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_logb.S | 314 ++++++++++++++++++++++++++-------------------- 1 file changed, 180 insertions(+), 134 deletions(-) (limited to 'sysdeps/ia64/fpu/s_logb.S') diff --git a/sysdeps/ia64/fpu/s_logb.S b/sysdeps/ia64/fpu/s_logb.S index 7ee898712b..76c4fe778e 100644 --- a/sysdeps/ia64/fpu/s_logb.S +++ b/sysdeps/ia64/fpu/s_logb.S @@ -1,10 +1,10 @@ .file "logb.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 @@ -20,43 +20,41 @@ // * 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. // // History //============================================================== -// 02/02/00 Initial version -// 02/16/00 Modified to conform to C9X -// 03/16/00 Improved speed -// 04/04/00 Unwind support added -// 05/30/00 Fixed bug when x double-extended denormal -// 08/15/00 Bundle added after call to __libm_error_support to properly +// 2/02/00 Initial version +// 2/16/00 Modified to conform to C9X +// 3/16/00 Improved speed +// 4/04/00 Unwind support added +// 5/30/00 Fixed bug when x double-extended denormal +// 8/15/00 Bundle added after call to __libm_error_support to properly // set [the previously overwritten] GR_Parameter_RESULT. -// 05/20/02 Cleaned up namespace and sf0 syntax -// 01/20/03 Improved performance // // API //============================================================== -// double logb( double x ); +// double logb( double x); // // Overview of operation //============================================================== -// The logb function extracts the exponent of x as an integer in -// floating-point format. +// The logb function extracts the exponent of x as an integer in +// floating-point format. // logb computes log2 of x as a double // // logb is similar to ilogb but differs in the following ways: @@ -73,169 +71,217 @@ // // Registers used //============================================================== -// general registers used: -// r26 -> r38 -// r35 -> r38 used as parameters to error path +// general registers used: +// ar.pfs r32 +// r33 -> r37 +// r38 -> r41 used as parameters to error path // -// predicate registers used: +// predicate registers used: // p6, p7, p8 -// floating-point registers used: +// floating-point registers used: // f9, f10, f11 // f8, input -rExpBias = r26 -rExpMask = r27 -rSignexp_x = r28 -rExp_x = r29 -rTrueExp_x = r30 -rExp_2to64 = r31 +#include "libm_support.h" +GR_SAVE_B0 = r34 +GR_SAVE_GP = r35 GR_SAVE_PFS = r32 -GR_SAVE_B0 = r33 -GR_SAVE_GP = r34 -GR_Parameter_X = r35 -GR_Parameter_Y = r36 -GR_Parameter_RESULT = r37 -GR_Parameter_TAG = r38 +GR_Parameter_X = r38 +GR_Parameter_Y = r39 +GR_Parameter_RESULT = r40 -fExp_in_signif = f9 -fNorm_x = f10 -fFloat_Exp = f10 -f2to64 = f11 +.align 32 +.global logb# .section .text -GLOBAL_LIBM_ENTRY(logb) +.proc logb# +.align 32 + +logb: + +// qnan snan inf norm unorm 0 -+ +// 0 0 0 0 1 0 11 +// 0 b +{ .mfi + alloc r32=ar.pfs,1,5,4,0 +(p0) fclass.m.unc p8,p0 = f8, 0x0b + nop.i 999 +} // X NORMAL -// TrueExp_x = exp(f8) - 0xffff -// sig = TrueExp_x +// r37 = exp(f8) - - 0xffff +// sig(f8) = r37 // f8 = convert_to_fp (sig)) { .mfi - getf.exp rSignexp_x = f8 - fclass.m p8,p0 = f8, 0x0b // Test for x unorm - mov rExpBias = 0xffff // Exponent bias +(p0) getf.exp r35 = f8 +(p0) fnorm f10=f8 + nop.i 999 ;; } -{ .mfi - nop.m 0 - fnorm.s1 fNorm_x = f8 - mov rExpMask = 0x1ffff // Exponent mask + +// qnan snan inf norm unorm 0 -+ +// 1 1 1 0 0 0 11 +// e 3 +{ .mmf +(p0) mov r33 = 0xffff +(p0) mov r34 = 0x1ffff +(p0) fclass.m.unc p6,p0 = f8, 0xe3 ;; } -;; -// Form signexp of 2^64 in case need to scale denormal { .mfb - mov rExp_2to64 = 0x1003f - fclass.m p6,p0 = f8, 0x1e3 // Test x natval, nan, inf -(p8) br.cond.spnt LOGB_DENORM // Branch if x unorm +(p0) and r36 = r35, r34 +(p0) fclass.m.unc p7,p0 = f8, 0x07 +(p8) br.cond.spnt L(LOGB_DENORM) ;; } -;; -LOGB_COMMON: -// Return here from LOGB_DENORM -{ .mfi - and rExp_x = rSignexp_x, rExpMask // Get biased exponent - fclass.m p7,p0 = f8, 0x07 // Test x zero - nop.i 0 -} -;; - -// X NAN or INFINITY, return f8 * f8 -{ .mfb - sub rTrueExp_x = rExp_x, rExpBias // Get true exponent -(p6) fma.d.s0 f8= f8,f8,f0 // Result if x natval, nan, inf -(p6) br.ret.spnt b0 // Exit if x natval, nan, inf +{ .mib +(p0) sub r37 = r36, r33 + nop.i 999 +(p6) br.cond.spnt L(LOGB_NAN_INF) ;; } -;; { .mib - setf.sig fExp_in_signif = rTrueExp_x // Exponent as integer in fp +(p0) setf.sig f9 = r37 nop.i 999 -(p7) br.cond.spnt LOGB_ZERO +(p7) br.cond.spnt L(LOGB_ZERO) ;; +} + +{ .mfi + nop.m 999 +(p0) fcvt.xf f10 = f9 + nop.i 999 ;; } -;; -// Result can be represented in less than 24 bits, so no precision completer -// is needed. { .mfb - nop.m 0 - fcvt.xf f8 = fExp_in_signif - br.ret.sptk b0 // Exit main path, 0 < |x| < inf + nop.m 999 +(p0) fnorm.d f8 = f10 +(p0) br.ret.sptk b0 ;; } -;; -LOGB_DENORM: -// Form 2^64 in case need to scale denormal +L(LOGB_DENORM): +// Form signexp of 2^64 in case need to scale denormal // Check to see if double-extended denormal { .mfi - setf.exp f2to64 = rExp_2to64 - fclass.m p8,p0 = fNorm_x, 0x0b - nop.i 0 +(p0) mov r38 = 0x1003f +(p0) fclass.m.unc p8,p0 = f10, 0x0b + nop.i 999 ;; } -;; +// Form 2^64 in case need to scale denormal { .mfi - nop.m 0 - fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag - nop.i 0 +(p0) setf.exp f11 = r38 + nop.f 999 + nop.i 999 ;; } -;; // If double-extended denormal add 64 to exponent bias for scaling // If double-extended denormal form x * 2^64 which is normal { .mfi -(p8) add rExpBias = 64, rExpBias -(p8) fmpy.s1 fNorm_x = fNorm_x, f2to64 - nop.i 0 +(p8) add r33 = 64, r33 +(p8) fmpy f10 = f10, f11 + nop.i 999 ;; } -;; // Logic is the same as normal path but use normalized input -{ .mib - getf.exp rSignexp_x = fNorm_x - nop.i 0 - br.cond.sptk LOGB_COMMON // Return to main path +{ .mmi +(p0) getf.exp r35 = f10 ;; + nop.m 999 + nop.i 999 ;; } -;; -LOGB_ZERO: -// Here if x zero -// f10 = -|f8| -// f9 = 1.0/f10 = -1.0/|f8| = -inf +{ .mmi +(p0) and r36 = r35, r34 ;; +(p0) sub r37 = r36, r33 + nop.i 999 ;; +} -{ .mmf - alloc r32=ar.pfs,1,2,4,0 - mov GR_Parameter_TAG = 151 // Error code - fmerge.ns f10 = f0,f8 +{ .mmi +(p0) setf.sig f9 = r37 + nop.m 999 + nop.i 999 ;; +} + +{ .mfi + nop.m 999 +(p0) fcvt.xf f10 = f9 + nop.i 999 ;; } -;; { .mfb - nop.m 0 - frcpa.s0 f9,p6 = f1,f10 // Produce -inf, Z flag - br.cond.sptk __libm_error_region // Call error support + nop.m 999 +(p0) fnorm.d f8 = f10 +(p0) br.ret.sptk b0 ;; } -;; -GLOBAL_LIBM_END(logb) +L(LOGB_NAN_INF): + +// X NAN or INFINITY, return f8 * f8 +{ .mfb + nop.m 999 +(p0) fma.d f8= f8,f8,f0 +(p0) br.ret.sptk b0 ;; +} +.endp logb# +ASM_SIZE_DIRECTIVE(logb) -LOCAL_LIBM_ENTRY(__libm_error_region) +// Stack operations when calling error support. +// (1) (2) (3) (call) (4) +// sp -> + psp -> + psp -> + sp -> + +// | | | | +// | | <- GR_Y R3 ->| <- GR_RESULT | -> f8 +// | | | | +// | <-GR_Y Y2->| Y2 ->| <- GR_Y | +// | | | | +// | | <- GR_X X1 ->| | +// | | | | +// sp-64 -> + sp -> + sp -> + + +// save ar.pfs save b0 restore gp +// save gp restore ar.pfs + + + +.proc __libm_error_region +__libm_error_region: +L(LOGB_ZERO): .prologue +// f9 = |f8| +// f10 = -f9 = -|f8| +// f9 = 1.0/f10 = -1.0/-|f8| + +{ .mfi + mov r41 = 151 // Error code +(p0) fmerge.s f9 = f0,f8 + nop.i 999 +} +;; + + { .mfi - add GR_Parameter_Y=-32,sp // Parameter 2 value - nop.f 0 + nop.m 999 + fmerge.ns f10 = f0,f9 + nop.i 999 +} +;; + +// (1) +{ .mfi + add GR_Parameter_Y=-32,sp // Parameter 2 value + frcpa f9,p6 = f1,f10 .save ar.pfs,GR_SAVE_PFS - mov GR_SAVE_PFS=ar.pfs // Save ar.pfs + mov GR_SAVE_PFS=ar.pfs // Save ar.pfs } { .mfi .fframe 64 - add sp=-64,sp // Create new stack + add sp=-64,sp // Create new stack nop.f 0 - mov GR_SAVE_GP=gp // Save gp + mov GR_SAVE_GP=gp // Save gp };; + +// (2) { .mmi stfd [GR_Parameter_Y] = f0,16 // STORE Parameter 2 on stack add GR_Parameter_X = 16,sp // Parameter 1 address @@ -244,38 +290,38 @@ LOCAL_LIBM_ENTRY(__libm_error_region) };; .body +// (3) { .mib stfd [GR_Parameter_X] = f8 // STORE Parameter 1 on stack add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address - nop.b 0 + nop.b 0 } { .mib stfd [GR_Parameter_Y] = f9 // 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 + nop.m 0 + add GR_Parameter_RESULT = 48,sp };; +// (4) { .mmi ldfd 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 };; -LOCAL_LIBM_END(__libm_error_region) - +.endp __libm_error_region +ASM_SIZE_DIRECTIVE(__libm_error_region) .type __libm_error_support#,@function .global __libm_error_support# -- cgit 1.4.1