Downgrade to softfloat 2, for improved hackability

This commit is contained in:
Theodore Dubois 2018-06-23 18:59:40 -07:00
parent 9cce538f49
commit 2e53637e5b
9 changed files with 6617 additions and 55 deletions

View File

@ -2,8 +2,8 @@
#define EMU_H
#include <stddef.h>
#include <softfloat.h>
#include "misc.h"
#include "emu/softfloat.h"
#include "emu/memory.h"
#include "emu/tlb.h"
@ -102,7 +102,7 @@ struct cpu_state {
};
// fpu
extFloat80_t fp[8];
floatx80 fp[8];
union {
word_t fsw;
struct {

View File

@ -16,7 +16,7 @@
union xmm_reg xmm_src; \
union xmm_reg xmm_dst; \
\
extFloat80_t ftmp;
floatx80 ftmp;
#define FINISH \
return -1 // everything is ok.

View File

@ -1,33 +1,24 @@
#include <softfloat.h>//0xffffb390
#include "emu/softfloat.h"
// a few extra things not included in the softfloat library
static inline extFloat80_t extF80_to_f80(extFloat80_t f) { return f; }
static inline extFloat80_t f80_to_extF80(extFloat80_t f) { return f; }
static inline extFloat80_t extF80_neg(extFloat80_t f) {
f.signExp ^= 1 << 15; // flip the sign bit
return f;
}
static inline extFloat80_t extF80_abs(extFloat80_t f) {
f.signExp &= ~(1 << 15); // clear the sign bit
return f;
}
// yay hack
static inline floatx80 floatx80_to_float80(floatx80 f) { return f; }
static inline floatx80 float80_to_floatx80(floatx80 f) { return f; }
#define ty_real(x) ty_real_##x
#define ty_real_16 float16_t
#define ty_real_32 float32_t
#define ty_real_64 float64_t
#define ty_real_80 extFloat80_t
#define ty_real_16 float16
#define ty_real_32 float32
#define ty_real_64 float64
#define ty_real_80 floatx80
#define mem_read_real(addr, size) mem_read_ts(addr, ty_real(size), size)
#define mem_write_real(addr, val, size) mem_write_ts(addr, val, ty_real(size), size)
#define get_mem_addr_real(size) mem_read_real(addr, size)
#define set_mem_addr_real(to, size) mem_write_real(addr, to, size)
#define extF80_to_f(f, z) glue(extF80_to_f, sz(z))(f)
#define f_to_extF80(f_, z) glue3(f, sz(z), _to_extF80)(f_)
#define extF80_to_i(i, round, exact, z) glue(extF80_to_i, sz(z))(i, round, exact)
#define i_to_extF80(i_, round, exact, z) glue3(i, sz(z), _to_extF80)(i_, round, exact)
#define floatx80_to_float(f, z) glue(floatx80_to_float, sz(z))(f)
#define float_to_floatx80(f_, z) glue3(float, sz(z), _to_floatx80)(f_)
#define floatx80_to_int(i, z) glue(floatx80_to_int, sz(z))(i)
#define int_to_floatx80(i_, z) glue3(int, sz(z), _to_floatx80)(i_)
#define ST(i) cpu->fp[cpu->top + i]
#define ST_i ST(modrm.rm_opcode)
@ -37,74 +28,74 @@ static inline extFloat80_t extF80_abs(extFloat80_t f) {
cpu->top++
#define FXCH() \
extFloat80_t ftmp = ST(0); ST(0) = ST_i; ST_i = ftmp
floatx80 ftmp = ST(0); ST(0) = ST_i; ST_i = ftmp
#define st_0 ST(0)
#define st_i ST(modrm.rm_opcode)
#define FADD(src, dst) \
dst = extF80_add(dst, src)
dst = floatx80_add(dst, src)
#define FIADD(val,z) \
ST(0) = extF80_add(ST(0), i64_to_extF80((sint(z)) get(val,z)))
ST(0) = floatx80_add(ST(0), int64_to_floatx80((sint(z)) get(val,z)))
#define FADDM(val,z) \
ST(0) = extF80_add(ST(0), f_to_extF80(get(val,z),z))
ST(0) = floatx80_add(ST(0), float_to_floatx80(get(val,z),z))
#define FSUB(src, dst) \
dst = extF80_sub(dst, src)
dst = floatx80_sub(dst, src)
#define FSUBM(val,z) \
ST(0) = extF80_sub(ST(0), f_to_extF80(get(val,z),z))
ST(0) = floatx80_sub(ST(0), float_to_floatx80(get(val,z),z))
#define FISUB(val,z) \
ST(0) = extF80_sub(ST(0), i64_to_extF80((sint(z)) get(val,z)))
ST(0) = floatx80_sub(ST(0), int64_to_floatx80((sint(z)) get(val,z)))
#define FMUL(src, dst) \
dst = extF80_mul(dst, src)
dst = floatx80_mul(dst, src)
#define FIMUL(val,z) \
ST(0) = extF80_mul(ST(0), i64_to_extF80((sint(z)) get(val,z)))
ST(0) = floatx80_mul(ST(0), int64_to_floatx80((sint(z)) get(val,z)))
#define FMULM(val,z) \
ST(0) = extF80_mul(ST(0), f_to_extF80(get(val,z),z))
ST(0) = floatx80_mul(ST(0), float_to_floatx80(get(val,z),z))
#define FDIV(src, dst) \
dst = extF80_div(dst, src)
dst = floatx80_div(dst, src)
#define FIDIV(val,z) \
ST(0) = extF80_div(ST(0), i64_to_extF80((sint(z)) get(val,z)))
ST(0) = floatx80_div(ST(0), int64_to_floatx80((sint(z)) get(val,z)))
#define FDIVM(val,z) \
ST(0) = extF80_div(ST(0), f_to_extF80(get(val,z),z))
ST(0) = floatx80_div(ST(0), float_to_floatx80(get(val,z),z))
#define FCHS() \
ST(0) = extF80_neg(ST(0))
floatx80_neg(ST(0))
#define FABS() \
ST(0) = extF80_abs(ST(0))
floatx80_abs(ST(0))
// FIXME this is the IEEE ABSOLUTELY CORRECT AND AWESOME REMAINDER which is
// computed by fprem1, not fprem
// only known case of intel naming an instruction by taking another instruction
// that does the same thing but wrong and adding a 1
#define FPREM() \
ST(0) = extF80_rem(ST(0), ST(1))
ST(0) = floatx80_rem(ST(0), ST(1))
#define FUCOMI() \
cpu->zf = extF80_eq(ST(0), ST_i); \
cpu->cf = extF80_lt(ST(0), ST_i); \
cpu->zf = floatx80_eq(ST(0), ST_i); \
cpu->cf = floatx80_lt(ST(0), ST_i); \
cpu->pf = 0; cpu->pf_res = 0
// not worrying about nans and shit yet
#define FUCOM() \
cpu->c0 = extF80_lt(ST(0), ST_i); \
cpu->c0 = floatx80_lt(ST(0), ST_i); \
cpu->c1 = 0; \
cpu->c2 = 0; /* again, not worrying about nans */ \
cpu->c3 = extF80_eq(ST(0), ST_i)
cpu->c3 = floatx80_eq(ST(0), ST_i)
#define FILD(val,z) \
FPUSH(i64_to_extF80((sint(z)) get(val,z)))
FPUSH(int64_to_floatx80((sint(z)) get(val,z)))
#define FLD() FPUSH(ST_i)
#define FLDM(val,z) \
FPUSH(f_to_extF80(get(val,z),z))
FPUSH(float_to_floatx80(get(val,z),z))
#define FLDC(what) FPUSH(fconst_##what)
#define fconst_one i64_to_extF80(1)
#define fconst_zero i64_to_extF80(0)
#define fconst_one int64_to_floatx80(1)
#define fconst_zero int64_to_floatx80(0)
#define FSTM(dst,z) \
set(dst, extF80_to_f(ST(0),z),z)
set(dst, floatx80_to_float(ST(0),z),z)
#define FIST(dst,z) \
set(dst, extF80_to_i(ST(0), softfloat_roundingMode, false, z),z)
set(dst, floatx80_to_int(ST(0), z),z)
#define FST() ST_i = ST(0)

View File

@ -49,6 +49,6 @@
#define MOVD(src, dst) \
set(dst, get(src,128).dw[0],32)
#include <softfloat.h>
#include "emu/softfloat.h"
#define CVTTSD2SI(src, dst) \
set(dst, f64_to_i32(get(src,64), softfloat_round_minMag, false),32)
set(dst, float64_to_int32(get(src,64)),32)

714
emu/softfloat-macros.h Normal file
View File

@ -0,0 +1,714 @@
// This file was originally part of Berkeley Softfloat. It has been modified for use in iSH.
/*============================================================================
This C source fragment is part of the Berkeley SoftFloat IEEE Floating-Point
Arithmetic Package, Release 2c, by John R. Hauser.
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 TOLERATE ALL LOSSES, COSTS, OR OTHER
PROBLEMS THEY INCUR DUE TO THE SOFTWARE WITHOUT RECOMPENSE FROM JOHN HAUSER OR
THE INTERNATIONAL COMPUTER SCIENCE INSTITUTE, 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, OR
INCURRED BY ANYONE DUE TO A DERIVATIVE WORK THEY CREATE USING ANY PART OF THE
SOFTWARE.
Derivative works require also that (1) the source code for the derivative work
includes prominent notice that the work is derivative, and (2) the source code
includes prominent notice of these three 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'.
*----------------------------------------------------------------------------*/
static inline void shift32RightJamming( uint32_t a, int16_t count, uint32_t *zPtr )
{
uint32_t 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'.
*----------------------------------------------------------------------------*/
static inline void shift64RightJamming( uint64_t a, int16_t count, uint64_t *zPtr )
{
uint64_t 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'.)
*----------------------------------------------------------------------------*/
static inline void
shift64ExtraRightJamming(
uint64_t a0, uint64_t a1, int16_t count, uint64_t *z0Ptr, uint64_t *z1Ptr )
{
uint64_t z0, z1;
int8_t 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'.
*----------------------------------------------------------------------------*/
static inline void
shift128Right(
uint64_t a0, uint64_t a1, int16_t count, uint64_t *z0Ptr, uint64_t *z1Ptr )
{
uint64_t z0, z1;
int8_t 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 < 128 ) ? ( 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'.
*----------------------------------------------------------------------------*/
static inline void
shift128RightJamming(
uint64_t a0, uint64_t a1, int16_t count, uint64_t *z0Ptr, uint64_t *z1Ptr )
{
uint64_t z0, z1;
int8_t 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'.)
*----------------------------------------------------------------------------*/
static inline void
shift128ExtraRightJamming(
uint64_t a0,
uint64_t a1,
uint64_t a2,
int16_t count,
uint64_t *z0Ptr,
uint64_t *z1Ptr,
uint64_t *z2Ptr
)
{
uint64_t z0, z1, z2;
int8_t 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'.
*----------------------------------------------------------------------------*/
static inline void
shortShift128Left(
uint64_t a0, uint64_t a1, int16_t count, uint64_t *z0Ptr, uint64_t *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'.
*----------------------------------------------------------------------------*/
static inline void
shortShift192Left(
uint64_t a0,
uint64_t a1,
uint64_t a2,
int16_t count,
uint64_t *z0Ptr,
uint64_t *z1Ptr,
uint64_t *z2Ptr
)
{
uint64_t z0, z1, z2;
int8_t 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'.
*----------------------------------------------------------------------------*/
static inline void
add128(
uint64_t a0, uint64_t a1, uint64_t b0, uint64_t b1, uint64_t *z0Ptr, uint64_t *z1Ptr )
{
uint64_t 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'.
*----------------------------------------------------------------------------*/
static inline void
add192(
uint64_t a0,
uint64_t a1,
uint64_t a2,
uint64_t b0,
uint64_t b1,
uint64_t b2,
uint64_t *z0Ptr,
uint64_t *z1Ptr,
uint64_t *z2Ptr
)
{
uint64_t z0, z1, z2;
int8_t 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'.
*----------------------------------------------------------------------------*/
static inline void
sub128(
uint64_t a0, uint64_t a1, uint64_t b0, uint64_t b1, uint64_t *z0Ptr, uint64_t *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'.
*----------------------------------------------------------------------------*/
static inline void
sub192(
uint64_t a0,
uint64_t a1,
uint64_t a2,
uint64_t b0,
uint64_t b1,
uint64_t b2,
uint64_t *z0Ptr,
uint64_t *z1Ptr,
uint64_t *z2Ptr
)
{
uint64_t z0, z1, z2;
int8_t 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'.
*----------------------------------------------------------------------------*/
static inline void mul64To128( uint64_t a, uint64_t b, uint64_t *z0Ptr, uint64_t *z1Ptr )
{
uint32_t aHigh, aLow, bHigh, bLow;
uint64_t z0, zMiddleA, zMiddleB, z1;
aLow = a;
aHigh = a>>32;
bLow = b;
bHigh = b>>32;
z1 = ( (uint64_t) aLow ) * bLow;
zMiddleA = ( (uint64_t) aLow ) * bHigh;
zMiddleB = ( (uint64_t) aHigh ) * bLow;
z0 = ( (uint64_t) aHigh ) * bHigh;
zMiddleA += zMiddleB;
z0 += ( ( (uint64_t) ( 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'.
*----------------------------------------------------------------------------*/
static inline void
mul128By64To192(
uint64_t a0,
uint64_t a1,
uint64_t b,
uint64_t *z0Ptr,
uint64_t *z1Ptr,
uint64_t *z2Ptr
)
{
uint64_t 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'.
*----------------------------------------------------------------------------*/
static inline void
mul128To256(
uint64_t a0,
uint64_t a1,
uint64_t b0,
uint64_t b1,
uint64_t *z0Ptr,
uint64_t *z1Ptr,
uint64_t *z2Ptr,
uint64_t *z3Ptr
)
{
uint64_t z0, z1, z2, z3;
uint64_t 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 uint64_t estimateDiv128To64( uint64_t a0, uint64_t a1, uint64_t b )
{
uint64_t b0, b1;
uint64_t rem0, rem1, term0, term1;
uint64_t 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 ( ( (int64_t) 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 uint32_t estimateSqrt32( int16_t aExp, uint32_t a )
{
static const uint16_t sqrtOddAdjustments[] = {
0x0004, 0x0022, 0x005D, 0x00B1, 0x011D, 0x019F, 0x0236, 0x02E0,
0x039C, 0x0468, 0x0545, 0x0631, 0x072B, 0x0832, 0x0946, 0x0A67
};
static const uint16_t sqrtEvenAdjustments[] = {
0x0A2D, 0x08AF, 0x075A, 0x0629, 0x051A, 0x0429, 0x0356, 0x029E,
0x0200, 0x0179, 0x0109, 0x00AF, 0x0068, 0x0034, 0x0012, 0x0002
};
int8_t index;
uint32_t 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 (uint32_t) ( ( (int32_t) a )>>1 );
}
return ( (uint32_t) ( ( ( (uint64_t) 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_t countLeadingZeros32( uint32_t a )
{
static const int8_t 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_t 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_t countLeadingZeros64( uint64_t a )
{
int8_t shiftCount;
shiftCount = 0;
if ( a < ( (uint64_t) 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.
*----------------------------------------------------------------------------*/
static inline bool eq128( uint64_t a0, uint64_t a1, uint64_t b0, uint64_t 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.
*----------------------------------------------------------------------------*/
static inline bool le128( uint64_t a0, uint64_t a1, uint64_t b0, uint64_t 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.
*----------------------------------------------------------------------------*/
static inline bool lt128( uint64_t a0, uint64_t a1, uint64_t b0, uint64_t 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.
*----------------------------------------------------------------------------*/
static inline bool ne128( uint64_t a0, uint64_t a1, uint64_t b0, uint64_t b1 )
{
return ( a0 != b0 ) || ( a1 != b1 );
}

426
emu/softfloat-specialize.h Normal file
View File

@ -0,0 +1,426 @@
// This file was originally part of Berkeley Softfloat. It has been modified for use in iSH.
/*============================================================================
This C source fragment is part of the Berkeley SoftFloat IEEE Floating-Point
Arithmetic Package, Release 2c, by John R. Hauser.
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 TOLERATE ALL LOSSES, COSTS, OR OTHER
PROBLEMS THEY INCUR DUE TO THE SOFTWARE WITHOUT RECOMPENSE FROM JOHN HAUSER OR
THE INTERNATIONAL COMPUTER SCIENCE INSTITUTE, 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, OR
INCURRED BY ANYONE DUE TO A DERIVATIVE WORK THEY CREATE USING ANY PART OF THE
SOFTWARE.
Derivative works require also that (1) the source code for the derivative work
includes prominent notice that the work is derivative, and (2) the source code
includes prominent notice of these three 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_t' type here.)
*----------------------------------------------------------------------------*/
int8_t 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_t flags )
{
float_exception_flags |= flags;
}
/*----------------------------------------------------------------------------
| Internal canonical NaN format.
*----------------------------------------------------------------------------*/
typedef struct {
bool sign;
uint64_t signExp, signif;
} 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.
*----------------------------------------------------------------------------*/
bool float32_is_nan( float32 a )
{
return ( 0xFF000000 < (uint32_t) ( a<<1 ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the single-precision floating-point value `a' is a signaling
| NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
bool 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.signif = 0;
z.signExp = ( (uint64_t) 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 ( ( (uint32_t) a.sign )<<31 ) | 0x7FC00000 | ( a.signExp>>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 )
{
bool 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.
*----------------------------------------------------------------------------*/
bool float64_is_nan( float64 a )
{
return ( LIT64( 0xFFE0000000000000 ) < (uint64_t) ( a<<1 ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the double-precision floating-point value `a' is a signaling
| NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
bool 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.signif = 0;
z.signExp = 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
( ( (uint64_t) a.sign )<<63 )
| LIT64( 0x7FF8000000000000 )
| ( a.signExp>>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 )
{
bool 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 double-extended-precision NaN.
| The `signExp' and `signif' 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 double-extended-precision floating-point value `a' is a
| NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
bool floatx80_is_nan( floatx80 a )
{
return ( ( a.signExp & 0x7FFF ) == 0x7FFF ) && (uint64_t) ( a.signif<<1 );
}
/*----------------------------------------------------------------------------
| Returns 1 if the double-extended-precision floating-point value `a' is a
| signaling NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
bool floatx80_is_signaling_nan( floatx80 a )
{
uint64_t aLow;
aLow = a.signif & ~ LIT64( 0x4000000000000000 );
return
( ( a.signExp & 0x7FFF ) == 0x7FFF )
&& (uint64_t) ( aLow<<1 )
&& ( a.signif == aLow );
}
/*----------------------------------------------------------------------------
| Returns the result of converting the double-extended-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.signExp>>15;
z.signif = 0;
z.signExp = a.signif<<1;
return z;
}
/*----------------------------------------------------------------------------
| Returns the result of converting the canonical NaN `a' to the double-
| extended-precision floating-point format.
*----------------------------------------------------------------------------*/
static floatx80 commonNaNToFloatx80( commonNaNT a )
{
floatx80 z;
z.signif = LIT64( 0xC000000000000000 ) | ( a.signExp>>1 );
z.signExp = ( ( (uint16_t) a.sign )<<15 ) | 0x7FFF;
return z;
}
/*----------------------------------------------------------------------------
| Takes two double-extended-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 )
{
bool 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.signif |= LIT64( 0xC000000000000000 );
b.signif |= 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 `signExp' and
| `signif' 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.
*----------------------------------------------------------------------------*/
bool float128_is_nan( float128 a )
{
return
( LIT64( 0xFFFE000000000000 ) <= (uint64_t) ( a.signExp<<1 ) )
&& ( a.signif || ( a.signExp & LIT64( 0x0000FFFFFFFFFFFF ) ) );
}
/*----------------------------------------------------------------------------
| Returns 1 if the quadruple-precision floating-point value `a' is a
| signaling NaN; otherwise returns 0.
*----------------------------------------------------------------------------*/
bool float128_is_signaling_nan( float128 a )
{
return
( ( ( a.signExp>>47 ) & 0xFFFF ) == 0xFFFE )
&& ( a.signif || ( a.signExp & 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.signExp>>63;
shortShift128Left( a.signExp, a.signif, 16, &z.signExp, &z.signif );
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.signExp, a.signif, 16, &z.signExp, &z.signif );
z.signExp |= ( ( (uint64_t) 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 )
{
bool 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.signExp |= LIT64( 0x0000800000000000 );
b.signExp |= LIT64( 0x0000800000000000 );
if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid );
if ( aIsNaN ) {
return ( aIsSignalingNaN & bIsNaN ) ? b : a;
}
else {
return b;
}
}
#endif

5161
emu/softfloat.c Normal file

File diff suppressed because it is too large Load Diff

270
emu/softfloat.h Normal file
View File

@ -0,0 +1,270 @@
// This file was originally part of Berkeley Softfloat. It has been modified for use in iSH.
/*============================================================================
This C header file template is part of the Berkeley SoftFloat IEEE Floating-
Point Arithmetic Package, Release 2c, by John R. Hauser.
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 TOLERATE ALL LOSSES, COSTS, OR OTHER
PROBLEMS THEY INCUR DUE TO THE SOFTWARE WITHOUT RECOMPENSE FROM JOHN HAUSER OR
THE INTERNATIONAL COMPUTER SCIENCE INSTITUTE, 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, OR
INCURRED BY ANYONE DUE TO A DERIVATIVE WORK THEY CREATE USING ANY PART OF THE
SOFTWARE.
Derivative works require also that (1) the source code for the derivative work
includes prominent notice that the work is derivative, and (2) the source code
includes prominent notice of these three paragraphs for those parts of this
code that are retained.
=============================================================================*/
#ifndef SOFTFLOAT_H
#define SOFTFLOAT_H
/*----------------------------------------------------------------------------
| The macro `FLOATX80' must be defined to enable the double-extended-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
#include <stdint.h>
#include <stdbool.h>
#define LIT64( a ) a##LL
/*----------------------------------------------------------------------------
| Software IEEE floating-point types.
*----------------------------------------------------------------------------*/
typedef uint32_t float32;
typedef uint64_t float64;
#ifdef FLOATX80
typedef struct {
uint16_t signExp;
uint64_t signif;
} floatx80;
#endif
#ifdef FLOAT128
typedef struct {
uint64_t signExp, signif;
} float128;
#endif
/*----------------------------------------------------------------------------
| Software IEEE floating-point underflow tininess-detection mode.
*----------------------------------------------------------------------------*/
extern int8_t float_detect_tininess;
enum {
float_tininess_after_rounding = 0,
float_tininess_before_rounding = 1
};
/*----------------------------------------------------------------------------
| Software IEEE floating-point rounding mode.
*----------------------------------------------------------------------------*/
extern int8_t float_rounding_mode;
enum {
float_round_nearest_even = 0,
float_round_to_zero = 1,
float_round_down = 2,
float_round_up = 3
};
/*----------------------------------------------------------------------------
| Software IEEE floating-point exception flags.
*----------------------------------------------------------------------------*/
extern int8_t 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 IEEE floating-point exception
| flags.
*----------------------------------------------------------------------------*/
void float_raise( int8_t );
/*----------------------------------------------------------------------------
| Software IEEE integer-to-floating-point conversion routines.
*----------------------------------------------------------------------------*/
float32 int32_to_float32( int32_t );
float64 int32_to_float64( int32_t );
#ifdef FLOATX80
floatx80 int32_to_floatx80( int32_t );
#endif
#ifdef FLOAT128
float128 int32_to_float128( int32_t );
#endif
float32 int64_to_float32( int64_t );
float64 int64_to_float64( int64_t );
#ifdef FLOATX80
floatx80 int64_to_floatx80( int64_t );
#endif
#ifdef FLOAT128
float128 int64_to_float128( int64_t );
#endif
/*----------------------------------------------------------------------------
| Software IEEE single-precision conversion routines.
*----------------------------------------------------------------------------*/
int32_t float32_to_int32( float32 );
int32_t float32_to_int32_round_to_zero( float32 );
int64_t float32_to_int64( float32 );
int64_t 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 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 );
bool float32_eq( float32, float32 );
bool float32_le( float32, float32 );
bool float32_lt( float32, float32 );
bool float32_eq_signaling( float32, float32 );
bool float32_le_quiet( float32, float32 );
bool float32_lt_quiet( float32, float32 );
bool float32_is_signaling_nan( float32 );
/*----------------------------------------------------------------------------
| Software IEEE double-precision conversion routines.
*----------------------------------------------------------------------------*/
int32_t float64_to_int32( float64 );
int32_t float64_to_int32_round_to_zero( float64 );
int64_t float64_to_int64( float64 );
int64_t 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 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 );
bool float64_eq( float64, float64 );
bool float64_le( float64, float64 );
bool float64_lt( float64, float64 );
bool float64_eq_signaling( float64, float64 );
bool float64_le_quiet( float64, float64 );
bool float64_lt_quiet( float64, float64 );
bool float64_is_signaling_nan( float64 );
#ifdef FLOATX80
/*----------------------------------------------------------------------------
| Software IEEE double-extended-precision conversion routines.
*----------------------------------------------------------------------------*/
int32_t floatx80_to_int32( floatx80 );
int32_t floatx80_to_int32_round_to_zero( floatx80 );
int64_t floatx80_to_int64( floatx80 );
int64_t 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
static inline void floatx80_neg(floatx80 f) {
f.signExp ^= 1 << 15; // flip the sign bit
}
static inline void floatx80_abs(floatx80 f) {
f.signExp &= ~(1 << 15); // clear the sign bit
}
/*----------------------------------------------------------------------------
| Software IEEE double-extended-precision rounding precision. Valid values
| are 32, 64, and 80.
*----------------------------------------------------------------------------*/
extern int8_t floatx80_rounding_precision;
/*----------------------------------------------------------------------------
| Software IEEE double-extended-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 );
bool floatx80_eq( floatx80, floatx80 );
bool floatx80_le( floatx80, floatx80 );
bool floatx80_lt( floatx80, floatx80 );
bool floatx80_eq_signaling( floatx80, floatx80 );
bool floatx80_le_quiet( floatx80, floatx80 );
bool floatx80_lt_quiet( floatx80, floatx80 );
bool floatx80_is_signaling_nan( floatx80 );
#endif
#ifdef FLOAT128
/*----------------------------------------------------------------------------
| Software IEEE quadruple-precision conversion routines.
*----------------------------------------------------------------------------*/
int32_t float128_to_int32( float128 );
int32_t float128_to_int32_round_to_zero( float128 );
int64_t float128_to_int64( float128 );
int64_t 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 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 );
bool float128_eq( float128, float128 );
bool float128_le( float128, float128 );
bool float128_lt( float128, float128 );
bool float128_eq_signaling( float128, float128 );
bool float128_le_quiet( float128, float128 );
bool float128_lt_quiet( float128, float128 );
bool float128_is_signaling_nan( float128 );
#endif
#endif

View File

@ -28,7 +28,6 @@ threads = dependency('threads')
librt = cc.find_library('rt', required: false)
gdbm = subproject('gdbm').get_variable('gdbm')
softfloat = subproject('softfloat').get_variable('softfloat_dep')
subdir('vdso') # ish depends on the vdso
@ -90,6 +89,7 @@ src = [
'emu/memory.c',
'emu/tlb.c',
'emu/softfloat.c',
cified_vdso,
]
if get_option('jit')
@ -114,10 +114,10 @@ endif
libish = library('ish', src,
include_directories: includes,
dependencies: [librt, threads, softfloat, gdbm])
dependencies: [librt, threads, gdbm])
ish = declare_dependency(
link_with: libish,
dependencies: [librt, threads, softfloat, gdbm],
dependencies: [librt, threads, gdbm],
include_directories: includes)
# ptraceomatic et al