openssl-1.x added optimized support for: sha1, sha256, sha512, aes, modmul Current openssl does not enable those. Patches are to be attached in a second. My box is armv5tel (marvell sheevapug). Tested successfully with FEATURES=test. Some stats in advance: Some stats (I took only AES): 'openssl speed' for linux-generic32: aes-128 ige 9599.54k 11060.98k 11493.29k 11621.89k 11419.48k aes-192 ige 8389.06k 9475.13k 9786.24k 9871.84k 9734.51k aes-256 ige 7443.99k 8285.13k 8524.63k 8589.27k 8482.42k 'openssl speed' for linux-armv4: aes-128 ige 14030.57k 16466.19k 17184.17k 17424.44k 17318.27k aes-192 ige 12427.90k 14249.97k 14818.38k 14917.63k 14904.51k aes-256 ige 11094.91k 12564.38k 12998.64k 13114.39k 13052.40k 46 to 49% speedup
Created attachment 243747 [details, diff] gentoo-config.patch - adds armv4+ targets as special
Created attachment 243749 [details, diff] openss-ebuild.patch - just changes gentoo.config name
looks fine, but there would be issues with arm-* targets
thinking about it some more, i think it'll get other targets incorrect as well. for example, arm6-xxx probably means the ARM6 family which is of the architecture ARMv3. the ARM naming convention is just way too f-ed up to make this an "opt-out" glob. perhaps the only safe way to handle this is: + armv[4-9]*b*) machine="armv4 -DB_ENDIAN";; + armv[4-9]*) machine="armv4 -DL_ENDIAN";; + arm*b*) machine="generic32 -DB_ENDIAN";; + arm*) machine="generic32 -DL_ENDIAN";; the openssl config itself does the opposite, but they're comparing against `uname -m`, so they can trust the output a lot more than we can. if this change works for your circumstances, i'll go ahead and commit it
(In reply to comment #4) > thinking about it some more, i think it'll get other targets incorrect as well. > for example, arm6-xxx probably means the ARM6 family which is of the > architecture ARMv3. the ARM naming convention is just way too f-ed up to make > this an "opt-out" glob. Ugh, I messed things even worse in initial patch then. > perhaps the only safe way to handle this is: > + armv[4-9]*b*) machine="armv4 -DB_ENDIAN";; > + armv[4-9]*) machine="armv4 -DL_ENDIAN";; > + arm*b*) machine="generic32 -DB_ENDIAN";; > + arm*) machine="generic32 -DL_ENDIAN";; > > the openssl config itself does the opposite, but they're comparing against > `uname -m`, so they can trust the output a lot more than we can. > > if this change works for your circumstances, i'll go ahead and commit it > Yes, it works: Operating system: armv5tel-whatever-linux2 Configuring for linux-armv4 * Use configuration linux-armv4 -DL_ENDIAN
i glanced through the source and noticed that there also seems to be a new s390x target, so i added that on top of bfin/avr and these arm changes. thanks for testing.
