[aakanksha555]

Copy the launch_bounds Catch tests from rocm-systems. Only launch_bounds.cc is built; the other sources are compiler and hiprtc inputs. Link hiprtc and exclude nothing, matching upstream on amd_linux.

Ports unit/launchBounds (__launch_bounds__ attribute) from projects/hip-tests in ROCm/rocm-systems at 851d4028cb.

1. Copy the sources verbatim, without upstream's CMakeLists.txt. launch_bounds.cc is the test, and launch_bounds_negative_kernels_rtc.hh is hiprtc input for its _RTC case. launch_bounds_compiler_error_kernels.cc and launch_bounds_parse_error_kernels.cc are currently unused, because the upstream compile-error checks that consume them are commented out.
2. Add an entry to HipCatchTestLists.cmake:
   • SOURCES launch_bounds.cc, so the kernel input files aren't built as tests.
   • LINK_LIBS hiprtc, matching upstream's LINKER_LIBS.
   • No EXCLUDE_TESTS: upstream disables two cases only for amd_windows and amd_wsl.

The directory tree in CATCH_TESTS_README.md is updated.

Tested on gfx1030 with ROCm 10.1.0: check-hip-catch-unit-launchBounds runs 3 test cases, all passing.

Assisted-by: Claude Opus 5.5

[kasaurov]

Looks good, thanks! The vendored files are byte-identical to rocm-systems (unchanged since our framework sync point), and SOURCES, LINK_LIBS hiprtc and the empty EXCLUDE_TESTS all match upstream: the YAML only disables two cases for amd_windows. Two small nits:

• launch_bounds_compiler_error_kernels.cc and launch_bounds_parse_error_kernels.cc are only used by upstream's compile-error checks, which are commented out in upstream's CMakeLists.txt (stale with clang 25+), so the README comment could say they're currently unused rather than "compiler input".
• Could you mention in the description what you tested it on (GPU and ROCm version)?