diff options
author | Ulrich Drepper <drepper@redhat.com> | 2004-12-22 20:10:10 +0000 |
---|---|---|
committer | Ulrich Drepper <drepper@redhat.com> | 2004-12-22 20:10:10 +0000 |
commit | a334319f6530564d22e775935d9c91663623a1b4 (patch) | |
tree | b5877475619e4c938e98757d518bb1e9cbead751 /sysdeps/ia64/fpu/s_ilogbf.S | |
parent | 0ecb606cb6cf65de1d9fc8a919bceb4be476c602 (diff) | |
download | glibc-a334319f6530564d22e775935d9c91663623a1b4.tar.gz glibc-a334319f6530564d22e775935d9c91663623a1b4.tar.xz glibc-a334319f6530564d22e775935d9c91663623a1b4.zip |
(CFLAGS-tst-align.c): Add -mpreferred-stack-boundary=4.
Diffstat (limited to 'sysdeps/ia64/fpu/s_ilogbf.S')
-rw-r--r-- | sysdeps/ia64/fpu/s_ilogbf.S | 305 |
1 files changed, 145 insertions, 160 deletions
diff --git a/sysdeps/ia64/fpu/s_ilogbf.S b/sysdeps/ia64/fpu/s_ilogbf.S index 9e971bc634..ffa6d3b672 100644 --- a/sysdeps/ia64/fpu/s_ilogbf.S +++ b/sysdeps/ia64/fpu/s_ilogbf.S @@ -1,10 +1,10 @@ .file "ilogbf.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 @@ -20,249 +20,234 @@ // * 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 //============================================================== -// 02/03/00 Initial version -// 05/26/00 Fix bug when x a double-extended denormal; +// 2/03/00 Initial version +// 5/26/00 Fix bug when x a double-extended denormal; // if x=0 call error routine, per C9X -// 08/15/00 Bundle added after call to __libm_error_support to properly +// 8/15/00 Bundle added after call to __libm_error_support to properly // set [the previously overwritten] GR_Parameter_RESULT. -// 01/20/01 Fixed result for x=0 -// 05/20/02 Cleaned up namespace and sf0 syntax -// 01/20/03 Improved performance -// +// 1/20/01 Fixed result for x=0 + +.align 32 +.global ilogbf# + +.section .text +.proc ilogbf# +.align 32 + // API //============================================================== -// int ilogbf( float x ); -// +// int = ilogbf(float) + // Overview of operation //============================================================== -// The ilogbf function extracts the exponent of x as an integer +// ilogbf computes log2(x) as an int // and returns it in r8 -// -// ilogbf is similar to logbf but differs in the following ways: + +// ilogbf is similar to logbf but differs in the following ways: // +-inf // ilogbf: returns INT_MAX // logbf: returns +inf -// Nan returns FP_LOGBNAN (which is either INT_MAX or INT_MIN) +// Nan returns FP_ILOGBNAN (which is either INT_MAX or INT_MIN) // ilogbf: returns INT_MAX (7fffffff) -// logbf: returns QNAN (quietized SNAN) +// logbf: returns QNAN (quieted SNAN) // 0 returns FP_ILOGB0 (which is either INT_MIN or -INT_MAX) -// ilogbf: returns -INT_MAX (80000001) -// logbf: returns -inf, raises the divide-by-zero exception, -// and calls libm_error_support to set domain error -// +// ilogbf: returns INT_MIN (80000000) +// logbf: returns -inf + // Registers used //============================================================== -// general registers used: -// r26 -> r39 -// r36 -> r39 used as parameters to error path -// -// predicate registers used: -// p6 -> p10 -// floating-point registers used: -// f9, f10, f11 -// f8, input -rExpBias = r26 -rExpMask = r27 -rSignexp_x = r28 -rExp_x = r29 -rIntMax = r30 -rExp_2to64 = r31 +// general local registers: +// ar.pfs r32 +// r33 -> r37 +// r38 -> r41 used as parameters to error path + +// predicate registers used: +// p6 - x nan, inf +// p7 - x 0 +// p8 - x norm, unorm +// p9 - x unorm + +// floating-point registers used: +// f8 - f10 + +#include "libm_support.h" GR_SAVE_PFS = r32 -rTrialResult = r33 GR_SAVE_B0 = r34 GR_SAVE_GP = r35 +GR_Parameter_X = r38 +GR_Parameter_Y = r39 +GR_Parameter_RESULT = r40 +GR_Parameter_TAG = r41 -GR_Parameter_X = r36 -GR_Parameter_Y = r37 -GR_Parameter_RESULT = r38 -GR_Parameter_TAG = r39 +FR_X = f8 +FR_Y = f0 +FR_RESULT = f0 -fTmp = f9 -fNorm_x = f10 -f2to64 = f11 -.section .text -GLOBAL_LIBM_ENTRY(ilogbf) +ilogbf: -// X NORMAL -// TrueExp_x = exp(f8) - 0xffff -// r8 = TrueExp_x -{ .mfi - getf.exp rSignexp_x = f8 - fclass.m p8,p0 = f8, 0x0b // Test for x unorm - mov rExpBias = 0xffff // Exponent bias +// Form signexp of 2^64 in case need to scale denormal +{ .mmf + alloc r32=ar.pfs,1,5,4,0 +(p0) mov r37 = 0x1003f +(p0) fnorm f9 = f8 ;; } + +// Form 2^64 in case need to scale denormal { .mfi - nop.m 0 - fnorm.s1 fNorm_x = f8 - mov rExpMask = 0x1ffff // Exponent mask +(p0) setf.exp f10 = r37 +(p0) fclass.m.unc p7, p8 = f8, 0xe3 +(p0) mov r34 = 0xffff ;; } -;; -// Form signexp of 2^64 in case need to scale denormal -{ .mfb - mov rExp_2to64 = 0x1003f - fclass.m p6,p9 = f8, 0x1e3 // Test x natval, nan, inf -(p8) br.cond.spnt ILOGB_DENORM // Branch if x unorm -} -;; +// qnan snan inf norm unorm 0 -+ +// 1 1 1 0 0 0 11 +// e 3 +// X ZERO, returns INT_MIN +// X INF or NAN, returns INT_MAX -ILOGB_COMMON: -// Return here from ILOGB_DENORM { .mfi - and rExp_x = rSignexp_x, rExpMask // Get biased exponent - fclass.m p7,p10 = f8, 0x07 // Test x zero - nop.i 0 +(p0) mov r35 = 0x1ffff +(p8) fclass.m.unc p6, p8 = f8, 0x07 + nop.i 999 ;; } { .mlx - nop.m 0 - movl rIntMax = 0x000000007fffffff // Form INT_MAX + nop.m 999 +(p7) movl r8 = 0x000000007fffffff ;; } -;; -.pred.rel "mutex",p6,p9 -{ .mfi -(p9) sub r8 = rExp_x, rExpBias // Get true exponent for normal path -(p6) fma.s0 fTmp = f8, f8, f0 // Dummy to set Invalid flag -(p6) mov r8 = rIntMax // If nan, inf, return INT_MAX -} -{ .mbb - nop.m 0 -(p7) br.cond.spnt ILOGB_ZERO // Branch if x zero -(p10) br.ret.sptk b0 // Exit if x not zero +{ .mib + nop.m 999 + nop.i 999 +(p6) br.cond.spnt L(ILOGB_ZERO) ;; } -;; - -ILOGB_DENORM: -// Form 2^64 in case need to scale denormal -// Check to see if double-extended denormal +// Test for denormal { .mfi - setf.exp f2to64 = rExp_2to64 - fclass.m p8,p0 = fNorm_x, 0x0b - nop.i 0 + nop.m 999 +(p8) fclass.m.unc p9, p0 = f9, 0x0b + nop.i 999 ;; } -;; -{ .mfi - nop.m 0 - fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag - nop.i 0 +L(ILOGB_COMMON): +// X NORMAL returns true exponent +{ .mmi + nop.m 999 +(p8) getf.exp r33 = f9 + 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 +// If denormal add 64 to exponent bias for scaling +{ .mfb +(p9) add r34 = 64, r34 + nop.f 999 +(p9) br.cond.spnt L(ILOGB_DENORM) ;; } -;; -// Logic is the same as normal path but use normalized input -{ .mib - getf.exp rSignexp_x = fNorm_x - nop.i 0 - br.cond.sptk ILOGB_COMMON // Return to main path +{ .mmi +(p8) and r36 = r35, r33 + nop.m 999 + nop.i 999 ;; } -;; -ILOGB_ZERO: -// Here if x zero -// Return INT_MIN, call error support - -{ .mlx - alloc r32=ar.pfs,1,3,4,0 - movl rTrialResult = 0x0000000080000000 -} { .mib - mov GR_Parameter_TAG = 158 // Error code - nop.i 0 - br.cond.sptk __libm_error_region // Call error support +(p8) sub r8 = r36, r34 + nop.i 999 +(p0) br.ret.sptk b0 ;; } -;; -GLOBAL_LIBM_END(ilogbf) +L(ILOGB_DENORM): +// Here if x denormal +// Form x * 2^64 which is normal +// Return to common code +{ .mfb + cmp.eq p8,p9 = r0,r0 + fmpy f9 = f9, f10 + br.cond.sptk L(ILOGB_COMMON) ;; +} +// X ZERO +// return INT_MIN, call error support +L(ILOGB_ZERO): +{.mlx + mov GR_Parameter_TAG = 158 +(p6) movl r33 = 0x0000000080000000 ;; +};; +.endp ilogbf +ASM_SIZE_DIRECTIVE(ilogbf) -LOCAL_LIBM_ENTRY(__libm_error_region) +.proc __libm_error_region +__libm_error_region: .prologue - { .mfi - add GR_Parameter_Y=-32,sp // Parameter 2 value + 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 + 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 };; - { .mmi - stfs [GR_Parameter_Y] = f0,16 // STORE Parameter 2 on stack - add GR_Parameter_X = 16,sp // Parameter 1 address + stfs [GR_Parameter_Y] = FR_Y,16 // Save 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 + mov GR_SAVE_B0=b0 // Save b0 };; - .body { .mib - stfs [GR_Parameter_X] = f8 // STORE Parameter 1 on stack - add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address - nop.b 0 + stfs [GR_Parameter_X] = FR_X // Store Parameter 1 on stack + add GR_Parameter_RESULT = 0,GR_Parameter_Y + nop.b 0 // Parameter 3 address } { .mib - stfs [GR_Parameter_Y] = f9 // Store Parameter 3 on stack + 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 + 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 };; - { .mmi - mov r8 = rTrialResult + mov r8 = r33 // Store result .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 + br.ret.sptk b0 // Return };; -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# |