sysdeps/ia64/fpu/s_logbf.S

   1 .file "logbf.s"
   2
   3
   4 // Copyright (c) 2000 - 2003, Intel Corporation
   5 // All rights reserved.
   6 //
   7 // Contributed 2000 by the Intel Numerics Group, Intel Corporation
   8 //
   9 // Redistribution and use in source and binary forms, with or without
  10 // modification, are permitted provided that the following conditions are
  11 // met:
  12 //
  13 // * Redistributions of source code must retain the above copyright
  14 // notice, this list of conditions and the following disclaimer.
  15 //
  16 // * Redistributions in binary form must reproduce the above copyright
  17 // notice, this list of conditions and the following disclaimer in the
  18 // documentation and/or other materials provided with the distribution.
  19 //
  20 // * The name of Intel Corporation may not be used to endorse or promote
  21 // products derived from this software without specific prior written
  22 // permission.
  23
  24 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  25 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  26 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  27 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
  28 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
  29 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  30 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
  31 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
  32 // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
  33 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35 //
  36 // Intel Corporation is the author of this code, and requests that all
  37 // problem reports or change requests be submitted to it directly at
  38 // http://www.intel.com/software/products/opensource/libraries/num.htm.
  39 //
  40 // History
  41 //==============================================================
  42 // 02/02/00 Initial version
  43 // 02/16/00 Modified to conform to C9X
  44 // 03/16/00 Improved speed
  45 // 04/04/00 Unwind support added
  46 // 05/30/00 Fixed bug when x double-extended denormal
  47 // 08/15/00 Bundle added after call to __libm_error_support to properly
  48 //          set [the previously overwritten] GR_Parameter_RESULT.
  49 // 05/20/02 Cleaned up namespace and sf0 syntax
  50 // 01/20/03 Improved performance
  51 //
  52 // API
  53 //==============================================================
  54 // float logbf( float x );
  55 //
  56 // Overview of operation
  57 //==============================================================
  58 // The logbf function extracts the exponent of x as an integer in
  59 // floating-point format.
  60 // logbf computes log2 of x as a float
  61 //
  62 // logbf is similar to ilogbf but differs in the following ways:
  63 //         +-inf
  64 //            ilogbf: returns INT_MAX
  65 //             logbf: returns +inf
  66 //         Nan  returns FP_LOGBNAN (which is either INT_MAX or INT_MIN)
  67 //            ilogbf: returns INT_MAX (7fffffff)
  68 //             logbf: returns QNAN (quietized SNAN)
  69 //         0    returns FP_ILOGB0 (which is either INT_MIN or -INT_MAX)
  70 //            ilogbf: returns -INT_MAX (80000001)
  71 //             logbf: returns -inf, raises the divide-by-zero exception,
  72 //                   and calls libm_error_support to set domain error
  73 //
  74 // Registers used
  75 //==============================================================
  76 // general registers used:
  77 // r26 -> r38
  78 // r35 -> r38 used as parameters to error path
  79 //
  80 // predicate registers used:
  81 // p6, p7, p8
  82 // floating-point registers used:
  83 // f9, f10, f11
  84 // f8, input
  85
  86 rExpBias            = r26
  87 rExpMask            = r27
  88 rSignexp_x          = r28
  89 rExp_x              = r29
  90 rTrueExp_x          = r30
  91 rExp_2to64          = r31
  92
  93 GR_SAVE_PFS         = r32
  94 GR_SAVE_B0          = r33
  95 GR_SAVE_GP          = r34
  96
  97 GR_Parameter_X      = r35
  98 GR_Parameter_Y      = r36
  99 GR_Parameter_RESULT = r37
 100 GR_Parameter_TAG    = r38
 101
 102 fExp_in_signif      = f9
 103 fNorm_x             = f10
 104 fFloat_Exp          = f10
 105 f2to64              = f11
 106
 107 .section .text
 108 GLOBAL_LIBM_ENTRY(logbf)
 109
 110 // X NORMAL
 111 // TrueExp_x = exp(f8) - 0xffff
 112 // sig = TrueExp_x
 113 // f8 = convert_to_fp (sig))
 114 { .mfi
 115       getf.exp        rSignexp_x = f8
 116       fclass.m        p8,p0 = f8, 0x0b   // Test for x unorm
 117       mov             rExpBias = 0xffff  // Exponent bias
 118 }
 119 { .mfi
 120       nop.m           0
 121       fnorm.s1        fNorm_x = f8
 122       mov             rExpMask = 0x1ffff // Exponent mask
 123 }
 124 ;;
 125
 126 // Form signexp of 2^64 in case need to scale denormal
 127 { .mfb
 128       mov             rExp_2to64 = 0x1003f
 129       fclass.m        p6,p0 = f8, 0x1e3  // Test x natval, nan, inf
 130 (p8)  br.cond.spnt    LOGB_DENORM        // Branch if x unorm
 131 }
 132 ;;
 133
 134 LOGB_COMMON:
 135 // Return here from LOGB_DENORM
 136 { .mfi
 137       and             rExp_x = rSignexp_x, rExpMask // Get biased exponent
 138       fclass.m        p7,p0 = f8, 0x07   // Test x zero
 139       nop.i           0
 140 }
 141 ;;
 142
 143 // X NAN or INFINITY, return f8 * f8
 144 { .mfb
 145       sub             rTrueExp_x = rExp_x, rExpBias // Get true exponent
 146 (p6)  fma.s.s0        f8= f8,f8,f0       // Result if x natval, nan, inf
 147 (p6)  br.ret.spnt     b0                 // Exit if x natval, nan, inf
 148 }
 149 ;;
 150
 151 { .mib
 152       setf.sig        fExp_in_signif = rTrueExp_x // Exponent as integer in fp
 153       nop.i 999
 154 (p7)  br.cond.spnt    LOGB_ZERO
 155 }
 156 ;;
 157
 158 // Result can be represented in less than 24 bits, so no precision completer
 159 // is needed.
 160 { .mfb
 161       nop.m           0
 162       fcvt.xf         f8 = fExp_in_signif
 163       br.ret.sptk     b0                 // Exit main path, 0 < |x| < inf
 164 }
 165 ;;
 166
 167 LOGB_DENORM:
 168 // Form 2^64 in case need to scale denormal
 169 // Check to see if double-extended denormal
 170 { .mfi
 171       setf.exp        f2to64 = rExp_2to64
 172       fclass.m        p8,p0 = fNorm_x, 0x0b
 173       nop.i           0
 174 }
 175 ;;
 176
 177 { .mfi
 178       nop.m           0
 179       fcmp.eq.s0      p7,p0 = f8, f0           // Dummy op to set denormal flag
 180       nop.i           0
 181 }
 182 ;;
 183
 184 // If double-extended denormal add 64 to exponent bias for scaling
 185 // If double-extended denormal form x * 2^64 which is normal
 186 { .mfi
 187 (p8)  add             rExpBias = 64, rExpBias
 188 (p8)  fmpy.s1         fNorm_x = fNorm_x, f2to64
 189       nop.i           0
 190 }
 191 ;;
 192
 193 // Logic is the same as normal path but use normalized input
 194 { .mib
 195       getf.exp        rSignexp_x = fNorm_x
 196       nop.i           0
 197       br.cond.sptk    LOGB_COMMON              // Return to main path
 198 }
 199 ;;
 200
 201 LOGB_ZERO:
 202 // Here if x zero
 203 // f10  = -|f8|
 204 // f9  = 1.0/f10 = -1.0/|f8| = -inf
 205
 206 { .mmf
 207       alloc           r32=ar.pfs,1,2,4,0
 208       mov             GR_Parameter_TAG = 152  // Error code
 209       fmerge.ns       f10 = f0,f8
 210 }
 211 ;;
 212
 213 { .mfb
 214       nop.m           0
 215       frcpa.s0        f9,p6 = f1,f10          // Produce -inf, Z flag
 216       br.cond.sptk    __libm_error_region     // Call error support
 217 }
 218 ;;
 219
 220 GLOBAL_LIBM_END(logbf)
 221
 222
 223 LOCAL_LIBM_ENTRY(__libm_error_region)
 224 .prologue
 225
 226 { .mfi
 227         add   GR_Parameter_Y=-32,sp           // Parameter 2 value
 228         nop.f 0
 229 .save   ar.pfs,GR_SAVE_PFS
 230         mov  GR_SAVE_PFS=ar.pfs               // Save ar.pfs
 231 }
 232 { .mfi
 233 .fframe 64
 234         add sp=-64,sp                         // Create new stack
 235         nop.f 0
 236         mov GR_SAVE_GP=gp                     // Save gp
 237 };;
 238
 239 { .mmi
 240         stfs [GR_Parameter_Y] = f0,16         // STORE Parameter 2 on stack
 241         add GR_Parameter_X    = 16,sp         // Parameter 1 address
 242 .save   b0, GR_SAVE_B0
 243         mov GR_SAVE_B0=b0                     // Save b0
 244 };;
 245
 246 .body
 247 { .mib
 248         stfs [GR_Parameter_X] = f8            // STORE Parameter 1 on stack
 249         add   GR_Parameter_RESULT = 0,GR_Parameter_Y    // Parameter 3 address
 250         nop.b 0
 251 }
 252 { .mib
 253         stfs [GR_Parameter_Y] = f9            // Store Parameter 3 on stack
 254         add   GR_Parameter_Y = -16,GR_Parameter_Y
 255         br.call.sptk b0=__libm_error_support# // Call error handling function
 256 };;
 257
 258 { .mmi
 259         add   GR_Parameter_RESULT = 48,sp
 260         nop.m 0
 261         nop.i 0
 262 };;
 263
 264 { .mmi
 265         ldfs  f8 = [GR_Parameter_RESULT]       // Get return result off stack
 266 .restore sp
 267         add   sp = 64,sp                       // Restore stack pointer
 268         mov   b0 = GR_SAVE_B0                  // Restore return address
 269 };;
 270
 271 { .mib
 272         mov   gp = GR_SAVE_GP                  // Restore gp
 273         mov   ar.pfs = GR_SAVE_PFS             // Restore ar.pfs
 274         br.ret.sptk   b0
 275 };;
 276
 277 LOCAL_LIBM_END(__libm_error_region)
 278
 279
 280 .type   __libm_error_support#,@function
 281 .global __libm_error_support#