fmt_conv_ops: Fix slow and corrupted default_all_to_all

1. While dividing `int32_t` by `size_t`, we should convert `size_t`
   to `int32_t` explicitly to prevent corrupted results.
2. Swap divide and summation order to preserve sample accuracy.
3. Reduce time complexity from O(N * M * Frames) to O((N + M) * Frames)
4. Add unit test `6chTo8chS16LE` which goes trough all int16_t cases (
   from INT16_MIN to INT16_MAX).

BUG=b:163052978
TEST=Unit tests
TEST=`speaker-test -c 6 -D hw:0,0 -f 440 -t sine=use` in Crostini and
     check output results from 8-channel device.

Change-Id: I08d93c6f4a6c33d9ee69c1ecfc93b9a8be9b95d2
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/third_party/adhd/+/2341187
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Reviewed-by: Cheng-Yi Chiang <cychiang@chromium.org>
Tested-by: Chih-Yang Hsia <paulhsia@chromium.org>
Commit-Queue: Chih-Yang Hsia <paulhsia@chromium.org>

1 files changed, 33 insertions, 3 deletions

diff --git a/cras/src/tests/fmt_conv_ops_unittest.cc b/cras/src/tests/fmt_conv_ops_unittest.cc
index a76087f6..d8feeab3 100644
--- a/cras/src/tests/fmt_conv_ops_unittest.cc
+++ b/cras/src/tests/fmt_conv_ops_unittest.cc
@@ -619,10 +619,10 @@ TEST(FormatConverterOpsTest, StereoTo3chS16LE) {
   EXPECT_EQ(ret, frames);
 
   for (size_t i = 0; i < frames; ++i) {
+    int32_t sum = 0;
     for (size_t k = 0; k < in_ch; ++k)
-      src[i * in_ch + k] /= in_ch;
-    for (size_t k = 1; k < in_ch; ++k)
-      src[i * in_ch + 0] += src[i * in_ch + k];
+      sum += (int32_t)src[i * in_ch + k];
+    src[i * in_ch + 0] = (int16_t)(sum / (int32_t)in_ch);
   }
   for (size_t i = 0; i < frames; ++i) {
     for (size_t k = 0; k < out_ch; ++k)
@@ -630,6 +630,36 @@ TEST(FormatConverterOpsTest, StereoTo3chS16LE) {
   }
 }
 
+// Test 6ch to 8ch conversion.  S16_LE.
+TEST(FormatConverterOpsTest, 6chTo8chS16LE) {
+  const size_t frames = 65536;
+  const size_t in_ch = 6;
+  const size_t out_ch = 8;
+  struct cras_audio_format fmt = {
+      .format = SND_PCM_FORMAT_S16_LE,
+      .frame_rate = 48000,
+      .num_channels = 8,
+  };
+
+  S16LEPtr src = CreateS16LE(frames * in_ch);
+  S16LEPtr dst = CreateS16LE(frames * out_ch);
+  for (size_t i = 0; i < frames; ++i) {
+    for (size_t k = 0; k < in_ch; k++) {
+      src[i * in_ch + k] = (k == 0) ? (INT16_MIN + (int16_t)i) : 0;
+    }
+  }
+
+  size_t ret = s16_default_all_to_all(&fmt, in_ch, out_ch, (uint8_t*)src.get(),
+                                      frames, (uint8_t*)dst.get());
+  EXPECT_EQ(ret, frames);
+
+  for (size_t i = 0; i < frames; ++i) {
+    src[i * in_ch + 0] /= (int16_t)in_ch;
+    for (size_t k = 0; k < out_ch; ++k)
+      EXPECT_EQ(src[i * in_ch + 0], dst[i * out_ch + k]);
+  }
+}
+
 // Test Multiply with Coef.  S16_LE.
 TEST(FormatConverterOpsTest, MultiplyWithCoefS16LE) {
   const size_t buf_size = 4096;