v1.7.5

GPUCompiler v1.7.5

Diff since v1.7.4

Merged pull requests:

• version bound LowerExcHandlersPass (#736) (@vchuravy)
• More robust error message on method compilation failure (#738) (@wsmoses)
• Add preference to enable Tracy and disable by default (#745) (@vchuravy)

Closed issues:

• LowerExcHandlersPass not defined in GPUCompiler (#735)
• Versions newer than 1.3.2 make Julia/CUDA/Flux slow (#742)

v1.7.4

GPUCompiler v1.7.4

Diff since v1.7.3

Merged pull requests:

• [CI] set 10 minute timeout for GH actions jobs (#739) (@simeonschaub)
• fix GPUInterpreter(interp; ...) constructor (#740) (@simeonschaub)
• use CC.MAX_INLINE_COST instead of typemax(Int) (#741) (@simeonschaub)

v1.7.3

GPUCompiler v1.7.3

Diff since v1.7.2

Merged pull requests:

• Add option for backends to run vectorization passes (#716) (@gbaraldi)
• Update Test workflows for 1.12 (#729) (@kshyatt)
• Enable Metal testing on 1.12. (#730) (@maleadt)
• Use ParallelTestRunner.jl (#731) (@vchuravy)
• Bump peter-evans/find-comment from 3 to 4 (#732) (@dependabot[bot])
• Bump peter-evans/create-or-update-comment from 4 to 5 (#733) (@dependabot[bot])
• [SPIRV] convert i128 allocas to <2 x i64> (#734) (@simeonschaub)

v1.7.2

GPUCompiler v1.7.2

Diff since v1.7.1

v1.7.1

GPUCompiler v1.7.1

Diff since v1.7.0

Merged pull requests:

• New interface: process_linked_module! (#727) (@maleadt)

v1.7.0

GPUCompiler v1.7.0

Diff since v1.6.3

Merged pull requests:

• Convert the kernel state back to a reference when needed. (#715) (@maleadt)

Closed issues:

• Call to codeinfo_for_const broken on nightly (#721)

v1.6.3

GPUCompiler v1.6.3

Diff since v1.6.2

Merged pull requests:

• Bump actions/checkout from 4 to 5 (#724) (@dependabot[bot])
• Fixes for Julia nightly (#725) (@KristofferC)

v1.6.2

GPUCompiler v1.6.2

Diff since v1.6.1

Merged pull requests:

• Fix deferred_codegen registration (#711) (@vchuravy)
• Simplify byval handling. (#714) (@maleadt)
• add_input_arguments! for other backends (#718) (@simeonschaub)

Closed issues:

• InitError(mod=:GPUcompiler, error=ErrorException("type Nothing has no field major" )) (#719)

v1.6.1

GPUCompiler v1.6.1

Diff since v1.6.0

Merged pull requests:

• fix erase! call in src/gcn.jl (#706) (@simeonschaub)

v1.6.0

GPUCompiler v1.6.0

Diff since v1.5.3

Feature changes

• Add StackedMethodTable to allow backends to share method tables. backends should implement GPUCompiler.method_table_view
• Support nested target and params for advanced nested compilation.

Merged pull requests:

• Allow for generic MethodTableView (#494) (@vchuravy)
• Allow for nested targets (#696) (@vchuravy)
• Allow the use of FileCheck (#703) (@vchuravy)