RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-27 12:07:45 +00:00
Code Issues Activity Actions

LibVideo/VP9: Make a lookup table for bit reversals

Browse source 
Bit reversals are used very often in intra-predicted frames. Turning
these into a constexpr lookup table reduces the branching needed for
block transforms significantly. This reduces the times spent decoding
an intra-heavy 1080p video by about 9% (~14.3s -> ~12.9s).
This commit is contained in:
Zaggy1024 2023-04-16 10:03:21 -05:00 • committed by Tim Flynn
parent f6764beead
commit 6e6cc1ddb2
2 changed files with 21 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

4
Userland/Libraries/LibVideo/VP9/Decoder.cpp
View file

@ -1400,7 +1400,7 @@ inline DecoderErrorOr<void> Decoder::inverse_discrete_cosine_transform_array_per
// 1.2. T[ i ] is set equal to copyT[ brev( n, i ) ] for i = 0..((1<<n) - 1). // 1.2. T[ i ] is set equal to copyT[ brev( n, i ) ] for i = 0..((1<<n) - 1).
for (auto i = 0u; i < block_size; i++) for (auto i = 0u; i < block_size; i++)
data[i] = data_copy[brev(log2_of_block_size, i)]; data[i] = data_copy[brev<log2_of_block_size>(i)];
return {}; return {};
} }
@ -1432,7 +1432,7 @@ inline DecoderErrorOr<void> Decoder::inverse_discrete_cosine_transform(Span<Inte
// 2.6 Invoke B( n1+i, n0-1-i, 32-brev( 5, n1+i), 0 ) for i = 0..(n2-1). // 2.6 Invoke B( n1+i, n0-1-i, 32-brev( 5, n1+i), 0 ) for i = 0..(n2-1).
for (auto i = 0u; i < quarter_block_size; i++) { for (auto i = 0u; i < quarter_block_size; i++) {
auto index = half_block_size + i; auto index = half_block_size + i;
butterfly_rotation_in_place(data, index, block_size - 1 - i, 32 - brev(5, index), false); butterfly_rotation_in_place(data, index, block_size - 1 - i, 32 - brev<5>(index), false);
} }
// 2.7 If n is greater than or equal to 3: // 2.7 If n is greater than or equal to 3:

25
Userland/Libraries/LibVideo/VP9/Utilities.h
View file

@ -34,15 +34,26 @@ u16 clip_1(u8 bit_depth, T x)
return x; return x;
} }
template<typename T, typename C> template<u8 bits>
inline T brev(C bit_count, T value) inline u8 brev(u8 value)
{ {
T result = 0; static_assert(bits <= 8, "brev() expects an 8-bit value.");
for (C i = 0; i < bit_count; i++) {
auto bit = (value >> i) & 1; static constexpr auto lookup_table = [] {
result |= bit << (bit_count - 1 - i); constexpr size_t value_count = 1 << bits;
Array<u8, value_count> the_table;
for (u8 lookup_value = 0; lookup_value < value_count; lookup_value++) {
u8 reversed = 0;
for (u8 bit_index = 0; bit_index < bits; bit_index++) {
auto bit = (lookup_value >> bit_index) & 1;
reversed |= bit << (bits - 1 - bit_index);
} }
return result; the_table[lookup_value] = reversed;
}
return the_table;
}();
return lookup_table[value];
} }
inline BlockSubsize get_subsampled_block_size(BlockSubsize size, bool subsampling_x, bool subsampling_y) inline BlockSubsize get_subsampled_block_size(BlockSubsize size, bool subsampling_x, bool subsampling_y)