Update Eigen to commit:b1f8b0c3b8db9c3711349144acf951226700ddb4

CHANGELOG
=========
b1f8b0c3b - Core: Support flexible indexers in `VectorwiseOp::replicate`
8f1a18ff7 - Core: Gate product_packet_cascade on MightVectorize and mark assignPacket device-callable
630f81dbf - RVV: Enable generic packet sinh, cosh, asinh, acosh, atan, atanh, log10, log1p, expm1, pow, cbrt, rsqrt and erfc
5b5b44a71 - LU: Unroll PartialPivLU for fixed sizes up to 12
32a4f127f - Eigenvalues: Defer far-column reflector updates in the Francis QR step
ab8a55665 - Householder: Apply two- and three-element reflectors from the right in one pass
97898f371 - AltiVec: Fix out-of-bounds tail accesses in VSX and MMA GEMM kernels
0013bcb18 - Geometry: Preserve single-rounding pivot and trace order in Quaternion(Matrix3)
07d4528c8 - MSVC: Fix the ILP64 BLAS DLL and the stencil test
2774a036e - StructuredMatrices: Identity and nested factors for KroneckerOperator
869eb9850 - Core: Pin the TRSM padded tail's view of the right-hand side to column-major
b6225b7dc - Core: Finish partial GEMV packets in one pass with masked segments
c541f8f43 - Core: Support flexible indexers in `DenseBase::replicate`
7f03d8469 - NEON: Classify floats with absolute comparisons and reduce allFinite() through masks
bc99fddc6 - Cholesky: Unblocked LLT up to 48 columns and 16-column minimum blocks
e3755c9c7 - Core: Scalar tails for assignments bounded to a few packets
ff16ae80a - Added specific value tests (x2) and random number generated tests for autodiff...
03d21656a - StructuredMatrices: Reuse Cauchy reciprocal columns across right-hand sides
477435767 - SME: Fold a real alpha into the complex tile-pair store
c8725af72 - StructuredMatrices: Drop extreme-scale hardening from DPR1EigenSolver
3346f1a2b - StructuredMatrices: Drop extreme-scale hardening from DiagonalPlusLowRank
a29e46e2e - StructuredMatrices: Solve Kronecker least squares by per-factor COD
036a81d7c - NEON: Clean up complex arithmetic and fix GCC ARM32 float compares
1c7d00c27 - Core: Fix out-of-bounds B-panel lookahead loads in AVX512 GEMM kernel
4be755d45 - Core: Decouple the fixed-size product threshold and retune it for SME
dee1af8bb - Householder: Restore AutoDiffScalar support in QR and JacobiSVD
f7a2dd7a9 - FindCoeff: Require HasCmp for the packet path of minCoeff/maxCoeff
a82afb7c7 - Geometry: Fix AngleAxis::fromRotationMatrix accuracy and axis sign near pi
7194105cc - StructuredMatrices: Drop extreme-scale hardening from the FFT operators
d11bb9121 - NEON: Optimize stride-two float gathers
7ac31bc91 - EulerAngles: Test the constructor from a data pointer
038961d50 - NumericalDiff: Implement Fornberg finite difference stencil algorithm
b4befd43f - SVD: add PreconditionSquareMatrix option.
982e195f9 - StructuredMatrices: Batch KroneckerOperator products and solves, drop extreme-scale hardening
221aa496a - StructuredMatrices: Drop extreme-scale hardening from Vandermonde and Cauchy
963fffeb0 - NVHPC: Fix the nvc++ build
19090299a - Core: Use k-masked packet segments on AVX-512 and predicated segments on SVE
062290ffb - Tests: Guard pabs and pmadd/pmsub/pnmadd with packet traits in packetmath
210e0a609 - Core: Return the cached outer stride from direct-access Block
3cf994918 - CI: Build the AVX512-FP16 tests with clang-19
bbefaffa6 - Tests: Compile split test parts with only the functions they call
8c653283e - SME: Keep pack_direct in streaming mode under GCC
c3123dc95 - BDCSVD: Cover the bidiagonal entry point in the switch-size test
ca0565a17 - CI: Pass the SME -march through CMAKE_CXX_FLAGS
e4505f831 - SME: Move the streaming primitives into arch/SME/PacketMath.h
1bd7cb0d1 - CI: Add an nvhpc-tests label for the NVHPC jobs
48b196e67 - Core: Drop constexpr from the noinline complex sqrt fallback
541f05781 - BDCSVD: Reallocate after setSwitchSize
dae92cd3a - BLAS: Add ILP64 (64-bit integer) build and test support
b79499c6f - AltiVec: Add pcmp_lt for Packet2l, fix the POWER7 build, and name the POWER8 check
6accef0ca - Core: Default-initialize storage in the expression constructors
0011719b1 - CMake: Disable if-conversion in the AVX512-FP16 tests on affected GCC
47bf0b657 - Tensor: Reject u == 0 in the normal random generator
859c035ba - Tests: Compile cheap split-test parts together
d22c0df89 - Tests: Build redux_bounded_compile_vectorized with the job's flags
d9ec2c9b9 - GPU: Copy views and adopt from a Context in the dense solvers' rvalue overloads
ae47f25fb - CI: Isolate Windows sccache daemon port by CI_JOB_ID and ensure script teardown
3d836af2d - Tests: Stop special_functions from failing on cancellation
ff16afbb1 - GPU: Give cuBLAS aligned copies of complex host scalars
cbda2af3e - Core: Fix vector shape validation when automatic resizing is disabled
077d29983 - Core: Clamp the exponent before vscalef in the AVX-512 pldexp
9a97f13f6 - CI: Stop stale sccache servers on Windows runners
cd1a734f2 - Core: Preserve bfloat16 packet sign under FTZ and DAZ
93933c899 - Core: Split pldexp's scale into three factors and round it once
356cdfd6e - Core: Preserve float and double sign under DAZ/FTZ
a63497bea - CI: Run the GPU jobs and the cuBLAS/cuSOLVER tests for GPU-module changes
493c3a3a8 - GPU: Check CUDA and library calls in release builds
8e08dc1f5 - Jacobi: Select bounded ratio operands before division
851458df8 - Core: Compute subnormal integer powers exactly on ARMv7 NEON
38c71b0cf - GPU: Reduce into a caller's DeviceScalar, and compute norm() from the dot product
a55314c6b - GPU: Stop solver result getters from waiting on unrelated GPU work
ab89b1ab1 - GPU: Fix complex device-scalar axpy/scal and empty-operand GEMM on DeviceMatrix
367032aa0 - Core: Add adjoint() to diagonal, permutation and transposition matrices
d17bdf5e9 - Tests: Index the FFT reference DFT's twiddles instead of calling exp
a44124c5a - Core: Preserve reduction bounds through unary expressions
86e6ae8e2 - SME: one disjoint part of the result per SME unit, minimum task size per scalar type
a1afbf4dd - Core: Vectorize isZero() and isApproxToConstant() checks
04bfe2287 - Core: Simplify packet traversal for strided coefficient assignments
1bd8e6f4d - SME: thread cap at the unit count, in-place LHS, ZA packers, kernels for narrow and tiny results
4df51993a - StructuredMatrices: Make DPR1 deflation exact for equal poles and option-independent
ef5a9d5c4 - Core: Refactor approximate comparisons and fix false positives
d142522b7 - Core: Fix the TRSM padded solve's link error in unoptimized C++14 builds
d1b87b4cd - Core: Solve whole TRSM diagonal blocks with the shared packet kernel
651678320 - Core: Respect resizing restrictions and streamline small product evaluation
ff1458835 - CI: Use '-' prefix for MSVC compiler flags so sccache caches Windows builds
6d293af5e - Core: Add lazy sums of dense, triangular and diagonal operands with diagonal and permutation matrices
563e4f40a - Compute integer powers by double-word repeated squaring, real and complex
ed56f0c1d - Eigenvalues: Keep subnormal couplings under FTZ on MSVC and fix the test build
92b9ed6b0 - StructuredMatrices: Apply alpha to x in non-finite sparse Kronecker products
a80a9cd50 - CI: Fix the GPU README docs failure and add a docs-build MR label
dd16a1d7f - CI: Run the lint checks in one job on a prebuilt lint image
38295af54 - Core: Fix RealView coefficient references and IndexedView/RealView data() constness
0ec724769 - CI: Run clang-tidy on a prebuilt, pruned Ubuntu image
af8d01b89 - Tests: Use explicit expected values for complex rsqrt corner cases
ac940d63c - CI: Build the Windows CUDA tests through ccache, not sccache
9c7cebf31 - Core: Keep pldexp's scale factors from being reassociated under fast math
ac63c7844 - CI: Pass failtest compiles through a missing launcher
de177ddbb - CI: Provision sccache on Linux test runners for failtest jobs
a7a0e35ee - Tests: Bound the scaled_permutation sum check by the terms, not the result
5fac38847 - Householder: Multiply a HouseholderSequence by a row vector
8746be0f2 - Core: Silence GCC 10 loop warnings in slice-vectorized reductions
91c8a4d8b - Householder: Apply a one-row reflector from the left
2dce125ec - CI: Fix sccache GCS credential JSON key and mask Windows token dump
53c33b64b - Core: Fix strides reported by vector IndexedView and strided RealView
9d194e7ad - Householder: Multiply HouseholderSequence by diagonal, triangular and permutation matrices
7681b6b7a - Core: Add permutation products with triangular and self-adjoint views
8a070f3e1 - CI: Add sccache with GCS remote backend with ccache fallback.
4b0b97830 - Core: Accelerate real vector operations with SME2
96aad0baa - Core: Fix scalar handling in mixed and self-adjoint products
59c28e04c - GPU: run Eigen's ConjugateGradient on the GPU types, with solveWithGuessInPlace
19292c62a - Docs: Fix the unresolved BSR section link in the GPU README
a1a19e6b2 - Core, SVD, StructuredMatrices: Scale all-subnormal data exactly on flush-to-zero hardware
4cba415f3 - CI: Pre-bake curl, ca-certificates, and python3 into smoke build Dockerfiles
087757ad5 - Docs: Use Doxygen inline-math delimiters in ScaledPermutationMatrix.h
e89236d8f - GPU: Add cuSPARSE BSR products for BlockSparseMatrix
d30a9e6d3 - Core: Fix reshape traversal order and unsafe packet assignments
6f4f522f5 - EventCount: Fix signal loss in CancelWait
b346103bd - Docs: Treat proprietary software as a black box
eef236cb4 - SME: numext::round_down for the non-streaming panel peels
776f395d2 - Core: Add ScaledPermutationMatrix, the product of a permutation and a diagonal matrix
ec8593a7d - Fix deprecated enum-enum conversion warning
cba8148e3 - Core: Add products of two triangular or self-adjoint views
45e88a56d - Core: Harden the structured-product dispatch for shapes without a kernel
34e0e0793 - Eigenvalues: Reorthogonalize globally unresolved block shifts
f26a444af - Core: Fix implicit Matrix conversions in std::variant
d5a19d2ca - Fix GCC array-bounds warnings in Core, Sparse and Tensor
ab4f7316d - CI: Auto-retry infrastructure drops on runner loss or preemption
14e4b5405 - Jacobi: Use power-of-two scaling and preserve subnormal real Givens rotations
d32e9702b - SME: NEON path for small blocks, 64-byte packed panels, generic crossovers
647bc59a9 - Core: Fix scalar rewrites of structured products
8c6b5162b - Tests: Fix redux_12 and structured_kronecker_12 on 32-bit ARM
82a7cb22a - Agents: Present benchmark measurements as tables in merge request descriptions
47a78413d - Eigen: Standardize LAPACK helper routines into native C++ implementations
399d29312 - StructuredMatrices: Accept sparse factors in KroneckerOperator
eddc25291 - SparseLU: Divide by subnormal pivots instead of forming their reciprocal
1b5286a2c - MSVC: Fix complex-component evaluator and tensor block-view builds
b5c3e3da6 - Core: Use four accumulators in the vectorized reductions
c7c04d737 - Core: Stabilize scalar complex square roots
7051b1f9a - Tests: Materialize the random vector before makeHouseholder in householder_essential_expressions
46a2d0c56 - Eigenvalues: Share and harden direct self-adjoint solver scaling
3ca5ab304 - CI: Allow runner-level CCACHE_* overrides via EIGEN_CI_CCACHE_* defaults
e3c8cb150 - Core: Respect packet access in in-place transpose
1eef8230d - Core: Fix reverse row and column iterator indexing
04cc35554 - Core: Fix packet twoprod residuals and correct its error bound
a076632bf - Performance: Avoid temporaries in non-aliasing product assignments
c1a957f96 - Core: Share inner product dispatch and retain a tiny-vector path
59d9e069a - Tensor: Fix the GPU warnings that overflow the build log
0039ee376 - CI: Run 128-bit SVE tests natively on dedicated ARM runner
33fd96d09 - Docs: Re-attach the Transform::rotate Doxygen block and fix the GMRES \param name
93a193c58 - Core: Vectorize triangular solves across right-hand sides
e1c311414 - Core: Enable linear counting and simplify visitor unrolling
e7e5178d1 - Core: Exclude unreachable wide reductions for bounded expressions
2396b9dc2 - Core: Keep complex and class operands of the PowerPC GCC barrier in memory
02f77fd76 - CI: Skip the ccache round trip when an affected build compiles nothing
ff285117c - Tests: Check backward error in dontalign QR solves
94f0b704f - Core: Reject Boolean subtraction and negation
8f8d4ed4c - Geometry: Support mixed-scalar operations in Transform::translate, pretranslate, and toRotationMatrix
dd755f75e - Householder: Suppress false-positive -Warray-bounds warnings in BlockHouseholder
bbcff8c67 - Eigenvalues: Avoid double promotion in scaling test
2800561ee - Core: Simplify and vectorize permutation operations
2a57319e6 - CMake: Compile GPU tests through CUDA and HIP language support
60c5525bb - Eigenvalues: Reuse QZ workspaces during compute
3f21c28a5 - CI: Parallelize the failtest suite and run native x86-64 tests on the large runner
154a40bd4 - Tensor: Avoid copying strided block operands
2002cae43 - Core: Avoid -Wfloat-equal warnings in SafeScaling for long double
2e7712e63 - CI: Use saas-linux-medium-amd64 for smoke build jobs
da8cda1da - Core: Vectorize complex component reductions
a14735e7e - Core: Add move support for DiagonalMatrix
4990b8359 - Eigenvalues: Break stalled ComplexQZ iterations with exceptional shifts
20b41e40b - QR: Document transposed solve limits and condition wide tests
99f1cccfd - Tests: Bound erfc comparisons in the subnormal range
cd38c5ddc - Core/QR: Vectorize complex partial norms and simplify QR initialization
6a0625b84 - Sparse: Share packet scatter updates in Cholesky and LU
1373c7a36 - CI: Build only the affected selection in the SVE affected jobs
2298fbd89 - CI: Raise the SME full build timeout to 3h
20f5a26ba - Core: Remove misc directory and simplify LU expressions
15f227178 - Eigenvalues: Avoid cache-conflicting strides in RealSchur workspaces
74601a9e3 - CI: Log machine configuration before builds and tests
4a6bf607a - Eigenvalues: Use safe scaling in tridiagonal eigensolvers
caac397f4 - Tensor: Enable double outer reductions with native sum atomics
27b0c620e - Tests: Allow for MSVC expm1 reference error
8facea8ba - SVD: Pack active secular terms for contiguous evaluation
6e86f64a0 - Core: Consolidate structured matrix norm utilities
5ad7d1788 - SVD: Reuse the saved row across blocked Jacobi rotations
ec5876775 - SVD: Reuse operands while constructing singular vectors
a4d943d1f - CI: Cross-compile the aarch64 builds on amd64
219e6c5a3 - HIP: Build tests for the requested GPU architectures
0f96e0f16 - Tests: Bound SVD normal-equation checks by backward error
7ec23fec7 - Core: Fix diagonal strides and subnormal identity scaling
68af15188 - Benchmarks: Add GPU kernel timing and validated baselines
50c7c77f3 - Tensor: Query attributes for the stream's owning GPU
4eaf42d05 - Docs: Ground performance changes in cost models and evidence
ddf4a5479 - Eigenvalues: Use safe scaling in the iterative self-adjoint solver
3f4260bfe - Eigenvalues: Fix inconsistent nonfinite input status in in-place eigensolvers
5df2459f1 - Polynomials: Refine PolynomialSolver roots with Ehrlich-Aberth iterations
99baace37 - Eigenvalues: Use power-of-two scaling in Schur decompositions
6a026803b - CI: Cap parallel compiles in the aarch64 gcc-10 affected build
e5b838361 - Geometry: Work around an LLVM ARM load/store miscompile in vectorwiseop
37e72f8df - Tests: Remove duplicate and near-duplicate test coverage
3bda1a485 - Warings: Fix some compiler warnings
133b7aa9d - StructuredMatrices: Vectorize scaling and reuse spectral work
f0a977a96 - Tests: Work around an LLVM ARM load/store miscompile in vectorwiseop
0b9759a73 - Eigenvalues: Support inplace decomposition through Ref<>
c0d6e9944 - GPU: Fix packet capabilities and simplify half operations
7a6afd466 - Geometry: Restore 4D vector assignment delegation in AngleAxis::operator=
b17fadee5 - Core: Avoid -Wfloat-equal warnings in ceil_power_of_two for long double
cae72577d - SparseCore: Silence unused variable warning under NDEBUG in triangular_solve_over_reach_iter
2d148dc46 - Tests: Add device packet-math coverage shared with the host suite
688bc7a5c - Core: Fix scalar complex GEMM accumulation on RVV

PiperOrigin-RevId: 995455097
Change-Id: Idbbc059f9c9e5e999ec6c62d97ec2a5392c935d5