hevc: cleanups in SSE2 and SSSE3 loop filters, use fewer instructions

cherry picked from commit f7843356253459e6010320292dbbc1e888a5249b
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Anton Khirnov
2014-07-19 14:18:03 +02:00
committed by Michael Niedermayer
parent e8581b17a8
commit b435043abb

View File

@@ -715,10 +715,9 @@ cglobal hevc_h_loop_filter_chroma_8, 3, 4, 7, pix, stride, tc, pix0
punpcklbw m2, m5 punpcklbw m2, m5
punpcklbw m3, m5 punpcklbw m3, m5
CHROMA_DEBLOCK_BODY 8 CHROMA_DEBLOCK_BODY 8
packuswb m1, m1 ; p0' packed in bytes on low quadword packuswb m1, m2
packuswb m2, m2 ; q0' packed in bytes on low quadword movh[pix0q+strideq], m1
movq [pix0q+strideq], m1 movhps [pixq], m1
movq [pixq], m2
RET RET
cglobal hevc_h_loop_filter_chroma_10, 3, 4, 7, pix, stride, tc, pix0 cglobal hevc_h_loop_filter_chroma_10, 3, 4, 7, pix, stride, tc, pix0
@@ -793,18 +792,15 @@ cglobal hevc_h_loop_filter_luma_8, 4, 15, 16, pix, stride, beta, tc, count, pix0
punpcklbw m7, m8 punpcklbw m7, m8
LUMA_DEBLOCK_BODY 8, h LUMA_DEBLOCK_BODY 8, h
.store: .store:
packuswb m1, m1; p2 packuswb m1, m2
packuswb m2, m2; p1 packuswb m3, m4
packuswb m3, m3; p0 packuswb m5, m6
packuswb m4, m4; q0 movh [r5 + r1], m1
packuswb m5, m5; q1 movhps [r5 + 2 * r1], m1
packuswb m6, m6; q2 movh [r5 + r6], m3
movq [r5+r1], m1; p2 movhps [r0 ], m3
movq [r5+2*r1], m2; p1 movh [r0 + r1], m5
movq [r5+r6], m3; p0 movhps [r0 + 2 * r1], m5
movq [r0], m4; q0
movq [r0+r1], m5; q1
movq [r0+2*r1], m6; q2
.bypassluma: .bypassluma:
RET RET