This fixes the handling of subregister extract copies. This
will allow AMDGPU to remove its implementation of
shouldRewriteCopySrc, which exists as a 10 year old workaround
to this bug. peephole-opt-fold-reg-sequence-subreg.mir will
show the expected improvement once the custom implementation
is removed.
The copy coalescing processing here is overly abstracted
from what's actually happening. Previously when visiting
coalescable copy-like instructions, we would parse the
sources one at a time and then pass the def of the root
instruction into findNextSource. This means that the
first thing the new ValueTracker constructed would do
is getVRegDef to find the instruction we are currently
processing. This adds an unnecessary step, placing
a useless entry in the RewriteMap, and required skipping
the no-op case where getNewSource would return the original
source operand. This was a problem since in the case
of a subregister extract, shouldRewriteCopySource would always
say that it is useful to rewrite and the use-def chain walk
would abort, returning the original operand. Move the process
to start looking at the source operand to begin with.
This does not fix the confused handling in the uncoalescable
copy case which is proving to be more difficult. Some currently
handled cases have multiple defs from a single source, and other
handled cases have 0 input operands. It would be simpler if
this was implemented with isCopyLikeInstr, rather than guessing
at the operand structure as it does now.
There are some improvements and some regressions. The
regressions appear to be downstream issues for the most part. One
of the uglier regressions is in PPC, where a sequence of insert_subrgs
is used to build registers. I opened #125502 to use reg_sequence instead,
which may help.
The worst regression is an absurd SPARC testcase using a <251 x fp128>,
which uses a very long chain of insert_subregs.
We need improved subregister handling locally in PeepholeOptimizer,
and other pasess like MachineCSE to fix some of the other regressions.
We should handle subregister composes and folding more indexes
into insert_subreg and reg_sequence.
Patch was reverted due to test case (added) exposing an infinite loop in
combiner, where (shl C1, C2) create by performSHLCombine isn't
constant-folded:
Combining: t14: i64 = shl t12, Constant:i64<1>
Creating new node: t36: i64 = shl
OpaqueConstant:i64<-2401053089408754003>, Constant:i64<1>
Creating new node: t37: i64 = shl t6, Constant:i64<1>
Creating new node: t38: i64 = and t37, t36
... into: t38: i64 = and t37, t36
...
Combining: t38: i64 = and t37, t36
Creating new node: t39: i64 = and t6,
OpaqueConstant:i64<-2401053089408754003>
Creating new node: t40: i64 = shl t39, Constant:i64<1>
... into: t40: i64 = shl t39, Constant:i64<1>
and subsequently gets simplified by DAGCombiner::visitAND:
// Simplify: (and (op x...), (op y...)) -> (op (and x, y))
if (N0.getOpcode() == N1.getOpcode())
if (SDValue V = hoistLogicOpWithSameOpcodeHands(N))
return V;
before being folded by performSHLCombine once again and so on.
The combine in performSHLCombine should only be done if (shl C1, C2) can
be constant-folded, it may otherwise be unsafe and generally have a
worse end result. Thanks to Dave Sherwood for his insight on this one.
This reverts commit f719771f251d7c30eca448133fe85730f19a6bd1.
This reverts commit 9c319d5bb40785c969d2af76535ca62448dfafa7.
Some issues were discovered with the bootstrap builds, which
seem like they were caused by this commit. I'm reverting to investigate.
The existing heuristics were assuming that every core behaves like an
Apple A7, where any extend/shift costs an extra micro-op... but in
reality, nothing else behaves like that.
On some older Cortex designs, shifts by 1 or 4 cost extra, but all other
shifts/extensions are free. On all other cores, as far as I can tell,
all shifts/extensions for integer loads are free (i.e. the same cost as
an unshifted load).
To reflect this, this patch:
- Enables aggressive folding of shifts into loads by default.
- Removes the old AddrLSLFast feature, since it applies to everything
except A7 (and even if you are explicitly targeting A7, we want to
assume extensions are free because the code will almost always run on a
newer core).
- Adds a new feature AddrLSLSlow14 that applies specifically to the
Cortex cores where shifts by 1 or 4 cost extra.
I didn't add support for AddrLSLSlow14 on the GlobalISel side because it
would require a bunch of refactoring to work correctly. Someone can pick
this up as a followup.
Refresh of the generic scheduling model to use A510 instead of A55.
Main benefits are to the little core, and introducing SVE scheduling information.
Changes tested on various OoO cores, no performance degradation is seen.
Differential Revision: https://reviews.llvm.org/D156799
We would like to start pushing -mcpu=generic towards enabling the set of
features that improves performance for some CPUs, without hurting any
others. A blend of the performance options hopefully beneficial to all
CPUs. The largest part of that is enabling in-order scheduling using the
Cortex-A55 schedule model. This is similar to the Arm backend change
from eecb353d0e25ba which made -mcpu=generic perform in-order scheduling
using the cortex-a8 schedule model.
The idea is that in-order cpu's require the most help in instruction
scheduling, whereas out-of-order cpus can for the most part out-of-order
schedule around different codegen. Our benchmarking suggests that
hypothesis holds. When running on an in-order core this improved
performance by 3.8% geomean on a set of DSP workloads, 2% geomean on
some other embedded benchmark and between 1% and 1.8% on a set of
singlecore and multicore workloads, all running on a Cortex-A55 cluster.
On an out-of-order cpu the results are a lot more noisy but show flat
performance or an improvement. On the set of DSP and embedded
benchmarks, run on a Cortex-A78 there was a very noisy 1% speed
improvement. Using the most detailed results I could find, SPEC2006 runs
on a Neoverse N1 show a small increase in instruction count (+0.127%),
but a decrease in cycle counts (-0.155%, on average). The instruction
count is very low noise, the cycle count is more noisy with a 0.15%
decrease not being significant. SPEC2k17 shows a small decrease (-0.2%)
in instruction count leading to a -0.296% decrease in cycle count. These
results are within noise margins but tend to show a small improvement in
general.
When specifying an Apple target, clang will set "-target-cpu apple-a7"
on the command line, so should not be affected by this change when
running from clang. This also doesn't enable more runtime unrolling like
-mcpu=cortex-a55 does, only changing the schedule used.
A lot of existing tests have updated. This is a summary of the important
differences:
- Most changes are the same instructions in a different order.
- Sometimes this leads to very minor inefficiencies, such as requiring
an extra mov to move variables into r0/v0 for the return value of a test
function.
- misched-fusion.ll was no longer fusing the pairs of instructions it
should, as per D110561. I've changed the schedule used in the test
for now.
- neon-mla-mls.ll now uses "mul; sub" as opposed to "neg; mla" due to
the different latencies. This seems fine to me.
- Some SVE tests do not always remove movprfx where they did before due
to different register allocation giving different destructive forms.
- The tests argument-blocks-array-of-struct.ll and arm64-windows-calls.ll
produce two LDR where they previously produced an LDP due to
store-pair-suppress kicking in.
- arm64-ldp.ll and arm64-neon-copy.ll are missing pre/postinc on LPD.
- Some tests such as arm64-neon-mul-div.ll and
ragreedy-local-interval-cost.ll have more, less or just different
spilling.
- In aarch64_generated_funcs.ll.generated.expected one part of the
function is no longer outlined. Interestingly if I switch this to use
any other scheduled even less is outlined.
Some of these are expected to happen, such as differences in outlining
or register spilling. There will be places where these result in worse
codegen, places where they are better, with the SPEC instruction counts
suggesting it is not a decrease overall, on average.
Differential Revision: https://reviews.llvm.org/D110830
Add a comment when there is a shifted value,
add x9, x0, #291, lsl #12 ; =1191936
but not when the immediate value is unshifted,
subs x9, x0, #256 ; =256
when the comment adds nothing additional to the reader.
Differential Revision: https://reviews.llvm.org/D107196
This is generally more readable due to the way the assembler aliases
work.
(This causes a lot of test changes, but it's not really as scary as it
looks at first glance; it's just mechanically changing a bunch of checks
for orr to check for mov instead.)
Differential Revision: https://reviews.llvm.org/D59720
llvm-svn: 356954
@bextr64_32_b1 is extracted from hotpath of real-world code
(RawSpeed BitStream<>::peekBitsNoFill()) after `clang -O3`.
@bextr64_32_b2/@bextr64_32_b0 is the same pattern,
but with trunc done last, showing how i think it can be handled:
https://rise4fun.com/Alive/K4Bhttps://rise4fun.com/Alive/qC9
It is possible that middle-end should do some of this, too.
https://bugs.llvm.org/show_bug.cgi?id=36419
llvm-svn: 349998
It would be best to introduce ISD::BitFieldExtract,
because clearly more than one backend faces the same problem.
But for now let's solve this in the x86-specific DAG combine.
https://bugs.llvm.org/show_bug.cgi?id=38938
llvm-svn: 342880
Summary: Also, adjust the check prefixes so that we actually get to check the BMI1-only-case.
Reviewers: craig.topper, RKSimon, spatel, javed.absar
Reviewed By: RKSimon
Subscribers: kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D48490
llvm-svn: 342623
