tsan: avoid extra call indirection in unaligned access functions

[dvyukov]

Summary

Currently unaligned access functions are defined in tsan_interface.cpp
and do a real call to MemoryAccess. This means we have a real call
and no read/write constant propagation.

Unaligned memory access can be quite hot for some programs
(observed on some compression algorithms with ~90% of unaligned accesses).

Move them to tsan_interface_inl.h to avoid the additional call
and enable constant propagation.
Also reorder the actual store and memory access handling for
__sanitizer_unaligned_store callbacks to enable tail calling
in MemoryAccess.

Depends on D107282.

[vitalybuka]

compiler-rt/lib/tsan/rtl/tsan_interface_inl.h

112

D107319 should help with warnings

[dvyukov]

rebased

compiler-rt/lib/tsan/rtl/tsan_interface.cpp

void __tsan_write16_pc(void *addr, void *pc) {
   uptr pc_no_pac = STRIP_PAC_PC(pc);
   ThreadState *thr = cur_thread();
   MemoryAccess(thr, pc_no_pac, (uptr)addr, 8, kAccessWrite);
   MemoryAccess(thr, pc_no_pac, (uptr)addr + 8, 8, kAccessWrite);
 }
 
 // __tsan_unaligned_read/write calls are emitted by compiler.
 
-void __tsan_unaligned_read2(const void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, kAccessRead);
-}
-
-void __tsan_unaligned_read4(const void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 4, kAccessRead);
-}
-
-void __tsan_unaligned_read8(const void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 8, kAccessRead);
-}
-
 void __tsan_unaligned_read16(const void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 16, kAccessRead);
-}
-
-void __tsan_unaligned_write2(void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, kAccessWrite);
-}
-
-void __tsan_unaligned_write4(void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 4, kAccessWrite);
-}
-
-void __tsan_unaligned_write8(void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 8, kAccessWrite);
+  uptr pc = CALLERPC;
+  ThreadState *thr = cur_thread();
+  UnalignedMemoryAccess(thr, pc, (uptr)addr, 8, kAccessRead);
+  UnalignedMemoryAccess(thr, pc, (uptr)addr + 8, 8, kAccessRead);
 }
 
 void __tsan_unaligned_write16(void *addr) {
-  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 16, kAccessWrite);
+  uptr pc = CALLERPC;
+  ThreadState *thr = cur_thread();
+  UnalignedMemoryAccess(thr, pc, (uptr)addr, 8, kAccessWrite);
+  UnalignedMemoryAccess(thr, pc, (uptr)addr + 8, 8, kAccessWrite);
 }
 
-// __sanitizer_unaligned_load/store are for user instrumentation.
-
 extern "C" {
 SANITIZER_INTERFACE_ATTRIBUTE
-u16 __sanitizer_unaligned_load16(const uu16 *addr) {
-  __tsan_unaligned_read2(addr);
-  return *addr;
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-u32 __sanitizer_unaligned_load32(const uu32 *addr) {
-  __tsan_unaligned_read4(addr);
-  return *addr;
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-u64 __sanitizer_unaligned_load64(const uu64 *addr) {
-  __tsan_unaligned_read8(addr);
-  return *addr;
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void __sanitizer_unaligned_store16(uu16 *addr, u16 v) {
-  __tsan_unaligned_write2(addr);
-  *addr = v;
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void __sanitizer_unaligned_store32(uu32 *addr, u32 v) {
-  __tsan_unaligned_write4(addr);
-  *addr = v;
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void __sanitizer_unaligned_store64(uu64 *addr, u64 v) {
-  __tsan_unaligned_write8(addr);
-  *addr = v;
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
 void *__tsan_get_current_fiber() {
   return cur_thread();
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE
 void *__tsan_create_fiber(unsigned flags) {
   return FiberCreate(cur_thread(), CALLERPC, flags);
 }

compiler-rt/lib/tsan/rtl/tsan_interface_inl.h

 void __tsan_write4_pc(void *addr, void *pc) {
   MemoryAccess(cur_thread(), STRIP_PAC_PC(pc), (uptr)addr, 4, kAccessWrite);
 }
 
 void __tsan_write8_pc(void *addr, void *pc) {
   MemoryAccess(cur_thread(), STRIP_PAC_PC(pc), (uptr)addr, 8, kAccessWrite);
 }
 
+ALWAYS_INLINE USED void __tsan_unaligned_read2(const void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, kAccessRead);
+}
+
+ALWAYS_INLINE USED void __tsan_unaligned_read4(const void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 4, kAccessRead);
+}
+
+ALWAYS_INLINE USED void __tsan_unaligned_read8(const void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 8, kAccessRead);
+}
+
+ALWAYS_INLINE USED void __tsan_unaligned_write2(void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, kAccessWrite);
+}
+
+ALWAYS_INLINE USED void __tsan_unaligned_write4(void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 4, kAccessWrite);
+}
+
+ALWAYS_INLINE USED void __tsan_unaligned_write8(void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 8, kAccessWrite);
+}
+
+extern "C" {
+// __sanitizer_unaligned_load/store are for user instrumentation.
+SANITIZER_INTERFACE_ATTRIBUTE
+u16 __sanitizer_unaligned_load16(const uu16 *addr) {
+  __tsan_unaligned_read2(addr);
+  return *addr;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+u32 __sanitizer_unaligned_load32(const uu32 *addr) {
+  __tsan_unaligned_read4(addr);
+  return *addr;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+u64 __sanitizer_unaligned_load64(const uu64 *addr) {
+  __tsan_unaligned_read8(addr);
+  return *addr;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __sanitizer_unaligned_store16(uu16 *addr, u16 v) {
+  *addr = v;
+  __tsan_unaligned_write2(addr);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __sanitizer_unaligned_store32(uu32 *addr, u32 v) {
+  *addr = v;
+  __tsan_unaligned_write4(addr);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __sanitizer_unaligned_store64(uu64 *addr, u64 v) {
+  *addr = v;
+  __tsan_unaligned_write8(addr);
+}
+}
+
 void __tsan_vptr_update(void **vptr_p, void *new_val) {
   if (*vptr_p == new_val)
     return;
   MemoryAccess(cur_thread(), CALLERPC, (uptr)vptr_p, sizeof(*vptr_p),
                kAccessWrite | kAccessVptr);
 }
 
 void __tsan_vptr_read(void **vptr_p) {