ROCm · ronlieb · Apr 14, 2026 · Apr 13, 2026
diff --git a/compiler-rt/CMakeLists.txt b/compiler-rt/CMakeLists.txt
@@ -362,6 +362,16 @@ if("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "hexagon")
   list(APPEND SANITIZER_COMMON_CFLAGS -fno-emulated-tls)
   string(APPEND COMPILER_RT_TEST_COMPILER_CFLAGS " -fno-emulated-tls")
 endif()
+if("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "amdgcn|nvptx")
+  append_list_if(UBSAN_COMMON_SUPPORTED_ARCH -ffreestanding SANITIZER_COMMON_CFLAGS)
+  append_list_if(COMPILER_RT_HAS_NOGPULIB_FLAG -nogpulib SANITIZER_COMMON_CFLAGS)
+  append_list_if(COMPILER_RT_HAS_FLTO_FLAG -flto SANITIZER_COMMON_CFLAGS)
+  if("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "amdgcn")
+    append_list_if(COMPILER_RT_HAS_CODE_OBJECT_VERSION_FLAG
+                   "SHELL:-Xclang -mcode-object-version=none" SANITIZER_COMMON_CFLAGS)
+  endif()
+endif()
+
 if(NOT WIN32)
   append_list_if(COMPILER_RT_HAS_FPIC_FLAG -fPIC SANITIZER_COMMON_CFLAGS)
 endif()
@@ -377,7 +387,9 @@ append_list_if(COMPILER_RT_HAS_FVISIBILITY_HIDDEN_FLAG -fvisibility=hidden SANIT
 if(NOT COMPILER_RT_HAS_FVISIBILITY_HIDDEN_FLAG)
   append_list_if(COMPILER_RT_HAS_FVISIBILITY_INLINES_HIDDEN_FLAG -fvisibility-inlines-hidden SANITIZER_COMMON_CFLAGS)
 endif()
-append_list_if(COMPILER_RT_HAS_FNO_LTO_FLAG -fno-lto SANITIZER_COMMON_CFLAGS)
+if(NOT "${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "amdgcn|nvptx")
+  append_list_if(COMPILER_RT_HAS_FNO_LTO_FLAG -fno-lto SANITIZER_COMMON_CFLAGS)
+endif()
 
 # By default do not instrument or use profdata for compiler-rt.
 if(NOT COMPILER_RT_ENABLE_PGO)
@@ -479,7 +491,7 @@ else()
   set(SANITIZER_LIMIT_FRAME_SIZE FALSE)
 endif()
 
-if(FUCHSIA OR UNIX)
+if(NOT COMPILER_RT_GPU_BUILD AND (FUCHSIA OR UNIX))
   set(SANITIZER_USE_SYMBOLS TRUE)
 else()
   set(SANITIZER_USE_SYMBOLS FALSE)
@@ -520,20 +532,20 @@ elseif(COMPILER_RT_HAS_G_FLAG)
   list(APPEND SANITIZER_COMMON_CFLAGS -g)
 endif()
 
-if(SANITIZER_AMDGPU)
-  list(APPEND SANITIZER_COMMON_CFLAGS -DSANITIZER_AMDGPU=1)
+if(SANITIZER_AMDHSA)
+  list(APPEND SANITIZER_COMMON_CFLAGS -DSANITIZER_AMDHSA=1)
   message(STATUS "Looking 'hsa' and 'amd_comgr' header")
   find_path(HSA_INCLUDE NAMES hsa.h HINTS ${SANITIZER_HSA_INCLUDE_PATH} /opt/rocm/include PATH_SUFFIXES hsa)
   if(NOT HSA_INCLUDE)
-    message(FATAL_ERROR "Required header 'hsa.h' not found in path ${HSA_INCLUDE}. Aborting SANITIZER_AMDGPU build")
+    message(FATAL_ERROR "Required header 'hsa.h' not found in path ${HSA_INCLUDE}. Aborting SANITIZER_AMDHSA build")
   endif()
   message(STATUS "Found 'hsa.h' in ${HSA_INCLUDE}")
   include_directories(${HSA_INCLUDE})
   find_path(COMgr_INCLUDE NAMES amd_comgr.h.in HINTS ${SANITIZER_COMGR_INCLUDE_PATH} PATH_SUFFIXES amd_comgr)
   if(NOT COMgr_INCLUDE)
     find_path(COMgr_INCLUDE NAMES amd_comgr.h HINTS /opt/rocm/include PATH_SUFFIXES amd_comgr)
     if(NOT COMgr_INCLUDE)
-      message(FATAL_ERROR "Required header 'amd_comgr.h/amd_comgr.h.in' not found in path ${COMgr_INCLUDE}. Aborting SANITIZER_AMDGPU build")
+      message(FATAL_ERROR "Required header 'amd_comgr.h/amd_comgr.h.in' not found in path ${COMgr_INCLUDE}. Aborting SANITIZER_AMDHSA build")
     endif()
   endif()
   message(STATUS "Found 'amd_comgr.h.in/amd_comgr.h' in ${COMgr_INCLUDE}")

diff --git a/compiler-rt/cmake/Modules/AddCompilerRT.cmake b/compiler-rt/cmake/Modules/AddCompilerRT.cmake
@@ -185,6 +185,12 @@ function(add_compiler_rt_runtime name type)
     set(NO_LTO_FLAGS "")
   endif()
 
+  # The GPU build does not support shared libraries, just suppress them here as
+  # there is no global config for this.
+  if(COMPILER_RT_GPU_BUILD AND type MATCHES "SHARED")
+    return()
+  endif()
+
   # By default do not instrument or use profdata for compiler-rt.
   set(NO_PGO_FLAGS "")
   if(NOT COMPILER_RT_ENABLE_PGO)

diff --git a/compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake b/compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake
@@ -103,7 +103,7 @@ endif()
 set(ALL_TYSAN_SUPPORTED_ARCH ${X86_64} ${ARM64} ${S390X})
 set(ALL_UBSAN_SUPPORTED_ARCH ${X86} ${X86_64} ${ARM32} ${ARM64} ${RISCV64}
     ${MIPS32} ${MIPS64} ${PPC64} ${S390X} ${SPARC} ${SPARCV9} ${HEXAGON}
-    ${LOONGARCH64})
+    ${LOONGARCH64} ${AMDGPU} ${NVPTX})
 if (OS_NAME MATCHES "FreeBSD")
   set(ALL_SAFESTACK_SUPPORTED_ARCH ${X86} ${X86_64} ${ARM64})
 else()

diff --git a/compiler-rt/cmake/caches/GPU.cmake b/compiler-rt/cmake/caches/GPU.cmake
@@ -5,7 +5,8 @@ set(COMPILER_RT_HAS_SAFESTACK OFF CACHE BOOL "")
 set(COMPILER_RT_BUILD_BUILTINS ON CACHE BOOL "")
 set(COMPILER_RT_BAREMETAL_BUILD ON CACHE BOOL "")
 set(COMPILER_RT_BUILD_CRT OFF CACHE BOOL "")
-set(COMPILER_RT_BUILD_SANITIZERS OFF CACHE BOOL "")
+set(COMPILER_RT_BUILD_SANITIZERS ON CACHE BOOL "")
+set(COMPILER_RT_SANITIZERS_TO_BUILD "ubsan_minimal" CACHE STRING "")
 set(COMPILER_RT_BUILD_XRAY OFF CACHE BOOL "")
 set(COMPILER_RT_BUILD_LIBFUZZER OFF CACHE BOOL "")
 set(COMPILER_RT_BUILD_PROFILE OFF CACHE BOOL "")

diff --git a/compiler-rt/cmake/config-ix.cmake b/compiler-rt/cmake/config-ix.cmake
@@ -702,7 +702,7 @@ else()
   filter_available_targets(LSAN_COMMON_SUPPORTED_ARCH
     ${SANITIZER_COMMON_SUPPORTED_ARCH})
   filter_available_targets(UBSAN_COMMON_SUPPORTED_ARCH
-    ${SANITIZER_COMMON_SUPPORTED_ARCH})
+    ${ALL_UBSAN_SUPPORTED_ARCH})
   filter_available_targets(ASAN_SUPPORTED_ARCH ${ALL_ASAN_SUPPORTED_ARCH})
   filter_available_targets(RTSAN_SUPPORTED_ARCH ${ALL_RTSAN_SUPPORTED_ARCH})
   filter_available_targets(FUZZER_SUPPORTED_ARCH ${ALL_FUZZER_SUPPORTED_ARCH})
@@ -888,8 +888,9 @@ else()
   set(COMPILER_RT_HAS_UBSAN FALSE)
 endif()
 
-if (COMPILER_RT_HAS_SANITIZER_COMMON AND UBSAN_SUPPORTED_ARCH AND
-    OS_NAME MATCHES "Linux|FreeBSD|NetBSD|Android|Darwin|SunOS")
+if (UBSAN_SUPPORTED_ARCH AND
+    (OS_NAME MATCHES "Linux|FreeBSD|NetBSD|Android|Darwin|SunOS" OR
+     "${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "amdgcn|nvptx"))
   set(COMPILER_RT_HAS_UBSAN_MINIMAL TRUE)
 else()
   set(COMPILER_RT_HAS_UBSAN_MINIMAL FALSE)

diff --git a/compiler-rt/lib/asan/asan_allocator.cpp b/compiler-rt/lib/asan/asan_allocator.cpp
@@ -395,7 +395,7 @@ struct Allocator {
 
   void InitLinkerInitialized(const AllocatorOptions &options) {
     SetAllocatorMayReturnNull(options.may_return_null);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     allocator.InitLinkerInitialized(options.release_to_os_interval_ms, 0, true);
 #else
     allocator.InitLinkerInitialized(options.release_to_os_interval_ms);
@@ -1469,7 +1469,7 @@ int __asan_update_allocation_context(void* addr) {
   return instance.UpdateAllocationStack((uptr)addr, &stack);
 }
 
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 DECLARE_REAL(hsa_status_t, hsa_init);
 DECLARE_REAL(hsa_status_t, hsa_amd_agents_allow_access, uint32_t num_agents,
   const hsa_agent_t *agents, const uint32_t *flags, const void *ptr)

diff --git a/compiler-rt/lib/asan/asan_allocator.h b/compiler-rt/lib/asan/asan_allocator.h
@@ -324,7 +324,7 @@ void AsanSoftRssLimitExceededCallback(bool exceeded);
 
 }  // namespace __asan
 
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 #include <hsa.h>
 #include <hsa_ext_amd.h>
 

diff --git a/compiler-rt/lib/asan/asan_errors.cpp b/compiler-rt/lib/asan/asan_errors.cpp
@@ -779,7 +779,7 @@ ErrorNonSelfAMDGPU::ErrorNonSelfAMDGPU(uptr *dev_callstack, u32 n_callstack,
 void ErrorNonSelfAMDGPU::PrintStack() {
   InternalScopedString source_location;
   source_location.AppendF("  #0 %p", (void *)callstack[0]);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
   source_location.Append(" in ");
   __sanitizer::AMDGPUCodeObjectSymbolizer symbolizer;
   symbolizer.Init(cb_loc.fd, cb_loc.offset, cb_loc.size);

diff --git a/compiler-rt/lib/asan/asan_globals.cpp b/compiler-rt/lib/asan/asan_globals.cpp
@@ -172,7 +172,7 @@ static u32 FindRegistrationSite(const Global *g) {
   return 0;
 }
 
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 static bool IsValidGlobal(const Global *g) {
   return
     *(u8 *)MEM_TO_SHADOW((uptr)g) == kAsanGlobalRedzoneMagic &&
@@ -190,7 +190,7 @@ int GetGlobalsForAddress(uptr addr, Global *globals, u32 *reg_sites,
   Lock lock(&mu_for_globals);
   int res = 0;
   for (const auto &l : list_of_all_globals) {
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     if (!IsValidGlobal(l.g))
       continue;
 #endif

diff --git a/compiler-rt/lib/asan/asan_interceptors.cpp b/compiler-rt/lib/asan/asan_interceptors.cpp
@@ -896,7 +896,7 @@ DEFINE_REAL(int, vfork, )
 DECLARE_EXTERN_INTERCEPTOR_AND_WRAPPER(int, vfork, )
 #  endif
 
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 void ENSURE_HSA_INITED();
 
 INTERCEPTOR(hsa_status_t, hsa_amd_memory_pool_allocate,
@@ -1170,7 +1170,7 @@ void InitializeAsanInterceptors() {
   ASAN_INTERCEPT_FUNC(vfork);
 #  endif
 
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
   InitializeAmdgpuInterceptors();
 #endif
 

diff --git a/compiler-rt/lib/asan/asan_report.cpp b/compiler-rt/lib/asan/asan_report.cpp
@@ -627,7 +627,7 @@ void ReportNonselfLeak(u64 alloc_pc, u64 alloc_size, int device_id,
 
       InternalScopedString source_location;
       source_location.AppendF("    #0 0x%llx", e->alloc_pc);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
       source_location.Append(" in ");
       __sanitizer::AMDGPUCodeObjectSymbolizer symbolizer;
       symbolizer.Init(e->fd, e->file_extent_start, e->file_extent_size);

diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_allocator.h b/compiler-rt/lib/sanitizer_common/sanitizer_allocator.h
@@ -24,7 +24,7 @@
 #include "sanitizer_procmaps.h"
 #include "sanitizer_type_traits.h"
 
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 #include <hsa.h>
 #include <hsa_ext_amd.h>
 #endif

diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_allocator_amdgpu.cpp b/compiler-rt/lib/sanitizer_common/sanitizer_allocator_amdgpu.cpp
@@ -9,7 +9,7 @@
 // Part of the Sanitizer Allocator.
 //
 //===----------------------------------------------------------------------===//
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 #  include <dlfcn.h>  // For dlsym
 #  include "sanitizer_allocator.h"
 #  include "sanitizer_atomic.h"
@@ -205,4 +205,4 @@ void AmdgpuMemFuncs::RegisterSystemEventHandlers() {
 
 uptr AmdgpuMemFuncs::GetPageSize() { return kPageSize_; }
 }  // namespace __sanitizer
-#endif  // SANITIZER_AMDGPU
+#endif  // SANITIZER_AMDHSA
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_allocator_amdgpu.h b/compiler-rt/lib/sanitizer_common/sanitizer_allocator_amdgpu.h
@@ -13,7 +13,7 @@
 #  error This file must be included inside sanitizer_allocator_device.h
 #endif
 
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 class AmdgpuMemFuncs {
  public:
   static bool Init();
@@ -45,4 +45,4 @@ struct AmdgpuAllocationInfo : public DeviceAllocationInfo {
   u32 flags;
   void *ptr;
 };
-#endif  // SANITIZER_AMDGPU
+#endif  // SANITIZER_AMDHSA
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_allocator_combined.h b/compiler-rt/lib/sanitizer_common/sanitizer_allocator_combined.h
@@ -28,7 +28,7 @@ class CombinedAllocator {
       LargeMmapAllocator<typename PrimaryAllocator::MapUnmapCallback,
                          LargeMmapAllocatorPtrArray,
                          typename PrimaryAllocator::AddressSpaceView>;
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
   using DeviceAllocator =
       DeviceAllocatorT<typename PrimaryAllocator::MapUnmapCallback>;
 #endif
@@ -38,7 +38,7 @@ class CombinedAllocator {
                              bool enable_device_allocator = false) {
     primary_.Init(release_to_os_interval_ms, heap_start);
     secondary_.InitLinkerInitialized();
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     device_.Init(enable_device_allocator, primary_.kMetadataSize);
 #endif
   }
@@ -48,7 +48,7 @@ class CombinedAllocator {
     stats_.Init();
     primary_.Init(release_to_os_interval_ms, heap_start);
     secondary_.Init();
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     device_.Init(enable_device_allocator, primary_.kMetadataSize);
 #endif
   }
@@ -78,7 +78,7 @@ class CombinedAllocator {
     // alignment without such requirement, and allocating 'size' would use
     // extraneous memory, so we employ 'original_size'.
     void *res;
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     if (da_info)
       res = device_.Allocate(&stats_, original_size, alignment, da_info);
     else
@@ -110,7 +110,7 @@ class CombinedAllocator {
       cache->Deallocate(&primary_, primary_.GetSizeClass(p), p);
     else if (secondary_.PointerIsMine(p))
       secondary_.Deallocate(&stats_, p);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     else if (device_.PointerIsMine(p))
       device_.Deallocate(&stats_, p);
 #endif
@@ -139,7 +139,7 @@ class CombinedAllocator {
       return true;
     if (secondary_.PointerIsMine(p))
       return true;
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     if (device_.PointerIsMine(p))
       return true;
 #endif
@@ -153,7 +153,7 @@ class CombinedAllocator {
       return primary_.GetMetaData(p);
     if (secondary_.PointerIsMine(p))
       return secondary_.GetMetaData(p);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     if (device_.PointerIsMine(p))
       return device_.GetMetaData(p);
 #endif
@@ -165,7 +165,7 @@ class CombinedAllocator {
       return primary_.GetBlockBegin(p);
     if (secondary_.PointerIsMine(p))
       return secondary_.GetBlockBegin(p);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     if (device_.PointerIsMine(p))
       return device_.GetBlockBegin(p);
 #endif
@@ -180,7 +180,7 @@ class CombinedAllocator {
       return primary_.GetBlockBegin(p);
     if ((beg = secondary_.GetBlockBeginFastLocked(p)))
       return beg;
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     if ((beg = device_.GetBlockBeginFastLocked(p)))
       return beg;
 #endif
@@ -192,7 +192,7 @@ class CombinedAllocator {
       return primary_.GetActuallyAllocatedSize(p);
     if (secondary_.PointerIsMine(p))
       return secondary_.GetActuallyAllocatedSize(p);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     if (device_.PointerIsMine(p))
       return device_.GetActuallyAllocatedSize(p);
 #endif
@@ -201,7 +201,7 @@ class CombinedAllocator {
 
   uptr TotalMemoryUsed() {
     return primary_.TotalMemoryUsed() + secondary_.TotalMemoryUsed()
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
       + device_.TotalMemoryUsed()
 #endif
       ;
@@ -228,15 +228,15 @@ class CombinedAllocator {
   void PrintStats() {
     primary_.PrintStats();
     secondary_.PrintStats();
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     device_.PrintStats();
 #endif
   }
 
   // ForceLock() and ForceUnlock() are needed to implement Darwin malloc zone
   // introspection API.
   void ForceLock() SANITIZER_NO_THREAD_SAFETY_ANALYSIS {
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     device_.ForceLock();
 #endif
     primary_.ForceLock();
@@ -246,7 +246,7 @@ class CombinedAllocator {
   void ForceUnlock() SANITIZER_NO_THREAD_SAFETY_ANALYSIS {
     secondary_.ForceUnlock();
     primary_.ForceUnlock();
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     device_.ForceUnlock();
 #endif
   }
@@ -256,15 +256,15 @@ class CombinedAllocator {
   void ForEachChunk(ForEachChunkCallback callback, void *arg) {
     primary_.ForEachChunk(callback, arg);
     secondary_.ForEachChunk(callback, arg);
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
     device_.ForEachChunk(callback, arg);
 #endif
   }
 
  private:
   PrimaryAllocator primary_;
   SecondaryAllocator secondary_;
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
   DeviceAllocator device_;
 #endif
   AllocatorGlobalStats stats_;

diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_allocator_device.h b/compiler-rt/lib/sanitizer_common/sanitizer_allocator_device.h
@@ -14,7 +14,7 @@
 #endif
 
 struct DeviceAllocationInfo;
-#if SANITIZER_AMDGPU
+#if SANITIZER_AMDHSA
 // Device memory allocation usually requires additional information, we can put
 // all the additional information into a data structure DeviceAllocationInfo.
 // This is only a parent structure since different vendors may require
@@ -348,4 +348,4 @@ class DeviceAllocatorT {
   } stats;
   mutable StaticSpinMutex mutex_;
 };
-#endif  // SANITIZER_AMDGPU
+#endif  // SANITIZER_AMDHSA