gdb.rocm: replace hipcc with amdclang++ as the HIP compiler

[spatrang]

Motivation

The GDB testsuite uses hipcc for HIP compiler discovery, naming, and
documentation. hipcc is deprecated and will be removed at some point, so this
switches the testsuite over to amdclang++.

What changes

Compiler discovery (lib/future.exp)

• Look for amdclang++ under $ROCM_PATH/lib/llvm/bin instead of hipcc.
• Rename HIPCC_FOR_TARGET to HIP_COMPILER_FOR_TARGET (the environment
  override).
• Rename gdb_find_hipcc / find_hipcc to gdb_find_hip_compiler /
  find_hip_compiler.

Compile and link flags (lib/gdb.exp, boards/hip.exp)

• amdclang++ does not infer the input language from the file extension, so
  tag each input explicitly: -x hip before C/C++/HIP sources and -x none
  before object files/archives. A single invocation can then handle a mix of
  sources and objects (e.g. set_unbuffered_mode.o / hip-driver.o injected
  via ldflags).
• Pass --hip-link only when producing an executable.
• Pass --rocm-path only when ROCM_PATH is set; otherwise defer to
  amdclang++'s own HIP discovery.
• Drop the explicit -O0 (amdclang++ already defaults to -O0).
• amdclang++ is clang-based and accepts -fdiagnostics-color, so stop
  skipping it for HIP; this keeps color escape sequences out of gdb.log.

Testcases

• Rename allow_hipcc_tests to allow_hip_tests in lib/rocm.exp,
  boards/hip.exp, and all gdb.rocm/ and gdb.perf/ testcases.
• Replace hipcc's --genco with amdclang++'s --cuda-device-only in the
  two code-object tests (code-object-load-while-breakpoint-hit.exp,
  snapshot-objfile-on-load.exp).

Documentation (gdb/doc/gdb.texinfo)

• Update the example compile command to amdclang++ -x hip --hip-link ...,
  keeping the note about the equivalent hipcc invocation.

[Copilot]

Pull request overview

This PR completes the ROCm/HIP testsuite migration from hipcc to amdclang++ by updating compiler discovery, build flag handling, naming, and test/doc references to align with the current ROCm toolchain.

Changes:

• Switch HIP compiler discovery to amdclang++ (with HIP_COMPILER_FOR_TARGET override and HIPCC_FOR_TARGET fallback) and keep backward-compatible aliases.
• Update HIP compilation/linking flags to correctly apply -x hip only when compiling sources and --hip-link only when producing executables.
• Rename testsuite gating helpers (allow_hipcc_tests → allow_hip_tests) and update testcases and documentation accordingly (with compatibility aliases retained).

Reviewed changes

Copilot reviewed 79 out of 79 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
gdb/testsuite/lib/rocm.exp	Renames HIP gating proc to allow_hip_tests and keeps allow_hipcc_tests as a compatibility alias.
gdb/testsuite/lib/gdb.exp	Updates HIP compile/link flag handling for amdclang++ (-x hip / --hip-link) and related comments.
gdb/testsuite/lib/future.exp	Switches HIP compiler discovery to amdclang++, adds HIP_COMPILER_FOR_TARGET, and introduces new/compat proc names.
gdb/testsuite/gdb.rocm/watchpoint-basic.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/watchpoint-at-end-of-shader.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/watch-gpu-global-from-host.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/update-thread-list.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/until-tests.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/unaligned-memory-access.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/step-schedlock-spurious-waves.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/step-over-kernel-exit.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/static-global.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/simple.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/simple-outside-debugger.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/show-info.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/shared-memory.exp	Updates require gate to allow_hip_tests (and keeps python requirement).
gdb/testsuite/gdb.rocm/scheduler-locking.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/runtime-core.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/resume-exception.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/register-watchpoint.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/program-execution.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/precise-memory.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/precise-memory-warning-watchpoint.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/precise-memory-warning-sigsegv.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/precise-memory-multi-inferiors.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/precise-memory-fork.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/precise-memory-exec.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/ocp_mx.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/nonstop-mode.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/nonstop-displaced.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/names.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/multi-inferior-run-spurious-waves.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/multi-inferior-gpu.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/multi-inferior-fork.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/mi-lanes.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/mi-attach.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/mi-aspace.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/load-core-remote-system.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/line-breakpoint-in-kernel.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/lane-pc-vega20.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/lane-info.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/lane-execution.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/instruction-stepping-commands.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/info-sharedlibrary.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/info-dispatches.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/illegal-insn-sigill.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/hip-builtin-variables.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/hip-builtin-completions.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/generic-address.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/gcore-after-attach.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/fork-exec-non-gpu-to-gpu.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/fork-exec-gpu-to-non-gpu.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/finish.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/exec-bit-extract.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/displaced-stepping.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/disassemble.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/devicecode-breakpoint.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/device-interrupt.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/device-barrier.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/device-attach.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/detach-while-breakpoints-inserted.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/deref-scoped-pointer.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/deep-stack.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/debugtrap.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/corefile.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/core-no-read-special-files.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/convenience_variables.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/continue-over-kernel-exit.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/breakpoint-after-exit.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/break-kernel-no-debug-info.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/branch-fault.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/bit-extract.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/aspace-watchpoint.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/aspace-user-input.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/alu-exceptions.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.rocm/addr-bp-gpu-no-deb-info.exp	Updates require gate to allow_hip_tests.
gdb/testsuite/gdb.perf/rocm-break-cond-false.exp	Updates require gate to allow_hip_tests for the perf testcase.
gdb/testsuite/boards/hip.exp	Updates HIP board config to use find_hip_compiler, renames allow proc, and adjusts amdclang++ detection.
gdb/doc/gdb.texinfo	Updates documentation example command line to amdclang++ -x hip --hip-link ....

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 81 out of 81 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[lumachad]

Thanks! Looks mostly good to me. Made some comments about small adjustments.

[lancesix]

I only skimmed thought the patch at this point. Note that some of what you are touching comes from upstream, and we might want to land the changes there as well. I'd probably advise for having this changed upstream first, then we can import it into amd-staging.

[lumachad]

I agree with upstreaming. We should keep the PR open for reviews and validation. Once we're happy with it, we can take it upstream, wait for approval and pick this up again, then close the PR.

[palves]

Another point is making sure this works on the Windows HIP SDK too. The compiler tool name and paths may be different there IIRC.

The boards/hip.exp changes need to be tested with --target_board=hip. A smoke test of running a few testcases like for example gdb.base/break.exp, instead of the full testsuite, is probably enough.

[palves]

The boards/hip.exp changes need to be tested with --target_board=hip. A smoke test of running a few testcases like for example gdb.base/break.exp, instead of the full testsuite, is probably enough.

Please make sure you test this. I tried it now with:

$ make check RUNTESTFLAGS="--target_board=hip" TESTS="gdb.base/break.exp"

and that fails with:

gdb compile failed, /home/pedro/rocm/gdb/build/gdb/testsuite/../../../src/gdb/testsuite/lib/hip/hip-test-wrapper.cc:24:10: fatal error: 
      'hip/hip_runtime.h' file not found
   24 | #include <hip/hip_runtime.h>
      |          ^~~~~~~~~~~~~~~~~~~
1 error generated.

                === gdb Summary ===

# of untested testcases         1

Without your patch, it compiles OK. (It then fails for other unrelated reasons.)

[spatrang]

The boards/hip.exp changes need to be tested with --target_board=hip. A smoke test of running a few testcases like for example gdb.base/break.exp, instead of the full testsuite, is probably enough.

Please make sure you test this. I tried it now with:

$ make check RUNTESTFLAGS="--target_board=hip" TESTS="gdb.base/break.exp"

and that fails with:

gdb compile failed, /home/pedro/rocm/gdb/build/gdb/testsuite/../../../src/gdb/testsuite/lib/hip/hip-test-wrapper.cc:24:10: fatal error: 
      'hip/hip_runtime.h' file not found
   24 | #include <hip/hip_runtime.h>
      |          ^~~~~~~~~~~~~~~~~~~
1 error generated.

                === gdb Summary ===

# of untested testcases         1

Without your patch, it compiles OK. (It then fails for other unrelated reasons.)

Fixed. The --target_board=hip path broke because amdclang++ (unlike hipcc) doesn't automatically discover the HIP headers -- it needs --rocm-path to locate them. Added --rocm-path=$rocm_path to the board's compilation flags, along with the -x hip / --hip-link split (compile vs link). Tested with make check RUNTESTFLAGS="--target_board=hip" TESTS="gdb.base/break.exp" -- compilation succeeds, zero hip_runtime.h errors.

[spatrang]

The boards/hip.exp changes need to be tested with --target_board=hip. A smoke test of running a few testcases like for example gdb.base/break.exp, instead of the full testsuite, is probably enough.

Please make sure you test this. I tried it now with:

$ make check RUNTESTFLAGS="--target_board=hip" TESTS="gdb.base/break.exp"

and that fails with:

gdb compile failed, /home/pedro/rocm/gdb/build/gdb/testsuite/../../../src/gdb/testsuite/lib/hip/hip-test-wrapper.cc:24:10: fatal error: 
      'hip/hip_runtime.h' file not found
   24 | #include <hip/hip_runtime.h>
      |          ^~~~~~~~~~~~~~~~~~~
1 error generated.

                === gdb Summary ===

# of untested testcases         1

Without your patch, it compiles OK. (It then fails for other unrelated reasons.)

@palves
Implemented your suggestion -- for HIP, gdb_compile now pre-compiles each .cpp/.c source to its own .o before the final link step, mirroring the pattern boards/hip.exp already uses. In the unbuffered-mode handler, we now:

Loop over $source and recursively call gdb_compile to produce a .o for each source file (with -x hip, since they're sources)
Append set_unbuffered_mode.o to the now all-.o source list
Strip -x hip from early_flags for the final link (since we only have .o inputs)
Verified with your hack (if { 1 || ...): before the fix, set_unbuffered_mode-hip.o:1:1: error: expected unqualified-id reproduced exactly as you reported. After the fix, the final link becomes amdclang++ --hip-link -O0 ... device_enumerator.cpp.hip.o set_unbuffered_mode-hip.o ... -o device_enumerator.x (only .o inputs, no -x hip) and succeeds

[palves]

Fixed. The --target_board=hip path broke because amdclang++ (unlike hipcc) doesn't automatically discover the HIP headers -- it needs --rocm-path to locate them. Added --rocm-path=$rocm_path to the board's compilation flags, along with the -x hip / --hip-link split (compile vs link). Tested with make check RUNTESTFLAGS="--target_board=hip" TESTS="gdb.base/break.exp" -- compilation succeeds, zero hip_runtime.h errors.

I'm confused. When running the normal gdb.rocm/ tests, I don't see "--rocm-path" passed to amdclang++. Why is it needed here, and not there?

[Copilot]

Pull request overview

Copilot reviewed 81 out of 81 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[lumachad]

Bringing this up here. @amd-shahab was asking questions on finding llvm-dwarfdump on a CI system. I think we should take the opportunity to make the tools search order correct in these changes (or a separate patch if desired).

See my previous comment here: #62 (comment).

I think we should honor PATH first. Then if ROCM_PATH is set, we go look for tools there. Then if everything fails, we search some hardcoded /opt/rocm path.

[palves]

Bringing this up here. @amd-shahab was asking questions on finding llvm-dwarfdump on a CI system. I think we should take the opportunity to make the tools search order correct in these changes (or a separate patch if desired).

See my previous comment here: #62 (comment).

I think we should honor PATH first. Then if ROCM_PATH is set, we go look for tools there. Then if everything fails, we search some hardcoded /opt/rocm path.

I agree we should look into this, though I'd prefer if any changes to the search path order was done in a separate patch. We have a little mess here, with search orders different for different tools. E.g., "${rocm_path}/llvm/bin/clang-offload-bundler" in rocm.exp (which now misses the leading lib/), Shahab's patch that searches PATH first, the searches for all the different tools lib/future.exp which search PATH last.

[palves]

I also removed the hipcc deprecation / equivalent-command block from gdb.texinfo

Well, I think that change went too far. I had only asked to remove the comments parts. hipcc still exists and is widely used by users, so I think that part of the manual should have stayed.

[palves]

What is that "--genco" commit about? If you do fixup commits, please make sure to add some rationale in their commit logs. That was needed because of these two tests, I guess:

testsuite/gdb.rocm/code-object-load-while-breakpoint-hit.exp: { debug hip additional_flags=--genco additional_flags=-DDEVICE } ] != "" } {
testsuite/gdb.rocm/snapshot-objfile-on-load.exp: { debug hip additional_flags=--genco } ] != "" } {

But why the mapping instead of just fixing the tests?

[spatrang]

Well, I think that change went too far. I had only asked to remove the comments parts. hipcc still exists and is widely used by users, so I think that part of the manual should have stayed.

Agreed, that was overreach on my part. I've restored the hipcc section in gdb.texinfo (the deprecation note plus the equivalent hipcc command); only the legacy-hipcc comparisons in code comments are dropped now.

What is that "--genco" commit about? If you do fixup commits, please make sure to add some rationale in their commit logs. [...] But why the mapping instead of just fixing the tests?

Good point on both counts. I've dropped the gdb_compile mapping and instead changed the two tests directly to use --cuda-device-only (the amdclang++ equivalent of hipcc's --genco):

• gdb.rocm/code-object-load-while-breakpoint-hit.exp
• gdb.rocm/snapshot-objfile-on-load.exp

I also folded everything back into the single commit (no more fixup commit), and the commit log now documents the --genco -> --cuda-device-only change.

[palves]

The patch itself LGTM now. Just a few comments on the commit log:

• The first paragraph has a few nits to pick:

The GDB testsuite used hipcc for HIP compiler discovery, naming, and
documentation, even though the underlying compiler is amdclang++.
Complete the switch to amdclang++ throughout the testsuite.

"The GDB testsuite used hipcc" => "uses hipcc"

The "the underlying compiler is amdclang++" isn't what motivates this. The real reason is that hipcc is deprecated and is going away at some point.

And then "complete the switch" implies the switch had been started already. It hasn't, this patch starts and finishes it.

• It doesn't mention the -fdiagnostics-color change in the "Compile and link flags" section. It should mention that amdclang++ is clang-based and accepts the flag.

• "Add HIP_COMPILER_FOR_TARGET" is not strictly correct, it's more rename "HIPCC_FOR_TARGET to HIP_COMPILER_FOR_TARGET".

• FYI, the PR text and the commit log are slightly out of sync, which is a bit confusing and made me wonder which one is up to date. Since with "squash and merge" the PR text doesn't make it to git (the git log history is the canonical one), then feel free to simplify the PR text to give the short intro and then say to look at the individual commit log for details. Alternatively, it's fine to have the info in both places, though I'd suggest using the gh tool to easily and automatically sync them.

[lancesix]

Sorry I have not had much time to look at this PR. I expect a fair amount of this can / should be upstream, right? Having this only downstream would cause some divergence we'll need to fix soon.

[palves]

Sorry I have not had much time to look at this PR. I expect a fair amount of this can / should be upstream, right? Having this only downstream would cause some divergence we'll need to fix soon.

Agreed, you had already mentioned this earlier, even. :-)

One issue is that upstream does not have any of the Windows bits in the testsuite (which needed adjustment in this patch), and it has fewer tests to adjust. This is why I kept pushing for finishing up the patch downstream first.

I'd vote for landing the patch downstream first, and then also send it upstream shortly after. The upstream version will end up slightly simplified in some parts.

[lancesix]

Agreed, you had already mentioned this earlier, even. :-)

oups, forgot I already mentioned that 😅

I'd vote for landing the patch downstream first, and then also send it upstream shortly after. The upstream version will end up slightly simplified in some parts.

cool, works for me.

[spatrang]

Just a few comments on the commit log

Thanks - addressed all of these in the latest force-push:

• Reworded the opening paragraph: "uses hipcc", motivation is now that hipcc is deprecated and going away, and dropped "complete the switch" since this patch both starts and finishes it.
• Added a note that amdclang++ is clang-based and accepts -fdiagnostics-color, so we no longer skip it for HIP (keeps color escapes out of gdb.log).
• Reworded to "rename HIPCC_FOR_TARGET to HIP_COMPILER_FOR_TARGET" instead of "add".
• Synced the PR description with the commit log.

[palves]

Thanks, LGTM. Let's merge this (with rebase and merge).

Next step is upstreaming this. Let me/us know if you'd like to work on that.

[lumachad]

Please don't merge. These tests are failing for both gcc and llvm:

2026-06-18 18:25:09,111 - INFO:        gdb.dwarf2/ada-thick-pointer.exp
2026-06-18 18:25:09,111 - INFO:        gdb.rocm/coop-group-grid-sync.exp

[lumachad]

For the group sync one...

ERROR: -------------------------------------------
ERROR: in testcase /__w/ROCgdb/ROCgdb/build/tests/rocgdb/gdb/testsuite/gdb.rocm/coop-group-grid-sync.exp
ERROR:  invalid command name "allow_hipcc_tests"
ERROR:  tcl error code TCL LOOKUP COMMAND allow_hipcc_tests
ERROR:  tcl error info:
invalid command name "allow_hipcc_tests"
    while executing
"::gdb_tcl_unknown allow_hipcc_tests"
    ("uplevel" body line 1)
    invoked from within
"uplevel 1 ::gdb_tcl_unknown $args"
    (procedure "::unknown" line 5)
    invoked from within
"allow_hipcc_tests"
    ("uplevel" body line 1)
    invoked from within
"uplevel 1 $fn"
    (procedure "require" line 11)
    invoked from within
"require allow_hipcc_tests"
    (file "/__w/ROCgdb/ROCgdb/build/tests/rocgdb/gdb/testsuite/gdb.rocm/coop-group-grid-sync.exp" line 34)
    invoked from within
"source /__w/ROCgdb/ROCgdb/build/tests/rocgdb/gdb/testsuite/gdb.rocm/coop-group-grid-sync.exp"
    ("uplevel" body line 1)
    invoked from within
"uplevel #0 source /__w/ROCgdb/ROCgdb/build/tests/rocgdb/gdb/testsuite/gdb.rocm/coop-group-grid-sync.exp"
    invoked from within
"catch "uplevel #0 source $test_file_name" msg"

[lumachad]

Don't think there's actually anything wrong with gdb.dwarf2/ada-thick-pointer.exp. It's just confusion from the CI script due to the group sync testcase errors trumpling over the gdb.dwarf2/ada-thick-pointer.exp testcase output.

[spatrang]

Rebased onto the latest amd-staging and force-pushed.

What changed since the last push:

• Rebased onto current amd-staging, which pulled in the new gdb.rocm/coop-group-grid-sync.exp test (and the supports_cooperative_groups helper it needs).
• Renamed that test's require allow_hipcc_tests to require allow_hip_tests, matching the rest of the testsuite. This was the actual cause of the red CI (invalid command name "allow_hipcc_tests"); gdb.dwarf2/ada-thick-pointer.exp was only collateral noise from that test's error trampling the log, so it clears too.

I built rocgdb from this branch and ran gdb.rocm/*.exp + gdb.dwarf2/*.exp via .github/scripts/test_rocgdb.py (GCC and LLVM) against that tree:

• All gdb.rocm tests pass (incl. coop-group-grid-sync, hip-lang-detect, interrupt-twice, and the code-object tests).
• The only remaining unexpected failures are in gdb.dwarf2 (debug-names-non-ascending-cu, macro-source-path, dw2-ranges). These are unrelated to this patch — it touches no gdb.dwarf2 or GDB source files — and reproduce on amd-staging independently.