AMDGPU.jl

AMDGPU v0.3.3

Diff since v0.3.2

Closed issues:

• Addition of AMDGPU.functional() (equivalent to CUDA.functional()) (#198)

Merged pull requests:

• utils: Add functional query (#200) (@jpsamaroo)

AMDGPU v0.3.2

Diff since v0.3.1

Merged pull requests:

• Define a map from gamma to tgamma (#196) (@vchuravy)

AMDGPU v0.3.1

Diff since v0.3.0

Merged pull requests:

• Function overrides lead to undefined functions in KA (#194) (@michel2323)

AMDGPU v0.3.0

Diff since v0.2.17

Closed issues:

• Embed/wrap lld linker (#27)
• Document math intrinsics usage (#17)
• Track kernel dependencies in ROCArray (#48)
• Error in build step: Inconsistency detected by ld.so (#68)
• Use Array{UInt8}(undef, len) instead of repeat(" ", len) (#132)
• Add method overrides (#135)
• CUDA vs AMDGPU (#137)
• Problems on 4750U mobile APU (#142)
• Unable to build AMDGPU.jl with roc* libraries (#154)
• wavefrontsize64_{on|off}.bc missing with JULIA_AMDGPU_DISABLE_ARTIFACTS=true (#172)
• Segfault on array or kernel operation (#178)
• Julia 1.7.0 breaks (#184)

Merged pull requests:

• at-roc: Support threads and blocks kwargs (#167) (@jpsamaroo)
• Add errorhandler for async queue errors (#174) (@vchuravy)
• rocprintf: Avoid reading bad data (#176) (@jpsamaroo)
• bump GPUCompiler to 0.13 (#181) (@vchuravy)
• Test multiple GPUArrays versions (#182) (@jpsamaroo)
• Propagate device to roccall (#186) (@jpsamaroo)
• LB Julia to 1.7, upgrade to ROCm 4.2 (#187) (@jpsamaroo)
• Detect ROCM_PATH environment variable (#188) (@michel2323)
• Add new DOCUMENTER_KEY (#189) (@jpsamaroo)

AMDGPU v0.2.17

Diff since v0.2.16

Closed issues:

• No warning when we fail to find ld.lld, and general build observability (#179)

Merged pull requests:

• Add support for GPUArrays 8 (#163) (@vchuravy)
• run GPUArrays test-suite (#173) (@vchuravy)
• Make build failure logging more verbose (#180) (@jpsamaroo)

AMDGPU v0.2.16

Diff since v0.2.15

Closed issues:

• No write to non-contiguous views (#140)

Merged pull requests:

• Specialize Dataype for broadcasting (#162) (@vchuravy)
• Revert "Specialize Dataype for broadcasting" (#168) (@vchuravy)
• Specialize Datatype for broadcasting -- Take 2 (#169) (@vchuravy)
• Support complex abs (#170) (@vchuravy)
• Fix norm (#171) (@vchuravy)

AMDGPU v0.2.15

Diff since v0.2.14

Closed issues:

• Generation specific device libraries (#164)
• Array + Diagonal failure (#165)

Merged pull requests:

• Make CI green again (#91) (@jpsamaroo)
• Fix logic error in gridDimWG (#166) (@vchuravy)

AMDGPU v0.2.14

Diff since v0.2.13

Merged pull requests:

• remove activate from deps/build.jl (#160) (@vchuravy)

AMDGPU v0.2.13

Diff since v0.2.12

Closed issues:

• Document JULIA_AMDGPU_DISABLE_ARTIFACTS (#155)
• has_rocm_gpu: a utility function that returns true if a GPU is available? (#158)

Merged pull requests:

• Add warning in quick start addressing #154. (#157) (@ali-ramadhan)
• Add has_rocm_gpu util function (#159) (@ali-ramadhan)

AMDGPU v0.2.12

Diff since v0.2.11

Closed issues:

• Missing active_kernels (#139)
• ROCArray : LoadError: UndefRefError: access to undefined reference (#153)

Merged pull requests:

• Add support for active_kernels (#141) (@michel2323)

AMDGPU v0.2.11

Diff since v0.2.10

Merged pull requests:

• GPU workshop changes! (#152) (@jpsamaroo)

AMDGPU v0.2.10

Diff since v0.2.9

Merged pull requests:

• Use HIP_jll and ROCmDeviceLibs_jll (#148) (@jpsamaroo)

AMDGPU v0.2.9

Diff since v0.2.8

Merged pull requests:

• Update to LLVM.jl 4 (#146) (@jpsamaroo)

AMDGPU v0.2.8

Diff since v0.2.7

Merged pull requests:

• fixes for delayed codegeneration (#124) (@vchuravy)
• Support GPUCompiler 0.12 (#145) (@vchuravy)

AMDGPU v0.2.7

Diff since v0.2.6

Closed issues:

• "Spills" from adjacent views of ROCVector (#130)

Merged pull requests:

• Fix copyto! on view; include offset (#131) (@tkf)
• Add support for features (#136) (@vchuravy)

AMDGPU v0.2.6

Diff since v0.2.5

Closed issues:

• ROCArrays matrix multiplication not working (#103)
• Data race in kernel packet writing? (#121)

Merged pull requests:

• Add mark/wait synchronization system (#116) (@jpsamaroo)
• CompatHelper: bump compat for "GPUCompiler" to "0.11" (#122) (@github-actions[bot])
• Replace arrays with Refs in ccall. (#123) (@chriselrod)
• Fix packet launch (#125) (@jpsamaroo)

AMDGPU v0.2.5

Diff since v0.2.4

Merged pull requests:

• Allow AMDGPU to load everywhere (#120) (@jpsamaroo)

AMDGPU v0.2.4

Diff since v0.2.3

Closed issues:

• Implement execution contexts (#16)
• Add/test broadcasting support to ROCArray (#12)
• Add queue/device/system sync functionality (#24)
• Support OpenCL.jl as device runtime (#23)
• Distribute ROCR/ROCT via artifacts (#6)
• Allow setting Private and Group segment sizes manually (#56)
• FATAL ERROR: Symbol "ccalllib_libhsa-runtime64445"not found on AMDGPU (#73)
• test failures and crashes on 580 (#92)
• Tests allocate memory indefinitely (#106)
• Check for invalid workgroup sizes (#110)
• Add example for gridsize usage and workgroup sizing (#113)

Merged pull requests:

• Silence queue destroy finalizer errors (#98) (@jpsamaroo)
• KernelAbstractions support (#100) (@jpsamaroo)
• Fix kernel argument alignment (#101) (@jpsamaroo)
• Fix alloc_local (#102) (@jpsamaroo)
• Add mapreducedim! (#104) (@jpsamaroo)
• Added support for HSA artifacts (#105) (@0x0f0f0f)
• Update to GPUCompiler 0.10 (#107) (@jpsamaroo)
• Fix CompatHelper config (#108) (@jpsamaroo)
• CompatHelper: bump compat for "AbstractFFTs" to "1.0" (#109) (@github-actions[bot])
• Fix loading of AMDGPU in other packages (#114) (@jpsamaroo)
• Check groupsize, add group/grid docs (#115) (@jpsamaroo)
• Fix memory leak (#118) (@jpsamaroo)

AMDGPU v0.2.3

Diff since v0.2.2

Closed issues:

• Add support for trap handlers (#8)
• Unreachable reached in SIISelLowering.cpp due to unhandled AS (#76)
• Ensure that CI tests all available external libraries (#85)
• Only load libhsa-runtime64 major version 1 (#93)

Merged pull requests:

• Disable unreliable pointerinfo tests (#86) (@jpsamaroo)
• Test extlibs are available under CI (#87) (@jpsamaroo)
• Enable device stacktraces (#88) (@jpsamaroo)
• Add ForwardDiff integrations (#90) (@jpsamaroo)
• Look for libhsa-runtime64.so.1 (#94) (@jpsamaroo)
• Remove annoying target-features for now (#95) (@jpsamaroo)
• Bump to 0.2.3, add MacroTools LB (#96) (@jpsamaroo)