zenithsiz/ftmemsim-valgrind
1
0
Fork 0
mirror of https://github.com/Zenithsiz/ftmemsim-valgrind.git synced 2026-02-04 10:21:20 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
ftmemsim-valgrind/include/pub_tool_libcbase.h
Julian Seward 50bb127b1d Bug 402781 - Redo the cache used to process indirect branch targets. 
[This commit contains an implementation for all targets except amd64-solaris
and x86-solaris, which will be completed shortly.]

In the baseline simulator, jumps to guest code addresses that are not known at
JIT time have to be looked up in a guest->host mapping table.  That means:
indirect branches, indirect calls and most commonly, returns.  Since there are
huge numbers of these (often 10+ million/second) the mapping mechanism needs
to be extremely cheap.

Currently, this is implemented using a direct-mapped cache, VG_(tt_fast), with
2^15 (guest_addr, host_addr) pairs.  This is queried in handwritten assembly
in VG_(disp_cp_xindir) in dispatch-<arch>-<os>.S.  If there is a miss in the
cache then we fall back out to C land, and do a slow lookup using
VG_(search_transtab).

Given that the size of the translation table(s) in recent years has expanded
significantly in order to keep pace with increasing application sizes, two bad
things have happened: (1) the cost of a miss in the fast cache has risen
significantly, and (2) the miss rate on the fast cache has also increased
significantly.  This means that large (~ one-million-basic-blocks-JITted)
applications that run for a long time end up spending a lot of time in
VG_(search_transtab).

The proposed fix is to increase associativity of the fast cache, from 1
(direct mapped) to 4.  Simulations of various cache configurations using
indirect-branch traces from a large application show that is the best of
various configurations.  In an extreme case with 5.7 billion indirect
branches:

* The increase of associativity from 1 way to 4 way, whilst keeping the
  overall cache size the same (32k guest/host pairs), reduces the miss rate by
  around a factor of 3, from 4.02% to 1.30%.

* The use of a slightly better hash function than merely slicing off the
  bottom 15 bits of the address, reduces the miss rate further, from 1.30% to
  0.53%.

Overall the VG_(tt_fast) miss rate is almost unchanged on small workloads, but
reduced by a factor of up to almost 8 on large workloads.

By implementing each (4-entry) cache set using a move-to-front scheme in the
case of hits in ways 1, 2 or 3, the vast majority of hits can be made to
happen in way 0.  Hence the cost of having this extra associativity is almost
zero in the case of a hit.  The improved hash function costs an extra 2 ALU
shots (a shift and an xor) but overall this seems performance neutral to a
win.
2019-01-25 09:14:56 +01:00

239 lines
11 KiB
C
Raw Blame History

/*--------------------------------------------------------------------*/
/*--- Standalone libc stuff. pub_tool_libcbase.h ---*/
/*--------------------------------------------------------------------*/
/*
This file is part of Valgrind, a dynamic binary instrumentation
framework.
Copyright (C) 2000-2017 Julian Seward
jseward@acm.org
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the
License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
02111-1307, USA.
The GNU General Public License is contained in the file COPYING.
*/
#ifndef __PUB_TOOL_LIBCBASE_H
#define __PUB_TOOL_LIBCBASE_H
#include "pub_tool_basics.h" // VG_ macro
/* ---------------------------------------------------------------------
Char functions.
------------------------------------------------------------------ */
extern Bool VG_(isspace) ( HChar c );
extern Bool VG_(isdigit) ( HChar c );
extern HChar VG_(tolower) ( HChar c );
/* ---------------------------------------------------------------------
Converting strings to numbers
------------------------------------------------------------------ */
// Convert strings to numbers according to various bases. Leading
// whitespace is ignored. A subsequent '-' or '+' is accepted. For strtoll16,
// accepts an initial "0x" or "0X" prefix, but only if it's followed by a
// hex digit (if not, the '0' will be read and then it will stop on the
// "x"/"X".) If 'endptr' isn't NULL, it gets filled in with the first
// non-digit char. Returns 0 if no number could be converted, and 'endptr'
// is set to the start of the string. None of them test that the number
// fits into 64 bits.
//
// Nb: we also don't provide VG_(atoll*); these functions are worse than
// useless because they don't do any error checking and so accept malformed
// numbers and non-numbers -- eg. "123xyz" gives 123, and "foo" gives 0!
// If you really want that behaviour, you can use "VG_(strtoll10)(str, NULL)".
extern Long VG_(strtoll10) ( const HChar* str, HChar** endptr );
extern Long VG_(strtoll16) ( const HChar* str, HChar** endptr );
extern ULong VG_(strtoull10) ( const HChar* str, HChar** endptr );
extern ULong VG_(strtoull16) ( const HChar* str, HChar** endptr );
// Convert a string to a double. After leading whitespace is ignored, a
// '+' or '-' is allowed, and then it accepts a non-empty sequence of
// decimal digits possibly containing a '.'. Hexadecimal floats are not
// accepted, nor are "fancy" floats (eg. "3.4e-5", "NAN").
extern double VG_(strtod) ( const HChar* str, HChar** endptr );
/* ---------------------------------------------------------------------
String functions and macros
------------------------------------------------------------------ */
/* Use this for normal null-termination-style string comparison. */
#define VG_STREQ(s1,s2) ( (s1 != NULL && s2 != NULL \
&& VG_(strcmp)((s1),(s2))==0) ? True : False )
#define VG_STREQN(n,s1,s2) ( (s1 != NULL && s2 != NULL \
&& VG_(strncmp)((s1),(s2),(n))==0) ? True : False )
extern SizeT VG_(strlen) ( const HChar* str );
extern SizeT VG_(strnlen) ( const HChar* str, SizeT n );
extern HChar* VG_(strcat) ( HChar* dest, const HChar* src );
extern HChar* VG_(strncat) ( HChar* dest, const HChar* src, SizeT n );
extern HChar* VG_(strpbrk) ( const HChar* s, const HChar* accpt );
extern HChar* VG_(strcpy) ( HChar* dest, const HChar* src );
extern HChar* VG_(strncpy) ( HChar* dest, const HChar* src, SizeT ndest );
extern SizeT VG_(strlcpy) ( HChar* dest, const HChar* src, SizeT n );
extern Int VG_(strcmp) ( const HChar* s1, const HChar* s2 );
extern Int VG_(strcasecmp) ( const HChar* s1, const HChar* s2 );
extern Int VG_(strncmp) ( const HChar* s1, const HChar* s2, SizeT nmax );
extern Int VG_(strncasecmp) ( const HChar* s1, const HChar* s2, SizeT nmax );
extern HChar* VG_(strstr) ( const HChar* haystack, const HChar* needle );
extern HChar* VG_(strcasestr) ( const HChar* haystack, const HChar* needle );
extern HChar* VG_(strchr) ( const HChar* s, HChar c );
extern HChar* VG_(strrchr) ( const HChar* s, HChar c );
extern SizeT VG_(strspn) ( const HChar* s, const HChar* accpt );
extern SizeT VG_(strcspn) ( const HChar* s, const HChar* reject );
/* strtok* functions and some parsing utilities. */
extern HChar* VG_(strtok_r) (HChar* s, const HChar* delim, HChar** saveptr);
extern HChar* VG_(strtok) (HChar* s, const HChar* delim);
/* Parse a 32- or 64-bit hex number, including leading 0x, from string
starting at *ppc, putting result in *result, advance *ppc past the
characters used, and return True. Or fail, in which case *ppc and
*result are undefined, and return False. */
extern Bool VG_(parse_Addr) ( const HChar** ppc, Addr* result );
/* Parse an unsigned 32 bit number, written using decimals only.
Calling conventions are the same as for VG_(parse_Addr). */
extern Bool VG_(parse_UInt) ( const HChar** ppc, UInt* result );
/* Parse an "enum set" made of one or more words comma separated.
The allowed word values are given in 'tokens', separated by comma.
If a word in 'tokens' is found in 'input', the corresponding bit
will be set in *enum_set (words in 'tokens' are numbered starting from 0).
Using in 'tokens' the special token "-" (a minus character) indicates that
the corresponding bit position cannot be set.
In addition to the words specified in 'tokens', VG_(parse_enum_set)
automatically accept the word "none" to indicate an empty enum_set (0).
If allow_all, VG_(parse_enum_set) automatically accept the word "all"
to indicate an enum_set with all bits corresponding to the words in tokens
set.
If "none" or "all" is present in 'input', no other word can be given
in 'input'.
If parsing is successful, returns True and sets *enum_set.
If parsing fails, returns False. */
extern Bool VG_(parse_enum_set) ( const HChar *tokens,
Bool allow_all,
const HChar *input,
UInt *enum_set);
/* ---------------------------------------------------------------------
mem* functions
------------------------------------------------------------------ */
extern void* VG_(memcpy) ( void *d, const void *s, SizeT sz );
extern void* VG_(memmove)( void *d, const void *s, SizeT sz );
extern void* VG_(memset) ( void *s, Int c, SizeT sz );
extern Int VG_(memcmp) ( const void* s1, const void* s2, SizeT n );
/* Zero out up to 12 words quickly in-line. Do not use this for blocks
of size which are unknown at compile time, since the whole point is
for it to be inlined, and then for gcc to remove all code except
for the relevant 'sz' case. */
inline __attribute__((always_inline))
static void VG_(bzero_inline) ( void* s, SizeT sz )
{
if (LIKELY(0 == (((Addr)sz) & (Addr)(sizeof(UWord)-1)))
&& LIKELY(0 == (((Addr)s) & (Addr)(sizeof(UWord)-1)))) {
UWord* p = (UWord*)s;
switch (sz / (SizeT)sizeof(UWord)) {
case 12: p[0] = p[1] = p[2] = p[3]
= p[4] = p[5] = p[6] = p[7]
= p[8] = p[9] = p[10] = p[11] = 0UL; return;
case 11: p[0] = p[1] = p[2] = p[3]
= p[4] = p[5] = p[6] = p[7]
= p[8] = p[9] = p[10] = 0UL; return;
case 10: p[0] = p[1] = p[2] = p[3]
= p[4] = p[5] = p[6] = p[7]
= p[8] = p[9] = 0UL; return;
case 9: p[0] = p[1] = p[2] = p[3]
= p[4] = p[5] = p[6] = p[7]
= p[8] = 0UL; return;
case 8: p[0] = p[1] = p[2] = p[3]
= p[4] = p[5] = p[6] = p[7] = 0UL; return;
case 7: p[0] = p[1] = p[2] = p[3]
= p[4] = p[5] = p[6] = 0UL; return;
case 6: p[0] = p[1] = p[2] = p[3]
= p[4] = p[5] = 0UL; return;
case 5: p[0] = p[1] = p[2] = p[3] = p[4] = 0UL; return;
case 4: p[0] = p[1] = p[2] = p[3] = 0UL; return;
case 3: p[0] = p[1] = p[2] = 0UL; return;
case 2: p[0] = p[1] = 0UL; return;
case 1: p[0] = 0UL; return;
case 0: return;
default: break;
}
}
VG_(memset)(s, 0, sz);
}
/* ---------------------------------------------------------------------
Address computation helpers
------------------------------------------------------------------ */
// Check if an address/whatever is aligned
#define VG_IS_2_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)0x1)))
#define VG_IS_4_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)0x3)))
#define VG_IS_8_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)0x7)))
#define VG_IS_16_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)0xf)))
#define VG_IS_32_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)0x1f)))
#define VG_IS_64_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)0x3f)))
#define VG_IS_WORD_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)(sizeof(Addr)-1))))
#define VG_IS_PAGE_ALIGNED(aaa_p) (0 == (((Addr)(aaa_p)) & ((Addr)(VKI_PAGE_SIZE-1))))
// 'a' -- the alignment -- must be a power of 2.
// The latter two require the vki-*.h header to be imported also.
#define VG_ROUNDDN(p, a) ((Addr)(p) & ~((Addr)(a)-1))
#define VG_ROUNDUP(p, a) VG_ROUNDDN((p)+(a)-1, (a))
#define VG_PGROUNDDN(p) VG_ROUNDDN(p, VKI_PAGE_SIZE)
#define VG_PGROUNDUP(p) VG_ROUNDUP(p, VKI_PAGE_SIZE)
/* ---------------------------------------------------------------------
Misc useful functions
------------------------------------------------------------------ */
/* Like qsort(). The name VG_(ssort) is for historical reasons -- it used
* to be a shell sort, but is now a quicksort. */
extern void VG_(ssort)( void* base, SizeT nmemb, SizeT size,
Int (*compar)(const void*, const void*) );
/* Returns the base-2 logarithm of a 32 bit unsigned number. Returns
-1 if it is not a power of two. Nb: VG_(log2)(1) == 0. */
extern Int VG_(log2) ( UInt x );
/* Ditto for 64 bit unsigned numbers. */
extern Int VG_(log2_64)( ULong x );
// A pseudo-random number generator returning a random UInt. If pSeed
// is NULL, it uses its own seed, which starts at zero. If pSeed is
// non-NULL, it uses and updates whatever pSeed points at.
extern UInt VG_(random) ( /*MOD*/UInt* pSeed );
/* Update a running Adler-32 checksum with the bytes buf[0..len-1] and
return the updated checksum. If buf is NULL, this function returns
the required initial value for the checksum. An Adler-32 checksum is
almost as reliable as a CRC32 but can be computed much faster. */
extern UInt VG_(adler32)( UInt adler, const UChar* buf, UInt len);
#endif // __PUB_TOOL_LIBCBASE_H
/*--------------------------------------------------------------------*/
/*--- end ---*/
/*--------------------------------------------------------------------*/
Reference in New Issue View Git Blame Copy Permalink