|
Packit |
6c4009 |
.file "rintf.s"
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// Copyright (c) 2000 - 2003, Intel Corporation
|
|
Packit |
6c4009 |
// All rights reserved.
|
|
Packit |
6c4009 |
//
|
|
Packit |
6c4009 |
// Contributed 2000 by the Intel Numerics Group, Intel Corporation
|
|
Packit |
6c4009 |
//
|
|
Packit |
6c4009 |
// Redistribution and use in source and binary forms, with or without
|
|
Packit |
6c4009 |
// modification, are permitted provided that the following conditions are
|
|
Packit |
6c4009 |
// met:
|
|
Packit |
6c4009 |
//
|
|
Packit |
6c4009 |
// * Redistributions of source code must retain the above copyright
|
|
Packit |
6c4009 |
// notice, this list of conditions and the following disclaimer.
|
|
Packit |
6c4009 |
//
|
|
Packit |
6c4009 |
// * Redistributions in binary form must reproduce the above copyright
|
|
Packit |
6c4009 |
// notice, this list of conditions and the following disclaimer in the
|
|
Packit |
6c4009 |
// documentation and/or other materials provided with the distribution.
|
|
Packit |
6c4009 |
//
|
|
Packit |
6c4009 |
// * The name of Intel Corporation may not be used to endorse or promote
|
|
Packit |
6c4009 |
// products derived from this software without specific prior written
|
|
Packit |
6c4009 |
// permission.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
Packit |
6c4009 |
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
Packit |
6c4009 |
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
Packit |
6c4009 |
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
|
|
Packit |
6c4009 |
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
|
Packit |
6c4009 |
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
|
Packit |
6c4009 |
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
|
Packit |
6c4009 |
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
|
Packit |
6c4009 |
// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
|
|
Packit |
6c4009 |
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
|
Packit |
6c4009 |
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
Packit |
6c4009 |
//
|
|
Packit |
6c4009 |
// Intel Corporation is the author of this code, and requests that all
|
|
Packit |
6c4009 |
// problem reports or change requests be submitted to it directly at
|
|
Packit |
6c4009 |
// http://www.intel.com/software/products/opensource/libraries/num.htm.
|
|
Packit |
6c4009 |
//
|
|
Packit |
6c4009 |
// History
|
|
Packit |
6c4009 |
//==============================================================
|
|
Packit |
6c4009 |
// 02/02/00 Initial version
|
|
Packit |
6c4009 |
// 02/08/01 Corrected behavior for all rounding modes.
|
|
Packit |
6c4009 |
// 05/20/02 Cleaned up namespace and sf0 syntax
|
|
Packit |
6c4009 |
// 01/20/03 Improved performance
|
|
Packit |
6c4009 |
//==============================================================
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// API
|
|
Packit |
6c4009 |
//==============================================================
|
|
Packit |
6c4009 |
// float rintf(float x)
|
|
Packit |
6c4009 |
//==============================================================
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// general input registers:
|
|
Packit |
6c4009 |
// r14 - r21
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
rSignexp = r14
|
|
Packit |
6c4009 |
rExp = r15
|
|
Packit |
6c4009 |
rExpMask = r16
|
|
Packit |
6c4009 |
rBigexp = r17
|
|
Packit |
6c4009 |
rM1 = r18
|
|
Packit |
6c4009 |
rFpsr = r19
|
|
Packit |
6c4009 |
rRcs0 = r20
|
|
Packit |
6c4009 |
rRcs0Mask = r21
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// floating-point registers:
|
|
Packit |
6c4009 |
// f8 - f11
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
fXInt = f9
|
|
Packit |
6c4009 |
fNormX = f10
|
|
Packit |
6c4009 |
fTmp = f11
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// predicate registers used:
|
|
Packit |
6c4009 |
// p6 - p10
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// Overview of operation
|
|
Packit |
6c4009 |
//==============================================================
|
|
Packit |
6c4009 |
// float rintf(float x)
|
|
Packit |
6c4009 |
// Return an integer value (represented as a float) that is x
|
|
Packit |
6c4009 |
// rounded to integer in current rounding mode
|
|
Packit |
6c4009 |
// Inexact is set if x != rint(x)
|
|
Packit |
6c4009 |
//==============================================================
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// double_extended
|
|
Packit |
6c4009 |
// if the exponent is > 1003e => 3F(true) = 63(decimal)
|
|
Packit |
6c4009 |
// we have a significand of 64 bits 1.63-bits.
|
|
Packit |
6c4009 |
// If we multiply by 2^63, we no longer have a fractional part
|
|
Packit |
6c4009 |
// So input is an integer value already.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// double
|
|
Packit |
6c4009 |
// if the exponent is >= 10033 => 34(true) = 52(decimal)
|
|
Packit |
6c4009 |
// 34 + 3ff = 433
|
|
Packit |
6c4009 |
// we have a significand of 53 bits 1.52-bits. (implicit 1)
|
|
Packit |
6c4009 |
// If we multiply by 2^52, we no longer have a fractional part
|
|
Packit |
6c4009 |
// So input is an integer value already.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// single
|
|
Packit |
6c4009 |
// if the exponent is > 10016 => 17(true) = 23(decimal)
|
|
Packit |
6c4009 |
// we have a significand of 24 bits 1.23-bits. (implicit 1)
|
|
Packit |
6c4009 |
// If we multiply by 2^23, we no longer have a fractional part
|
|
Packit |
6c4009 |
// So input is an integer value already.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
.section .text
|
|
Packit |
6c4009 |
GLOBAL_IEEE754_ENTRY(rintf)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
getf.exp rSignexp = f8 // Get signexp, recompute if unorm
|
|
Packit |
6c4009 |
fclass.m p7,p0 = f8, 0x0b // Test x unorm
|
|
Packit |
6c4009 |
addl rBigexp = 0x10016, r0 // Set exponent at which is integer
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
mov rM1 = -1 // Set all ones
|
|
Packit |
6c4009 |
fcvt.fx.s1 fXInt = f8 // Convert to int in significand
|
|
Packit |
6c4009 |
mov rExpMask = 0x1FFFF // Form exponent mask
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
mov rFpsr = ar40 // Read fpsr -- check rc.s0
|
|
Packit |
6c4009 |
fclass.m p6,p0 = f8, 0x1e3 // Test x natval, nan, inf
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
{ .mfb
|
|
Packit |
6c4009 |
setf.sig fTmp = rM1 // Make const for setting inexact
|
|
Packit |
6c4009 |
fnorm.s1 fNormX = f8 // Normalize input
|
|
Packit |
6c4009 |
(p7) br.cond.spnt RINT_UNORM // Branch if x unorm
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
RINT_COMMON:
|
|
Packit |
6c4009 |
// Return here from RINT_UNORM
|
|
Packit |
6c4009 |
{ .mfb
|
|
Packit |
6c4009 |
and rExp = rSignexp, rExpMask // Get biased exponent
|
|
Packit |
6c4009 |
(p6) fma.s.s0 f8 = f8, f1, f0 // Result if x natval, nan, inf
|
|
Packit |
6c4009 |
(p6) br.ret.spnt b0 // Exit if x natval, nan, inf
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
mov rRcs0Mask = 0x0c00 // Mask for rc.s0
|
|
Packit |
6c4009 |
fcvt.xf f8 = fXInt // Result assume |x| < 2^23
|
|
Packit |
6c4009 |
cmp.ge p7,p8 = rExp, rBigexp // Is |x| >= 2^23?
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// We must correct result if |x| >= 2^23
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
(p7) fma.s.s0 f8 = fNormX, f1, f0 // If |x| >= 2^23, result x
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
fcmp.eq.unc.s1 p0, p9 = f8, fNormX // Is result = x ?
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
(p8) fmerge.s f8 = fNormX, f8 // Make sure sign rint(x) = sign x
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
(p8) and rRcs0 = rFpsr, rRcs0Mask // Get rounding mode for sf0
|
|
Packit |
6c4009 |
nop.f 0
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// If |x| < 2^23 we must test for other rounding modes
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
(p8) cmp.ne.unc p10,p0 = rRcs0, r0 // Test for other rounding modes
|
|
Packit |
6c4009 |
(p9) fmpy.s0 fTmp = fTmp, fTmp // Dummy to set inexact
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
{ .mbb
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
(p10) br.cond.spnt RINT_NOT_ROUND_NEAREST // Branch if not round nearest
|
|
Packit |
6c4009 |
br.ret.sptk b0 // Exit main path if round nearest
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
RINT_UNORM:
|
|
Packit |
6c4009 |
// Here if x unorm
|
|
Packit |
6c4009 |
{ .mfb
|
|
Packit |
6c4009 |
getf.exp rSignexp = fNormX // Get signexp, recompute if unorm
|
|
Packit |
6c4009 |
fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag
|
|
Packit |
6c4009 |
br.cond.sptk RINT_COMMON // Return to main path
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
RINT_NOT_ROUND_NEAREST:
|
|
Packit |
6c4009 |
// Here if not round to nearest, and |x| < 2^23
|
|
Packit |
6c4009 |
// Set rounding mode of s2 to that of s0, and repeat the conversion using s2
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
fsetc.s2 0x7f, 0x40
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
fcvt.fx.s2 fXInt = fNormX // Convert to int in significand
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
{ .mfi
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
fcvt.xf f8 = fXInt // Expected result
|
|
Packit |
6c4009 |
nop.i 0
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
// Be sure sign of result = sign of input. Fixes cases where result is 0.
|
|
Packit |
6c4009 |
{ .mfb
|
|
Packit |
6c4009 |
nop.m 0
|
|
Packit |
6c4009 |
fmerge.s f8 = fNormX, f8
|
|
Packit |
6c4009 |
br.ret.sptk b0 // Exit main path
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
;;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
GLOBAL_IEEE754_END(rintf)
|
|
Packit |
6c4009 |
libm_alias_float_other (__rint, rint)
|