Skip to content

Commit

Permalink
fix and optimize bswap
Browse files Browse the repository at this point in the history 
Also allow FLOATTYPE_16PPC, but this format seems to be double-double, not __float128
  • Loading branch information
Reini Urban committed Sep 9, 2012
1 parent b0ab7d2 commit 5851213
Show file tree
Hide file tree
Showing 3 changed files with 88 additions and 52 deletions.
47 changes: 26 additions & 21 deletions include/parrot/bswap.h
View file
Expand Up @@ -5,48 +5,53 @@
*
*/

#ifdef HAS_HEADER_BYTESWAP
#include "parrot/config.h"

#ifdef PARROT_HAS_HEADER_BYTESWAP
# include <byteswap.h> /* GNU */
# define bswap_16(x) __bswap_16(x)
# define bswap_32(x) __bswap_32(x)
# if __WORDSIZE == 64
/* we want fast bswap on 64bit cpus with -m32 */
# ifdef HAS_LONGLONG
# define bswap_64(x) __bswap_64(x)
# endif
#else
# ifdef HAS_HEADER_ENDIAN /* linux */
# ifdef PARROT_HAS_HEADER_ENDIAN /* linux */
# include <endian.h>
# elif defined(HAS_HEADER_SYS_ENDIAN) /* FreeBSD */
# elif defined(PARROT_HAS_HEADER_SYS_ENDIAN) /* FreeBSD */
# include <sys/endian.h>
# endif
# if defined(HAS_HEADER_ENDIAN) \
|| defined(HAS_HEADER_SYS_ENDIAN)
# if defined(PARROT_HAS_HEADER_ENDIAN) \
|| defined(PARROT_HAS_HEADER_SYS_ENDIAN)
# define bswap_16(x) __bswap_16(x)
# define bswap_32(x) __bswap_32(x)
# if __WORDSIZE == 64
# ifdef HAS_LONGLONG
# define bswap_64(x) __bswap_64(x)
# endif
# else
# ifdef HAS_HEADER_LIBKERN_OSBYTEORDER
# ifdef PARROT_HAS_HEADER_LIBKERN_OSBYTEORDER
# include <libkern/OSByteOrder.h>
# define bswap_16(x) OSSwapInt16(x)
# define bswap_32(x) OSSwapInt32(x)
# if __WORDSIZE == 64
# ifdef HAS_LONGLONG
# define bswap_64(x) OSSwapInt64(x)
# endif
# else
# ifdef HAS_HEADER_SYS_BYTEORDER
# ifdef PARROT_HAS_HEADER_SYS_BYTEORDER
# define bswap_16(x) BSWAP_16(x)
# define bswap_32(x) BSWAP_32(x)
# if __WORDSIZE == 64
# ifdef HAS_LONGLONG
# define bswap_64(x) BSWAP_64(x)
# endif
# else
# ifdef __MSC_VER
# define bswap_16(x) _byteswap_ushort(x)
# define bswap_32(x) _byteswap_ulong(x)
# define bswap_64(x) _byteswap_uint64(x)
# ifdef HAS_LONGLONG
# define bswap_64(x) _byteswap_uint64(x)
# endif
# else
/* lost */
/* no native bswap */
# define bswap_16(x) \
({ \
Parrot_UInt2 __x = (x); \
Expand All @@ -63,7 +68,7 @@
(((Parrot_UInt4)(__x) & (Parrot_UInt4)0x00ff0000UL) >> 8) | \
(((Parrot_UInt4)(__x) & (Parrot_UInt4)0xff000000UL) >> 24) )); \
})
# if __WORDSIZE == 64
# ifdef HAS_LONGLONG
# define bswap_64(x) \
({ \
Parrot_UInt8 __x = (x); \
Expand All @@ -78,11 +83,11 @@
(Parrot_UInt8)(((Parrot_UInt8)(__x) & (Parrot_UInt8)0xff00000000000000ULL) >> 56) )); \
})
# else
# define bswap_64(x) ({ \
unsigned char rb[16]; \
# define bswap_64(x) ({ \
unsigned char rb[8]; \
const unsigned char *c = &x; \
SWAB_16(rb,c); \
memcpy(x,rb,16); })
SWAB_8(rb, c); \
(Parrot_UInt8)rb; })
# endif
# endif
# endif
Expand Down Expand Up @@ -187,10 +192,10 @@ static inline Parrot_UInt8 bswap64(Parrot_UInt8 x)
#if defined(bswap_64)
return bswap_64(x);
#else
unsigned char rb[16];
unsigned char rb[8];
const unsigned char *c = &x;
SWAB_16(rb, c);
return rb;
SWAB_8(rb, c);
return (Parrot_UInt8)rb;
#endif
}

Expand Down
11 changes: 6 additions & 5 deletions include/parrot/packfile.h
View file
Expand Up @@ -32,20 +32,21 @@
#define FLOATTYPE_4 3
#define FLOATTYPE_4_NAME "IEEE-754 4-byte single float"

#define FLOATTYPE_16PPC 4
#define FLOATTYPE_16PPC_NAME "PPC64 16 byte double-double"

static
int PF_floattype_size[] = { 8,12,16,4,2,16,16,16 };
int PF_floattype_size[] = { 8,12,16,4,16,2,16,16 };

/* Supported until here. */
#define FLOATTYPE_MAX 3
#define FLOATTYPE_MAX 4

/* Reserved for later */
#define FLOATTYPE_2 4
#define FLOATTYPE_2 5
#define FLOATTYPE_2_NAME "IEEE-754 2-byte half-precision float"

/* Non IEEE-754 versions, yet unsupported. */
/* NaN and other minor differences, but patches welcome */
#define FLOATTYPE_16PPC 5
#define FLOATTYPE_16PPC_NAME "PPC64 16 byte double-double"

#define FLOATTYPE_16MIPS 6
#define FLOATTYPE_16MIPS_NAME "MIPS 16 byte long double"
Expand Down
82 changes: 56 additions & 26 deletions src/packfile/pf_items.c
View file
Expand Up @@ -285,9 +285,9 @@ static opcode_t fetch_op_le_8(ARGIN(const unsigned char *b))
* Floattype 2 = IEEE-754 128 bit quad precision stored in 16 byte,
* Sparc64 quad-float or __float128, gcc since 4.3 (binary128)
* Floattype 3 = IEEE-754 4 byte float (binary32)
* Floattype 4 = PowerPC 16 byte double-double
* not yet:
* Floattype 4 = IEEE-754 2 byte half-precision float (binary16)
* Floattype 5 = PowerPC64 16 byte double-double
* Floattype 6 = MIPS64 16 byte long double
* Floattype 7 = AIX 16 byte long double
*
Expand Down Expand Up @@ -1055,12 +1055,16 @@ PF_fetch_number(ARGIN_NULLOK(PackFile *pf), ARGIN(const opcode_t **stream))
else if (floatsize == 4) {
*((const unsigned char **)stream) = bswap32((Parrot_UInt4)*stream);
}
else if (floatsize == 2) {
*((const unsigned char **)stream) = bswap16((Parrot_UInt2)*stream);
}
else if (floatsize == 16) {
#if 0
/* TODO 64bit CPU: bswap64 with temp */
Parrot_UInt8 tmp = bswap64((Parrot_UInt8)*stream);
*((const unsigned char **)stream) =
bswap64((Parrot_UInt8)*((const unsigned char *)stream+8));
*(((const unsigned char **)stream+8)) = tmp;
#endif
unsigned char rb[16];
const unsigned char *c = (const unsigned char *) *stream;
const unsigned char *c = (const unsigned char *)*stream;
SWAB_16(rb, c);
memcpy(*((const unsigned char **)stream), rb, 16);
}
Expand Down Expand Up @@ -1508,68 +1512,96 @@ PackFile_assign_transforms(ARGMOD(PackFile *pf))
pf->fetch_iv = pf->fetch_op;

switch (pf->header->floattype) {
#if NUMVAL_SIZE == 8
#if FLOATTYPE == FLOATTYPE_8
case FLOATTYPE_8:
pf->fetch_nv = pf->header->byteorder ? fetch_buf_be_8 : fetch_buf_le_8;
break;
case FLOATTYPE_10:
pf->fetch_nv = cvt_num10_num8;
break;
case FLOATTYPE_16:
case FLOATTYPE_16PPC:
pf->fetch_nv = cvt_num16_num8;
break;
case FLOATTYPE_4:
pf->fetch_nv = cvt_num4_num8;
break;
#endif
#if NUMVAL_SIZE == 12
#if FLOATTYPE == FLOATTYPE_10
case FLOATTYPE_8:
pf->fetch_nv = cvt_num8_num10;
break;
case FLOATTYPE_10:
pf->fetch_nv = pf->header->byteorder ? fetch_buf_be_12 : fetch_buf_le_12;
pf->fetch_nv = pf->header->wordsize == 8
? fetch_buf_le_16;
? fetch_buf_le_12;
break;
case FLOATTYPE_16:
case FLOATTYPE_16PPC:
pf->fetch_nv = cvt_num16_num10;
break;
case FLOATTYPE_4:
pf->fetch_nv = cvt_num4_num10;
break;
#endif
#if NUMVAL_SIZE == 16
#if FLOATTYPE == FLOATTYPE_16
case FLOATTYPE_8:
pf->fetch_nv = cvt_num8_num16;
break;
case FLOATTYPE_10:
pf->fetch_nv = cvt_num10_num16;
break;
case FLOATTYPE_16:
case FLOATTYPE_16PPC:
pf->fetch_nv = pf->header->byteorder ? fetch_buf_be_16 : fetch_buf_le_16;
break;
case FLOATTYPE_4:
pf->fetch_nv = cvt_num4_num16;
break;
#endif
#if NUMVAL_SIZE == 4
#if FLOATTYPE == FLOATTYPE_16PPC
case FLOATTYPE_8:
pf->fetch_nv = cvt_num8_num16;
break;
case FLOATTYPE_10:
pf->fetch_nv = cvt_num10_num16;
break;
case FLOATTYPE_16:
case FLOATTYPE_16PPC:
pf->fetch_nv = pf->header->byteorder ? fetch_buf_be_16 : fetch_buf_le_16;
break;
case FLOATTYPE_4:
pf->fetch_nv = cvt_num4_num16;
break;
#endif
#if FLOATTYPE == FLOATTYPE_4
case FLOATTYPE_8:
pf->fetch_nv = cvt_num8_num4;
break;
case FLOATTYPE_10:
pf->fetch_nv = cvt_num10_num4;
break;
case FLOATTYPE_16:
case FLOATTYPE_16PPC:
pf->fetch_nv = cvt_num16_num4;
break;
case FLOATTYPE_4:
pf->fetch_nv = pf->header->byteorder ? fetch_buf_be_4 : fetch_buf_le_4;
break;
#endif
default:
{
int floatsize = PF_floattype_size[ pf->header->floattype ];
/* Intel x86_64 has FLOATTYPE_10 aligned to size 16. */
if ( floatsize == 12 && pf->header->wordsize == 8 )
floatsize = 16;
Parrot_x_force_error_exit(NULL, 1,
"PackFile_unpack: unsupported float conversion %d to %d, "
"PARROT_BIGENDIAN=%d\n",
NUMVAL_SIZE, pf->header->floattype, PARROT_BIGENDIAN);
"size=%d, bigendian=%d\n",
pf->header->floattype, FLOATTYPE,
floatsize, pf->header->byteorder );
break;
}
}
return;
}
Expand Down Expand Up @@ -1729,8 +1761,7 @@ fetch_buf_be_4(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
#if PARROT_BIGENDIAN
memcpy(rb, b, 4);
#else
memcpy(rb, bswap16((Parrot_UInt2)*b), 4);
//SWAB_4(rb,b);
*(Parrot_UInt4*)rb = bswap32(*(Parrot_UInt4*)b);
#endif
}

Expand All @@ -1752,8 +1783,7 @@ fetch_buf_le_4(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
#if !PARROT_BIGENDIAN
memcpy(rb, b, 4);
#else
memcpy(rb, bswap16((Parrot_UInt2)*b), 4);
//SWAB_4(rb,b);
*(Parrot_UInt4*)rb = bswap32(*(Parrot_UInt4*)b);
#endif
}

Expand All @@ -1774,9 +1804,10 @@ fetch_buf_be_8(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
ASSERT_ARGS(fetch_buf_be_8)
#if PARROT_BIGENDIAN
memcpy(rb, b, 8);
#elif defined(HAS_LONGLONG)
*(Parrot_UInt8*)rb = bswap64(*(Parrot_UInt8*)b);
#else
memcpy(rb, bswap32((Parrot_UInt4)*b), 8);
//SWAB_8(rb,b);
SWAB_8(rb, b);
#endif
}

Expand All @@ -1797,9 +1828,10 @@ fetch_buf_le_8(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
ASSERT_ARGS(fetch_buf_le_8)
#if !PARROT_BIGENDIAN
memcpy(rb, b, 8);
#elif defined(HAS_LONGLONG)
*(Parrot_UInt8*)rb = bswap64(*(Parrot_UInt8*)b);
#else
memcpy(rb, bswap32((Parrot_UInt4)*b), 8);
//SWAB_8(rb,b);
SWAB_8(rb, b);
#endif
}

Expand Down Expand Up @@ -1880,8 +1912,7 @@ fetch_buf_le_16(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
#if !PARROT_BIGENDIAN
memcpy(rb, b, 16);
#else
memcpy(rb, bswap64((Parrot_UInt8)*b), 16);
//SWAB_16(rb,b);
SWAB_16(rb, b);
#endif
}

Expand All @@ -1903,8 +1934,7 @@ fetch_buf_be_16(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
#if PARROT_BIGENDIAN
memcpy(rb, b, 16);
#else
memcpy(rb, bswap64((Parrot_UInt8)*b), 16);
//SWAB_16(rb,b);
SWAB_16(rb, b);
#endif
}

Expand All @@ -1926,7 +1956,7 @@ fetch_buf_le_32(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
#if !PARROT_BIGENDIAN
memcpy(rb, b, 32);
#else
SWAB_32(rb,b);
SWAB_32(rb, b);
#endif
}

Expand All @@ -1948,7 +1978,7 @@ fetch_buf_be_32(ARGOUT(unsigned char *rb), ARGIN(const unsigned char *b))
#if PARROT_BIGENDIAN
memcpy(rb, b, 32);
#else
SWAB_32(rb,b);
SWAB_32(rb, b);
#endif
}

Expand Down

0 comments on commit 5851213

Please sign in to comment.