[putting mailing lists on cc]

On Fri, Sep 30, 2011 at 08:01:36AM +0200, Joakim Tjernlund wrote:

(Just happen to see this patch in the archives)

- This is basically an copy of Bobs crc32 work and duplicates code, this
  code needs to move into /lib/crc32.c and use the existing framework.

Which framework are you talking about?  lib/crc32.c appears to be a simple
module that exports a utility function.  Do you mean that you want to merge the
crc32{,c}defs.h and gen_crc32{,c}table.c code?  Do you want a build script that
starts with only a crc${ALG}_defs.h file and stamps out gencrc${ALG}_table.c
and crc${ALG}.c boilerplate code and then builds it?

I really don't know; from my perspective there was a slow implementation in
crypto/crc32c.c and I wanted to speed it up.  crc32c seems to be in crypto/ and
not lib/ so that the implementation can be replaced with a hardware accelerated
version at runtime (crc32c-intel).

For crc32 which has no such hw replacement (as far as I know), moving it into
crypto/ would incur the overhead of going through the cryptoapi for not much
benefit.  On the other hand it wouldn't be hard to put the crc32 code into
crypto/.

- Slice by 8 is just half the speed on my ppc32 compared to slice by 4 so
  it can't be enabled for all archs. Best to start with all 64 bit archs

<shrug> I suppose I could make CRC32C_BITS configurable.  What is the hardware
profile of your ppc32 processor?  How much L1D/L2 cache?  slice-by-8 does have
a big cache footprint.  On the other hand it's faster than the slice-by-4
(crc32) and Sarwate (crc32c) code in the kernel, even on old slow 32-bit x86
processors (PII, PIII, P4).

- Last time I tested Bobs slice by 8 on ppc32 it didn't work.

... is crc32c broken *now*?  It seems fine on x86/amd64/ppc64.

--D