diff options
| author | James Almer <jamrial@gmail.com> | 2014-02-27 14:40:28 -0300 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2014-02-27 23:28:15 +0100 |
| commit | 2163a40a4693f309dfff14ae64aaa80f841d8a22 (patch) | |
| tree | 81b0b72679e5ce5a5e751c1938565c19816a8e05 | |
| parent | fbf98375e481980d7c19fdb9ccf6f9134b9050bb (diff) | |
| download | ffmpeg-2163a40a4693f309dfff14ae64aaa80f841d8a22.tar.gz | |
x86/imdct36: use sse3 instructions in the last BUTTERF step when possible
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
| -rw-r--r-- | libavcodec/x86/imdct36.asm | 21 |
1 files changed, 15 insertions, 6 deletions
diff --git a/libavcodec/x86/imdct36.asm b/libavcodec/x86/imdct36.asm index 1d58d6bcb9..ce30b42103 100644 --- a/libavcodec/x86/imdct36.asm +++ b/libavcodec/x86/imdct36.asm @@ -50,7 +50,7 @@ ps_cosh_sse3: dd 1.0, -0.50190991877167369479, 1.0, -5.73685662283492756461 dd 1.0, -0.51763809020504152469, 1.0, -1.93185165257813657349 dd 1.0, -0.55168895948124587824, -1.0, 1.18310079157624925896 dd 1.0, -0.61038729438072803416, -1.0, 0.87172339781054900991 - dd 1.0, 0.70710678118654752439, 0.0, 0.0 + dd 1.0, -0.70710678118654752439, 0.0, 0.0 costabs: times 4 dd 0.98480773 times 4 dd 0.93969262 @@ -129,6 +129,19 @@ SECTION_TEXT %endif %endmacro +%macro BUTTERF2 3 +%if cpuflag(sse3) + mulps %1, %1, [ps_cosh_sse3 + %3] + PSHUFD %2, %1, 0xe1 + addsubps %1, %1, %2 +%else + mulps %1, [ps_cosh + %3] + PSHUFD %2, %1, 0xe1 + xorps %1, [ps_p1m1p1m1] + addps %1, %2 +%endif +%endmacro + %macro STORE 4 movhlps %2, %1 movss [%3 ], %1 @@ -279,11 +292,7 @@ cglobal imdct36_float, 4,4,9, out, buf, in, win BUTTERF m7, m2, 16 BUTTERF m3, m6, 32 BUTTERF m4, m1, 48 - - mulps m5, m5, [ps_cosh + 64] - PSHUFD m1, m5, 0xe1 - xorps m5, m5, [ps_p1m1p1m1] - addps m5, m5, m1 + BUTTERF2 m5, m1, 64 ; permutates: ; m0 0 1 2 3 => 2 6 10 14 m1 |