peterbell10 55430f8673 [NVPTX] Customize getScalarizationOverhead (#128077) ...

We've observed that the SLPVectorizer is too conservative on NVPTX
because it over-estimates the cost to build a vector. PTX has a single
`mov` instruction that can build e.g. `<2 x half>` vectors from scalars,
however the SLPVectorizer over-estimates it as the cost of 2 insert
elements.

To fix this I customize `getScalarizationOverhead` to lower the cost for
building 2x16 types.

2025-03-29 01:31:33 +00:00

..

AArch64

Revert "[SLP] Make getSameOpcode support interchangeable instructions. (#132887)"

2025-03-26 10:24:25 +02:00

AMDGPU

[SLP] Sort PHIs by ExtractElements when relevant (#131229)

2025-03-17 14:19:46 -07:00

ARM

[SLP]Check the MainOp matches the requirements for the instructions

2025-01-28 06:00:52 -08:00

NVPTX

[NVPTX] Customize getScalarizationOverhead (#128077)

2025-03-29 01:31:33 +00:00

PowerPC

…

RISCV

Revert "[SLP] Make getSameOpcode support interchangeable instructions. (#132887)"

2025-03-26 10:24:25 +02:00

SystemZ

[SLP][REVEC] Fix scalar mask is passed to getScalarizationOverhead but the type is vector. (#128476)

2025-02-24 23:43:27 +08:00

VE

…

WebAssembly

[InstCombine] Infer nusw + nneg -> nuw for getelementptr (#111144)

2024-12-05 14:36:40 +01:00

X86

[SLP]Fix spill cost analysis for split vectorized nodes

2025-03-28 12:45:53 -07:00

XCore

…

abs-overflow-incorrect-minbws.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

addsub.ll

[SLP]Reduce number of alternate instruction, where possible

2025-03-12 08:18:51 -07:00

alt-cmp-vectorize.ll

…

alternate-cmp-swapped-pred-parent.ll

[SLP]Add cost estimation for gather node reshuffling

2024-12-24 15:35:29 -05:00

alternate-non-profitable.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

alternate-opcode-sindle-bv.ll

Revert "[SLP] Make getSameOpcode support interchangeable instructions. (#132887)"

2025-03-26 10:24:25 +02:00

arith-div-undef.ll

…

bool-logical-op-reduction-with-poison.ll

…

buildvector-insert-mask-size.ll

…

buildvector-nodes-dependency.ll

…

call-arg-reduced-by-minbitwidth.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

catchswitch.ll

…

consecutive-access.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

const-bool-logical-or-reduction.ll

…

crash_exceed_scheduling.ll

…

crash_extractelement_poison.ll

…

diamond_broadcast.ll

…

empty-struct.ll

…

ext-after-phi-node.ll

…

ext-int-reduced-not-operand.ll

…

extended-vectorized-gathered-inst.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

external-user-instruction-minbitwidth.ll

…

extract_with_non_const_index.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

extract-many-users-buildvector.ll

[SLP]Initial non-power-of-2 (but still whole register) for remaining nodes

2025-01-21 10:33:03 -05:00

extract-subvector-long-input.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

extract-vectorized-operand.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

extractelement-insertpoint.ll

…

extractlements-gathered-first-node.ll

…

extracts-with-undefs.ll

…

fix-113880.ll

[SLP] NFC. Replace TreeEntry::setOperandsInOrder with VLOperands. (#118949)

2024-12-11 10:09:23 +08:00

freeze-signedness-missed.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

full-overlap-non-schedulable.ll

[SLP]Add cost estimation for gather node reshuffling

2024-12-24 15:35:29 -05:00

gather_extract_from_vectorbuild.ll

…

gathered-consecutive-loads-different-types.ll

[SLP]Add cost estimation for gather node reshuffling

2024-12-24 15:35:29 -05:00

gep-with-extractelement-many-users.ll

…

icmp-altopcode-after-reordering.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

insert-crash-index.ll

…

insert-element-build-vector-const.ll

[SLP][NFC]Replace undef with constant zero in tests, NFC

2025-01-17 09:48:03 -08:00

insert-element-build-vector-inseltpoison.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

insert-element-build-vector.ll

[SLP][NFC]Replace undef with constant zero in tests, NFC

2025-01-17 09:48:03 -08:00

insert-element-multiple-uses.ll

…

insert-shuffle.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

insertelement-across-zero.ll

[SLP]Do not ignore undefs when trying to replace with "poisonous" shuffles

2024-10-24 07:47:23 -07:00

insertelement-postpone.ll

…

insertelement-uses-vectorized-index.ll

…

InstructionsState-is-invalid-2.ll

[SLP] Replace MainOp and AltOp in TreeEntry with InstructionsState. (#122443)

2025-01-18 10:23:20 +08:00

int_sideeffect.ll

…

int-bitcast-minbitwidth.ll

…

is_fpclass.ll

…

jumbled_store_crash.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

limit.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

logical-ops-poisonous-repeated.ll

[SLP]Followup fix for the poisonous logical op in reductions

2024-12-26 05:11:26 -08:00

materialize-vector-of-consts.ll

[AArch64][NFC] NFC for const vector as Instruction operand (#116790)

2024-11-21 10:23:05 +05:30

minbitwidth-multiuse-with-insertelement.ll

…

minbitwidth-node-with-multi-users.ll

…

minbitwidth-user-not-min.ll

…

multi-node-vectorized-insts.ll

…

multi-uses-with-deps-in-first.ll

…

one-element-vector.ll

…

operand-is-reduced-val.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

orig-btiwidth-les-projected.ll

…

partial-register-extract.ll

[SLP]Improved reduction cost/codegen

2025-02-14 11:03:33 -08:00

peek-through-shuffle.ll

…

phi-node-bitwidt-op-not.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

phi-not-ordered.ll

…

phi-undef-input.ll

…

postponed_gathers.ll

…

pr31599-inseltpoison.ll

…

pr31599.ll

…

rdar128092379.ll

…

reduced-gathered-vectorized.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

reduction_loads.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

reduction-gather-non-scheduled-extracts.ll

[SLP]Use TTI::getScalarizationOverhead where possible

2025-02-04 18:49:43 -05:00

reduction-min-select.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

reduction-modified-values.ll

[SLP]Improved reduction cost/codegen

2025-02-14 11:03:33 -08:00

reduction-whole-regs-loads.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

reorder-clustered-node.ll

[SLP]Add cost estimation for gather node reshuffling

2024-12-24 15:35:29 -05:00

reorder-node.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

reordered-top-scalars.ll

…

reordering-single-phi.ll

…

reschedule.ll

…

resized-alt-shuffle-after-minbw.ll

Revert "[SLP] Make getSameOpcode support interchangeable instructions. (#132887)"

2025-03-26 10:24:25 +02:00

return.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

reudction-or-non-poisoned.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

reused-buildvector-matching-vectorized-node.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

reused-scalar-in-multi-node.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

revec-fix-109835.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

revec-reduction-logical.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

revec-shufflevector.ll

[SLP][REVEC] Add ExtractSubvector cost for ExternalUses. (#132761)

2025-03-25 18:58:54 +08:00

revec.ll

[SLP][REVEC] reorderNodeWithReuses should not be called if all users of a TreeEntry are ShuffleVectorInst. (#118260)

2024-12-03 09:04:04 +08:00

root-trunc-extract-reuse.ll

[SLP]Make PHICompare comparator follow weak strict ordering requirement

2024-10-04 14:23:48 -04:00

same-scalar-in-same-phi-extract.ll

…

same-scalars-reordered-in-reduction.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

scalarazied-result.ll

[SLP]Use TTI::getScalarizationOverhead where possible

2025-02-04 18:49:43 -05:00

scalarization-overhead.ll

[SLP]Improved reduction cost/codegen

2025-02-14 11:03:33 -08:00

several_store_chains.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

shrink_after_reorder2.ll

[LLVM][IR] Use splat syntax when printing Constant[Data]Vector. (#112548)

2024-11-06 11:53:33 +00:00

shuffle-mask-resized.ll

Revert "[SLP] Make getSameOpcode support interchangeable instructions. (#132887)"

2025-03-26 10:24:25 +02:00

shuffle-multivector.ll

[SLP]Fix shuffling of entries of the different sizes

2024-11-21 13:08:27 -08:00

shufflebuilder-bug.ll

…

slp-deleted-inst.ll

[SLP] Check if instructions exist after vectorization (#120434)

2024-12-19 06:21:57 +08:00

slp-hr-with-reuse.ll

…

slp-max-phi-size.ll

…

slp-max-reg-size.ll

…

slp-non-pow-2-insertelement.ll

…

slp-umax-rdx-matcher-crash.ll

[llvm] Remove br i1 undef from some regression tests [NFC] (#118419)

2024-12-03 20:54:36 +00:00

slp-variable-insertelement.ll

…

splat-buildvector.ll

[SLP]Do not ignore undefs when trying to replace with "poisonous" shuffles

2024-10-24 07:47:23 -07:00

stores-non-ordered.ll

…

unknown-entries.ll

…

vectorizable-functions-inseltpoison.ll

…

vectorizable-functions.ll

…

vectorize-reorder-alt-shuffle.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

vectorize-reordered-list.ll

[SLP] Move more X86 tests to common directory (#111134)

2024-10-04 13:18:56 +01:00

zext-incoming-for-neg-icmp.ll

[SLP]Model reduction_add(ext(<n x i1>)) as ext(ctpop(bitcast <n x i1> to int n))

2024-11-22 06:50:25 -08:00