breaking: polish @clz() @ctz() @popcount() @bswap() @bitReverse() #2134
Conversation
|
Why are the new files added? |
Whoops, thanks for spotting that. |
shawnl
force-pushed
the
builtins-consistancy
branch
7 times, most recently
from
7e25928 to
a401388
Compare
|
I'm now working on exanding the vector features sufficiently to get this algorithm ported to zig https://github.com/cyb70289/utf8 using llvm vectors, rather than assembly intrinsics. |
shawnl
force-pushed
the
builtins-consistancy
branch
from
f1c765d to
157ade9
Compare
…d add @fshl() and @fshr() Closes: ziglang#2119 breaking: @bitreverse -> @bitreverse, @bswap -> @bswap Closes: ziglang#2120 Comes with tests The implicit casting fixups are not in here due to technical issues. see ziglang#2121
The optimizer can only optimize that code if it can guarantee that source and dest do not overlap.
…int_fits_in_bits code Also fix a few nasty bugs in bigint_fits_in_bits, such as handling of the value 0.
Some platforms don't support vectors, and it should be easy to skip these tests for those platforms
shawnl
force-pushed
the
builtins-consistancy
branch
from
3e43f36 to
c005271
Compare
no safety checks no 0-width type support
shawnl
force-pushed
the
builtins-consistancy
branch
from
c005271 to
1430867
Compare
Also, change the interface to accept an appropiately ranged type. While this is technically breaking, no users used this useless feature.
| var neg32: i32 = -16773785; | ||
| expect(@bitreverse(i32, -16773785) == @bitreverse(i32, neg32)); | ||
| var neg40: i40 = minInt(i40) + 12345; | ||
| expect(@bitreverse(i40, minInt(i40) + 12345) == @bitreverse(i40, neg40)); |
There was a problem hiding this comment.
Why remove these?
There was a problem hiding this comment.
because they are tautologies. They are not testing bitreverse.
There was a problem hiding this comment.
At the time that code was written, I'm not sure about now, the comment at the top of the block explained correctly what was happening. The LHS values were calculated by comptime code in the Zig compiler while the RHS values are calculated on the LLVM side. These tests were used to make sure the two match.
There was a problem hiding this comment.
@vegecode is correct, and it's still true. @bitreverse(i32, -16773785) exercises comptime evaluation; @bitreverse(i32, neg32) with neg32 as a var exercises runtime evaluation.
shawnl
force-pushed
the
builtins-consistancy
branch
from
1292aa0 to
e90f46e
Compare
|
This is too much. I can't keep up with this ever-growing pull request. @shawnl I need you to make one pull request at a time, which only tackles one thing. |
Closes: #2119
breaking: @bitreverse -> @bitreverse, @bswap -> @bswap
Closes: #2120
Comes with tests
OK. This has changed quite a bit since it was opened. It gained
@fshland@fshr, and then I added@memMoveand merged a differn't pull request so I could flesh out@memMove. I am working on #767