diff --git a/test/acm_random.h b/test/acm_random.h
index ff5c93e..a29ced2 100644
--- a/test/acm_random.h
+++ b/test/acm_random.h
@@ -32,6 +32,12 @@
     return (value >> 15) & 0xffff;
   }
 
+  int16_t Rand9Signed(void) {
+    // Use 9 bits: values between 255 (0x0FF) and -256 (0x100).
+    const uint32_t value = random_.Generate(512);
+    return static_cast<int16_t>(value - 256);
+  }
+
   uint8_t Rand8(void) {
     const uint32_t value =
         random_.Generate(testing::internal::Random::kMaxRange);
diff --git a/test/hadamard_test.cc b/test/hadamard_test.cc
new file mode 100644
index 0000000..6a428ff
--- /dev/null
+++ b/test/hadamard_test.cc
@@ -0,0 +1,122 @@
+/*
+ *  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <algorithm>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/register_state_check.h"
+
+namespace {
+
+using ::libvpx_test::ACMRandom;
+
+typedef void (*Hadamard8x8Func)(const int16_t *a, int a_stride,
+                                int16_t *b);
+
+class HadamardTest : public ::testing::TestWithParam<Hadamard8x8Func> {
+ public:
+  virtual void SetUp() {
+    h_func_ = GetParam();
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+  }
+
+ protected:
+  Hadamard8x8Func h_func_;
+  ACMRandom rnd_;
+};
+
+void hadamard_loop(const int16_t *a, int a_stride, int16_t *out) {
+  int16_t b[8];
+  for (int i = 0; i < 8; i += 2) {
+    b[i + 0] = a[i * a_stride] + a[(i + 1) * a_stride];
+    b[i + 1] = a[i * a_stride] - a[(i + 1) * a_stride];
+  }
+  int16_t c[8];
+  for (int i = 0; i < 8; i += 4) {
+    c[i + 0] = b[i + 0] + b[i + 2];
+    c[i + 1] = b[i + 1] + b[i + 3];
+    c[i + 2] = b[i + 0] - b[i + 2];
+    c[i + 3] = b[i + 1] - b[i + 3];
+  }
+  out[0] = c[0] + c[4];
+  out[7] = c[1] + c[5];
+  out[3] = c[2] + c[6];
+  out[4] = c[3] + c[7];
+  out[2] = c[0] - c[4];
+  out[6] = c[1] - c[5];
+  out[1] = c[2] - c[6];
+  out[5] = c[3] - c[7];
+}
+
+void reference_hadamard(const int16_t *a, int a_stride, int16_t *b) {
+  int16_t buf[64];
+  for (int i = 0; i < 8; i++) {
+    hadamard_loop(a + i, a_stride, buf + i * 8);
+  }
+
+  for (int i = 0; i < 8; i++) {
+    hadamard_loop(buf + i, 8, b + i * 8);
+  }
+}
+
+TEST_P(HadamardTest, CompareReferenceRandom) {
+  int16_t a[64], b[64], b_ref[64];
+  for (int i = 0; i < 64; i++) {
+    a[i] = rnd_.Rand9Signed();
+  }
+  memset(b, 0, sizeof(b));
+  memset(b_ref, 0, sizeof(b_ref));
+
+  reference_hadamard(a, 8, b_ref);
+  ASM_REGISTER_STATE_CHECK(h_func_(a, 8, b));
+
+  // The order of the output is not important. Sort before checking.
+  std::sort(b, b + 64);
+  std::sort(b_ref, b_ref + 64);
+  EXPECT_EQ(0, memcmp(b, b_ref, sizeof(b)));
+}
+
+TEST_P(HadamardTest, VaryStride) {
+  int16_t a[64 * 8], b[64], b_ref[64];
+  for (int i = 0; i < 64 * 8; i++) {
+    a[i] = rnd_.Rand9Signed();
+  }
+
+  for (int i = 8; i < 64; i += 8) {
+    memset(b, 0, sizeof(b));
+    memset(b_ref, 0, sizeof(b_ref));
+
+    reference_hadamard(a, i, b_ref);
+    ASM_REGISTER_STATE_CHECK(h_func_(a, i, b));
+
+    // The order of the output is not important. Sort before checking.
+    std::sort(b, b + 64);
+    std::sort(b_ref, b_ref + 64);
+    EXPECT_EQ(0, memcmp(b, b_ref, sizeof(b)));
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(C, HadamardTest,
+                        ::testing::Values(&vpx_hadamard_8x8_c));
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, HadamardTest,
+                        ::testing::Values(&vpx_hadamard_8x8_sse2));
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3 && CONFIG_USE_X86INC && ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSSE3, HadamardTest,
+                        ::testing::Values(&vpx_hadamard_8x8_ssse3));
+#endif  // HAVE_SSSE3 && CONFIG_USE_X86INC && ARCH_X86_64
+}  // namespace
diff --git a/test/test.mk b/test/test.mk
index 6931853..d28ab11 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -143,6 +143,7 @@
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct32x32_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct4x4_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += hadamard_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_error_block_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_quantize_test.cc
diff --git a/vpx_dsp/avg.c b/vpx_dsp/avg.c
index a44c52e..a8c9966 100644
--- a/vpx_dsp/avg.c
+++ b/vpx_dsp/avg.c
@@ -62,6 +62,8 @@
   coeff[5] = c3 - c7;
 }
 
+// The order of the output coeff of the hadamard is not important. For
+// optimization purposes the final transpose may be skipped.
 void vpx_hadamard_8x8_c(const int16_t *src_diff, int src_stride,
                         int16_t *coeff) {
   int idx;