diff options
author | Christophe Gisquet <christophe.gisquet@gmail.com> | 2014-05-17 14:22:39 +0400 |
---|---|---|
committer | Michael Niedermayer <michaelni@gmx.at> | 2014-05-28 19:39:34 +0400 |
commit | 0810608e23456eeff75b691c7c444762352ec196 (patch) | |
tree | 051bc242267b35fe353f96238c34d44c9fcbb884 /libavcodec/x86 | |
parent | ec33f59fed82e0a92912639c79ac1b3ac0f23c4a (diff) |
x86: hevc_mc: better register allocation
The xmm reg count was incorrect, and manual loading of the gprs
furthermore allows to noticeable reduce the number needed.
The modified functions are used in weighted prediction, so only a
few samples like WP_* exhibit a change. For this one and Win64
(some widths removed because of too few occurrences):
WP_A_Toshiba_3.bit, ff_hevc_put_hevc_uni_w
16 32
before: 2194 3872
after: 2119 3767
WP_B_Toshiba_3.bit, ff_hevc_put_hevc_bi_w
16 32 64
before: 2819 4960 9396
after: 2617 4788 9150
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec/x86')
-rw-r--r-- | libavcodec/x86/hevc_mc.asm | 79 |
1 files changed, 48 insertions, 31 deletions
diff --git a/libavcodec/x86/hevc_mc.asm b/libavcodec/x86/hevc_mc.asm index 8a605d759f..0c1a431bf1 100644 --- a/libavcodec/x86/hevc_mc.asm +++ b/libavcodec/x86/hevc_mc.asm @@ -1103,19 +1103,32 @@ cglobal hevc_put_hevc_bi_qpel_hv%1_%2, 9, 11, 16, dst, dststride, src, srcstride %endmacro %macro WEIGHTING_FUNCS 2 -cglobal hevc_put_hevc_uni_w%1_%2, 8, 10, 11, dst, dststride, src, srcstride, height, denom, wx, ox, shift - lea shiftd, [denomd+14-%2] ; shift = 14 - bitd + denom - shl oxd, %2-8 ; ox << (bitd - 8) - movd m2, wxd ; WX - movd m3, oxd ; OX - movd m4, shiftd ; shift +%if WIN64 || ARCH_X86_32 +cglobal hevc_put_hevc_uni_w%1_%2, 4, 5, 7, dst, dststride, src, srcstride, height, denom, wx, ox + mov r4d, denomm +%define SHIFT r4d +%else +cglobal hevc_put_hevc_uni_w%1_%2, 6, 6, 7, dst, dststride, src, srcstride, height, denom, wx, ox +%define SHIFT denomd +%endif + lea SHIFT, [SHIFT+14-%2] ; shift = 14 - bitd + denom + movd m2, wxm ; WX + movd m4, SHIFT ; shift punpcklwd m2, m2 - pshufd m3, m3, 0 - pshufd m2, m2, 0 - sub shiftd, 1 - movd m6, shiftd + dec SHIFT movdqu m5, [one_per_32] + movd m6, SHIFT + pshufd m2, m2, 0 + mov SHIFT, oxm pslld m5, m6 +%if %2 != 8 + shl SHIFT, %2-8 ; ox << (bitd - 8) +%endif + movd m3, SHIFT ; OX + pshufd m3, m3, 0 +%if WIN64 || ARCH_X86_32 + mov SHIFT, heightm +%endif .loop SIMPLE_LOAD %1, 10, srcq, m0 pmulhw m6, m0, m2 @@ -1141,38 +1154,42 @@ cglobal hevc_put_hevc_uni_w%1_%2, 8, 10, 11, dst, dststride, src, srcstride, hei jnz .loop ; height loop RET -cglobal hevc_put_hevc_bi_w%1_%2, 12, 14, 14, dst, dststride, src, srcstride, src2, src2stride, height, denom, wx0, wx1, ox0, ox1, shift, temp - shl ox0d, %2-8 ; ox << (bitd - 8) - shl ox1d, %2-8 ; ox << (bitd - 8) - lea shiftd, [denomd+14-%2] ; shift = 14 - bitd + denom - movd m2, wx0d ; WX0 - movd m3, wx1d ; WX1 +cglobal hevc_put_hevc_bi_w%1_%2, 6, 7, 10, dst, dststride, src, srcstride, src2, src2stride, height, denom, wx0, wx1, ox0, ox1 + mov r6d, denomm + movd m2, wx0m ; WX0 + lea r6d, [r6d+14-%2] ; shift = 14 - bitd + denom + movd m3, wx1m ; WX1 + movd m0, r6d ; shift punpcklwd m2, m2 + inc r6d punpcklwd m3, m3 + movd m5, r6d ; shift+1 pshufd m2, m2, 0 + mov r6d, ox0m pshufd m3, m3, 0 - add ox0d, ox1d - add ox0d, 1 - movd m4, ox0d ; offset + add r6d, ox1m +%if %2 != 8 + shl r6d, %2-8 ; ox << (bitd - 8) +%endif + inc r6d + movd m4, r6d ; offset pshufd m4, m4, 0 - movd m5, shiftd ; shift - pslld m4, m5 - add shiftd, 1 - movd m5, shiftd ; shift + mov r6d, heightm + pslld m4, m0 .loop SIMPLE_LOAD %1, 10, srcq, m0 - SIMPLE_LOAD %1, 10, src2q, m10 + SIMPLE_LOAD %1, 10, src2q, m8 pmulhw m6, m0, m3 pmullw m0, m3 - pmulhw m7, m10, m2 - pmullw m10, m2 + pmulhw m7, m8, m2 + pmullw m8, m2 punpckhwd m1, m0, m6 punpcklwd m0, m6 - punpckhwd m11, m10, m7 - punpcklwd m10, m7 - paddd m0, m10 - paddd m1, m11 + punpckhwd m9, m8, m7 + punpcklwd m8, m7 + paddd m0, m8 + paddd m1, m9 paddd m0, m4 paddd m1, m4 psrad m0, m5 @@ -1187,7 +1204,7 @@ cglobal hevc_put_hevc_bi_w%1_%2, 12, 14, 14, dst, dststride, src, srcstride, src lea dstq, [dstq+dststrideq] ; dst += dststride lea srcq, [srcq+2*srcstrideq] ; src += srcstride lea src2q, [src2q+2*src2strideq] ; src2 += srcstride - dec heightd ; cmp height + dec r6d ; cmp height jnz .loop ; height loop RET %endmacro |