qt5base-lts/util/qfloat16-tables/gen_qfloat16_tables.cpp
Allan Sandfeld Jensen d3ff95dcb8 Round float->qfloat16 to even
Calibrated to match F16C and ARM-FP16 hardware conversions.

Change-Id: I3bdd4d3db3046fee4aeb24e4ce8b9bc9a06e0397
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
2020-09-06 12:35:12 +02:00

192 lines
7.0 KiB
C++

/****************************************************************************
**
** Copyright (C) 2016 by Southwest Research Institute (R)
** Copyright (C) 2019 Intel Corporation.
** Copyright (C) 2020 The Qt Company Ltd.
** Contact: http://www.qt-project.org/legal
**
** This file is part of the QtCore module of the Qt Toolkit.
**
** $QT_BEGIN_LICENSE:LGPL$
** Commercial License Usage
** Licensees holding valid commercial Qt licenses may use this file in
** accordance with the commercial license agreement provided with the
** Software or, alternatively, in accordance with the terms contained in
** a written agreement between you and The Qt Company. For licensing terms
** and conditions see https://www.qt.io/terms-conditions. For further
** information use the contact form at https://www.qt.io/contact-us.
**
** GNU Lesser General Public License Usage
** Alternatively, this file may be used under the terms of the GNU Lesser
** General Public License version 3 as published by the Free Software
** Foundation and appearing in the file LICENSE.LGPL3 included in the
** packaging of this file. Please review the following information to
** ensure the GNU Lesser General Public License version 3 requirements
** will be met: https://www.gnu.org/licenses/lgpl-3.0.html.
**
** GNU General Public License Usage
** Alternatively, this file may be used under the terms of the GNU
** General Public License version 2.0 or (at your option) the GNU General
** Public license version 3 or any later version approved by the KDE Free
** Qt Foundation. The licenses are as published by the Free Software
** Foundation and appearing in the file LICENSE.GPL2 and LICENSE.GPL3
** included in the packaging of this file. Please review the following
** information to ensure the GNU General Public License requirements will
** be met: https://www.gnu.org/licenses/gpl-2.0.html and
** https://www.gnu.org/licenses/gpl-3.0.html.
**
** $QT_END_LICENSE$
**
****************************************************************************/
#include <stdint.h>
#include <stdio.h>
/*
* This tool generates the tables used by qfloat16 to implement a
* software-emulated version of IEEE 754 binary16. qfloat16 automatically uses
* CPU instructions to convert to and from float (IEEE 754 binary32), but if
* the CPU is not guaranteed to have those instructions available at compile
* time, then qfloat16 needs the tables to perform the conversion with
* reasonable performance.
*
* Because Qt requires float to be IEEE 754 binary32, these tables are
* platform-independent and will never change.
*/
uint32_t convertmantissa(int32_t i)
{
uint32_t m = i << 13; // Zero pad mantissa bits
uint32_t e = 0; // Zero exponent
while (!(m & 0x00800000)) { // While not normalized
e -= 0x00800000; // Decrement exponent (1<<23)
m <<= 1; // Shift mantissa
}
m &= ~0x00800000; // Clear leading 1 bit
e += 0x38800000; // Adjust bias ((127-14)<<23)
return m | e; // Return combined number
}
// we first build these tables up and then print them out as a separate step in order
// to more closely map the implementation given in the paper.
uint32_t basetable[512];
uint32_t shifttable[512];
uint32_t roundtable[512];
int main()
{
uint32_t i;
printf("/* This file was generated by util/qfloat16-tables/gen_qfloat16_tables.cpp */\n\n");
printf("#include <QtCore/qfloat16.h>\n\n");
printf("QT_BEGIN_NAMESPACE\n\n");
printf("#if !defined(__ARM_FP16_FORMAT_IEEE)\n\n");
printf("const quint32 qfloat16::mantissatable[2048] = {\n");
printf("0,\n");
for (i = 1; i < 1024; i++)
printf("0x%XU,\n", convertmantissa(i));
for (i = 1024; i < 2048; i++)
printf("0x%XU,\n", 0x38000000U + ((i - 1024) << 13));
printf("};\n\n");
printf("const quint32 qfloat16::exponenttable[64] = {\n");
printf("0,\n");
for (i = 1; i < 31; i++)
printf("0x%XU,\n", i << 23);
printf("0x47800000U,\n"); // 31
printf("0x80000000U,\n"); // 32
for (i = 33; i < 63; i++)
printf("0x%XU,\n", 0x80000000U + ((i - 32) << 23));
printf("0xC7800000U,\n"); // 63
printf("};\n\n");
printf("const quint32 qfloat16::offsettable[64] = {\n");
printf("0,\n");
for (i = 1; i < 32; i++)
printf("1024U,\n");
printf("0,\n");
for (i = 33; i < 64; i++)
printf("1024U,\n");
printf("};\n\n");
int32_t e;
for (i = 0; i < 256; ++i) {
e = i - 127;
if (e < -25) { // Very small numbers map to zero
basetable[i | 0x000] = 0x0000;
basetable[i | 0x100] = 0x8000;
shifttable[i | 0x000] = 24;
shifttable[i | 0x100] = 24;
roundtable[i | 0x000] = 0;
roundtable[i | 0x100] = 0;
} else if (e < -14) { // Small numbers map to denorms
basetable[i | 0x000] = (0x0400 >> (-e - 14));
basetable[i | 0x100] = (0x0400 >> (-e - 14)) | 0x8000;
shifttable[i | 0x000] = -e - 1;
shifttable[i | 0x100] = -e - 1;
if (e == -25) {
// rounds up
roundtable[i | 0x000] = (1 << 24);
roundtable[i | 0x100] = (1 << 24);
} else if (e == -24) {
// rounds half up
roundtable[i | 0x000] = (1 << 22) + 1;
roundtable[i | 0x100] = (1 << 22) + 1;
} else {
roundtable[i | 0x000] = (1 << (-e - 2));
roundtable[i | 0x100] = (1 << (-e - 2));
}
} else if (e <= 15) { // Normal numbers just lose precision
basetable[i | 0x000] = ((e + 15) << 10);
basetable[i | 0x100] = ((e + 15) << 10) | 0x8000;
shifttable[i | 0x000] = 13;
shifttable[i | 0x100] = 13;
roundtable[i | 0x000] = (1 << 12);
roundtable[i | 0x100] = (1 << 12);
} else if (e < 128) { // Large numbers map to Infinity
basetable[i | 0x000] = 0x7C00;
basetable[i | 0x100] = 0xFC00;
shifttable[i | 0x000] = 24;
shifttable[i | 0x100] = 24;
roundtable[i | 0x000] = 0;
roundtable[i | 0x100] = 0;
} else { // Infinity and NaN's stay Infinity and NaN's
basetable[i | 0x000] = 0x7C00;
basetable[i | 0x100] = 0xFC00;
shifttable[i | 0x000] = 13;
shifttable[i | 0x100] = 13;
roundtable[i | 0x000] = 0;
roundtable[i | 0x100] = 0;
}
}
printf("const quint16 qfloat16::basetable[512] = {\n");
for (i = 0; i < 512; i++)
printf("0x%XU,\n", basetable[i]);
printf("};\n\n");
printf("const quint16 qfloat16::shifttable[512] = {\n");
for (i = 0; i < 512; i++)
printf("0x%XU,\n", shifttable[i]);
printf("};\n\n");
printf("const quint32 qfloat16::roundtable[512] = {\n");
for (i = 0; i < 512; i++)
printf("0x%XU,\n", roundtable[i]);
printf("};\n\n");
printf("#endif // !__ARM_FP16_FORMAT_IEEE\n\n");
printf("QT_END_NAMESPACE\n");
return 0;
}