Added a bit-exact, integer, reversible, orthonormal 32-point DCT based on Loeffler's factorization with 215 adds, 38 shifts and 87 fixed-point multiplies. This patch does not enable testing of the 64-point DCT because does not use truncated constants and is known to overflow.
MSE: 0.00019605494433023361528292682898211
cc32e01 Add integer 64-point DCT.
src/dct.c | 3157 +++++++++++++++++++++++++++++++++++++++++++---
tools/gen_dct_template.c | 204 +++
2 files changed, 3169 insertions(+), 192 deletions(-)
Upstream: git.xiph.org