Skip to content

Highlights:

  • Added support for D3D12 compute in a new mode of the command line tool (-s dx12) :
    • Compile D3D12 compute pipelines through the command line tool to generate:
      • GCN/RDNA ISA disassembly
      • HW resource usage statistics
      • Live register analysis
      • Control flow graphs
    • Target a variety of devices, regardless of the ones physically installed on the system.
    • Both HLSL source code and binary DXBC/DXIL are supported as an input.
    • More accurate disassembly and hardware resource usage statistics, as the new mode leverages the live driver and uses root signatures.
    • For more information about the new D3D12 compute mode, visit this page.
  • Added support for the Navi architecture in the following modes: Vulkan, Vulkan offline, DX12, DX11 and OpenGL.

Note: HLSL mode (-s hlsl) was renamed DX11 (-s dx11).

Assets 7

@AmitBM AmitBM released this Mar 15, 2019 · 10 commits to master since this release

Optimizing your Vulkan® code for GCN targets is easier than ever with RGA 2.1:

  • Added support for Vulkan in the graphical UI application.
  • ISA disassembly and HW resource usage information are now generated using the installed driver's compiler, which matches closely what you will see in your application. You can still target a variety of GPUs, independent of the GPU that is physically installed on your system.
  • Compile SPIR-V™ binaries, GLSL source code or mix thereof in a single pipeline.
  • Edit SPIR-V binaries in the app. SPIR-V binaries get disassembled and reassembled for you automatically.
  • Edit the Vulkan pipeline state for your shaders in the UI.
  • Force the compilation to go through an older or newer driver (amdvlk64) by using the Custom ICD feature.
  • Get detailed information about build errors by enabling the Vulkan validation layers.
  • Plug-and-play your own package of glslang, disassembler and assembler to be used in the compilation process.
  • Beta feature: serialize shaders and pipeline state from your Vulkan application using the RGA layer.
  • Supports both Windows® (7, 10) and Linux® (Ubuntu 18.04, RHEL 7).
  • Runs on machines without AMD hardware or driver. The 2.1 package contains a copy of AMD's Vulkan driver to be used on machines without a proper configuration.

This release also provides new features for OpenGL® in the command line tool:

  • Support for Vega targets in OpenGL mode.
  • Generate AMDIL disassembly for OpenGL shaders.

In addition to the above, gfx906 (Radeon VII) support was added to all modes.

For more details, see the RGA-2.1 Release Notes PDF document.

Assets 7

@AmitBM AmitBM released this Aug 17, 2018 · 16 commits to master since this release

Highlights:

  • Added support for pre-Vega GPUs in ROCm OpenCL mode
  • Updated the default LLVM-based Lightning Compiler package that ships with RGA
  • You can now extend the target GPUs list when using the Lightning Compiler plug & play feature. For more information about this feature, see the “Alternative ROCm OpenCL compiler” section in the help manual for rocm-cl mode (–s rocm-cl –h).
  • Updated the Vulkan offline backend
  • Bug fixes
Assets 8

@AmitBM AmitBM released this Mar 19, 2018 · 30 commits to master since this release

Analyzing and optimizing OpenCL™ code has never been as easy as it is with the first Radeon GPU Analyzer GUI app, an offline compiler and code analysis tool for OpenCL™ on top of AMD’s LLVM-based Lightning Compiler, the OpenCL™ compiler for the ROCm platform.

The main features of this release are:

  • Graphical user interface for rapid analysis and optimization
  • Build OpenCL™ source files into a HSA Code Object, using AMD’s LLVM-based Lightning Compiler
  • Generate GCN ISA disassembly for the Vega architecture (gfx900, gfx902)
  • Fine-tune optimization and build settings
  • Correlation of high-level source code to GCN ISA disassembly
  • Identify areas in your code with high memory pressure through disassembly color-coding
  • Be notified about potential performance hazards in your code due to GPU resource overuse
  • Plug & Play your own LLVM-based compiler

Source-to-disassembly correlation in RGA GUI app

GUI app's build view

Notification about potential performance hazards in the GUI app

GUI app's resource usage view

For more details, see the RGA-2.0 Release Notes PDF document.

Assets 6

@AmitBM AmitBM released this Dec 22, 2017 · 33 commits to master since this release

Highlights:

Ready to ROCK: introducing the new “rocm-cl” mode that supports OpenCL™ on top of AMD’s LLVM-based Lightning Compiler, the OpenCL™ compiler for the ROCm platform:

  • Compile and link one or more OpenCL™ source files into a HSA Code Object.
  • Generate disassembly for the Vega architecture:
    o GCN ISA disassembly
    o Mapping high-level source code to GCN ISA disassembly (--line-numbers option)
    o LLVM IR disassembly (--il option)
  • Extract the Code Object meta-data (--metadata option).
  • Gather static HW resource usage: SGPR and VGPR consumption, LDS usage, and more.
  • Perform live register analysis at the assembly level.
  • Generate control-flow graphs at the assembly level.
  • Cross-platform, no need for a ROCm-capable machine.

For more details, see the RGA-1.4 Release Notes PDF document.

Assets 6

@AmitBM AmitBM released this Nov 2, 2017 · 34 commits to master since this release

This release provides bug fixes and backend improvements:

  • Fixed an issue which prevented live VGPR analysis reports from being generated in SPIR-V mode (#14).
  • Fixed an issue that caused RGA to generate a live VGPR analysis report even though --livereg was not mentioned in the invocation command (DX11 mode).
  • Fixed an issue that caused RGA to output incorrect error messages in SPIR-V and OpenGL modes (this issue is related to #15).
  • Vulkan SPIR-V backend: improvements and stabilization.

For more details, see the RGA-1.3 Release Notes PDF document.

Assets 6

@AmitBM AmitBM released this Aug 17, 2017 · 38 commits to master since this release

Highlights:

  • You can now use public GPU names, and RGA will detect the relevant hardware architecture.
  • Added support for multiple-GPU system configurations in the DirectX mode.
  • SPIR-V modes (both binary and textual) can now accept a single input file. RGA will deduce the present pipeline stages automatically.
  • Added support for more SPIR-V extensions.
  • Building RGA: added CMake post-build scripts that copy dependency binaries.

For more details, see the RGA-1.2 Release Notes PDF document.

Assets 7

@AmitBM AmitBM released this Jun 27, 2017 · 43 commits to master since this release

This is a new version of Radeon GPU Analyzer (RGA).

Highlights:

  • Support for the Vega architecture.
  • Support for static analysis of VOP2 and VOP3 instruction groups.
  • Beta feature: support for SPIR-V (both in binary and textual forms).
  • OpenGL mode now works on Linux through the amdgpu-pro driver.
  • Finding your relevant target device name is easier with this version: the "-l" and "--list-asics" options now provide a mapping between the GPU codename, architecture name and the names of different GPU variants.
  • The Vulkan modes (GLSL, SPIR-V binary and SPIR-V textual) are now independent of the graphics hardware and the installed drivers, and can work on any x86-based system.
  • DirectX, OpenGL and Vulkan modes are no longer dependent on the OpenCL runtime.

Notes:

  • Windows users can use either the installer or the zip archive. For Linux, a tgz archive is provided.
  • Some features require the latest Radeon Software version. More details can be found in the Release Notes file.
Assets 6

@AmitBM AmitBM released this Feb 3, 2017 · 58 commits to master since this release

This is the first release of the Radeon GPU Analyzer (RGA) command line tool.
Windows users can use either the installer or the zip archive.
For Linux, a tgz archive is provided.

Assets 6
You can’t perform that action at this time.