Even without knl, clang does some wacky things at -O3, that someone should maybe look into.https://godbolt.org/z/kE2tiK
-
-
The -Os codegen is considerably more sane, but also depressing in its own way.
1 reply 0 retweets 2 likes -
We really need -O2.5 that's like "yeah, go ahead and vectorize, but let's not go *completely* nuts, ok?" Or we could just kill -O3, that would be OK with me.
2 replies 0 retweets 7 likes -
new optimization levels: -O0 -O1 -Os -O2 -OGentoo
2 replies 1 retweet 21 likes -
-
-
-Olordhecomin
1 reply 0 retweets 7 likes -
Replying to @stephentyrone @jeeger and
My original question is really about optimizing for maximum number of opcodes, or -Oops
2 replies 0 retweets 9 likes -
Replying to @moyix @stephentyrone and
brb building a superoptimizer that saves opcodes by doing incredibly slow and horrible things that thrash the pipeline
2 replies 0 retweets 9 likes -
Has someone filed a bug on getting LLVM to emit XLAT where possible yet
-
-
Replying to @pcwalton @stephentyrone and
my favorite x86 instruction, to be quite honest. right up there with xadd
2 replies 0 retweets 4 likes -
Replying to @iximeow @stephentyrone and
I wanted to "suggest" using AAD with arbitrary base to save code size but Intel is apparently hell bent on ruining all my funpic.twitter.com/bJfVCLVOiT
1 reply 0 retweets 5 likes - 3 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.