Added highbitdepth sse2 acceleration for quantize

Also includes block error.

(This patch is mostly cherry picked from
commit db7192e0b014a331a1dcb102c8a1148e9f0e1081)

Change-Id: Idef18f90b111a0d0c9546543d3347e551908fd78

test/test.mk

@@ -137,6 +137,8 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += lpf_8_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_avg_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_error_block_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_quantize_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9)         += vp9_intrapred_test.cc
 
 ifeq ($(CONFIG_VP9_ENCODER),yes)

test/vp9_error_block_test.cc

+/*
+ *  Copyright (c) 2014 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 <cmath>
+#include <cstdlib>
+#include <string>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+#if CONFIG_VP9_HIGHBITDEPTH
+const int kNumIterations = 1000;
+
+typedef int64_t (*ErrorBlockFunc)(const tran_low_t *coeff,
+                                  const tran_low_t *dqcoeff,
+                                  intptr_t block_size,
+                                  int64_t *ssz, int bps);
+
+typedef std::tr1::tuple<ErrorBlockFunc, ErrorBlockFunc, vpx_bit_depth_t>
+                        ErrorBlockParam;
+
+class ErrorBlockTest
+  : public ::testing::TestWithParam<ErrorBlockParam> {
+ public:
+  virtual ~ErrorBlockTest() {}
+  virtual void SetUp() {
+    error_block_op_     = GET_PARAM(0);
+    ref_error_block_op_ = GET_PARAM(1);
+    bit_depth_  = GET_PARAM(2);
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  vpx_bit_depth_t bit_depth_;
+  ErrorBlockFunc error_block_op_;
+  ErrorBlockFunc ref_error_block_op_;
+};
+
+TEST_P(ErrorBlockTest, OperationCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff,   4096);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, dqcoeff, 4096);
+  int err_count_total = 0;
+  int first_failure = -1;
+  intptr_t block_size;
+  int64_t ssz;
+  int64_t ret;
+  int64_t ref_ssz;
+  int64_t ref_ret;
+  for (int i = 0; i < kNumIterations; ++i) {
+    int err_count = 0;
+    block_size = 16 << (i % 9);  // All block sizes from 4x4, 8x4 ..64x64
+    for (int j = 0; j < block_size; j++) {
+      coeff[j]   = rnd(2 << 20) - (1 << 20);
+      dqcoeff[j] = rnd(2 << 20) - (1 << 20);
+    }
+    ref_ret = ref_error_block_op_(coeff, dqcoeff, block_size, &ref_ssz,
+                                  bit_depth_);
+    ASM_REGISTER_STATE_CHECK(ret = error_block_op_(coeff, dqcoeff, block_size,
+                                                   &ssz, bit_depth_));
+    err_count += (ref_ret != ret) | (ref_ssz != ssz);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Error Block Test, C output doesn't match SSE2 output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(ErrorBlockTest, ExtremeValues) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff,   4096);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, dqcoeff, 4096);
+  int err_count_total = 0;
+  int first_failure = -1;
+  intptr_t block_size;
+  int64_t ssz;
+  int64_t ret;
+  int64_t ref_ssz;
+  int64_t ref_ret;
+  int max_val = ((1 << 20) - 1);
+  for (int i = 0; i < kNumIterations; ++i) {
+    int err_count = 0;
+    int k = (i / 9) % 5;
+
+    // Change the maximum coeff value, to test different bit boundaries
+    if ( k == 4 && (i % 9) == 0 ) {
+      max_val >>= 1;
+    }
+    block_size = 16 << (i % 9);  // All block sizes from 4x4, 8x4 ..64x64
+    for (int j = 0; j < block_size; j++) {
+      if (k < 4) {  // Test at maximum values
+        coeff[j]   = k % 2 ? max_val : -max_val;
+        dqcoeff[j] = (k >> 1) % 2 ? max_val : -max_val;
+      } else {
+        coeff[j]   = rnd(2 << 14) - (1 << 14);
+        dqcoeff[j] = rnd(2 << 14) - (1 << 14);
+      }
+    }
+    ref_ret = ref_error_block_op_(coeff, dqcoeff, block_size, &ref_ssz,
+                                  bit_depth_);
+    ASM_REGISTER_STATE_CHECK(ret = error_block_op_(coeff, dqcoeff, block_size,
+                                                   &ssz, bit_depth_));
+    err_count += (ref_ret != ret) | (ref_ssz != ssz);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Error Block Test, C output doesn't match SSE2 output. "
+      << "First failed at test case " << first_failure;
+}
+
+using std::tr1::make_tuple;
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, ErrorBlockTest,
+    ::testing::Values(
+        make_tuple(&vp9_highbd_block_error_sse2,
+                   &vp9_highbd_block_error_c, VPX_BITS_10),
+        make_tuple(&vp9_highbd_block_error_sse2,
+                   &vp9_highbd_block_error_c, VPX_BITS_12),
+        make_tuple(&vp9_highbd_block_error_sse2,
+                   &vp9_highbd_block_error_c, VPX_BITS_8)));
+#endif  // HAVE_SSE2
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+}  // namespace

test/vp9_quantize_test.cc

+/*
+ *  Copyright (c) 2014 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 <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+#if CONFIG_VP9_HIGHBITDEPTH
+const int number_of_iterations = 100;
+
+typedef void (*QuantizeFunc)(const tran_low_t *coeff, intptr_t count,
+                             int skip_block, const int16_t *zbin,
+                             const int16_t *round, const int16_t *quant,
+                             const int16_t *quant_shift,
+                             tran_low_t *qcoeff, tran_low_t *dqcoeff,
+                             const int16_t *dequant, int zbin_oq_value,
+                             uint16_t *eob, const int16_t *scan,
+                             const int16_t *iscan);
+typedef std::tr1::tuple<QuantizeFunc, QuantizeFunc, vpx_bit_depth_t>
+    QuantizeParam;
+
+class VP9QuantizeTest : public ::testing::TestWithParam<QuantizeParam> {
+ public:
+  virtual ~VP9QuantizeTest() {}
+  virtual void SetUp() {
+    quantize_op_   = GET_PARAM(0);
+    ref_quantize_op_ = GET_PARAM(1);
+    bit_depth_  = GET_PARAM(2);
+    mask_ = (1 << bit_depth_) - 1;
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  vpx_bit_depth_t bit_depth_;
+  int mask_;
+  QuantizeFunc quantize_op_;
+  QuantizeFunc ref_quantize_op_;
+};
+
+class VP9Quantize32Test : public ::testing::TestWithParam<QuantizeParam> {
+ public:
+  virtual ~VP9Quantize32Test() {}
+  virtual void SetUp() {
+    quantize_op_   = GET_PARAM(0);
+    ref_quantize_op_ = GET_PARAM(1);
+    bit_depth_  = GET_PARAM(2);
+    mask_ = (1 << bit_depth_) - 1;
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  vpx_bit_depth_t bit_depth_;
+  int mask_;
+  QuantizeFunc quantize_op_;
+  QuantizeFunc ref_quantize_op_;
+};
+
+TEST_P(VP9QuantizeTest, OperationCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int zbin_oq_value = 0;
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, zbin_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, round_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_shift_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, qcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, dqcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_qcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_dqcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, dequant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, eob_ptr, 1);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_eob_ptr, 1);
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < number_of_iterations; ++i) {
+    const int skip_block = i == 0;
+    const TX_SIZE sz = (TX_SIZE)(i % 3);  // TX_4X4, TX_8X8 TX_16X16
+    const TX_TYPE tx_type = (TX_TYPE)((i >> 2) % 3);
+    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
+    const int count = (4 << sz) * (4 << sz);  // 16, 64, 256
+    int err_count = 0;
+    *eob_ptr = rnd.Rand16();
+    *ref_eob_ptr = *eob_ptr;
+    for (int j = 0; j < count; j++) {
+      coeff_ptr[j] = rnd.Rand16()&mask_;
+    }
+    for (int j = 0; j < 2; j++) {
+      zbin_ptr[j] = rnd.Rand16()&mask_;
+      round_ptr[j] = rnd.Rand16();
+      quant_ptr[j] = rnd.Rand16();
+      quant_shift_ptr[j] = rnd.Rand16();
+      dequant_ptr[j] = rnd.Rand16();
+    }
+    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
+                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
+                     ref_dqcoeff_ptr, dequant_ptr, zbin_oq_value,
+                     ref_eob_ptr, scan_order->scan, scan_order->iscan);
+    ASM_REGISTER_STATE_CHECK(quantize_op_(coeff_ptr, count, skip_block,
+                                          zbin_ptr, round_ptr, quant_ptr,
+                                          quant_shift_ptr, qcoeff_ptr,
+                                          dqcoeff_ptr, dequant_ptr,
+                                          zbin_oq_value, eob_ptr,
+                                          scan_order->scan, scan_order->iscan));
+    for (int j = 0; j < sz; ++j) {
+      err_count += (ref_qcoeff_ptr[j]  != qcoeff_ptr[j]) |
+          (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
+    }
+    err_count += (*ref_eob_ptr != *eob_ptr);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Quantization Test, C output doesn't match SSE2 output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(VP9Quantize32Test, OperationCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int zbin_oq_value = 0;
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, zbin_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, round_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_shift_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, qcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, dqcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_qcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_dqcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, dequant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, eob_ptr, 1);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_eob_ptr, 1);
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < number_of_iterations; ++i) {
+    const int skip_block = i == 0;
+    const TX_SIZE sz = TX_32X32;
+    const TX_TYPE tx_type = (TX_TYPE)(i % 4);
+    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
+    const int count = (4 << sz) * (4 << sz);  // 1024
+    int err_count = 0;
+    *eob_ptr = rnd.Rand16();
+    *ref_eob_ptr = *eob_ptr;
+    for (int j = 0; j < count; j++) {
+      coeff_ptr[j] = rnd.Rand16()&mask_;
+    }
+    for (int j = 0; j < 2; j++) {
+      zbin_ptr[j] = rnd.Rand16()&mask_;
+      round_ptr[j] = rnd.Rand16();
+      quant_ptr[j] = rnd.Rand16();
+      quant_shift_ptr[j] = rnd.Rand16();
+      dequant_ptr[j] = rnd.Rand16();
+    }
+    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
+                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
+                     ref_dqcoeff_ptr, dequant_ptr, zbin_oq_value,
+                     ref_eob_ptr, scan_order->scan, scan_order->iscan);
+    ASM_REGISTER_STATE_CHECK(quantize_op_(coeff_ptr, count, skip_block,
+                                          zbin_ptr, round_ptr, quant_ptr,
+                                          quant_shift_ptr, qcoeff_ptr,
+                                          dqcoeff_ptr, dequant_ptr,
+                                          zbin_oq_value, eob_ptr,
+                                          scan_order->scan, scan_order->iscan));
+    for (int j = 0; j < sz; ++j) {
+      err_count += (ref_qcoeff_ptr[j]  != qcoeff_ptr[j]) |
+          (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
+    }
+    err_count += (*ref_eob_ptr != *eob_ptr);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Quantization Test, C output doesn't match SSE2 output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(VP9QuantizeTest, EOBCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int zbin_oq_value = 0;
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, zbin_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, round_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_shift_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, qcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, dqcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_qcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_dqcoeff_ptr, 256);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, dequant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, eob_ptr, 1);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_eob_ptr, 1);
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < number_of_iterations; ++i) {
+    int skip_block = i == 0;
+    TX_SIZE sz = (TX_SIZE)(i % 3);  // TX_4X4, TX_8X8 TX_16X16
+    TX_TYPE tx_type = (TX_TYPE)((i >> 2) % 3);
+    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
+    int count = (4 << sz) * (4 << sz);  // 16, 64, 256
+    int err_count = 0;
+    *eob_ptr = rnd.Rand16();
+    *ref_eob_ptr = *eob_ptr;
+    // Two random entries
+    for (int j = 0; j < count; j++) {
+      coeff_ptr[j] = 0;
+    }
+    coeff_ptr[rnd(count)] = rnd.Rand16()&mask_;
+    coeff_ptr[rnd(count)] = rnd.Rand16()&mask_;
+    for (int j = 0; j < 2; j++) {
+      zbin_ptr[j] = rnd.Rand16()&mask_;
+      round_ptr[j] = rnd.Rand16();
+      quant_ptr[j] = rnd.Rand16();
+      quant_shift_ptr[j] = rnd.Rand16();
+      dequant_ptr[j] = rnd.Rand16();
+    }
+
+    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
+                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
+                     ref_dqcoeff_ptr, dequant_ptr, zbin_oq_value,
+                     ref_eob_ptr, scan_order->scan, scan_order->iscan);
+    ASM_REGISTER_STATE_CHECK(quantize_op_(coeff_ptr, count, skip_block,
+                                          zbin_ptr, round_ptr, quant_ptr,
+                                          quant_shift_ptr, qcoeff_ptr,
+                                          dqcoeff_ptr, dequant_ptr,
+                                          zbin_oq_value, eob_ptr,
+                                          scan_order->scan, scan_order->iscan));
+
+    for (int j = 0; j < sz; ++j) {
+      err_count += (ref_qcoeff_ptr[j]  != qcoeff_ptr[j]) |
+          (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
+    }
+    err_count += (*ref_eob_ptr != *eob_ptr);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Quantization Test, C output doesn't match SSE2 output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(VP9Quantize32Test, EOBCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int zbin_oq_value = 0;
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, zbin_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, round_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, quant_shift_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, qcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, dqcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_qcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_dqcoeff_ptr, 1024);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, dequant_ptr, 2);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, eob_ptr, 1);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_eob_ptr, 1);
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < number_of_iterations; ++i) {
+    int skip_block = i == 0;
+    TX_SIZE sz = TX_32X32;
+    TX_TYPE tx_type = (TX_TYPE)(i % 4);
+    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
+    int count = (4 << sz) * (4 << sz);  // 1024
+    int err_count = 0;
+    *eob_ptr = rnd.Rand16();
+    *ref_eob_ptr = *eob_ptr;
+    for (int j = 0; j < count; j++) {
+      coeff_ptr[j] = 0;
+    }
+    // Two random entries
+    coeff_ptr[rnd(count)] = rnd.Rand16()&mask_;
+    coeff_ptr[rnd(count)] = rnd.Rand16()&mask_;
+    for (int j = 0; j < 2; j++) {
+      zbin_ptr[j] = rnd.Rand16()&mask_;
+      round_ptr[j] = rnd.Rand16();
+      quant_ptr[j] = rnd.Rand16();
+      quant_shift_ptr[j] = rnd.Rand16();
+      dequant_ptr[j] = rnd.Rand16();
+    }
+
+    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
+                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
+                     ref_dqcoeff_ptr, dequant_ptr, zbin_oq_value,
+                     ref_eob_ptr, scan_order->scan, scan_order->iscan);
+    ASM_REGISTER_STATE_CHECK(quantize_op_(coeff_ptr, count, skip_block,
+                                          zbin_ptr, round_ptr, quant_ptr,
+                                          quant_shift_ptr, qcoeff_ptr,
+                                          dqcoeff_ptr, dequant_ptr,
+                                          zbin_oq_value, eob_ptr,
+                                          scan_order->scan, scan_order->iscan));
+
+    for (int j = 0; j < sz; ++j) {
+      err_count += (ref_qcoeff_ptr[j]  != qcoeff_ptr[j]) |
+          (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
+    }
+    err_count += (*ref_eob_ptr != *eob_ptr);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Quantization Test, C output doesn't match SSE2 output. "
+      << "First failed at test case " << first_failure;
+}
+using std::tr1::make_tuple;
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, VP9QuantizeTest,
+    ::testing::Values(
+        make_tuple(&vp9_highbd_quantize_b_sse2,
+                   &vp9_highbd_quantize_b_c, VPX_BITS_8),
+        make_tuple(&vp9_highbd_quantize_b_sse2,
+                   &vp9_highbd_quantize_b_c, VPX_BITS_10),
+        make_tuple(&vp9_highbd_quantize_b_sse2,
+                   &vp9_highbd_quantize_b_c, VPX_BITS_12)));
+INSTANTIATE_TEST_CASE_P(
+    SSE2, VP9Quantize32Test,
+    ::testing::Values(
+        make_tuple(&vp9_highbd_quantize_b_32x32_sse2,
+                   &vp9_highbd_quantize_b_32x32_c, VPX_BITS_8),
+        make_tuple(&vp9_highbd_quantize_b_32x32_sse2,
+                   &vp9_highbd_quantize_b_32x32_c, VPX_BITS_10),
+        make_tuple(&vp9_highbd_quantize_b_32x32_sse2,
+                   &vp9_highbd_quantize_b_32x32_c, VPX_BITS_12)));
+#endif  // HAVE_SSE2
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+}  // namespace

vp9/common/vp9_rtcd_defs.pl

@@ -1855,7 +1855,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   # ENCODEMB INVOKE
 
   add_proto qw/int64_t vp9_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
-  specialize qw/vp9_highbd_block_error/;
+  specialize qw/vp9_highbd_block_error sse2/;
 
   add_proto qw/void vp9_highbd_subtract_block/, "int rows, int cols, int16_t *diff_ptr, ptrdiff_t diff_stride, const uint8_t *src_ptr, ptrdiff_t src_stride, const uint8_t *pred_ptr, ptrdiff_t pred_stride, int bd";
   specialize qw/vp9_highbd_subtract_block/;
@@ -1867,10 +1867,10 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vp9_highbd_quantize_fp_32x32/;
 
   add_proto qw/void vp9_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vp9_highbd_quantize_b/;
+  specialize qw/vp9_highbd_quantize_b sse2/;
 
   add_proto qw/void vp9_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vp9_highbd_quantize_b_32x32/;
+  specialize qw/vp9_highbd_quantize_b_32x32 sse2/;
 
   #
   # Structured Similarity (SSIM)

vp9/encoder/x86/vp9_highbd_block_error_intrin_sse2.c

+/*
+ *  Copyright (c) 2014 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 <emmintrin.h>
+#include <stdio.h>
+
+#include "vp9/common/vp9_common.h"
+
+int64_t vp9_highbd_block_error_sse2(tran_low_t *coeff, tran_low_t *dqcoeff,
+                                    intptr_t block_size, int64_t *ssz,
+                                    int bps) {
+  int i, j, test;
+  uint32_t temp[4];
+  __m128i max, min, cmp0, cmp1, cmp2, cmp3;
+  int64_t error = 0, sqcoeff = 0;
+  const int shift = 2 * (bps - 8);
+  const int rounding = shift > 0 ? 1 << (shift - 1) : 0;
+
+  for (i = 0; i < block_size; i+=8) {
+    // Load the data into xmm registers
+    __m128i mm_coeff = _mm_load_si128((__m128i*) (coeff + i));
+    __m128i mm_coeff2 = _mm_load_si128((__m128i*) (coeff + i + 4));
+    __m128i mm_dqcoeff = _mm_load_si128((__m128i*) (dqcoeff + i));
+    __m128i mm_dqcoeff2 = _mm_load_si128((__m128i*) (dqcoeff + i + 4));
+    // Check if any values require more than 15 bit
+    max = _mm_set1_epi32(0x3fff);
+    min = _mm_set1_epi32(0xffffc000);
+    cmp0 = _mm_xor_si128(_mm_cmpgt_epi32(mm_coeff, max),
+            _mm_cmplt_epi32(mm_coeff, min));
+    cmp1 = _mm_xor_si128(_mm_cmpgt_epi32(mm_coeff2, max),
+            _mm_cmplt_epi32(mm_coeff2, min));
+    cmp2 = _mm_xor_si128(_mm_cmpgt_epi32(mm_dqcoeff, max),
+            _mm_cmplt_epi32(mm_dqcoeff, min));
+    cmp3 = _mm_xor_si128(_mm_cmpgt_epi32(mm_dqcoeff2, max),
+            _mm_cmplt_epi32(mm_dqcoeff2, min));
+    test = _mm_movemask_epi8(_mm_or_si128(_mm_or_si128(cmp0, cmp1),
+            _mm_or_si128(cmp2, cmp3)));
+
+    if (!test) {
+      __m128i mm_diff, error_sse2, sqcoeff_sse2;;
+      mm_coeff = _mm_packs_epi32(mm_coeff, mm_coeff2);
+      mm_dqcoeff = _mm_packs_epi32(mm_dqcoeff, mm_dqcoeff2);
+      mm_diff = _mm_sub_epi16(mm_coeff, mm_dqcoeff);
+      error_sse2 = _mm_madd_epi16(mm_diff, mm_diff);
+      sqcoeff_sse2 = _mm_madd_epi16(mm_coeff, mm_coeff);
+      _mm_storeu_si128((__m128i*)temp, error_sse2);
+      error = error + temp[0] + temp[1] + temp[2] + temp[3];