x86: sbrdsp: Implement SSE2 qmf_deint_bfly

Sandybridge: 47 cycles

Having a loop counter is a 7 cycle gain.
Unrolling is another 7 cycle gain.
Working in reverse scan is another 6 cycles.
Signed-off-by: Diego Biurrun <diego@biurrun.de>

libavcodec/x86/sbrdsp.asm

@@ -245,3 +245,31 @@ cglobal sbr_neg_odd_64, 1,2,4,z
     cmp         zq, r1q
     jne      .loop
     REP_RET
+
+INIT_XMM sse2
+; 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              m2, [src0q+cq+mmsize]
+    mova              m3, [src1q+mmsize]
+    pshufd            m4, m0, q0123
+    pshufd            m5, m1, q0123
+    pshufd            m6, m2, q0123
+    pshufd            m7, m3, q0123
+    addps             m3, m4
+    subps             m0, m7
+    addps             m1, m6
+    subps             m2, m5
+    mova         [vrevq], m1
+    mova  [vrevq+mmsize], m3
+    mova         [vq+cq], m0
+    mova  [vq+cq+mmsize], m2
+    add            src1q, 2*mmsize
+    add            vrevq, 2*mmsize
+    sub               cq, 2*mmsize
+    jge            .loop
+    REP_RET

libavcodec/x86/sbrdsp_init.c

@@ -34,6 +34,7 @@ void ff_sbr_hf_gen_sse(float (*X_high)[2], const float (*X_low)[2],
                        float bw, int start, int end);
 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_deint_bfly_sse2(float *v, const float *src0, const float *src1);
 
 av_cold void ff_sbrdsp_init_x86(SBRDSPContext *s)
 {
@@ -47,4 +48,8 @@ av_cold void ff_sbrdsp_init_x86(SBRDSPContext *s)
         s->hf_gen     = ff_sbr_hf_gen_sse;
         s->qmf_post_shuffle = ff_sbr_qmf_post_shuffle_sse;
     }
+
+    if (EXTERNAL_SSE2(mm_flags)) {
+        s->qmf_deint_bfly   = ff_sbr_qmf_deint_bfly_sse2;
+    }
 }