shithub: libvpx

Info  •  Files  •  Log  •  Branches

All zero coeff skip in IDCT 32x32

When all coefficients are zeros, skip the corresponding 1-D inverse
transform. This practice has been used in the SSE2 implementation of
inverse 32x32 DCT. This commit imports this algorithm into the C code.

Change-Id: I0f58bfcb183a569fab85d524d5d9cf8ae8653f86

--- a/vp9/common/vp9_idct.c

+++ b/vp9/common/vp9_idct.c

@@ -1253,7 +1253,20 @@

   // Rows

   for (i = 0; i < 32; ++i) {

-    idct32_1d(input, outptr);

+    int16_t zero_coeff[16];

+    for (j = 0; j < 16; ++j)

+      zero_coeff[j] = input[2 * j] | input[2 * j + 1];

+    for (j = 0; j < 8; ++j)

+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];

+    for (j = 0; j < 4; ++j)

+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];

+    for (j = 0; j < 2; ++j)

+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];

+

+    if (zero_coeff[0] | zero_coeff[1])

+      idct32_1d(input, outptr);

+    else

+      vpx_memset(outptr, 0, sizeof(int16_t) * 32);

     input += 32;

     outptr += 32;

   }