From a334319f6530564d22e775935d9c91663623a1b4 Mon Sep 17 00:00:00 2001 From: Ulrich Drepper Date: Wed, 22 Dec 2004 20:10:10 +0000 Subject: (CFLAGS-tst-align.c): Add -mpreferred-stack-boundary=4. --- sysdeps/ia64/fpu/s_ceilf.S | 274 ++++++++++++++++++++++++++------------------- 1 file changed, 156 insertions(+), 118 deletions(-) (limited to 'sysdeps/ia64/fpu/s_ceilf.S') diff --git a/sysdeps/ia64/fpu/s_ceilf.S b/sysdeps/ia64/fpu/s_ceilf.S index 051534a202..d1011052e8 100644 --- a/sysdeps/ia64/fpu/s_ceilf.S +++ b/sysdeps/ia64/fpu/s_ceilf.S @@ -1,10 +1,10 @@ .file "ceilf.s" - -// Copyright (c) 2000 - 2003, Intel Corporation +// Copyright (C) 2000, 2001, Intel Corporation // All rights reserved. -// -// Contributed 2000 by the Intel Numerics Group, Intel Corporation +// +// 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. // // Redistribution and use in source and binary forms, with or without // modification, are permitted provided that the following conditions are @@ -20,67 +20,90 @@ // * 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 +// +// 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 +// 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 +// 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. -// +// 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. +// problem reports or change requests be submitted to it directly at +// http://developer.intel.com/opensource. // + +#include "libm_support.h" + +.align 32 +.global ceilf# + +.section .text +.proc ceilf# +.align 32 + // History //============================================================== -// 02/02/00 Initial version -// 06/13/00 Improved speed -// 06/27/00 Eliminated incorrect invalid flag setting -// 05/20/02 Cleaned up namespace and sf0 syntax -// 01/28/03 Improved performance -//============================================================== +// 2/02/00: Initial version +// 6/13/00: Improved speed +// 6/27/00: Eliminated incorrect invalid flag setting // API //============================================================== // float ceilf(float x) -//============================================================== -// general input registers: -// r14 - r19 +// general input registers: + +ceil_GR_FFFF = r14 +ceil_GR_signexp = r15 +ceil_GR_exponent = r16 +ceil_GR_expmask = r17 +ceil_GR_bigexp = r18 + -rSignexp = r14 -rExp = r15 -rExpMask = r16 -rBigexp = r17 -rM1 = r18 -rSignexpM1 = r19 +// predicate registers used: -// floating-point registers: -// f8 - f13 +// p6 ==> Input is NaN, infinity, zero +// p7 ==> Input is denormal +// p8 ==> Input is <0 +// p9 ==> Input is >=0 +// p10 ==> Input is already an integer (bigger than largest integer) +// p11 ==> Input is not a large integer +// p12 ==> Input is a smaller integer +// p13 ==> Input is not an even integer, so inexact must be set +// p14 ==> Input is between -1 and 0, so result will be -0 and inexact -fXInt = f9 -fNormX = f10 -fTmp = f11 -fAdj = f12 -fPreResult = f13 -// predicate registers used: -// p6 - p10 +// floating-point registers used: + +CEIL_SIGNED_ZERO = f7 +CEIL_NORM_f8 = f9 +CEIL_FFFF = f10 +CEIL_INEXACT = f11 +CEIL_FLOAT_INT_f8 = f12 +CEIL_INT_f8 = f13 +CEIL_adj = f14 +CEIL_MINUS_ONE = f15 // Overview of operation //============================================================== + // float ceilf(float x) -// Return an integer value (represented as a float) that is the smallest +// Return an integer value (represented as a float) that is the smallest // value not less than x // This is x rounded toward +infinity to an integral value. // Inexact is set if x != ceilf(x) -//============================================================== +// ************************************************************************** + +// Set denormal flag for denormal input and +// and take denormal fault if necessary. + +// Is the input an integer value already? // double_extended // if the exponent is > 1003e => 3F(true) = 63(decimal) @@ -101,124 +124,139 @@ fPreResult = f13 // If we multiply by 2^23, we no longer have a fractional part // So input is an integer value already. +// If x is NAN, ZERO, or INFINITY, then return + +// qnan snan inf norm unorm 0 -+ +// 1 1 1 0 0 1 11 0xe7 -.section .text -GLOBAL_LIBM_ENTRY(ceilf) + +ceilf: { .mfi - getf.exp rSignexp = f8 // Get signexp, recompute if unorm - fclass.m p7,p0 = f8, 0x0b // Test x unorm - addl rBigexp = 0x10016, r0 // Set exponent at which is integer + getf.exp ceil_GR_signexp = f8 + fcvt.fx.trunc.s1 CEIL_INT_f8 = f8 + addl ceil_GR_bigexp = 0x10016, r0 } { .mfi - mov rM1 = -1 // Set all ones - fcvt.fx.trunc.s1 fXInt = f8 // Convert to int in significand - mov rExpMask = 0x1FFFF // Form exponent mask + addl ceil_GR_FFFF = -1,r0 + fcmp.lt.s1 p8,p9 = f8,f0 + mov ceil_GR_expmask = 0x1FFFF ;; } -;; +// p7 ==> denorm { .mfi - mov rSignexpM1 = 0x2FFFF // Form signexp of -1 - fcmp.lt.s1 p8,p9 = f8, f0 // Test x < 0 - nop.i 0 + setf.sig CEIL_FFFF = ceil_GR_FFFF + fclass.m p7,p0 = f8, 0x0b + nop.i 999 } -{ .mfb - setf.sig fTmp = rM1 // Make const for setting inexact - fnorm.s1 fNormX = f8 // Normalize input -(p7) br.cond.spnt CEIL_UNORM // Branch if x unorm +{ .mfi + nop.m 999 + fnorm CEIL_NORM_f8 = f8 + nop.i 999 ;; } -;; -CEIL_COMMON: -// Return here from CEIL_UNORM +// Form 0 with sign of input in case negative zero is needed +{ .mfi + nop.m 999 + fmerge.s CEIL_SIGNED_ZERO = f8, f0 + nop.i 999 +} { .mfi - nop.m 0 - fclass.m p6,p0 = f8, 0x1e7 // Test x natval, nan, inf, 0 - nop.i 0 + nop.m 999 + fsub.s1 CEIL_MINUS_ONE = f0, f1 + nop.i 999 ;; +} + +// p6 ==> NAN, INF, ZERO +{ .mfb + nop.m 999 + fclass.m p6,p10 = f8, 0xe7 +(p7) br.cond.spnt L(CEIL_DENORM) ;; } -;; +L(CEIL_COMMON): .pred.rel "mutex",p8,p9 +// Set adjustment to add to trunc(x) for result +// If x>0, adjustment is 1.0 +// If x<=0, adjustment is 0.0 { .mfi - nop.m 0 -(p8) fma.s1 fAdj = f0, f0, f0 // If x < 0, adjustment is 0 - nop.i 0 + and ceil_GR_exponent = ceil_GR_signexp, ceil_GR_expmask +(p9) fadd.s1 CEIL_adj = f1,f0 + nop.i 999 } { .mfi - nop.m 0 -(p9) fma.s1 fAdj = f1, f1, f0 // If x > 0, adjustment is +1 - nop.i 0 + nop.m 999 +(p8) fadd.s1 CEIL_adj = f0,f0 + nop.i 999 ;; } -;; { .mfi - nop.m 0 - fcvt.xf fPreResult = fXInt // trunc(x) - nop.i 0 +(p10) cmp.ge.unc p10,p11 = ceil_GR_exponent, ceil_GR_bigexp +(p6) fnorm.s f8 = f8 + nop.i 999 ;; } -{ .mfb - nop.m 0 -(p6) fma.s.s0 f8 = f8, f1, f0 // Result if x natval, nan, inf, 0 -(p6) br.ret.spnt b0 // Exit if x natval, nan, inf, 0 + +{ .mfi + nop.m 999 +(p11) fcvt.xf CEIL_FLOAT_INT_f8 = CEIL_INT_f8 + nop.i 999 ;; } -;; -{ .mmi - and rExp = rSignexp, rExpMask // Get biased exponent -;; - cmp.ge p7,p6 = rExp, rBigexp // Is |x| >= 2^23? -(p8) cmp.lt.unc p10,p0 = rSignexp, rSignexpM1 // Is -1 < x < 0? +{ .mfi + nop.m 999 +(p10) fnorm.s f8 = CEIL_NORM_f8 + nop.i 999 ;; } -;; -// If -1 < x < 0, we turn off p6 and compute result as -0 +// Is -1 < x < 0? If so, result will be -0. Special case it with p14 set. { .mfi -(p10) cmp.ne p6,p0 = r0,r0 -(p10) fmerge.s f8 = fNormX, f0 - nop.i 0 + nop.m 999 +(p8) fcmp.gt.unc.s1 p14,p0 = CEIL_NORM_f8, CEIL_MINUS_ONE + nop.i 999 ;; } -;; -.pred.rel "mutex",p6,p7 { .mfi - nop.m 0 -(p6) fma.s.s0 f8 = fPreResult, f1, fAdj // Result if !int, |x| < 2^23 - nop.i 0 +(p14) cmp.ne p11,p0 = r0,r0 +(p14) fnorm.s f8 = CEIL_SIGNED_ZERO + nop.i 999 } { .mfi - nop.m 0 -(p7) fma.s.s0 f8 = fNormX, f1, f0 // Result, if |x| >= 2^23 -(p10) cmp.eq p6,p0 = r0,r0 // If -1 < x < 0, turn on p6 again + nop.m 999 +(p14) fmpy.s0 CEIL_INEXACT = CEIL_FFFF,CEIL_FFFF + nop.i 999 ;; } -;; { .mfi - nop.m 0 -(p6) fcmp.eq.unc.s1 p8, p9 = fPreResult, fNormX // Is trunc(x) = x ? - nop.i 0 + nop.m 999 +(p11) fadd.s f8 = CEIL_FLOAT_INT_f8,CEIL_adj + nop.i 999 ;; +} +{ .mfi + nop.m 999 +(p11) fcmp.eq.unc.s1 p12,p13 = CEIL_FLOAT_INT_f8, CEIL_NORM_f8 + nop.i 999 ;; } -;; +// Set inexact if result not equal to input { .mfi - nop.m 0 -(p9) fmpy.s0 fTmp = fTmp, fTmp // Dummy to set inexact - nop.i 0 + nop.m 999 +(p13) fmpy.s0 CEIL_INEXACT = CEIL_FFFF,CEIL_FFFF + nop.i 999 } +// Set result to input if integer { .mfb - nop.m 0 -(p8) fma.s.s0 f8 = fNormX, f1, f0 // If x int, result normalized x - br.ret.sptk b0 // Exit main path, 0 < |x| < 2^23 + nop.m 999 +(p12) fnorm.s f8 = CEIL_NORM_f8 + br.ret.sptk b0 ;; } -;; - -CEIL_UNORM: -// Here if x unorm +// Here if input denorm +L(CEIL_DENORM): { .mfb - getf.exp rSignexp = fNormX // Get signexp, recompute if unorm - fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag - br.cond.sptk CEIL_COMMON // Return to main path + getf.exp ceil_GR_signexp = CEIL_NORM_f8 + fcvt.fx.trunc.s1 CEIL_INT_f8 = CEIL_NORM_f8 + br.cond.sptk L(CEIL_COMMON) ;; } -;; -GLOBAL_LIBM_END(ceilf) +.endp ceilf +ASM_SIZE_DIRECTIVE(ceilf) -- cgit v1.2.3