Vulkan RHI: High overhead of ICommandList::SetProgramBindings in multi-threaded rendering scenario

[egorodet]

Describe the bug "Asteroids" sample and "Parallel Rendering" tutorial built for Vulkan API has significantly lower FPS than for DirectX API in multi-threaded rendering scenario.

Steps To Reproduce Steps to reproduce the behavior:

1. Run "Parallel Rendering" tutorial built for DirectX, set grid size to 25 or higher => get ~400 FPS
2. Run "Parallel Rendering" tutorial built for Vulkan, set grid size to 25 or higher => get ~150 FPS
3. Run "Asteroids" sample built for DirectX, set scene complexity to 8 (35000 asteroids) => get ~294 FPS
4. Run "Asteroids" sample built for Vulkan, set scene complexity to 8 (35000 asteroids) => get ~77 FPS

Expected behavior Comparable FPS is expected for DirectX and Vulkan version for the same application with equal parameters.

Desktop (please complete the following information):

• OS: Windows 11
• Version 23H2
• HW: Intel Core i7-7820X, NVidia GeForce 2070 SUPER
• Build environment Visual Studio 2022

[egorodet]

Rootcause Multi-threaded rendering performance became limited by std::mutex lock in Vulkan::Program::GetNativePipelineLayout called from Vulkan::ProgramBindings::Apply, as seen in Tracy:

[egorodet]

Fixed in commit f57fcc84847a557de063e69698cbd2810130867f