loop twice. Gave a modest speedup.

Rev: nettle/x86/arcfour-crypt.asm:1.2

loop with movl.	Eliminated redundant movl.
* x86/aes-decrypt-internal.asm: Likewise.

Rev: nettle/x86/aes-decrypt-internal.asm:1.5
Rev: nettle/x86/aes-encrypt-internal.asm:1.5

Rev: nettle/x86/aes.m4:1.6

three times through the substitution loop.
* x86/aes-decrypt-internal.asm: Likewise.

Rev: nettle/x86/aes-decrypt-internal.asm:1.4
Rev: nettle/x86/aes-encrypt-internal.asm:1.4

significant byte here.

Rev: nettle/x86/aes.m4:1.5

decl for outer loop.
* x86/aes-decrypt-internal.asm: Likewise.

Rev: nettle/x86/aes-decrypt-internal.asm:1.3
Rev: nettle/x86/aes-encrypt-internal.asm:1.3

Rev: nettle/x86/aes.m4:1.4

(AES_SUBST_BYTE): Likewise.

Rev: nettle/x86/aes.m4:1.3

* x86/aes-decrypt-internal.asm: Likewise.
* x86/aes.m4 (BYTEREG): New macro.
(AES_SUBST_BYTE): Take state registers as argument. Use BYTEREG to
get the corresponding byte register.

Rev: nettle/x86/aes-decrypt-internal.asm:1.2
Rev: nettle/x86/aes-encrypt-internal.asm:1.2
Rev: nettle/x86/aes.m4:1.2

Rev: src/nettle/x86/aes.asm:1.14(DEAD)

* x86/aes-encrypt-internal.asm: Likewise.
* x86/arcfour-crypt.asm: Likewise.
* x86/md5-compress.asm: Likewise.
* x86/sha1-compress.asm: Likewise.

Rev: src/nettle/x86/aes-decrypt-internal.asm:1.3
Rev: src/nettle/x86/aes-encrypt-internal.asm:1.4
Rev: src/nettle/x86/arcfour-crypt.asm:1.15
Rev: src/nettle/x86/md5-compress.asm:1.6
Rev: src/nettle/x86/sha1-compress.asm:1.13

* asm.m4 (ARCFOUR): New struct.

Rev: src/nettle/asm.m4:1.15
Rev: src/nettle/x86/arcfour-crypt.asm:1.14

* x86/aes-decrypt-internal.asm: Likewise.

Rev: src/nettle/x86/aes-decrypt-internal.asm:1.2
Rev: src/nettle/x86/aes-encrypt-internal.asm:1.3

* x86/machine.m4: ... removed aes macros.

Rev: src/nettle/x86/aes.m4:1.1
Rev: src/nettle/x86/machine.m4:1.10

Rev: src/nettle/x86/aes-decrypt.asm:1.22(DEAD)
Rev: src/nettle/x86/aes-encrypt.asm:1.23(DEAD)

Rev: src/nettle/x86/aes-decrypt-internal.asm:1.1

Rev: src/nettle/x86/aes-encrypt-internal.asm:1.2

Also take more of the used registers as argument.

Rev: src/nettle/x86/machine.m4:1.9

with the table pointer as an argument. Unlike the old code, this
should really be position independent.

Rev: src/nettle/x86/aes-encrypt-internal.asm:1.1

Rev: src/nettle/x86/md5-compress.asm:1.5

(F1): Fixed bug.

Rev: src/nettle/x86/md5-compress.asm:1.4

Rev: src/nettle/x86/md5-compress.asm:1.3

Rev: src/nettle/x86/md5-compress.asm:1.2

* x86/sha1-compress.asm (OFFSET): ... removed macro.

Rev: src/nettle/x86/machine.m4:1.8
Rev: src/nettle/x86/sha1-compress.asm:1.12

Rev: src/nettle/x86/md5-compress.asm:1.1

using the AES_SUBST_BYTE macro. Use PROLOGUE and EPILOGUE.
* x86/sha1-compress.asm: Use PROLOGUE and EPILOGUE.
* x86/arcfour-crypt.asm: Likewise.
* x86/aes-encrypt.asm: Likewise.

Rev: src/nettle/x86/aes-encrypt.asm:1.22
Rev: src/nettle/x86/arcfour-crypt.asm:1.13
Rev: src/nettle/x86/sha1-compress.asm:1.11

using the AES_SUBST_BYTE macro. Use PROLOGUE and EPILOGUE.

Rev: src/nettle/x86/aes-decrypt.asm:1.21

Rev: src/nettle/x86/aes_tables.asm:1.2(DEAD)

override aes.c, which isn't needed for the x86 implementation.

Rev: src/nettle/x86/aes.asm:1.13

compute offsets mod 16, since m4 on FreeBSD 49.RELEASE and NetBSD
doesn't implement & correctly in eval.

Rev: src/nettle/x86/sha1-compress.asm:1.10

(F3): Eliminated a movl.
(ROUNd): New argument, for k. When using F3, it's TMP3, on the
stack, otherwise, it is kept in TMP2, a register.

Rev: src/nettle/x86/sha1-compress.asm:1.9

conversion, and store the input data on the stack. This leaves one
more register free for other uses.

Rev: src/nettle/x86/sha1-compress.asm:1.8

Rev: src/nettle/x86/sha1-compress.asm:1.7

* x86/aes-encrypt.asm: Likewise.
* x86/aes-decrypt.asm: Likewise.
* x86/arcfour-crypt.asm: Likewise.

Rev: src/nettle/x86/aes-decrypt.asm:1.20
Rev: src/nettle/x86/aes-encrypt.asm:1.21
Rev: src/nettle/x86/arcfour-crypt.asm:1.12
Rev: src/nettle/x86/sha1-compress.asm:1.6

Rev: src/nettle/x86/arcfour-crypt.asm:1.11

bost src and dst pointers in the loop, and use plain addb when
updating j. These two previous changes slowed the code down on AMD
Duron.

Rev: src/nettle/x86/arcfour-crypt.asm:1.10

precomputing the offset between src and dst.

Rev: src/nettle/x86/arcfour-crypt.asm:1.9

addl + andl $0xff, improving speed on PPro by another 15%.

Rev: src/nettle/x86/arcfour-crypt.asm:1.8

incl + andl, to improve speed on PPro and PII. Suggested by
Fredrik Olsson.

Rev: src/nettle/x86/arcfour-crypt.asm:1.7

Rev: src/nettle/x86/sha1-compress.asm:1.5