LLVM Bump to c27444ab4976dd9ff131212f87463f9945ab28d7

[jorickert]

AMD changes: Update lowering and tests for onnx->tosa conversions that are not upstream

Partial cherry-pick of f03b287

LLVM update 43d71ba (onnx#3086)

• update float types, tosa, other misc changes

• fix buildOnnxToTosaPaddingConstOp

• fix lit tests (wip)

• updte doc

• use stablehlo tagged version

• fixed more lit tests

• fix .clang-format

• fix lit (wip)

• revert .clang-format change

• fix lit tests

• fix formatting

• lit tests pass (except jni -- not tested)

• manually fix formatting; can't get clang-format to do it on any of my machines

• revert lit test changes unrelated to update

• update llvm and stablhlo shas, misc minor updates

• remove non-existent passes

• lit updates (wip)

• Bump Upsample to Opset 10 and change the opset versioning to allow to skip over opset versions if a newer, backwards compatible one exists. (Bump Upsample to Opset 10 and change the opset versioning to allow to skip over opset versions if a newer, backwards compatible one exists. onnx/onnx-mlir#3065)

• Bump Upsample to Opset 10

This is a non-functional change, the only difference is that Upsample was marked as deprecated with Opset 10

• Use a map of the available opset versions in onnx to select the node opset to use.

Introduces a new built-time generated map that contains all versions of an operation as defined by onnx. To determine the opset version for a node/op:

1. Determine the latest valid opset version. This is the newest version in this opset-version-map that is older or equal to the current graph opset.

2. Select the newest version from the versions supported by onnx-mlir that is equal or newer to the latest valid opset version. This allows it to skip over opset versions, that have a newer backwards compatible version. Example:
   Versions in onnx and supported by onnx-mlir: [3, 5].
   Graph opset version to node version: 3 -> 3, 4 -> 3, 5 -> 5

   Versions in onnx: [7, 9, 10]. Version 10 is backwards compatible to version 9.
   Version supported by onnx-mlir: [7, 10].
   Graph opset version to node version: 7 -> 7, 8 -> 7, 9 -> 10, 10 -> 10

---

• Improve scripts (Improve scripts onnx/onnx-mlir#3089)

• Bump various ops to opset 21, adding int4/uint4 and 8 bit float support. (Bump various ops to opset 21, adding int4/uint4 and 8 bit float support. onnx/onnx-mlir#3064)

• Add support for TensorProto::UINT4/INT4

• Upgrade onnx.Cast to opset 21

• Bump various ops to opset 21.

These are all backwards compatibel version bumps, only adding support for int/uint4.

Bumped ops:
Flatten
Identity
If
Loop
Pad
Reshape
Scan
Shape
Size
Squeeze
Transpose
Unsqueeze

---

• Added minimal support to do some timing of OM Runtime functionality (Added minimal support to do some timing of OM Runtime functionality onnx/onnx-mlir#3095)

• adding __errno_location call for mvs (Including __errno_location call for MVS onnx/onnx-mlir#3099)

• Rewriting pattern to remove WhereOp and EqualOp. (Rewriting pattern to remove WhereOp and EqualOp.  onnx/onnx-mlir#3094)

Remove ONNXWhereOp and ONNXEqualOp into newly created ConcatOp.

---

• Enable NNPA saturation by default and change the option to --nnpa-disable-saturation (Enable NNPA saturation by default and change the option to --nnpa-disable-saturation onnx/onnx-mlir#3101)

• Enable NNPA saturation by default and change the option to --nnpa-disable-saturation

---

• removing weak attribute of errorno (removing weak attribute of errorno onnx/onnx-mlir#3103)

• Fix the custom build link for docs/Docker.md (Fix the custom build link for docs/Docker.md onnx/onnx-mlir#3104)

• Python driver for torch model (Python driver for torch model onnx/onnx-mlir#3093)

• implementation

• format

• test

• py format

• torch.compile

• refine

• add debug

• respond

• response

• format

---

• implement (Improve the c++ driver build-run-onnx-lib.sh onnx/onnx-mlir#3108)

• Followups for torch model driver (Followups for torch model driver onnx/onnx-mlir#3106)

• simplify

• complete

• fix

• fix

---

• Fix an error in ZHighConstantPropagation for QuantizedStick ([NNPA] Fix an error in ZHighConstantPropagation for QuantizedStick onnx/onnx-mlir#3112)

• Add z17 for -march (Add z17 for -march onnx/onnx-mlir#3113)

• done

• convert

• fix

• format

---

• Decompose Hardswish into simpler ONNX ops (Decompose Hardswish into simpler ONNX ops onnx/onnx-mlir#3107)

• Decompose and lower Hardswish

• Providing the decomposition as compile time option with krnl dialect lowering as default

---

• Reorder relu to maxpool optimization pass in ONNX dialect (Reorder relu to maxpool optimization pass in ONNX dialect onnx/onnx-mlir#3109)

• Reorder Relu and maxpool optimization

• Swap Relu and maxpool only when Relu is not a consumer of conv

---

• Move onnx.Constant before the root op when fusing onnx ops (Fix "operand #0 does not dominate this use" when fusing onnx.ConstantOp onnx/onnx-mlir#3119)

• Support QLinearMatMul on CPU (Support QLinearMatMul on CPU onnx/onnx-mlir#3117)

• Support QLinearMatMul on CPU

---

• Update black-format-check.yml (Update black-format-check.yml onnx/onnx-mlir#3118)

• Merge nested concat Ops optimization pass in ONNX dialect (Merge nested concat Ops optimization pass in ONNX dialect onnx/onnx-mlir#3111)

• Merging nested concat ops

---

• Enhance shape inference for ONNX Reshape (Enhance shape inference for ONNX Reshape onnx/onnx-mlir#3122)

• Add a special case in shape inference for reshape

---

• update zdnn1.1.2 (update zdnn-v1.1.2 onnx/onnx-mlir#3130)

• Updating supported ops on NNPA md for z17. (Updating supported ops on NNPA md for z17.  onnx/onnx-mlir#3120)

• starting to update new z17 NNPA ops

---

• fix CVE-2025-32434 (fix CVE-2025-32434 onnx/onnx-mlir#3135)

• Fuse consecutive clips pattern (Fuse consecutive clips pattern onnx/onnx-mlir#3132)

• Fuse consecutive clips pattern

---

• Replace deprecated applyPatternsAndFoldGreedily with applyPatternsGreedily. This functions also folds by default, so it is an NFC

• Fix clang-format

• Replace bufferization::createOwnershipBasedBufferDeallocationPass with mlir::createConvertBufferizationToMemRefPass

• Update onnx-to-tosa reshape lit test

• Move gemm_to_fc tests to gemm_to_matmul

• Change tosaBuilder::mul function signature to make clear that the shift is an int8

• Disable buffer_loop_hoisting test as it gets completly optimized away

• Guard against dynamic dim in result

• Use resize operaton input and output type to calculate the border, instead of using the calculated numerator/denominator

• Guard against linear interpolation of integer types

• Add test for disallowed onnx.Resize on its with linear interpolation to tosa

• Add 'Pure' annotation to some krnl ops and recreate documentation

• Build stablehlo with static libs

• Disable memref.prefetch since it does not work with the new bufferization

• Conv add const where the constant is a scalar (Conv add const where the constant is a scalar onnx/onnx-mlir#3145)

• added support for Celu op (added support for Celu op onnx/onnx-mlir#3139)

• Fix some warnings related to stickification for NNPA (Fix some warnings related to stickification for NNPA onnx/onnx-mlir#3147)

• Removing duplicate file (Removing duplicate file docs/SupportedONNXOps-NNPA-supplement.md onnx/onnx-mlir#3146)

• migrated instance/group normalization from decompose to canonicalize (Instance and Group norm needs shape inference onnx/onnx-mlir#3148)

• Fusion of Matmul add covering the stacked/unstacked/bcast1/bcast23 patterns (Fusion of Matmul add covering the stacked/unstacked/bcast1/bcast23 patterns onnx/onnx-mlir#3140)

• Support --march=native (Support --march=native onnx/onnx-mlir#3134)

• changes

• format

• linkage

• lib

---

• fix another error on s390x

• lower Ub to LLVM since vector.shape_cast is lowered to UB

---

[jorickert]

Targeting Xilinx/llvm-project#584