PR #19886 [NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent

coldav requested a review 260 days ago

coldav changed the title ~~DRAFT: [NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent~~ DRAFT: NOT FOR REVIEW [NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent 260 days ago

coldav force pushed from e6122d84 to 5f8574c8 260 days ago

coldav force pushed from 5f8574c8 to 729e7b51 259 days ago

coldav requested a review 259 days ago

coldav force pushed from 729e7b51 to 63d7eb88 257 days ago

coldav force pushed from 63d7eb88 to 0b99891d 257 days ago

coldav force pushed from 0b99891d to 087ecb6e 253 days ago

coldav force pushed from 087ecb6e to dd1785fe 253 days ago

coldav force pushed from dd1785fe to a473b5fb 253 days ago

coldav commented on 2025-09-03

uwedolinsky commented on 2025-09-03

coldav force pushed from a473b5fb to 312d82bc 252 days ago

coldav changed the title ~~DRAFT: NOT FOR REVIEW [NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent~~ [NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent 252 days ago

coldav assigned

steffenlarsen 252 days ago

coldav requested a review from

steffenlarsen 252 days ago

coldav unassigned

steffenlarsen 252 days ago

coldav requested a review from

aelovikov-intel 252 days ago

coldav force pushed from 312d82bc to 4cd30134 252 days ago

coldav force pushed from 4cd30134 to 86b21b55 252 days ago

sarnex commented on 2025-09-03

[vecz] Move to compiler module

8874bb2a

[multi_llvm] Move to compiler module

4e8a584f

[lit] Have lit configure tools in-house

41376e61

[compiler] Replace all uses of llvm::Optional

6e408436

[multi_llvm] Add Triple.h wrapper

5eabd4d1

Run clang-format-9 on everything

f01fae7c

[vecz] Update vsetvli intrinsic for LLVM 17

f64652a6

[compiler] Adapt to removal of llvm.dbg.addr in LLVM 17+

148a3867

[compiler] Drop last traces of LLVM 14 support

1c957a04

[multi_llvm] Provide implicit conversion from Optional to std::optional

7a0e6550

[compiler] Remove need for passing Module to NameMangler

0dafdeb0

Subgroup broadcast of uniform value becomes a NOP (#59)

29945d21

[vecz] Update some lit tests to account for LLVM 17

84807258

[compiler] Enable mux subgroup/workgroup builtins

5a543943

[compiler] Provide methods for common group operation checks

33501902

[compiler] Add more helper methods for mux group operations

3d006339

[vecz] Migrate vecz lit tests to mux builtins

106d3b93

[compiler] Update to build with LLVM 17

91ef73cf

[compiler] Fix a couple of lit tests with LLVM 17

b3d22762

Amy/refactor work group collectives 2 (#98)

e5904ee6

[vecz] Do not vectorize llvm.debug.value intrinsics

10b9d78f

[compiler] Delete dead CLBuiltinInfo code

c93b31fe

[vecz] Remove checks for non-i8 memcpy/memsets

73bd8bd7

[vecz] Remove dead/discarded function demangle

427b2ef6

[vecz] Switch from AssertingVH to PoisoningVH

0ed8ebd6

[multi_llvm] Remove opaque pointer helpers

fdbb6b1a

[compiler] Remove all but one use of multi_llvm::Optional

58585729

[vecz] Vectorize sub-group local ID on top of mux sub-groups

c0e10f56

[vecz] Vectorize sub-group reductions on top of mux sub-groups

5c08537b

[vecz] Vectorize sub-group broadcasts on top of mux sub-groups

f54110b7

[vecz] Vectorize sub-group scans on top of mux sub-groups

b839b764

[vecz] Packetize the sub-group size on top of mux sub-groups

cc242ab6

[compiler] Add mux sub-group shuffle builtins

1005ff9e

[compiler] Vectorize to any required sub-group size

0c377894

[compiler] Preserve 'entry point' on vectorized kernels

5c547665

[vecz] Vectorize to the reqd sub-group size under 'auto'

a6b43d58

[riscv] Vectorize sub-group functions to a device size

3bae73be

[vecz] Fix alignment of load operations

449c0fe5

[vecz] Fix invalid dangling AssumptionCaches

1887b3f3

[compiler] Add work-group scan support in vecz/work-item-loops

753793ed

[vecz] Use correct alignment for memcpy source

a21a7afa

[vecz] Don't mask work-group collective operations

69289946

[vecz] Packetize sub-group shuffles with uniform indices

0d38f0b3

[vecz] Packetize sub-group shuffle_xor builtins

14225d93

[multi_llvm] Remove ArrayRef helpers

3871885e

[multi_llvm] Remove TypeSize helpers

c22c69f1

[multi_llvm] Remove insertInto helpers

6b4ec570

[multi_llvm] Remove InlineFunction helper

e4dc6c4f

[multi_llvm] Remove addVectorizableFunctionsFromVecLib helper

b88164f0

[multi_llvm] Clean up unused includes

41dad0f9

[multi_llvm] Remove Optional helper

f1132928

Remove old LLVM 15 code

7009d6a6

[multi_llvm] Remove getStructTypeByName helper

48113797

[multi_llvm] Remove getDILocation helper

85623271

[multi_llvm] Move some vectorization helpers into the vectorizer

3da0921b

[multi_llvm] Remove creation_apis_helper.h

d47f463d

[multi_llvm] Move createBinOpForRecurKind helper

a8ae40fe

[multi_llvm] Clean up assorted 'inline' versioning

23584d52

[vecz] Support vector-predicated reductions natively

6b5b08de

[vecz] Fix vectorization of sub-group broadcasts

e2eddfb4

[vecz] Packetize sub-group shuffle_(up|down) builtins

437baf78

[vecz] Fix SquashSmallVectorsPass on larger vectors

643bc629

[LLVM] Multi_llvm support for current tip changes

45a665b5

[vecz] Update tests for LLVM tip and fix alignment optimization

319e7dc5

[vecz] Suppress cert-err33-c on error-handling fprintf

b8678688

Move LLVM 18 fixups.

f9883739

[vecz] Provide more context when CFG conversion fails

6b193d6e

[vecz] Run SROA as part of the vecz pipeline

74d0cf51

[vecz] Simplify pass construction process

b6a7155e

More LLVM 18 fixups.

9e232b93

[vecz] Suppress cert-err33-c on error-handling fprintf

b8678688

Move LLVM 18 fixups.

f9883739

[vecz] Simplify pass construction process

b6a7155e

[vecz] Add support for masking cmpxchg instructions

1a9491e8

[NFC] Change startswith/endswith to starts_with/ends_with.

783cd9ba

[vecz] Fix missing CHECKs in LIT tests

bcae693b

[vecz] Fix missing CHECK in LIT test

eb3737ff

[vecz] Enable XFAIL tests

153037f9

[vecz] Merge and fix call instantiation LIT tests

f3764ec1

[vecz] Add nounwind/norecurse attributes to internal vecz builtins

c46bae64

[vecz] Ensure inactive lanes don't contribute to branch conditions

2f7e00c7

[compiler] Ensure createLoop always creates a loop

3bddeaa6

[NFC] Merge OpaquePointers with main tests.

73fd0060

[NFC] Pick a style for const.

39f8f465

[tests] Update for LLVM 18.

f993b320

[NFC] Upgrade to clang-format-17.

8fe1f818

[compiler] Fix unchecked std::optional accesses

0f682622

[NFC] Add const.

345e8b75

[NFC] Address clang-tidy-17 readability warnings.

c39749dc

[compiler] Improve analysis of 'true uniform' values

5b304da0

Update findDbgDeclare for LLVM 18

3798c49e

Enable extra warnings, adjust code.

0549866b

Upgrade to clang-tidy-17.

0ce030eb

Update tests for LLVM 19.

ebb0a0d2

[vecz] Use LLVM helper for testing for powers of 2

fb17e5f7

[vecz] Add a pass to print StrideAnalysis results

e0e0fa94

[vecz][NFC] Fix a couple of doxygen issues in OffsetInfo

e2ecdab2

[vecz] Fix dropped value analysis in OffsetInfo

dfe08c65

[NFC] buildAfter: return IRBuilder<>.

c421b7ce

Keep debug information.

9e84524b

Update for LLVM 19.

42b425a8

Update for LLVM 19.

715f8dc9

Update for LLVM 19.

8e00b3f0

LLVM 19 update: trunc to i1.

61d21131

Accept nuw/nsw flags on trunc.

e278c605

[LLVM 19] Adjust tests for llvm.ct* intrinsics.

7580493f

Update lit check for vector.splice

17d65f8d

Moved compiler-pipeline out of utils so it can be included directly.

4eeda203

Use LLVM passes to output modules

cca9b239

Remove all code specific to LLVM 16

2dc22836

Fix replace_local_module_scope_variables mempcy bug

80d71ac1

Avoid StringRef.equals()

d1f526d3

Update tests for LLVM 19.

8ebf0b78

[LLVM 19] Drop debug info format conversions

f4aaca57

Fix build with LLVM 19 again.

c5841ed6

Fix build with LLVM 19 again.

bad94d18

[LLVM 20] Adjust irreducible_loop test

2007e534

[LLVM 20] Adjust for non-experimental builtins

8f8bef4e

[LLVM 20] Adjust packetize_mask_varying test

a93c2cca

[vecz] Avoid scalarization using large ints

2e6fae47

Fix compiler breakage on llvm 20 for createSimpleTargetReduction.

86798e64

[LLVM20] Handle USubCond, USubSat.

d3668079

[vecz] Clear getelementptr flags.

41658d9d

Upgrade clang-format & clang-tidy.

18194f12

Remove LLVM 17 support.

293dbccf

Fix build with LLVM 20.

0cc33f72

Add support for ConstantAray in replace-module-scope pass.

1d452d5d

[NFC] Avoid deprecated Type::getPointerTo.

52f70b89

Fix vecz lit test for splat issues after llvm 20 changes to output

626b71d3

[LLVM 20] Update tests.

91d164ef

LLVM 20: Update lit tests.

d899e8f4

[LLVM 20] Allow getelementptr nuw flag.

3524c8cf

[compiler] Handle skipped vectorized functions.

a9bb9655

[compiler] Avoid out of bounds access.

acfd588d

[mux] Remove degenerate subgroup support.

9e7b4794

[compiler] Do not mix kernels with different sub-group sizes.

da04975e

[vecz] Handle missing gather/scatter functions.

94c6f949

Update for LLVM 20.

15fdf9fd

[vecz] Handle vectors of size 1.

e57031f4

LLVM 20: Update for getFirstNonPHIOrDbg.

40b912c7

LLVM 21: Update for Attribute::NoCapture removal.

a5ca431e

Update for LLVM 20, 21.

04d65a4f

Update for LLVM 21.

99b1f039

Address clang & clang-tidy warnings.

d9efef43

Address more clang & clang-tidy warnings.

9555a737

[vecz] Fix packetization of PHIs of literal structs.

56538f37

[LLVM 21] Take address space into account for legality.

79c9b402

Remove all conditional LLVM code for LLVM 18

a52b7d9b

Do not vectorize NoInline kernels.

e35e1993

[LLVM 21] Avoid calling spir_kernel functions.

a337576a

Fix up Instrinsic::getAttributes to work across LLVM versions

1168a189

Update to address PointerType deprecation.

0423986f

[vecz] Track barriers by ID.

3020f678

[builtins] Improve robustness.

4fe6c8ef

[NFC] Fix comments.

4869bb45

[LLVM 21] Update for CreateTargetInfo API change.

61d55e80

[LLVM 21] Add FMaximum, FMinimum.

6099cc0f

[NFC] Address LLVM 21's BinOp additions in a better way.

c5ce5ce6

Removal of Vulkan API support

f2e2aa41

[NFC] Remove Abacus extras.

71d18ce4

[LLVM 21] Avoid computeKnownBits.

1f16d4d7

[LLVM 21] Remove VectorizationFactor, use CreateElementCount.

76d0c25d

[NFC] Remove code for old debug info format.

dedf80ac

[LLVM 21] Allow nuw in more tests.

75f885bd

Replace getNextNonDebugInstruction() with getNextNode()

56b5011a

[LLVM TIP] Fix use of ConvertDebugDeclareToDebugValue

d93bf5e1

[NFC] Remove more old debug info handling.

f2e46fef

Fix check for GEPs with scalar indices

5cddabcd

[NFC] Remove some old LLVM support from tests.

94a47191

Remove CA references under compiler_pipeline and vecz

91e71048

[NFC] Remove some old LLVM support from tests.

94a47191

Changed copyright license to point to correct license file path

b7087e55

Use poison rather than undef.

e1663a0f

[vecz] Avoid mishandling poison.

cfc296e2

More poison/undef updates.

fc3e3f27

[NFC] Remove unused %pp-llvm-ver.

a951cdee

Switch to LLVM versions 20/21 as supported versions.

3720900b

Remove outdated test.

ac9ee79f

[NFC] Remove leftover older LLVM handling.

93336499

Re-enable RISC-V lit tests.

d3a5a699

[SYCL][NATIVE_CPU] Resolve formatting issues on compiler_pipeline

25f934d9

[NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline

477f0cd0

coldav force pushed from 86b21b55 to 477f0cd0 250 days ago

aelovikov-intel dismissed these changes on 2025-09-05

aelovikov-intel removed review request from

steffenlarsen 250 days ago

aelovikov-intel requested a review from

steffenlarsen 250 days ago

aelovikov-intel dismissed their stale review 250 days ago

This needs to be merged and not squashed. Removing my approval to prevent accidental squash.

uwedolinsky approved these changes on 2025-09-05

aelovikov-intel approved these changes on 2025-09-05

aelovikov-intel merged bbae036e into sycl 250 days ago

llvm
[NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent
#19886

Merged

[NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent #19886

llvm [NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent #19886 Merged

[NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent #19886

llvm
[NATIVE_CPU][SYCL] Switch to using native_cpu compiler pipeline inline instead of OCK fetchContent
#19886

Merged