lavu/tx: invert permutation lookups

out[lut[i]] = in[i] lookups were 4.04 times(!) slower than
out[i] = in[lut[i]] lookups for an out-of-place FFT of length 4096.

The permutes remain unchanged for anything but out-of-place monolithic
FFT, as those benefit quite a lot from the current order (it means
there's only 1 lookup necessary to add to an offset, rather than
a full gather).

The code was based around non-power-of-two FFTs, so this wasn't
benchmarked early on.

1 files changed, 5 insertions, 2 deletions

diff --git a/libavutil/tx.c b/libavutil/tx.c
index ac67b354be..1161df3285 100644
--- a/libavutil/tx.c
+++ b/libavutil/tx.c
@@ -91,7 +91,7 @@ int ff_tx_gen_compound_mapping(AVTXContext *s)
     return 0;
 }
 
-int ff_tx_gen_ptwo_revtab(AVTXContext *s)
+int ff_tx_gen_ptwo_revtab(AVTXContext *s, int invert_lookup)
 {
     const int m = s->m, inv = s->inv;
 
@@ -101,7 +101,10 @@ int ff_tx_gen_ptwo_revtab(AVTXContext *s)
     /* Default */
     for (int i = 0; i < m; i++) {
         int k = -split_radix_permutation(i, m, inv) & (m - 1);
-        s->revtab[k] = i;
+        if (invert_lookup)
+            s->revtab[i] = k;
+        else
+            s->revtab[k] = i;
     }
 
     return 0;