This allows it to produce a more accurate cost for the shuffle, using
the more accurate calls to getShuffleCost in getInstructionCost. It
helps fix some of the regressions from vector combine a little while
ago, now that we have better subvector extract costs.
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.
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.
<u,1,u,u> 'inplace' single src shuffles can be treated as free identity shuffles - ignore any shuffle cost (similar to what we already do in other folds like foldShuffleOfShuffles) - eventually getShuffleCost should just return TCC_Free in these cases but in a lot of the targets' shuffle cost logic this currently ends up treated as a generic SK_PermuteSingleSrc.
We still want to generate the shuffle as it will help further shuffle folds with the additional PoisonMaskElem 'undemanded' elements.
Some patterns (in particular horizontal style patterns) can end up with shuffles straddling both sides of a binop/cmp.
Where individually the folds aren't worth it, by merging the (oneuse) shuffles we can notably reduce the net instruction count and cost.
One of the final steps towards finally addressing #34072
`-passes='default<O3>'` isn't correctly parsed on DOS, so when update_test_checks.py runs a system call on the opt RUN line, it fails to evaluate properly - use `-passes="default<O3>"` instead.
Currently available intrinsics are only ld2/st2, which don't support interleaving factor > 2.
This patch teaches the LV to use ld2/st2 recursively to support high
interleaving factors.
We already canonicalized an undef base vector to the RHS to improve further folding, this extends this to improve the shuffle cost estimate of the single src shuffle
foldShuffleOfShuffles currently only folds unary shuffles to ensure we don't end up with a merged shuffle with more than 2 sources, but this prevented cases where both shuffles were sharing sources.
This patch generalizes the merge process to find up to 2 sources as it merges with the inner shuffles, it also moves the undef/poison handling stages into the merge loop as well.
Fixes#120764
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.
Add tests for horizontal add patterns with missing/undemanded elements - which typically prevents folding to the (add (shuffle a, b),(shuffle a, b)) optimal pattern
This was originally done to reduce the diff for the change. Remove it
and update the remaining tests. NFC modulo reordering of incoming
values.
Clean up after https://github.com/llvm/llvm-project/pull/114292.
As a first step to move towards modeling the full skeleton in VPlan,
start by wrapping IR blocks created during legacy skeleton creation in
VPIRBasicBlocks and hook them into the VPlan. This means the skeleton
CFG is represented in VPlan, just before execute. This allows moving
parts of skeleton creation into recipes in the VPBBs gradually.
Note that this allows retiring some manual DT updates, as this will be
handled automatically during VPlan execution.
PR: https://github.com/llvm/llvm-project/pull/114292
Mask/Bool vectors are often bitcast to/from scalar integers, in particular when concatenating mask results, often this is due to the difficulties of working with vector of bools on C/C++. On x86 this typically involves the MOVMSK/KMOV instructions.
To concatenate bool masks, these are typically cast to scalars, which are then zero-extended, shifted and OR'd together.
This patch attempts to match these scalar concatenation patterns and convert them to vector shuffles instead. This in turn often assists with further vector combines, depending on the cost model.
Fixes#111431
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.
foldShuffleOfShuffles already handles "shuffle (shuffle x, undef), (shuffle y, undef)" patterns, this patch relaxes the requirement so it can handle cases where only a single operand is a shuffle (and the other can be any other value and will be kept in place).
Fixes#86068
When we have a gep inbounds from the base of an object (e.g. alloca or
global), we know that the index cannot be negative, as this would go out
of bounds. As such, we can infer nuw as well.
The implementation is a bit stricter than necessary, we could also
accept one unknown index followed by known-non-negative indices.
Proof: https://alive2.llvm.org/ce/z/Hp7-6w (Note that alive2 currently
incorrectly doesn't require the inbounds for the alloca case, see
https://github.com/AliveToolkit/alive2/issues/1138).
SK_PermuteTwoSrc legalization has to assume any of the legalised source registers could be referenced in split shuffles, but if we already know that each 128-bit lane only references elements from the same lane of the source operands, then this scaling won't occur.
Hopefully this can help with #113356 without us having to get full processShuffleMasks canonicalization finished first.
When the size is an appropriate constant, __memcpy_chk will turn into a
memcpy that gets folded away by InstCombine. Therefore this patch avoids
counting these as calls for purposes of inlining costs.
This is only really relevant on platforms whose headers redirect memcpy
to __memcpy_chk (such as Darwin). On platforms that use intrinsics,
memcpy and similar functions are already exempt from call penalties.
Preserve llvm.access.group metadata on the replacement instruction, if
it does not move. In that case, the program would be UB, if the parallel
property encoded in the metadata does not hold.
This matches the LangRef recently updated in #116220
PR https://github.com/llvm/llvm-project/pull/115868
- Consider MainLoopVF * IC when determining whether Epilogue
Vectorization is profitable
- Allow the same VF for the Epilogue as for the main loop
- Use an upper bound for the trip count of the Epilogue when choosing
the Epilogue VF
PR: https://github.com/llvm/llvm-project/pull/108190
---------
Co-authored-by: Florian Hahn <flo@fhahn.com>
I noticed that the two C functions emitted different IR:
```
int switch_duplicate_arms(int switch_val, int v, int w) {
switch (switch_val) {
default:
break;
case 0:
w = v;
break;
case 1:
w = v;
break;
}
return w;
}
int if_duplicate_arms(int switch_val, int v, int w) {
if (switch_val == 0)
w = v;
else if (switch_val == 1)
w = v;
return v0;
}
```
We generate IR that looks like this:
```
define i32 @switch_duplicate_arms(i32 %0, i32 %1, i32 %2, i32 %3) {
switch i32 %1, label %7 [
i32 0, label %5
i32 1, label %6
]
5:
br label %7
6:
br label %7
7:
%8 = phi i32 [ %3, %4 ], [ %2, %6 ], [ %2, %5 ]
ret i32 %8
}
define i32 @if_duplicate_arms(i32 %0, i32 %1, i32 %2, i32 %3) {
%5 = icmp ult i32 %1, 2
%6 = select i1 %5, i32 %2, i32 %3
ret i32 %6
}
```
For `switch_duplicate_arms`, taking case 0 and 1 are the same since %5
and %6
branch to the same location and the incoming values for %8 are the same
from
those blocks. We could remove one on the duplicate switch targets and
update
the switch with the single target.
On RISC-V, prior to this patch, we generate the following code:
```
switch_duplicate_arms:
li a4, 1
beq a1, a4, .LBB0_2
mv a0, a3
bnez a1, .LBB0_3
.LBB0_2:
mv a0, a2
.LBB0_3:
ret
if_duplicate_arms:
li a4, 2
mv a0, a2
bltu a1, a4, .LBB1_2
mv a0, a3
.LBB1_2:
ret
```
After this patch, the O3 code is optimized to the icmp + select pair,
which
gives us the same code gen as `if_duplicate_arms`, as desired. This
results
is one less branch instruction in the final assembly.
This may help with both code size and further switch simplification. I
found
that this patch causes no significant impact to spec2006/int/ref and
spec2017/intrate/ref.
---------
Co-authored-by: Min Hsu <min@myhsu.dev>
Run ArgumentPromotion before IPSCCP in the LTO pipeline, to expose more
constants to be propagated. We also run PostOrderFunctionAttrs to
improve the information available to ArgumentPromotion's alias analysis,
and SROA to clean up allocas.
Relands #111163.
Most of the x86 shuffle instructions operate within each 128-bit subvector lane, but our shuffle costs struggle to handle this and have to fallback to worst case shuffles that reference elements from any lane.
This patch detects shuffle masks that we know are "inlane" and enable us to assume a cheaper shuffle cost.
