x86/vorbisdsp: optimize ff_vorbis_inverse_coupling_sse

About 7% faster.

libavcodec/x86/vorbisdsp.asm

@@ -57,13 +57,17 @@ cglobal vorbis_inverse_coupling, 3, 3, 6, mag, ang, block_size
 %endif
 
 INIT_XMM sse
-cglobal vorbis_inverse_coupling, 3, 4, 6, mag, ang, block_size, cntr
+cglobal vorbis_inverse_coupling, 3, 3, 6, mag, ang, block_size
     mova                     m5, [pdw_80000000]
-    xor                   cntrq, cntrq
+    shl             block_sized, 2
+    add                    magq, block_sizeq
+    add                    angq, block_sizeq
+    neg             block_sizeq
+
 align 16
 .loop:
-    mova                     m0, [magq+cntrq*4]
-    mova                     m1, [angq+cntrq*4]
+    mova                     m0, [magq+block_sizeq]
+    mova                     m1, [angq+block_sizeq]
     xorps                    m2, m2
     xorps                    m3, m3
     cmpleps                  m2, m0     ; m <= 0.0
@@ -75,9 +79,8 @@ align 16
     andnps                   m4, m1
     addps                    m3, m0     ; a = m + ((a < 0) & (a ^ sign(m)))
     subps                    m0, m4     ; m = m + ((a > 0) & (a ^ sign(m)))
-    mova         [angq+cntrq*4], m3
-    mova         [magq+cntrq*4], m0
-    add                   cntrq, 4
-    cmp                   cntrq, block_sizeq
+    mova     [angq+block_sizeq], m3
+    mova     [magq+block_sizeq], m0
+    add             block_sizeq, mmsize
     jl .loop
     RET