diff --git a/compiler-rt/lib/tsan/CMakeLists.txt b/compiler-rt/lib/tsan/CMakeLists.txt
--- a/compiler-rt/lib/tsan/CMakeLists.txt
+++ b/compiler-rt/lib/tsan/CMakeLists.txt
@@ -51,6 +51,7 @@
   rtl/tsan_suppressions.cpp
   rtl/tsan_symbolize.cpp
   rtl/tsan_sync.cpp
+  rtl/tsan_vector_clock.cpp
   )
 
 set(TSAN_CXX_SOURCES
@@ -105,6 +106,7 @@
   rtl/tsan_sync.h
   rtl/tsan_trace.h
   rtl/tsan_update_shadow_word_inl.h
+  rtl/tsan_vector_clock.h
   )
 
 set(TSAN_RUNTIME_LIBRARIES)
diff --git a/compiler-rt/lib/tsan/rtl/tsan_defs.h b/compiler-rt/lib/tsan/rtl/tsan_defs.h
--- a/compiler-rt/lib/tsan/rtl/tsan_defs.h
+++ b/compiler-rt/lib/tsan/rtl/tsan_defs.h
@@ -18,6 +18,24 @@
 #include "sanitizer_common/sanitizer_mutex.h"
 #include "ubsan/ubsan_platform.h"
 
+#ifndef TSAN_VECTORIZE
+#  define TSAN_VECTORIZE __SSE4_2__
+#endif
+
+#if TSAN_VECTORIZE
+// <emmintrin.h> transitively includes <stdlib.h>,
+// and it's prohibited to include std headers into tsan runtime.
+// So we do this dirty trick.
+#  define _MM_MALLOC_H_INCLUDED
+#  define __MM_MALLOC_H
+#  include <emmintrin.h>
+#  include <smmintrin.h>
+#  define VECTOR_ALIGNED ALIGNED(16)
+typedef __m128i m128;
+#else
+#  define VECTOR_ALIGNED
+#endif
+
 // Setup defaults for compile definitions.
 #ifndef TSAN_NO_HISTORY
 # define TSAN_NO_HISTORY 0
@@ -33,6 +51,14 @@
 
 namespace __tsan {
 
+// Thread slot ID.
+enum class Sid : u8 {};
+constexpr uptr kThreadSlotCount = 256;
+
+// Abstract time unit, vector clock element.
+enum class Epoch : u16 {};
+constexpr Epoch kEpochZero = static_cast<Epoch>(0);
+
 const int kClkBits = 42;
 const unsigned kMaxTidReuse = (1 << (64 - kClkBits)) - 1;
 
diff --git a/compiler-rt/lib/tsan/rtl/tsan_rtl.cpp b/compiler-rt/lib/tsan/rtl/tsan_rtl.cpp
--- a/compiler-rt/lib/tsan/rtl/tsan_rtl.cpp
+++ b/compiler-rt/lib/tsan/rtl/tsan_rtl.cpp
@@ -28,16 +28,6 @@
 #include "tsan_symbolize.h"
 #include "ubsan/ubsan_init.h"
 
-#ifdef __SSE3__
-// <emmintrin.h> transitively includes <stdlib.h>,
-// and it's prohibited to include std headers into tsan runtime.
-// So we do this dirty trick.
-#define _MM_MALLOC_H_INCLUDED
-#define __MM_MALLOC_H
-#include <emmintrin.h>
-typedef __m128i m128;
-#endif
-
 volatile int __tsan_resumed = 0;
 
 extern "C" void __tsan_resume() {
@@ -779,10 +769,11 @@
   return false;
 }
 
-#if defined(__SSE3__)
-#define SHUF(v0, v1, i0, i1, i2, i3) _mm_castps_si128(_mm_shuffle_ps( \
-    _mm_castsi128_ps(v0), _mm_castsi128_ps(v1), \
-    (i0)*1 + (i1)*4 + (i2)*16 + (i3)*64))
+#if TSAN_VECTORIZE
+#  define SHUF(v0, v1, i0, i1, i2, i3)                    \
+    _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(v0), \
+                                    _mm_castsi128_ps(v1), \
+                                    (i0)*1 + (i1)*4 + (i2)*16 + (i3)*64))
 ALWAYS_INLINE
 bool ContainsSameAccessFast(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
   // This is an optimized version of ContainsSameAccessSlow.
@@ -839,7 +830,7 @@
 
 ALWAYS_INLINE
 bool ContainsSameAccess(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
-#if defined(__SSE3__)
+#if TSAN_VECTORIZE
   bool res = ContainsSameAccessFast(s, a, sync_epoch, is_write);
   // NOTE: this check can fail if the shadow is concurrently mutated
   // by other threads. But it still can be useful if you modify
diff --git a/compiler-rt/lib/tsan/rtl/tsan_vector_clock.h b/compiler-rt/lib/tsan/rtl/tsan_vector_clock.h
new file mode 100644
--- /dev/null
+++ b/compiler-rt/lib/tsan/rtl/tsan_vector_clock.h
@@ -0,0 +1,51 @@
+//===-- tsan_vector_clock.h -------------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+#ifndef TSAN_VECTOR_CLOCK_H
+#define TSAN_VECTOR_CLOCK_H
+
+#include "tsan_defs.h"
+
+namespace __tsan {
+
+// Fixed-size vector clock, used both for threads and sync objects.
+class VectorClock {
+ public:
+  VectorClock();
+
+  Epoch Get(Sid sid) const;
+  void Set(Sid sid, Epoch v);
+
+  void Reset();
+  void Acquire(const VectorClock* src);
+  void Release(VectorClock** dstp) const;
+  void ReleaseStore(VectorClock** dstp) const;
+  void ReleaseStoreAcquire(VectorClock** dstp);
+  void ReleaseAcquire(VectorClock** dstp);
+
+  VectorClock& operator=(const VectorClock& other);
+
+ private:
+  Epoch clk_[kThreadSlotCount] VECTOR_ALIGNED;
+};
+
+ALWAYS_INLINE Epoch VectorClock::Get(Sid sid) const {
+  return clk_[static_cast<u8>(sid)];
+}
+
+ALWAYS_INLINE void VectorClock::Set(Sid sid, Epoch v) {
+  DCHECK_GE(v, clk_[static_cast<u8>(sid)]);
+  clk_[static_cast<u8>(sid)] = v;
+}
+
+}  // namespace __tsan
+
+#endif  // TSAN_VECTOR_CLOCK_H
diff --git a/compiler-rt/lib/tsan/rtl/tsan_vector_clock.cpp b/compiler-rt/lib/tsan/rtl/tsan_vector_clock.cpp
new file mode 100644
--- /dev/null
+++ b/compiler-rt/lib/tsan/rtl/tsan_vector_clock.cpp
@@ -0,0 +1,123 @@
+//===-- tsan_vector_clock.cpp ---------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+#include "tsan_vector_clock.h"
+
+#include "sanitizer_common/sanitizer_placement_new.h"
+#include "tsan_mman.h"
+
+namespace __tsan {
+
+#if TSAN_VECTORIZE
+const uptr kVectorClockSize = kThreadSlotCount * sizeof(Epoch) / sizeof(m128);
+#endif
+
+VectorClock::VectorClock() { Reset(); }
+
+void VectorClock::Reset() {
+#if !TSAN_VECTORIZE
+  for (uptr i = 0; i < kMaxSid; i++) clk_[i] = kEpochZero;
+#else
+  m128 z = _mm_setzero_si128();
+  m128* vclk = reinterpret_cast<m128*>(clk_);
+  for (uptr i = 0; i < kVectorClockSize; i++) _mm_store_si128(&vclk[i], z);
+#endif
+}
+
+void VectorClock::Acquire(const VectorClock* src) {
+  if (!src)
+    return;
+#if !TSAN_VECTORIZE
+  for (uptr i = 0; i < kMaxSid; i++) clk_[i] = max(clk_[i], src->clk_[i]);
+#else
+  m128* __restrict vdst = reinterpret_cast<m128*>(clk_);
+  m128 const* __restrict vsrc = reinterpret_cast<m128 const*>(src->clk_);
+  for (uptr i = 0; i < kVectorClockSize; i++) {
+    m128 s = _mm_load_si128(&vsrc[i]);
+    m128 d = _mm_load_si128(&vdst[i]);
+    m128 m = _mm_max_epu16(s, d);
+    _mm_store_si128(&vdst[i], m);
+  }
+#endif
+}
+
+static VectorClock* AllocClock(VectorClock** dstp) {
+  if (UNLIKELY(!*dstp))
+    *dstp = New<VectorClock>();
+  return *dstp;
+}
+
+void VectorClock::Release(VectorClock** dstp) const {
+  VectorClock* dst = AllocClock(dstp);
+  dst->Acquire(this);
+}
+
+void VectorClock::ReleaseStore(VectorClock** dstp) const {
+  VectorClock* dst = AllocClock(dstp);
+  *dst = *this;
+}
+
+VectorClock& VectorClock::operator=(const VectorClock& other) {
+#if !TSAN_VECTORIZE
+  for (uptr i = 0; i < kMaxSid; i++) clk_[i] = other.clk_[i];
+#else
+  m128* __restrict vdst = reinterpret_cast<m128*>(clk_);
+  m128 const* __restrict vsrc = reinterpret_cast<m128 const*>(other.clk_);
+  for (uptr i = 0; i < kVectorClockSize; i++) {
+    m128 s = _mm_load_si128(&vsrc[i]);
+    _mm_store_si128(&vdst[i], s);
+  }
+#endif
+  return *this;
+}
+
+void VectorClock::ReleaseStoreAcquire(VectorClock** dstp) {
+  VectorClock* dst = AllocClock(dstp);
+#if !TSAN_VECTORIZE
+  for (uptr i = 0; i < kMaxSid; i++) {
+    Epoch tmp = dst->clk_[i];
+    dst->clk_[i] = clk_[i];
+    clk_[i] = max(clk_[i], tmp);
+  }
+#else
+  m128* __restrict vdst = reinterpret_cast<m128*>(dst->clk_);
+  m128* __restrict vclk = reinterpret_cast<m128*>(clk_);
+  for (uptr i = 0; i < kVectorClockSize; i++) {
+    m128 t = _mm_load_si128(&vdst[i]);
+    m128 c = _mm_load_si128(&vclk[i]);
+    m128 m = _mm_max_epu16(c, t);
+    _mm_store_si128(&vdst[i], c);
+    _mm_store_si128(&vclk[i], m);
+  }
+#endif
+}
+
+void VectorClock::ReleaseAcquire(VectorClock** dstp) {
+  VectorClock* dst = AllocClock(dstp);
+#if !TSAN_VECTORIZE
+  for (uptr i = 0; i < kMaxSid; i++) {
+    dst->clk_[i] = max(dst->clk_[i], clk_[i]);
+    clk_[i] = dst->clk_[i];
+  }
+#else
+  m128* __restrict vdst = reinterpret_cast<m128*>(dst->clk_);
+  m128* __restrict vclk = reinterpret_cast<m128*>(clk_);
+  for (uptr i = 0; i < kVectorClockSize; i++) {
+    m128 c = _mm_load_si128(&vclk[i]);
+    m128 d = _mm_load_si128(&vdst[i]);
+    m128 m = _mm_max_epu16(c, d);
+    _mm_store_si128(&vdst[i], m);
+    _mm_store_si128(&vclk[i], m);
+  }
+#endif
+}
+
+}  // namespace __tsan
diff --git a/compiler-rt/lib/tsan/tests/unit/CMakeLists.txt b/compiler-rt/lib/tsan/tests/unit/CMakeLists.txt
--- a/compiler-rt/lib/tsan/tests/unit/CMakeLists.txt
+++ b/compiler-rt/lib/tsan/tests/unit/CMakeLists.txt
@@ -8,6 +8,7 @@
   tsan_stack_test.cpp
   tsan_sync_test.cpp
   tsan_unit_test_main.cpp
+  tsan_vector_clock_test.cpp
   )
 
 add_tsan_unittest(TsanUnitTest
diff --git a/compiler-rt/lib/tsan/tests/unit/tsan_vector_clock_test.cpp b/compiler-rt/lib/tsan/tests/unit/tsan_vector_clock_test.cpp
new file mode 100644
--- /dev/null
+++ b/compiler-rt/lib/tsan/tests/unit/tsan_vector_clock_test.cpp
@@ -0,0 +1,101 @@
+//===-- tsan_clock_test.cpp -----------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+#include "tsan_vector_clock.h"
+
+#include "gtest/gtest.h"
+#include "tsan_rtl.h"
+
+namespace __tsan {
+
+TEST(VectorClock, GetSet) {
+  // Compiler won't ensure alignment on stack.
+  VectorClock *vc = New<VectorClock>();
+  for (uptr i = 0; i < kThreadSlotCount; i++)
+    ASSERT_EQ(vc->Get(static_cast<Sid>(i)), kEpochZero);
+  for (uptr i = 0; i < kThreadSlotCount; i++)
+    vc->Set(static_cast<Sid>(i), static_cast<Epoch>(i));
+  for (uptr i = 0; i < kThreadSlotCount; i++)
+    ASSERT_EQ(vc->Get(static_cast<Sid>(i)), static_cast<Epoch>(i));
+  vc->Reset();
+  for (uptr i = 0; i < kThreadSlotCount; i++)
+    ASSERT_EQ(vc->Get(static_cast<Sid>(i)), kEpochZero);
+  DestroyAndFree(vc);
+}
+
+TEST(VectorClock, VectorOps) {
+  VectorClock *vc1 = New<VectorClock>();
+  VectorClock *vc2 = nullptr;
+  VectorClock *vc3 = nullptr;
+
+  vc1->Acquire(vc2);
+  for (uptr i = 0; i < kThreadSlotCount; i++)
+    ASSERT_EQ(vc1->Get(static_cast<Sid>(i)), kEpochZero);
+  vc1->Release(&vc2);
+  EXPECT_NE(vc2, nullptr);
+  vc1->Acquire(vc2);
+  for (uptr i = 0; i < kThreadSlotCount; i++)
+    ASSERT_EQ(vc1->Get(static_cast<Sid>(i)), kEpochZero);
+
+  for (uptr i = 0; i < kThreadSlotCount; i++) {
+    vc1->Set(static_cast<Sid>(i), static_cast<Epoch>(i));
+    vc2->Set(static_cast<Sid>(i), static_cast<Epoch>(kThreadSlotCount - i));
+  }
+  vc1->Acquire(vc2);
+  for (uptr i = 0; i < kThreadSlotCount; i++) {
+    ASSERT_EQ(vc1->Get(static_cast<Sid>(i)),
+              static_cast<Epoch>(i < kThreadSlotCount / 2 ? kThreadSlotCount - i
+                                                          : i));
+    ASSERT_EQ(vc2->Get(static_cast<Sid>(i)),
+              static_cast<Epoch>(kThreadSlotCount - i));
+  }
+  vc2->ReleaseStore(&vc3);
+  for (uptr i = 0; i < kThreadSlotCount; i++) {
+    ASSERT_EQ(vc3->Get(static_cast<Sid>(i)),
+              static_cast<Epoch>(kThreadSlotCount - i));
+    ASSERT_EQ(vc2->Get(static_cast<Sid>(i)),
+              static_cast<Epoch>(kThreadSlotCount - i));
+  }
+
+  vc1->Reset();
+  vc2->Reset();
+  for (uptr i = 0; i < kThreadSlotCount; i++) {
+    vc1->Set(static_cast<Sid>(i), static_cast<Epoch>(i));
+    vc2->Set(static_cast<Sid>(i), static_cast<Epoch>(kThreadSlotCount - i));
+  }
+  vc1->ReleaseAcquire(&vc2);
+  for (uptr i = 0; i < kThreadSlotCount; i++) {
+    Epoch expect =
+        static_cast<Epoch>(i < kThreadSlotCount / 2 ? kThreadSlotCount - i : i);
+    ASSERT_EQ(vc1->Get(static_cast<Sid>(i)), expect);
+    ASSERT_EQ(vc2->Get(static_cast<Sid>(i)), expect);
+  }
+
+  vc1->Reset();
+  vc2->Reset();
+  for (uptr i = 0; i < kThreadSlotCount; i++) {
+    vc1->Set(static_cast<Sid>(i), static_cast<Epoch>(i));
+    vc2->Set(static_cast<Sid>(i), static_cast<Epoch>(kThreadSlotCount - i));
+  }
+  vc1->ReleaseStoreAcquire(&vc2);
+  for (uptr i = 0; i < kThreadSlotCount; i++) {
+    ASSERT_EQ(vc1->Get(static_cast<Sid>(i)),
+              static_cast<Epoch>(i < kThreadSlotCount / 2 ? kThreadSlotCount - i
+                                                          : i));
+    ASSERT_EQ(vc2->Get(static_cast<Sid>(i)), static_cast<Epoch>(i));
+  }
+
+  DestroyAndFree(vc1);
+  DestroyAndFree(vc2);
+  DestroyAndFree(vc3);
+}
+
+}  // namespace __tsan