The vectoriser sometimes generates predicated vector loops using
the llvm.get.active.lane.mask intrinsic so it's important that we
are able to calculate a valid cost for the call instruction. When
SVE is enabled we are able to use a single whilelo instruction
for some vector types - in such cases I've marked the cost as 1.
For all other cases I've set the cost according to how the intrinsic
will be expanded.
Tests added here:
Analysis/CostModel/AArch64/sve-intrinsics.ll
Analysis/CostModel/ARM/active_lane_mask.ll
Analysis/CostModel/RISCV/active_lane_mask.ll
Differential Revision: https://reviews.llvm.org/D121109
Currently the cost model under-estimates the cost of certain
FP16 conversions.
This patch updates getCastInstrCost to return more accurate costs for
the cases improved in c2ed9fd054.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D113700
The costs of vector shifts was 2 as opposed to 1, as the nodes are
marked custom. Fix this like the others and mark the nodes as cheap.
Differential Revision: https://reviews.llvm.org/D120773
The current cost-model overestimates the cost of vector compares &
selects for ordered floating point compares. This patch fixes that by
extending the existing logic for integer predicates.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D118256
We have some bitcasts which we know will be simplified,
so their cost is zero.
Reviewed By: david-arm, sdesmalen
Differential Revision: https://reviews.llvm.org/D118019
If we are inserting into or extracting from a scalable vector we do
not know the number of elements at runtime, so we can only let the
index wrap for fixed-length vectors.
Tests added here:
Analysis/CostModel/AArch64/sve-insert-extract.ll
Differential Revision: https://reviews.llvm.org/D117099
The code size cost model for most targets uses the legalization cost for the type of the pointer of a load. If this load is followed directly by a trunc instruction, and is the only use of the result of the load, only one instruction is generated in the target assembly language. This adds a check for this case, and uses the target type of the trunc instruction if so.
This did not show any changes in CTMark code size benchmarks.
Reviewers: paquette, samparker, dmgreen
Differential Revision: https://reviews.llvm.org/D109388
Similar to D116732, this adds basic scalar sadd_with_overflow,
uadd_with_overflow, ssub_with_overflow and usub_with_overflow costs for
aarch64, which are usually quite efficiently lowered.
Differential Revision: https://reviews.llvm.org/D116734
This adds some AArch64 specific smul_with_overflow and umul_with_overflow
costs, overriding the default costs. The code generation for these mul
with overflow intrinsics is usually better than the default expansion on
AArch64. The costs come from https://godbolt.org/z/zEzYhMWqo with various
types, or llvm/test/CodeGen/AArch64/arm64-xaluo.ll.
Differential Revision: https://reviews.llvm.org/D116732
This patch adds on an overhead cost for gathers and scatters, which
is a rough estimate based on performance investigations I have
performed on SVE hardware for various micro-benchmarks.
Differential Revision: https://reviews.llvm.org/D115143
The default for min is changed to 1. The behaviour of -mvscale-{min,max}
in Clang is also changed such that 16 is the max vscale when targeting
SVE and no max is specified.
Reviewed By: sdesmalen, paulwalker-arm
Differential Revision: https://reviews.llvm.org/D113294
This patch introduces a new function:
AArch64Subtarget::getVScaleForTuning
that returns a value for vscale that can be used for tuning the cost
model when using scalable vectors. The VScaleForTuning option in
AArch64Subtarget is initialised according to the following rules:
1. If the user has specified the CPU to tune for we use that, else
2. If the target CPU was specified we use that, else
3. The tuning is set to "generic".
For CPUs of type "generic" I have assumed that vscale=2.
New tests added here:
Analysis/CostModel/AArch64/sve-gather.ll
Analysis/CostModel/AArch64/sve-scatter.ll
Transforms/LoopVectorize/AArch64/sve-strict-fadd-cost.ll
Differential Revision: https://reviews.llvm.org/D110259
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 eecb353d0e 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
The expansion for these was updated in https://reviews.llvm.org/D47927 but the cost model was not adjusted.
I believe the cost model was also incorrect for the old expansion.
The expansion prior to D47927 used 3 icmps using LHS, RHS, and Result
to calculate theirs signs. Then 2 icmps to compare the signs. Followed
by an And. The previous cost model was using 3 icmps and 2 selects.
Digging back through git blame, those 2 selects in the cost model used to
be 2 icmps, but were changed in https://reviews.llvm.org/D90681
Differential Revision: https://reviews.llvm.org/D110739
We previously didn't have any tests to defend the cost model
for gathers and scatters using SVE without a vscale_range
attribute. I've added tests to existing files:
Analysis/CostModel/AArch64/sve-gather.ll
Analysis/CostModel/AArch64/sve-scatter.ll
Differential Revision: https://reviews.llvm.org/D109055
Several FP instructions (fadd, fsub, etc.) were incorrectly assigned
a higher cost for SVE because they have custom lowering, however we
know they are legal. This patch explicitly assigns a cost of 2 to
these opcodes.
Tests added here:
Analysis/CostModel/AArch64/arith-fp-sve.ll
Differential Revision: https://reviews.llvm.org/D108993
Tell the cost model to use the scalable calculation for non-neon fixed vector.
This results in a cheaper cost for fixed-length SVE masked gathers/scatters
allowing the vectorizor to emit them more frequently.
For tight loops like this:
float r = 0;
for (int i = 0; i < n; i++) {
r += a[i];
}
it's better not to vectorise at -O3 using fixed-width ordered reductions
on AArch64 targets. Although the resulting number of instructions in the
generated code ends up being comparable to not vectorising at all, there
may be additional costs on some CPUs, for example perhaps the scheduling
is worse. It makes sense to deter vectorisation in tight loops.
Differential Revision: https://reviews.llvm.org/D108292
Removed AArch64 usage of the getMaxVScale interface, replacing it with
the vscale_range(min, max) IR Attribute.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D106277
This takes the existing SVE costing for the various min/max reduction
intrinsics and expands it to NEON, where I believe it applies equally
well.
In the process it changes the lowering to use min/max cost, as opposed
to summing up the cost of ICmp+Select.
Differential Revision: https://reviews.llvm.org/D106239
This expands the cost model test for min/max to many more types,
including floating point minnum/maxnum and minimum/maximum, and FP16
with and without fullfp16. The old llc run lines are removed, as those
are better tested by CodeGen tests.
I have added a new FastMathFlags parameter to getArithmeticReductionCost
to indicate what type of reduction we are performing:
1. Tree-wise. This is the typical fast-math reduction that involves
continually splitting a vector up into halves and adding each
half together until we get a scalar result. This is the default
behaviour for integers, whereas for floating point we only do this
if reassociation is allowed.
2. Ordered. This now allows us to estimate the cost of performing
a strict vector reduction by treating it as a series of scalar
operations in lane order. This is the case when FP reassociation
is not permitted. For scalable vectors this is more difficult
because at compile time we do not know how many lanes there are,
and so we use the worst case maximum vscale value.
I have also fixed getTypeBasedIntrinsicInstrCost to pass in the
FastMathFlags, which meant fixing up some X86 tests where we always
assumed the vector.reduce.fadd/mul intrinsics were 'fast'.
New tests have been added here:
Analysis/CostModel/AArch64/reduce-fadd.ll
Analysis/CostModel/AArch64/sve-intrinsics.ll
Transforms/LoopVectorize/AArch64/strict-fadd-cost.ll
Transforms/LoopVectorize/AArch64/sve-strict-fadd-cost.ll
Differential Revision: https://reviews.llvm.org/D105432
When BasicTTIImpl::getCastInstrCost can't determine the cost of a
vector cast operation when the types need legalization, it falls
back to calculating scalarization costs. Instead of crashing on
`cast<FixedVectorType>(DstVTy)` when the type is a scalable vector,
return an Invalid cost.
Reviewed By: david-arm
Differential Revision: https://reviews.llvm.org/D106655
This adds some missing single source shuffle costs for AArch64, of i16
and i8 vectors. v4i16 are the same as v4i32 with a worse case cost of 3
coming from the perfect shuffle tables. The larger vector sizes expand
into a constant pool, plus a load (and adrp) and a tbl. I arbitrarily
chose 8 for the cost to be expensive but not too expensive.
Differential Revision: https://reviews.llvm.org/D106241
This changes the cost to (LT.first-1) * cost(add) + 2, where the cost of
an add is assumed to be 1. This brings it inline with the other
reductions.
Differential Revision: https://reviews.llvm.org/D106240
David Sherwood