If the vectorized insertelements instructions form indentity subvector
(the subvector at the beginning of the long vector), it is just enough
to extend the vector itself, no need to generate inserting subvector
shuffle.
Differential Revision: https://reviews.llvm.org/D107494
If the vectorized insertelements instructions form indentity subvector
(the subvector at the beginning of the long vector), it is just enough
to extend the vector itself, no need to generate inserting subvector
shuffle.
Differential Revision: https://reviews.llvm.org/D107344
This reverts commit e408d1dfab42b27d0aa51b221e50fa6390fb5ed1 and
2 other (4b25c113210e579a5346ca0abc0717ab1ce5d9df and
c2deb2afafee991c06cc96dc5beecb6de448b9fc) related to fix the problem with the
reordering shuffles.
Replace insertelement instructions for splats with just single
insertelement + broadcast shuffle. Also, try to merge these instructions
if they come from the same/shuffled gather node.
Differential Revision: https://reviews.llvm.org/D107104
For the nodes with reused scalars the user may be not only of the size
of the final shuffle but also of the size of the scalars themselves,
need to check for this. It is safe to just modify the check here, since
the order of the scalars themselves is preserved, only indeces of the
reused scalars are changed. So, the users with the same size as the
number of scalars in the node, will not be affected, they still will get
the operands in the required order.
Reported by @mstorsjo in D105020.
Differential Revision: https://reviews.llvm.org/D107080
If the instruction was previously deleted, it should not be treated as
an external user. This fixes cost estimation and removes dead
extractelement instructions.
Differential Revision: https://reviews.llvm.org/D107106
Need to check that the minimum acceptable vector factor is at least 2,
not 0, to avoid compiler crash during gathered loads analysis.
Differential Revision: https://reviews.llvm.org/D107058
Reworked reordering algorithm. Originally, the compiler just tried to
detect the most common order in the reordarable nodes (loads, stores,
extractelements,extractvalues) and then fully rebuilding the graph in
the best order. This was not effecient, since it required an extra
memory and time for building/rebuilding tree, double the use of the
scheduling budget, which could lead to missing vectorization due to
exausted scheduling resources.
Patch provide 2-way approach for graph reodering problem. At first, all
reordering is done in-place, it doe not required tree
deleting/rebuilding, it just rotates the scalars/orders/reuses masks in
the graph node.
The first step (top-to bottom) rotates the whole graph, similarly to the previous
implementation. Compiler counts the number of the most used orders of
the graph nodes with the same vectorization factor and then rotates the
subgraph with the given vectorization factor to the most used order, if
it is not empty. Then repeats the same procedure for the subgraphs with
the smaller vectorization factor. We can do this because we still need
to reshuffle smaller subgraph when buildiong operands for the graph
nodes with lasrger vectorization factor, we can rotate just subgraph,
not the whole graph.
The second step (bottom-to-top) scans through the leaves and tries to
detect the users of the leaves which can be reordered. If the leaves can
be reorder in the best fashion, they are reordered and their user too.
It allows to remove double shuffles to the same ordering of the operands in
many cases and just reorder the user operations instead. Plus, it moves
the final shuffles closer to the top of the graph and in many cases
allows to remove extra shuffle because the same procedure is repeated
again and we can again merge some reordering masks and reorder user nodes
instead of the operands.
Also, patch improves cost model for gathering of loads, which improves
x264 benchmark in some cases.
Gives about +2% on AVX512 + LTO (more expected for AVX/AVX2) for {625,525}x264,
+3% for 508.namd, improves most of other benchmarks.
The compile and link time are almost the same, though in some cases it
should be better (we're not doing an extra instruction scheduling
anymore) + we may vectorize more code for the large basic blocks again
because of saving scheduling budget.
Differential Revision: https://reviews.llvm.org/D105020
Need to fix several cost-related problems. The final type may be defined
incorrectly because of to early definition (we may end up with the wider
type), the CommonCost should not be redefined in ExtractElements
cost related calculations and the shuffle of the final insertelements
vectors should be calculated as a cost of single vector permutations
+ costs of two vector permutations for other n-1 incoming vectors.
Differential Revision: https://reviews.llvm.org/D106578
Need to fix several cost-related problems. The final type may be defined
incorrectly because of to early definition (we may end up with the wider
type), the CommonCost should not be redefined in ExtractElements
cost related calculations and the shuffle of the final insertelements
vectors should be calculated as a cost of single vector permutations
+ costs of two vector permutations for other n-1 incoming vectors.
Differential Revision: https://reviews.llvm.org/D106578
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
The incoming values for PHI nodes may come from unreachable BasicBlocks,
need to handle this case.
Differential Revision: https://reviews.llvm.org/D106264
Part of D105020. Also, fixed FIXMEs that need to use wider vector type
when trying to calculate the cost of reused scalars. This may cause
regressions unless D100486 is landed to improve the cost estimations
for long vectors shuffling.
Differential Revision: https://reviews.llvm.org/D106060
The cost of the InsertSubvector shuffle kind cost is not complete and
may end up with just extracts + inserts costs in many cases. Added
a workaround to represent it as a generic PermuteSingleSrc, which is
still pessimistic but better than InsertSubvector.
Differential Revision: https://reviews.llvm.org/D105827
This bug was introduced with D105730 / 25ee55c0baff .
If we are not converting all of the operations of a reduction
into a vector op, we need to preserve the existing select form
of the remaining ops. Otherwise, we are potentially leaking
poison where it did not in the original code.
Alive2 agrees that the version that freezes some inputs
and then falls back to scalar is correct:
https://alive2.llvm.org/ce/z/erF4K2
The cost of the InsertSubvector shuffle kind cost is not complete and
may end up with just extracts + inserts costs in many cases. Added
a workaround to represent it as a generic PermuteSingleSrc, which is
still pessimistic but better than InsertSubvector.
Differential Revision: https://reviews.llvm.org/D105827
This has been a work-in-progress for a long time...we finally have all of
the pieces in place to handle vectorization of compare code as shown in:
https://llvm.org/PR41312
To do this (see PhaseOrdering tests), we converted SimplifyCFG and
InstCombine to the poison-safe (select) forms of the logic ops, so now we
need to have SLP recognize those patterns and insert a freeze op to make
a safe reduction:
https://alive2.llvm.org/ce/z/NH54Ah
We get the minimal patterns with this patch, but the PhaseOrdering tests
show that we still need adjustments to get the ideal IR in some or all of
the motivating cases.
Differential Revision: https://reviews.llvm.org/D105730
We know that "CVTTPS2SI" returns 0x80000000 for out of range inputs (and for FP_TO_UINT, negative float values are undefined). We can use this to make unsigned conversions from vXf32 to vXi32 more efficient, particularly on targets without blend using the following logic:
small := CVTTPS2SI(x);
fp_to_ui(x) := small | (CVTTPS2SI(x - 2^31) & ARITHMETIC_RIGHT_SHIFT(small, 31))
Even on targets where "PBLENDVPS"/"PBLENDVB" exists, it is often a latency 2, low throughput instruction so this logic is applied there too (in particular for AVX2 also). It furthermore gets rid of one high latency floating point comparison in the previous lowering.
@TomHender checked the correctness of this for all possible floats between -1 and 2^32 (both ends excluded).
Original Patch by @TomHender (Tom Hender)
Differential Revision: https://reviews.llvm.org/D89697
Update (mainly) vXf32/vXf64 -> vXi8/vXi16 fptosi/fptoui costs based on the worst case costs from the script in D103695.
Move to using legalized types wherever possible, which allows us to prune the cost tables.
Update truncation costs based on the worst case costs from the script in D103695.
Move to using legalized types wherever possible, which allows us to prune the cost tables.
Patch tries to improve the vectorization of stores. Originally, we just
check the type and the base pointer of the store.
Patch adds some extra checks to avoid non-profitable vectorization
cases. It includes analysis of the scalar values to be stored and
triggers the vectorization attempt only if the scalar values have
same/alt opcode and are from same basic block, i.e. we don't end up
immediately with the gather node, which is not profitable.
This also improves compile time by filtering out non-profitable cases.
Part of D57059.
Differential Revision: https://reviews.llvm.org/D104122
Revived D101297 in its original form + added some changes in X86
legalization cehcking for masked gathers.
This solution is the most stable and the most correct one. We have to
check the legality before trying to build the masked gather in SLP.
Without this check we have incorrect cost (for SLP) in case if the masked gather
is not legal/slower than the gather. And we're missing some
vectorization opportunities.
This can be fixed in the cost model, but in this case we need to add
special checks for the cost of GEPs for ScatterVectorize node, add
special check for small trees, etc., i.e. there are a lot of corner
cases here and there, which insrease code base and make it harder to
maintain the code.
> Can't we rely on cost model to deal with this? This can be profitable for futher vectorization, when we can start from such gather loads as seed.
The question from D101297. Actually, no, it can't. Actually, simple
gather may give us better result, especially after we started
vectorization of insertelements. Plus, like I said before, the cost for
non-legal masked gathers leads to missed vectorization opportunities.
Differential Revision: https://reviews.llvm.org/D105042
Update costs based on the worst case costs from the script in D103695.
Move to using legalized types wherever possible, which allows us to prune the cost tables.
Update (mainly) vXi8/vXi16 -> vXf32/vXf64 sitofp/uitofp costs based on the worst case costs from the script in D103695.
Move to using legalized types wherever possible, which allows us to prune the cost tables.
Compare type IDs and DFS numbering for basic block instead of addresses
to fix non-determinism.
Differential Revision: https://reviews.llvm.org/D105031
Provide a generic fallback that extends sub-i32 scalars before using the existing sitofp instructions.
These numbers can be tweaked for specific sse levels, but we should get the default handling in place first.
We get the extension for free for non-vector loads.
The function vectorizeChainsInBlock does not support scalable vector,
because function like canReuseExtract and isCommutative in the code
path assert with scalable vectors.
This patch avoids vectorizing blocks that have extract instructions with scalable
vector..
Differential Revision: https://reviews.llvm.org/D104809
Loads of <4 x i8> vectors were modeled as extremely expensive. And while we
don't have a load instruction that supports this, it isn't that expensive to
create a vector of i8 elements. The codegen for this was fixed/optimised in
D105110. This now tweaks the cost model and enables SLP vectorisation of my
motivating case loadi8.ll.
Differential Revision: https://reviews.llvm.org/D103629
Update v4i64 -> v4f32/v4f64 uitofp costs based on the worst case costs from the script in D103695.
Fixes a few regressions before we start adding AVX costs for legalized types.
The compiler should not ignore UndefValue when gathering the scalars,
otherwise the resulting code may be less defined than the original one.
Also, grouped scalars to insert them at first to reduce the analysis in
further passes.
Differential Revision: https://reviews.llvm.org/D105275
Building on rG2a1ef8784ad9a, adjust the SSE cost tables to use the legalized types based on the worst case costs from the script in D103695.
To account for different numbers of src/dst legalized type registers we must scale the cost by maximum of the src/dst, not just use src
Based off the worse case numbers generated by D103695, the AVX1/2/512 sitofp/uitofp/fptosi/fptoui costs were higher than necessary (based off instruction counts instead of actual throughput).
The SSE costs still need further fixes, but I hit an issue with the order in which SSE costs are checked - we need to check CUSTOM costs (with non-legal types) first, and then fallback to LEGALIZED types. I'm looking at this now, and this should let us start thinning out a lot of the duplicates in the costs tables.
Then we can finally start work on vXi64 / vXi16 / vXi8 / vXi1 integers, which should let us look at sub-128-bit vectorization (D103925).
