[x265] [PATCH] asm: chroma_hpp[8x8] for colorspace i420 in avx2 improve 530c->373c
aasaipriya at multicorewareinc.com
aasaipriya at multicorewareinc.com
Tue Dec 9 05:42:35 CET 2014
# HG changeset patch
# User Aasaipriya Chandran <aasaipriya at multicorewareinc.com>
# Date 1418099615 -19800
# Tue Dec 09 10:03:35 2014 +0530
# Node ID c7eada9b23500a37c7bc8dccb39bcb59f04ea73d
# Parent b376435b31c1e25078ad47034f826f06703cce88
asm: chroma_hpp[8x8] for colorspace i420 in avx2 improve 530c->373c
diff -r b376435b31c1 -r c7eada9b2350 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Mon Dec 08 21:35:14 2014 -0600
+++ b/source/common/x86/asm-primitives.cpp Tue Dec 09 10:03:35 2014 +0530
@@ -1873,6 +1873,8 @@
p.luma_hpp[LUMA_48x64] = x265_interp_8tap_horiz_pp_48x64_avx2;
+ p.chroma[X265_CSP_I420].filter_hpp[CHROMA_8x8] = x265_interp_4tap_horiz_pp_8x8_avx2;
+
p.luma_vpp[LUMA_4x4] = x265_interp_8tap_vert_pp_4x4_avx2;
p.luma_vpp[LUMA_8x4] = x265_interp_8tap_vert_pp_8x4_avx2;
diff -r b376435b31c1 -r c7eada9b2350 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm Mon Dec 08 21:35:14 2014 -0600
+++ b/source/common/x86/ipfilter8.asm Tue Dec 09 10:03:35 2014 +0530
@@ -475,6 +475,9 @@
RET
+ALIGN 32
+const interp_4tap_8x8_horiz_shuf, dd 0, 4, 1, 5, 2, 6, 3, 7
+
%macro FILTER_H4_w6 3
movu %1, [srcq - 1]
@@ -1408,6 +1411,72 @@
IPFILTER_LUMA 12, 16, pp
IPFILTER_LUMA 4, 16, pp
+INIT_YMM avx2
+cglobal interp_4tap_horiz_pp_8x8, 4,6,6
+ mov r4d, r4m
+
+%ifdef PIC
+ lea r5, [tab_ChromaCoeff]
+ vpbroadcastd m0, [r5 + r4 * 4]
+%else
+ vpbroadcastd m0, [tab_ChromaCoeff + r4 * 4]
+%endif
+
+ movu m1, [tab_Tm]
+ vpbroadcastd m2, [pw_1]
+
+ ; register map
+ ; m0 - interpolate coeff
+ ; m1 - shuffle order table
+ ; m2 - constant word 1
+
+ sub r0, 1
+ mov r4d, 2
+
+.loop:
+ ; Row 0
+ vbroadcasti128 m3, [r0] ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+ pshufb m3, m1
+ pmaddubsw m3, m0
+ pmaddwd m3, m2
+
+ ; Row 1
+ vbroadcasti128 m4, [r0 + r1] ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+ pshufb m4, m1
+ pmaddubsw m4, m0
+ pmaddwd m4, m2
+ packssdw m3, m4
+ pmulhrsw m3, [pw_512]
+ lea r0, [r0 + r1 * 2]
+
+ ; Row 2
+ vbroadcasti128 m4, [r0 ] ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+ pshufb m4, m1
+ pmaddubsw m4, m0
+ pmaddwd m4, m2
+
+ ; Row 3
+ vbroadcasti128 m5, [r0 + r1] ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
+ pshufb m5, m1
+ pmaddubsw m5, m0
+ pmaddwd m5, m2
+ packssdw m4, m5
+ pmulhrsw m4, [pw_512]
+
+ packuswb m3, m4
+ mova m5, [interp_4tap_8x8_horiz_shuf]
+ vpermd m3, m5, m3
+ vextracti128 xm4, m3, 1
+ movq [r2], xm3
+ movhps [r2 + r3], xm3
+ lea r2, [r2 + r3 * 2]
+ movq [r2], xm4
+ movhps [r2 + r3], xm4
+ lea r2, [r2 + r3 * 2]
+ lea r0, [r0 + r1*2]
+ dec r4d
+ jnz .loop
+ RET
IPFILTER_LUMA_AVX2 16, 4
IPFILTER_LUMA_AVX2 16, 8
More information about the x265-devel
mailing list