(-)
 delete mode 100644 sysdeps/ia64/fpu/e_exp10f.S
 delete mode 100644 sysdeps/ia64/fpu/e_exp2f_data.c
 delete mode 100644 sysdeps/ia64/fpu/math_errf.c
 delete mode 100644 sysdeps/ia64/fpu/w_exp10f.c
 delete mode 100644 sysdeps/ia64/fpu/w_exp10f_compat.c

Comments

Joseph Myers April 9, 2020, 9:18 p.m. UTC | #1
For most ia64 functions there would be the issue that the existing ia64 
libm interface uses matherrf and matherrl in addition to matherr for SVID 
error handling, so replacing an existing float implementation with an 
architecture-independent one would break the ABI for code using those 
functions.  However, it looks like there is no such error handling support 
for exp10f in ia64 libm, so maybe that isn't an issue; any existing 
programs expecting SVID error handling and using exp10f would already fail 
to function correctly.
Adhemerval Zanella April 16, 2020, 8:52 p.m. UTC | #2
On 09/04/2020 18:18, Joseph Myers wrote:
> For most ia64 functions there would be the issue that the existing ia64 
> libm interface uses matherrf and matherrl in addition to matherr for SVID 
> error handling, so replacing an existing float implementation with an 
> architecture-independent one would break the ABI for code using those 
> functions.  However, it looks like there is no such error handling support 
> for exp10f in ia64 libm, so maybe that isn't an issue; any existing 
> programs expecting SVID error handling and using exp10f would already fail 
> to function correctly.
> 

Indeed, I have the following note on the commit message:

  The existing ia64 libm interface uses matherrf and matherrl in addition
  to matherr for SVID error handling. However, there is no such error
  handling support for exp10f in ia64 libm.  So replacing it with the
  generic implementation should be fine.

Regarding ia64 libm functions, I am seeing a lot of issues on math tests,
specially in non-default rounding modes. And I think it would be way simpler
to just replace them with the generic interface instead of fiddling with
ia64 assembly. 

For the compat symbols, one possibility might to reimplement a ia64
sysdeps/ieee754/k_standard.c with the expected semantic (and replace
libm_error.c).
Adhemerval Zanella May 20, 2020, 8:19 p.m. UTC | #3
Ping (x2).

On 29/04/2020 14:11, Adhemerval Zanella wrote:
> Ping.
> 
> On 09/04/2020 16:59, Adhemerval Zanella wrote:
>> The generic implementation is slight worse (Itanium(R) Processor 9020):
>>
>> Before new code:
>>   "exp10f": {
>>    "": {
>>     "duration": 3.52876e+08,
>>     "iterations": 2.1456e+07,
>>     "max": 232.907,
>>     "min": 7.766,
>>     "mean": 16.4465
>>    }
>>   }
>>
>> With new code:
>>   "exp10f": {
>>    "": {
>>     "duration": 3.67279e+08,
>>     "iterations": 2.1456e+07,
>>     "max": 220.718,
>>     "min": 15.503,
>>     "mean": 17.1178
>>    }
>>   }
>>
>> However it fixes all the issues on both:
>>
>>   math/test-float-exp10
>>   math/test-float32-exp10
>>
>> (all the issues wrong results for non default rounding modes).
>>
>> Checked on ia64-linux-gnu.
>> ---
>>  sysdeps/ia64/fpu/e_exp10f.S        | 568 -----------------------------
>>  sysdeps/ia64/fpu/e_exp2f_data.c    |   1 -
>>  sysdeps/ia64/fpu/libm-test-ulps    |   3 +
>>  sysdeps/ia64/fpu/math_errf.c       |   1 -
>>  sysdeps/ia64/fpu/w_exp10f.c        |   1 -
>>  sysdeps/ia64/fpu/w_exp10f_compat.c |   1 -
>>  6 files changed, 3 insertions(+), 572 deletions(-)
>>  delete mode 100644 sysdeps/ia64/fpu/e_exp10f.S
>>  delete mode 100644 sysdeps/ia64/fpu/e_exp2f_data.c
>>  delete mode 100644 sysdeps/ia64/fpu/math_errf.c
>>  delete mode 100644 sysdeps/ia64/fpu/w_exp10f.c
>>  delete mode 100644 sysdeps/ia64/fpu/w_exp10f_compat.c
>>
>> diff --git a/sysdeps/ia64/fpu/e_exp10f.S b/sysdeps/ia64/fpu/e_exp10f.S
>> deleted file mode 100644
>> index e82394b725..0000000000
>> --- a/sysdeps/ia64/fpu/e_exp10f.S
>> +++ /dev/null
>> @@ -1,568 +0,0 @@
>> -.file "exp10f.s"
>> -
>> -
>> -// Copyright (c) 2000 - 2005, Intel Corporation
>> -// All rights reserved.
>> -//
>> -// Contributed 2000 by the Intel Numerics Group, Intel Corporation
>> -//
>> -// Redistribution and use in source and binary forms, with or without
>> -// modification, are permitted provided that the following conditions are
>> -// met:
>> -//
>> -// * Redistributions of source code must retain the above copyright
>> -// notice, this list of conditions and the following disclaimer.
>> -//
>> -// * Redistributions in binary form must reproduce the above copyright
>> -// notice, this list of conditions and the following disclaimer in the
>> -// documentation and/or other materials provided with the distribution.
>> -//
>> -// * The name of Intel Corporation may not be used to endorse or promote
>> -// products derived from this software without specific prior written
>> -// permission.
>> -
>> -// 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://www.intel.com/software/products/opensource/libraries/num.htm.
>> -//
>> -// History
>> -//==============================================================
>> -// 08/25/00 Initial version
>> -// 05/20/02 Cleaned up namespace and sf0 syntax
>> -// 09/06/02 Improved performance and accuracy; no inexact flags on exact cases
>> -// 01/29/03 Added missing } to bundle templates
>> -// 12/16/04 Call error handling on underflow.
>> -// 03/31/05 Reformatted delimiters between data tables
>> -//
>> -// API
>> -//==============================================================
>> -// float exp10f(float)
>> -//
>> -// Overview of operation
>> -//==============================================================
>> -// Background
>> -//
>> -// Implementation
>> -//
>> -// Let x= (K + fh + fl + r)/log2(10), where
>> -// K is an integer, fh= 0.b1 b2 b3 b4 b5,
>> -// fl= 2^{-5}* 0.b6 b7 b8 b8 b10 (fh, fl >= 0),
>> -// and |r|<2^{-11}
>> -// Th is a table that stores 2^fh (32 entries) rounded to
>> -// double extended precision (only mantissa is stored)
>> -// Tl is a table that stores 2^fl (32 entries) rounded to
>> -// double extended precision (only mantissa is stored)
>> -//
>> -// 10^x is approximated as
>> -// 2^K * Th [ f ] * Tl [ f ] * (1+c1*r+c2*r^2)
>> -
>> -// Note there are only 10 non-zero values that produce an exact result:
>> -//  1.0, 2.0, ... 10.0.
>> -// We test for these cases and use s1 to avoid setting the inexact flag.
>> -
>> -// Special values
>> -//==============================================================
>> -// exp10(0)= 1
>> -// exp10(+inf)= inf
>> -// exp10(-inf)= 0
>> -//
>> -
>> -// Registers used
>> -//==============================================================
>> -// r2-r3, r14-r40
>> -// f6-f15, f32-f52
>> -// p6-p12
>> -//
>> -
>> -#include <shlib-compat.h>
>> -
>> -
>> -GR_TBL_START        = r2
>> -GR_LOG_TBL          = r3
>> -
>> -GR_OF_LIMIT         = r14
>> -GR_UF_LIMIT         = r15
>> -GR_EXP_CORR         = r16
>> -GR_F_low            = r17
>> -GR_F_high           = r18
>> -GR_K                = r19
>> -GR_Flow_ADDR        = r20
>> -
>> -GR_BIAS             = r21
>> -GR_Fh               = r22
>> -GR_Fh_ADDR          = r23
>> -GR_EXPMAX           = r24
>> -
>> -GR_ROUNDVAL         = r26
>> -GR_SNORM_LIMIT      = r26
>> -GR_MASK             = r27
>> -GR_KF0              = r28
>> -GR_MASK_low         = r29
>> -GR_COEFF_START      = r30
>> -GR_exact_limit      = r31
>> -
>> -GR_SAVE_B0          = r33
>> -GR_SAVE_PFS         = r34
>> -GR_SAVE_GP          = r35
>> -GR_SAVE_SP          = r36
>> -
>> -GR_Parameter_X      = r37
>> -GR_Parameter_Y      = r38
>> -GR_Parameter_RESULT = r39
>> -GR_Parameter_TAG    = r40
>> -
>> -
>> -FR_X                = f10
>> -FR_Y                = f1
>> -FR_RESULT           = f8
>> -
>> -
>> -FR_COEFF1           = f6
>> -FR_COEFF2           = f7
>> -FR_R                = f9
>> -FR_LOG2_10          = f10
>> -
>> -FR_2P53             = f11
>> -FR_KF0              = f12
>> -FR_COEFF3           = f13
>> -FR_COEFF4           = f14
>> -FR_UF_LIMIT         = f15
>> -
>> -FR_OF_LIMIT         = f32
>> -FR_DX_L210          = f33
>> -FR_ROUNDVAL         = f34
>> -FR_KF               = f35
>> -
>> -FR_2_TO_K           = f36
>> -FR_T_low            = f37
>> -FR_T_high           = f38
>> -
>> -FR_P12              = f41
>> -FR_T_low_K          = f42
>> -FR_T                = f44
>> -FR_P                = f45
>> -
>> -FR_E                = f49
>> -FR_exact_limit      = f50
>> -
>> -FR_int_x            = f51
>> -FR_SNORM_LIMIT      = f52
>> -
>> -
>> -// Data tables
>> -//==============================================================
>> -
>> -RODATA
>> -
>> -.align 16
>> -
>> -LOCAL_OBJECT_START(poly_coeffs)
>> -
>> -data8 0xd49a784bcd1b8afe, 0x00003fcb // log2(10)*2^(10-63)
>> -data8 0xb17217f7d1cf79ab, 0x00004033 // C_1 * 2^53
>> -data8 0xf5fdeffc162c7541, 0x00004066 // C_2 * 2^106
>> -LOCAL_OBJECT_END(poly_coeffs)
>> -
>> -
>> -LOCAL_OBJECT_START(T_table)
>> -
>> -// 2^{0.00000 b6 b7 b8 b9 b10}
>> -data8 0x8000000000000000, 0x8016302f17467628
>> -data8 0x802c6436d0e04f50, 0x80429c17d77c18ed
>> -data8 0x8058d7d2d5e5f6b0, 0x806f17687707a7af
>> -data8 0x80855ad965e88b83, 0x809ba2264dada76a
>> -data8 0x80b1ed4fd999ab6c, 0x80c83c56b50cf77f
>> -data8 0x80de8f3b8b85a0af, 0x80f4e5ff089f763e
>> -data8 0x810b40a1d81406d4, 0x81219f24a5baa59d
>> -data8 0x813801881d886f7b, 0x814e67cceb90502c
>> -data8 0x8164d1f3bc030773, 0x817b3ffd3b2f2e47
>> -data8 0x8191b1ea15813bfd, 0x81a827baf7838b78
>> -data8 0x81bea1708dde6055, 0x81d51f0b8557ec1c
>> -data8 0x81eba08c8ad4536f, 0x820225f44b55b33b
>> -data8 0x8218af4373fc25eb, 0x822f3c7ab205c89a
>> -data8 0x8245cd9ab2cec048, 0x825c62a423d13f0c
>> -data8 0x8272fb97b2a5894c, 0x828998760d01faf3
>> -data8 0x82a0393fe0bb0ca8, 0x82b6ddf5dbc35906
>> -//
>> -// 2^{0.b1 b2 b3 b4 b5}
>> -data8 0x8000000000000000, 0x82cd8698ac2ba1d7
>> -data8 0x85aac367cc487b14, 0x88980e8092da8527
>> -data8 0x8b95c1e3ea8bd6e6, 0x8ea4398b45cd53c0
>> -data8 0x91c3d373ab11c336, 0x94f4efa8fef70961
>> -data8 0x9837f0518db8a96f, 0x9b8d39b9d54e5538
>> -data8 0x9ef5326091a111ad, 0xa27043030c496818
>> -data8 0xa5fed6a9b15138ea, 0xa9a15ab4ea7c0ef8
>> -data8 0xad583eea42a14ac6, 0xb123f581d2ac258f
>> -data8 0xb504f333f9de6484, 0xb8fbaf4762fb9ee9
>> -data8 0xbd08a39f580c36be, 0xc12c4cca66709456
>> -data8 0xc5672a115506dadd, 0xc9b9bd866e2f27a2
>> -data8 0xce248c151f8480e3, 0xd2a81d91f12ae45a
>> -data8 0xd744fccad69d6af4, 0xdbfbb797daf23755
>> -data8 0xe0ccdeec2a94e111, 0xe5b906e77c8348a8
>> -data8 0xeac0c6e7dd24392e, 0xefe4b99bdcdaf5cb
>> -data8 0xf5257d152486cc2c, 0xfa83b2db722a033a
>> -LOCAL_OBJECT_END(T_table)
>> -
>> -
>> -
>> -.section .text
>> -GLOBAL_IEEE754_ENTRY(__exp10f)
>> -
>> -
>> -{.mfi
>> -       alloc r32= ar.pfs, 1, 4, 4, 0
>> -       // will continue only for non-zero normal/denormal numbers
>> -       fclass.nm.unc p12, p7= f8, 0x1b
>> -       nop.i 0
>> -}
>> -{.mlx
>> -       // GR_TBL_START= pointer to log2(10), C_1...C_4 followed by T_table
>> -       addl GR_TBL_START= @ltoff(poly_coeffs), gp
>> -       movl GR_ROUNDVAL= 0x3fc00000             // 1.5 (SP)
>> -}
>> -;;
>> -
>> -{.mfi
>> -       ld8 GR_COEFF_START= [ GR_TBL_START ]     // Load pointer to coeff table
>> -       fcmp.lt.s1 p6, p8= f8, f0                // X<0 ?
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mlx
>> -       nop.m 0
>> -       movl GR_UF_LIMIT= 0xc2349e35             // (-2^7-22) / log2(10)
>> -}
>> -{.mlx
>> -       setf.s FR_ROUNDVAL= GR_ROUNDVAL
>> -       movl GR_OF_LIMIT= 0x421a209a             // Overflow threshold
>> -}
>> -;;
>> -
>> -{.mlx
>> -       ldfe FR_LOG2_10= [ GR_COEFF_START ], 16  // load log2(10)*2^(10-63)
>> -       movl GR_SNORM_LIMIT= 0xc217b818          // Smallest normal threshold
>> -}
>> -{.mib
>> -       nop.m 0
>> -       nop.i 0
>> - (p12) br.cond.spnt SPECIAL_exp10               // Branch if nan, inf, zero
>> -}
>> -;;
>> -
>> -{.mfi
>> -       setf.s FR_OF_LIMIT= GR_OF_LIMIT           // Set overflow limit
>> -       fma.s0 f8= f8, f1, f0                     // normalize x
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mfi
>> -       setf.s FR_SNORM_LIMIT= GR_SNORM_LIMIT      // Set smallest normal limit
>> - (p8)  fcvt.fx.s1 FR_int_x = f8                   // Convert x to integer
>> -       nop.i 0
>> -}
>> -{.mfi
>> -       setf.s FR_UF_LIMIT= GR_UF_LIMIT            // Set underflow limit
>> -       fma.s1 FR_KF0= f8, FR_LOG2_10, FR_ROUNDVAL // y= (x*log2(10)*2^10 +
>> -                                                  //    1.5*2^63) * 2^(-63)
>> -       mov GR_EXP_CORR= 0xffff-126
>> -}
>> -;;
>> -
>> -{.mfi
>> -       ldfe FR_COEFF1= [ GR_COEFF_START ], 16    // load C_1
>> -       fms.s1 FR_KF= FR_KF0, f1, FR_ROUNDVAL     // (K+f)*2^(10-63)
>> -       mov GR_MASK= 1023
>> -}
>> -;;
>> -
>> -{.mfi
>> -       ldfe FR_COEFF2= [ GR_COEFF_START ], 16    // load C_2
>> -       nop.f 0
>> -       mov GR_MASK_low= 31
>> -}
>> -;;
>> -
>> -{.mlx
>> -       getf.sig GR_KF0= FR_KF0                   // (K+f)*2^10= round_to_int(y)
>> - (p8)  movl GR_exact_limit= 0x41200000           // Largest x for exact result,
>> -                                                 //  +10.0
>> -}
>> -;;
>> -
>> -{.mfi
>> -       add GR_LOG_TBL= 256, GR_COEFF_START       // Pointer to high T_table
>> -       fcmp.gt.s1 p12, p7= f8, FR_OF_LIMIT       // x>overflow threshold ?
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mfi
>> - (p8)  setf.s FR_exact_limit = GR_exact_limit    // Largest x for exact result
>> - (p8)  fcvt.xf FR_int_x = FR_int_x               // Integral part of x
>> -       shr GR_K= GR_KF0, 10                      // K
>> -}
>> -{.mfi
>> -       and GR_F_high= GR_MASK, GR_KF0            // f_high*32
>> -       fms.s1 FR_R= f8, FR_LOG2_10, FR_KF        // r*2^(-53)= [ x*log2(10)-
>> -                                                 //           (K+f) ] *2^{10-63}
>> -       and GR_F_low= GR_KF0, GR_MASK_low         // f_low
>> -}
>> -;;
>> -
>> -{.mmi
>> -       shladd GR_Flow_ADDR= GR_F_low, 3, GR_COEFF_START // address of 2^{f_low}
>> -       add GR_BIAS= GR_K, GR_EXP_CORR            // K= bias-2*63
>> -       shr GR_Fh= GR_F_high, 5                   // f_high
>> -}
>> -;;
>> -
>> -{.mfi
>> -       setf.exp FR_2_TO_K= GR_BIAS               // 2^{K-126}
>> - (p7)  fcmp.lt.s1 p12, p7= f8, FR_UF_LIMIT       // x<underflow threshold ?
>> -       shladd GR_Fh_ADDR= GR_Fh, 3, GR_LOG_TBL   // address of 2^{f_high}
>> -}
>> -{.mfi
>> -       ldf8 FR_T_low= [ GR_Flow_ADDR ]           // load T_low= 2^{f_low}
>> -       nop.f 0
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mfb
>> -       ldf8 FR_T_high= [ GR_Fh_ADDR ]            // load T_high= 2^{f_high}
>> -       fcmp.ge.s1 p11, p0= f8, FR_SNORM_LIMIT    // Test x for normal range
>> - (p12) br.cond.spnt OUT_RANGE_exp10
>> -}
>> -;;
>> -
>> -{.mfi
>> -       nop.m 0
>> -       fma.s1 FR_P12= FR_COEFF2, FR_R, FR_COEFF1 // P12= C_1+C_2*r
>> -       cmp.eq p7,p9= r0,r0                       // Assume inexact result
>> -}
>> -;;
>> -
>> -{.mfi
>> -       nop.m 0
>> - (p8)  fcmp.eq.s1 p9,p7= FR_int_x, f8            // Test x positive integer
>> -       nop.i 0
>> -}
>> -{.mfi
>> -       nop.m 0
>> -       fma.s1 FR_T_low_K= FR_T_low, FR_2_TO_K, f0 // T= 2^{K-126}*T_low
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mfi
>> -       nop.m 0
>> -       fma.s1 FR_P= FR_P12, FR_R, f0              // P= P12*r
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -// If x a positive integer, will it produce an exact result?
>> -//   p7 result will be inexact
>> -//   p9 result will be exact
>> -{.mfi
>> -       nop.m 0
>> - (p9)  fcmp.le.s1 p9,p7= f8, FR_exact_limit       // Test x gives exact result
>> -       nop.i 0
>> -}
>> -{.mfi
>> -       nop.m 0
>> -       fma.s1 FR_T= FR_T_low_K, FR_T_high, f0     // T= T*T_high
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -.pred.rel "mutex",p7,p9
>> -{.mfi
>> -       nop.m 0
>> - (p7)  fma.s.s0 f8= FR_P, FR_T, FR_T              // result= T+T*P, inexact set
>> -       nop.i 0
>> -}
>> -{.mfb
>> -       nop.m 0
>> - (p9)  fma.s.s1 f8= FR_P, FR_T, FR_T              // result= T+T*P, exact use s1
>> - (p11) br.ret.sptk b0                             // return, if result normal
>> -}
>> -;;
>> -
>> -// Here if result in denormal range (and not zero)
>> -{.mib
>> -       nop.m 0
>> -       mov GR_Parameter_TAG= 266
>> -       br.cond.sptk __libm_error_region           // Branch to error handling
>> -}
>> -;;
>> -
>> -SPECIAL_exp10:
>> -{.mfi
>> -       nop.m 0
>> -       fclass.m p6, p0= f8, 0x22                  // x= -Infinity ?
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mfi
>> -       nop.m 0
>> -       fclass.m p7, p0= f8, 0x21                  // x= +Infinity ?
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mfi
>> -       nop.m 0
>> -       fclass.m p8, p0= f8, 0x7                   // x= +/-Zero ?
>> -       nop.i 0
>> -}
>> -{.mfb
>> -       nop.m 0
>> - (p6)  mov f8= f0                                 // exp10(-Infinity)= 0
>> - (p6)  br.ret.spnt b0
>> -}
>> -;;
>> -
>> -{.mfb
>> -       nop.m 0
>> -       nop.f 0
>> - (p7)  br.ret.spnt b0                             // exp10(+Infinity)= +Infinity
>> -}
>> -;;
>> -
>> -{.mfb
>> -       nop.m 0
>> - (p8)  mov f8= f1                                 // exp10(+/-0)= 1
>> - (p8)  br.ret.spnt b0
>> -}
>> -;;
>> -
>> -{.mfb
>> -       nop.m 0
>> -       fma.s.s0 f8= f8, f1, f0                    // Remaining cases: NaNs
>> -       br.ret.sptk b0
>> -}
>> -;;
>> -
>> -
>> -OUT_RANGE_exp10:
>> -
>> -// underflow: p6= 1
>> -// overflow: p8= 1
>> -
>> -.pred.rel "mutex",p6,p8
>> -{.mmi
>> - (p8)  mov GR_EXPMAX= 0x1fffe
>> - (p6)  mov GR_EXPMAX= 1
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mii
>> -       setf.exp FR_R= GR_EXPMAX
>> - (p8)  mov GR_Parameter_TAG= 167
>> - (p6)  mov GR_Parameter_TAG= 266
>> -}
>> -;;
>> -
>> -{.mfb
>> -       nop.m 0
>> -       fma.s.s0 f8= FR_R, FR_R, f0                // Create overflow/underflow
>> -       br.cond.sptk __libm_error_region           // Branch to error handling
>> -}
>> -;;
>> -
>> -GLOBAL_IEEE754_END(__exp10f)
>> -libm_alias_float_other (__exp10, exp10)
>> -#if SHLIB_COMPAT (libm, GLIBC_2_1, GLIBC_2_27)
>> -strong_alias (__exp10f, __pow10f_compat)
>> -compat_symbol (libm, __pow10f_compat, pow10f, GLIBC_2_2)
>> -#endif
>> -#if SHLIB_COMPAT (libm, GLIBC_2_1, GLIBC_2_32)
>> -strong_alias (__exp10f, __exp10f_compat)
>> -compat_symbol (libm, __exp10f_compat, exp10f, GLIBC_2_1)
>> -#endif
>> -versioned_symbol (libm, __exp10f, exp10f, GLIBC_2_32)
>> -
>> -
>> -LOCAL_LIBM_ENTRY(__libm_error_region)
>> -
>> -.prologue
>> -{.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
>> -}
>> -;;
>> -
>> -{.mmi
>> -       stfs [ GR_Parameter_Y ]= FR_Y, 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
>> -{.mib
>> -       stfs [ GR_Parameter_X ]= FR_X              // STORE Parameter 1 on stack
>> -       add GR_Parameter_RESULT= 0, GR_Parameter_Y // Parameter 3 address
>> -       nop.b 0
>> -}
>> -{.mib
>> -       stfs [ GR_Parameter_Y ]= FR_RESULT         // 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
>> -       add GR_Parameter_RESULT= 48, sp
>> -       nop.m 0
>> -       nop.i 0
>> -}
>> -;;
>> -
>> -{.mmi
>> -       ldfs 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
>> -}
>> -;;
>> -
>> -
>> -LOCAL_LIBM_END(__libm_error_region)
>> -
>> -.type __libm_error_support#, @function
>> -.global __libm_error_support#
>> diff --git a/sysdeps/ia64/fpu/e_exp2f_data.c b/sysdeps/ia64/fpu/e_exp2f_data.c
>> deleted file mode 100644
>> index 1cc8931700..0000000000
>> --- a/sysdeps/ia64/fpu/e_exp2f_data.c
>> +++ /dev/null
>> @@ -1 +0,0 @@
>> -/* Not needed.  */
>> diff --git a/sysdeps/ia64/fpu/libm-test-ulps b/sysdeps/ia64/fpu/libm-test-ulps
>> index d0cab46c3b..249dd9e7e5 100644
>> --- a/sysdeps/ia64/fpu/libm-test-ulps
>> +++ b/sysdeps/ia64/fpu/libm-test-ulps
>> @@ -1057,12 +1057,15 @@ float: 1
>>  float128: 2
>>  
>>  Function: "exp10_downward":
>> +float: 1
>>  float128: 3
>>  
>>  Function: "exp10_towardzero":
>> +float: 1
>>  float128: 3
>>  
>>  Function: "exp10_upward":
>> +float: 1
>>  float128: 3
>>  
>>  Function: "exp2":
>> diff --git a/sysdeps/ia64/fpu/math_errf.c b/sysdeps/ia64/fpu/math_errf.c
>> deleted file mode 100644
>> index 1cc8931700..0000000000
>> --- a/sysdeps/ia64/fpu/math_errf.c
>> +++ /dev/null
>> @@ -1 +0,0 @@
>> -/* Not needed.  */
>> diff --git a/sysdeps/ia64/fpu/w_exp10f.c b/sysdeps/ia64/fpu/w_exp10f.c
>> deleted file mode 100644
>> index 1cc8931700..0000000000
>> --- a/sysdeps/ia64/fpu/w_exp10f.c
>> +++ /dev/null
>> @@ -1 +0,0 @@
>> -/* Not needed.  */
>> diff --git a/sysdeps/ia64/fpu/w_exp10f_compat.c b/sysdeps/ia64/fpu/w_exp10f_compat.c
>> deleted file mode 100644
>> index 41254ae60a..0000000000
>> --- a/sysdeps/ia64/fpu/w_exp10f_compat.c
>> +++ /dev/null
>> @@ -1 +0,0 @@
>> -/* Not needed. */
>>
diff mbox series

Patch

diff --git a/sysdeps/ia64/fpu/e_exp10f.S b/sysdeps/ia64/fpu/e_exp10f.S
deleted file mode 100644
index e82394b725..0000000000
--- a/sysdeps/ia64/fpu/e_exp10f.S
+++ /dev/null
@@ -1,568 +0,0 @@ 
-.file "exp10f.s"
-
-
-// Copyright (c) 2000 - 2005, Intel Corporation
-// All rights reserved.
-//
-// Contributed 2000 by the Intel Numerics Group, Intel Corporation
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-// * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//
-// * Redistributions in binary form must reproduce the above copyright
-// notice, this list of conditions and the following disclaimer in the
-// documentation and/or other materials provided with the distribution.
-//
-// * The name of Intel Corporation may not be used to endorse or promote
-// products derived from this software without specific prior written
-// permission.
-
-// 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://www.intel.com/software/products/opensource/libraries/num.htm.
-//
-// History
-//==============================================================
-// 08/25/00 Initial version
-// 05/20/02 Cleaned up namespace and sf0 syntax
-// 09/06/02 Improved performance and accuracy; no inexact flags on exact cases
-// 01/29/03 Added missing } to bundle templates
-// 12/16/04 Call error handling on underflow.
-// 03/31/05 Reformatted delimiters between data tables
-//
-// API
-//==============================================================
-// float exp10f(float)
-//
-// Overview of operation
-//==============================================================
-// Background
-//
-// Implementation
-//
-// Let x= (K + fh + fl + r)/log2(10), where
-// K is an integer, fh= 0.b1 b2 b3 b4 b5,
-// fl= 2^{-5}* 0.b6 b7 b8 b8 b10 (fh, fl >= 0),
-// and |r|<2^{-11}
-// Th is a table that stores 2^fh (32 entries) rounded to
-// double extended precision (only mantissa is stored)
-// Tl is a table that stores 2^fl (32 entries) rounded to
-// double extended precision (only mantissa is stored)
-//
-// 10^x is approximated as
-// 2^K * Th [ f ] * Tl [ f ] * (1+c1*r+c2*r^2)
-
-// Note there are only 10 non-zero values that produce an exact result:
-//  1.0, 2.0, ... 10.0.
-// We test for these cases and use s1 to avoid setting the inexact flag.
-
-// Special values
-//==============================================================
-// exp10(0)= 1
-// exp10(+inf)= inf
-// exp10(-inf)= 0
-//
-
-// Registers used
-//==============================================================
-// r2-r3, r14-r40
-// f6-f15, f32-f52
-// p6-p12
-//
-
-#include <shlib-compat.h>
-
-
-GR_TBL_START        = r2
-GR_LOG_TBL          = r3
-
-GR_OF_LIMIT         = r14
-GR_UF_LIMIT         = r15
-GR_EXP_CORR         = r16
-GR_F_low            = r17
-GR_F_high           = r18
-GR_K                = r19
-GR_Flow_ADDR        = r20
-
-GR_BIAS             = r21
-GR_Fh               = r22
-GR_Fh_ADDR          = r23
-GR_EXPMAX           = r24
-
-GR_ROUNDVAL         = r26
-GR_SNORM_LIMIT      = r26
-GR_MASK             = r27
-GR_KF0              = r28
-GR_MASK_low         = r29
-GR_COEFF_START      = r30
-GR_exact_limit      = r31
-
-GR_SAVE_B0          = r33
-GR_SAVE_PFS         = r34
-GR_SAVE_GP          = r35
-GR_SAVE_SP          = r36
-
-GR_Parameter_X      = r37
-GR_Parameter_Y      = r38
-GR_Parameter_RESULT = r39
-GR_Parameter_TAG    = r40
-
-
-FR_X                = f10
-FR_Y                = f1
-FR_RESULT           = f8
-
-
-FR_COEFF1           = f6
-FR_COEFF2           = f7
-FR_R                = f9
-FR_LOG2_10          = f10
-
-FR_2P53             = f11
-FR_KF0              = f12
-FR_COEFF3           = f13
-FR_COEFF4           = f14
-FR_UF_LIMIT         = f15
-
-FR_OF_LIMIT         = f32
-FR_DX_L210          = f33
-FR_ROUNDVAL         = f34
-FR_KF               = f35
-
-FR_2_TO_K           = f36
-FR_T_low            = f37
-FR_T_high           = f38
-
-FR_P12              = f41
-FR_T_low_K          = f42
-FR_T                = f44
-FR_P                = f45
-
-FR_E                = f49
-FR_exact_limit      = f50
-
-FR_int_x            = f51
-FR_SNORM_LIMIT      = f52
-
-
-// Data tables
-//==============================================================
-
-RODATA
-
-.align 16
-
-LOCAL_OBJECT_START(poly_coeffs)
-
-data8 0xd49a784bcd1b8afe, 0x00003fcb // log2(10)*2^(10-63)
-data8 0xb17217f7d1cf79ab, 0x00004033 // C_1 * 2^53
-data8 0xf5fdeffc162c7541, 0x00004066 // C_2 * 2^106
-LOCAL_OBJECT_END(poly_coeffs)
-
-
-LOCAL_OBJECT_START(T_table)
-
-// 2^{0.00000 b6 b7 b8 b9 b10}
-data8 0x8000000000000000, 0x8016302f17467628
-data8 0x802c6436d0e04f50, 0x80429c17d77c18ed
-data8 0x8058d7d2d5e5f6b0, 0x806f17687707a7af
-data8 0x80855ad965e88b83, 0x809ba2264dada76a
-data8 0x80b1ed4fd999ab6c, 0x80c83c56b50cf77f
-data8 0x80de8f3b8b85a0af, 0x80f4e5ff089f763e
-data8 0x810b40a1d81406d4, 0x81219f24a5baa59d
-data8 0x813801881d886f7b, 0x814e67cceb90502c
-data8 0x8164d1f3bc030773, 0x817b3ffd3b2f2e47
-data8 0x8191b1ea15813bfd, 0x81a827baf7838b78
-data8 0x81bea1708dde6055, 0x81d51f0b8557ec1c
-data8 0x81eba08c8ad4536f, 0x820225f44b55b33b
-data8 0x8218af4373fc25eb, 0x822f3c7ab205c89a
-data8 0x8245cd9ab2cec048, 0x825c62a423d13f0c
-data8 0x8272fb97b2a5894c, 0x828998760d01faf3
-data8 0x82a0393fe0bb0ca8, 0x82b6ddf5dbc35906
-//
-// 2^{0.b1 b2 b3 b4 b5}
-data8 0x8000000000000000, 0x82cd8698ac2ba1d7
-data8 0x85aac367cc487b14, 0x88980e8092da8527
-data8 0x8b95c1e3ea8bd6e6, 0x8ea4398b45cd53c0
-data8 0x91c3d373ab11c336, 0x94f4efa8fef70961
-data8 0x9837f0518db8a96f, 0x9b8d39b9d54e5538
-data8 0x9ef5326091a111ad, 0xa27043030c496818
-data8 0xa5fed6a9b15138ea, 0xa9a15ab4ea7c0ef8
-data8 0xad583eea42a14ac6, 0xb123f581d2ac258f
-data8 0xb504f333f9de6484, 0xb8fbaf4762fb9ee9
-data8 0xbd08a39f580c36be, 0xc12c4cca66709456
-data8 0xc5672a115506dadd, 0xc9b9bd866e2f27a2
-data8 0xce248c151f8480e3, 0xd2a81d91f12ae45a
-data8 0xd744fccad69d6af4, 0xdbfbb797daf23755
-data8 0xe0ccdeec2a94e111, 0xe5b906e77c8348a8
-data8 0xeac0c6e7dd24392e, 0xefe4b99bdcdaf5cb
-data8 0xf5257d152486cc2c, 0xfa83b2db722a033a
-LOCAL_OBJECT_END(T_table)
-
-
-
-.section .text
-GLOBAL_IEEE754_ENTRY(__exp10f)
-
-
-{.mfi
-       alloc r32= ar.pfs, 1, 4, 4, 0
-       // will continue only for non-zero normal/denormal numbers
-       fclass.nm.unc p12, p7= f8, 0x1b
-       nop.i 0
-}
-{.mlx
-       // GR_TBL_START= pointer to log2(10), C_1...C_4 followed by T_table
-       addl GR_TBL_START= @ltoff(poly_coeffs), gp
-       movl GR_ROUNDVAL= 0x3fc00000             // 1.5 (SP)
-}
-;;
-
-{.mfi
-       ld8 GR_COEFF_START= [ GR_TBL_START ]     // Load pointer to coeff table
-       fcmp.lt.s1 p6, p8= f8, f0                // X<0 ?
-       nop.i 0
-}
-;;
-
-{.mlx
-       nop.m 0
-       movl GR_UF_LIMIT= 0xc2349e35             // (-2^7-22) / log2(10)
-}
-{.mlx
-       setf.s FR_ROUNDVAL= GR_ROUNDVAL
-       movl GR_OF_LIMIT= 0x421a209a             // Overflow threshold
-}
-;;
-
-{.mlx
-       ldfe FR_LOG2_10= [ GR_COEFF_START ], 16  // load log2(10)*2^(10-63)
-       movl GR_SNORM_LIMIT= 0xc217b818          // Smallest normal threshold
-}
-{.mib
-       nop.m 0
-       nop.i 0
- (p12) br.cond.spnt SPECIAL_exp10               // Branch if nan, inf, zero
-}
-;;
-
-{.mfi
-       setf.s FR_OF_LIMIT= GR_OF_LIMIT           // Set overflow limit
-       fma.s0 f8= f8, f1, f0                     // normalize x
-       nop.i 0
-}
-;;
-
-{.mfi
-       setf.s FR_SNORM_LIMIT= GR_SNORM_LIMIT      // Set smallest normal limit
- (p8)  fcvt.fx.s1 FR_int_x = f8                   // Convert x to integer
-       nop.i 0
-}
-{.mfi
-       setf.s FR_UF_LIMIT= GR_UF_LIMIT            // Set underflow limit
-       fma.s1 FR_KF0= f8, FR_LOG2_10, FR_ROUNDVAL // y= (x*log2(10)*2^10 +
-                                                  //    1.5*2^63) * 2^(-63)
-       mov GR_EXP_CORR= 0xffff-126
-}
-;;
-
-{.mfi
-       ldfe FR_COEFF1= [ GR_COEFF_START ], 16    // load C_1
-       fms.s1 FR_KF= FR_KF0, f1, FR_ROUNDVAL     // (K+f)*2^(10-63)
-       mov GR_MASK= 1023
-}
-;;
-
-{.mfi
-       ldfe FR_COEFF2= [ GR_COEFF_START ], 16    // load C_2
-       nop.f 0
-       mov GR_MASK_low= 31
-}
-;;
-
-{.mlx
-       getf.sig GR_KF0= FR_KF0                   // (K+f)*2^10= round_to_int(y)
- (p8)  movl GR_exact_limit= 0x41200000           // Largest x for exact result,
-                                                 //  +10.0
-}
-;;
-
-{.mfi
-       add GR_LOG_TBL= 256, GR_COEFF_START       // Pointer to high T_table
-       fcmp.gt.s1 p12, p7= f8, FR_OF_LIMIT       // x>overflow threshold ?
-       nop.i 0
-}
-;;
-
-{.mfi
- (p8)  setf.s FR_exact_limit = GR_exact_limit    // Largest x for exact result
- (p8)  fcvt.xf FR_int_x = FR_int_x               // Integral part of x
-       shr GR_K= GR_KF0, 10                      // K
-}
-{.mfi
-       and GR_F_high= GR_MASK, GR_KF0            // f_high*32
-       fms.s1 FR_R= f8, FR_LOG2_10, FR_KF        // r*2^(-53)= [ x*log2(10)-
-                                                 //           (K+f) ] *2^{10-63}
-       and GR_F_low= GR_KF0, GR_MASK_low         // f_low
-}
-;;
-
-{.mmi
-       shladd GR_Flow_ADDR= GR_F_low, 3, GR_COEFF_START // address of 2^{f_low}
-       add GR_BIAS= GR_K, GR_EXP_CORR            // K= bias-2*63
-       shr GR_Fh= GR_F_high, 5                   // f_high
-}
-;;
-
-{.mfi
-       setf.exp FR_2_TO_K= GR_BIAS               // 2^{K-126}
- (p7)  fcmp.lt.s1 p12, p7= f8, FR_UF_LIMIT       // x<underflow threshold ?
-       shladd GR_Fh_ADDR= GR_Fh, 3, GR_LOG_TBL   // address of 2^{f_high}
-}
-{.mfi
-       ldf8 FR_T_low= [ GR_Flow_ADDR ]           // load T_low= 2^{f_low}
-       nop.f 0
-       nop.i 0
-}
-;;
-
-{.mfb
-       ldf8 FR_T_high= [ GR_Fh_ADDR ]            // load T_high= 2^{f_high}
-       fcmp.ge.s1 p11, p0= f8, FR_SNORM_LIMIT    // Test x for normal range
- (p12) br.cond.spnt OUT_RANGE_exp10
-}
-;;
-
-{.mfi
-       nop.m 0
-       fma.s1 FR_P12= FR_COEFF2, FR_R, FR_COEFF1 // P12= C_1+C_2*r
-       cmp.eq p7,p9= r0,r0                       // Assume inexact result
-}
-;;
-
-{.mfi
-       nop.m 0
- (p8)  fcmp.eq.s1 p9,p7= FR_int_x, f8            // Test x positive integer
-       nop.i 0
-}
-{.mfi
-       nop.m 0
-       fma.s1 FR_T_low_K= FR_T_low, FR_2_TO_K, f0 // T= 2^{K-126}*T_low
-       nop.i 0
-}
-;;
-
-{.mfi
-       nop.m 0
-       fma.s1 FR_P= FR_P12, FR_R, f0              // P= P12*r
-       nop.i 0
-}
-;;
-
-// If x a positive integer, will it produce an exact result?
-//   p7 result will be inexact
-//   p9 result will be exact
-{.mfi
-       nop.m 0
- (p9)  fcmp.le.s1 p9,p7= f8, FR_exact_limit       // Test x gives exact result
-       nop.i 0
-}
-{.mfi
-       nop.m 0
-       fma.s1 FR_T= FR_T_low_K, FR_T_high, f0     // T= T*T_high
-       nop.i 0
-}
-;;
-
-.pred.rel "mutex",p7,p9
-{.mfi
-       nop.m 0
- (p7)  fma.s.s0 f8= FR_P, FR_T, FR_T              // result= T+T*P, inexact set
-       nop.i 0
-}
-{.mfb
-       nop.m 0
- (p9)  fma.s.s1 f8= FR_P, FR_T, FR_T              // result= T+T*P, exact use s1
- (p11) br.ret.sptk b0                             // return, if result normal
-}
-;;
-
-// Here if result in denormal range (and not zero)
-{.mib
-       nop.m 0
-       mov GR_Parameter_TAG= 266
-       br.cond.sptk __libm_error_region           // Branch to error handling
-}
-;;
-
-SPECIAL_exp10:
-{.mfi
-       nop.m 0
-       fclass.m p6, p0= f8, 0x22                  // x= -Infinity ?
-       nop.i 0
-}
-;;
-
-{.mfi
-       nop.m 0
-       fclass.m p7, p0= f8, 0x21                  // x= +Infinity ?
-       nop.i 0
-}
-;;
-
-{.mfi
-       nop.m 0
-       fclass.m p8, p0= f8, 0x7                   // x= +/-Zero ?
-       nop.i 0
-}
-{.mfb
-       nop.m 0
- (p6)  mov f8= f0                                 // exp10(-Infinity)= 0
- (p6)  br.ret.spnt b0
-}
-;;
-
-{.mfb
-       nop.m 0
-       nop.f 0
- (p7)  br.ret.spnt b0                             // exp10(+Infinity)= +Infinity
-}
-;;
-
-{.mfb
-       nop.m 0
- (p8)  mov f8= f1                                 // exp10(+/-0)= 1
- (p8)  br.ret.spnt b0
-}
-;;
-
-{.mfb
-       nop.m 0
-       fma.s.s0 f8= f8, f1, f0                    // Remaining cases: NaNs
-       br.ret.sptk b0
-}
-;;
-
-
-OUT_RANGE_exp10:
-
-// underflow: p6= 1
-// overflow: p8= 1
-
-.pred.rel "mutex",p6,p8
-{.mmi
- (p8)  mov GR_EXPMAX= 0x1fffe
- (p6)  mov GR_EXPMAX= 1
-       nop.i 0
-}
-;;
-
-{.mii
-       setf.exp FR_R= GR_EXPMAX
- (p8)  mov GR_Parameter_TAG= 167
- (p6)  mov GR_Parameter_TAG= 266
-}
-;;
-
-{.mfb
-       nop.m 0
-       fma.s.s0 f8= FR_R, FR_R, f0                // Create overflow/underflow
-       br.cond.sptk __libm_error_region           // Branch to error handling
-}
-;;
-
-GLOBAL_IEEE754_END(__exp10f)
-libm_alias_float_other (__exp10, exp10)
-#if SHLIB_COMPAT (libm, GLIBC_2_1, GLIBC_2_27)
-strong_alias (__exp10f, __pow10f_compat)
-compat_symbol (libm, __pow10f_compat, pow10f, GLIBC_2_2)
-#endif
-#if SHLIB_COMPAT (libm, GLIBC_2_1, GLIBC_2_32)
-strong_alias (__exp10f, __exp10f_compat)
-compat_symbol (libm, __exp10f_compat, exp10f, GLIBC_2_1)
-#endif
-versioned_symbol (libm, __exp10f, exp10f, GLIBC_2_32)
-
-
-LOCAL_LIBM_ENTRY(__libm_error_region)
-
-.prologue
-{.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
-}
-;;
-
-{.mmi
-       stfs [ GR_Parameter_Y ]= FR_Y, 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
-{.mib
-       stfs [ GR_Parameter_X ]= FR_X              // STORE Parameter 1 on stack
-       add GR_Parameter_RESULT= 0, GR_Parameter_Y // Parameter 3 address
-       nop.b 0
-}
-{.mib
-       stfs [ GR_Parameter_Y ]= FR_RESULT         // 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
-       add GR_Parameter_RESULT= 48, sp
-       nop.m 0
-       nop.i 0
-}
-;;
-
-{.mmi
-       ldfs 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
-}
-;;
-
-
-LOCAL_LIBM_END(__libm_error_region)
-
-.type __libm_error_support#, @function
-.global __libm_error_support#
diff --git a/sysdeps/ia64/fpu/e_exp2f_data.c b/sysdeps/ia64/fpu/e_exp2f_data.c
deleted file mode 100644
index 1cc8931700..0000000000
--- a/sysdeps/ia64/fpu/e_exp2f_data.c
+++ /dev/null
@@ -1 +0,0 @@ 
-/* Not needed.  */
diff --git a/sysdeps/ia64/fpu/libm-test-ulps b/sysdeps/ia64/fpu/libm-test-ulps
index d0cab46c3b..249dd9e7e5 100644
--- a/sysdeps/ia64/fpu/libm-test-ulps
+++ b/sysdeps/ia64/fpu/libm-test-ulps
@@ -1057,12 +1057,15 @@  float: 1
 float128: 2
 
 Function: "exp10_downward":
+float: 1
 float128: 3
 
 Function: "exp10_towardzero":
+float: 1
 float128: 3
 
 Function: "exp10_upward":
+float: 1
 float128: 3
 
 Function: "exp2":
diff --git a/sysdeps/ia64/fpu/math_errf.c b/sysdeps/ia64/fpu/math_errf.c
deleted file mode 100644
index 1cc8931700..0000000000
--- a/sysdeps/ia64/fpu/math_errf.c
+++ /dev/null
@@ -1 +0,0 @@ 
-/* Not needed.  */
diff --git a/sysdeps/ia64/fpu/w_exp10f.c b/sysdeps/ia64/fpu/w_exp10f.c
deleted file mode 100644
index 1cc8931700..0000000000
--- a/sysdeps/ia64/fpu/w_exp10f.c
+++ /dev/null
@@ -1 +0,0 @@ 
-/* Not needed.  */
diff --git a/sysdeps/ia64/fpu/w_exp10f_compat.c b/sysdeps/ia64/fpu/w_exp10f_compat.c
deleted file mode 100644
index 41254ae60a..0000000000
--- a/sysdeps/ia64/fpu/w_exp10f_compat.c
+++ /dev/null
@@ -1 +0,0 @@ 
-/* Not needed. */