libvpx: Pull from upstream

source/libvpx/test/dct16x16_test.cc

 /*
  *  Copyright (c) 2012 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.
  */
 
@@ skipping 338 matching lines @@
   virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
 
   virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
 
   void RunAccuracyCheck() {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     uint32_t max_error = 0;
     int64_t total_error = 0;
     const int count_test_block = 10000;
     for (int i = 0; i < count_test_block; ++i) {
-      DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
-      DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, kNumCoeffs);
-      DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
-      DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+      DECLARE_ALIGNED(16, int16_t, test_input_block[kNumCoeffs]);
+      DECLARE_ALIGNED(16, tran_low_t, test_temp_block[kNumCoeffs]);
+      DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
+      DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
 #if CONFIG_VP9_HIGHBITDEPTH
-      DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
-      DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+      DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
+      DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
 #endif
 
       // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
         if (bit_depth_ == VPX_BITS_8) {
           src[j] = rnd.Rand8();
           dst[j] = rnd.Rand8();
           test_input_block[j] = src[j] - dst[j];
 #if CONFIG_VP9_HIGHBITDEPTH
         } else {
@@ skipping 33 matching lines @@
     EXPECT_GE(1u  << 2 * (bit_depth_ - 8), max_error)
         << "Error: 16x16 FHT/IHT has an individual round trip error > 1";
 
     EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
         << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
   }
 
   void RunCoeffCheck() {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 1000;
-    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
+    DECLARE_ALIGNED(16, int16_t, input_block[kNumCoeffs]);
+    DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
+    DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
 
     for (int i = 0; i < count_test_block; ++i) {
       // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j)
         input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
 
       fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
 
       // The minimum quant value is 4.
       for (int j = 0; j < kNumCoeffs; ++j)
         EXPECT_EQ(output_block[j], output_ref_block[j]);
     }
   }
 
   void RunMemCheck() {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 1000;
-    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
+    DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
+    DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
+    DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
 
     for (int i = 0; i < count_test_block; ++i) {
       // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
         input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
       }
       if (i == 0) {
         for (int j = 0; j < kNumCoeffs; ++j)
           input_extreme_block[j] = mask_;
       } else if (i == 1) {
         for (int j = 0; j < kNumCoeffs; ++j)
           input_extreme_block[j] = -mask_;
       }
 
       fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
                                           output_block, pitch_));
 
       // The minimum quant value is 4.
       for (int j = 0; j < kNumCoeffs; ++j) {
         EXPECT_EQ(output_block[j], output_ref_block[j]);
         EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
             << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
       }
     }
   }
 
   void RunQuantCheck(int dc_thred, int ac_thred) {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 100000;
-    DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
+    DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
+    DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
 
-    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint8_t, ref, kNumCoeffs);
+    DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
 #if CONFIG_VP9_HIGHBITDEPTH
-    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint16_t, ref16, kNumCoeffs);
+    DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
 #endif
 
     for (int i = 0; i < count_test_block; ++i) {
       // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        if (bit_depth_ == VPX_BITS_8)
-          input_block[j] = rnd.Rand8() - rnd.Rand8();
-        else
-          input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
         input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
       }
       if (i == 0)
         for (int j = 0; j < kNumCoeffs; ++j)
           input_extreme_block[j] = mask_;
       if (i == 1)
         for (int j = 0; j < kNumCoeffs; ++j)
           input_extreme_block[j] = -mask_;
 
       fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
 
       // clear reconstructed pixel buffers
-      vpx_memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
-      vpx_memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
+      memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
+      memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
 #if CONFIG_VP9_HIGHBITDEPTH
-      vpx_memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
-      vpx_memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
+      memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
+      memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
 #endif
 
       // quantization with maximum allowed step sizes
       output_ref_block[0] = (output_ref_block[0] / dc_thred) * dc_thred;
       for (int j = 1; j < kNumCoeffs; ++j)
         output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
       if (bit_depth_ == VPX_BITS_8) {
         inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
         ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
 #if CONFIG_VP9_HIGHBITDEPTH
@@ skipping 12 matching lines @@
         for (int j = 0; j < kNumCoeffs; ++j)
           EXPECT_EQ(ref16[j], dst16[j]);
 #endif
       }
     }
   }
 
   void RunInvAccuracyCheck() {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 1000;
-    DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+    DECLARE_ALIGNED(16, int16_t, in[kNumCoeffs]);
+    DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
 #if CONFIG_VP9_HIGHBITDEPTH
-    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+    DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
     for (int i = 0; i < count_test_block; ++i) {
       double out_r[kNumCoeffs];
 
       // Initialize a test block with input range [-255, 255].
       for (int j = 0; j < kNumCoeffs; ++j) {
         if (bit_depth_ == VPX_BITS_8) {
           src[j] = rnd.Rand8();
           dst[j] = rnd.Rand8();
@@ skipping 33 matching lines @@
             << " at index " << j;
       }
     }
   }
 
   void CompareInvReference(IdctFunc ref_txfm, int thresh) {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 10000;
     const int eob = 10;
     const int16_t *scan = vp9_default_scan_orders[TX_16X16].scan;
-    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint8_t, ref, kNumCoeffs);
+    DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
 #if CONFIG_VP9_HIGHBITDEPTH
-    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, uint16_t, ref16, kNumCoeffs);
+    DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
+    DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
     for (int i = 0; i < count_test_block; ++i) {
       for (int j = 0; j < kNumCoeffs; ++j) {
         if (j < eob) {
           // Random values less than the threshold, either positive or negative
           coeff[scan[j]] = rnd(thresh) * (1 - 2 * (i % 2));
         } else {
           coeff[scan[j]] = 0;
         }
@@ skipping 309 matching lines @@
         make_tuple(&idct16x16_10_add_10_c,
                    &idct16x16_10_add_10_sse2, 3167, VPX_BITS_10),
         make_tuple(&idct16x16_10,
                    &idct16x16_256_add_10_sse2, 3167, VPX_BITS_10),
         make_tuple(&idct16x16_10_add_12_c,
                    &idct16x16_10_add_12_sse2, 3167, VPX_BITS_12),
         make_tuple(&idct16x16_12,
                    &idct16x16_256_add_12_sse2, 3167, VPX_BITS_12)));
 #endif  // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 
-#if HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(
-    SSSE3, Trans16x16DCT,
+    MSA, Trans16x16DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_ssse3, 0,
-                   VPX_BITS_8)));
-#endif  // HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_msa, 0, VPX_BITS_8)));
+INSTANTIATE_TEST_CASE_P(
+    MSA, Trans16x16HT,
+    ::testing::Values(
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 3, VPX_BITS_8)));
+#endif  // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 }  // namespace