Skip to content

Releases: JuliaGPU/GPUCompiler.jl

v0.21.3

29 Jun 13:30
c0c9ab7
Compare
Choose a tag to compare

GPUCompiler v0.21.3

Diff since v0.21.2

Merged pull requests:

  • Update manifest (#478) (@github-actions[bot])
  • Version the compilecache. (#481) (@maleadt)

v0.21.2

23 Jun 14:09
Compare
Choose a tag to compare

GPUCompiler v0.21.2

Diff since v0.21.1

v0.21.1

23 Jun 12:32
52b6b25
Compare
Choose a tag to compare

GPUCompiler v0.21.1

Diff since v0.21.0

Closed issues:

  • UndefVarError: compile_cache not defined (#470)
  • Adapt to specsig changes (#472)

Merged pull requests:

v0.21.0

13 Jun 08:31
d84aea2
Compare
Choose a tag to compare

GPUCompiler v0.21.0

Diff since v0.20.3

Closed issues:

  • GPUCompiler.compile replaces dots with underscores in extern functions (#464)

Merged pull requests:

  • Don't rename function declarations, only definitions. (#465) (@maleadt)
  • PTX: Lower unreachable control flow to avoid bad CFG reconstruction (#467) (@maleadt)
  • Update manifest (#468) (@github-actions[bot])

v0.20.3

01 Jun 17:30
87f0aeb
Compare
Choose a tag to compare

GPUCompiler v0.20.3

Diff since v0.20.2

Closed issues:

  • LLVM IR linking warning due to mismatching datalayout (#459)
  • PTX: LLVM assertion error during function merging (#461)

Merged pull requests:

  • Align GPUInterpreter and NativeInterpreter constructors. (#457) (@maleadt)
  • Perform library linking after deferred codegen. (#458) (@maleadt)
  • PTX: Don't use a 32-bit pointer index type. (#462) (@maleadt)

v0.20.2

22 May 13:34
5536c45
Compare
Choose a tag to compare

GPUCompiler v0.20.2

Diff since v0.20.1

Merged pull requests:

  • Improvements for Metal (#436) (@maleadt)
  • Port some intrinsic lowering code from LLVM/C++. (#452) (@maleadt)
  • Simplify interfaces (#453) (@maleadt)
  • Metal: Inline functions to avoid unreachable function attributes. (#454) (@maleadt)
  • Metal: Don't use deprecated LLVM.jl functionality. (#455) (@maleadt)

v0.20.1

16 May 16:07
87f1818
Compare
Choose a tag to compare

GPUCompiler v0.20.1

Diff since v0.20.0

Merged pull requests:

v0.20.0

15 May 18:36
3de799a
Compare
Choose a tag to compare

GPUCompiler v0.20.0

Diff since v0.19.4

What's Changed

  • Rework cached compilation; remove invalidation generator by @maleadt in #445

Full Changelog: v0.19.4...v0.20.0

v0.19.4

15 May 15:05
8385e56
Compare
Choose a tag to compare

GPUCompiler v0.19.4

Diff since v0.19.3

Closed issues:

  • llvmcall attributes are lost (#437)
  • Handle (Type, MI) for invoke in backedge invalidation (#442)

Merged pull requests:

  • Save and restore function attributes across optimization. (#438) (@maleadt)
  • Update manifest (#439) (@github-actions[bot])
  • Update manifest (#441) (@github-actions[bot])
  • Clarify invoke backedge invalidation. (#443) (@maleadt)
  • PTX: Default to 32-bit indexing of pointers. (#444) (@maleadt)
  • Update for LLVM 15. (#446) (@maleadt)
  • Refuse opaque closures. (#447) (@maleadt)
  • Optimize the execution macro helper functionality. (#448) (@maleadt)
  • Update manifest (#449) (@github-actions[bot])

v0.19.3

19 Apr 15:35
d5086fb
Compare
Choose a tag to compare

GPUCompiler v0.19.3

Diff since v0.19.2

Closed issues:

  • Installation fails on Mac M1 with v0.19.{1, 2} (#430)
  • min_enabled_level is at risk of invalidations (#431)
  • Failed to precompile (#433)

Merged pull requests:

  • Prevent invalidations of safe logging functionality by custom loggers. (#432) (@maleadt)
  • Only use CachedMethodTable when available. (#434) (@maleadt)