v1.5.1

GPUCompiler v1.5.1

Diff since v1.5.0

Make things work on Julia 1.13 (nightly).

Merged pull requests:

• Fixes for 1.13 (#693) (@maleadt)

v1.5.0

GPUCompiler v1.5.0

Diff since v1.4.1

Support for the non-unified LLVM SPIR-V Translator JLL.

Merged pull requests:

• Add support for the non-unified SPIRV Translator JLL. (#692) (@maleadt)
• Add OpenCL to CI + other improvements (#694) (@maleadt)

Closed issues:

• Broken with Julia 1.12-beta3 (#695)

v1.4.1

GPUCompiler v1.4.1

Diff since v1.4.0

Merged pull requests:

• Fix typo in invalid IR error message (#691) (@giordano)

v1.4.0

GPUCompiler v1.4.0

Diff since v1.3.2

Merged pull requests:

• Pass optimize kwarg from code_llvm to CompilerConfig. (#682) (@maleadt)
• Avoid closure in @safe_* macros (#686) (@danielwe)
• use Tracy instead of TimerOutputs (#688) (@vchuravy)
• Bump LLVM.jl. (#690) (@maleadt)

Closed issues:

• 1.12: inference cache CI failure (#646)
• Pygments now supports PTX and GCN (#655)
• Safe logging macros don't work in generated functions due to with_logger closure (#685)
• Julia 1.12 LLVM error: Instruction Combining did not reach a fixpoint after 1 iterations (#687)

v1.3.2

GPUCompiler v1.3.2

Diff since v1.3.1

Merged pull requests:

• Derive compile config for runtime correctly (#684) (@pxl-th)

Closed issues:

• fails to precompile on Julia nightly, the precompile workload throws an AssertionError (#667)

v1.3.1

GPUCompiler v1.3.1

Diff since v1.3.0

Merged pull requests:

• Adapt to latest binding change in 1.12 (#681) (@gbaraldi)
• also test with 1.12 nightly on CI (#683) (@KristofferC)

v1.3.0

GPUCompiler v1.3.0

Diff since v1.2.0

Merged pull requests:

• Move arguments to compile/codegen into the CompilerConfig struct (#668) (@maleadt)
• Fix Github Actions badge and add new GPU backends (#669) (@christiangnrd)
• Fix some edge cases for mangling (#672) (@Keluaa)
• Force emission of const return functions using new API (#677) (@gbaraldi)
• Prune unused constant expressions recursively. (#678) (@maleadt)

Closed issues:

• fails to precompile on 1.11 (#670)
• Wrong mangling for some function signatures (#671)
• Cannot use GPUCompiler on 1.12-nightly (#673)
• Support for older GPUs (#675)
• ERROR: old function still has uses (via a constant expr) (#680)

v1.2.0

GPUCompiler v1.2.0

Diff since v1.1.1

Merged pull requests:

• Use Pygments' highlighting for PTX and GCN (#659) (@Keluaa)
• Resolve binding MethodError directly on 1.12 (#664) (@vchuravy)
• Add support for the LLVM SPIR-V back-end. (#665) (@maleadt)

v1.1.1

GPUCompiler v1.1.1

Diff since v1.1.0

Merged pull requests:

• Revert back to our hand-rolled test runner. (#658) (@maleadt)
• Support code emission of nvptx on Apple (#660) (@wsmoses)
• [CI] Don't run on PRs from forks a job which requires accessing a secret (#661) (@giordano)
• Don't include codeconv files (#662) (@wsmoses)

v1.1.0

GPUCompiler v1.1.0

Diff since v1.0.1

• Validator: it is now permitted to pass fieldless objects to kernels, even if they are mutable (#650). This should make it possible to e.g. pass symbols.
• Metal: fix emission of global constants (#648)
• Compatibility with Julia 1.12 (#651, #656)

Merged pull requests:

• [Metal] Add correct addrspace to global constants (#648) (@tgymnich)
• Validator: Permit all fieldless objects. (#650) (@maleadt)
• Fixes for nightly (#651) (@maleadt)
• Avoid calling specialize_method again for typeinf_type inference (#653) (@wsmoses)
• Bump codecov/codecov-action from 4 to 5 (#654) (@dependabot[bot])
• Updates for Julia 1.12 (#656) (@maleadt)