Burndown list of HSA/ROCR issues for initial milestone

[benvanik]

There's a few issues with the HSA/ROCR/LLVM mix that we're running into or new features that we need. This issue tracks the filed issues so we can track what is still required.

ROCR

• Fixing HostQueue to obey the alignment requirement on calls to aligned_alloc. ROCm/ROCR-Runtime#276
• Fixing non-portable inline attribute on hsa_flag_* utilities. ROCm/ROCR-Runtime#254
• [Issue]: any dynamic library linking ROCR cannot be dlopen'ed multiple times due to RuntimeCleanup. ROCm/ROCR-Runtime#275
• rocr: Add WaitMultiple to core Signal ROCm/ROCR-Runtime#250
• [Feature]: add a hsa_amd_signal_wait_all API. ROCm/ROCR-Runtime#241
• Add API flag for whether an AQL queue should be allocated in device memory. ROCm/ROCR-Runtime#269
• [Feature]: allow 0/NULL signal handles for wait operations to be no-op. ROCm/ROCR-Runtime#252
• [Feature]: add a batched version of hsa_amd_profiling_convert_tick_to_system_domain. ROCm/ROCR-Runtime#243
• [Feature]: add a method for querying the AMDKFD_IOC_GET_CLOCK_COUNTERS values. ROCm/ROCR-Runtime#278
• Fixing SVM profiler QUEUE_RESTORE parsing. ROCm/ROCR-Runtime#283

LLVM

• AMDGPUTargetStreamer generates .kd symbols, breaking LTO requirement, may be discarded by --gc-sections llvm/llvm-project#119479

[benvanik]

The current workaround to llvm/llvm-project#119479 is:

diff --git a/lld/ELF/MarkLive.cpp b/lld/ELF/MarkLive.cpp
index b6c22884d917..08d9ecedf2cc 100644
--- a/lld/ELF/MarkLive.cpp
+++ b/lld/ELF/MarkLive.cpp
@@ -221,7 +221,7 @@ template <class ELFT> void MarkLive<ELFT>::run() {
   // Preserve externally-visible symbols if the symbols defined by this
   // file can interpose other ELF file's symbols at runtime.
   for (Symbol *sym : ctx.symtab->getSymbols())
-    if (sym->isExported && sym->partition == partition)
+    if (sym->includeInDynsym(ctx) && sym->partition == partition)
       markSymbol(sym);
 
   // If this isn't the main partition, that's all that we need to preserve.