[FFmpeg-cvslog] x86: hevc_mc: better register allocation

[Christophe Gisquet]

ffmpeg | branch: master | Christophe Gisquet <christophe.gisquet at gmail.com> | Sat May 17 12:22:39 2014 +0200| [0810608e23456eeff75b691c7c444762352ec196] | committer: Michael Niedermayer

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 at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=0810608e23456eeff75b691c7c444762352ec196
---

 libavcodec/x86/hevc_mc.asm |   79 +++++++++++++++++++++++++++-----------------
 1 file changed, 48 insertions(+), 31 deletions(-)

diff --git a/libavcodec/x86/hevc_mc.asm b/libavcodec/x86/hevc_mc.asm
index 8a605d7..0c1a431 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