testing/zlib
SHA256
3
0
Fork 0
forked from pool/zlib
Code Pull Requests Activity

Accepting request 1009574 from home:dspinella:branches:devel:libraries:c_c++

Browse Source 
- Add Power8 optimizations:
  * zlib-1.2.12-add-optimized-slide_hash-for-power.patch
  * zlib-1.2.12-add-vectorized-longest_match-for-power.patch
  * zlib-1.2.12-adler32-vector-optimizations-for-power.patch
  * zlib-1.2.12-fix-invalid-memory-access-on-ppc-and-ppc64.patch
- Update zlib-1.2.12-IBM-Z-hw-accelerated-deflate-s390x.patch

OBS-URL: https://build.opensuse.org/request/show/1009574
OBS-URL: https://build.opensuse.org/package/show/devel:libraries:c_c++/zlib?expand=0&rev=82
This commit is contained in:
Dirk Mueller 2022-10-11 12:40:29 +00:00 committed by Git OBS Bridge
parent 91abbb8bfa
commit f6c5a01beb
7 changed files with 1157 additions and 109 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

307
zlib-1.2.12-IBM-Z-hw-accelerated-deflate-s390x.patch
View File

@ -1,8 +1,99 @@
From e1d7e6dc9698968a8536b00ebd9e9b4e429b4306 Mon Sep 17 00:00:00 2001 From 171d0ff3c9ed40da0ac14085ab16b766b1162069 Mon Sep 17 00:00:00 2001
From: Ilya Leoshkevich <iii@linux.ibm.com> From: Ilya Leoshkevich <iii@linux.ibm.com>
Date: Wed, 27 Apr 2022 14:37:24 +0200 Date: Wed, 18 Jul 2018 13:14:07 +0200
Subject: [PATCH] zlib-1.2.12-IBM-Z-hw-accelrated-deflate-s390x.patch Subject: [PATCH] Add support for IBM Z hardware-accelerated deflate
IBM Z mainframes starting from version z15 provide DFLTCC instruction,
which implements deflate algorithm in hardware with estimated
compression and decompression performance orders of magnitude faster
than the current zlib and ratio comparable with that of level 1.
This patch adds DFLTCC support to zlib. In order to enable it, the
following build commands should be used:
$ ./configure --dfltcc
$ make
When built like this, zlib would compress in hardware on level 1, and in
software on all other levels. Decompression will always happen in
hardware. In order to enable DFLTCC compression for levels 1-6 (i.e. to
make it used by default) one could either configure with
--dfltcc-level-mask=0x7e or set the environment variable
DFLTCC_LEVEL_MASK to 0x7e at run time.
Two DFLTCC compression calls produce the same results only when they
both are made on machines of the same generation, and when the
respective buffers have the same offset relative to the start of the
page. Therefore care should be taken when using hardware compression
when reproducible results are desired. One such use case - reproducible
software builds - is handled explicitly: when SOURCE_DATE_EPOCH
environment variable is set, the hardware compression is disabled.
DFLTCC does not support every single zlib feature, in particular:
* inflate(Z_BLOCK) and inflate(Z_TREES)
* inflateMark()
* inflatePrime()
* inflateSyncPoint()
When used, these functions will either switch to software, or, in case
this is not possible, gracefully fail.
This patch tries to add DFLTCC support in the least intrusive way.
All SystemZ-specific code is placed into a separate file, but
unfortunately there is still a noticeable amount of changes in the
main zlib code. Below is the summary of these changes.
DFLTCC takes as arguments a parameter block, an input buffer, an output
buffer and a window. Since DFLTCC requires parameter block to be
doubleword-aligned, and it's reasonable to allocate it alongside
deflate and inflate states, ZALLOC_STATE, ZFREE_STATE and ZCOPY_STATE
macros were introduced in order to encapsulate the allocation details.
The same is true for window, for which ZALLOC_WINDOW and
TRY_FREE_WINDOW macros were introduced.
While for inflate software and hardware window formats match, this is
not the case for deflate. Therefore, deflateSetDictionary and
deflateGetDictionary need special handling, which is triggered using the
new DEFLATE_SET_DICTIONARY_HOOK and DEFLATE_GET_DICTIONARY_HOOK macros.
deflateResetKeep() and inflateResetKeep() now update the DFLTCC
parameter block, which is allocated alongside zlib state, using
the new DEFLATE_RESET_KEEP_HOOK and INFLATE_RESET_KEEP_HOOK macros.
The new DEFLATE_PARAMS_HOOK switches between hardware and software
deflate implementations when deflateParams() arguments demand this.
The new INFLATE_PRIME_HOOK, INFLATE_MARK_HOOK and
INFLATE_SYNC_POINT_HOOK macros make the respective unsupported calls
gracefully fail.
The algorithm implemented in hardware has different compression ratio
than the one implemented in software. In order for deflateBound() to
return the correct results for the hardware implementation, the new
DEFLATE_BOUND_ADJUST_COMPLEN and DEFLATE_NEED_CONSERVATIVE_BOUND macros
were introduced.
Actual compression and decompression are handled by the new DEFLATE_HOOK
and INFLATE_TYPEDO_HOOK macros. Since inflation with DFLTCC manages the
window on its own, calling updatewindow() is suppressed using the new
INFLATE_NEED_UPDATEWINDOW() macro.
In addition to compression, DFLTCC computes CRC-32 and Adler-32
checksums, therefore, whenever it's used, software checksumming needs to
be suppressed using the new DEFLATE_NEED_CHECKSUM and
INFLATE_NEED_CHECKSUM macros.
DFLTCC will refuse to write an End-of-block Symbol if there is no input
data, thus in some cases it is necessary to do this manually. In order
to achieve this, send_bits, bi_reverse, bi_windup and flush_pending
were promoted from local to ZLIB_INTERNAL. Furthermore, since block and
stream termination must be handled in software as well, block_state enum
was moved to deflate.h.
Since the first call to dfltcc_inflate already needs the window, and it
might be not allocated yet, inflate_ensure_window was factored out of
updatewindow and made ZLIB_INTERNAL.
--- ---
Makefile.in | 8 + Makefile.in | 8 +
compress.c | 14 +- compress.c | 14 +-
@ -27,10 +118,10 @@ Subject: [PATCH] zlib-1.2.12-IBM-Z-hw-accelrated-deflate-s390x.patch
create mode 100644 contrib/s390/dfltcc.h create mode 100644 contrib/s390/dfltcc.h
create mode 100644 contrib/s390/dfltcc_deflate.h create mode 100644 contrib/s390/dfltcc_deflate.h
Index: zlib-1.2.12/Makefile.in diff --git a/Makefile.in b/Makefile.in
=================================================================== index fd28bbfbf..66e3a8057 100644
--- zlib-1.2.12.orig/Makefile.in --- a/Makefile.in
+++ zlib-1.2.12/Makefile.in +++ b/Makefile.in
@@ -143,6 +143,14 @@ match.lo: match.S @@ -143,6 +143,14 @@ match.lo: match.S
mv _match.o match.lo mv _match.o match.lo
rm -f _match.s rm -f _match.s
@ -46,10 +137,10 @@ Index: zlib-1.2.12/Makefile.in
example.o: $(SRCDIR)test/example.c $(SRCDIR)zlib.h zconf.h example.o: $(SRCDIR)test/example.c $(SRCDIR)zlib.h zconf.h
$(CC) $(CFLAGS) $(ZINCOUT) -c -o $@ $(SRCDIR)test/example.c $(CC) $(CFLAGS) $(ZINCOUT) -c -o $@ $(SRCDIR)test/example.c
Index: zlib-1.2.12/compress.c diff --git a/compress.c b/compress.c
=================================================================== index e2db404ab..78fc6568f 100644
--- zlib-1.2.12.orig/compress.c --- a/compress.c
+++ zlib-1.2.12/compress.c +++ b/compress.c
@@ -5,9 +5,15 @@ @@ -5,9 +5,15 @@
/* @(#) $Id$ */ /* @(#) $Id$ */
@ -67,7 +158,7 @@ Index: zlib-1.2.12/compress.c
/* =========================================================================== /* ===========================================================================
Compresses the source buffer into the destination buffer. The level Compresses the source buffer into the destination buffer. The level
parameter has the same meaning as in deflateInit. sourceLen is the byte parameter has the same meaning as in deflateInit. sourceLen is the byte
@@ -81,6 +87,12 @@ int ZEXPORT compress (dest, destLen, sou @@ -81,6 +87,12 @@ int ZEXPORT compress (dest, destLen, source, sourceLen)
uLong ZEXPORT compressBound (sourceLen) uLong ZEXPORT compressBound (sourceLen)
uLong sourceLen; uLong sourceLen;
{ {
@ -80,10 +171,10 @@ Index: zlib-1.2.12/compress.c
return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) + return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
(sourceLen >> 25) + 13; (sourceLen >> 25) + 13;
} }
Index: zlib-1.2.12/configure diff --git a/configure b/configure
=================================================================== index fbaf25357..02e325e22 100755
--- zlib-1.2.12.orig/configure --- a/configure
+++ zlib-1.2.12/configure +++ b/configure
@@ -115,6 +115,7 @@ case "$1" in @@ -115,6 +115,7 @@ case "$1" in
echo ' configure [--const] [--zprefix] [--prefix=PREFIX] [--eprefix=EXPREFIX]' | tee -a configure.log echo ' configure [--const] [--zprefix] [--prefix=PREFIX] [--eprefix=EXPREFIX]' | tee -a configure.log
echo ' [--static] [--64] [--libdir=LIBDIR] [--sharedlibdir=LIBDIR]' | tee -a configure.log echo ' [--static] [--64] [--libdir=LIBDIR] [--sharedlibdir=LIBDIR]' | tee -a configure.log
@ -109,7 +200,7 @@ Index: zlib-1.2.12/configure
*) *)
echo "unknown option: $1" | tee -a configure.log echo "unknown option: $1" | tee -a configure.log
echo "$0 --help for help" | tee -a configure.log echo "$0 --help for help" | tee -a configure.log
@@ -833,6 +844,19 @@ EOF @@ -836,6 +847,19 @@ EOF
fi fi
fi fi
@ -129,11 +220,11 @@ Index: zlib-1.2.12/configure
# show the results in the log # show the results in the log
echo >> configure.log echo >> configure.log
echo ALL = $ALL >> configure.log echo ALL = $ALL >> configure.log
Index: zlib-1.2.12/contrib/README.contrib diff --git a/contrib/README.contrib b/contrib/README.contrib
=================================================================== index 335e43508..130a28bdb 100644
--- zlib-1.2.12.orig/contrib/README.contrib --- a/contrib/README.contrib
+++ zlib-1.2.12/contrib/README.contrib +++ b/contrib/README.contrib
@@ -46,6 +46,10 @@ puff/ by Mark Adler <madler@alumni @@ -46,6 +46,10 @@ puff/ by Mark Adler <madler@alumni.caltech.edu>
Small, low memory usage inflate. Also serves to provide an Small, low memory usage inflate. Also serves to provide an
unambiguous description of the deflate format. unambiguous description of the deflate format.
@ -144,10 +235,11 @@ Index: zlib-1.2.12/contrib/README.contrib
testzlib/ by Gilles Vollant <info@winimage.com> testzlib/ by Gilles Vollant <info@winimage.com>
Example of the use of zlib Example of the use of zlib
Index: zlib-1.2.12/contrib/s390/README.txt diff --git a/contrib/s390/README.txt b/contrib/s390/README.txt
=================================================================== new file mode 100644
index 000000000..48be008bd
--- /dev/null --- /dev/null
+++ zlib-1.2.12/contrib/s390/README.txt +++ b/contrib/s390/README.txt
@@ -0,0 +1,17 @@ @@ -0,0 +1,17 @@
+IBM Z mainframes starting from version z15 provide DFLTCC instruction, +IBM Z mainframes starting from version z15 provide DFLTCC instruction,
+which implements deflate algorithm in hardware with estimated +which implements deflate algorithm in hardware with estimated
@ -166,10 +258,11 @@ Index: zlib-1.2.12/contrib/s390/README.txt
+make it used by default) one could either configure with +make it used by default) one could either configure with
+--dfltcc-level-mask=0x7e or set the environment variable +--dfltcc-level-mask=0x7e or set the environment variable
+DFLTCC_LEVEL_MASK to 0x7e at run time. +DFLTCC_LEVEL_MASK to 0x7e at run time.
Index: zlib-1.2.12/contrib/s390/dfltcc.c diff --git a/contrib/s390/dfltcc.c b/contrib/s390/dfltcc.c
=================================================================== new file mode 100644
index 000000000..cd959290d
--- /dev/null --- /dev/null
+++ zlib-1.2.12/contrib/s390/dfltcc.c +++ b/contrib/s390/dfltcc.c
@@ -0,0 +1,996 @@ @@ -0,0 +1,996 @@
+/* dfltcc.c - SystemZ DEFLATE CONVERSION CALL support. */ +/* dfltcc.c - SystemZ DEFLATE CONVERSION CALL support. */
+ +
@ -797,7 +890,7 @@ Index: zlib-1.2.12/contrib/s390/dfltcc.c
+ state->bits = param->sbb; + state->bits = param->sbb;
+ state->whave = param->hl; + state->whave = param->hl;
+ state->wnext = (param->ho + param->hl) & ((1 << HB_BITS) - 1); + state->wnext = (param->ho + param->hl) & ((1 << HB_BITS) - 1);
+ state->check = state->flags ? ZSWAP32(param->cv) : param->cv; + strm->adler = state->check = state->flags ? ZSWAP32(param->cv) : param->cv;
+ if (cc == DFLTCC_CC_OP2_CORRUPT && param->oesc != 0) { + if (cc == DFLTCC_CC_OP2_CORRUPT && param->oesc != 0) {
+ /* Report an error if stream is corrupted */ + /* Report an error if stream is corrupted */
+ state->mode = BAD; + state->mode = BAD;
@ -1167,10 +1260,11 @@ Index: zlib-1.2.12/contrib/s390/dfltcc.c
+ *dict_length = param->hl; + *dict_length = param->hl;
+ return Z_OK; + return Z_OK;
+} +}
Index: zlib-1.2.12/contrib/s390/dfltcc.h diff --git a/contrib/s390/dfltcc.h b/contrib/s390/dfltcc.h
=================================================================== new file mode 100644
index 000000000..da26612ca
--- /dev/null --- /dev/null
+++ zlib-1.2.12/contrib/s390/dfltcc.h +++ b/contrib/s390/dfltcc.h
@@ -0,0 +1,81 @@ @@ -0,0 +1,81 @@
+#ifndef DFLTCC_H +#ifndef DFLTCC_H
+#define DFLTCC_H +#define DFLTCC_H
@ -1253,10 +1347,11 @@ Index: zlib-1.2.12/contrib/s390/dfltcc.h
+ } while (0) + } while (0)
+ +
+#endif +#endif
Index: zlib-1.2.12/contrib/s390/dfltcc_deflate.h diff --git a/contrib/s390/dfltcc_deflate.h b/contrib/s390/dfltcc_deflate.h
=================================================================== new file mode 100644
index 000000000..46acfc550
--- /dev/null --- /dev/null
+++ zlib-1.2.12/contrib/s390/dfltcc_deflate.h +++ b/contrib/s390/dfltcc_deflate.h
@@ -0,0 +1,55 @@ @@ -0,0 +1,55 @@
+#ifndef DFLTCC_DEFLATE_H +#ifndef DFLTCC_DEFLATE_H
+#define DFLTCC_DEFLATE_H +#define DFLTCC_DEFLATE_H
@ -1313,10 +1408,10 @@ Index: zlib-1.2.12/contrib/s390/dfltcc_deflate.h
+#define DEFLATE_NEED_CHECKSUM(strm) (!dfltcc_can_deflate((strm))) +#define DEFLATE_NEED_CHECKSUM(strm) (!dfltcc_can_deflate((strm)))
+ +
+#endif +#endif
Index: zlib-1.2.12/deflate.c diff --git a/deflate.c b/deflate.c
=================================================================== index 7f421e4da..a56c1783c 100644
--- zlib-1.2.12.orig/deflate.c --- a/deflate.c
+++ zlib-1.2.12/deflate.c +++ b/deflate.c
@@ -61,15 +61,30 @@ const char deflate_copyright[] = @@ -61,15 +61,30 @@ const char deflate_copyright[] =
*/ */
@ -1355,7 +1450,7 @@ Index: zlib-1.2.12/deflate.c
typedef block_state (*compress_func) OF((deflate_state *s, int flush)); typedef block_state (*compress_func) OF((deflate_state *s, int flush));
/* Compression function. Returns the block state after the call. */ /* Compression function. Returns the block state after the call. */
@@ -85,7 +100,6 @@ local block_state deflate_rle OF((def @@ -85,7 +100,6 @@ local block_state deflate_rle OF((deflate_state *s, int flush));
local block_state deflate_huff OF((deflate_state *s, int flush)); local block_state deflate_huff OF((deflate_state *s, int flush));
local void lm_init OF((deflate_state *s)); local void lm_init OF((deflate_state *s));
local void putShortMSB OF((deflate_state *s, uInt b)); local void putShortMSB OF((deflate_state *s, uInt b));
@ -1363,7 +1458,7 @@ Index: zlib-1.2.12/deflate.c
local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
#ifdef ASMV #ifdef ASMV
# pragma message("Assembler code may have bugs -- use at your own risk") # pragma message("Assembler code may have bugs -- use at your own risk")
@@ -294,7 +308,7 @@ int ZEXPORT deflateInit2_(strm, level, m @@ -299,7 +313,7 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
return Z_STREAM_ERROR; return Z_STREAM_ERROR;
} }
if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
@ -1372,7 +1467,7 @@ Index: zlib-1.2.12/deflate.c
if (s == Z_NULL) return Z_MEM_ERROR; if (s == Z_NULL) return Z_MEM_ERROR;
strm->state = (struct internal_state FAR *)s; strm->state = (struct internal_state FAR *)s;
s->strm = strm; s->strm = strm;
@@ -311,7 +325,7 @@ int ZEXPORT deflateInit2_(strm, level, m @@ -316,7 +330,7 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
s->hash_mask = s->hash_size - 1; s->hash_mask = s->hash_size - 1;
s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
@ -1381,7 +1476,7 @@ Index: zlib-1.2.12/deflate.c
s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
@@ -429,6 +443,7 @@ int ZEXPORT deflateSetDictionary (strm, @@ -434,6 +448,7 @@ int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
/* when using zlib wrappers, compute Adler-32 for provided dictionary */ /* when using zlib wrappers, compute Adler-32 for provided dictionary */
if (wrap == 1) if (wrap == 1)
strm->adler = adler32(strm->adler, dictionary, dictLength); strm->adler = adler32(strm->adler, dictionary, dictLength);
@ -1389,7 +1484,7 @@ Index: zlib-1.2.12/deflate.c
s->wrap = 0; /* avoid computing Adler-32 in read_buf */ s->wrap = 0; /* avoid computing Adler-32 in read_buf */
/* if dictionary would fill window, just replace the history */ /* if dictionary would fill window, just replace the history */
@@ -487,6 +502,7 @@ int ZEXPORT deflateGetDictionary (strm, @@ -492,6 +507,7 @@ int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
if (deflateStateCheck(strm)) if (deflateStateCheck(strm))
return Z_STREAM_ERROR; return Z_STREAM_ERROR;
@ -1397,7 +1492,7 @@ Index: zlib-1.2.12/deflate.c
s = strm->state; s = strm->state;
len = s->strstart + s->lookahead; len = s->strstart + s->lookahead;
if (len > s->w_size) if (len > s->w_size)
@@ -533,6 +549,8 @@ int ZEXPORT deflateResetKeep (strm) @@ -538,6 +554,8 @@ int ZEXPORT deflateResetKeep (strm)
_tr_init(s); _tr_init(s);
@ -1406,7 +1501,7 @@ Index: zlib-1.2.12/deflate.c
return Z_OK; return Z_OK;
} }
@@ -608,6 +626,7 @@ int ZEXPORT deflateParams(strm, level, s @@ -613,6 +631,7 @@ int ZEXPORT deflateParams(strm, level, strategy)
{ {
deflate_state *s; deflate_state *s;
compress_func func; compress_func func;
@ -1414,7 +1509,7 @@ Index: zlib-1.2.12/deflate.c
if (deflateStateCheck(strm)) return Z_STREAM_ERROR; if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
s = strm->state; s = strm->state;
@@ -620,15 +639,18 @@ int ZEXPORT deflateParams(strm, level, s @@ -625,15 +644,18 @@ int ZEXPORT deflateParams(strm, level, strategy)
if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
return Z_STREAM_ERROR; return Z_STREAM_ERROR;
} }
@ -1437,7 +1532,7 @@ Index: zlib-1.2.12/deflate.c
return Z_BUF_ERROR; return Z_BUF_ERROR;
} }
if (s->level != level) { if (s->level != level) {
@@ -695,6 +717,7 @@ uLong ZEXPORT deflateBound(strm, sourceL @@ -700,6 +722,7 @@ uLong ZEXPORT deflateBound(strm, sourceLen)
/* conservative upper bound for compressed data */ /* conservative upper bound for compressed data */
complen = sourceLen + complen = sourceLen +
((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5; ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
@ -1445,7 +1540,7 @@ Index: zlib-1.2.12/deflate.c
/* if can't get parameters, return conservative bound plus zlib wrapper */ /* if can't get parameters, return conservative bound plus zlib wrapper */
if (deflateStateCheck(strm)) if (deflateStateCheck(strm))
@@ -736,7 +759,8 @@ uLong ZEXPORT deflateBound(strm, sourceL @@ -741,7 +764,8 @@ uLong ZEXPORT deflateBound(strm, sourceLen)
} }
/* if not default parameters, return conservative bound */ /* if not default parameters, return conservative bound */
@ -1455,7 +1550,7 @@ Index: zlib-1.2.12/deflate.c
return complen + wraplen; return complen + wraplen;
/* default settings: return tight bound for that case */ /* default settings: return tight bound for that case */
@@ -763,7 +787,7 @@ local void putShortMSB (s, b) @@ -768,7 +792,7 @@ local void putShortMSB (s, b)
* applications may wish to modify it to avoid allocating a large * applications may wish to modify it to avoid allocating a large
* strm->next_out buffer and copying into it. (See also read_buf()). * strm->next_out buffer and copying into it. (See also read_buf()).
*/ */
@ -1464,7 +1559,7 @@ Index: zlib-1.2.12/deflate.c
z_streamp strm; z_streamp strm;
{ {
unsigned len; unsigned len;
@@ -1035,7 +1059,8 @@ int ZEXPORT deflate (strm, flush) @@ -1040,7 +1064,8 @@ int ZEXPORT deflate (strm, flush)
(flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
block_state bstate; block_state bstate;
@ -1474,7 +1569,7 @@ Index: zlib-1.2.12/deflate.c
s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
s->strategy == Z_RLE ? deflate_rle(s, flush) : s->strategy == Z_RLE ? deflate_rle(s, flush) :
(*(configuration_table[s->level].func))(s, flush); (*(configuration_table[s->level].func))(s, flush);
@@ -1082,7 +1107,6 @@ int ZEXPORT deflate (strm, flush) @@ -1087,7 +1112,6 @@ int ZEXPORT deflate (strm, flush)
} }
if (flush != Z_FINISH) return Z_OK; if (flush != Z_FINISH) return Z_OK;
@ -1482,7 +1577,7 @@ Index: zlib-1.2.12/deflate.c
/* Write the trailer */ /* Write the trailer */
#ifdef GZIP #ifdef GZIP
@@ -1098,7 +1122,7 @@ int ZEXPORT deflate (strm, flush) @@ -1103,7 +1127,7 @@ int ZEXPORT deflate (strm, flush)
} }
else else
#endif #endif
@ -1491,7 +1586,7 @@ Index: zlib-1.2.12/deflate.c
putShortMSB(s, (uInt)(strm->adler >> 16)); putShortMSB(s, (uInt)(strm->adler >> 16));
putShortMSB(s, (uInt)(strm->adler & 0xffff)); putShortMSB(s, (uInt)(strm->adler & 0xffff));
} }
@@ -1107,7 +1131,11 @@ int ZEXPORT deflate (strm, flush) @@ -1112,7 +1136,11 @@ int ZEXPORT deflate (strm, flush)
* to flush the rest. * to flush the rest.
*/ */
if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
@ -1504,7 +1599,7 @@ Index: zlib-1.2.12/deflate.c
} }
/* ========================================================================= */ /* ========================================================================= */
@@ -1124,9 +1152,9 @@ int ZEXPORT deflateEnd (strm) @@ -1129,9 +1157,9 @@ int ZEXPORT deflateEnd (strm)
TRY_FREE(strm, strm->state->pending_buf); TRY_FREE(strm, strm->state->pending_buf);
TRY_FREE(strm, strm->state->head); TRY_FREE(strm, strm->state->head);
TRY_FREE(strm, strm->state->prev); TRY_FREE(strm, strm->state->prev);
@ -1516,7 +1611,7 @@ Index: zlib-1.2.12/deflate.c
strm->state = Z_NULL; strm->state = Z_NULL;
return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
@@ -1156,13 +1184,13 @@ int ZEXPORT deflateCopy (dest, source) @@ -1161,13 +1189,13 @@ int ZEXPORT deflateCopy (dest, source)
zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream)); zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
@ -1533,7 +1628,7 @@ Index: zlib-1.2.12/deflate.c
ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
ds->pending_buf = (uchf *) ZALLOC(dest, ds->lit_bufsize, 4); ds->pending_buf = (uchf *) ZALLOC(dest, ds->lit_bufsize, 4);
@@ -1209,7 +1237,8 @@ local unsigned read_buf(strm, buf, size) @@ -1214,7 +1242,8 @@ local unsigned read_buf(strm, buf, size)
strm->avail_in -= len; strm->avail_in -= len;
zmemcpy(buf, strm->next_in, len); zmemcpy(buf, strm->next_in, len);
@ -1543,11 +1638,11 @@ Index: zlib-1.2.12/deflate.c
strm->adler = adler32(strm->adler, buf, len); strm->adler = adler32(strm->adler, buf, len);
} }
#ifdef GZIP #ifdef GZIP
Index: zlib-1.2.12/deflate.h diff --git a/deflate.h b/deflate.h
=================================================================== index 1a06cd5f2..f92750ca6 100644
--- zlib-1.2.12.orig/deflate.h --- a/deflate.h
+++ zlib-1.2.12/deflate.h +++ b/deflate.h
@@ -299,6 +299,7 @@ void ZLIB_INTERNAL _tr_flush_bits OF((de @@ -299,6 +299,7 @@ void ZLIB_INTERNAL _tr_flush_bits OF((deflate_state *s));
void ZLIB_INTERNAL _tr_align OF((deflate_state *s)); void ZLIB_INTERNAL _tr_align OF((deflate_state *s));
void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf, void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf,
ulg stored_len, int last)); ulg stored_len, int last));
@ -1555,7 +1650,7 @@ Index: zlib-1.2.12/deflate.h
#define d_code(dist) \ #define d_code(dist) \
((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)])
@@ -343,4 +344,15 @@ void ZLIB_INTERNAL _tr_stored_block OF(( @@ -343,4 +344,15 @@ void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf,
flush = _tr_tally(s, distance, length) flush = _tr_tally(s, distance, length)
#endif #endif
@ -1571,10 +1666,10 @@ Index: zlib-1.2.12/deflate.h
+void ZLIB_INTERNAL flush_pending OF((z_streamp strm)); +void ZLIB_INTERNAL flush_pending OF((z_streamp strm));
+ +
#endif /* DEFLATE_H */ #endif /* DEFLATE_H */
Index: zlib-1.2.12/gzguts.h diff --git a/gzguts.h b/gzguts.h
=================================================================== index 57faf3716..581f2b631 100644
--- zlib-1.2.12.orig/gzguts.h --- a/gzguts.h
+++ zlib-1.2.12/gzguts.h +++ b/gzguts.h
@@ -153,7 +153,11 @@ @@ -153,7 +153,11 @@
/* default i/o buffer size -- double this for output when reading (this and /* default i/o buffer size -- double this for output when reading (this and
@ -1587,10 +1682,10 @@ Index: zlib-1.2.12/gzguts.h
/* gzip modes, also provide a little integrity check on the passed structure */ /* gzip modes, also provide a little integrity check on the passed structure */
#define GZ_NONE 0 #define GZ_NONE 0
Index: zlib-1.2.12/inflate.c diff --git a/inflate.c b/inflate.c
=================================================================== index 2a3c4fe98..ca0f8c9a4 100644
--- zlib-1.2.12.orig/inflate.c --- a/inflate.c
+++ zlib-1.2.12/inflate.c +++ b/inflate.c
@@ -85,6 +85,24 @@ @@ -85,6 +85,24 @@
#include "inflate.h" #include "inflate.h"
#include "inffast.h" #include "inffast.h"
@ -1633,7 +1728,7 @@ Index: zlib-1.2.12/inflate.c
state->window = Z_NULL; state->window = Z_NULL;
} }
@@ -219,7 +238,7 @@ int stream_size; @@ -222,7 +241,7 @@ int stream_size;
strm->zfree = zcfree; strm->zfree = zcfree;
#endif #endif
state = (struct inflate_state FAR *) state = (struct inflate_state FAR *)
@ -1642,7 +1737,7 @@ Index: zlib-1.2.12/inflate.c
if (state == Z_NULL) return Z_MEM_ERROR; if (state == Z_NULL) return Z_MEM_ERROR;
Tracev((stderr, "inflate: allocated\n")); Tracev((stderr, "inflate: allocated\n"));
strm->state = (struct internal_state FAR *)state; strm->state = (struct internal_state FAR *)state;
@@ -228,7 +247,7 @@ int stream_size; @@ -231,7 +250,7 @@ int stream_size;
state->mode = HEAD; /* to pass state test in inflateReset2() */ state->mode = HEAD; /* to pass state test in inflateReset2() */
ret = inflateReset2(strm, windowBits); ret = inflateReset2(strm, windowBits);
if (ret != Z_OK) { if (ret != Z_OK) {
@ -1651,7 +1746,7 @@ Index: zlib-1.2.12/inflate.c
strm->state = Z_NULL; strm->state = Z_NULL;
} }
return ret; return ret;
@@ -250,6 +269,7 @@ int value; @@ -253,6 +272,7 @@ int value;
struct inflate_state FAR *state; struct inflate_state FAR *state;
if (inflateStateCheck(strm)) return Z_STREAM_ERROR; if (inflateStateCheck(strm)) return Z_STREAM_ERROR;
@ -1659,7 +1754,7 @@ Index: zlib-1.2.12/inflate.c
state = (struct inflate_state FAR *)strm->state; state = (struct inflate_state FAR *)strm->state;
if (bits < 0) { if (bits < 0) {
state->hold = 0; state->hold = 0;
@@ -377,6 +397,27 @@ void makefixed() @@ -380,6 +400,27 @@ void makefixed()
} }
#endif /* MAKEFIXED */ #endif /* MAKEFIXED */
@ -1687,7 +1782,7 @@ Index: zlib-1.2.12/inflate.c
/* /*
Update the window with the last wsize (normally 32K) bytes written before Update the window with the last wsize (normally 32K) bytes written before
returning. If window does not exist yet, create it. This is only called returning. If window does not exist yet, create it. This is only called
@@ -401,20 +442,7 @@ unsigned copy; @@ -404,20 +445,7 @@ unsigned copy;
state = (struct inflate_state FAR *)strm->state; state = (struct inflate_state FAR *)strm->state;
@ -1709,7 +1804,7 @@ Index: zlib-1.2.12/inflate.c
/* copy state->wsize or less output bytes into the circular window */ /* copy state->wsize or less output bytes into the circular window */
if (copy >= state->wsize) { if (copy >= state->wsize) {
@@ -857,6 +885,7 @@ int flush; @@ -861,6 +889,7 @@ int flush;
if (flush == Z_BLOCK || flush == Z_TREES) goto inf_leave; if (flush == Z_BLOCK || flush == Z_TREES) goto inf_leave;
/* fallthrough */ /* fallthrough */
case TYPEDO: case TYPEDO:
@ -1717,7 +1812,7 @@ Index: zlib-1.2.12/inflate.c
if (state->last) { if (state->last) {
BYTEBITS(); BYTEBITS();
state->mode = CHECK; state->mode = CHECK;
@@ -1218,7 +1247,7 @@ int flush; @@ -1222,7 +1251,7 @@ int flush;
out -= left; out -= left;
strm->total_out += out; strm->total_out += out;
state->total += out; state->total += out;
@ -1726,7 +1821,7 @@ Index: zlib-1.2.12/inflate.c
strm->adler = state->check = strm->adler = state->check =
UPDATE_CHECK(state->check, put - out, out); UPDATE_CHECK(state->check, put - out, out);
out = left; out = left;
@@ -1273,8 +1302,9 @@ int flush; @@ -1277,8 +1306,9 @@ int flush;
*/ */
inf_leave: inf_leave:
RESTORE(); RESTORE();
@ -1738,7 +1833,7 @@ Index: zlib-1.2.12/inflate.c
if (updatewindow(strm, strm->next_out, out - strm->avail_out)) { if (updatewindow(strm, strm->next_out, out - strm->avail_out)) {
state->mode = MEM; state->mode = MEM;
return Z_MEM_ERROR; return Z_MEM_ERROR;
@@ -1284,7 +1314,7 @@ int flush; @@ -1288,7 +1318,7 @@ int flush;
strm->total_in += in; strm->total_in += in;
strm->total_out += out; strm->total_out += out;
state->total += out; state->total += out;
@ -1747,7 +1842,7 @@ Index: zlib-1.2.12/inflate.c
strm->adler = state->check = strm->adler = state->check =
UPDATE_CHECK(state->check, strm->next_out - out, out); UPDATE_CHECK(state->check, strm->next_out - out, out);
strm->data_type = (int)state->bits + (state->last ? 64 : 0) + strm->data_type = (int)state->bits + (state->last ? 64 : 0) +
@@ -1302,8 +1332,8 @@ z_streamp strm; @@ -1306,8 +1336,8 @@ z_streamp strm;
if (inflateStateCheck(strm)) if (inflateStateCheck(strm))
return Z_STREAM_ERROR; return Z_STREAM_ERROR;
state = (struct inflate_state FAR *)strm->state; state = (struct inflate_state FAR *)strm->state;
@ -1758,7 +1853,7 @@ Index: zlib-1.2.12/inflate.c
strm->state = Z_NULL; strm->state = Z_NULL;
Tracev((stderr, "inflate: end\n")); Tracev((stderr, "inflate: end\n"));
return Z_OK; return Z_OK;
@@ -1482,6 +1512,7 @@ z_streamp strm; @@ -1486,6 +1516,7 @@ z_streamp strm;
struct inflate_state FAR *state; struct inflate_state FAR *state;
if (inflateStateCheck(strm)) return Z_STREAM_ERROR; if (inflateStateCheck(strm)) return Z_STREAM_ERROR;
@ -1766,7 +1861,7 @@ Index: zlib-1.2.12/inflate.c
state = (struct inflate_state FAR *)strm->state; state = (struct inflate_state FAR *)strm->state;
return state->mode == STORED && state->bits == 0; return state->mode == STORED && state->bits == 0;
} }
@@ -1502,21 +1533,22 @@ z_streamp source; @@ -1506,21 +1537,22 @@ z_streamp source;
/* allocate space */ /* allocate space */
copy = (struct inflate_state FAR *) copy = (struct inflate_state FAR *)
@ -1793,7 +1888,7 @@ Index: zlib-1.2.12/inflate.c
copy->strm = dest; copy->strm = dest;
if (state->lencode >= state->codes && if (state->lencode >= state->codes &&
state->lencode <= state->codes + ENOUGH - 1) { state->lencode <= state->codes + ENOUGH - 1) {
@@ -1573,6 +1605,7 @@ z_streamp strm; @@ -1577,6 +1609,7 @@ z_streamp strm;
if (inflateStateCheck(strm)) if (inflateStateCheck(strm))
return -(1L << 16); return -(1L << 16);
@ -1801,20 +1896,20 @@ Index: zlib-1.2.12/inflate.c
state = (struct inflate_state FAR *)strm->state; state = (struct inflate_state FAR *)strm->state;
return (long)(((unsigned long)((long)state->back)) << 16) + return (long)(((unsigned long)((long)state->back)) << 16) +
(state->mode == COPY ? state->length : (state->mode == COPY ? state->length :
Index: zlib-1.2.12/inflate.h diff --git a/inflate.h b/inflate.h
=================================================================== index f127b6b1f..519ed3535 100644
--- zlib-1.2.12.orig/inflate.h --- a/inflate.h
+++ zlib-1.2.12/inflate.h +++ b/inflate.h
@@ -124,3 +124,5 @@ struct inflate_state { @@ -124,3 +124,5 @@ struct inflate_state {
int back; /* bits back of last unprocessed length/lit */ int back; /* bits back of last unprocessed length/lit */
unsigned was; /* initial length of match */ unsigned was; /* initial length of match */
}; };
+ +
+int ZLIB_INTERNAL inflate_ensure_window OF((struct inflate_state *state)); +int ZLIB_INTERNAL inflate_ensure_window OF((struct inflate_state *state));
Index: zlib-1.2.12/test/infcover.c diff --git a/test/infcover.c b/test/infcover.c
=================================================================== index 2be01646c..a208219dc 100644
--- zlib-1.2.12.orig/test/infcover.c --- a/test/infcover.c
+++ zlib-1.2.12/test/infcover.c +++ b/test/infcover.c
@@ -373,7 +373,7 @@ local void cover_support(void) @@ -373,7 +373,7 @@ local void cover_support(void)
mem_setup(&strm); mem_setup(&strm);
strm.avail_in = 0; strm.avail_in = 0;
@ -1833,10 +1928,10 @@ Index: zlib-1.2.12/test/infcover.c
{ {
static unsigned int next = 0; static unsigned int next = 0;
static unsigned char dat[] = {0x63, 0, 2, 0}; static unsigned char dat[] = {0x63, 0, 2, 0};
Index: zlib-1.2.12/test/minigzip.c diff --git a/test/minigzip.c b/test/minigzip.c
=================================================================== index e22fb08c0..4b5f4efed 100644
--- zlib-1.2.12.orig/test/minigzip.c --- a/test/minigzip.c
+++ zlib-1.2.12/test/minigzip.c +++ b/test/minigzip.c
@@ -132,7 +132,11 @@ static void pwinerror (s) @@ -132,7 +132,11 @@ static void pwinerror (s)
#endif #endif
#define SUFFIX_LEN (sizeof(GZ_SUFFIX)-1) #define SUFFIX_LEN (sizeof(GZ_SUFFIX)-1)
@ -1849,11 +1944,11 @@ Index: zlib-1.2.12/test/minigzip.c
#define MAX_NAME_LEN 1024 #define MAX_NAME_LEN 1024
#ifdef MAXSEG_64K #ifdef MAXSEG_64K
Index: zlib-1.2.12/trees.c diff --git a/trees.c b/trees.c
=================================================================== index 72b521fb0..534f29c98 100644
--- zlib-1.2.12.orig/trees.c --- a/trees.c
+++ zlib-1.2.12/trees.c +++ b/trees.c
@@ -149,8 +149,6 @@ local void send_all_trees OF((deflate_st @@ -149,8 +149,6 @@ local void send_all_trees OF((deflate_state *s, int lcodes, int dcodes,
local void compress_block OF((deflate_state *s, const ct_data *ltree, local void compress_block OF((deflate_state *s, const ct_data *ltree,
const ct_data *dtree)); const ct_data *dtree));
local int detect_data_type OF((deflate_state *s)); local int detect_data_type OF((deflate_state *s));
@ -1894,11 +1989,11 @@ Index: zlib-1.2.12/trees.c
deflate_state *s; deflate_state *s;
{ {
if (s->bi_valid > 8) { if (s->bi_valid > 8) {
Index: zlib-1.2.12/zutil.h diff --git a/zutil.h b/zutil.h
=================================================================== index d9a20ae1b..bc83f59d0 100644
--- zlib-1.2.12.orig/zutil.h --- a/zutil.h
+++ zlib-1.2.12/zutil.h +++ b/zutil.h
@@ -87,6 +87,8 @@ extern z_const char * const z_errmsg[10] @@ -87,6 +87,8 @@ extern z_const char * const z_errmsg[10]; /* indexed by 2-zlib_error */
#define PRESET_DICT 0x20 /* preset dictionary flag in zlib header */ #define PRESET_DICT 0x20 /* preset dictionary flag in zlib header */

219
zlib-1.2.12-add-optimized-slide_hash-for-power.patch Normal file
View File

@ -0,0 +1,219 @@
From 4a8d89ae49aa17d1634a2816c8d159f533a07eae Mon Sep 17 00:00:00 2001
From: Matheus Castanho <msc@linux.ibm.com>
Date: Wed, 27 Nov 2019 10:18:10 -0300
Subject: [PATCH] Add optimized slide_hash for Power
Considerable time is spent on deflate.c:slide_hash() during
deflate. This commit introduces a new slide_hash function that
uses VSX vector instructions to slide 8 hash elements at a time,
instead of just one as the standard code does.
The choice between the optimized and default versions is made only
on the first call to the function, enabling a fallback to standard
behavior if the host processor does not support VSX instructions,
so the same binary can be used for multiple Power processor
versions.
Author: Matheus Castanho <msc@linux.ibm.com>
---
CMakeLists.txt | 3 +-
Makefile.in | 8 ++++
configure | 4 +-
contrib/power/power.h | 3 ++
contrib/power/slide_hash_power8.c | 63 +++++++++++++++++++++++++++++
contrib/power/slide_hash_resolver.c | 15 +++++++
deflate.c | 12 ++++++
7 files changed, 105 insertions(+), 3 deletions(-)
create mode 100644 contrib/power/slide_hash_power8.c
create mode 100644 contrib/power/slide_hash_resolver.c
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 44de486f6..8208c626b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -186,7 +186,8 @@ if(CMAKE_COMPILER_IS_GNUCC)
add_definitions(-DZ_POWER8)
set(ZLIB_POWER8
contrib/power/adler32_power8.c
- contrib/power/crc32_z_power8.c)
+ contrib/power/crc32_z_power8.c
+ contrib/power/slide_hash_power8.c)
set_source_files_properties(
${ZLIB_POWER8}
diff --git a/Makefile.in b/Makefile.in
index 9ef9fa9b5..f71c6eae0 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -183,6 +183,9 @@ crc32_z_power8.o: $(SRCDIR)contrib/power/crc32_z_power8.c
deflate.o: $(SRCDIR)deflate.c
$(CC) $(CFLAGS) $(ZINC) -c -o $@ $(SRCDIR)deflate.c
+slide_hash_power8.o: $(SRCDIR)contrib/power/slide_hash_power8.c
+ $(CC) $(CFLAGS) -mcpu=power8 $(ZINC) -c -o $@ $(SRCDIR)contrib/power/slide_hash_power8.c
+
infback.o: $(SRCDIR)infback.c
$(CC) $(CFLAGS) $(ZINC) -c -o $@ $(SRCDIR)infback.c
@@ -245,6 +248,11 @@ deflate.lo: $(SRCDIR)deflate.c
$(CC) $(SFLAGS) $(ZINC) -DPIC -c -o objs/deflate.o $(SRCDIR)deflate.c
-@mv objs/deflate.o $@
+slide_hash_power8.lo: $(SRCDIR)contrib/power/slide_hash_power8.c
+ -@mkdir objs 2>/dev/null || test -d objs
+ $(CC) $(SFLAGS) -mcpu=power8 $(ZINC) -DPIC -c -o objs/slide_hash_power8.o $(SRCDIR)contrib/power/slide_hash_power8.c
+ -@mv objs/slide_hash_power8.o $@
+
infback.lo: $(SRCDIR)infback.c
-@mkdir objs 2>/dev/null || test -d objs
$(CC) $(SFLAGS) $(ZINC) -DPIC -c -o objs/infback.o $(SRCDIR)infback.c
diff --git a/configure b/configure
index 810a7404d..d0dacf9c2 100755
--- a/configure
+++ b/configure
@@ -879,8 +879,8 @@ if tryboth $CC -c $CFLAGS $test.c; then
if tryboth $CC -c $CFLAGS -mcpu=power8 $test.c; then
POWER8="-DZ_POWER8"
- PIC_OBJC="${PIC_OBJC} adler32_power8.lo crc32_z_power8.lo"
- OBJC="${OBJC} adler32_power8.o crc32_z_power8.o"
+ PIC_OBJC="${PIC_OBJC} adler32_power8.lo crc32_z_power8.lo slide_hash_power8.lo"
+ OBJC="${OBJC} adler32_power8.o crc32_z_power8.o slide_hash_power8.o"
echo "Checking for -mcpu=power8 support... Yes." | tee -a configure.log
else
echo "Checking for -mcpu=power8 support... No." | tee -a configure.log
diff --git a/contrib/power/power.h b/contrib/power/power.h
index f57c76167..28c8f78ca 100644
--- a/contrib/power/power.h
+++ b/contrib/power/power.h
@@ -4,7 +4,10 @@
*/
#include "../../zconf.h"
#include "../../zutil.h"
+#include "../../deflate.h"
uLong _adler32_power8(uLong adler, const Bytef* buf, uInt len);
unsigned long _crc32_z_power8(unsigned long, const Bytef *, z_size_t);
+
+void _slide_hash_power8(deflate_state *s);
diff --git a/contrib/power/slide_hash_power8.c b/contrib/power/slide_hash_power8.c
new file mode 100644
index 000000000..c5a0eb5a6
--- /dev/null
+++ b/contrib/power/slide_hash_power8.c
@@ -0,0 +1,63 @@
+ /* Copyright (C) 2019 Matheus Castanho <msc@linux.ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#include <altivec.h>
+#include "../../deflate.h"
+
+local inline void slide_hash_power8_loop OF((deflate_state *s,
+ unsigned n_elems, Posf *table_end)) __attribute__((always_inline));
+
+local void slide_hash_power8_loop(
+ deflate_state *s,
+ unsigned n_elems,
+ Posf *table_end)
+{
+ vector unsigned short vw, vm, *vp;
+ unsigned chunks;
+
+ /* Each vector register (chunk) corresponds to 128 bits == 8 Posf,
+ * so instead of processing each of the n_elems in the hash table
+ * individually, we can do it in chunks of 8 with vector instructions.
+ *
+ * This function is only called from slide_hash_power8(), and both calls
+ * pass n_elems as a power of 2 higher than 2^7, as defined by
+ * deflateInit2_(), so n_elems will always be a multiple of 8. */
+ chunks = n_elems >> 3;
+ Assert(n_elems % 8 == 0, "Weird hash table size!");
+
+ /* This type casting is safe since s->w_size is always <= 64KB
+ * as defined by deflateInit2_() and Posf == unsigned short */
+ vw[0] = (Posf) s->w_size;
+ vw = vec_splat(vw,0);
+
+ vp = (vector unsigned short *) table_end;
+
+ do {
+ /* Processing 8 elements at a time */
+ vp--;
+ vm = *vp;
+
+ /* This is equivalent to: m >= w_size ? m - w_size : 0
+ * Since we are using a saturated unsigned subtraction, any
+ * values that are > w_size will be set to 0, while the others
+ * will be subtracted by w_size. */
+ *vp = vec_subs(vm,vw);
+ } while (--chunks);
+};
+
+void ZLIB_INTERNAL _slide_hash_power8(deflate_state *s)
+{
+ unsigned n;
+ Posf *p;
+
+ n = s->hash_size;
+ p = &s->head[n];
+ slide_hash_power8_loop(s,n,p);
+
+#ifndef FASTEST
+ n = s->w_size;
+ p = &s->prev[n];
+ slide_hash_power8_loop(s,n,p);
+#endif
+}
diff --git a/contrib/power/slide_hash_resolver.c b/contrib/power/slide_hash_resolver.c
new file mode 100644
index 000000000..54fa1eb21
--- /dev/null
+++ b/contrib/power/slide_hash_resolver.c
@@ -0,0 +1,15 @@
+/* Copyright (C) 2019 Matheus Castanho <msc@linux.ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#include "../gcc/zifunc.h"
+#include "power.h"
+
+Z_IFUNC(slide_hash) {
+#ifdef Z_POWER8
+ if (__builtin_cpu_supports("arch_2_07"))
+ return _slide_hash_power8;
+#endif
+
+ return slide_hash_default;
+}
diff --git a/deflate.c b/deflate.c
index 799fb93cc..b2db576dc 100644
--- a/deflate.c
+++ b/deflate.c
@@ -196,6 +196,13 @@ local const config configuration_table[10] = {
(unsigned)(s->hash_size-1)*sizeof(*s->head)); \
} while (0)
+#ifdef Z_POWER_OPT
+/* Rename function so resolver can use its symbol. The default version will be
+ * returned by the resolver if the host has no support for an optimized version.
+ */
+#define slide_hash slide_hash_default
+#endif /* Z_POWER_OPT */
+
/* ===========================================================================
* Slide the hash table when sliding the window down (could be avoided with 32
* bit values at the expense of memory usage). We slide even when level == 0 to
@@ -227,6 +234,11 @@ local void slide_hash(s)
#endif
}
+#ifdef Z_POWER_OPT
+#undef slide_hash
+#include "contrib/power/slide_hash_resolver.c"
+#endif /* Z_POWER_OPT */
+
/* ========================================================================= */
int ZEXPORT deflateInit_(strm, level, version, stream_size)
z_streamp strm;

338
zlib-1.2.12-add-vectorized-longest_match-for-power.patch Normal file
View File

@ -0,0 +1,338 @@
From aecdff0646c7e188b48f6db285d8d63a74f246c1 Mon Sep 17 00:00:00 2001
From: Matheus Castanho <msc@linux.ibm.com>
Date: Tue, 29 Oct 2019 18:04:11 -0300
Subject: [PATCH] Add vectorized longest_match for Power
This commit introduces an optimized version of the longest_match
function for Power processors. It uses VSX instructions to match
16 bytes at a time on each comparison, instead of one by one.
Author: Matheus Castanho <msc@linux.ibm.com>
---
CMakeLists.txt | 3 +-
Makefile.in | 8 +
configure | 4 +-
contrib/power/longest_match_power9.c | 194 +++++++++++++++++++++++++
contrib/power/longest_match_resolver.c | 15 ++
contrib/power/power.h | 2 +
deflate.c | 13 ++
7 files changed, 236 insertions(+), 3 deletions(-)
create mode 100644 contrib/power/longest_match_power9.c
create mode 100644 contrib/power/longest_match_resolver.c
Index: zlib-1.2.12/CMakeLists.txt
===================================================================
--- zlib-1.2.12.orig/CMakeLists.txt
+++ zlib-1.2.12/CMakeLists.txt
@@ -199,7 +199,8 @@ if(CMAKE_COMPILER_IS_GNUCC)
if(POWER9)
add_definitions(-DZ_POWER9)
- set(ZLIB_POWER9 )
+ set(ZLIB_POWER9
+ contrib/power/longest_match_power9.c)
set_source_files_properties(
${ZLIB_POWER9}
Index: zlib-1.2.12/Makefile.in
===================================================================
--- zlib-1.2.12.orig/Makefile.in
+++ zlib-1.2.12/Makefile.in
@@ -189,6 +189,9 @@ crc32-vx.o: $(SRCDIR)contrib/s390/crc32-
deflate.o: $(SRCDIR)deflate.c
$(CC) $(CFLAGS) $(ZINC) -c -o $@ $(SRCDIR)deflate.c
+longest_match_power9.o: $(SRCDIR)contrib/power/longest_match_power9.c
+ $(CC) $(CFLAGS) -mcpu=power9 $(ZINC) -c -o $@ $(SRCDIR)contrib/power/longest_match_power9.c
+
slide_hash_power8.o: $(SRCDIR)contrib/power/slide_hash_power8.c
$(CC) $(CFLAGS) -mcpu=power8 $(ZINC) -c -o $@ $(SRCDIR)contrib/power/slide_hash_power8.c
@@ -259,6 +262,11 @@ deflate.lo: $(SRCDIR)deflate.c
$(CC) $(SFLAGS) $(ZINC) -DPIC -c -o objs/deflate.o $(SRCDIR)deflate.c
-@mv objs/deflate.o $@
+longest_match_power9.lo: $(SRCDIR)contrib/power/longest_match_power9.c
+ -@mkdir objs 2>/dev/null || test -d objs
+ $(CC) $(SFLAGS) -mcpu=power9 $(ZINC) -DPIC -c -o objs/longest_match_power9.o $(SRCDIR)contrib/power/longest_match_power9.c
+ -@mv objs/longest_match_power9.o $@
+
slide_hash_power8.lo: $(SRCDIR)contrib/power/slide_hash_power8.c
-@mkdir objs 2>/dev/null || test -d objs
$(CC) $(SFLAGS) -mcpu=power8 $(ZINC) -DPIC -c -o objs/slide_hash_power8.o $(SRCDIR)contrib/power/slide_hash_power8.c
Index: zlib-1.2.12/configure
===================================================================
--- zlib-1.2.12.orig/configure
+++ zlib-1.2.12/configure
@@ -915,8 +915,8 @@ if tryboth $CC -c $CFLAGS $test.c; then
if tryboth $CC -c $CFLAGS -mcpu=power9 $test.c; then
POWER9="-DZ_POWER9"
- PIC_OBJC="${PIC_OBJC}"
- OBJC="${OBJC}"
+ PIC_OBJC="$PIC_OBJC longest_match_power9.lo"
+ OBJC="$OBJC longest_match_power9.o"
echo "Checking for -mcpu=power9 support... Yes." | tee -a configure.log
else
echo "Checking for -mcpu=power9 support... No." | tee -a configure.log
Index: zlib-1.2.12/contrib/power/longest_match_power9.c
===================================================================
--- /dev/null
+++ zlib-1.2.12/contrib/power/longest_match_power9.c
@@ -0,0 +1,194 @@
+/* Copyright (C) 2019 Matheus Castanho <msc@linux.ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#include <altivec.h>
+#include "../../deflate.h"
+
+local inline int vec_match OF((Bytef* scan, Bytef* match))
+ __attribute__((always_inline));
+
+local inline int vec_match(Bytef* scan, Bytef* match)
+{
+ vector unsigned char vscan, vmatch, vc;
+ int len;
+
+ vscan = *((vector unsigned char *) scan);
+ vmatch = *((vector unsigned char *) match);
+
+ /* Compare 16 bytes at a time.
+ * Each byte of vc will be either all ones or all zeroes,
+ * depending on the result of the comparison
+ */
+ vc = (vector unsigned char) vec_cmpne(vscan,vmatch);
+
+ /* Since the index of matching bytes will contain only zeroes
+ * on vc (since we used cmpne), counting the number of consecutive
+ * bytes where LSB == 0 is the same as counting the length of the match.
+ *
+ * There was an issue in the way the vec_cnttz_lsbb builtin was implemented
+ * that got fixed on GCC 12, but now we have to use different builtins
+ * depending on the compiler version. To avoid that, let's use inline asm to
+ * generate the exact instruction we need.
+ */
+ #ifdef __LITTLE_ENDIAN__
+ asm volatile("vctzlsbb %0, %1\n\t" : "=r" (len) : "v" (vc));
+ #else
+ asm volatile("vclzlsbb %0, %1\n\t" : "=r" (len) : "v" (vc));
+ #endif
+
+ return len;
+}
+
+uInt ZLIB_INTERNAL _longest_match_power9(deflate_state *s, IPos cur_match)
+{
+ unsigned chain_length = s->max_chain_length;/* max hash chain length */
+ register Bytef *scan = s->window + s->strstart; /* current string */
+ register Bytef *match; /* matched string */
+ register int len; /* length of current match */
+ int best_len = (int)s->prev_length; /* best match length so far */
+ int nice_match = s->nice_match; /* stop if match long enough */
+ int mbytes; /* matched bytes inside loop */
+ IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
+ s->strstart - (IPos)MAX_DIST(s) : 0;
+ /* Stop when cur_match becomes <= limit. To simplify the code,
+ * we prevent matches with the string of window index 0.
+ */
+ Posf *prev = s->prev;
+ uInt wmask = s->w_mask;
+
+#if (MAX_MATCH == 258)
+ /* Compare the last two bytes at once. */
+ register Bytef *strend2 = s->window + s->strstart + MAX_MATCH - 2;
+ register ush scan_end = *(ushf*)(scan+best_len-1);
+#else
+ register Bytef *strend = s->window + s->strstart + MAX_MATCH;
+ register Byte scan_end1 = scan[best_len-1];
+ register Byte scan_end = scan[best_len];
+#endif
+
+ /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
+ * It is easy to get rid of this optimization if necessary.
+ */
+ Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
+
+ /* Do not waste too much time if we already have a good match: */
+ if (s->prev_length >= s->good_match) {
+ chain_length >>= 2;
+ }
+ /* Do not look for matches beyond the end of the input. This is necessary
+ * to make deflate deterministic.
+ */
+ if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
+
+ Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
+
+ do {
+ Assert(cur_match < s->strstart, "no future");
+ match = s->window + cur_match;
+
+ /* Skip to next match if the match length cannot increase
+ * or if the match length is less than 2. Note that the checks below
+ * for insufficient lookahead only occur occasionally for performance
+ * reasons. Therefore uninitialized memory will be accessed, and
+ * conditional jumps will be made that depend on those values.
+ * However the length of the match is limited to the lookahead, so
+ * the output of deflate is not affected by the uninitialized values.
+ */
+
+/* MAX_MATCH - 2 should be a multiple of 16 for this optimization to work. */
+#if (MAX_MATCH == 258)
+
+ /* Compare ending (2 bytes) and beginning of potential match.
+ *
+ * On Power processors, loading a 16-byte vector takes only 1 extra
+ * cycle compared to a regular byte load. So instead of comparing the
+ * first two bytes and then the rest later if they match, we can compare
+ * the first 16 at once, and when we have a match longer than 2, we will
+ * already have the result of comparing the first 16 bytes saved in mbytes.
+ */
+ if (*(ushf*)(match+best_len-1) != scan_end ||
+ (mbytes = vec_match(scan,match)) < 3) continue;
+
+ scan += mbytes;
+ match += mbytes;
+
+ /* In case when we may have a match longer than 16, we perform further
+ * comparisons in chunks of 16 and keep going while all bytes match.
+ */
+ while(mbytes == 16) {
+ mbytes = vec_match(scan,match);
+ scan += mbytes;
+ match += mbytes;
+
+ /* We also have to limit the maximum match based on MAX_MATCH.
+ * Since we are comparing 16 bytes at a time and MAX_MATCH == 258 (to
+ * comply with default implementation), we should stop comparing when
+ * we have matched 256 bytes, which happens when scan == strend2.
+ * In this ("rare") case, we have to check the remaining 2 bytes
+ * individually using common load and compare operations.
+ */
+ if(scan >= strend2) {
+ if(*scan == *match) {
+ if(*++scan == *++match)
+ scan++;
+ }
+ break;
+ }
+ }
+
+ Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
+
+ len = (MAX_MATCH - 2) - (int)(strend2 - scan);
+ scan = strend2 - (MAX_MATCH - 2);
+
+#else /* MAX_MATCH == 258 */
+
+ if (match[best_len] != scan_end ||
+ match[best_len-1] != scan_end1 ||
+ *match != *scan ||
+ *++match != scan[1]) continue;
+
+ /* The check at best_len-1 can be removed because it will be made
+ * again later. (This heuristic is not always a win.)
+ * It is not necessary to compare scan[2] and match[2] since they
+ * are always equal when the other bytes match, given that
+ * the hash keys are equal and that HASH_BITS >= 8.
+ */
+ scan += 2, match++;
+ Assert(*scan == *match, "match[2]?");
+
+ /* We check for insufficient lookahead only every 8th comparison;
+ * the 256th check will be made at strstart+258.
+ */
+ do {
+ } while (*++scan == *++match && *++scan == *++match &&
+ *++scan == *++match && *++scan == *++match &&
+ *++scan == *++match && *++scan == *++match &&
+ *++scan == *++match && *++scan == *++match &&
+ scan < strend);
+
+ Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
+
+ len = MAX_MATCH - (int)(strend - scan);
+ scan = strend - MAX_MATCH;
+
+#endif /* MAX_MATCH == 258 */
+
+ if (len > best_len) {
+ s->match_start = cur_match;
+ best_len = len;
+ if (len >= nice_match) break;
+#if (MAX_MATCH == 258)
+ scan_end = *(ushf*)(scan+best_len-1);
+#else
+ scan_end1 = scan[best_len-1];
+ scan_end = scan[best_len];
+#endif
+ }
+ } while ((cur_match = prev[cur_match & wmask]) > limit
+ && --chain_length != 0);
+
+ if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
+ return s->lookahead;
+}
Index: zlib-1.2.12/contrib/power/longest_match_resolver.c
===================================================================
--- /dev/null
+++ zlib-1.2.12/contrib/power/longest_match_resolver.c
@@ -0,0 +1,15 @@
+/* Copyright (C) 2019 Matheus Castanho <msc@linux.ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#include "../gcc/zifunc.h"
+#include "power.h"
+
+Z_IFUNC(longest_match) {
+#ifdef Z_POWER9
+ if (__builtin_cpu_supports("arch_3_00"))
+ return _longest_match_power9;
+#endif
+
+ return longest_match_default;
+}
Index: zlib-1.2.12/contrib/power/power.h
===================================================================
--- zlib-1.2.12.orig/contrib/power/power.h
+++ zlib-1.2.12/contrib/power/power.h
@@ -10,4 +10,6 @@ uLong _adler32_power8(uLong adler, const
unsigned long _crc32_z_power8(unsigned long, const Bytef *, z_size_t);
+uInt _longest_match_power9(deflate_state *s, IPos cur_match);
+
void _slide_hash_power8(deflate_state *s);
Index: zlib-1.2.12/deflate.c
===================================================================
--- zlib-1.2.12.orig/deflate.c
+++ zlib-1.2.12/deflate.c
@@ -1309,6 +1309,14 @@ local void lm_init (s)
/* For 80x86 and 680x0, an optimized version will be provided in match.asm or
* match.S. The code will be functionally equivalent.
*/
+
+#ifdef Z_POWER_OPT
+/* Rename function so resolver can use its symbol. The default version will be
+ * returned by the resolver if the host has no support for an optimized version.
+ */
+#define longest_match longest_match_default
+#endif /* Z_POWER_OPT */
+
local uInt longest_match(s, pcur_match)
deflate_state *s;
IPos pcur_match; /* current match */
@@ -1454,6 +1462,11 @@ local uInt longest_match(s, pcur_match)
}
#endif /* ASMV */
+#ifdef Z_POWER_OPT
+#undef longest_match
+#include "contrib/power/longest_match_resolver.c"
+#endif /* Z_POWER_OPT */
+
#else /* FASTEST */
/* ---------------------------------------------------------------------------

342
zlib-1.2.12-adler32-vector-optimizations-for-power.patch Normal file
View File

@ -0,0 +1,342 @@
From 772f4bd0f880c4c193ab7da78728f38821572a02 Mon Sep 17 00:00:00 2001
From: Rogerio Alves <rcardoso@linux.ibm.com>
Date: Mon, 9 Dec 2019 14:40:53 -0300
Subject: [PATCH] Adler32 vector optimization for Power.
This commit implements a Power (POWER8+) vector optimization for Adler32
checksum using VSX (vector) instructions. The VSX adler32 checksum is up
to 10x fast than the adler32 baseline code.
Author: Rogerio Alves <rcardoso@linux.ibm.com>
---
CMakeLists.txt | 1 +
Makefile.in | 8 ++
adler32.c | 11 ++
configure | 4 +-
contrib/power/adler32_power8.c | 196 +++++++++++++++++++++++++++++++
contrib/power/adler32_resolver.c | 15 +++
contrib/power/power.h | 4 +-
7 files changed, 236 insertions(+), 3 deletions(-)
create mode 100644 contrib/power/adler32_power8.c
create mode 100644 contrib/power/adler32_resolver.c
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 581e1fa6d..c6296ee68 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -185,6 +185,7 @@ if(CMAKE_COMPILER_IS_GNUCC)
if(POWER8)
add_definitions(-DZ_POWER8)
set(ZLIB_POWER8
+ contrib/power/adler32_power8.c
contrib/power/crc32_z_power8.c)
set_source_files_properties(
diff --git a/Makefile.in b/Makefile.in
index 16943044e..a0ffac860 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -165,6 +165,9 @@ minigzip64.o: $(SRCDIR)test/minigzip.c $(SRCDIR)zlib.h zconf.h
adler32.o: $(SRCDIR)adler32.c
$(CC) $(CFLAGS) $(ZINC) -c -o $@ $(SRCDIR)adler32.c
+adler32_power8.o: $(SRCDIR)contrib/power/adler32_power8.c
+ $(CC) $(CFLAGS) -mcpu=power8 $(ZINC) -c -o $@ $(SRCDIR)contrib/power/adler32_power8.c
+
crc32.o: $(SRCDIR)crc32.c
$(CC) $(CFLAGS) $(ZINC) -c -o $@ $(SRCDIR)crc32.c
@@ -216,6 +219,11 @@ adler32.lo: $(SRCDIR)adler32.c
$(CC) $(SFLAGS) $(ZINC) -DPIC -c -o objs/adler32.o $(SRCDIR)adler32.c
-@mv objs/adler32.o $@
+adler32_power8.lo: $(SRCDIR)contrib/power/adler32_power8.c
+ -@mkdir objs 2>/dev/null || test -d objs
+ $(CC) $(SFLAGS) -mcpu=power8 $(ZINC) -DPIC -c -o objs/adler32_power8.o $(SRCDIR)contrib/power/adler32_power8.c
+ -@mv objs/adler32_power8.o $@
+
crc32.lo: $(SRCDIR)crc32.c
-@mkdir objs 2>/dev/null || test -d objs
$(CC) $(SFLAGS) $(ZINC) -DPIC -c -o objs/crc32.o $(SRCDIR)crc32.c
diff --git a/adler32.c b/adler32.c
index d0be4380a..4bde0fa18 100644
--- a/adler32.c
+++ b/adler32.c
@@ -131,6 +131,12 @@ uLong ZEXPORT adler32_z(adler, buf, len)
}
/* ========================================================================= */
+
+#ifdef Z_POWER_OPT
+/* Rename the default function to avoid naming conflicts */
+#define adler32 adler32_default
+#endif /* Z_POWER_OPT */
+
uLong ZEXPORT adler32(adler, buf, len)
uLong adler;
const Bytef *buf;
@@ -139,6 +145,11 @@ uLong ZEXPORT adler32(adler, buf, len)
return adler32_z(adler, buf, len);
}
+#ifdef Z_POWER_OPT
+#undef adler32
+#include "contrib/power/adler32_resolver.c"
+#endif /* Z_POWER_OPT */
+
/* ========================================================================= */
local uLong adler32_combine_(adler1, adler2, len2)
uLong adler1;
diff --git a/configure b/configure
index 914d9f4aa..810a7404d 100755
--- a/configure
+++ b/configure
@@ -879,8 +879,8 @@ if tryboth $CC -c $CFLAGS $test.c; then
if tryboth $CC -c $CFLAGS -mcpu=power8 $test.c; then
POWER8="-DZ_POWER8"
- PIC_OBJC="${PIC_OBJC} crc32_z_power8.lo"
- OBJC="${OBJC} crc32_z_power8.o"
+ PIC_OBJC="${PIC_OBJC} adler32_power8.lo crc32_z_power8.lo"
+ OBJC="${OBJC} adler32_power8.o crc32_z_power8.o"
echo "Checking for -mcpu=power8 support... Yes." | tee -a configure.log
else
echo "Checking for -mcpu=power8 support... No." | tee -a configure.log
diff --git a/contrib/power/adler32_power8.c b/contrib/power/adler32_power8.c
new file mode 100644
index 000000000..473c39457
--- /dev/null
+++ b/contrib/power/adler32_power8.c
@@ -0,0 +1,196 @@
+/*
+ * Adler32 for POWER 8+ using VSX instructions.
+ *
+ * Calculate adler32 checksum for 16 bytes at once using POWER8+ VSX (vector)
+ * instructions.
+ *
+ * If adler32 do 1 byte at time on the first iteration s1 is s1_0 (_n means
+ * iteration n) is the initial value of adler - at start _0 is 1 unless
+ * adler initial value is different than 1. So s1_1 = s1_0 + c[0] after
+ * the first calculation. For the iteration s1_2 = s1_1 + c[1] and so on.
+ * Hence, for iteration N, s1_N = s1_(N-1) + c[N] is the value of s1 on
+ * after iteration N.
+ *
+ * Therefore, for s2 and iteration N, s2_N = s2_0 + N*s1_N + N*c[0] +
+ * N-1*c[1] + ... + c[N]
+ *
+ * In a more general way:
+ *
+ * s1_N = s1_0 + sum(i=1 to N)c[i]
+ * s2_N = s2_0 + N*s1 + sum (i=1 to N)(N-i+1)*c[i]
+ *
+ * Where s1_N, s2_N are the values for s1, s2 after N iterations. So if we
+ * can process N-bit at time we can do this at once.
+ *
+ * Since VSX can support 16-bit vector instructions, we can process
+ * 16-bit at time using N = 16 we have:
+ *
+ * s1 = s1_16 = s1_(16-1) + c[16] = s1_0 + sum(i=1 to 16)c[i]
+ * s2 = s2_16 = s2_0 + 16*s1 + sum(i=1 to 16)(16-i+1)*c[i]
+ *
+ * After the first iteration we calculate the adler32 checksum for 16 bytes.
+ *
+ * For more background about adler32 please check the RFC:
+ * https://www.ietf.org/rfc/rfc1950.txt
+ *
+ * Copyright (C) 2019 Rogerio Alves <rcardoso@linux.ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ *
+ */
+
+#include "../../zutil.h"
+#include <altivec.h>
+
+/* Largest prime smaller than 65536. */
+#define BASE 65521U
+#define NMAX 5552
+/* NMAX is the largest n such that 255n(n+1)/2 + (n+1)(BASE-1) <= 2^32-1. */
+
+#define DO1(s1,s2,buf,i) {(s1) += buf[(i)]; (s2) += (s1);}
+#define DO2(s1,s2,buf,i) {DO1(s1,s2,buf,i); DO1(s1,s2,buf,i+1);}
+#define DO4(s1,s2,buf,i) {DO2(s1,s2,buf,i); DO2(s1,s2,buf,i+2);}
+#define DO8(s1,s2,buf,i) {DO4(s1,s2,buf,i); DO4(s1,s2,buf,i+4);}
+#define DO16(s1,s2,buf) {DO8(s1,s2,buf,0); DO8(s1,s2,buf,8);}
+
+/* Vector across sum unsigned int (saturate). */
+inline vector unsigned int vec_sumsu (vector unsigned int __a,
+ vector unsigned int __b)
+{
+ __b = vec_sld(__a, __a, 8);
+ __b = vec_add(__b, __a);
+ __a = vec_sld(__b, __b, 4);
+ __a = vec_add(__a, __b);
+
+ return __a;
+}
+
+uLong ZLIB_INTERNAL _adler32_power8 (uLong adler, const Bytef* buf, uInt len)
+{
+ /* If buffer is empty or len=0 we need to return adler initial value. */
+ if (buf == NULL)
+ return 1;
+
+ unsigned int s1 = adler & 0xffff;
+ unsigned int s2 = (adler >> 16) & 0xffff;
+
+ /* in case user likes doing a byte at a time, keep it fast */
+ if (len == 1) {
+ s1 += buf[0];
+ if (s1 >= BASE)
+ s1 -= BASE;
+ s2 += s1;
+ if (s2 >= BASE)
+ s2 -= BASE;
+ return (s2 << 16) | s1;
+ }
+
+ /* Keep it fast for short length buffers. */
+ if (len < 16) {
+ while (len--) {
+ s1 += *buf++;
+ s2 += s1;
+ }
+ if (s1 >= BASE)
+ s1 -= BASE;
+ s2 %= BASE;
+ return (s2 << 16) | s1;
+ }
+
+ /* This is faster than VSX code for len < 64. */
+ if (len < 64) {
+ while (len >= 16) {
+ len -= 16;
+ DO16(s1,s2,buf);
+ buf += 16;
+ }
+ } else {
+ /* Use POWER VSX instructions for len >= 64. */
+ const vector unsigned int v_zeros = { 0 };
+ const vector unsigned char v_mul = {16, 15, 14, 13, 12, 11, 10, 9, 8, 7,
+ 6, 5, 4, 3, 2, 1};
+ const vector unsigned char vsh = vec_splat_u8(4);
+ const vector unsigned int vmask = {0xffffffff, 0x0, 0x0, 0x0};
+ vector unsigned int vs1 = vec_xl(0, &s1);
+ vector unsigned int vs2 = vec_xl(0, &s2);
+ vector unsigned int vs1_save = { 0 };
+ vector unsigned int vsum1, vsum2;
+ vector unsigned char vbuf;
+ int n;
+
+ /* Zeros the undefined values of vectors vs1, vs2. */
+ vs1 = vec_and(vs1, vmask);
+ vs2 = vec_and(vs2, vmask);
+
+ /* Do length bigger than NMAX in blocks of NMAX size. */
+ while (len >= NMAX) {
+ len -= NMAX;
+ n = NMAX / 16;
+ do {
+ vbuf = vec_xl(0, (unsigned char *) buf);
+ vsum1 = vec_sum4s(vbuf, v_zeros); /* sum(i=1 to 16) buf[i]. */
+ /* sum(i=1 to 16) buf[i]*(16-i+1). */
+ vsum2 = vec_msum(vbuf, v_mul, v_zeros);
+ /* Save vs1. */
+ vs1_save = vec_add(vs1_save, vs1);
+ /* Accumulate the sums. */
+ vs1 = vec_add(vsum1, vs1);
+ vs2 = vec_add(vsum2, vs2);
+
+ buf += 16;
+ } while (--n);
+ /* Once each block of NMAX size. */
+ vs1 = vec_sumsu(vs1, vsum1);
+ vs1_save = vec_sll(vs1_save, vsh); /* 16*vs1_save. */
+ vs2 = vec_add(vs1_save, vs2);
+ vs2 = vec_sumsu(vs2, vsum2);
+
+ /* vs1[0] = (s1_i + sum(i=1 to 16)buf[i]) mod 65521. */
+ vs1[0] = vs1[0] % BASE;
+ /* vs2[0] = s2_i + 16*s1_save +
+ sum(i=1 to 16)(16-i+1)*buf[i] mod 65521. */
+ vs2[0] = vs2[0] % BASE;
+
+ vs1 = vec_and(vs1, vmask);
+ vs2 = vec_and(vs2, vmask);
+ vs1_save = v_zeros;
+ }
+
+ /* len is less than NMAX one modulo is needed. */
+ if (len >= 16) {
+ while (len >= 16) {
+ len -= 16;
+
+ vbuf = vec_xl(0, (unsigned char *) buf);
+
+ vsum1 = vec_sum4s(vbuf, v_zeros); /* sum(i=1 to 16) buf[i]. */
+ /* sum(i=1 to 16) buf[i]*(16-i+1). */
+ vsum2 = vec_msum(vbuf, v_mul, v_zeros);
+ /* Save vs1. */
+ vs1_save = vec_add(vs1_save, vs1);
+ /* Accumulate the sums. */
+ vs1 = vec_add(vsum1, vs1);
+ vs2 = vec_add(vsum2, vs2);
+
+ buf += 16;
+ }
+ /* Since the size will be always less than NMAX we do this once. */
+ vs1 = vec_sumsu(vs1, vsum1);
+ vs1_save = vec_sll(vs1_save, vsh); /* 16*vs1_save. */
+ vs2 = vec_add(vs1_save, vs2);
+ vs2 = vec_sumsu(vs2, vsum2);
+ }
+ /* Copy result back to s1, s2 (mod 65521). */
+ s1 = vs1[0] % BASE;
+ s2 = vs2[0] % BASE;
+ }
+
+ /* Process tail (len < 16). */
+ while (len--) {
+ s1 += *buf++;
+ s2 += s1;
+ }
+ s1 %= BASE;
+ s2 %= BASE;
+
+ return (s2 << 16) | s1;
+}
diff --git a/contrib/power/adler32_resolver.c b/contrib/power/adler32_resolver.c
new file mode 100644
index 000000000..07a1a2cb2
--- /dev/null
+++ b/contrib/power/adler32_resolver.c
@@ -0,0 +1,15 @@
+/* Copyright (C) 2019 Rogerio Alves <rcardoso@linux.ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#include "../gcc/zifunc.h"
+#include "power.h"
+
+Z_IFUNC(adler32) {
+#ifdef Z_POWER8
+ if (__builtin_cpu_supports("arch_2_07"))
+ return _adler32_power8;
+#endif
+
+ return adler32_default;
+}
diff --git a/contrib/power/power.h b/contrib/power/power.h
index 79123aa90..f57c76167 100644
--- a/contrib/power/power.h
+++ b/contrib/power/power.h
@@ -2,7 +2,9 @@
* 2019 Rogerio Alves <rogerio.alves@ibm.com>, IBM
* For conditions of distribution and use, see copyright notice in zlib.h
*/
-
#include "../../zconf.h"
+#include "../../zutil.h"
+
+uLong _adler32_power8(uLong adler, const Bytef* buf, uInt len);
unsigned long _crc32_z_power8(unsigned long, const Bytef *, z_size_t);

34
zlib-1.2.12-fix-invalid-memory-access-on-ppc-and-ppc64.patch Normal file
View File

@ -0,0 +1,34 @@
From 11b722e4ae91b611f605221587ec8e0829c27949 Mon Sep 17 00:00:00 2001
From: Matheus Castanho <msc@linux.ibm.com>
Date: Tue, 23 Jun 2020 10:26:19 -0300
Subject: [PATCH] Fix invalid memory access on ppc and ppc64
---
contrib/power/adler32_power8.c | 9 ++++-----
1 file changed, 4 insertions(+), 5 deletions(-)
diff --git a/contrib/power/adler32_power8.c b/contrib/power/adler32_power8.c
index 473c39457..fdd086453 100644
--- a/contrib/power/adler32_power8.c
+++ b/contrib/power/adler32_power8.c
@@ -110,16 +110,15 @@ uLong ZLIB_INTERNAL _adler32_power8 (uLong adler, const Bytef* buf, uInt len)
6, 5, 4, 3, 2, 1};
const vector unsigned char vsh = vec_splat_u8(4);
const vector unsigned int vmask = {0xffffffff, 0x0, 0x0, 0x0};
- vector unsigned int vs1 = vec_xl(0, &s1);
- vector unsigned int vs2 = vec_xl(0, &s2);
+ vector unsigned int vs1 = { 0 };
+ vector unsigned int vs2 = { 0 };
vector unsigned int vs1_save = { 0 };
vector unsigned int vsum1, vsum2;
vector unsigned char vbuf;
int n;
- /* Zeros the undefined values of vectors vs1, vs2. */
- vs1 = vec_and(vs1, vmask);
- vs2 = vec_and(vs2, vmask);
+ vs1[0] = s1;
+ vs2[0] = s2;
/* Do length bigger than NMAX in blocks of NMAX size. */
while (len >= NMAX) {

10
zlib.changes
View File

@ -1,3 +1,13 @@
-------------------------------------------------------------------
Mon Oct 10 10:08:02 UTC 2022 - Danilo Spinella <danilo.spinella@suse.com>
- Add Power8 optimizations:
* zlib-1.2.12-add-optimized-slide_hash-for-power.patch
* zlib-1.2.12-add-vectorized-longest_match-for-power.patch
* zlib-1.2.12-adler32-vector-optimizations-for-power.patch
* zlib-1.2.12-fix-invalid-memory-access-on-ppc-and-ppc64.patch
- Update zlib-1.2.12-IBM-Z-hw-accelerated-deflate-s390x.patch
------------------------------------------------------------------- -------------------------------------------------------------------
Tue Aug 23 16:22:59 UTC 2022 - Danilo Spinella <danilo.spinella@suse.com> Tue Aug 23 16:22:59 UTC 2022 - Danilo Spinella <danilo.spinella@suse.com>

16
zlib.spec
View File

@ -44,12 +44,18 @@ Patch6: minizip-dont-install-crypt-header.patch
# The following patches are taken from https://github.com/iii-i/zlib/commits/crc32vx-v3 # The following patches are taken from https://github.com/iii-i/zlib/commits/crc32vx-v3
Patch7: zlib-1.2.5-minizip-fixuncrypt.patch Patch7: zlib-1.2.5-minizip-fixuncrypt.patch
Patch8: zlib-1.2.11-optimized-s390.patch Patch8: zlib-1.2.11-optimized-s390.patch
# https://github.com/iii-i/zlib/commit/171d0ff3c9ed40da0ac14085ab16b766b1162069
Patch9: zlib-1.2.12-IBM-Z-hw-accelerated-deflate-s390x.patch Patch9: zlib-1.2.12-IBM-Z-hw-accelerated-deflate-s390x.patch
Patch10: zlib-1.2.11-covscan-issues.patch Patch10: zlib-1.2.11-covscan-issues.patch
Patch11: zlib-1.2.11-covscan-issues-rhel9.patch Patch11: zlib-1.2.11-covscan-issues-rhel9.patch
Patch12: zlib-1.2.12-optimized-crc32-power8.patch Patch12: zlib-1.2.12-optimized-crc32-power8.patch
Patch13: zlib-1.2.12-fix-configure.patch Patch13: zlib-1.2.12-fix-configure.patch
Patch14: zlib-1.2.12-s390-vectorize-crc32.patch Patch14: zlib-1.2.12-s390-vectorize-crc32.patch
# The following patches are taken from https://github.com/mscastanho/zlib/commits/power-optimizations-1.2.12
Patch15: zlib-1.2.12-adler32-vector-optimizations-for-power.patch
Patch16: zlib-1.2.12-fix-invalid-memory-access-on-ppc-and-ppc64.patch
Patch17: zlib-1.2.12-add-optimized-slide_hash-for-power.patch
Patch18: zlib-1.2.12-add-vectorized-longest_match-for-power.patch
BuildRequires: autoconf BuildRequires: autoconf
BuildRequires: automake BuildRequires: automake
BuildRequires: libtool BuildRequires: libtool
@ -148,6 +154,10 @@ It should exit 0
%patch12 -p1 %patch12 -p1
%patch13 -p1 %patch13 -p1
%patch14 -p1 %patch14 -p1
%patch15 -p1
%patch16 -p1
%patch17 -p1
%patch18 -p1
cp %{SOURCE4} . cp %{SOURCE4} .
%build %build
@ -167,10 +177,10 @@ CC="cc" ./configure \
# Profiling flags breaks tests, as of 1.2.12 # Profiling flags breaks tests, as of 1.2.12
# In particular, gzseek does not work as intended # In particular, gzseek does not work as intended
#%if %{do_profiling} #%if %{do_profiling}
# #make %{?_smp_mflags} CFLAGS="%{optflags} %{cflags_profile_generate}" # make %{?_smp_mflags} CFLAGS="%{optflags} %{cflags_profile_generate}"
# make check %{?_smp_mflags} # make check %{?_smp_mflags}
# #make %{?_smp_mflags} clean # make %{?_smp_mflags} clean
# #make %{?_smp_mflags} CFLAGS="%{optflags} %{cflags_profile_feedback}" # make %{?_smp_mflags} CFLAGS="%{optflags} %{cflags_profile_feedback}"
#%else #%else
make %{?_smp_mflags} make %{?_smp_mflags}
#%endif #%endif