Skip to content

x86: Increase precision of IDCT intermediates

Henrik Gramner requested to merge gramner/dav1d:avx2_idct_precision into master

Fixes #299 (closed) for the SSSE3 and AVX2 asm at a slight performance hit:

AVX2       4x4    8x8  16x16  32x32  64x64
Before:   41.0  103.8  221.7 1001.3 3031.2
After:    45.8  108.6  251.7 1097.0 3293.4
Edited by Henrik Gramner

Merge request reports