Merge "ppc: Add vpx_sadnxmx4d_vsx for n,m = {8, 16, 32 ,64}"

[libvpx] / test / partial_idct_test.cc

diff --git a/test/partial_idct_test.cc b/test/partial_idct_test.cc
index 2dda7da4520075fbef59c3bcba483ac90c6cf119..c76308bbb7f12ba7ab0e9ee30db433df68df9944 100644 (file)
--- a/test/partial_idct_test.cc
+++ b/test/partial_idct_test.cc
@@ -128,12 +128,12 @@ class PartialIDctTest : public ::testing::TestWithParam<PartialInvTxfmParam> {
   }
 
   void InitInput() {
-    const int max_coeff = 32766 / 4;
-    int max_energy_leftover = max_coeff * max_coeff;
+    const int64_t max_coeff = (32766 << (bit_depth_ - 8)) / 4;
+    int64_t max_energy_leftover = max_coeff * max_coeff;
     for (int j = 0; j < last_nonzero_; ++j) {
-      int16_t coeff = static_cast<int16_t>(sqrt(1.0 * max_energy_leftover) *
-                                           (rnd_.Rand16() - 32768) / 65536);
-      max_energy_leftover -= coeff * coeff;
+      tran_low_t coeff = static_cast<tran_low_t>(
+          sqrt(1.0 * max_energy_leftover) * (rnd_.Rand16() - 32768) / 65536);
+      max_energy_leftover -= static_cast<int64_t>(coeff) * coeff;
       if (max_energy_leftover < 0) {
         max_energy_leftover = 0;
         coeff = 0;
@@ -142,6 +142,36 @@ class PartialIDctTest : public ::testing::TestWithParam<PartialInvTxfmParam> {
     }
   }
 
+  void PrintDiff() {
+    if (memcmp(output_block_ref_, output_block_,
+               pixel_size_ * output_block_size_)) {
+      uint16_t ref, opt;
+      for (int y = 0; y < size_; y++) {
+        for (int x = 0; x < size_; x++) {
+          if (pixel_size_ == 1) {
+            ref = output_block_ref_[y * stride_ + x];
+            opt = output_block_[y * stride_ + x];
+          } else {
+            ref = reinterpret_cast<uint16_t *>(
+                output_block_ref_)[y * stride_ + x];
+            opt = reinterpret_cast<uint16_t *>(output_block_)[y * stride_ + x];
+          }
+          if (ref != opt) {
+            printf("dest[%d][%d] diff:%6d (ref),%6d (opt)\n", y, x, ref, opt);
+          }
+        }
+      }
+
+      printf("\ninput_block_:\n");
+      for (int y = 0; y < size_; y++) {
+        for (int x = 0; x < size_; x++) {
+          printf("%6d,", input_block_[y * size_ + x]);
+        }
+        printf("\n");
+      }
+    }
+  }
+
  protected:
   int last_nonzero_;
   TX_SIZE tx_size_;
@@ -162,23 +192,32 @@ class PartialIDctTest : public ::testing::TestWithParam<PartialInvTxfmParam> {
 };
 
 TEST_P(PartialIDctTest, RunQuantCheck) {
+  const int count_test_block = (size_ != 4) ? kCountTestBlock : 65536;
   DECLARE_ALIGNED(16, int16_t, input_extreme_block[kMaxNumCoeffs]);
   DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kMaxNumCoeffs]);
 
   InitMem();
-  for (int i = 0; i < kCountTestBlock; ++i) {
+
+  for (int i = 0; i < count_test_block; ++i) {
     // Initialize a test block with input range [-mask_, mask_].
-    if (i == 0) {
-      for (int k = 0; k < input_block_size_; ++k) {
-        input_extreme_block[k] = mask_;
-      }
-    } else if (i == 1) {
-      for (int k = 0; k < input_block_size_; ++k) {
-        input_extreme_block[k] = -mask_;
+    if (size_ != 4) {
+      if (i == 0) {
+        for (int k = 0; k < input_block_size_; ++k) {
+          input_extreme_block[k] = mask_;
+        }
+      } else if (i == 1) {
+        for (int k = 0; k < input_block_size_; ++k) {
+          input_extreme_block[k] = -mask_;
+        }
+      } else {
+        for (int k = 0; k < input_block_size_; ++k) {
+          input_extreme_block[k] = rnd_.Rand8() % 2 ? mask_ : -mask_;
+        }
       }
     } else {
+      // Try all possible combinations.
       for (int k = 0; k < input_block_size_; ++k) {
-        input_extreme_block[k] = rnd_.Rand8() % 2 ? mask_ : -mask_;
+        input_extreme_block[k] = (i & (1 << k)) ? mask_ : -mask_;
       }
     }
 
@@ -277,9 +316,9 @@ TEST_P(PartialIDctTest, DISABLED_Speed) {
   vpx_usec_timer_mark(&timer);
   const int elapsed_time =
       static_cast<int>(vpx_usec_timer_elapsed(&timer) / 1000);
-  printf("idct%dx%d_%d (bitdepth %d) time: %5d ms\n", size_, size_,
-         last_nonzero_, bit_depth_, elapsed_time);
-
+  printf("idct%dx%d_%d (%s %d) time: %5d ms\n", size_, size_, last_nonzero_,
+         (pixel_size_ == 1) ? "bitdepth" : "high bitdepth", bit_depth_,
+         elapsed_time);
   ASSERT_EQ(0, memcmp(output_block_ref_, output_block_,
                       pixel_size_ * output_block_size_))
       << "Error: partial inverse transform produces different results";
@@ -619,6 +658,15 @@ const PartialInvTxfmParam sse2_partial_idct_tests[] = {
   make_tuple(
       &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
       &highbd_wrapper<vpx_highbd_idct16x16_10_add_sse2>, TX_16X16, 10, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_sse2>, TX_16X16, 1, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_sse2>, TX_16X16, 1, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_sse2>, TX_16X16, 1, 12, 2),
   make_tuple(&vpx_highbd_fdct8x8_c,
              &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
              &highbd_wrapper<vpx_highbd_idct8x8_64_add_sse2>, TX_8X8, 64, 8, 2),
@@ -637,6 +685,12 @@ const PartialInvTxfmParam sse2_partial_idct_tests[] = {
   make_tuple(
       &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_12_add_c>,
       &highbd_wrapper<vpx_highbd_idct8x8_12_add_sse2>, TX_8X8, 12, 12, 2),
+  make_tuple(&vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct8x8_1_add_sse2>, TX_8X8, 1, 8, 2),
+  make_tuple(&vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct8x8_1_add_sse2>, TX_8X8, 1, 10, 2),
+  make_tuple(&vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct8x8_1_add_sse2>, TX_8X8, 1, 12, 2),
   make_tuple(&vpx_highbd_fdct4x4_c,
              &highbd_wrapper<vpx_highbd_idct4x4_16_add_c>,
              &highbd_wrapper<vpx_highbd_idct4x4_16_add_sse2>, TX_4X4, 16, 8, 2),
@@ -646,6 +700,12 @@ const PartialInvTxfmParam sse2_partial_idct_tests[] = {
   make_tuple(
       &vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_16_add_c>,
       &highbd_wrapper<vpx_highbd_idct4x4_16_add_sse2>, TX_4X4, 16, 12, 2),
+  make_tuple(&vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct4x4_1_add_sse2>, TX_4X4, 1, 8, 2),
+  make_tuple(&vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct4x4_1_add_sse2>, TX_4X4, 1, 10, 2),
+  make_tuple(&vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct4x4_1_add_sse2>, TX_4X4, 1, 12, 2),
 #endif  // CONFIG_VP9_HIGHBITDEPTH
   make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
              &wrapper<vpx_idct32x32_1024_add_sse2>, TX_32X32, 1024, 8, 1),