Sprinter/mame
2
0
Fork 0
mirror of https://github.com/holub/mame synced 2025-04-19 15:11:37 +03:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

Add MAME-ified version of SoftFloat library [John R. Hauser]

Browse Source 
As previously discussed, this will be used for the MC680x0 FPU in 
order to handle high-precision floats portably.  The license is 
included in README.txt and is MAME compatible.
This commit is contained in:
R. Belmont 2010-03-02 04:26:48 +00:00
parent 396b49bf1b
commit beeffeb5e7
10 changed files with 6812 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
.gitattributes vendored
View File

@ -1089,6 +1089,13 @@ src/lib/expat/xmltok_impl.c svneol=native#text/plain
src/lib/expat/xmltok_impl.h svneol=native#text/plain
src/lib/expat/xmltok_ns.c svneol=native#text/plain
src/lib/lib.mak svneol=native#text/plain
src/lib/softfloat/README.txt svneol=native#text/plain
src/lib/softfloat/mamesf.h svneol=native#text/plain
src/lib/softfloat/milieu.h svneol=native#text/plain
src/lib/softfloat/softfloat-macros svneol=native#text/plain
src/lib/softfloat/softfloat-specialize svneol=native#text/plain
src/lib/softfloat/softfloat.c svneol=native#text/plain
src/lib/softfloat/softfloat.h svneol=native#text/plain
src/lib/util/astring.c svneol=native#text/plain
src/lib/util/astring.h svneol=native#text/plain
src/lib/util/avcomp.c svneol=native#text/plain

7
makefile
View File

@ -566,7 +566,8 @@ LIBS += -lz
ZLIB =
endif
# add SoftFloat floating point emulation library
SOFTFLOAT = $(OBJ)/libsoftfloat.a
#-------------------------------------------------
# 'default' target needs to go here, before the
@ -656,9 +657,9 @@ $(sort $(OBJDIRS)):
ifndef EXECUTABLE_DEFINED
# always recompile the version string
$(VERSIONOBJ): $(DRVLIBS) $(LIBOSD) $(LIBEMU) $(LIBCPU) $(LIBSOUND) $(LIBUTIL) $(EXPAT) $(ZLIB) $(LIBOCORE) $(RESFILE)
$(VERSIONOBJ): $(DRVLIBS) $(LIBOSD) $(LIBEMU) $(LIBCPU) $(LIBSOUND) $(LIBUTIL) $(EXPAT) $(ZLIB) $(SOFTFLOAT) $(LIBOCORE) $(RESFILE)
$(EMULATOR): $(VERSIONOBJ) $(DRVLIBS) $(LIBOSD) $(LIBEMU) $(LIBCPU) $(LIBDASM) $(LIBSOUND) $(LIBUTIL) $(EXPAT) $(ZLIB) $(LIBOCORE) $(RESFILE)
$(EMULATOR): $(VERSIONOBJ) $(DRVLIBS) $(LIBOSD) $(LIBEMU) $(LIBCPU) $(LIBDASM) $(LIBSOUND) $(LIBUTIL) $(EXPAT) $(SOFTFLOAT) $(ZLIB) $(LIBOCORE) $(RESFILE)
@echo Linking $@...
$(LD) $(LDFLAGS) $(LDFLAGSEMULATOR) $^ $(LIBS) -o $@

16
src/lib/lib.mak
View File

@ -17,6 +17,7 @@ OBJDIRS += \
$(LIBOBJ)/util \
$(LIBOBJ)/expat \
$(LIBOBJ)/zlib \
$(LIBOBJ)/softfloat \
@ -93,3 +94,18 @@ $(OBJ)/libz.a: $(ZLIBOBJS)
$(LIBOBJ)/zlib/%.o: $(LIBSRC)/zlib/%.c | $(OSPREBUILD)
@echo Compiling $<...
$(CC) $(CDEFS) $(CCOMFLAGS) $(CONLYFLAGS) -c $< -o $@
#-------------------------------------------------
# SoftFloat library objects
#-------------------------------------------------
PROCESSOR_H = $(LIBSRC)/softfloat/processors/mamesf.h
SOFTFLOAT_MACROS = $(LIBSRC)/softfloat/softfloat/bits64/softfloat-macros
SOFTFLOATOBJS = \
$(LIBOBJ)/softfloat/softfloat.o
$(OBJ)/libsoftfloat.a: $(SOFTFLOATOBJS)
$(LIBOBJ)/softfloat/softfloat.o: $(LIBSRC)/softfloat/softfloat.c $(LIBSRC)/softfloat/softfloat.h $(LIBSRC)/softfloat/softfloat-macros $(LIBSRC)/softfloat/softfloat-specialize

78
src/lib/softfloat/README.txt Normal file
View File

@ -0,0 +1,78 @@
MAME note: this package is derived from the following original SoftFloat
package and has been "re-packaged" to work with MAME's conventions and
build system. The source files come from bits64/ and bits64/templates
in the original distribution as MAME requires a compiler with a 64-bit
integer type.
Package Overview for SoftFloat Release 2b
John R. Hauser
2002 May 27
----------------------------------------------------------------------------
Overview
SoftFloat is a software implementation of floating-point that conforms to
the IEC/IEEE Standard for Binary Floating-Point Arithmetic. SoftFloat is
distributed in the form of C source code. Compiling the SoftFloat sources
generates two things:
-- A SoftFloat object file (typically `softfloat.o') containing the complete
set of IEC/IEEE floating-point routines.
-- A `timesoftfloat' program for evaluating the speed of the SoftFloat
routines. (The SoftFloat module is linked into this program.)
The SoftFloat package is documented in four text files:
SoftFloat.txt Documentation for using the SoftFloat functions.
SoftFloat-source.txt Documentation for compiling SoftFloat.
SoftFloat-history.txt History of major changes to SoftFloat.
timesoftfloat.txt Documentation for using `timesoftfloat'.
Other files in the package comprise the source code for SoftFloat.
Please be aware that some work is involved in porting this software to other
targets. It is not just a matter of getting `make' to complete without
error messages. I would have written the code that way if I could, but
there are fundamental differences between systems that can't be hidden.
You should not attempt to compile SoftFloat without first reading both
`SoftFloat.txt' and `SoftFloat-source.txt'.
----------------------------------------------------------------------------
Legal Notice
SoftFloat was written by me, John R. Hauser. This work was made possible in
part by the International Computer Science Institute, located at Suite 600,
1947 Center Street, Berkeley, California 94704. Funding was partially
provided by the National Science Foundation under grant MIP-9311980. The
original version of this code was written as part of a project to build
a fixed-point vector processor in collaboration with the University of
California at Berkeley, overseen by Profs. Nelson Morgan and John Wawrzynek.
THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort
has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO
PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL
LOSSES, COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO
FURTHERMORE EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER
SCIENCE INSTITUTE (possibly via similar legal warning) AGAINST ALL LOSSES,
COSTS, OR OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE
SOFTWARE.
Derivative works are acceptable, even for commercial purposes, provided
that the minimal documentation requirements stated in the source code are
satisfied.
----------------------------------------------------------------------------
Contact Information
At the time of this writing, the most up-to-date information about
SoftFloat and the latest release can be found at the Web page `http://
www.cs.berkeley.edu/~jhauser/arithmetic/SoftFloat.html'.

66
src/lib/softfloat/mamesf.h Normal file
View File

@ -0,0 +1,66 @@
/*----------------------------------------------------------------------------
| One of the macros `BIGENDIAN' or `LITTLEENDIAN' must be defined.
*----------------------------------------------------------------------------*/
#ifdef LSB_FIRST
#define LITTLEENDIAN
#else
#define BIGENDIAN
#endif
/*----------------------------------------------------------------------------
| The macro `BITS64' can be defined to indicate that 64-bit integer types are
| supported by the compiler.
*----------------------------------------------------------------------------*/
#define BITS64
/*----------------------------------------------------------------------------
| Each of the following `typedef's defines the most convenient type that holds
| integers of at least as many bits as specified. For example, `uint8' should
| be the most convenient type that can hold unsigned integers of as many as
| 8 bits. The `flag' type must be able to hold either a 0 or 1. For most
| implementations of C, `flag', `uint8', and `int8' should all be `typedef'ed
| to the same as `int'.
*----------------------------------------------------------------------------*/
#include "emu.h"
typedef INT8 flag;
typedef UINT8 uint8;
typedef INT8 int8;
typedef UINT16 uint16;
typedef INT16 int16;
typedef UINT32 uint32;
typedef INT32 int32;
typedef UINT64 uint64;
typedef INT64 int64;
/*----------------------------------------------------------------------------
| Each of the following `typedef's defines a type that holds integers
| of _exactly_ the number of bits specified. For instance, for most
| implementation of C, `bits16' and `sbits16' should be `typedef'ed to
| `unsigned short int' and `signed short int' (or `short int'), respectively.
*----------------------------------------------------------------------------*/
typedef UINT8 bits8;
typedef INT8 sbits8;
typedef UINT16 bits16;
typedef INT16 sbits16;
typedef UINT32 bits32;
typedef INT32 sbits32;
typedef UINT64 bits64;
typedef INT64 sbits64;
/*----------------------------------------------------------------------------
| The `LIT64' macro takes as its argument a textual integer literal and
| if necessary ``marks'' the literal as having a 64-bit integer type.
| For example, the GNU C Compiler (`gcc') requires that 64-bit literals be
| appended with the letters `LL' standing for `long long', which is `gcc's
| name for the 64-bit integer type. Some compilers may allow `LIT64' to be
| defined as the identity macro: `#define LIT64( a ) a'.
*----------------------------------------------------------------------------*/
#define LIT64( a ) a##LL
/*----------------------------------------------------------------------------
| The macro `INLINE' can be used before functions that should be inlined. If
| a compiler does not support explicit inlining, this macro should be defined
| to be `static'.
*----------------------------------------------------------------------------*/
// MAME defines INLINE

42
src/lib/softfloat/milieu.h Normal file
View File

@ -0,0 +1,42 @@
/*============================================================================
This C header file is part of the SoftFloat IEC/IEEE Floating-point Arithmetic
Package, Release 2b.
Written by John R. Hauser. This work was made possible in part by the
International Computer Science Institute, located at Suite 600, 1947 Center
Street, Berkeley, California 94704. Funding was partially provided by the
National Science Foundation under grant MIP-9311980. The original version
of this code was written as part of a project to build a fixed-point vector
processor in collaboration with the University of California at Berkeley,
overseen by Profs. Nelson Morgan and John Wawrzynek. More information
is available through the Web page `http://www.cs.berkeley.edu/~jhauser/
arithmetic/SoftFloat.html'.
THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has
been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES
RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS
AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL LOSSES,
COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO FURTHERMORE
EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER SCIENCE
INSTITUTE (possibly via similar legal warning) AGAINST ALL LOSSES, COSTS, OR
OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE SOFTWARE.
Derivative works are acceptable, even for commercial purposes, so long as
(1) the source code for the derivative work includes prominent notice that
the work is derivative, and (2) the source code includes prominent notice with
these four paragraphs for those parts of this code that are retained.
=============================================================================*/
/*----------------------------------------------------------------------------
| Include common integer types and flags.
*----------------------------------------------------------------------------*/
#include "mamesf.h"
/*----------------------------------------------------------------------------
| Symbolic Boolean literals.
*----------------------------------------------------------------------------*/
#define FALSE 0
#define TRUE 1

720
src/lib/softfloat/softfloat-macros Normal file
View File

@ -0,0 +1,720 @@
/*============================================================================
This C source fragment is part of the SoftFloat IEC/IEEE Floating-point
Arithmetic Package, Release 2b.
Written by John R. Hauser. This work was made possible in part by the
International Computer Science Institute, located at Suite 600, 1947 Center
Street, Berkeley, California 94704. Funding was partially provided by the
National Science Foundation under grant MIP-9311980. The original version
of this code was written as part of a project to build a fixed-point vector
processor in collaboration with the University of California at Berkeley,
overseen by Profs. Nelson Morgan and John Wawrzynek. More information
is available through the Web page `http://www.cs.berkeley.edu/~jhauser/
arithmetic/SoftFloat.html'.
THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has
been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES
RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS
AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL LOSSES,
COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO FURTHERMORE
EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER SCIENCE
INSTITUTE (possibly via similar legal notice) AGAINST ALL LOSSES, COSTS, OR
OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE SOFTWARE.
Derivative works are acceptable, even for commercial purposes, so long as
(1) the source code for the derivative work includes prominent notice that
the work is derivative, and (2) the source code includes prominent notice with
these four paragraphs for those parts of this code that are retained.
=============================================================================*/
/*----------------------------------------------------------------------------
| Shifts `a' right by the number of bits given in `count'. If any nonzero
| bits are shifted off, they are ``jammed'' into the least significant bit of
| the result by setting the least significant bit to 1. The value of `count'
| can be arbitrarily large; in particular, if `count' is greater than 32, the
| result will be either 0 or 1, depending on whether `a' is zero or nonzero.
| The result is stored in the location pointed to by `zPtr'.
*----------------------------------------------------------------------------*/
INLINE void shift32RightJamming( bits32 a, int16 count, bits32 *zPtr )
{
bits32 z;
if ( count == 0 ) {
z = a;
}
else if ( count < 32 ) {
z = ( a>>count ) | ( ( a<<( ( - count ) & 31 ) ) != 0 );
}
else {
z = ( a != 0 );
}
*zPtr = z;
}
/*----------------------------------------------------------------------------
| Shifts `a' right by the number of bits given in `count'. If any nonzero
| bits are shifted off, they are ``jammed'' into the least significant bit of
| the result by setting the least significant bit to 1. The value of `count'
| can be arbitrarily large; in particular, if `count' is greater than 64, the
| result will be either 0 or 1, depending on whether `a' is zero or nonzero.
| The result is stored in the location pointed to by `zPtr'.
*----------------------------------------------------------------------------*/
INLINE void shift64RightJamming( bits64 a, int16 count, bits64 *zPtr )
{
bits64 z;
if ( count == 0 ) {
z = a;
}
else if ( count < 64 ) {
z = ( a>>count ) | ( ( a<<( ( - count ) & 63 ) ) != 0 );
}
else {
z = ( a != 0 );
}
*zPtr = z;
}
/*----------------------------------------------------------------------------
| Shifts the 128-bit value formed by concatenating `a0' and `a1' right by 64
| _plus_ the number of bits given in `count'. The shifted result is at most
| 64 nonzero bits; this is stored at the location pointed to by `z0Ptr'. The
| bits shifted off form a second 64-bit result as follows: The _last_ bit
| shifted off is the most-significant bit of the extra result, and the other
| 63 bits of the extra result are all zero if and only if _all_but_the_last_
| bits shifted off were all zero. This extra result is stored in the location
| pointed to by `z1Ptr'. The value of `count' can be arbitrarily large.
| (This routine makes more sense if `a0' and `a1' are considered to form
| a fixed-point value with binary point between `a0' and `a1'. This fixed-
| point value is shifted right by the number of bits given in `count', and
| the integer part of the result is returned at the location pointed to by
| `z0Ptr'. The fractional part of the result may be slightly corrupted as
| described above, and is returned at the location pointed to by `z1Ptr'.)
*----------------------------------------------------------------------------*/
INLINE void
shift64ExtraRightJamming(
bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr )
{
bits64 z0, z1;
int8 negCount = ( - count ) & 63;
if ( count == 0 ) {
z1 = a1;
z0 = a0;
}
else if ( count < 64 ) {
z1 = ( a0<<negCount ) | ( a1 != 0 );
z0 = a0>>count;
}
else {
if ( count == 64 ) {
z1 = a0 | ( a1 != 0 );
}
else {
z1 = ( ( a0 | a1 ) != 0 );
}
z0 = 0;
}
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Shifts the 128-bit value formed by concatenating `a0' and `a1' right by the
| number of bits given in `count'. Any bits shifted off are lost. The value
| of `count' can be arbitrarily large; in particular, if `count' is greater
| than 128, the result will be 0. The result is broken into two 64-bit pieces
| which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
shift128Right(
bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr )
{
bits64 z0, z1;
int8 negCount = ( - count ) & 63;
if ( count == 0 ) {
z1 = a1;
z0 = a0;
}
else if ( count < 64 ) {
z1 = ( a0<<negCount ) | ( a1>>count );
z0 = a0>>count;
}
else {
z1 = ( count < 64 ) ? ( a0>>( count & 63 ) ) : 0;
z0 = 0;
}
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Shifts the 128-bit value formed by concatenating `a0' and `a1' right by the
| number of bits given in `count'. If any nonzero bits are shifted off, they
| are ``jammed'' into the least significant bit of the result by setting the
| least significant bit to 1. The value of `count' can be arbitrarily large;
| in particular, if `count' is greater than 128, the result will be either
| 0 or 1, depending on whether the concatenation of `a0' and `a1' is zero or
| nonzero. The result is broken into two 64-bit pieces which are stored at
| the locations pointed to by `z0Ptr' and `z1Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
shift128RightJamming(
bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr )
{
bits64 z0, z1;
int8 negCount = ( - count ) & 63;
if ( count == 0 ) {
z1 = a1;
z0 = a0;
}
else if ( count < 64 ) {
z1 = ( a0<<negCount ) | ( a1>>count ) | ( ( a1<<negCount ) != 0 );
z0 = a0>>count;
}
else {
if ( count == 64 ) {
z1 = a0 | ( a1 != 0 );
}
else if ( count < 128 ) {
z1 = ( a0>>( count & 63 ) ) | ( ( ( a0<<negCount ) | a1 ) != 0 );
}
else {
z1 = ( ( a0 | a1 ) != 0 );
}
z0 = 0;
}
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Shifts the 192-bit value formed by concatenating `a0', `a1', and `a2' right
| by 64 _plus_ the number of bits given in `count'. The shifted result is
| at most 128 nonzero bits; these are broken into two 64-bit pieces which are
| stored at the locations pointed to by `z0Ptr' and `z1Ptr'. The bits shifted
| off form a third 64-bit result as follows: The _last_ bit shifted off is
| the most-significant bit of the extra result, and the other 63 bits of the
| extra result are all zero if and only if _all_but_the_last_ bits shifted off
| were all zero. This extra result is stored in the location pointed to by
| `z2Ptr'. The value of `count' can be arbitrarily large.
| (This routine makes more sense if `a0', `a1', and `a2' are considered
| to form a fixed-point value with binary point between `a1' and `a2'. This
| fixed-point value is shifted right by the number of bits given in `count',
| and the integer part of the result is returned at the locations pointed to
| by `z0Ptr' and `z1Ptr'. The fractional part of the result may be slightly
| corrupted as described above, and is returned at the location pointed to by
| `z2Ptr'.)
*----------------------------------------------------------------------------*/
INLINE void
shift128ExtraRightJamming(
bits64 a0,
bits64 a1,
bits64 a2,
int16 count,
bits64 *z0Ptr,
bits64 *z1Ptr,
bits64 *z2Ptr
)
{
bits64 z0, z1, z2;
int8 negCount = ( - count ) & 63;
if ( count == 0 ) {
z2 = a2;
z1 = a1;
z0 = a0;
}
else {
if ( count < 64 ) {
z2 = a1<<negCount;
z1 = ( a0<<negCount ) | ( a1>>count );
z0 = a0>>count;
}
else {
if ( count == 64 ) {
z2 = a1;
z1 = a0;
}
else {
a2 |= a1;
if ( count < 128 ) {
z2 = a0<<negCount;
z1 = a0>>( count & 63 );
}
else {
z2 = ( count == 128 ) ? a0 : ( a0 != 0 );
z1 = 0;
}
}
z0 = 0;
}
z2 |= ( a2 != 0 );
}
*z2Ptr = z2;
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Shifts the 128-bit value formed by concatenating `a0' and `a1' left by the
| number of bits given in `count'. Any bits shifted off are lost. The value
| of `count' must be less than 64. The result is broken into two 64-bit
| pieces which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
shortShift128Left(
bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr )
{
*z1Ptr = a1<<count;
*z0Ptr =
( count == 0 ) ? a0 : ( a0<<count ) | ( a1>>( ( - count ) & 63 ) );
}
/*----------------------------------------------------------------------------
| Shifts the 192-bit value formed by concatenating `a0', `a1', and `a2' left
| by the number of bits given in `count'. Any bits shifted off are lost.
| The value of `count' must be less than 64. The result is broken into three
| 64-bit pieces which are stored at the locations pointed to by `z0Ptr',
| `z1Ptr', and `z2Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
shortShift192Left(
bits64 a0,
bits64 a1,
bits64 a2,
int16 count,
bits64 *z0Ptr,
bits64 *z1Ptr,
bits64 *z2Ptr
)
{
bits64 z0, z1, z2;
int8 negCount;
z2 = a2<<count;
z1 = a1<<count;
z0 = a0<<count;
if ( 0 < count ) {
negCount = ( ( - count ) & 63 );
z1 |= a2>>negCount;
z0 |= a1>>negCount;
}
*z2Ptr = z2;
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Adds the 128-bit value formed by concatenating `a0' and `a1' to the 128-bit
| value formed by concatenating `b0' and `b1'. Addition is modulo 2^128, so
| any carry out is lost. The result is broken into two 64-bit pieces which
| are stored at the locations pointed to by `z0Ptr' and `z1Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
add128(
bits64 a0, bits64 a1, bits64 b0, bits64 b1, bits64 *z0Ptr, bits64 *z1Ptr )
{
bits64 z1;
z1 = a1 + b1;
*z1Ptr = z1;
*z0Ptr = a0 + b0 + ( z1 < a1 );
}
/*----------------------------------------------------------------------------
| Adds the 192-bit value formed by concatenating `a0', `a1', and `a2' to the
| 192-bit value formed by concatenating `b0', `b1', and `b2'. Addition is
| modulo 2^192, so any carry out is lost. The result is broken into three
| 64-bit pieces which are stored at the locations pointed to by `z0Ptr',
| `z1Ptr', and `z2Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
add192(
bits64 a0,
bits64 a1,
bits64 a2,
bits64 b0,
bits64 b1,
bits64 b2,
bits64 *z0Ptr,
bits64 *z1Ptr,
bits64 *z2Ptr
)
{
bits64 z0, z1, z2;
int8 carry0, carry1;
z2 = a2 + b2;
carry1 = ( z2 < a2 );
z1 = a1 + b1;
carry0 = ( z1 < a1 );
z0 = a0 + b0;
z1 += carry1;
z0 += ( z1 < carry1 );
z0 += carry0;
*z2Ptr = z2;
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Subtracts the 128-bit value formed by concatenating `b0' and `b1' from the
| 128-bit value formed by concatenating `a0' and `a1'. Subtraction is modulo
| 2^128, so any borrow out (carry out) is lost. The result is broken into two
| 64-bit pieces which are stored at the locations pointed to by `z0Ptr' and
| `z1Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
sub128(
bits64 a0, bits64 a1, bits64 b0, bits64 b1, bits64 *z0Ptr, bits64 *z1Ptr )
{
*z1Ptr = a1 - b1;
*z0Ptr = a0 - b0 - ( a1 < b1 );
}
/*----------------------------------------------------------------------------
| Subtracts the 192-bit value formed by concatenating `b0', `b1', and `b2'
| from the 192-bit value formed by concatenating `a0', `a1', and `a2'.
| Subtraction is modulo 2^192, so any borrow out (carry out) is lost. The
| result is broken into three 64-bit pieces which are stored at the locations
| pointed to by `z0Ptr', `z1Ptr', and `z2Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
sub192(
bits64 a0,
bits64 a1,
bits64 a2,
bits64 b0,
bits64 b1,
bits64 b2,
bits64 *z0Ptr,
bits64 *z1Ptr,
bits64 *z2Ptr
)
{
bits64 z0, z1, z2;
int8 borrow0, borrow1;
z2 = a2 - b2;
borrow1 = ( a2 < b2 );
z1 = a1 - b1;
borrow0 = ( a1 < b1 );
z0 = a0 - b0;
z0 -= ( z1 < borrow1 );
z1 -= borrow1;
z0 -= borrow0;
*z2Ptr = z2;
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Multiplies `a' by `b' to obtain a 128-bit product. The product is broken
| into two 64-bit pieces which are stored at the locations pointed to by
| `z0Ptr' and `z1Ptr'.
*----------------------------------------------------------------------------*/
INLINE void mul64To128( bits64 a, bits64 b, bits64 *z0Ptr, bits64 *z1Ptr )
{
bits32 aHigh, aLow, bHigh, bLow;
bits64 z0, zMiddleA, zMiddleB, z1;
aLow = a;
aHigh = a>>32;
bLow = b;
bHigh = b>>32;
z1 = ( (bits64) aLow ) * bLow;
zMiddleA = ( (bits64) aLow ) * bHigh;
zMiddleB = ( (bits64) aHigh ) * bLow;
z0 = ( (bits64) aHigh ) * bHigh;
zMiddleA += zMiddleB;
z0 += ( ( (bits64) ( zMiddleA < zMiddleB ) )<<32 ) + ( zMiddleA>>32 );
zMiddleA <<= 32;
z1 += zMiddleA;
z0 += ( z1 < zMiddleA );
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Multiplies the 128-bit value formed by concatenating `a0' and `a1' by
| `b' to obtain a 192-bit product. The product is broken into three 64-bit
| pieces which are stored at the locations pointed to by `z0Ptr', `z1Ptr', and
| `z2Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
mul128By64To192(
bits64 a0,
bits64 a1,
bits64 b,
bits64 *z0Ptr,
bits64 *z1Ptr,
bits64 *z2Ptr
)
{
bits64 z0, z1, z2, more1;
mul64To128( a1, b, &z1, &z2 );
mul64To128( a0, b, &z0, &more1 );
add128( z0, more1, 0, z1, &z0, &z1 );
*z2Ptr = z2;
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Multiplies the 128-bit value formed by concatenating `a0' and `a1' to the
| 128-bit value formed by concatenating `b0' and `b1' to obtain a 256-bit
| product. The product is broken into four 64-bit pieces which are stored at
| the locations pointed to by `z0Ptr', `z1Ptr', `z2Ptr', and `z3Ptr'.
*----------------------------------------------------------------------------*/
INLINE void
mul128To256(
bits64 a0,
bits64 a1,
bits64 b0,
bits64 b1,
bits64 *z0Ptr,
bits64 *z1Ptr,
bits64 *z2Ptr,
bits64 *z3Ptr
)
{
bits64 z0, z1, z2, z3;
bits64 more1, more2;
mul64To128( a1, b1, &z2, &z3 );
mul64To128( a1, b0, &z1, &more2 );
add128( z1, more2, 0, z2, &z1, &z2 );
mul64To128( a0, b0, &z0, &more1 );
add128( z0, more1, 0, z1, &z0, &z1 );
mul64To128( a0, b1, &more1, &more2 );
add128( more1, more2, 0, z2, &more1, &z2 );
add128( z0, z1, 0, more1, &z0, &z1 );
*z3Ptr = z3;
*z2Ptr = z2;
*z1Ptr = z1;
*z0Ptr = z0;
}
/*----------------------------------------------------------------------------
| Returns an approximation to the 64-bit integer quotient obtained by dividing
| `b' into the 128-bit value formed by concatenating `a0' and `a1'. The
| divisor `b' must be at least 2^63. If q is the exact quotient truncated
| toward zero, the approximation returned lies between q and q + 2 inclusive.
| If the exact quotient q is larger than 64 bits, the maximum positive 64-bit
| unsigned integer is returned.
*----------------------------------------------------------------------------*/
static bits64 estimateDiv128To64( bits64 a0, bits64 a1, bits64 b )
{
bits64 b0, b1;
bits64 rem0, rem1, term0, term1;
bits64 z;
if ( b <= a0 ) return LIT64( 0xFFFFFFFFFFFFFFFF );
b0 = b>>32;
z = ( b0<<32 <= a0 ) ? LIT64( 0xFFFFFFFF00000000 ) : ( a0 / b0 )<<32;
mul64To128( b, z, &term0, &term1 );
sub128( a0, a1, term0, term1, &rem0, &rem1 );
while ( ( (sbits64) rem0 ) < 0 ) {
z -= LIT64( 0x100000000 );
b1 = b<<32;
add128( rem0, rem1, b0, b1, &rem0, &rem1 );
}
rem0 = ( rem0<<32 ) | ( rem1>>32 );
z |= ( b0<<32 <= rem0 ) ? 0xFFFFFFFF : rem0 / b0;
return z;
}
/*----------------------------------------------------------------------------
| Returns an approximation to the square root of the 32-bit significand given
| by `a'. Considered as an integer, `a' must be at least 2^31. If bit 0 of
| `aExp' (the least significant bit) is 1, the integer returned approximates
| 2^31*sqrt(`a'/2^31), where `a' is considered an integer. If bit 0 of `aExp'
| is 0, the integer returned approximates 2^31*sqrt(`a'/2^30). In either
| case, the approximation returned lies strictly within +/-2 of the exact
| value.
*----------------------------------------------------------------------------*/
static bits32 estimateSqrt32( int16 aExp, bits32 a )
{
static const bits16 sqrtOddAdjustments[] = {
0x0004, 0x0022, 0x005D, 0x00B1, 0x011D, 0x019F, 0x0236, 0x02E0,
0x039C, 0x0468, 0x0545, 0x0631, 0x072B, 0x0832, 0x0946, 0x0A67
};
static const bits16 sqrtEvenAdjustments[] = {
0x0A2D, 0x08AF, 0x075A, 0x0629, 0x051A, 0x0429, 0x0356, 0x029E,
0x0200, 0x0179, 0x0109, 0x00AF, 0x0068, 0x0034, 0x0012, 0x0002
};
int8 index;
bits32 z;
index = ( a>>27 ) & 15;
if ( aExp & 1 ) {
z = 0x4000 + ( a>>17 ) - sqrtOddAdjustments[ index ];
z = ( ( a / z )<<14 ) + ( z<<15 );
a >>= 1;
}
else {
z = 0x8000 + ( a>>17 ) - sqrtEvenAdjustments[ index ];
z = a / z + z;
z = ( 0x20000 <= z ) ? 0xFFFF8000 : ( z<<15 );
if ( z <= a ) return (bits32) ( ( (sbits32) a )>>1 );
}
return ( (bits32) ( ( ( (bits64) a )<<31 ) / z ) ) + ( z>>1 );
}
/*----------------------------------------------------------------------------
| Returns the number of leading 0 bits before the most-significant 1 bit of
| `a'. If `a' is zero, 32 is returned.
*----------------------------------------------------------------------------*/
static int8 countLeadingZeros32( bits32 a )
{
static const int8 countLeadingZerosHigh[] = {
8, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
};
int8 shiftCount;
shiftCount = 0;
if ( a < 0x10000 ) {
shiftCount += 16;
a <<= 16;
}
if ( a < 0x1000000 ) {
shiftCount += 8;
a <<= 8;
}
shiftCount += countLeadingZerosHigh[ a>>24 ];
return shiftCount;
}
/*----------------------------------------------------------------------------
| Returns the number of leading 0 bits before the most-significant 1 bit of
| `a'. If `a' is zero, 64 is returned.
*----------------------------------------------------------------------------*/
static int8 countLeadingZeros64( bits64 a )
{
int8 shiftCount;
shiftCount = 0;
if ( a < ( (bits64) 1 )<<32 ) {
shiftCount += 32;
}
else {
a >>= 32;
}
shiftCount += countLeadingZeros32( a );
return shiftCount;
}
/*----------------------------------------------------------------------------
| Returns 1 if the 128-bit value formed by concatenating `a0' and `a1'
| is equal to the 128-bit value formed by concatenating `b0' and `b1'.
| Otherwise, returns 0.
*----------------------------------------------------------------------------*/
INLINE flag eq128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 )
{
return ( a0 == b0 ) && ( a1 == b1 );
}
/*----------------------------------------------------------------------------
| Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is less
| than or equal to the 128-bit value formed by concatenating `b0' and `b1'.
| Otherwise, returns 0.
*----------------------------------------------------------------------------*/
INLINE flag le128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 )
{
return ( a0 < b0 ) || ( ( a0 == b0 ) && ( a1 <= b1 ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is less
| than the 128-bit value formed by concatenating `b0' and `b1'. Otherwise,
| returns 0.
*----------------------------------------------------------------------------*/
INLINE flag lt128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 )
{
return ( a0 < b0 ) || ( ( a0 == b0 ) && ( a1 < b1 ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is
| not equal to the 128-bit value formed by concatenating `b0' and `b1'.
| Otherwise, returns 0.
*----------------------------------------------------------------------------*/
INLINE flag ne128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 )
{
return ( a0 != b0 ) || ( a1 != b1 );
}

432
src/lib/softfloat/softfloat-specialize Normal file
View File

@ -0,0 +1,432 @@
/*============================================================================
This C source fragment is part of the SoftFloat IEC/IEEE Floating-point
Arithmetic Package, Release 2b.
Written by John R. Hauser. This work was made possible in part by the
International Computer Science Institute, located at Suite 600, 1947 Center
Street, Berkeley, California 94704. Funding was partially provided by the
National Science Foundation under grant MIP-9311980. The original version
of this code was written as part of a project to build a fixed-point vector
processor in collaboration with the University of California at Berkeley,
overseen by Profs. Nelson Morgan and John Wawrzynek. More information
is available through the Web page `http://www.cs.berkeley.edu/~jhauser/
arithmetic/SoftFloat.html'.
THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has
been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES
RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS
AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL LOSSES,
COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO FURTHERMORE
EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER SCIENCE
INSTITUTE (possibly via similar legal warning) AGAINST ALL LOSSES, COSTS, OR
OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE SOFTWARE.
Derivative works are acceptable, even for commercial purposes, so long as
(1) the source code for the derivative work includes prominent notice that
the work is derivative, and (2) the source code includes prominent notice with
these four paragraphs for those parts of this code that are retained.
=============================================================================*/
/*----------------------------------------------------------------------------
| Underflow tininess-detection mode, statically initialized to default value.
| (The declaration in `softfloat.h' must match the `int8' type here.)
*----------------------------------------------------------------------------*/
int8 float_detect_tininess = float_tininess_after_rounding;
/*----------------------------------------------------------------------------
| Raises the exceptions specified by `flags'. Floating-point traps can be
| defined here if desired. It is currently not possible for such a trap to
| substitute a result value. If traps are not implemented, this routine
| should be simply `float_exception_flags |= flags;'.
*----------------------------------------------------------------------------*/
void float_raise( int8 flags )
{
float_exception_flags |= flags;
}
/*----------------------------------------------------------------------------
| Internal canonical NaN format.
*----------------------------------------------------------------------------*/
typedef struct {
flag sign;
bits64 high, low;
} commonNaNT;
/*----------------------------------------------------------------------------
| The pattern for a default generated single-precision NaN.
*----------------------------------------------------------------------------*/
#define float32_default_nan 0xFFFFFFFF
/*----------------------------------------------------------------------------
| Returns 1 if the single-precision floating-point value `a' is a NaN;
| otherwise returns 0.
*----------------------------------------------------------------------------*/
flag float32_is_nan( float32 a )
{
return ( 0xFF000000 < (bits32) ( a<<1 ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the single-precision floating-point value `a' is a signaling
| NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
flag float32_is_signaling_nan( float32 a )
{
return ( ( ( a>>22 ) & 0x1FF ) == 0x1FE ) && ( a & 0x003FFFFF );
}
/*----------------------------------------------------------------------------
| Returns the result of converting the single-precision floating-point NaN
| `a' to the canonical NaN format. If `a' is a signaling NaN, the invalid
| exception is raised.
*----------------------------------------------------------------------------*/
static commonNaNT float32ToCommonNaN( float32 a )
{
commonNaNT z;
if ( float32_is_signaling_nan( a ) ) float_raise( float_flag_invalid );
z.sign = a>>31;
z.low = 0;
z.high = ( (bits64) a )<<41;
return z;
}
/*----------------------------------------------------------------------------
| Returns the result of converting the canonical NaN `a' to the single-
| precision floating-point format.
*----------------------------------------------------------------------------*/
static float32 commonNaNToFloat32( commonNaNT a )
{
return ( ( (bits32) a.sign )<<31 ) | 0x7FC00000 | ( a.high>>41 );
}
/*----------------------------------------------------------------------------
| Takes two single-precision floating-point values `a' and `b', one of which
| is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a
| signaling NaN, the invalid exception is raised.
*----------------------------------------------------------------------------*/
static float32 propagateFloat32NaN( float32 a, float32 b )
{
flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN;
aIsNaN = float32_is_nan( a );
aIsSignalingNaN = float32_is_signaling_nan( a );
bIsNaN = float32_is_nan( b );
bIsSignalingNaN = float32_is_signaling_nan( b );
a |= 0x00400000;
b |= 0x00400000;
if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid );
if ( aIsNaN ) {
return ( aIsSignalingNaN & bIsNaN ) ? b : a;
}
else {
return b;
}
}
/*----------------------------------------------------------------------------
| The pattern for a default generated double-precision NaN.
*----------------------------------------------------------------------------*/
#define float64_default_nan LIT64( 0xFFFFFFFFFFFFFFFF )
/*----------------------------------------------------------------------------
| Returns 1 if the double-precision floating-point value `a' is a NaN;
| otherwise returns 0.
*----------------------------------------------------------------------------*/
flag float64_is_nan( float64 a )
{
return ( LIT64( 0xFFE0000000000000 ) < (bits64) ( a<<1 ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the double-precision floating-point value `a' is a signaling
| NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
flag float64_is_signaling_nan( float64 a )
{
return
( ( ( a>>51 ) & 0xFFF ) == 0xFFE )
&& ( a & LIT64( 0x0007FFFFFFFFFFFF ) );
}
/*----------------------------------------------------------------------------
| Returns the result of converting the double-precision floating-point NaN
| `a' to the canonical NaN format. If `a' is a signaling NaN, the invalid
| exception is raised.
*----------------------------------------------------------------------------*/
static commonNaNT float64ToCommonNaN( float64 a )
{
commonNaNT z;
if ( float64_is_signaling_nan( a ) ) float_raise( float_flag_invalid );
z.sign = a>>63;
z.low = 0;
z.high = a<<12;
return z;
}
/*----------------------------------------------------------------------------
| Returns the result of converting the canonical NaN `a' to the double-
| precision floating-point format.
*----------------------------------------------------------------------------*/
static float64 commonNaNToFloat64( commonNaNT a )
{
return
( ( (bits64) a.sign )<<63 )
| LIT64( 0x7FF8000000000000 )
| ( a.high>>12 );
}
/*----------------------------------------------------------------------------
| Takes two double-precision floating-point values `a' and `b', one of which
| is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a
| signaling NaN, the invalid exception is raised.
*----------------------------------------------------------------------------*/
static float64 propagateFloat64NaN( float64 a, float64 b )
{
flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN;
aIsNaN = float64_is_nan( a );
aIsSignalingNaN = float64_is_signaling_nan( a );
bIsNaN = float64_is_nan( b );
bIsSignalingNaN = float64_is_signaling_nan( b );
a |= LIT64( 0x0008000000000000 );
b |= LIT64( 0x0008000000000000 );
if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid );
if ( aIsNaN ) {
return ( aIsSignalingNaN & bIsNaN ) ? b : a;
}
else {
return b;
}
}
#ifdef FLOATX80
/*----------------------------------------------------------------------------
| The pattern for a default generated extended double-precision NaN. The
| `high' and `low' values hold the most- and least-significant bits,
| respectively.
*----------------------------------------------------------------------------*/
#define floatx80_default_nan_high 0xFFFF
#define floatx80_default_nan_low LIT64( 0xFFFFFFFFFFFFFFFF )
/*----------------------------------------------------------------------------
| Returns 1 if the extended double-precision floating-point value `a' is a
| NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
flag floatx80_is_nan( floatx80 a )
{
return ( ( a.high & 0x7FFF ) == 0x7FFF ) && (bits64) ( a.low<<1 );
}
/*----------------------------------------------------------------------------
| Returns 1 if the extended double-precision floating-point value `a' is a
| signaling NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
flag floatx80_is_signaling_nan( floatx80 a )
{
bits64 aLow;
aLow = a.low & ~ LIT64( 0x4000000000000000 );
return
( ( a.high & 0x7FFF ) == 0x7FFF )
&& (bits64) ( aLow<<1 )
&& ( a.low == aLow );
}
/*----------------------------------------------------------------------------
| Returns the result of converting the extended double-precision floating-
| point NaN `a' to the canonical NaN format. If `a' is a signaling NaN, the
| invalid exception is raised.
*----------------------------------------------------------------------------*/
static commonNaNT floatx80ToCommonNaN( floatx80 a )
{
commonNaNT z;
if ( floatx80_is_signaling_nan( a ) ) float_raise( float_flag_invalid );
z.sign = a.high>>15;
z.low = 0;
z.high = a.low<<1;
return z;
}
/*----------------------------------------------------------------------------
| Returns the result of converting the canonical NaN `a' to the extended
| double-precision floating-point format.
*----------------------------------------------------------------------------*/
static floatx80 commonNaNToFloatx80( commonNaNT a )
{
floatx80 z;
z.low = LIT64( 0xC000000000000000 ) | ( a.high>>1 );
z.high = ( ( (bits16) a.sign )<<15 ) | 0x7FFF;
return z;
}
/*----------------------------------------------------------------------------
| Takes two extended double-precision floating-point values `a' and `b', one
| of which is a NaN, and returns the appropriate NaN result. If either `a' or
| `b' is a signaling NaN, the invalid exception is raised.
*----------------------------------------------------------------------------*/
static floatx80 propagateFloatx80NaN( floatx80 a, floatx80 b )
{
flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN;
aIsNaN = floatx80_is_nan( a );
aIsSignalingNaN = floatx80_is_signaling_nan( a );
bIsNaN = floatx80_is_nan( b );
bIsSignalingNaN = floatx80_is_signaling_nan( b );
a.low |= LIT64( 0xC000000000000000 );
b.low |= LIT64( 0xC000000000000000 );
if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid );
if ( aIsNaN ) {
return ( aIsSignalingNaN & bIsNaN ) ? b : a;
}
else {
return b;
}
}
#endif
#ifdef FLOAT128
/*----------------------------------------------------------------------------
| The pattern for a default generated quadruple-precision NaN. The `high' and
| `low' values hold the most- and least-significant bits, respectively.
*----------------------------------------------------------------------------*/
#define float128_default_nan_high LIT64( 0xFFFFFFFFFFFFFFFF )
#define float128_default_nan_low LIT64( 0xFFFFFFFFFFFFFFFF )
/*----------------------------------------------------------------------------
| Returns 1 if the quadruple-precision floating-point value `a' is a NaN;
| otherwise returns 0.
*----------------------------------------------------------------------------*/
flag float128_is_nan( float128 a )
{
return
( LIT64( 0xFFFE000000000000 ) <= (bits64) ( a.high<<1 ) )
&& ( a.low || ( a.high & LIT64( 0x0000FFFFFFFFFFFF ) ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the quadruple-precision floating-point value `a' is a
| signaling NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
flag float128_is_signaling_nan( float128 a )
{
return
( ( ( a.high>>47 ) & 0xFFFF ) == 0xFFFE )
&& ( a.low || ( a.high & LIT64( 0x00007FFFFFFFFFFF ) ) );
}
/*----------------------------------------------------------------------------
| Returns the result of converting the quadruple-precision floating-point NaN
| `a' to the canonical NaN format. If `a' is a signaling NaN, the invalid
| exception is raised.
*----------------------------------------------------------------------------*/
static commonNaNT float128ToCommonNaN( float128 a )
{
commonNaNT z;
if ( float128_is_signaling_nan( a ) ) float_raise( float_flag_invalid );
z.sign = a.high>>63;
shortShift128Left( a.high, a.low, 16, &z.high, &z.low );
return z;
}
/*----------------------------------------------------------------------------
| Returns the result of converting the canonical NaN `a' to the quadruple-
| precision floating-point format.
*----------------------------------------------------------------------------*/
static float128 commonNaNToFloat128( commonNaNT a )
{
float128 z;
shift128Right( a.high, a.low, 16, &z.high, &z.low );
z.high |= ( ( (bits64) a.sign )<<63 ) | LIT64( 0x7FFF800000000000 );
return z;
}
/*----------------------------------------------------------------------------
| Takes two quadruple-precision floating-point values `a' and `b', one of
| which is a NaN, and returns the appropriate NaN result. If either `a' or
| `b' is a signaling NaN, the invalid exception is raised.
*----------------------------------------------------------------------------*/
static float128 propagateFloat128NaN( float128 a, float128 b )
{
flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN;
aIsNaN = float128_is_nan( a );
aIsSignalingNaN = float128_is_signaling_nan( a );
bIsNaN = float128_is_nan( b );
bIsSignalingNaN = float128_is_signaling_nan( b );
a.high |= LIT64( 0x0000800000000000 );
b.high |= LIT64( 0x0000800000000000 );
if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid );
if ( aIsNaN ) {
return ( aIsSignalingNaN & bIsNaN ) ? b : a;
}
else {
return b;
}
}
#endif

5188
src/lib/softfloat/softfloat.c Normal file
View File

File diff suppressed because it is too large Load Diff

259
src/lib/softfloat/softfloat.h Normal file
View File

@ -0,0 +1,259 @@
/*============================================================================
This C header file is part of the SoftFloat IEC/IEEE Floating-point Arithmetic
Package, Release 2b.
Written by John R. Hauser. This work was made possible in part by the
International Computer Science Institute, located at Suite 600, 1947 Center
Street, Berkeley, California 94704. Funding was partially provided by the
National Science Foundation under grant MIP-9311980. The original version
of this code was written as part of a project to build a fixed-point vector
processor in collaboration with the University of California at Berkeley,
overseen by Profs. Nelson Morgan and John Wawrzynek. More information
is available through the Web page `http://www.cs.berkeley.edu/~jhauser/
arithmetic/SoftFloat.html'.
THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has
been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES
RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS
AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL LOSSES,
COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO FURTHERMORE
EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER SCIENCE
INSTITUTE (possibly via similar legal warning) AGAINST ALL LOSSES, COSTS, OR
OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE SOFTWARE.
Derivative works are acceptable, even for commercial purposes, so long as
(1) the source code for the derivative work includes prominent notice that
the work is derivative, and (2) the source code includes prominent notice with
these four paragraphs for those parts of this code that are retained.
=============================================================================*/
/*----------------------------------------------------------------------------
| The macro `FLOATX80' must be defined to enable the extended double-precision
| floating-point format `floatx80'. If this macro is not defined, the
| `floatx80' type will not be defined, and none of the functions that either
| input or output the `floatx80' type will be defined. The same applies to
| the `FLOAT128' macro and the quadruple-precision format `float128'.
*----------------------------------------------------------------------------*/
#define FLOATX80
#define FLOAT128
/*----------------------------------------------------------------------------
| Software IEC/IEEE floating-point types.
*----------------------------------------------------------------------------*/
typedef bits32 float32;
typedef bits64 float64;
#ifdef FLOATX80
typedef struct {
bits16 high;
bits64 low;
} floatx80;
#endif
#ifdef FLOAT128
typedef struct {
bits64 high, low;
} float128;
#endif
/*----------------------------------------------------------------------------
| Software IEC/IEEE floating-point underflow tininess-detection mode.
*----------------------------------------------------------------------------*/
extern int8 float_detect_tininess;
enum {
float_tininess_after_rounding = 0,
float_tininess_before_rounding = 1
};
/*----------------------------------------------------------------------------
| Software IEC/IEEE floating-point rounding mode.
*----------------------------------------------------------------------------*/
extern int8 float_rounding_mode;
enum {
float_round_nearest_even = 0,
float_round_to_zero = 1,
float_round_down = 2,
float_round_up = 3
};
/*----------------------------------------------------------------------------
| Software IEC/IEEE floating-point exception flags.
*----------------------------------------------------------------------------*/
extern int8 float_exception_flags;
enum {
float_flag_inexact = 1,
float_flag_underflow = 2,
float_flag_overflow = 4,
float_flag_divbyzero = 8,
float_flag_invalid = 16
};
/*----------------------------------------------------------------------------
| Routine to raise any or all of the software IEC/IEEE floating-point
| exception flags.
*----------------------------------------------------------------------------*/
void float_raise( int8 );
/*----------------------------------------------------------------------------
| Software IEC/IEEE integer-to-floating-point conversion routines.
*----------------------------------------------------------------------------*/
float32 int32_to_float32( int32 );
float64 int32_to_float64( int32 );
#ifdef FLOATX80
floatx80 int32_to_floatx80( int32 );
#endif
#ifdef FLOAT128
float128 int32_to_float128( int32 );
#endif
float32 int64_to_float32( int64 );
float64 int64_to_float64( int64 );
#ifdef FLOATX80
floatx80 int64_to_floatx80( int64 );
#endif
#ifdef FLOAT128
float128 int64_to_float128( int64 );
#endif
/*----------------------------------------------------------------------------
| Software IEC/IEEE single-precision conversion routines.
*----------------------------------------------------------------------------*/
int32 float32_to_int32( float32 );
int32 float32_to_int32_round_to_zero( float32 );
int64 float32_to_int64( float32 );
int64 float32_to_int64_round_to_zero( float32 );
float64 float32_to_float64( float32 );
#ifdef FLOATX80
floatx80 float32_to_floatx80( float32 );
#endif
#ifdef FLOAT128
float128 float32_to_float128( float32 );
#endif
/*----------------------------------------------------------------------------
| Software IEC/IEEE single-precision operations.
*----------------------------------------------------------------------------*/
float32 float32_round_to_int( float32 );
float32 float32_add( float32, float32 );
float32 float32_sub( float32, float32 );
float32 float32_mul( float32, float32 );
float32 float32_div( float32, float32 );
float32 float32_rem( float32, float32 );
float32 float32_sqrt( float32 );
flag float32_eq( float32, float32 );
flag float32_le( float32, float32 );
flag float32_lt( float32, float32 );
flag float32_eq_signaling( float32, float32 );
flag float32_le_quiet( float32, float32 );
flag float32_lt_quiet( float32, float32 );
flag float32_is_signaling_nan( float32 );
/*----------------------------------------------------------------------------
| Software IEC/IEEE double-precision conversion routines.
*----------------------------------------------------------------------------*/
int32 float64_to_int32( float64 );
int32 float64_to_int32_round_to_zero( float64 );
int64 float64_to_int64( float64 );
int64 float64_to_int64_round_to_zero( float64 );
float32 float64_to_float32( float64 );
#ifdef FLOATX80
floatx80 float64_to_floatx80( float64 );
#endif
#ifdef FLOAT128
float128 float64_to_float128( float64 );
#endif
/*----------------------------------------------------------------------------
| Software IEC/IEEE double-precision operations.
*----------------------------------------------------------------------------*/
float64 float64_round_to_int( float64 );
float64 float64_add( float64, float64 );
float64 float64_sub( float64, float64 );
float64 float64_mul( float64, float64 );
float64 float64_div( float64, float64 );
float64 float64_rem( float64, float64 );
float64 float64_sqrt( float64 );
flag float64_eq( float64, float64 );
flag float64_le( float64, float64 );
flag float64_lt( float64, float64 );
flag float64_eq_signaling( float64, float64 );
flag float64_le_quiet( float64, float64 );
flag float64_lt_quiet( float64, float64 );
flag float64_is_signaling_nan( float64 );
#ifdef FLOATX80
/*----------------------------------------------------------------------------
| Software IEC/IEEE extended double-precision conversion routines.
*----------------------------------------------------------------------------*/
int32 floatx80_to_int32( floatx80 );
int32 floatx80_to_int32_round_to_zero( floatx80 );
int64 floatx80_to_int64( floatx80 );
int64 floatx80_to_int64_round_to_zero( floatx80 );
float32 floatx80_to_float32( floatx80 );
float64 floatx80_to_float64( floatx80 );
#ifdef FLOAT128
float128 floatx80_to_float128( floatx80 );
#endif
/*----------------------------------------------------------------------------
| Software IEC/IEEE extended double-precision rounding precision. Valid
| values are 32, 64, and 80.
*----------------------------------------------------------------------------*/
extern int8 floatx80_rounding_precision;
/*----------------------------------------------------------------------------
| Software IEC/IEEE extended double-precision operations.
*----------------------------------------------------------------------------*/
floatx80 floatx80_round_to_int( floatx80 );
floatx80 floatx80_add( floatx80, floatx80 );
floatx80 floatx80_sub( floatx80, floatx80 );
floatx80 floatx80_mul( floatx80, floatx80 );
floatx80 floatx80_div( floatx80, floatx80 );
floatx80 floatx80_rem( floatx80, floatx80 );
floatx80 floatx80_sqrt( floatx80 );
flag floatx80_eq( floatx80, floatx80 );
flag floatx80_le( floatx80, floatx80 );
flag floatx80_lt( floatx80, floatx80 );
flag floatx80_eq_signaling( floatx80, floatx80 );
flag floatx80_le_quiet( floatx80, floatx80 );
flag floatx80_lt_quiet( floatx80, floatx80 );
flag floatx80_is_signaling_nan( floatx80 );
#endif
#ifdef FLOAT128
/*----------------------------------------------------------------------------
| Software IEC/IEEE quadruple-precision conversion routines.
*----------------------------------------------------------------------------*/
int32 float128_to_int32( float128 );
int32 float128_to_int32_round_to_zero( float128 );
int64 float128_to_int64( float128 );
int64 float128_to_int64_round_to_zero( float128 );
float32 float128_to_float32( float128 );
float64 float128_to_float64( float128 );
#ifdef FLOATX80
floatx80 float128_to_floatx80( float128 );
#endif
/*----------------------------------------------------------------------------
| Software IEC/IEEE quadruple-precision operations.
*----------------------------------------------------------------------------*/
float128 float128_round_to_int( float128 );
float128 float128_add( float128, float128 );
float128 float128_sub( float128, float128 );
float128 float128_mul( float128, float128 );
float128 float128_div( float128, float128 );
float128 float128_rem( float128, float128 );
float128 float128_sqrt( float128 );
flag float128_eq( float128, float128 );
flag float128_le( float128, float128 );
flag float128_lt( float128, float128 );
flag float128_eq_signaling( float128, float128 );
flag float128_le_quiet( float128, float128 );
flag float128_lt_quiet( float128, float128 );
flag float128_is_signaling_nan( float128 );
#endif