Florian Hahn 2c8836c899

[LV] Don't consider predicated insts as invariant unconditionally in CM. ...

Predicated instructions cannot hoisted trivially, so don't treat them as
uniform value in the cost model.

This fixes a difference between legacy and VPlan-based cost model.

Fixes https://github.com/llvm/llvm-project/issues/110295.

2024-09-29 20:31:24 +01:00

..

already-vectorized.ll

[LV] Disable runtime unrolling for vectorized loops.

2023-01-06 10:56:17 +00:00

amdlibm-calls-finite.ll

Adding support of AMDLIBM vector library (#78560)

2024-02-15 12:13:07 +05:30

amdlibm-calls.ll

[Analysis] getIntrinsicForCallSite - add vectorization support for acos/asin/atan and cosh/sinh/tanh libcalls (#106844)

2024-09-03 10:05:56 +01:00

avx1.ll

…

avx512.ll

BlockFrequencyInfoImpl: Avoid big numbers, increase precision for small spreads

2023-10-24 20:27:39 -07:00

consecutive-ptr-cg-bug.ll

…

consecutive-ptr-uniforms.ll

[LV] Check isPredInst instead of isScalarWithPred in uniform analysis. (#98892)

2024-07-19 12:02:25 +01:00

constant-fold.ll

[LV] Use IRBuilder to create and optimize middle-block compare.

2023-08-29 11:42:18 +01:00

constant-vector-operand.ll

…

conversion-cost.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

cost-constant-known-via-scev.ll

[LV] Use SCEV to analyze second operand for cost query.

2024-08-29 12:08:27 +01:00

cost-model-assert.ll

…

cost-model.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

divs-with-tail-folding.ll

[VPlan] Generalize VPValue::isDefinedOutsideLoopRegions.

2024-09-20 15:34:00 +01:00

drop-poison-generating-flags.ll

[VPlan] Remove unused first mask op from VPBlendRecipe. (#87770)

2024-04-09 11:14:05 +01:00

ephemeral-recipes.ll

[VPlan] Don't add live-outs for IV phis.

2024-07-14 20:49:03 +01:00

epilog-vectorization-inductions.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

fixed-order-recurrence.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

float-induction-x86.ll

[LoopUnroll] Use poison instead of undef for preheader value

2024-06-25 12:09:58 +02:00

fneg-cost.ll

…

fp32_to_uint32-cost-model.ll

…

fp64_to_uint32-cost-model.ll

…

fp80-widest-type.ll

…

fp_to_sint8-cost-model.ll

…

funclet.ll

[LV] Check for vector instruction in main vector loop.

2023-07-03 14:16:47 +01:00

gather_scatter.ll

[VPlan] Explicitly handle scalar pointer inductions. (#83068)

2024-03-26 16:01:57 +01:00

gather-cost.ll

…

gather-vs-interleave.ll

…

gcc-examples.ll

…

gep-use-outside-loop.ll

[LV] Check pointer user are in loop when checking for uniform pointers.

2024-08-13 09:23:44 +01:00

i1-reg-usage.ll

…

illegal-parallel-loop-uniform-write.ll

[LV] Optimize trip count SCEV.

2023-04-12 16:17:58 -07:00

imprecise-through-phis.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

induction-costs.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

int128_no_gather.ll

…

interleave_short_tc.ll

[LV] Remove loop trip count threshold for deciding whether to interleave a loop (#67725)

2024-02-05 17:23:58 -08:00

interleave-cost.ll

[CostModel][X86] Improve add/sub/mul overflow intrinsic costs

2024-07-25 16:01:05 +01:00

interleave-opaque-pointers.ll

[VPlan] Explicitly handle scalar pointer inductions. (#83068)

2024-03-26 16:01:57 +01:00

interleaved-accesses-hoist-load-across-store.ll

[VPlan] Preserve original induction order when creating scalar steps.

2024-01-31 13:31:28 +00:00

interleaved-accesses-large-gap.ll

…

interleaved-accesses-sink-store-across-load.ll

[LV] Complete load groups and release store groups. Try 2.

2023-08-08 18:10:23 -04:00

interleaved-accesses-use-after-free.ll

[LV] Move interleaved test to X86 directory

2023-08-09 16:03:33 -04:00

interleaved-accesses-waw-dependency.ll

…

interleaving.ll

[VPlan] Implement unrolling as VPlan-to-VPlan transform. (#95842)

2024-09-21 19:47:37 +01:00

intrinsiccost.ll

[InstCombine] Infer nsw/nuw for trunc (#87910)

2024-04-11 19:10:53 +08:00

invariant-load-gather.ll

[VPlan] Add initial loop-invariant code motion transform. (#107894)

2024-09-20 11:22:03 +01:00

invariant-store-vectorization.ll

[InstCombine] Remove some of the complexity-based canonicalization (#91185)

2024-08-21 12:02:54 +02:00

iv-live-outs.ll

[VPlan] Unify mayWriteToMemory and mayHaveSideEffects logic for VPInst.

2024-09-26 19:16:43 +01:00

libm-vector-calls-finite.ll

…

libm-vector-calls-VF2-VF8.ll

[x86] Add tan intrinsic part 4 (#90503)

2024-06-05 15:01:33 -04:00

libm-vector-calls.ll

…

limit-vf-by-tripcount.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

lit.local.cfg

[NFC][Py Reformat] Reformat lit.local.cfg python files in llvm

2023-05-17 17:03:15 +02:00

load-deref-pred.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

masked_load_store.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

masked-store-cost.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

max-mstore.ll

…

metadata-enable.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

min-trip-count-switch.ll

…

mul_slm_16bit.ll

…

multi-exit-cost.ll

[VPlan] Skip users outside loop in check for exit pre-compute candidates

2024-07-11 22:04:39 +01:00

no_fpmath_with_hotness.ll

BlockFrequencyInfoImpl: Avoid big numbers, increase precision for small spreads

2023-10-24 20:27:39 -07:00

no_fpmath.ll

…

no-vector.ll

…

nontemporal.ll

…

optsize.ll

[LV] Add additional test coverage for cost modeling.

2024-06-26 10:18:01 +01:00

outer_loop_test1_no_explicit_vect_width.ll

[IRBuilder] Use canonical i64 type for insertelement index used by vector splats.

2023-01-11 14:08:06 +00:00

parallel-loops-after-reg2mem.ll

…

parallel-loops.ll

[InstCombine] Set disjoint flag when turning Add into Or. (#72702)

2023-11-27 12:54:11 -08:00

pointer-runtime-checks-unprofitable.ll

[LV] Add uses of result to pointer-runtime-checks-unprofitable.ll test.

2024-07-15 09:59:46 +01:00

powof2div.ll

…

pr23997.ll

[InstCombine] Remove some of the complexity-based canonicalization (#91185)

2024-08-21 12:02:54 +02:00

pr34438.ll

[LV] Use IRBuilder to create and optimize middle-block compare.

2023-08-29 11:42:18 +01:00

pr35432.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

pr36524.ll

[VPlan] Preserve original induction order when creating scalar steps.

2024-01-31 13:31:28 +00:00

pr39160.ll

[VPlan] Don't create live-outs for induction increments.

2024-07-25 21:34:55 +01:00

pr42674.ll

[ConstantFolding] Canonicalize constexpr GEPs to i8 (#89872)

2024-05-20 11:47:30 +02:00

pr47437.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

pr48340.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

pr51366-sunk-instruction-used-outside-of-loop.ll

[LV] Use IRBuilder to create and optimize middle-block compare.

2023-08-29 11:42:18 +01:00

pr54413-select-interleave-count-loop-with-cost-zero.ll

…

pr54634.ll

[CostModel][X86] Improve add/sub/mul overflow intrinsic costs

2024-07-25 16:01:05 +01:00

pr55096-scalarize-add.ll

LoopVectorize: guard appending InstsToScalarize; fix bug (#88720)

2024-04-18 10:03:07 +01:00

pr56319-vector-exit-cond-optimization-epilogue-vectorization.ll

[LV] Consider if scalar epilogue is required in getMaximizedVFForTarget.

2023-07-06 13:35:35 +01:00

pr72969.ll

[VPlan] Introduce ResumePhi VPInstruction, use to create phi for FOR. (#94760)

2024-07-11 16:08:04 +01:00

pr81872.ll

Recommit "[VPlan] Replace disjoint or with add instead of dropping disjoint. (#83821)"

2024-03-27 19:11:18 +00:00

pr109581-unused-blend.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

predicate-switch.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

predicated-instruction-cost.ll

[LV] Don't consider predicated insts as invariant unconditionally in CM.

2024-09-29 20:31:24 +01:00

propagate-metadata.ll

[LoopVectorize] Convert some tests to opaque pointers (NFC)

2023-01-04 17:25:42 +01:00

ptr-indvar-crash.ll

[LoopVectorize] Convert some tests to opaque pointers (NFC)

2023-01-04 17:25:42 +01:00

rauw-bug.ll

…

reduction-crash.ll

…

reduction-fastmath.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

reduction-small-size.ll

…

redundant-vf2-cost.ll

…

reg-usage-debug.ll

[LV] Reland "Update logic for calculating register usage due to invariants"

2023-02-28 17:32:39 +05:30

reg-usage.ll

…

register-assumption.ll

…

runtime-limit.ll

…

scatter_crash.ll

Reland "[X86] Remove knl/knm specific ISAs supports (#92883)" (#93136)

2024-05-24 13:46:34 +08:00

slm-no-vectorize.ll

[LoopVectorize][X86] Regenerate slm-no-vectorize.ll

2023-06-13 14:15:37 +01:00

small-size.ll

[VPlan] Explicitly handle scalar pointer inductions. (#83068)

2024-03-26 16:01:57 +01:00

strided_load_cost.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

struct-store.ll

…

svml-calls-finite.ll

…

svml-calls.ll

[x86] Add tan intrinsic part 4 (#90503)

2024-06-05 15:01:33 -04:00

tail_folding_and_assume_safety.ll

…

tail_loop_folding.ll

[VPlan] Model branch cond to enter scalar epilogue in VPlan. (#92651)

2024-07-05 10:08:42 +01:00

tripcount.ll

…

uint64_to_fp64-cost-model.ll

…

uniform_load.ll

…

uniform_mem_op.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

uniform-phi.ll

…

uniformshift.ll

…

unroll_selection.ll

…

unroll-pm.ll

…

unroll-small-loops.ll

[LV] Remove loop trip count threshold for deciding whether to interleave a loop (#67725)

2024-02-05 17:23:58 -08:00

veclib-calls.ll

[X86][LoopVectorize] Add support for arc and hyperbolic trig functions (#99383)

2024-07-28 20:57:43 -04:00

vect.omp.force.ll

…

vect.omp.force.small-tc.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

vector_max_bandwidth.ll

…

vector_ptr_load_store.ll

…

vector-scalar-select-cost.ll

…

vectorization-remarks-loopid-dbg.ll

[LV] Remove loop trip count threshold for deciding whether to interleave a loop (#67725)

2024-02-05 17:23:58 -08:00

vectorization-remarks-missed.ll

[LoopVectorize] Teach LoopVectorizationLegality about more early exits (#107004)

2024-09-19 09:41:25 +01:00

vectorization-remarks-profitable.ll

…

vectorization-remarks.ll

[LV] Remove loop trip count threshold for deciding whether to interleave a loop (#67725)

2024-02-05 17:23:58 -08:00

vectorize-force-tail-with-evl.ll

[VPlan] Run DCE after unrolling.

2024-09-22 22:08:46 +01:00

vectorize-interleaved-accesses-gap.ll

[IRBuilder] Use canonical i64 type for insertelement index used by vector splats.

2023-01-11 14:08:06 +00:00

vectorize-only-for-real.ll

…

vplan-native-inner-loop-only.ll

[llvm][LV] Move new test into X86 subfolder

2024-01-25 17:04:34 +00:00

vplan-vp-intrinsics.ll

[LV, VP]VP intrinsics support for the Loop Vectorizer + adding new tail-folding mode using EVL. (#76172)

2024-04-04 18:30:17 -04:00

widened-value-used-as-scalar-and-first-lane.ll

[VPlan] Unify mayWriteToMemory and mayHaveSideEffects logic for VPInst.

2024-09-26 19:16:43 +01:00

x86_fp80-interleaved-access.ll

…

x86_fp80-vector-store.ll

[InstCombine] Set disjoint flag when turning Add into Or. (#72702)

2023-11-27 12:54:11 -08:00

x86-interleaved-accesses-masked-group.ll

[LV] Create block in mask up-front if needed. (#76635)

2024-01-09 10:50:08 +00:00

x86-interleaved-store-accesses-with-gaps.ll

[InstCombine] Canonicalize constant GEPs to i8 source element type (#68882)

2024-01-24 15:25:29 +01:00

x86-pr39099.ll

…

x86-predication.ll

[VPlan] Remove VPPredInstPHIRecipes without users after region merging.

2024-07-20 13:21:32 +01:00