There's quite a few places where the segment offset appending is open-coded
throughout the opcode dispatcher, but we can pull these out into a few
helpers to make the sites a little more compact and declarative.
This is something the new shift flag code will do. Backporting the opt since
that's stalled and this reduces the diff.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
single unified implementation for ROL & ROR (instead of 4 cases). no more
deferred flags because it's easy to shoot ourselves in the foot with deferred
flags w.r.t the new RA design, and rotates are rare enough with very efficient
flag calculations such that the extra JIT overhead should be minimal to DCE the
resulting calculations later.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
It has been a long time coming that FEX no longer needed to leak IR
implementation details to the frontend, this was legacy due to IR CI and
various other problems.
Now that the last bits of IR leaking has been removed, move everything
that we can internally to the implementation.
We still have a couple of minor details in the exposed IR.h to the
frontend, but these are limited to a few enums and some thunking struct
information rather than all the implementation details.
No functional change with this, just moving headers around.
I was looking at some other JIT overheads and this cropped up as some
overhead. Instead of materializing a constant using mov+movk+movk+movk,
load it from the named vector constant array.
In a micro-benchmark this improved performance by 34%.
In bytemark this improved on subbench by 0.82%
This reduces pointless constant usage. For now, it's no net change to
instcountci, but it should make it easier to get wins later. Hopefully.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
The usual tricks, also requires introducing a bare adc op to optimize adcs to,
but we wanted that anyway!
Also support a zero source, so we can calculate "foo + CF" in one instruction to
optimize the "lock adc" cases.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
we don't need the result, we can use subs and come out ahead in practice. also a
step towards better fusion
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
- Res is unused
- SrcSize doesn't matter since we ignore the high bits, might as well always use
32-bit, it doesn't matter
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
We only need each part of W extracted in the corresponding round, so sink the
extract into the round to reduce pressure.
Further, W and E are added and then never used again. So, by reassociating we
can do the add upfront, killing W and E at the start and further reducing
pressure.
Eliminates spilling in sha1rnds4.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
as far as flags go, they're identical: set ZF for zero output, set CF for output
= DestSize, undef the rest. merge the impls, so we get the optimized lzcnt impl
for tzcnt.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Paulo Matos