Enable unit test for partial 16x16 inverse 2D-DCT

This commit enables unit test for SSSE3 16x16 inverse 2D-DCT with
10 non-zero coefficients. It includes a new test condition to
cover the potential overflow issue due to extremely coarse quantization.

Change-Id: I945e16f05dfbe19500f0da5f15990feba8e26d99

diff --git a/test/partial_idct_test.cc b/test/partial_idct_test.cc
index b63e07b..f2171b2 100644 (file)
--- a/test/partial_idct_test.cc
+++ b/test/partial_idct_test.cc
@@ -28,7 +28,8 @@ using libvpx_test::ACMRandom;
 namespace {
 typedef void (*fwd_txfm_t)(const int16_t *in, int16_t *out, int stride);
 typedef void (*inv_txfm_t)(const int16_t *in, uint8_t *out, int stride);
-typedef std::tr1::tuple<inv_txfm_t,
+typedef std::tr1::tuple<fwd_txfm_t,
+                        inv_txfm_t,
                         inv_txfm_t,
                         TX_SIZE, int> partial_itxfm_param_t;
 const int kMaxNumCoeffs = 1024;
@@ -36,10 +37,11 @@ class PartialIDctTest : public ::testing::TestWithParam<partial_itxfm_param_t> {
  public:
   virtual ~PartialIDctTest() {}
   virtual void SetUp() {
-    full_itxfm_ = GET_PARAM(0);
-    partial_itxfm_ = GET_PARAM(1);
-    tx_size_  = GET_PARAM(2);
-    last_nonzero_ = GET_PARAM(3);
+    ftxfm_ = GET_PARAM(0);
+    full_itxfm_ = GET_PARAM(1);
+    partial_itxfm_ = GET_PARAM(2);
+    tx_size_  = GET_PARAM(3);
+    last_nonzero_ = GET_PARAM(4);
   }
 
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
@@ -47,10 +49,90 @@ class PartialIDctTest : public ::testing::TestWithParam<partial_itxfm_param_t> {
  protected:
   int last_nonzero_;
   TX_SIZE tx_size_;
+  fwd_txfm_t ftxfm_;
   inv_txfm_t full_itxfm_;
   inv_txfm_t partial_itxfm_;
 };
 
+TEST_P(PartialIDctTest, RunQuantCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int size;
+  switch (tx_size_) {
+    case TX_4X4:
+      size = 4;
+      break;
+    case TX_8X8:
+      size = 8;
+      break;
+    case TX_16X16:
+      size = 16;
+      break;
+    case TX_32X32:
+      size = 32;
+      break;
+    default:
+      FAIL() << "Wrong Size!";
+      break;
+  }
+  DECLARE_ALIGNED_ARRAY(16, int16_t, test_coef_block1, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, test_coef_block2, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, dst1, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, dst2, kMaxNumCoeffs);
+
+  const int count_test_block = 1000;
+  const int block_size = size * size;
+
+  DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kMaxNumCoeffs);
+
+  int max_error = 0;
+  for (int i = 0; i < count_test_block; ++i) {
+    // clear out destination buffer
+    memset(dst1, 0, sizeof(*dst1) * block_size);
+    memset(dst2, 0, sizeof(*dst2) * block_size);
+    memset(test_coef_block1, 0, sizeof(*test_coef_block1) * block_size);
+    memset(test_coef_block2, 0, sizeof(*test_coef_block2) * block_size);
+
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with input range [-255, 255].
+      if (i == 0) {
+        for (int j = 0; j < block_size; ++j)
+          input_extreme_block[j] = 255;
+      } else if (i == 1) {
+        for (int j = 0; j < block_size; ++j)
+          input_extreme_block[j] = -255;
+      } else {
+        for (int j = 0; j < block_size; ++j) {
+          input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
+        }
+      }
+
+      ftxfm_(input_extreme_block, output_ref_block, size);
+
+      // quantization with maximum allowed step sizes
+      test_coef_block1[0] = (output_ref_block[0] / 1336) * 1336;
+      for (int j = 1; j < last_nonzero_; ++j)
+        test_coef_block1[vp9_default_scan_orders[tx_size_].scan[j]]
+                         = (output_ref_block[j] / 1828) * 1828;
+    }
+
+    REGISTER_STATE_CHECK(full_itxfm_(test_coef_block1, dst1, size));
+    REGISTER_STATE_CHECK(partial_itxfm_(test_coef_block1, dst2, size));
+
+    for (int j = 0; j < block_size; ++j) {
+      const int diff = dst1[j] - dst2[j];
+      const int error = diff * diff;
+      if (max_error < error)
+        max_error = error;
+    }
+  }
+
+  EXPECT_EQ(0, max_error)
+      << "Error: partial inverse transform produces different results";
+}
+
 TEST_P(PartialIDctTest, ResultsMatch) {
   ACMRandom rnd(ACMRandom::DeterministicSeed());
   int size;
@@ -119,47 +201,60 @@ using std::tr1::make_tuple;
 INSTANTIATE_TEST_CASE_P(
     C, PartialIDctTest,
     ::testing::Values(
-        make_tuple(&vp9_idct32x32_1024_add_c,
+        make_tuple(&vp9_fdct32x32_c,
+                   &vp9_idct32x32_1024_add_c,
                    &vp9_idct32x32_34_add_c,
                    TX_32X32, 34),
-        make_tuple(&vp9_idct32x32_1024_add_c,
+        make_tuple(&vp9_fdct32x32_c,
+                   &vp9_idct32x32_1024_add_c,
                    &vp9_idct32x32_1_add_c,
                    TX_32X32, 1),
-        make_tuple(&vp9_idct16x16_256_add_c,
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_c,
                    &vp9_idct16x16_10_add_c,
                    TX_16X16, 10),
-        make_tuple(&vp9_idct16x16_256_add_c,
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_c,
                    &vp9_idct16x16_1_add_c,
                    TX_16X16, 1),
-        make_tuple(&vp9_idct8x8_64_add_c,
+        make_tuple(&vp9_fdct8x8_c,
+                   &vp9_idct8x8_64_add_c,
                    &vp9_idct8x8_12_add_c,
                    TX_8X8, 12),
-        make_tuple(&vp9_idct8x8_64_add_c,
+        make_tuple(&vp9_fdct8x8_c,
+                   &vp9_idct8x8_64_add_c,
                    &vp9_idct8x8_1_add_c,
                    TX_8X8, 1),
-        make_tuple(&vp9_idct4x4_16_add_c,
+        make_tuple(&vp9_fdct4x4_c,
+                   &vp9_idct4x4_16_add_c,
                    &vp9_idct4x4_1_add_c,
                    TX_4X4, 1)));
 #if HAVE_NEON_ASM
 INSTANTIATE_TEST_CASE_P(
     NEON, PartialIDctTest,
     ::testing::Values(
-        make_tuple(&vp9_idct32x32_1024_add_c,
+        make_tuple(&vp9_fdct32x32_c,
+                   &vp9_idct32x32_1024_add_c,
                    &vp9_idct32x32_1_add_neon,
                    TX_32X32, 1),
-        make_tuple(&vp9_idct16x16_256_add_c,
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_c,
                    &vp9_idct16x16_10_add_neon,
                    TX_16X16, 10),
-        make_tuple(&vp9_idct16x16_256_add_c,
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_c,
                    &vp9_idct16x16_1_add_neon,
                    TX_16X16, 1),
-        make_tuple(&vp9_idct8x8_64_add_c,
+        make_tuple(&vp9_fdct8x8_c,
+                   &vp9_idct8x8_64_add_c,
                    &vp9_idct8x8_12_add_neon,
                    TX_8X8, 12),
-        make_tuple(&vp9_idct8x8_64_add_c,
+        make_tuple(&vp9_fdct8x8_c,
+                   &vp9_idct8x8_64_add_c,
                    &vp9_idct8x8_1_add_neon,
                    TX_8X8, 1),
-        make_tuple(&vp9_idct4x4_16_add_c,
+        make_tuple(&vp9_fdct4x4_c,
+                   &vp9_idct4x4_16_add_c,
                    &vp9_idct4x4_1_add_neon,
                    TX_4X4, 1)));
 #endif
@@ -168,25 +263,32 @@ INSTANTIATE_TEST_CASE_P(
 INSTANTIATE_TEST_CASE_P(
     SSE2, PartialIDctTest,
     ::testing::Values(
-        make_tuple(&vp9_idct32x32_1024_add_c,
+        make_tuple(&vp9_fdct32x32_c,
+                   &vp9_idct32x32_1024_add_c,
                    &vp9_idct32x32_34_add_sse2,
                    TX_32X32, 34),
-        make_tuple(&vp9_idct32x32_1024_add_c,
+        make_tuple(&vp9_fdct32x32_c,
+                   &vp9_idct32x32_1024_add_c,
                    &vp9_idct32x32_1_add_sse2,
                    TX_32X32, 1),
-        make_tuple(&vp9_idct16x16_256_add_c,
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_c,
                    &vp9_idct16x16_10_add_sse2,
                    TX_16X16, 10),
-        make_tuple(&vp9_idct16x16_256_add_c,
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_c,
                    &vp9_idct16x16_1_add_sse2,
                    TX_16X16, 1),
-        make_tuple(&vp9_idct8x8_64_add_c,
+        make_tuple(&vp9_fdct8x8_c,
+                   &vp9_idct8x8_64_add_c,
                    &vp9_idct8x8_12_add_sse2,
                    TX_8X8, 12),
-        make_tuple(&vp9_idct8x8_64_add_c,
+        make_tuple(&vp9_fdct8x8_c,
+                   &vp9_idct8x8_64_add_c,
                    &vp9_idct8x8_1_add_sse2,
                    TX_8X8, 1),
-        make_tuple(&vp9_idct4x4_16_add_c,
+        make_tuple(&vp9_fdct4x4_c,
+                   &vp9_idct4x4_16_add_c,
                    &vp9_idct4x4_1_add_sse2,
                    TX_4X4, 1)));
 #endif
@@ -195,8 +297,19 @@ INSTANTIATE_TEST_CASE_P(
 INSTANTIATE_TEST_CASE_P(
     SSSE3_64, PartialIDctTest,
     ::testing::Values(
-        make_tuple(&vp9_idct8x8_64_add_c,
+        make_tuple(&vp9_fdct8x8_c,
+                   &vp9_idct8x8_64_add_c,
                    &vp9_idct8x8_12_add_ssse3,
                    TX_8X8, 12)));
 #endif
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, PartialIDctTest,
+    ::testing::Values(
+        make_tuple(&vp9_fdct16x16_c,
+                   &vp9_idct16x16_256_add_c,
+                   &vp9_idct16x16_10_add_ssse3,
+                   TX_16X16, 10)));
+#endif
 }  // namespace