There is a narrow special-case in isImpliedCondICmps that can benefit
from being taught about samesign. Since it costs us nothing to implement
it, teach it about samesign, for completeness. This patch marks the
completion of the effort to teach ValueTracking about samesign.
Move isImplied{True,False}ByMatchingCmp from CmpInst to ICmpInst, so
that it can operate on CmpPredicate instead of CmpInst::Predicate, and
teach it about samesign. There are two callers of this function, and we
choose to migrate the one in ValueTracking, namely
isImpliedCondMatchingOperands to CmpPredicate, hence teaching it about
samesign, with visible test impact.
We have a textual representation of contextual profiles for test scenarios, mainly. This patch moves that to YAML instead of JSON. YAML is more succinct and readable (some of the .ll tests should be illustrative). In addition, JSON is parse-able by the YAML reader.
A subsequent patch will address deserialization.
(thanks, @kazutakahirata, for showing me how to use the llvm YAML reader/writer APIs, which I incorrectly thought to be more low-level than the JSON ones!)
isImpliedCondICmps() and its callers in ValueTracking can greatly
benefit from being taught about samesign. As a first step, teach one
caller, namely isImpliedCondOperands(). Very minimal changes are
required for this, as CmpPredicate::getMatching() does most of the work.
I added some CodeGen test cases related to reduce. To maintain
consistency, I also added cases for instructions like
`vector.reduce.or`.
For cases where `v1i1` type generates `VFIRST`, please refer to:
https://reviews.llvm.org/D139512.
These can generally be emitted using an ext instruction or mov from the
high half. The half half extracts can be free depending on the users,
but that is not handled here, just the basic costs. It originally
included all subvector extracts, but that was toned-down to just
half-vector extracts to try and help the mid end not breakup high/low
extracts without having the SLP vectorizer create a mess using other
shuffles.
the `ptx_kernel` calling convention is a more idiomatic and standard way
of specifying a NVPTX kernel than using the metadata which is not
supposed to change the meaning of the program. Further, checking the
calling convention is significantly faster than traversing the metadata,
improving compile time.
This change updates the clang and mlir frontends as well as the
NVPTXCtorDtorLowering pass to emit kernels using the calling convention.
In addition, this updates all NVPTX unit tests to use the calling
convention as well.
This is just the MOVSS instruction (SSE41 INSERTPS is still necessary for index != 0)
This exposed an issue in VectorCombine::foldInsExtFNeg - we need to use the more general SK_PermuteTwoSrc shuffle kind to allow getShuffleCost to match other shuffle kinds (not just SK_Select).
improveShuffleKindFromMask matches this as a SK_InsertSubvector of a v1f32 (which legalises to f32) into a v4f32 base vector, making it easy to recognise. MOVSS is limited to index0.
This expands the recently added fp16 fpext and fpround costs to bf16.
Some of the costs are taken from the rough number of instructions
needed, some are a little aspirational.
https://godbolt.org/z/bGEEd1vsW
Avoid always assuming the worst for v4f32 2 input shuffles, and match the SHUFPS pattern where possible - each pair of output elements must come from the same source register.
Now that processShuffleMasks can correctly handle 2 src shuffles, we can completely remove the shuffle kind limits and correctly recognize the number of active subvectors per legalized shuffle - improveShuffleKindFromMask will determine the shuffle kind for each split subvector.
This splits the shuffle-select CostModel test into a seperate CodeGen test and
removes the codegen from the CostModel version. An extra fp16 test is added
too.
When `collectFromBlock` is called without a predecessor (in particular
for loops that don't have a unique predecessor outside the loop) we
never start climbing the predecessor chain, and thus don't mark the
starting block as visited.
Fixes https://github.com/llvm/llvm-project/issues/120615.
A signed min-max clamp is the sequence of smin and smax intrinsics,
which constrain a signed value into the range: smin <= value <= smax.
The patch improves the calculation of KnownBits for a value subjected to
the signed clamping.
Adds cost estimation for the variants of the permutations of the scalar
values, used in gather nodes. Currently, SLP just unconditionally emits
shuffles for the reused buildvectors, but in some cases better to leave
them as buildvectors rather than shuffles, if the cost of such
buildvectors is better.
X86, AVX512, -O3+LTO
Metric: size..text
Program size..text
results results0 diff
test-suite :: External/SPEC/CINT2006/445.gobmk/445.gobmk.test 912998.00 913238.00 0.0%
test-suite :: MultiSource/Benchmarks/MiBench/consumer-lame/consumer-lame.test 203070.00 203102.00 0.0%
test-suite :: External/SPEC/CFP2017speed/638.imagick_s/638.imagick_s.test 1396320.00 1396448.00 0.0%
test-suite :: External/SPEC/CFP2017rate/538.imagick_r/538.imagick_r.test 1396320.00 1396448.00 0.0%
test-suite :: MultiSource/Benchmarks/Bullet/bullet.test 309790.00 309678.00 -0.0%
test-suite :: External/SPEC/CFP2017rate/526.blender_r/526.blender_r.test 12477607.00 12470807.00 -0.1%
CINT2006/445.gobmk - extra code vectorized
MiBench/consumer-lame - small variations
CFP2017speed/638.imagick_s
CFP2017rate/538.imagick_r - extra vectorized code
Benchmarks/Bullet - extra code vectorized
CFP2017rate/526.blender_r - extra vector code
RISC-V, sifive-p670, -O3+LTO
CFP2006/433.milc - regressions, should be fixed by https://github.com/llvm/llvm-project/pull/115173
CFP2006/453.povray - extra vectorized code
CFP2017rate/508.namd_r - better vector code
CFP2017rate/510.parest_r - extra vectorized code
SPEC/CFP2017rate - extra/better vector code
CFP2017rate/526.blender_r - extra vectorized code
CFP2017rate/538.imagick_r - extra vectorized code
CINT2006/403.gcc - extra vectorized code
CINT2006/445.gobmk - extra vectorized code
CINT2006/464.h264ref - extra vectorized code
CINT2006/483.xalancbmk - small variations
CINT2017rate/525.x264_r - better vectorization
Reviewers: RKSimon
Reviewed By: RKSimon
Pull Request: https://github.com/llvm/llvm-project/pull/115201
When assumptions are present `Terms.size()` does not actually count the
number of conditions collected from dominating branches; introduce a
separate counter.
Fixes https://github.com/llvm/llvm-project/issues/120237
processShuffleMasks can now correctly handle 2 src shuffles, so we can use the existing SK_PermuteSingleSrc splitting cost logic to handle SK_PermuteTwoSrc as well and correctly recognise the number of active subvectors per legalised shuffle.
processShuffleMasks can now correctly handle 2 src shuffles, so we can use the existing SK_PermuteSingleSrc splitting cost logic to handle SK_PermuteTwoSrc as well and correctly recognise the number of active subvectors per legalised shuffle.
If the shuffle split results in referencing a single legalised whole vector (i.e. no permutation), then this can be treated as free.
We already do something similar for broadcasts / whole subvector insertion + extraction - its purely an issue for register allocation.
== We were previously returning an invalid cost when truncating
anything to <vscale x 2 x i1>, which is incorrect since we can
generate perfectly good code for this.
== The costs for truncating legal or unpacked types to predicates
seemed overly optimistic. For example, when truncating
<vscale x 8 x i16> to <vscale x 8 x i1> we typically do
something like
and z0.h, z0.h, #0x1
cmpne p0.h, p0/z, z0.h, #0
I guess it might depend upon whether the input value is
generated in the same block or not and if we can avoid the
inreg zero-extend. However, it feels safe to take the more
conservative cost here.
== The costs for some truncates such as
trunc <vscale x 2 x i32> %a to <vscale x 2 x i16>
were 1, whereas in actual fact they are free and no instructions
are required.
== Also, for this
trunc <vscale x 8 x i32> %a to <vscale x 8 x i16>
it's just a single uzp1 instruction so I reduced the cost to 1.
In general, I've added costs for all cases where the destination
type is legal or unpacked. One unfortunate side effect of this
is the costs for some fixed-width truncates when using SVE now
look too optimistic.
This splits the DXILResourceAnalysis pass into TypeAnalysis and
BindingAnalysis passes. The type analysis pass is made immutable and
populated lazily so that it can be used earlier in the pipeline without
needing to carefully maintain the invariants of the binding analysis.
Fixes#118400
Instead of storing an auxilliary structure with the information from the
DXIL resource target extension types duplicated, access the information
that we can via the type itself.
This also means we need to handle some of the target extension types we
haven't fully defined yet, like Texture and CBuffer. For now we make an
educated guess to what those should look like based on llvm/wg-hlsl#76,
and we can update them fairly easily when we've defined them more
thoroughly.
First part of #118400
A SCEVWrapPredicate A implies B, if
* they have the same flag,
* both steps are positive and
* B's start and step are ULE/SLE (for NSUW/NSSW) than A's.
See https://alive2.llvm.org/ce/z/n2T4ss (first pair with known constants
as strides, second pair with variable strides).
Note that this is limited to steps of the same size, due to NSUW having
slightly different semantics than regular NUW. We should be able to
remove this restriction for NSSW (which matches NSW) in the future.
PR: https://github.com/llvm/llvm-project/pull/118184
BasicAA currently tries to support addrspacecasts that change the index
width by performing the decomposition in the maximum of all index widths
and then trying to fix this up with in-place sign extends to get correct
overflow behavior if the actual index width is smaller.
However, even in the case where we don't mix different index widths and
just have an index width that is smaller than the maximum, the behavior
is incorrect (see test), because we only perform the index width
adjustment during decomposition and not any of the later logic -- and we
don't do anything at all for variable offsets. I'm sure that the case
where we actually mix different index widths is even more broken than
that.
Fix this by not allowing decomposition through index width changes. If
the pointers have different index widths, fall back to a base object
comparison, ignoring the offsets.
This patch implements the cost when the size of the vector need to split
into multiple groups and the index exceed single vector group.
For extract element, we need to store split vectors to stack and load
the target element.
For insert element, we need to store split vectors to stack and store
the target element and load vectors back.
After this patch, the cost of insert/extract element will close to the
generated assembly.
The base cost approximates the expansion code in SelectionDAGBuilder.
For the AArch64 cases that don't need generic expansion, fixed-length
search vectors have a higher cost than scalable vectors due to the extra
instructions to convert the boolean mask.
This adds some basic costs for fpext and fpround, many of which were
already handled by the generic costing routines but this does make some
adjustments for larger vector types that can use fcvtn+fcvtn2, as
opposed to fcvtn+fcvtn+concat.
These should now more closely match the codegen from
https://godbolt.org/z/r3P9Mf8ez, for example.
In cases where the base/sub vector type in an insert_subvector pattern legalize to the same width through splitting, we can assume that the shuffle becomes free as the legalized vectors will not overlap.
Note this isn't true if the vectors have been widened during legalization (e.g. v2f32 insertion into v4f32 would legalize to v4f32 into v4f32).
Noticed while working on adding processShuffleMasks handling for SK_PermuteTwoSrc.
Some of the scalable tests have been split off to make the tests more
managable. AArch64TTIImpl::getCastInstrCost is also formatted to avoid the need
to fight against CI.
