>From 713 to 209 cycles on Penrynn.
Having a loop counter is a 7 cycle gain.
Unrolling is another 7 cycle gain.
Working in reverse scan is another 6 cycles.
---
 libavcodec/x86/sbrdsp.asm    |   31 +++++++++++++++++++++++++++++++
 libavcodec/x86/sbrdsp_init.c |    2 ++
 2 files changed, 33 insertions(+), 0 deletions(-)

diff --git a/libavcodec/x86/sbrdsp.asm b/libavcodec/x86/sbrdsp.asm
index 49dd78c..cfbd6e8 100644
--- a/libavcodec/x86/sbrdsp.asm
+++ b/libavcodec/x86/sbrdsp.asm
@@ -287,3 +287,34 @@ cglobal sbr_qmf_deint_neg, 2,3,4,v,src,vrev
   cmp        vq, vrevq
   jl      .loop
   REP_RET
+
+; sbr_qmf_deint_bfly(float *v, const float *src0, const float *src1)
+cglobal sbr_qmf_deint_bfly, 3,5,8, v,src0,src1,vrev,c
+  mov        cq, 64*4-2*mmsize
+  lea     vrevq, [vq + 64*4]
+.loop:
+  mova       m0, [src0q+cq]
+  mova       m1, [src1q]
+  mova       m4, [src0q+cq+mmsize]
+  mova       m5, [src1q+mmsize]
+  mova       m2, m0
+  mova       m3, m1
+  shufps     m2, m2, 11011b
+  shufps     m3, m3, 11011b
+  mova       m6, m4
+  mova       m7, m5
+  shufps     m6, m6, 11011b
+  shufps     m7, m7, 11011b
+  addps      m5, m2
+  subps      m0, m7
+  addps      m1, m6
+  subps      m4, m3
+  mova  [vrevq], m1
+  mova  [vrevq+mmsize], m5
+  mova  [vq+cq], m0
+  mova  [vq+cq+mmsize], m4
+  add     src1q, 2*mmsize
+  add     vrevq, 2*mmsize
+  sub        cq, 2*mmsize
+  jge     .loop
+  REP_RET
diff --git a/libavcodec/x86/sbrdsp_init.c b/libavcodec/x86/sbrdsp_init.c
index 1ac64aa..5e3e131 100644
--- a/libavcodec/x86/sbrdsp_init.c
+++ b/libavcodec/x86/sbrdsp_init.c
@@ -35,6 +35,7 @@ void ff_sbr_neg_odd_64_sse(float *z);
 void ff_sbr_qmf_post_shuffle_sse(float W[32][2], const float *z);
 void ff_sbr_qmf_pre_shuffle_sse(float *z);
 void ff_sbr_qmf_deint_neg_sse(float *v, const float *src);
+void ff_sbr_qmf_deint_bfly_sse(float *v, const float *src0, const float *src1);
 
 void ff_sbrdsp_init_x86(SBRDSPContext *s)
 {
@@ -49,5 +50,6 @@ void ff_sbrdsp_init_x86(SBRDSPContext *s)
         s->qmf_post_shuffle = ff_sbr_qmf_post_shuffle_sse;
         s->qmf_pre_shuffle = ff_sbr_qmf_pre_shuffle_sse;
         s->qmf_deint_neg = ff_sbr_qmf_deint_neg_sse;
+        s->qmf_deint_bfly = ff_sbr_qmf_deint_bfly_sse;
     }
 }
-- 
1.7.7.msysgit.0

_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to