mirror of
git://sourceware.org/git/glibc.git
synced 2024-12-15 04:20:28 +08:00
aa1142c593
Continuing the preparation for additional _FloatN / _FloatNx function aliases, this patch makes ia64 libm function implementations use libm_alias_float to define function aliases. The same approach is followed as with the corresponding long double and double patches: the ia64-specific macros are left unchanged, with calls to libm_alias_float_other being added in most cases and libm_alias_float itself being used in only a few places. Tested with build-many-glibcs.py for ia64-linux-gnu that installed stripped shared libraries are unchanged by the patch. * sysdeps/ia64/fpu/libm-symbols.h: Include <libm-alias-float.h>. * sysdeps/ia64/fpu/e_acosf.S (acosf): Use libm_alias_float_other. * sysdeps/ia64/fpu/e_acoshf.S (acoshf): Likewise. * sysdeps/ia64/fpu/e_asinf.S (asinf): Likewise. * sysdeps/ia64/fpu/e_atan2f.S (atan2f): Likewise. * sysdeps/ia64/fpu/e_atanhf.S (atanhf): Likewise. * sysdeps/ia64/fpu/e_coshf.S (coshf): Likewise. * sysdeps/ia64/fpu/e_exp10f.S (exp10f): Likewise. * sysdeps/ia64/fpu/e_exp2f.S (exp2f): Likewise. * sysdeps/ia64/fpu/e_expf.S (expf): Likewise. * sysdeps/ia64/fpu/e_fmodf.S (fmodf): Likewise. * sysdeps/ia64/fpu/e_hypotf.S (hypotf): Likewise. * sysdeps/ia64/fpu/e_lgammaf_r.c (lgammaf_r): Define using libm_alias_float_r. * sysdeps/ia64/fpu/e_log2f.S (log2f): Use libm_alias_float_other. * sysdeps/ia64/fpu/e_logf.S (log10f): Likewise. (logf): Likewise. * sysdeps/ia64/fpu/e_powf.S (powf): Likewise. * sysdeps/ia64/fpu/e_remainderf.S (remainderf): Likewise. * sysdeps/ia64/fpu/e_sinhf.S (sinhf): Likewise. * sysdeps/ia64/fpu/e_sqrtf.S (sqrtf): Likewise. * sysdeps/ia64/fpu/libm_sincosf.S (sincosf): Likewise. * sysdeps/ia64/fpu/s_asinhf.S (asinhf): Likewise. * sysdeps/ia64/fpu/s_atanf.S (atanf): Likewise. * sysdeps/ia64/fpu/s_cbrtf.S (cbrtf): Likewise. * sysdeps/ia64/fpu/s_ceilf.S (ceilf): Likewise. * sysdeps/ia64/fpu/s_copysign.S (copysignf): Define using libm_alias_float. * sysdeps/ia64/fpu/s_cosf.S (sinf): Use libm_alias_float_other. (cosf): Likewise. * sysdeps/ia64/fpu/s_erfcf.S (erfcf): Likewise. * sysdeps/ia64/fpu/s_erff.S (erff): Likewise. * sysdeps/ia64/fpu/s_expm1f.S (expm1f): Likewise. * sysdeps/ia64/fpu/s_fabsf.S (fabsf): Likewise. * sysdeps/ia64/fpu/s_fdimf.S (fdimf): Likewise. * sysdeps/ia64/fpu/s_floorf.S (floorf): Likewise. * sysdeps/ia64/fpu/s_fmaf.S (fmaf): Likewise. * sysdeps/ia64/fpu/s_fmaxf.S (fmaxf): Likewise. * sysdeps/ia64/fpu/s_frexpf.c (frexpf): Likewise. * sysdeps/ia64/fpu/s_ldexpf.c (ldexpf): Likewise. * sysdeps/ia64/fpu/s_log1pf.S (log1pf): Likewise. * sysdeps/ia64/fpu/s_logbf.S (logbf): Likewise. * sysdeps/ia64/fpu/s_modff.S (modff): Likewise. * sysdeps/ia64/fpu/s_nearbyintf.S (nearbyintf): Likewise. * sysdeps/ia64/fpu/s_nextafterf.S (nextafterf): Likewise. * sysdeps/ia64/fpu/s_rintf.S (rintf): Likewise. * sysdeps/ia64/fpu/s_roundf.S (roundf): Likewise. * sysdeps/ia64/fpu/s_scalblnf.c (scalblnf): Likewise. * sysdeps/ia64/fpu/s_scalbnf.c (scalbnf): Define using libm_alias_float. * sysdeps/ia64/fpu/s_tanf.S (tanf): Use libm_alias_float_other. * sysdeps/ia64/fpu/s_tanhf.S (tanhf): Likewise. * sysdeps/ia64/fpu/s_truncf.S (truncf): Likewise. * sysdeps/ia64/fpu/w_lgammaf_main.c [BUILD_LGAMMA && !USE_AS_COMPAT] (lgammaf): Likewise. * sysdeps/ia64/fpu/w_tgammaf_compat.S (tgammaf): Likewise.
235 lines
7.0 KiB
ArmAsm
235 lines
7.0 KiB
ArmAsm
.file "roundf.s"
|
|
|
|
|
|
// Copyright (c) 2000 - 2003, 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
|
|
//==============================================================
|
|
// 10/25/00 Initial version
|
|
// 06/14/01 Changed cmp to an equivalent form
|
|
// 05/20/02 Cleaned up namespace and sf0 syntax
|
|
// 01/20/03 Improved performance and reduced code size
|
|
// 04/18/03 Eliminate possible WAW dependency warning
|
|
// 09/03/03 Improved performance
|
|
//==============================================================
|
|
|
|
// API
|
|
//==============================================================
|
|
// float roundf(float x)
|
|
//==============================================================
|
|
|
|
// general input registers:
|
|
// r14 - r18
|
|
|
|
rSignexp = r14
|
|
rExp = r15
|
|
rExpMask = r16
|
|
rBigexp = r17
|
|
rExpHalf = r18
|
|
|
|
// floating-point registers:
|
|
// f8 - f13
|
|
|
|
fXtruncInt = f9
|
|
fNormX = f10
|
|
fHalf = f11
|
|
fInc = f12
|
|
fRem = f13
|
|
|
|
// predicate registers used:
|
|
// p6 - p10
|
|
|
|
// Overview of operation
|
|
//==============================================================
|
|
// float roundf(float x)
|
|
// Return an integer value (represented as a float) that is x
|
|
// rounded to nearest integer, halfway cases rounded away from
|
|
// zero.
|
|
// if x>0 result = trunc(x+0.5)
|
|
// if x<0 result = trunc(x-0.5)
|
|
//
|
|
//==============================================================
|
|
|
|
// double_extended
|
|
// if the exponent is > 1003e => 3F(true) = 63(decimal)
|
|
// we have a significand of 64 bits 1.63-bits.
|
|
// If we multiply by 2^63, we no longer have a fractional part
|
|
// So input is an integer value already.
|
|
|
|
// double
|
|
// if the exponent is >= 10033 => 34(true) = 52(decimal)
|
|
// 34 + 3ff = 433
|
|
// we have a significand of 53 bits 1.52-bits. (implicit 1)
|
|
// If we multiply by 2^52, we no longer have a fractional part
|
|
// So input is an integer value already.
|
|
|
|
// single
|
|
// if the exponent is > 10016 => 17(true) = 23(decimal)
|
|
// we have a significand of 24 bits 1.23-bits. (implicit 1)
|
|
// If we multiply by 2^23, we no longer have a fractional part
|
|
// So input is an integer value already.
|
|
|
|
|
|
.section .text
|
|
GLOBAL_LIBM_ENTRY(roundf)
|
|
|
|
{ .mfi
|
|
getf.exp rSignexp = f8 // Get signexp, recompute if unorm
|
|
fcvt.fx.trunc.s1 fXtruncInt = f8 // Convert to int in significand
|
|
addl rBigexp = 0x10016, r0 // Set exponent at which is integer
|
|
}
|
|
{ .mfi
|
|
mov rExpHalf = 0x0FFFE // Form sign and exponent of 0.5
|
|
fnorm.s1 fNormX = f8 // Normalize input
|
|
mov rExpMask = 0x1FFFF // Form exponent mask
|
|
}
|
|
;;
|
|
|
|
{ .mfi
|
|
setf.exp fHalf = rExpHalf // Form 0.5
|
|
fclass.m p7,p0 = f8, 0x0b // Test x unorm
|
|
nop.i 0
|
|
}
|
|
;;
|
|
|
|
{ .mfb
|
|
nop.m 0
|
|
fclass.m p6,p0 = f8, 0x1e3 // Test x natval, nan, inf
|
|
(p7) br.cond.spnt ROUND_UNORM // Branch if x unorm
|
|
}
|
|
;;
|
|
|
|
ROUND_COMMON:
|
|
// Return here from ROUND_UNORM
|
|
{ .mfb
|
|
nop.m 0
|
|
fcmp.lt.s1 p8,p9 = f8, f0 // Test if x < 0
|
|
(p6) br.cond.spnt ROUND_SPECIAL // Exit if x natval, nan, inf
|
|
}
|
|
;;
|
|
|
|
{ .mfi
|
|
nop.m 0
|
|
fcvt.xf f8 = fXtruncInt // Pre-Result if 0.5 <= |x| < 2^23
|
|
nop.i 0
|
|
}
|
|
;;
|
|
|
|
{ .mfi
|
|
and rExp = rSignexp, rExpMask // Get biased exponent
|
|
fmerge.s fInc = fNormX, f1 // Form increment if |rem| >= 0.5
|
|
nop.i 0
|
|
}
|
|
;;
|
|
|
|
{ .mmi
|
|
cmp.lt p6,p0 = rExp, rExpHalf // Is |x| < 0.5?
|
|
cmp.ge p7,p0 = rExp, rBigexp // Is |x| >= 2^23?
|
|
cmp.lt p10,p0 = rExp, rExpHalf // Is |x| < 0.5?
|
|
}
|
|
;;
|
|
|
|
// We must correct result if |x| < 0.5, or |x| >= 2^23
|
|
.pred.rel "mutex",p6,p7
|
|
{ .mfi
|
|
nop.m 0
|
|
(p6) fmerge.s f8 = fNormX, f0 // If |x| < 0.5, result sgn(x)*0
|
|
nop.i 0
|
|
}
|
|
{ .mfb
|
|
(p7) cmp.eq p10,p0 = r0, r0 // Also turn on p10 if |x| >= 2^23
|
|
(p7) fma.s.s0 f8 = fNormX, f1, f0 // If |x| >= 2^23, result x
|
|
(p10) br.ret.spnt b0 // Exit |x| < 0.5 or |x| >= 2^23
|
|
}
|
|
;;
|
|
|
|
// Here if 0.5 <= |x| < 2^23
|
|
{ .mfi
|
|
nop.m 0
|
|
(p9) fms.s1 fRem = fNormX, f1, f8 // Get remainder = x - trunc(x)
|
|
nop.i 0
|
|
}
|
|
{ .mfi
|
|
nop.m 0
|
|
(p8) fms.s1 fRem = f8, f1, fNormX // Get remainder = trunc(x) - x
|
|
nop.i 0
|
|
}
|
|
;;
|
|
|
|
{ .mfi
|
|
nop.m 0
|
|
fcmp.ge.s1 p9,p0 = fRem, fHalf // Test |rem| >= 0.5
|
|
nop.i 0
|
|
}
|
|
;;
|
|
|
|
// If x < 0 and remainder <= -0.5, then subtract 1 from result
|
|
// If x > 0 and remainder >= +0.5, then add 1 to result
|
|
{ .mfb
|
|
nop.m 0
|
|
(p9) fma.s.s0 f8 = f8, f1, fInc
|
|
br.ret.sptk b0
|
|
}
|
|
;;
|
|
|
|
|
|
ROUND_SPECIAL:
|
|
// Here if x natval, nan, inf
|
|
{ .mfb
|
|
nop.m 0
|
|
fma.s.s0 f8 = f8, f1, f0
|
|
br.ret.sptk b0
|
|
}
|
|
;;
|
|
|
|
ROUND_UNORM:
|
|
// Here if x unorm
|
|
{ .mfi
|
|
getf.exp rSignexp = fNormX // Get signexp, recompute if unorm
|
|
fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag
|
|
nop.i 0
|
|
}
|
|
{ .mfb
|
|
nop.m 0
|
|
fcvt.fx.trunc.s1 fXtruncInt = fNormX // Convert to int in significand
|
|
br.cond.sptk ROUND_COMMON // Return to main path
|
|
}
|
|
;;
|
|
|
|
GLOBAL_LIBM_END(roundf)
|
|
libm_alias_float_other (round, round)
|