[x265] [PATCH 04 of 12] asm: interp_4tap_vert_ps_6xN sse2
dtyx265 at gmail.com
dtyx265 at gmail.com
Tue May 19 02:24:31 CEST 2015
# HG changeset patch
# User David T Yuen <dtyx265 at gmail.com>
# Date 1431991674 25200
# Node ID 2f44ae6d6250677b87c31ca21a0098e44b19ee98
# Parent 660aa8e3a00e3c22543f2fcbc61ff0d81287f9cd
asm: interp_4tap_vert_ps_6xN sse2
Converted vert_pp_6xN macro to also create ps primitives. This replaces c code for ps with minimal impact on pp.
64-bit
./test/TestBench --testbench interp | grep vp | grep " 6x"
chroma_vpp[ 6x8] 2.93x 2162.50 6330.30
chroma_vps[ 6x8] 2.48x 1997.49 4950.24
chroma_vpp[ 6x16] 2.99x 4184.12 12520.14
chroma_vps[ 6x16] 2.45x 4041.30 9897.54
diff -r 660aa8e3a00e -r 2f44ae6d6250 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Mon May 18 16:16:18 2015 -0700
+++ b/source/common/x86/asm-primitives.cpp Mon May 18 16:27:54 2015 -0700
@@ -1520,6 +1520,8 @@
p.chroma[X265_CSP_I444].pu[LUMA_64x32].filter_vpp = x265_interp_4tap_vert_pp_64x32_sse2;
p.chroma[X265_CSP_I444].pu[LUMA_64x48].filter_vpp = x265_interp_4tap_vert_pp_64x48_sse2;
p.chroma[X265_CSP_I444].pu[LUMA_64x64].filter_vpp = x265_interp_4tap_vert_pp_64x64_sse2;
+ p.chroma[X265_CSP_I420].pu[CHROMA_420_6x8].filter_vps = x265_interp_4tap_vert_ps_6x8_sse2;
+ p.chroma[X265_CSP_I422].pu[CHROMA_422_6x16].filter_vps = x265_interp_4tap_vert_ps_6x16_sse2;
#endif
ALL_LUMA_PU(luma_hpp, interp_8tap_horiz_pp, sse2);
diff -r 660aa8e3a00e -r 2f44ae6d6250 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm Mon May 18 16:16:18 2015 -0700
+++ b/source/common/x86/ipfilter8.asm Mon May 18 16:27:54 2015 -0700
@@ -1282,12 +1282,11 @@
FILTER_V4_W4_H4_sse2 ps, 32
;-----------------------------------------------------------------------------
-;void interp_4tap_vert_pp_6x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
-;-----------------------------------------------------------------------------
-%macro FILTER_V4_W6_H4_sse2 1
+;void interp_4tap_vert_%1_6x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+%macro FILTER_V4_W6_H4_sse2 2
INIT_XMM sse2
-cglobal interp_4tap_vert_pp_6x%1, 4, 7, 10
-
+cglobal interp_4tap_vert_%1_6x%2, 4, 7, 10
mov r4d, r4m
sub r0, r1
shl r4d, 5
@@ -1302,11 +1301,17 @@
mova m5, [tab_ChromaCoeffV + r4 + 16]
%endif
+%ifidn %1,pp
mova m4, [pw_32]
+%elifidn %1,ps
+ mova m4, [pw_2000]
+ add r3d, r3d
+%endif
lea r5, [3 * r1]
+ lea r4, [3 * r3]
%assign x 1
-%rep %1/4
+%rep %2/4
movq m0, [r0]
movq m1, [r0 + r1]
movq m2, [r0 + 2 * r1]
@@ -1333,12 +1338,20 @@
paddw m0, m7
+%ifidn %1,pp
paddw m0, m4
psraw m0, 6
packuswb m0, m0
+
movd [r2], m0
pextrw r6d, m0, 2
mov [r2 + 4], r6w
+%elifidn %1,ps
+ psubw m0, m4
+ movh [r2], m0
+ pshufd m0, m0, 2
+ movd [r2 + 8], m0
+%endif
lea r0, [r0 + 4 * r1]
@@ -1362,12 +1375,21 @@
paddw m1, m7
+%ifidn %1,pp
paddw m1, m4
psraw m1, 6
packuswb m1, m1
+
movd [r2 + r3], m1
pextrw r6d, m1, 2
mov [r2 + r3 + 4], r6w
+%elifidn %1,ps
+ psubw m1, m4
+ movh [r2 + r3], m1
+ pshufd m1, m1, 2
+ movd [r2 + r3 + 8], m1
+%endif
+
movq m1, [r0 + r1]
punpcklbw m7, m0, m1
@@ -1387,13 +1409,19 @@
paddw m2, m7
+%ifidn %1,pp
paddw m2, m4
psraw m2, 6
packuswb m2, m2
- lea r2, [r2 + 2 * r3]
- movd [r2], m2
+ movd [r2 + 2 * r3], m2
pextrw r6d, m2, 2
- mov [r2 + 4], r6w
+ mov [r2 + 2 * r3 + 4], r6w
+%elifidn %1,ps
+ psubw m2, m4
+ movh [r2 + 2 * r3], m2
+ pshufd m2, m2, 2
+ movd [r2 + 2 * r3 + 8], m2
+%endif
movq m2, [r0 + 2 * r1]
punpcklbw m1, m2
@@ -1414,17 +1442,25 @@
paddw m3, m1
+%ifidn %1,pp
paddw m3, m4
psraw m3, 6
packuswb m3, m3
- movd [r2 + r3], m3
- pextrw r6d, m3, 2
- mov [r2 + r3 + 4], r6w
-
-%if x < %1/4
- lea r2, [r2 + 2 * r3]
-%endif
+ movd [r2 + r4], m3
+ pextrw r6d, m3, 2
+ mov [r2 + r4 + 4], r6w
+%elifidn %1,ps
+ psubw m3, m4
+ movh [r2 + r4], m3
+ pshufd m3, m3, 2
+ movd [r2 + r4 + 8], m3
+%endif
+
+%if x < %2/4
+ lea r2, [r2 + 4 * r3]
+%endif
+
%assign x x+1
%endrep
RET
@@ -1432,8 +1468,10 @@
%endmacro
%if ARCH_X86_64
- FILTER_V4_W6_H4_sse2 8
- FILTER_V4_W6_H4_sse2 16
+ FILTER_V4_W6_H4_sse2 pp, 8
+ FILTER_V4_W6_H4_sse2 pp, 16
+ FILTER_V4_W6_H4_sse2 ps, 8
+ FILTER_V4_W6_H4_sse2 ps, 16
%endif
;-----------------------------------------------------------------------------
diff -r 660aa8e3a00e -r 2f44ae6d6250 source/common/x86/ipfilter8.h
--- a/source/common/x86/ipfilter8.h Mon May 18 16:16:18 2015 -0700
+++ b/source/common/x86/ipfilter8.h Mon May 18 16:27:54 2015 -0700
@@ -954,6 +954,8 @@
void x265_interp_4tap_vert_pp_64x32_sse2(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx);
void x265_interp_4tap_vert_pp_64x48_sse2(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx);
void x265_interp_4tap_vert_pp_64x64_sse2(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx);
+void x265_interp_4tap_vert_ps_6x8_sse2(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);
+void x265_interp_4tap_vert_ps_6x16_sse2(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);
#endif
#undef LUMA_FILTERS
#undef LUMA_SP_FILTERS
More information about the x265-devel
mailing list