[DeviceMSAN] Enable origin tracking

clang/lib/Driver/SanitizerArgs.cpp

@@ -1317,6 +1317,12 @@ void SanitizerArgs::addArgs(const ToolChain &TC, const llvm::opt::ArgList &Args,
 
       CmdArgs.push_back("-mllvm");
       CmdArgs.push_back("-msan-poison-stack-with-call=1");
+
+      if (MsanTrackOrigins) {
+        CmdArgs.push_back("-mllvm");
+        CmdArgs.push_back(Args.MakeArgString("-msan-track-origins=" +
+                                             Twine(MsanTrackOrigins)));
+      }
     } else if (Sanitizers.has(SanitizerKind::Thread)) {
       CmdArgs.push_back("-fsanitize=thread");
       // The tsan function entry/exit builtins are used to record stack

libdevice/atomic.hpp

@@ -75,6 +75,9 @@ extern DEVICE_EXTERNAL void __spirv_AtomicStore(int *, int, int, int) noexcept;
 extern DEVICE_EXTERNAL int __spirv_AtomicIAdd(SPIR_GLOBAL int *, int, int,
                                               int) noexcept;
 
+extern DEVICE_EXTERNAL int __spirv_AtomicIAdd(SPIR_GLOBAL unsigned int *, int,
+                                              int, int) noexcept;
+
 /// Atomically set the value in *Ptr with Desired if and only if it is Expected
 /// Return the value which already was in *Ptr
 static inline int atomicCompareAndSet(SPIR_GLOBAL int *Ptr, int Desired,
@@ -113,4 +116,10 @@ static inline int atomicAdd(SPIR_GLOBAL int *Ptr, int V) {
                             V);
 }
 
+static inline int atomicAdd(SPIR_GLOBAL unsigned int *Ptr, int V) {
+  return __spirv_AtomicIAdd(Ptr, __spv::Scope::Device,
+                            __spv::MemorySemanticsMask::SequentiallyConsistent,
+                            V);
+}
+
 #endif // __SPIR__ || __SPIRV__

libdevice/cmake/modules/SYCLLibdevice.cmake

@@ -246,7 +246,7 @@ if (NOT MSVC AND UR_SANITIZER_INCLUDE_DIR)
     include/asan_rtl.hpp
     include/sanitizer_defs.hpp
     include/spir_global_var.hpp
-    include/group_utils.hpp
+    include/sanitizer_utils.hpp
     ${sycl-compiler_deps})
 
   set(sanitizer_generic_compile_opts ${compile_opts}
@@ -305,7 +305,7 @@ if (NOT MSVC AND UR_SANITIZER_INCLUDE_DIR)
     include/msan_rtl.hpp
     include/sanitizer_defs.hpp
     include/spir_global_var.hpp
-    include/group_utils.hpp
+    include/sanitizer_utils.hpp
     sycl-compiler)
 
   set(tsan_obj_deps
@@ -314,7 +314,7 @@ if (NOT MSVC AND UR_SANITIZER_INCLUDE_DIR)
     include/tsan_rtl.hpp
     include/sanitizer_defs.hpp
     include/spir_global_var.hpp
-    include/group_utils.hpp
+    include/sanitizer_utils.hpp
     sycl-compiler)
 endif()
 

libdevice/include/asan_rtl.hpp

@@ -8,6 +8,7 @@
 #pragma once
 
 #include "sanitizer_defs.hpp"
+#include "sanitizer_utils.hpp"
 #include "spir_global_var.hpp"
 #include <cstdint>
 

libdevice/include/msan_rtl.hpp

@@ -8,6 +8,7 @@
 #pragma once
 
 #include "sanitizer_defs.hpp"
+#include "sanitizer_utils.hpp"
 #include "spir_global_var.hpp"
 
 // Treat this header as system one to workaround frontend's restriction

libdevice/include/sanitizer_defs.hpp

@@ -8,7 +8,6 @@
 #pragma once
 
 #include "atomic.hpp"
-#include "group_utils.hpp"
 #include "spir_global_var.hpp"
 #include <cstdint>
 
@@ -30,6 +29,11 @@ enum ADDRESS_SPACE : uint32_t {
   ADDRESS_SPACE_GENERIC = 4,
 };
 
+#define __SYCL_GLOBAL__ __attribute__((opencl_global))
+#define __SYCL_LOCAL__ __attribute__((opencl_local))
+#define __SYCL_PRIVATE__ __attribute__((opencl_private))
+#define __SYCL_CONSTANT__ __attribute__((opencl_constant))
+
 #define LIKELY(x) __builtin_expect(!!(x), 1)
 #define UNLIKELY(x) __builtin_expect(!!(x), 0)
 #define NORETURN __declspec(noreturn)
@@ -56,14 +60,4 @@ __spirv_ControlBarrier(int32_t Execution, int32_t Memory,
 
 extern "C" SYCL_EXTERNAL void __devicelib_exit();
 
-__SYCL_GLOBAL__ void *ToGlobal(void *ptr) {
-  return __spirv_GenericCastToPtrExplicit_ToGlobal(ptr, 5);
-}
-__SYCL_LOCAL__ void *ToLocal(void *ptr) {
-  return __spirv_GenericCastToPtrExplicit_ToLocal(ptr, 4);
-}
-__SYCL_PRIVATE__ void *ToPrivate(void *ptr) {
-  return __spirv_GenericCastToPtrExplicit_ToPrivate(ptr, 7);
-}
-
 #endif // __SPIR__ || __SPIRV__

libdevice/include/group_utils.hpp → libdevice/include/sanitizer_utils.hpp

@@ -7,11 +7,12 @@
 //===----------------------------------------------------------------------===//
 #pragma once
 
+#include "sanitizer_defs.hpp"
 #include "spirv_vars.h"
 
 #if defined(__SPIR__) || defined(__SPIRV__)
 
-static inline size_t WorkGroupLinearId() {
+inline size_t WorkGroupLinearId() {
   return __spirv_BuiltInWorkgroupId.x * __spirv_BuiltInNumWorkgroups.y *
              __spirv_BuiltInNumWorkgroups.z +
          __spirv_BuiltInWorkgroupId.y * __spirv_BuiltInNumWorkgroups.z +
@@ -26,15 +27,54 @@ static inline size_t LocalLinearId() {
 }
 
 // For GPU device, each sub group is a hardware thread
-static inline size_t SubGroupLinearId() {
+inline size_t SubGroupLinearId() {
   return __spirv_BuiltInGlobalLinearId / __spirv_BuiltInSubgroupSize;
 }
 
-static inline void SubGroupBarrier() {
+inline void SubGroupBarrier() {
   __spirv_ControlBarrier(__spv::Scope::Subgroup, __spv::Scope::Subgroup,
                          __spv::MemorySemanticsMask::SequentiallyConsistent |
                              __spv::MemorySemanticsMask::CrossWorkgroupMemory |
                              __spv::MemorySemanticsMask::WorkgroupMemory);
 }
 
+inline __SYCL_GLOBAL__ void *ToGlobal(void *ptr) {
+  return __spirv_GenericCastToPtrExplicit_ToGlobal(ptr, 5);
+}
+inline __SYCL_LOCAL__ void *ToLocal(void *ptr) {
+  return __spirv_GenericCastToPtrExplicit_ToLocal(ptr, 4);
+}
+inline __SYCL_PRIVATE__ void *ToPrivate(void *ptr) {
+  return __spirv_GenericCastToPtrExplicit_ToPrivate(ptr, 7);
+}
+
+template <typename T> T Memset(T ptr, int value, size_t size) {
+  for (size_t i = 0; i < size; i++) {
+    ptr[i] = value;
+  }
+  return ptr;
+}
+
+template <typename DstT, typename SrcT>
+DstT Memcpy(DstT dst, SrcT src, size_t size) {
+  for (size_t i = 0; i < size; i++) {
+    dst[i] = src[i];
+  }
+  return dst;
+}
+
+template <typename DstT, typename SrcT>
+DstT Memmove(DstT dst, SrcT src, size_t size) {
+  if ((uptr)dst < (uptr)src) {
+    for (size_t i = 0; i < size; i++) {
+      dst[i] = src[i];
+    }
+  } else {
+    for (size_t i = size; i > 0; i--) {
+      dst[i - 1] = src[i - 1];
+    }
+  }
+  return dst;
+}
+
 #endif // __SPIR__ || __SPIRV__

libdevice/include/spir_global_var.hpp

@@ -38,11 +38,6 @@ class
   T val;
 };
 
-#define __SYCL_GLOBAL__ __attribute__((opencl_global))
-#define __SYCL_LOCAL__ __attribute__((opencl_local))
-#define __SYCL_PRIVATE__ __attribute__((opencl_private))
-#define __SYCL_CONSTANT__ __attribute__((opencl_constant))
-
 #ifndef SPIR_GLOBAL_VAR
 #ifdef __SYCL_DEVICE_ONLY__
 #define SPIR_GLOBAL_VAR __attribute__((sycl_global_var))

libdevice/include/tsan_rtl.hpp

@@ -8,6 +8,7 @@
 #pragma once
 
 #include "sanitizer_defs.hpp"
+#include "sanitizer_utils.hpp"
 #include "spir_global_var.hpp"
 #include "tsan/tsan_libdevice.hpp"