+ CHROMA_H_STORE r5
+%if mmsize == 8
+ lea r0, [r0+r1*(mmsize/4)]
+ dec r4
+ jg .loop
+%endif
+ RET
+
+;-----------------------------------------------------------------------------
+; void deblock_h_chroma_mbaff( uint16_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
+;-----------------------------------------------------------------------------
+cglobal deblock_h_chroma_mbaff, 5,7,8
+ add r1, r1
+ lea r6, [r1*3]
+%if mmsize == 8
+ mov r5, 16/mmsize
+.loop:
+%endif
+ CHROMA_H_LOAD r6
+ LOAD_AB m4, m5, r2d, r3d
+ LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
+ movd m6, [r4]
+ punpcklbw m6, m6
+ psraw m6, 8
+ punpcklwd m6, m6
+ pand m7, m6
+ DEBLOCK_P0_Q0 m1, m2, m0, m3, m7, m5, m6
+ CHROMA_H_STORE r6
+%if mmsize == 8
+ lea r0, [r0+r1*(mmsize/4)]
+ add r4, mmsize/4
+ dec r5
+ jg .loop
+%endif
+ RET
+
+;-----------------------------------------------------------------------------
+; void deblock_h_chroma_422_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
+;-----------------------------------------------------------------------------
+cglobal deblock_h_chroma_422_intra, 4,6,8
+ add r1, r1
+ mov r4, 64/mmsize
+%if mmsize == 16
+ lea r5, [r1*3]
+%endif
+.loop:
+ CHROMA_H_LOAD r5
+ call deblock_intra_body
+ CHROMA_H_STORE r5
+ lea r0, [r0+r1*(mmsize/4)]
+ dec r4
+ jg .loop
+ RET
+
+;-----------------------------------------------------------------------------
+; void deblock_h_chroma_422( uint16_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
+;-----------------------------------------------------------------------------
+cglobal deblock_h_chroma_422, 5,7,8
+ add r1, r1
+ mov r5, 64/mmsize
+ lea r6, [r1*3]
+.loop:
+ CHROMA_H_LOAD r6
+ LOAD_AB m4, m5, r2m, r3d
+ LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
+ pxor m4, m4
+ movd m6, [r4-1]
+ psraw m6, 8
+ SPLATW m6, m6
+ pmaxsw m6, m4
+ pand m7, m6
+ DEBLOCK_P0_Q0 m1, m2, m0, m3, m7, m5, m6
+ CHROMA_H_STORE r6
+ lea r0, [r0+r1*(mmsize/4)]
+%if mmsize == 16
+ inc r4
+%else
+ mov r2, r5
+ and r2, 1
+ add r4, r2 ; increment once every 2 iterations
+%endif
+ dec r5
+ jg .loop
+ RET
+%endmacro ; DEBLOCK_CHROMA