From 6e6cc1ddb2dc6b273d9e761eef9e6fb56dbbe455 Mon Sep 17 00:00:00 2001
From: Zaggy1024 <zaggy1024@gmail.com>
Date: Sun, 16 Apr 2023 10:03:21 -0500
Subject: [PATCH] LibVideo/VP9: Make a lookup table for bit reversals

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).
---
 Userland/Libraries/LibVideo/VP9/Decoder.cpp |  4 +--
 Userland/Libraries/LibVideo/VP9/Utilities.h | 27 +++++++++++++++------
 2 files changed, 21 insertions(+), 10 deletions(-)

diff --git a/Userland/Libraries/LibVideo/VP9/Decoder.cpp b/Userland/Libraries/LibVideo/VP9/Decoder.cpp
index a58526c0ae..1a88774f22 100644
--- a/Userland/Libraries/LibVideo/VP9/Decoder.cpp
+++ b/Userland/Libraries/LibVideo/VP9/Decoder.cpp
@@ -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).
     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 {};
 }
@@ -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).
     for (auto i = 0u; i < quarter_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:
diff --git a/Userland/Libraries/LibVideo/VP9/Utilities.h b/Userland/Libraries/LibVideo/VP9/Utilities.h
index ff79929fa9..726331e05b 100644
--- a/Userland/Libraries/LibVideo/VP9/Utilities.h
+++ b/Userland/Libraries/LibVideo/VP9/Utilities.h
@@ -34,15 +34,26 @@ u16 clip_1(u8 bit_depth, T x)
     return x;
 }
 
-template<typename T, typename C>
-inline T brev(C bit_count, T value)
+template<u8 bits>
+inline u8 brev(u8 value)
 {
-    T result = 0;
-    for (C i = 0; i < bit_count; i++) {
-        auto bit = (value >> i) & 1;
-        result |= bit << (bit_count - 1 - i);
-    }
-    return result;
+    static_assert(bits <= 8, "brev() expects an 8-bit value.");
+
+    static constexpr auto lookup_table = [] {
+        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);
+            }
+            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)