View release highlights for previous versions of AMD Vitis™ here.

AMD Vitis™ Software Platform 2026.2 Release Highlights

Enhanced Design Flow with AMD Versal™ AI Engines

New Vitis™ Library Functions: Least Squares Cholesky Solver
  • End-to-end Cholesky solving with forward and backward substitution on AI Engines
  • Configurable solver IPs for channel estimation, equalization, and beamforming weights
  • Available in open-source Vitis Libraries and Vitis Tutorials (28-solver_library)
New Vitis Library Functions: Packet Switching I/O for Compute-Bound IPs
  • Shared PLIO connections for compute-bound DSP IPs using packet switching
  • Performance improvement:  Fewer PLIOs and  less block RAM, with unchanged DSP performance when compared to a conventional channelizer implementation using AMD Vitis 2026.1 tools
  • Freed interface resources enable larger designs within existing PLIO limits
New Vitis Library Functions: ML-Based Ease-of-Use Configuration Helper (EA)
  • ML-based config_helper() recommends optimized parameters from performance targets and resource constraints
  • Simplified DSP IP configuration without detailed tile architecture knowledge
New FP8 Data Type Support for AIE-MLv2 (Versal™ AI Edge Series Gen 2 Devices)
  • Production-ready bfloat8 (E5M2) and float8 (E4M3) support for AIE-ML v2
  • New caccfloat accumulator for floating-point matrix multiplication
  • Support across AI Engine API operations, matrix multiply, sliding_mul_ops, and ADF graphs
  • Low memory bandwidth for efficient AI inference and DSP processing
Improved AI Engine Compiler: Per-Kernel Xchess Pragma Control
  • Per-kernel Xchess pragma control directly in kernel source code
  • Kernel-level constraints override -Xchess flags for the specified kernel
  • Lower compile overhead with precise per-kernel performance tuning
Independent Partitioning for Versal AI Engines
  • Faster iteration by rebuilding only modified AI Engine partitions
  • Runtime partition switching for flexible AI Engine reconfiguration
  • Host-controlled runtime partition switching using three separate .xclbin files
AI Engine Rotating Margin API for Local Tile Memory (available on Versal AI Edge Series Gen 2)
  • Ring-buffer margin updates using adf::rotating_margin<> and advanceHead, without data copying
  • Lower AIE-ML tile memory use and faster processing versus 2026.1 and earlier

Enhanced Verification Capabilities for Versal Adaptive SoCs

x86sim Support for Encrypted Source Code (Early Access)
  • Functional simulation of encrypted kernel source code using x86sim
  • Protected IP distribution with customer validation using x86sim
  • Faster encrypted third-party IP verification versus Vitis 2026.1 and earlier
  • Complex encrypted designs: minutes in x86sim versus hours in aiesim (2026.1 and earlier)
Code Coverage Visualization in Vitis Analyzer (Early Access)
  • Code coverage viewing and analysis directly in Vitis Analyzer
  • Visual highlighting of kernel code lines exercised during x86sim simulation
  • Faster identification of untested code paths for targeted verification
  • Greater confidence in AI Engine kernels before hardware validation
Vitis Functional Simulation: Expanded Protocol, Data Type, and Graph Support
  • AXI4-Stream side-channel verification (TLAST, TUSER, TDEST) using Python™ or MATLAB®
  • Packet-switching simulation for multi-destination AI Engine routing verification
  • Bit-exact AP float verification using standard float arrays, without conversion code
  • AXI4-Lite coefficient updates between simulation runs, without rebuilding
  • Output-only AI Engine graph simulation for NCOs and tone generators
Vitis Functional Simulation: Versal RF Series Hard IP Support
  • Versal RF LDPC decoder, poly block channelizer, and arbitrary resampler models (Early Access)
  • Python/MATLAB spectrum analyzer tutorial combining AI Engine channelization, HLS windowing, and Versal RF FFT
  • Reference: Vitis Functional Simulation User Guide (UG1864)
Vitis Hardware in the Loop (Vitis HIL) Updates
  • Automatic Vitis HIL server startup after board boot
  • Production support for VRK160 and VEK385 evaluation boards

Enhancements to Vitis Model Composer

Simplified Super-Sample-Rate (SSR) Connectivity
  • Automatic multi-lane SSR replication for AI Engine Kernel and Class Kernel blocks
  • Single matrix port for SSR lanes instead of individually wired ports
  • HLS SSR wrapper import using a single matrix-style port
  • Automatic SSR-enabled IP reconnection when lane counts change
  • Simpler design review and less rewiring when scaling SSR lanes
New AI Engine DSP Solver Library Blocks
  • QRD (Gram-Schmidt/Householder), SVD, and Cholesky blocks for AIE, AIE-ML, and AIE-ML v2
  • Details: Vitis Solver Library
FIR Compiler Updates and HLS Library Changes
  • Versal complex-data FIR support: single-channel systolic MAC, single-rate/integer-rate interpolation
  • Floating-point input mapping to DSPFP32 primitives on Versal devices
  • Selectable INT8 multiplication using DSP58 9x8 three-element dot-product mode
  • Vector xFFT HDL block with automatic SSR wrapping of FFT 9.1 IP
  • HLS Library deprecated in 2027.1; migrate to the HLS Kernel block
Faster Simulation with SystemC/TLM Models for Versal RF Series Hard IP (Early Access)
  • SystemC/TLM models for Versal RF FFT simulation (Early Access)
  • Bit-accurate functional simulation faster than cycle-accurate RTL simulation
  • Rapid algorithm iteration with hardware-equivalent arithmetic accuracy
  • Additional Versal RF Series hard IP TLM models planned for future releases

Updates to Vitis IDE for Embedded Development

Enhanced Theia AI Framework for AI-Assisted Vitis Workflow
  • Automatic MCP tool invocation and reusable skills, with cloud-based and local AI models
  • Model choice: Anthropic, OpenAI, DeepSeek, GLM and Qwen, selectable directly in AI Chat
  • Skills in standard folders (e.g., ~/.claude/skills) are detected automatically
  • Persistent chat history with file, image, and editor-context attachments
  • Selectable models with 1M/200K context windows and configurable reasoning effort
  • Architect, Coder, and User Guide agents by default; others are configurable
  • Slash commands for Vitis tool-specific help, via AMD Ross™ repository skills set in AI Settings
  • Ross documentation-search MCP server integrated in chat, with per-chat skill and server controls
  • Interactive Shell lets AI run commands step by step and use the immediate feedback
Debug Embedded Targets Directly from VS Code
  • Versal and Zynq™ device debugging in VS Code using two bundled Vitis software extensions
  • Ready-made Versal/Zynq device launch configurations for Linux® board/QEMU and bare-metal JTAG debugging
Vitis Python Console and Enhanced Query APIs
  • Integrated Python console with live server connection, GUI updates, and autocomplete
  • Build query APIs for component status, artifacts, timestamps, and dependencies
  • Platform inspection and configuration APIs for AI-assisted workflows, without duplicate properties
Enhanced Debugging Experience in Vitis IDE
  • Hover-to-inspect values for scalars, pointers, structs, and arrays during debugging
  • Right-click watchpoints on variables or address ranges halt execution on memory access
  • Inline Watch expansion, synchronized formats, prioritized errors, and persistent symbol paths
Zephyr IDE Extension: New Device Support and Device Tree Creation
  • Zephyr IDE support for Versal AI Edge Series Gen 2 APU/RPU and MicroBlaze™ V Processor
  • One-click device tree generation using SDTGEN and west lopper
  • Integrated Vitis debugging for any Zephyr binary using User Managed Mode

Streamlined Platform and System Deployment for Versal Adaptive SoCs

EDF/Yocto Platform Migration
  • Production EDF/Yocto® support for complete Versal portfolio, with an EDF-based APU Linux stack
  • Data-driven device tree and BOOT.bin generation
  • AMD authored CED base platforms for Versal development kits: VCK190 and VEK280 (flat), VRK160 and VEK385 (segmented), for hardware and emulation
  • Boot Metadata Archive (.bma) with user-defined naming
  • Hierarchical .dtsi device trees, with no need to merge them into a single file
DFX & Reconfiguration
  • DFX Abstract Shell is now in production in the Vitis DFX flow
  • Faster RM iteration, multi-team scaling and IP protection, since RM developers see only the interface boundary
  • Limitation: hardware emulation is not supported with Abstract Shell
  • v++ --package embeds rm.dtbo per reconfigurable module in the .xclbin
  • XRT delegates PL loading and decoupler management to libdfx/fpga_manager
  • Unified Image Loading (Early Access)
  • Single XRT image loading API (Production) for PL and AI Engine overlays across flat, segmented and DFX configurations
  • Smaller segmented xclbins: LOAD_PDI is no longer set, and the PDI is programmed separately
  • Driver-agnostic fpga_accelerator device tree node supports AI Engine soft partition and runtime overlays
  • Sysfs hardware handoff access, and custom user drivers alongside XRT
Vitis Software & AI System Integration for Versal AI Edge Series Gen 2
  • NoC QoS tuning for ML workloads via v++ --connectivity.noc.* directives or Vivado™ Design Suite set_noc_phase
  • Back-annotation of QoS settings from a post-placer .nts file
  • First-class training-libadf.a NoC training handoff in v++ --link, replacing gmio_train

Vitis What's New by Category

Expand the sections below to learn more about the new features and enhancements in AMD Vitis software platform 2026.2. For information on supported platforms, changed behavior, and known issues, please refer to the Vitis software platform 2026.2 Release Notes for the Application Acceleration Flow and Embedded Software Development Flow.

New Vitis Software Library Functions: Least Squares Cholesky Solver
  • End-to-end Cholesky solving with forward and backward substitution on AI Engines
  • Configurable solver IPs for channel estimation, equalization, and beamforming weights
  • Available in open-source Vitis Libraries and Vitis Software Tutorials (28-solver_library)
New Vitis Library Functions: Packet Switching I/O for Compute-Bound IPs
  • Shared PLIO connections for compute-bound DSP IPs using packet switching
  • Channelizer: ~2.5x fewer PLIOs and ~2x less block RAM, with unchanged DSP performance
  • Freed interface resources enable larger designs within existing PLIO limits
New Vitis Library Functions: ML-Based Ease-of-Use Configuration Helper (EA)
  • ML-based config_helper() recommends optimized parameters from performance targets and resource constraints
  • Simplified DSP IP configuration without detailed tile architecture knowledge

Applies to:

  • Versal™ AI Core Series with AIE
  • Versal AI Edge Series with AIE-ML
  • Versal AI Edge Series Gen 2 with AIE-ML v2

x86sim Support for Encrypted Source Code (Early Access)
  • Functional simulation of encrypted kernel source code using x86sim
  • Protected IP distribution with customer validation using x86sim
  • Faster encrypted third-party IP verification versus Vitis Software 2026.1 and earlier
  • Complex encrypted designs; minutes in x86sim versus hours in aiesim (2026.1 and earlier)
Code Coverage Visualization in Vitis Analyzer (Early Access)
  • Code coverage viewing and analysis directly in Vitis Analyzer
  • Visual highlighting of kernel code lines exercised during x86sim simulation
  • Faster identification of untested code paths for targeted verification
  • Greater confidence in AI Engine kernels before hardware validation
Vitis Functional Simulation: Expanded Protocol, Data-Type, and Graph Support
  • AXI4-Stream side-channel verification (TLAST, TUSER, TDEST) using Python™ or MATLAB®
  • Packet-switching simulation for multi-destination AI Engine routing verification
  • Bit-exact AP float verification using standard float arrays, without conversion code
  • AXI4-Lite coefficient updates between simulation runs, without rebuilding
  • Output-only AI Engine graph simulation for NCOs and tone generators
Vitis Functional Simulation: Versal RF Series Hard IP Support
  • Versal RF LDPC decoder, poly block channelizer, and arbitrary resampler models (Early Access)
  • Python/MATLAB spectrum analyzer tutorial combining AI Engine channelization, HLS windowing, and Versal RF Series FFT
  • Reference: Vitis Functional Simulation User Guide (UG1864)
Vitis Hardware in the Loop (Vitis HIL) Updates
  • Automatic Vitis HIL server startup after board boot
  • Production support for VRK160 and VEK385 evaluation boards

Simplified Super-Sample-Rate (SSR) Connectivity
  • Automatic multi-lane SSR replication for AI Engine Kernel and Class Kernel blocks
  • Single matrix port for SSR lanes instead of individually wired ports
  • HLS SSR wrapper import using a single matrix-style port
  • Automatic SSR-enabled IP reconnection when lane counts change
  • Simpler design review and less rewiring when scaling SSR lanes
New AI Engine DSP Solver Library Blocks
  • QRD (Gram-Schmidt/Householder), SVD, and Cholesky blocks for AIE, AIE-ML, and AIE-ML v2
  • Vitis Solver Library
FIR Compiler Updates and HLS Library Changes
  • Complex-data FIR support for Versal devices: single-channel systolic MAC, single-rate/integer-rate interpolation
  • Floating-point input mapping to DSPFP32 primitives on Versal devices
  • Selectable INT8 multiplication using DSP58 9x8 three-element dot-product mode
  • Vector xFFT HDL block with automatic SSR wrapping of FFT 9.1 IP
  • HLS Library deprecated in 2027.1; migrate to the HLS Kernel block
Faster Simulation with SystemC/TLM Models for Versal RF Series Hard IP (Early Access)
  • SystemC/TLM models for Versal RF FFT simulation (Early Access)
  • Bit-accurate functional simulation faster than cycle-accurate RTL simulation
  • Rapid algorithm iteration with hardware-equivalent arithmetic accuracy
  • Additional Versal RF Series hard IP TLM models planned for future releases

QoR Enhancements 
  • Vitis HLS scheduler displays information on why ‘II’ target is not met, which helps users understand the details without expert knowledge 
  • Auto-Restart feature enables support for finite or infinite iterations and integrated Auto-Restart + Direct I/O during co-simulation 
Direct I/O during co-simulation 
LogiCORE™ IP Enhancements 
  • Automatic import of LogiCORE IP parameters to create an auto-generated C++ header file 
  • Support for Versal RF LDPC IP core and DSP IP feature updates for FIR Compiler 
Vitis HLS Agentic AI Flow Released with AMD Ross™ Agentic AI Assistant 
Ease of Use Enhancements 
  • Interactive RTL-to-C++ cross-probing built into the Vitis software
  • Performance Pragma enhancements
  • Pragma Success Report coverage significantly expanded to cover more pragma types