v1.12.0

GPUCompiler v1.12.0

Diff since v1.11.1

Merged pull requests:

• Bump julia-actions/setup-julia from 2 to 3 (#786) (@dependabot[bot])
• PTX: Add support for selecting the GPU feature set. (#798) (@maleadt)

v1.11.1

GPUCompiler v1.11.1

Diff since v1.11.0

Merged pull requests:

• Revert "Fix always_inline via inlining policy override" (#797) (@maleadt)

Closed issues:

• Pass registration broken on recent release? (#789)

v1.11.0

GPUCompiler v1.11.0

Diff since v1.10.0

Merged pull requests:

• Turn instcombine into a flag (and use instsimplify otherwise) (#791) (@wsmoses)
• Adjust to upcoming sparam_vals representation change (#792) (@Keno)
• Consider assertions for runtime module (#793) (@vchuravy)
• Fix always_inline via inlining policy override (#795) (@maleadt)
• Adapt to LLVM.jl change. (#796) (@maleadt)

Closed issues:

• always_inline is broken on 1.11 (#527)

v1.10.0

GPUCompiler v1.10.0

Diff since v1.9.1

Merged pull requests:

• [SPIRV] BFloat16 support (#778) (@michel2323)
• Bump codecov/codecov-action from 5 to 6 (#779) (@dependabot[bot])
• Bump julia-actions/cache from 2 to 3 (#780) (@dependabot[bot])
• Metal: Configure TTI so that InferAddressSpaces works. (#782) (@maleadt)
• Minor compiler perf improvements (#783) (@maleadt)
• Deprecate 3-arg link_libraries! in favor of lazy loading/linking (#784) (@maleadt)
• Use invokelatest for deferred_codegen (#787) (@vchuravy)
• Revert "Enable Ahead-of-Time Compilation by hiding the runtime functions in the GLOBAL_METHOD_TABLE" (#788) (@maleadt)

Closed issues:

• Precompilation broken on 1.13/nightly (#748)
• Nightly codegen hang (#777)
• Metal kernel reads zeros from constant globals with dynamic indexing (#781)

v1.9.1

GPUCompiler v1.9.1

Diff since v1.9.0

v1.9.0

GPUCompiler v1.9.0

Diff since v1.8.2

Merged pull requests:

• Enable Ahead-of-Time Compilation by hiding the runtime functions in the GLOBAL_METHOD_TABLE (#749) (@apozharski)
• [NFC] Typos (#763) (@christiangnrd)
• Switch to FileCheck.jl. (#764) (@maleadt)
• [Metal] Add attributes for some simd permute intrinsics (#766) (@simsurace)
• Fix some tests (#767) (@christiangnrd)
• Move to ParallelTestRunner v2 (#769) (@christiangnrd)
• Fixup test filtering (#770) (@christiangnrd)
• Fix nightly (#771) (@christiangnrd)
• GCN: use byref instead of byval+lower_byval for kernel arguments (#772) (@gbaraldi)
• Default to the job's optimize parameter in code_llvm (#773) (@christiangnrd)
• Remove deprecation warning in native test (#774) (@christiangnrd)
• Adjust broken test predicate so 1.13 test pass (#775) (@christiangnrd)
• Derive the deferred world from the parent one. (#776) (@maleadt)

Closed issues:

• Dwarf warnings with Julia 1.13 (#765)

v1.8.2

GPUCompiler v1.8.2

Diff since v1.8.1

Merged pull requests:

• Revert "Allow semi-concrete interpretation" (#760) (@christiangnrd)

Closed issues:

• Metal.jl sincos intrinsic produces invalid LLVM IR on Julia 1.11+ (GPUCompiler 1.7.6+) (#761)

v1.8.1

GPUCompiler v1.8.1

Diff since v1.8.0

Merged pull requests:

• Fix typed pointer gv initialization (#758) (@wsmoses)

v1.8.0

GPUCompiler v1.8.0

Diff since v1.7.6

Feature changes

• Add new meta field gv_to_value that allows mapping GlobalVariables to their Julia Values for downstream consumers #752

Closed issues:

• LLVM IR linking warning due to mismatching Dwarf Version type (#392)

v1.7.6

GPUCompiler v1.7.6

Diff since v1.7.5

Merged pull requests:

• [Metal] Add some simdgroup and barrier intrinsic attributes (#744) (@christiangnrd)
• Bump actions/checkout from 5 to 6 (#747) (@dependabot[bot])
• Allow semi-concrete interpretation (#750) (@maleadt)
• Enable users to map from GV to Julia value and initialize them whenever possible (#752) (@wsmoses)
• Fix codeinstance mapping when using nested interpretation (#755) (@wsmoses)
• Fixes for 1.13 and 1.14 (#757) (@maleadt)

Closed issues:

• Re-enable semi-concrete evaluation (#382)