[x265] [PATCH] asm: assembly code for pixel_satd_32x64 and pixel_satd_48x64
yuvaraj at multicorewareinc.com
yuvaraj at multicorewareinc.com
Thu Nov 14 14:41:56 CET 2013
# HG changeset patch
# User Yuvaraj Venkatesh <yuvaraj at multicorewareinc.com>
# Date 1384436305 -19800
# Thu Nov 14 19:08:25 2013 +0530
# Node ID 95ed43fcd3f3f94c8bfe2f0ccb4e4b0fe38e2731
# Parent cb15dab6333f3ce23083274718754ca588596547
asm: assembly code for pixel_satd_32x64 and pixel_satd_48x64
diff -r cb15dab6333f -r 95ed43fcd3f3 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Thu Nov 14 17:34:34 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp Thu Nov 14 19:08:25 2013 +0530
@@ -63,9 +63,9 @@
p.satd[LUMA_24x32] = x265_pixel_satd_24x32_ ## cpu; \
p.satd[LUMA_64x64] = x265_pixel_satd_64x64_ ## cpu; \
p.satd[LUMA_64x32] = x265_pixel_satd_64x32_ ## cpu; \
- p.satd[LUMA_32x64] = cmp<32, 64, 16, 16, x265_pixel_satd_16x16_ ## cpu>; \
+ p.satd[LUMA_32x64] = x265_pixel_satd_32x64_ ## cpu; \
p.satd[LUMA_64x48] = x265_pixel_satd_64x48_ ## cpu; \
- p.satd[LUMA_48x64] = cmp<48, 64, 16, 16, x265_pixel_satd_16x16_ ## cpu>; \
+ p.satd[LUMA_48x64] = x265_pixel_satd_48x64_ ## cpu; \
p.satd[LUMA_64x16] = x265_pixel_satd_64x16_ ## cpu
#define ASSGN_SSE(cpu) \
diff -r cb15dab6333f -r 95ed43fcd3f3 source/common/x86/pixel-a.asm
--- a/source/common/x86/pixel-a.asm Thu Nov 14 17:34:34 2013 +0530
+++ b/source/common/x86/pixel-a.asm Thu Nov 14 19:08:25 2013 +0530
@@ -1880,83 +1880,47 @@
call pixel_satd_16x4_internal
SATD_END_SSE2 m10
-cglobal pixel_satd_64x16, 4,8,8 ;if WIN64 && notcpuflag(avx)
+cglobal pixel_satd_32x64, 4,8,8 ;if WIN64 && notcpuflag(avx)
SATD_START_SSE2 m10, m7
mov r6, r0
mov r7, r2
%if vertical
mova m7, [pw_00ff]
%endif
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
lea r0, [r6 + 16]
lea r2, [r7 + 16]
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- lea r0, [r6 + 32]
- lea r2, [r7 + 32]
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- lea r0, [r6 + 48]
- lea r2, [r7 + 48]
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- call pixel_satd_16x4_internal
- SATD_END_SSE2 m10
-
-cglobal pixel_satd_64x32, 4,8,8 ;if WIN64 && notcpuflag(avx)
- SATD_START_SSE2 m10, m7
- mov r6, r0
- mov r7, r2
-%if vertical
- mova m7, [pw_00ff]
-%endif
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- lea r0, [r6 + 16]
- lea r2, [r7 + 16]
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- lea r0, [r6 + 32]
- lea r2, [r7 + 32]
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- lea r0, [r6 + 48]
- lea r2, [r7 + 48]
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
-
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
pxor m9, m9
movhlps m9, m10
paddd m10, m9
@@ -1965,7 +1929,7 @@
movd eax, m10
RET
-cglobal pixel_satd_64x48, 4,8,8 ;if WIN64 && notcpuflag(avx)
+cglobal pixel_satd_48x64, 4,8,8 ;if WIN64 && notcpuflag(avx)
SATD_START_SSE2 m10, m7
mov r6, r0
mov r7, r2
@@ -1984,6 +1948,10 @@
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
lea r0, [r6 + 16]
lea r2, [r7 + 16]
call pixel_satd_16x4_internal2
@@ -1998,6 +1966,10 @@
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
lea r0, [r6 + 32]
lea r2, [r7 + 32]
call pixel_satd_16x4_internal2
@@ -2012,21 +1984,10 @@
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
- lea r0, [r6 + 48]
- lea r2, [r7 + 48]
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
-
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
pxor m9, m9
movhlps m9, m10
paddd m10, m9
@@ -2035,75 +1996,74 @@
movd eax, m10
RET
-cglobal pixel_satd_64x64, 4,8,8 ;if WIN64 && notcpuflag(avx)
+cglobal pixel_satd_64x16, 4,8,8 ;if WIN64 && notcpuflag(avx)
SATD_START_SSE2 m10, m7
mov r6, r0
mov r7, r2
%if vertical
mova m7, [pw_00ff]
%endif
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
lea r0, [r6 + 16]
lea r2, [r7 + 16]
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
lea r0, [r6 + 32]
lea r2, [r7 + 32]
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
lea r0, [r6 + 48]
lea r2, [r7 + 48]
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
- call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ call pixel_satd_16x4_internal
+ SATD_END_SSE2 m10
+
+cglobal pixel_satd_64x32, 4,8,8 ;if WIN64 && notcpuflag(avx)
+ SATD_START_SSE2 m10, m7
+ mov r6, r0
+ mov r7, r2
+%if vertical
+ mova m7, [pw_00ff]
+%endif
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 16]
+ lea r2, [r7 + 16]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 32]
+ lea r2, [r7 + 32]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 48]
+ lea r2, [r7 + 48]
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
call pixel_satd_16x4_internal2
@@ -2121,6 +2081,162 @@
movd eax, m10
RET
+cglobal pixel_satd_64x48, 4,8,8 ;if WIN64 && notcpuflag(avx)
+ SATD_START_SSE2 m10, m7
+ mov r6, r0
+ mov r7, r2
+%if vertical
+ mova m7, [pw_00ff]
+%endif
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 16]
+ lea r2, [r7 + 16]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 32]
+ lea r2, [r7 + 32]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 48]
+ lea r2, [r7 + 48]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+
+ pxor m9, m9
+ movhlps m9, m10
+ paddd m10, m9
+ pshufd m9, m10, 1
+ paddd m10, m9
+ movd eax, m10
+ RET
+
+cglobal pixel_satd_64x64, 4,8,8 ;if WIN64 && notcpuflag(avx)
+ SATD_START_SSE2 m10, m7
+ mov r6, r0
+ mov r7, r2
+%if vertical
+ mova m7, [pw_00ff]
+%endif
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 16]
+ lea r2, [r7 + 16]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 32]
+ lea r2, [r7 + 32]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ lea r0, [r6 + 48]
+ lea r2, [r7 + 48]
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+ call pixel_satd_16x4_internal2
+
+ pxor m9, m9
+ movhlps m9, m10
+ paddd m10, m9
+ pshufd m9, m10, 1
+ paddd m10, m9
+ movd eax, m10
+ RET
+
%else
cglobal pixel_satd_32x8, 4,6,8 ;if !WIN64
@@ -2308,6 +2424,263 @@
%endif
%if WIN64
+cglobal pixel_satd_32x64, 4,8,8 ;if WIN64 && cpuflag(avx)
+ SATD_START_SSE2 m6, m7
+ mov r6, r0
+ mov r7, r2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 8]
+ lea r2, [r7 + 8]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 16]
+ lea r2, [r7 + 16]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 24]
+ lea r2, [r7 + 24]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ pxor m7, m7
+ movhlps m7, m6
+ paddd m6, m7
+ pshufd m7, m6, 1
+ paddd m6, m7
+ movd eax, m6
+ RET
+%else
+cglobal pixel_satd_32x64, 4,6,8 ;if !WIN64
+ SATD_START_SSE2 m6, m7
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 8
+ add r2, 8
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 16
+ add r2, 16
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 24
+ add r2, 24
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ pxor m7, m7
+ movhlps m7, m6
+ paddd m6, m7
+ pshufd m7, m6, 1
+ paddd m6, m7
+ movd eax, m6
+ RET
+%endif
+
+%if WIN64
+cglobal pixel_satd_48x64, 4,8,8 ;if WIN64 && cpuflag(avx)
+ SATD_START_SSE2 m6, m7
+ mov r6, r0
+ mov r7, r2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 8]
+ lea r2, [r7 + 8]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 16]
+ lea r2, [r7 + 16]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 24]
+ lea r2, [r7 + 24]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 32]
+ lea r2, [r7 + 32]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ lea r0, [r6 + 40]
+ lea r2, [r7 + 40]
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ pxor m7, m7
+ movhlps m7, m6
+ paddd m6, m7
+ pshufd m7, m6, 1
+ paddd m6, m7
+ movd eax, m6
+ RET
+%else
+cglobal pixel_satd_48x64, 4,6,8 ;if !WIN64
+ SATD_START_SSE2 m6, m7
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 8
+ add r2, 8
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 16
+ add r2, 16
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 24
+ add r2, 24
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 32
+ add r2, 32
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ mov r0, r0mp
+ mov r2, r2mp
+ add r0, 40
+ add r2, 40
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ call pixel_satd_8x8_internal2
+ pxor m7, m7
+ movhlps m7, m6
+ paddd m6, m7
+ pshufd m7, m6, 1
+ paddd m6, m7
+ movd eax, m6
+ RET
+%endif
+
+
+%if WIN64
cglobal pixel_satd_64x16, 4,8,8 ;if WIN64 && cpuflag(avx)
SATD_START_SSE2 m6, m7
mov r6, r0
More information about the x265-devel
mailing list