bug#34918: [PATCH] Add support for IBM Z hardware-accelerated deflate

Future versions of IBM Z mainframes will provide DFLTCC instruction,
which implements deflate algorithm in hardware with estimated
compression and decompression performance orders of magnitude faster
than the current gzip and ratio comparable with that of level 1.

This patch adds DFLTCC support to gzip. In order to enable it, the
following build commands should be used:

    $ ./configure --enable-dfltcc
    $ make

When built like this, gzip 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 add -DDFLTCC_LEVEL_MASK=0x7e
at compile time, 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.

This patch tries to add DFLTCC support in a least intrusive way.
All SystemZ-specific code was placed into a separate file, but
unfortunately there is still a noticeable amount of changes in the
main gzip code. Below is the summary of those changes.

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, bi_buf and bi_valid were promoted to extern variables.

lm_init() function moves the input buffer into the window, which is not
desirable for DFLTCC. Therefore, its invocation was moved to
software-only deflate(). In addition to initializing the window, this
function also used to convert compression level to flags, which is still
needed for DFLTCC. This responsibility was handed off to zip() function.

To achieve maximum performance with DFLTCC, inbuf and outbuf must be
256k big and page-aligned. Additionally, for DFLTCC to work at all, the
window must be page-aligned.

In addition to compression, DFLTCC computes CRC-32 checksum, therefore,
whenever it's used, software checksumming needs to be suppressed and its
results replaced by those of dfltcc. This is achieved by introducing the
new getcrc() and setcrc() functions.

Unlike the current software implementation, DFLTCC decompresses data
into the output buffer, and not the window. Therefore, just like
flushing the window, flushing the output buffer must honor the test
flag.

Finally, znew-k test assumes that "znew -K" would not convert the test
.Z file to .gz, which is not the case with DFLTCC. Since this is not the
main point of the test, this assumption was relaxed.

1 files changed, 4 insertions, 0 deletions

diff --git a/unzip.c b/unzip.c
index a7255d4..86ef664 100644
--- a/unzip.c
+++ b/unzip.c
@@ -129,7 +129,11 @@ int unzip(in, out)
     /* Decompress */
     if (method == DEFLATED)  {
 
+#ifdef IBM_Z_DFLTCC
+        int res = dfltcc_inflate();
+#else
         int res = inflate();
+#endif
 
         if (res == 3) {
             xalloc_die ();