1.file "ilogbl.s" 2 3 4// Copyright (c) 2000 - 2003, Intel Corporation 5// All rights reserved. 6// 7// 8// Redistribution and use in source and binary forms, with or without 9// modification, are permitted provided that the following conditions are 10// met: 11// 12// * Redistributions of source code must retain the above copyright 13// notice, this list of conditions and the following disclaimer. 14// 15// * Redistributions in binary form must reproduce the above copyright 16// notice, this list of conditions and the following disclaimer in the 17// documentation and/or other materials provided with the distribution. 18// 19// * The name of Intel Corporation may not be used to endorse or promote 20// products derived from this software without specific prior written 21// permission. 22 23// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 26// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS 27// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 28// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 29// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 30// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY 31// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING 32// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 33// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 34// 35// Intel Corporation is the author of this code, and requests that all 36// problem reports or change requests be submitted to it directly at 37// http://www.intel.com/software/products/opensource/libraries/num.htm. 38// 39// History 40//============================================================== 41// 02/03/00 Initial version 42// 05/26/00 Fix bug when x a double-extended denormal; 43// if x=0 call error routine, per C9X 44// 08/15/00 Bundle added after call to __libm_error_support to properly 45// set [the previously overwritten] GR_Parameter_RESULT. 46// 01/20/01 Fixed result for x=0 47// 05/20/02 Cleaned up namespace and sf0 syntax 48// 01/20/03 Improved performance 49// 50// API 51//============================================================== 52// int ilogbl( long double x ); 53// 54// Overview of operation 55//============================================================== 56// The ilogbl function extracts the exponent of x as an integer 57// and returns it in r8 58// 59// ilogbl is similar to logbl but differs in the following ways: 60// +-inf 61// ilogbl: returns INT_MAX 62// logbl: returns +inf 63// Nan returns FP_LOGBNAN (which is either INT_MAX or INT_MIN) 64// ilogbl: returns INT_MAX (7fffffff) 65// logbl: returns QNAN (quietized SNAN) 66// 0 returns FP_ILOGB0 (which is either INT_MIN or -INT_MAX) 67// ilogbl: returns -INT_MAX (80000001) 68// logbl: returns -inf, raises the divide-by-zero exception, 69// and calls libm_error_support to set domain error 70// 71// Registers used 72//============================================================== 73// general registers used: 74// r26 -> r39 75// r36 -> r39 used as parameters to error path 76// 77// predicate registers used: 78// p6 -> p10 79// floating-point registers used: 80// f9, f10, f11 81// f8, input 82 83rExpBias = r26 84rExpMask = r27 85rSignexp_x = r28 86rExp_x = r29 87rIntMax = r30 88rExp_2to64 = r31 89 90GR_SAVE_PFS = r32 91rTrialResult = r33 92GR_SAVE_B0 = r34 93GR_SAVE_GP = r35 94 95GR_Parameter_X = r36 96GR_Parameter_Y = r37 97GR_Parameter_RESULT = r38 98GR_Parameter_TAG = r39 99 100fTmp = f9 101fNorm_x = f10 102f2to64 = f11 103 104.section .text 105GLOBAL_LIBM_ENTRY(__ieee754_ilogbl) 106 107// X NORMAL 108// TrueExp_x = exp(f8) - 0xffff 109// r8 = TrueExp_x 110{ .mfi 111 getf.exp rSignexp_x = f8 112 fclass.m p8,p0 = f8, 0x0b // Test for x unorm 113 mov rExpBias = 0xffff // Exponent bias 114} 115{ .mfi 116 nop.m 0 117 fnorm.s1 fNorm_x = f8 118 mov rExpMask = 0x1ffff // Exponent mask 119} 120;; 121 122// Form signexp of 2^64 in case need to scale denormal 123{ .mfb 124 mov rExp_2to64 = 0x1003f 125 fclass.m p6,p9 = f8, 0x1e3 // Test x natval, nan, inf 126(p8) br.cond.spnt ILOGB_DENORM // Branch if x unorm 127} 128;; 129 130ILOGB_COMMON: 131// Return here from ILOGB_DENORM 132{ .mfi 133 and rExp_x = rSignexp_x, rExpMask // Get biased exponent 134 fclass.m p7,p10 = f8, 0x07 // Test x zero 135 nop.i 0 136} 137{ .mlx 138 nop.m 0 139 movl rIntMax = 0x000000007fffffff // Form INT_MAX 140} 141;; 142 143.pred.rel "mutex",p6,p9 144{ .mfi 145(p9) sub r8 = rExp_x, rExpBias // Get true exponent for normal path 146(p6) fma.s0 fTmp = f8, f8, f0 // Dummy to set Invalid flag 147(p6) mov r8 = rIntMax // If nan, inf, return INT_MAX 148} 149{ .mbb 150 nop.m 0 151(p7) br.cond.spnt ILOGB_ZERO // Branch if x zero 152(p10) br.ret.sptk b0 // Exit if x not zero 153} 154;; 155 156 157ILOGB_DENORM: 158// Form 2^64 in case need to scale denormal 159// Check to see if double-extended denormal 160{ .mfi 161 setf.exp f2to64 = rExp_2to64 162 fclass.m p8,p0 = fNorm_x, 0x0b 163 nop.i 0 164} 165;; 166 167{ .mfi 168 nop.m 0 169 fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag 170 nop.i 0 171} 172;; 173 174// If double-extended denormal add 64 to exponent bias for scaling 175// If double-extended denormal form x * 2^64 which is normal 176{ .mfi 177(p8) add rExpBias = 64, rExpBias 178(p8) fmpy.s1 fNorm_x = fNorm_x, f2to64 179 nop.i 0 180} 181;; 182 183// Logic is the same as normal path but use normalized input 184{ .mib 185 getf.exp rSignexp_x = fNorm_x 186 nop.i 0 187 br.cond.sptk ILOGB_COMMON // Return to main path 188} 189;; 190 191ILOGB_ZERO: 192// Here if x zero 193// Return INT_MIN, call error support 194 195{ .mlx 196 alloc r32=ar.pfs,1,3,4,0 197 movl rTrialResult = 0x0000000080000000 198} 199{ .mib 200 mov GR_Parameter_TAG = 156 // Error code 201 nop.i 0 202 br.cond.sptk __libm_error_region // Call error support 203} 204;; 205 206GLOBAL_LIBM_END(__ieee754_ilogbl) 207 208 209LOCAL_LIBM_ENTRY(__libm_error_region) 210.prologue 211 212{ .mfi 213 add GR_Parameter_Y=-32,sp // Parameter 2 value 214 nop.f 0 215.save ar.pfs,GR_SAVE_PFS 216 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs 217} 218{ .mfi 219.fframe 64 220 add sp=-64,sp // Create new stack 221 nop.f 0 222 mov GR_SAVE_GP=gp // Save gp 223};; 224 225{ .mmi 226 stfe [GR_Parameter_Y] = f0,16 // STORE Parameter 2 on stack 227 add GR_Parameter_X = 16,sp // Parameter 1 address 228.save b0, GR_SAVE_B0 229 mov GR_SAVE_B0=b0 // Save b0 230};; 231 232.body 233{ .mib 234 stfe [GR_Parameter_X] = f8 // STORE Parameter 1 on stack 235 add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address 236 nop.b 0 237} 238{ .mib 239 stfe [GR_Parameter_Y] = f9 // Store Parameter 3 on stack 240 add GR_Parameter_Y = -16,GR_Parameter_Y 241 br.call.sptk b0=__libm_error_support# // Call error handling function 242};; 243 244{ .mmi 245 add GR_Parameter_RESULT = 48,sp 246 nop.m 0 247 nop.i 0 248};; 249 250{ .mmi 251 mov r8 = rTrialResult 252.restore sp 253 add sp = 64,sp // Restore stack pointer 254 mov b0 = GR_SAVE_B0 // Restore return address 255};; 256 257{ .mib 258 mov gp = GR_SAVE_GP // Restore gp 259 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs 260 br.ret.sptk b0 261};; 262 263LOCAL_LIBM_END(__libm_error_region) 264 265 266.type __libm_error_support#,@function 267.global __libm_error_support# 268