[x264-devel] Yet another updated threaded slicetype patch (v14)

[Steven Walters]

> I seriously doubt this, as I have never found a single function that
> goes faster in 3.4 than in 4.3, but I have found a huge number of
> places where 3.4 is phenomenally braindead in terms of code
> generation.
>
> Dark Shikari

here's a diff (-U0) of the log of the runs comparing gcc 3.4.5 and
4.4.0 ran on the foreman cif
copying settings and such from make fprofile (which totals 8 runs)
--- x264_pentium2_3.4.5_bench.log       2009-05-11 12:35:52 -0400
+++ x264_pentium2_4.4.0_bench.log       2009-05-11 12:35:52 -0400
@@ -2 +2 @@
-built on May 11 2009, gcc: 3.4.5 (mingw-vista special r3)
+built on May 11 2009, gcc: 4.4.0
@@ -43 +43 @@
-encoded 300 frames, 128.87 fps, 1334.11 kb/s
+encoded 300 frames, 127.12 fps, 1334.11 kb/s
@@ -62 +62 @@
-encoded 300 frames, 135.20 fps, 277.43 kb/s
+encoded 300 frames, 134.29 fps, 277.43 kb/s
@@ -84 +84 @@
-encoded 300 frames, 27.16 fps, 897.61 kb/s
+encoded 300 frames, 26.41 fps, 897.61 kb/s
@@ -104 +104 @@
-encoded 300 frames, 19.07 fps, 455.75 kb/s
+encoded 300 frames, 19.10 fps, 455.75 kb/s
@@ -124 +124 @@
-encoded 50 frames, 28.84 fps, 327.25 kb/s
+encoded 50 frames, 29.36 fps, 327.25 kb/s
@@ -151 +151 @@
-encoded 50 frames, 160.26 fps, 12437.70 kb/s
+encoded 50 frames, 177.94 fps, 12437.70 kb/s


and a comparison of a 640x480 CG video i have:
--- x264_pentium2_3.4.5_bench.log       2009-05-11 12:44:33 -0400
+++ x264_pentium2_4.4.0_bench.log       2009-05-11 12:44:32 -0400
@@ -2 +2 @@
-built on May 11 2009, gcc: 3.4.5 (mingw-vista special r3)
+built on May 11 2009, gcc: 4.4.0
@@ -23 +23 @@
-encoded 500 frames, 133.90 fps, 521.32 kb/s
+encoded 500 frames, 133.33 fps, 521.32 kb/s
@@ -43 +43 @@
-encoded 500 frames, 55.27 fps, 2010.80 kb/s
+encoded 500 frames, 54.70 fps, 2010.80 kb/s
@@ -62 +62 @@
-encoded 500 frames, 55.94 fps, 707.92 kb/s
+encoded 500 frames, 55.27 fps, 707.92 kb/s
@@ -84 +84 @@
-encoded 500 frames, 13.66 fps, 1448.21 kb/s
+encoded 500 frames, 13.10 fps, 1448.21 kb/s
@@ -104 +104 @@
-encoded 500 frames, 11.30 fps, 839.88 kb/s
+encoded 500 frames, 11.31 fps, 839.88 kb/s
@@ -124 +124 @@
-encoded 50 frames, 12.08 fps, 661.90 kb/s
+encoded 50 frames, 12.90 fps, 661.90 kb/s
@@ -138 +138 @@
-encoded 50 frames, 18.71 fps, 12647.28 kb/s
+encoded 50 frames, 18.82 fps, 12647.28 kb/s
@@ -151 +151 @@
-encoded 50 frames, 74.40 fps, 15368.83 kb/s
+encoded 50 frames, 76.22 fps, 15368.83 kb/s

tells me that the braindead code is running faster.
either way, people are going to need to update once this patch goes
through if they want a correct bitstream...
and the differences are IMO a minor % difference so i don't think
it'll be THAT noticeable to most people.