AMDGPU v2.7.3 Diff since v2.7.2 Merged pull requests: Thread safety fixes (#1016) (@maleadt) Missing module qualifier for rdiv (#1017) (@kshyatt) Bump LLVM_jll in / (#1018) (@dependabot[bot]) Remove dead LLVM_jll deps (#1019) (@luraess) Fix sysimage compilation fail (#1022) (@doraemonho) Add triangular loop regression test (#1026) (@luraess) add regression test for #931 (#1029) (@simeonschaub) CSCS MI300 CI: select the ROCm provider explicitly (#1031) (@luraess) Fix generic_matmul handling of size zero arrays (#1032) (@kshyatt) MIOpen: fix some issues found running tests on fedora (#1034) (@simeonschaub) Improve the artifact machinery (#1035) (@luraess) Narrow the trsm data operand to ROC arrays. (#1036) (@maleadt) Closed issues: conj seems to compile incorrectly on 1.10 (#931) Memory access fault on gfx90a/gfx942 (upstream LLVM AMDGPU miscompile) (#1015) @atomic kernel fails with hipErrorNoBinaryForGpu under --check-bounds=no (ROCm 7, Julia 1.12.6, v2.7.2) (#1020) AMDGPU.jl causes native compilation to fail when building a sysimage (#1021) [AMDGPU_LLVM_Backend] Backport llvm (#1025)