Link does not seem to work?
I've implemented some of the XBitmanip special functions in Verilog and compared them in size with a simple 32-bit rotate shift and the Rocket MulDiv core (default config: 5 cycle multiply, 32+ cycles divide). https://raw.githubusercontent.com/cliffordwolf/xbitmanip/master/xbitmanip-draft.pdf …pic.twitter.com/bVg5tsfJXj
-
-
- End of conversation
New conversation -
-
-
Very interesting read. A few drive-by comments. s/This/These in end of 2.9. More intended usage and examples would be nice. Also, one of the shift operations is SLOW!
-
This->These: Thanks! Fixed in LaTeX. (Will be in PDF when I update it next time.) More Example: Is planned. Also: I accept pull requests. :) Re Shift-Left-Ones-Word: Is this a problem? :D
- Show replies
New conversation -
-
-
Impressive. Shaping up quite nicely. I like that you have included decode_cj using the new intrinsics. Observation: Comparison with x86 Bit Manipulation ISAs is missing TZCOUNT “rep bsf” but has LZCOUNT “rep bsr” https://bugs.chromium.org/p/nativeclient/issues/detail?id=2869 …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Is it worth including TBM in the x86 comparison given they only replace simple 2 or 3 instruction sequences and AMD removed them from Jaguar and Zen. It’s not likely compilers will emit them given they exist on only 3 AMD micro-architectures (piledriver, steamroller, excavator)
-
The point is exactly to show that all of them are just 2 or 3 instructions and that none of those functions would be worth adding a dedicated xbitmanip instruction.
End of conversation
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.