AOCL 5.2.2 Release

AOCL-BLAS 5.2.2 Release Notes

Overview

AOCL-BLAS 5.2.2 is an incremental release building on the 5.2 GA release, delivering performance optimizations, bug fixes, improved threading stability, and expanded test coverage.

Performance Optimizations

• Optimized SGEMM rd kernels on Zen3
• Improved SGEMM rd kernel on Zen4/Zen5
• SGEMM tiny path tuning for Zen4 and Zen5
• Added tiny path for SGEMM
• Added fast path for single-threaded AVX512 DGEMV kernel
• Replaced intrinsics with inline assembly for bli_saxpyv_zen4_int and bli_saxpyf_zen_int_5
• Improved fringe case handling for AXPYV kernel
• Disabled small_gemm for Zen4/Zen5 and added single-thread check for tiny path

Bug Fixes

• Fixed memory leak in DGEMV kernel
• Fixed extreme values handling in GEMV
• Fixed integer division in GEMV that was supposed to be a double operation
• Fixed Integer Overflow issue in TPSV
• Fixed out-of-bound access in F32 matrix add/mul ops
• Bugfix: BF16 to F32 conversion in AVX2 F32 codepath
• Bug fix in BF16 AVX2 conversion path
• Fix for F32 to BF16 conversion and AVX512 ISA support checks
• Fixed cblas_ctrmm invalid diag handling
• Coverity issue fix for ZTRSM
• Fixed Coverity static analysis issue in DTRSM
• Fixed high priority Coverity issues in LPGEMM
• Resolved operator precedence warning in Zen5 DCOMPLEX threshold logic
• Modified AXPY kernel to ensure consistency of numerical results

Threading & Stability

• Fixed data race in native code-path
• Add OpenMP barrier before releasing threadinfo & global communicator to avoid race
• Replaced OMP barrier with bli_thread_barrier and added similar fixes
• Global communicator is now freed outside the parallel region
• Thread: free global communicator after parallel region completes
• Initialize mem_t structures safely and handle NULL communicator in threading
• Fix DTL dynamic thread logging in BLAS operations
• Added dynamic threads and actual threads in the DTL log of SAXPY
• Enabled disable-sba-pools feature in AOCL-BLAS

Build System & Infrastructure

• Updates to the build systems (CMake and Make) for LPGEMM compilation
• CMake: Adding targets and aliases so that BLIS works with FetchContent
• Set security flags default enable
• DTL Windows getpid support
• Add compiler information to make showconfig and bench_getlibraryInfo
• Make all bench applications consistent
• Standardize Zen kernel names

Test Suite (GTestSuite)

• Added Banded API tests: gbmv, hbmv, sbmv, tbmv, tbsv

• Added Packed API tests: hpmv, spmv, tpmv, tpsv, hpr, hpr2, spr, spr2

• Added conjugate dot and ger IIT_ERS tests

• Added data pool support

• Moved data generator definitions to a cpp file

• Computediff improvements

• Fix in swap

• Break up tests for better organization

• Multiple miscellaneous test fixes

• Code tidying

AOCL 5.2 GA Release

AOCL-BLAS 5.2 Release Notes

Overview

This release includes significant performance improvements, new features, and critical bug fixes for the AOCL - BLAS linear algebra library, with optimizations specifically targeting AMD Zen4 and Zen5 architectures.

---

Performance Improvements

GEMM Improvements

• Tuned ZGEMM thresholds for Zen4 and Zen5 architectures
• Optimized AVX512 ZGEMM kernel and edge-case handling
• Improved ZGEMM packing kernel for M-dimension edge cases
• Developed Optimal thread selection logic for ZGEMM on Zen5

GEMV Enhancements

• Added DGEMV no-transpose multithreaded implementations
• Exported AVX512 DGEMV kernels
• DGEMV bug fixes and code cleanup
• Added ability to handle non-unit incx in GEMV transpose kernel
• Improved numerical precision in ZGEMV API

DCOPY Optimization

• Tuned DCOPY aocl_dynamic logic for Zen4/Zen5 architectures

---

New Features

• Additional build options to disable optimized code paths for smaller matrices in GEMM and TRSM

  • Useful for testing and benchmarking
  • Reduces numerical rounding differences when repeating calculations with different core counts

• Complete set of GEMMTR APIs implemented

---

Bug Fixes

Critical Fixes

• Fixed probable integer overflow in TPSV
• Fixed ZTRSM accuracy for conjugate transpose
• Fixed DTRSM small threshold for extremely skinny sizes on Zen5

---

Acknowledgments

This release is the result of contributions from the AOCL team at AMD and the broader BLIS community.

---

Release Date: January 2026
Version: 5.2 GA

AOCL-BLAS 5.1 GA

Performance Optimizations

• DGEMM, DTRSM, DGEMV, ZGEMM, DTRSV, DCOPYV on Zen4/5
• DSCALV, DDOTV on Zen3
• Benchmark support for ASUMV
• Minor Bug Fixes.

Aocl-gemm Add-on Module updates

• AOCL_ENABLE_INSTRUCTIONS support
• batch_gemm support for all data types
• New Output Datatype for Integer APIs
• BF16 Support on AVX2 Platforms
• WOQ with/without Group Quantization
• Threading Framework Optimizations
• Reference Kernels for all Reorder APIs
• Performance Optimizations for all APIs
• Additional APIs and Post-Ops support in addition to the improved performance for the existing APIs

AOCL-BLAS 5.0

AOCL-BLAS 5.0 Release Highlights

• Added zen5 support
• Turin specific tuning for the APIs: D/ZGEMM, DTRSM and DNRM2
• AVX512 made improvements for the APIs: ZGEMV, D/ZAXPYF, D/ZDOTXF, ZDOTV, C/ZSCALV, DNRM2, S/D/ZCOPY, S/D/C/ZAXPBYV, DTRSV, DGEMMT, D/ZTRSM, and D/ZGEMM
• Improvements to the AOCL_ENABLE_INSTRUCTIONS functionality
• Additional APIs and Post-Ops support in addition to the improved performance for the existing APIs in aocl_gemm add-on

AOCL-BLAS 4.2

AOCL-BLAS 4.2 Release Highlights

• Added uint8 output and zero-point support in int8 API’s in aocl_gemm addon (Low Precision GEMM / LPGEMM)
• Improved performance for all downscaled versions of all API’s in aocl_gemm addon (Low Precision GEMM / LPGEMM)
• Multithread performance improved across API’s in aocl_gemm addon (Low Precision GEMM / LPGEMM)
• Introduced AOCL_ENABLE_INSTRUCTIONS environment variable as an alternative to BLIS_ARCH_TYPE, but with slightly different semantics.
• Improved functionality of XERBLA error handling routine in AOCL-BLAS.
• Performance optimizations for the following APIs:
  - DGEMM for tiny sizes
  - S/ZGEMM, D/ZTRSM, ZAXPBYV, Z/ZDSCALV, S/D/ZGEMV, and D/DZNRM2
• Following BLAS extension APIs have been added only for AMD “Zen” code paths:
  - sgemm_pack_get_size(), sgemm_pack(), and sgemm_compute()
  - dgemm_pack_get_size(), dgemm_pack(), and dgemm_compute()

AOCL-BLAS 4.1

AOCL-BLAS 4.1 Release Highlights

• Additional APIs and Post-Ops support in addition to the improved performance for the existing APIs in aocl_gemm add-on
• Dynamic dispatch and amdzen configuration support added to aocl_gemm add-on
• Dynamic dispatch feature enhancements.
• AVX 512-based optimizations for AMD “Zen4” platform:
  - SGEMM, DGEMM, and ZGEMM
  - DTRSM, D/ZAXPY, ZGEMV, DDOTV, and D/ZSCALV
• Improved support for OpenMP nested parallelism.

AOCL-BLIS 4.0

Highlights of AOCL-BLIS 4.0

• The following LPGEMM (Low Precision GEMM) variants are added along with post-ops support:
  • aocl_gemm_u8s8s32os32 and aocl_gemm_u8s8s32os8 routines are added and optimized using AVX-512-VNNI
  • aocl_gemm_u8s8s16os16 and aocl_gemm_u8s8s16os8 routines are added and optimized using AVX2
  • aocl_gemm_bf16bf16f32of32 and aocl_gemm_bf16bf16f32obf16 routines are added and optimized using AVX-512
• SGEMM with packed/reorder buffer support (aocl_gemm_f32f32f32f32)
• AMD “Zen4” support for BLIS
• Dynamic dispatch supports AMD “Zen4” configuration
• Optimizations and performance improvements for DGEMM, SGEMM, ZGEMM, DGEMMT, and DTRSM
• Framework design changes

AOCL-BLIS 3.2

New features:

• Extended BLAS function - DZGEMM
• Progress feature for xGEMM and xTRSM APIs: Time taken to complete the mathematical operations tends to increase exponentially with large input problem sizes; this feature provides users a periodic update on the operation progress.
• Runtime Threading control using OpenMP APIs
• Dynamic Dispatch covers APUs
• Improved detection of standard x86-64 feature support
• Minor bug fixes

Performance improvements in the following single-threaded and multi-threaded functions:

• DGEMM, SGEMM, ZGEMM, and CGEMM
• DTRSM, DGEMMT, ZTRSM, CTRSM, and DTRMM
• SGEMV, DHER2, ZTRSV, and DSYMV
• ?AXPBYV, SSCALV, DSCALV, ?DOTXV, and ZAXPY2V

AMD Optimized BLIS Version 3.1

AMD Optimized BLIS Version 3.1

Highlights of improvements on AMD EPYC^TM processor family CPUs

• Supports Dynamic Dispatch and AOCL Dynamic feature
• Improvements in DGEMM, ZGEMM, DTRSM, DSYRK, xGEMV, and DOTV

AMD Optimized BLIS Version 3.0.1

AMD Optimized BLIS Version 3.0.1

Highlights of improvements on AMD EPYC^TM processor family CPUs

• Improved performance of DGEMM for skinny matrix shapes.
• Improvements in SGEMM and ZGEMM
• Improved performance of Level-1 and Level2 BLAS routines, GEMV, DOT and AXPY routines
• Improvements in DTRSM for small matrix sizes