Oclgrind 26.03.1

This release incorporates the following changes:

• Enable OpenCL 3.0 by default
• Added support for LLVM 18, 19, 20, 21, and 22
• Dropped support for all older versions of LLVM
• Bump C++ requirement to C++17

Note 1: A patch number has been introduced to the version number. This will allow for future releases that either fix minor bugs or add support for newer versions of LLVM, without bumping the major/minor version numbers (which are based on the release date).

Note 2: Precompiled header support is disabled for LLVM 21 and 22 due to a bug in those versions. This may impact OpenCL kernel compilation speed.

Oclgrind 21.10

This release incorporates the following changes:

• Added initial support for OpenCL 3.0
• Added support for LLVM 11, 12, and 13
• Dropped support for older versions of LLVM
• Various minor bug fixes

Oclgrind 19.10

This release incorporates the following changes:

• Added support for executing commands across multiple command queues
• Added support for out-of-order command queues
• Added support for LLVM 7.0, 8.0 and 9.0
• Dropped support for LLVM 3.9 and 4.0
• Various minor bug fixes

Oclgrind 18.3

Added support for the following OpenCL 2.0 features:

• Program-scope global variables
• Non-uniform work-group sizes
• Several miscellaneous API and kernel functions

Other changes:

• Switched to using Clang's builtin OpenCL header (opencl-c.h)
• Added support for LLVM 4.0, 5.0 and 6.0
• Dropped support for LLVM 3.6, 3.7 and 3.8
• Added --{global,constant,local}-mem-size and --max-wgsize options
• Removed autotools build system
• Improved support for big-endian systems
• Various minor bug fixes

Oclgrind 16.10

This release incorporates the following changes:

• Added plugin to detect use of uninitialized values (from Moritz Pflanzer)
• Added memoryMap and memoryUnmap plugin callbacks
• Added support for LLVM 3.7, 3.8, and 3.9
• Added oclgrind.exe command on Windows
• Report invalid uses of mapped buffers inside kernels
• Report invalid indices when accessing statically sized arrays
• Improved coverage of race detection plugin
• Fixed memcheck false-positive when writing to a write-only vector array
• Oclgrind will now appear with device type (CPU | GPU | ACCELERATOR | DEFAULT)
• Various minor bug fixes

Oclgrind 15.5

This release updates to LLVM 3.6, which improves the OpenCL C compiler and provides some additional performance enhancements. See README for revised instructions on how to build Oclgrind from source.

• Fixed race conditions in atomic operations
• Interactive debugger breaks on Ctrl+C
• Various other minor bug fixes

Oclgrind 15.2

This release significantly improves simulation performance, and fixes
several bugs impacting on usage and stability.

• Added detection for violations of read-only/write-only attributes
• Added --build-options argument to append additional compiler flags
• Added hostMemoryLoad and hostMemoryStore callbacks
• Added workGroupBegin and workItemBegin callbacks
• Split atomic callbacks into separate load and store
• Multi-threaded simulation to improve performance
• Various other performance improvements
• Several general bug fixes and stability improvements

Oclgrind 14.12

This release incorporates a new plugin system, to allow third-party
developers to build tools that utilise Oclgrind. More information can
be found on the Wiki:
https://github.com/jrprice/Oclgrind/wiki/Creating-Plugins

In addition, this release contains the following changes:

• Interactive debugger now has a command history
• Detection for unaligned memory accesses
• Limit the number of error messages printed to avoid flooding output
• Various other bug fixes and improvements

Oclgrind 14.5

Initial release (beta).

Implements a SPIR 1.2 interpreter which can be targeted either via an
OpenCL 1.2 runtime API implementation or using a standalone kernel
interface.

Provides the following utilities:

• Memory access error detection
• Work-group divergence detection (barriers, async-copies)
• Data-race detection (--data-races)
• Simple interactive debugger (--interactive)
• Instruction histograms (--inst-counts)
• OpenCL runtime API error reporting (--check-api)