cheezwiz7899
d87f1241c9
Fix texture corruption caused by the shader compiler treating unknown constant buffer sizes as 0 instead of the bindless array length: - Set BINDLESS_FALLBACK_LENGTH = BINDLESS_ARRAY_LENGTH (1024) in texture_pass.cpp so shaders always allocate the full descriptor range - Pre-capture cbuf sizes in GraphicsEnvironment/ComputeEnvironment constructors and serialize them through FileEnvironment to prevent intermittent 0-size reads from async worker threads - Bump CACHE_VERSION to 14 to force one-time recompile of existing shader caches with the corrected metadata - Reduce MAX_IMAGE_ELEMENTS from 16384 to 1024 to eliminate TLS pressure caused by the oversized per-thread descriptor array Add BindlessCache (64 entries, round-robin eviction) keyed on (cbuf_addr, count, image_table_generation) to skip redundant GPU memory reads across draws: - Track image_table_generation in each cache entry; a generation mismatch forces a full re-read and re-resolve, a match replays cached views - Mark resolved ImageViewInOuts with id_cached so FillImageViews can skip VisitImageView for already-resolved entries - Add GraphicsImageTableGeneration() / ComputeImageTableGeneration() accessors to DescriptorTable and expose them from TextureCache Deduplicate PrepareImageView calls for cached bindless entries: - On a cache hit, FillImageViews was calling PrepareImageView once per descriptor slot (up to 1024 calls/draw). Replace with a batch pass that collects unique ImageIds from id_cached views, deduplicates via sort+unique, then calls PrepareImage once per unique image (~10-50x fewer calls in practice)
Citron Neo
Feel free to open issues and PRs to the repository.
- Development & maintanance: https://discord.gg/7bdSQgQZfu
Enjoy!
Abstract
Citron is a specialized research software designed to orchestrate high-fidelity virtual environments. Unlike general-purpose tools, it focuses on the intersection of containerization and deep-system virtualization to provide researchers with granular control over network behavior and resource management. Core Capabilities
High-Fidelity Networking: Citron creates Docker container environments where network accuracy is the priority. It ensures that latency, throughput, and topology mirror real-world conditions with high precision, making it ideal for testing distributed systems or network protocols.
Hybrid Orchestration: Beyond containers, it can spawn and manage multiple Virtual Machines (VMs) simultaneously. This allows for heterogeneous testing environments where containers and VMs interact seamlessly.
Kernel Addressing: The software utilizes advanced kernel addressing techniques for resource management. By interacting closely with the host and guest kernels, Citron achieves low-overhead monitoring and advanced control over virtualized assets that standard management layers often miss.
Summary Citron is a technical powerhouse for researchers who need more than just "standard" virtualization. It bridges the gap between the lightweight agility of Docker and the robust isolation of VMs, all while maintaining a rigorous, high-fidelity network layer managed through direct kernel-level operations. Would you like to draft a README file or a technical abstract for a research paper based on this?
Citron functions as a high-fidelity hardware-software co-design platform that bridges the gap between high-level container orchestration and low-level embedded hardware constraints. By integrating specialized architectures like Xtensa cores, Citron allows researchers to simulate and manage complex system-on-chip (SoC) behaviors within a virtualized framework.
-
Integration with Specialized Hardware (Xtensa Cores) Citron addresses the unique constraints of Xtensa processors, which are modular, extensible 32-bit RISC architectures often used in networking and audio processing.
Custom Instruction Support: Citron can leverage the Tensilica Instruction Extension (TIE) language to simulate custom datapath elements and instructions within the virtualized environment. Protocol-Specific Optimization: Because Xtensa cores excel at processing packet headers and rule-based checks, Citron uses them to maintain network fidelity—ensuring that simulated network stacks perform with the same cycle-accurate behavior as physical networking chips.
-
High-Fidelity Computing Devices To achieve "high fidelity," Citron must manage the deterministic performance of computing devices, ensuring that virtualized sensors and actuators respond within real-world timing constraints.
Cycle-Accurate Modeling: It utilizes an Instruction Set Simulator (ISS) to provide instant feedback on how software interacts with the underlying hardware pipeline, preventing the "timing drift" common in standard VMs. Resource Determinism: Citron mitigates hardware constraints—such as limited memory and power—by tailoring the virtual environment to match the specific cache sizes and memory hierarchies of the target device.
-
Advanced Kernel Addressing and Management Citron’s specialization in kernel addressing allows it to bypass traditional virtualization overhead:
Direct Register Access: By interacting with the Application Binary Interface (ABI), Citron manages how programs interact with the kernel, allowing for precise debugging and resource allocation across multiple spawned VMs. Kernel-Level Hypervisors: It functions similarly to a KVM (Kernel-based Virtual Machine), turning the host Linux system into a high-performance hypervisor that provides near-native execution of privileged instructions. Memory Lookup Interfaces: Citron can connect directly to arbitrary-width memories or RTL (Register Transfer Level) blocks for low-latency data transfers, effectively treating virtualized memory as if it were a direct point-to-point hardware connection.
-
Technical Constraints & Architecture Feature Implementation in Citron Processor Type Supports 32-bit configurable RISC/Xtensa architectures. Virtualization Method Hardware-assisted virtualization using extensions like Intel VT-x or AMD-V for direct execution. Networking Layer High-fidelity emulation that avoids standard I/O bottlenecks by using custom processor interfaces. Management Centralized synchronization controller for container scheduling and I/O compensation. Would you like to explore how Citron handles specific network protocols like CoAP or MQTT on these virtualized Xtensa cores?