ref: 04b70ea56d3258bedef3002ea877cc90277e5ab2
parent: a146437d8810faac27b0033d08793a6311dc8f01
author: Ronald S. Bultje <rsbultje@gmail.com>
date: Fri Oct 5 13:51:18 EDT 2018
Add vertical loopfilter AVX2 SIMD
--- a/src/lf_mask.c
+++ b/src/lf_mask.c
@@ -347,8 +347,8 @@
int level;
// set E/I/H values from loopfilter level
+ const int sharp = filter_sharpness;
for (level = 0; level < 64; level++) {- const int sharp = filter_sharpness;
int limit = level;
if (sharp > 0) {@@ -360,6 +360,8 @@
lim_lut->i[level] = limit;
lim_lut->e[level] = 2 * (level + 2) + limit;
}
+ lim_lut->sharp[0] = (sharp + 3) >> 2;
+ lim_lut->sharp[1] = sharp ? 9 - sharp : 0xff;
}
static void dav1d_calc_lf_value(uint8_t (*const lflvl_values)[2],
--- a/src/lf_mask.h
+++ b/src/lf_mask.h
@@ -36,6 +36,7 @@
typedef struct Av1FilterLUT {uint8_t e[64];
uint8_t i[64];
+ uint64_t sharp[2];
} Av1FilterLUT;
typedef struct Av1RestorationUnit {--- a/src/loopfilter.c
+++ b/src/loopfilter.c
@@ -231,4 +231,8 @@
c->loop_filter_sb128y = loop_filter_v_sb128y_c;
c->loop_filter_sb128uv = loop_filter_v_sb128uv_c;
+
+#if HAVE_ASM && ARCH_X86
+ bitfn(dav1d_loop_filter_dsp_init_x86)(c);
+#endif
}
--- a/src/loopfilter.h
+++ b/src/loopfilter.h
@@ -62,4 +62,7 @@
void dav1d_loop_filter_dsp_init_8bpc(Dav1dLoopFilterDSPContext *c);
void dav1d_loop_filter_dsp_init_10bpc(Dav1dLoopFilterDSPContext *c);
+void dav1d_loop_filter_dsp_init_x86_8bpc(Dav1dLoopFilterDSPContext *c);
+void dav1d_loop_filter_dsp_init_x86_10bpc(Dav1dLoopFilterDSPContext *c);
+
#endif /* __DAV1D_SRC_LOOPFILTER_H__ */
--- a/src/meson.build
+++ b/src/meson.build
@@ -99,6 +99,7 @@
libdav1d_tmpl_sources += files(
'x86/itx_init.c',
+ 'x86/loopfilter_init.c',
'x86/mc_init.c',
)
@@ -106,6 +107,7 @@
libdav1d_sources_asm = files(
'x86/cpuid.asm',
'x86/itx.asm',
+ 'x86/loopfilter.asm',
'x86/mc.asm',
)
--- /dev/null
+++ b/src/x86/loopfilter.asm
@@ -1,0 +1,829 @@
+; Copyright © 2018, VideoLAN and dav1d authors
+; Copyright © 2018, Two Orioles, LLC
+; All rights reserved.
+;
+; Redistribution and use in source and binary forms, with or without
+; modification, are permitted provided that the following conditions are met:
+;
+; 1. Redistributions of source code must retain the above copyright notice, this
+; list of conditions and the following disclaimer.
+;
+; 2. Redistributions in binary form must reproduce the above copyright notice,
+; this list of conditions and the following disclaimer in the documentation
+; and/or other materials provided with the distribution.
+;
+; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+; ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+; WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+; DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+; ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+; (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+; ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+%include "config.asm"
+%include "ext/x86/x86inc.asm"
+
+%if ARCH_X86_64
+
+SECTION_RODATA 32
+
+pb_4x1_4x5_4x9_4x13: times 2 db 0, 0, 0, 0, 4, 4, 4, 4, 8, 8, 8, 8, 12, 12, 12, 12
+pb_7_1: times 16 db 7, 1
+pb_3_1: times 16 db 3, 1
+pb_2_1: times 16 db 2, 1
+pb_m1_0: times 16 db -1, 0
+pb_m1_1: times 16 db -1, 1
+pb_m1_2: times 16 db -1, 2
+pb_1: times 32 db 1
+pb_2: times 32 db 2
+pb_3: times 32 db 3
+pb_4: times 32 db 4
+pb_16: times 32 db 16
+pb_63: times 32 db 63
+pb_64: times 32 db 64
+pb_128: times 32 db 0x80
+pb_129: times 32 db 0x81
+pb_240: times 32 db 0xf0
+pb_248: times 32 db 0xf8
+pb_254: times 32 db 0xfe
+
+pw_2048: times 16 dw 2048
+pw_4096: times 16 dw 4096
+
+pb_mask: dd 1, 2, 4, 8, 16, 32, 64, 128
+
+SECTION .text
+
+%macro ABSSUB 4 ; dst, a, b, tmp
+ psubusb %1, %2, %3
+ psubusb %4, %3, %2
+ por %1, %4
+%endmacro
+
+%macro FILTER 1 ; width
+ movu m1, [lq]
+ movu m0, [lq+l_strideq]
+ pxor m2, m2
+ pcmpeqb m3, m2, m0
+ pand m1, m3
+ por m0, m1 ; l[x][] ? l[x][] : l[x-stride][]
+ pshufb m0, [pb_4x1_4x5_4x9_4x13] ; l[x][1]
+ pcmpeqb m10, m2, m0 ; !L
+ pand m1, m0, [pb_240]
+ psrlq m1, 4 ; H
+ psrlq m2, m0, [lutq+128]
+ pand m2, [pb_63]
+ vpbroadcastb m4, [lutq+136]
+ pminub m2, m4
+ pmaxub m2, [pb_1] ; I
+ paddb m0, [pb_2]
+ paddb m0, m0
+ paddb m0, m2 ; E
+ pxor m1, [pb_128]
+ pxor m2, [pb_128]
+ pxor m0, [pb_128]
+
+%if %1 == 4
+ lea tmpq, [dstq+mstrideq*2]
+ mova m3, [tmpq+strideq*0] ; p1
+ mova m4, [tmpq+strideq*1] ; p0
+ mova m5, [tmpq+strideq*2] ; q0
+ mova m6, [tmpq+stride3q] ; q1
+%else
+ lea tmpq, [dstq+mstrideq*4]
+ mova m3, [tmpq+strideq*2]
+ mova m4, [tmpq+stride3q]
+ mova m5, [dstq+strideq*0]
+ mova m6, [dstq+strideq*1]
+%endif
+
+ ABSSUB m8, m3, m4, m9 ; abs(p1-p0)
+ pmaxub m8, m10
+ ABSSUB m9, m5, m6, m10 ; abs(q1-q0)
+ pmaxub m8, m9
+%if %1 == 4
+ pxor m8, [pb_128]
+ pcmpgtb m7, m8, m1 ; hev
+%else
+ pxor m7, m8, [pb_128]
+ pcmpgtb m7, m1 ; hev
+
+%if %1 != 6
+ mova m12, [tmpq+strideq*0]
+%endif
+ mova m13, [tmpq+strideq*1]
+ mova m14, [dstq+strideq*2]
+%if %1 != 6
+ mova m15, [dstq+stride3q]
+%endif
+
+%if %1 == 6
+ ABSSUB m9, m13, m4, m10 ; abs(p2-p0)
+ pmaxub m9, m8
+%else
+ ABSSUB m9, m12, m4, m10 ; abs(p3-p0)
+ pmaxub m9, m8
+ ABSSUB m10, m13, m4, m11 ; abs(p2-p0)
+ pmaxub m9, m10
+%endif
+ ABSSUB m10, m5, m14, m11 ; abs(q2-q0)
+ pmaxub m9, m10
+%if %1 != 6
+ ABSSUB m10, m5, m15, m11 ; abs(q3-q0)
+ pmaxub m9, m10
+%endif
+ pxor m9, [pb_128]
+ pcmpgtb m9, [pb_129] ; !flat8in
+
+%if %1 == 6
+ ABSSUB m10, m13, m3, m1 ; abs(p2-p1)
+%else
+ ABSSUB m10, m12, m13, m11 ; abs(p3-p2)
+ ABSSUB m11, m13, m3, m1 ; abs(p2-p1)
+ pmaxub m10, m11
+ ABSSUB m11, m14, m15, m1 ; abs(q3-q2)
+ pmaxub m10, m11
+%endif
+ ABSSUB m11, m14, m6, m1 ; abs(q2-q1)
+ pmaxub m10, m11
+%if %1 == 16
+ vpbroadcastd m11, [maskq+8]
+ vpbroadcastd m1, [maskq+4]
+ por m11, m1
+ pand m11, [pb_mask]
+ pcmpeqd m11, [pb_mask]
+ pand m10, m11
+%else
+ vpbroadcastd m11, [maskq+4]
+ pand m11, [pb_mask]
+ pcmpeqd m11, [pb_mask]
+ pand m10, m11 ; only apply fm-wide to wd>4 blocks
+%endif
+ pmaxub m8, m10
+
+ pxor m8, [pb_128]
+%endif
+ pcmpgtb m8, m2
+
+ ABSSUB m10, m3, m6, m11 ; abs(p1-q1)
+ ABSSUB m11, m4, m5, m2 ; abs(p0-q0)
+ paddusb m11, m11
+ pand m10, [pb_254]
+ psrlq m10, 1
+ paddusb m10, m11 ; abs(p0-q0)*2+(abs(p1-q1)>>1)
+ pxor m10, [pb_128]
+ pcmpgtb m10, m0 ; abs(p0-q0)*2+(abs(p1-q1)>>1) > E
+ por m8, m10
+
+%if %1 == 16
+ lea tmpq, [dstq+mstrideq*8]
+ mova m0, [tmpq+strideq*1]
+ ABSSUB m1, m0, m4, m2
+ mova m0, [tmpq+strideq*2]
+ ABSSUB m2, m0, m4, m10
+ pmaxub m1, m2
+ mova m0, [tmpq+stride3q]
+ ABSSUB m2, m0, m4, m10
+ pmaxub m1, m2
+ lea tmpq, [dstq+strideq*4]
+ mova m0, [tmpq+strideq*0]
+ ABSSUB m2, m0, m5, m10
+ pmaxub m1, m2
+ mova m0, [tmpq+strideq*1]
+ ABSSUB m2, m0, m5, m10
+ pmaxub m1, m2
+ mova m0, [tmpq+strideq*2]
+ ABSSUB m2, m0, m5, m10
+ pmaxub m1, m2
+ pxor m1, [pb_128]
+ pcmpgtb m1, [pb_129] ; !flat8out
+ por m1, m9 ; !flat8in | !flat8out
+ vpbroadcastd m2, [maskq+8]
+ pand m10, m2, [pb_mask]
+ pcmpeqd m10, [pb_mask]
+ pandn m1, m10 ; flat16
+ pandn m1, m8, m1 ; flat16 & fm
+
+ vpbroadcastd m10, [maskq+4]
+ por m10, m2
+ pand m2, m10, [pb_mask]
+ pcmpeqd m2, [pb_mask]
+ pandn m9, m2 ; flat8in
+ pandn m9, m8, m9
+ vpbroadcastd m2, [maskq+0]
+ por m2, m10
+ pand m2, [pb_mask]
+ pcmpeqd m2, [pb_mask]
+ pandn m8, m2
+ pandn m8, m9, m8 ; fm & !flat8 & !flat16
+ pandn m9, m1, m9 ; flat8 & !flat16
+%elif %1 != 4
+ vpbroadcastd m0, [maskq+4]
+ pand m2, m0, [pb_mask]
+ pcmpeqd m2, [pb_mask]
+ pandn m9, m2
+ pandn m9, m8, m9 ; flat8 & fm
+ vpbroadcastd m2, [maskq+0]
+ por m0, m2
+ pand m0, [pb_mask]
+ pcmpeqd m0, [pb_mask]
+ pandn m8, m0
+ pandn m8, m9, m8 ; fm & !flat8
+%else
+ vpbroadcastd m0, [maskq+0]
+ pand m0, [pb_mask]
+ pcmpeqd m0, [pb_mask]
+ pandn m8, m0 ; fm
+%endif
+
+ ; short filter
+
+ pxor m3, [pb_128]
+ pxor m6, [pb_128]
+ psubsb m10, m3, m6 ; iclip_diff(p1-q1)
+ pand m10, m7 ; f=iclip_diff(p1-q1)&hev
+ pxor m4, [pb_128]
+ pxor m5, [pb_128]
+ psubsb m11, m5, m4
+ paddsb m10, m11
+ paddsb m10, m11
+ paddsb m10, m11 ; f=iclip_diff(3*(q0-p0)+f)
+ pand m8, m10 ; f&=fm
+ paddsb m10, m8, [pb_3]
+ paddsb m8, [pb_4]
+ pand m10, [pb_248]
+ pand m8, [pb_248]
+ psrlq m10, 3
+ psrlq m8, 3
+ pxor m10, [pb_16]
+ pxor m8, [pb_16]
+ psubb m10, [pb_16] ; f2
+ psubb m8, [pb_16] ; f1
+ paddsb m4, m10
+ psubsb m5, m8
+ pxor m4, [pb_128]
+ pxor m5, [pb_128]
+
+ pxor m8, [pb_128]
+ pxor m10, m10
+ pavgb m8, m10 ; f=(f1+1)>>1
+ psubb m8, [pb_64]
+ pandn m8, m7, m8 ; f&=!hev
+ paddsb m3, m8
+ psubsb m6, m8
+ pxor m3, [pb_128]
+ pxor m6, [pb_128]
+
+%if %1 == 16
+ ; flat16 filter
+ lea tmpq, [dstq+mstrideq*8]
+ mova m0, [tmpq+strideq*1] ; p6
+ mova m2, [tmpq+strideq*2] ; p5
+ mova m7, [tmpq+stride3q] ; p4
+
+ mova [rsp+0*32], m9
+ mova [rsp+1*32], m14
+ mova [rsp+2*32], m15
+
+ ; p6*7+p5*2+p4*2+p3+p2+p1+p0+q0 [p5/p4/p2/p1/p0/q0][p6/p3] A
+ ; write -6
+ punpcklbw m14, m0, m12
+ punpckhbw m15, m0, m12
+ pmaddubsw m10, m14, [pb_7_1]
+ pmaddubsw m11, m15, [pb_7_1] ; p6*7+p3
+ punpcklbw m8, m2, m7
+ punpckhbw m9, m2, m7
+ pmaddubsw m8, [pb_2]
+ pmaddubsw m9, [pb_2]
+ paddw m10, m8
+ paddw m11, m9 ; p6*7+p5*2+p4*2+p3
+ punpcklbw m8, m13, m3
+ punpckhbw m9, m13, m3
+ pmaddubsw m8, [pb_1]
+ pmaddubsw m9, [pb_1]
+ paddw m10, m8
+ paddw m11, m9 ; p6*7+p5*2+p4*2+p3+p2+p1
+ punpcklbw m8, m4, m5
+ punpckhbw m9, m4, m5
+ pmaddubsw m8, [pb_1]
+ pmaddubsw m9, [pb_1]
+ paddw m10, m8
+ paddw m11, m9 ; p6*7+p5*2+p4*2+p3+p2+p1+p0+q0
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, m2
+ por m8, m9
+ mova [tmpq+strideq*2], m8 ; p5
+
+ ; sub p6*2, add p3/q1 [reuse p6/p3 from A][-p6,+q1|save] B
+ ; write -5
+ pmaddubsw m14, [pb_m1_1]
+ pmaddubsw m15, [pb_m1_1]
+ paddw m10, m14
+ paddw m11, m15 ; p6*6+p5*2+p4*2+p3*2+p2+p1+p0+q0
+ punpcklbw m8, m0, m6
+ punpckhbw m9, m0, m6
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ mova [rsp+3*32], m8
+ mova [rsp+4*32], m9
+ paddw m10, m8
+ paddw m11, m9 ; p6*5+p5*2+p4*2+p3*2+p2+p1+p0+q0+q1
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, m7
+ por m8, m9
+ mova [tmpq+stride3q], m8 ; p4
+
+ ; sub p6/p5, add p2/q2 [-p6,+p2][-p5,+q2|save] C
+ ; write -4
+ mova m14, [rsp+1*32]
+ punpcklbw m8, m0, m13
+ punpckhbw m9, m0, m13
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m9 ; p6*4+p5*2+p4*2+p3*2+p2*2+p1+p0+q0+q1
+ punpcklbw m8, m2, m14
+ punpckhbw m2, m14
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m2, [pb_m1_1]
+ mova [rsp+1*32], m8
+ paddw m10, m8
+ paddw m11, m2 ; p6*4+p5+p4*2+p3*2+p2*2+p1+p0+q0+q1+q2
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, m12
+ por m8, m9
+ mova [tmpq+strideq*4], m8 ; p3
+
+ ; sub p6/p4, add p1/q3 [-p6,+p1][-p4,+q3|save] D
+ ; write -3
+ mova m15, [rsp+2*32]
+ punpcklbw m8, m0, m3
+ punpckhbw m9, m0, m3
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m9 ; p6*3+p5+p4*2+p3*2+p2*2+p1*2+p0+q0+q1+q2
+ punpcklbw m8, m7, m15
+ punpckhbw m7, m15
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m7, [pb_m1_1]
+ mova [rsp+2*32], m8
+ paddw m10, m8
+ paddw m11, m7 ; p6*3+p5+p4+p3*2+p2*2+p1*2+p0+q0+q1+q2+q3
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, m13
+ por m8, m9
+ mova [rsp+6*32], m8 ; don't clobber p2/m13 since we need it in F
+
+ ; sub p6/p3, add p0/q4 [-p6,+p0][-p3,+q4|save] E
+ ; write -2
+ lea tmpq, [dstq+strideq*4]
+ punpcklbw m8, m0, m4
+ punpckhbw m9, m0, m4
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m9 ; p6*2+p5+p4+p3*2+p2*2+p1*2+p0*2+q0+q1+q2+q3
+ mova m9, [tmpq+strideq*0] ; q4
+ punpcklbw m8, m12, m9
+ punpckhbw m9, m12, m9
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ mova [rsp+7*32], m8
+ mova [rsp+5*32], m9
+ paddw m10, m8
+ paddw m11, m9 ; p6*2+p5+p4+p3+p2*2+p1*2+p0*2+q0+q1+q2+q3+q4
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, m3
+ por m8, m9
+ mova [rsp+8*32], m8 ; don't clobber p1/m3 since we need it in G
+
+ ; sub p6/p2, add q0/q5 [-p6,+q0][-p2,+q5|save] F
+ ; write -1
+ mova m9, [tmpq+strideq*1] ; q5
+ punpcklbw m8, m0, m5
+ punpckhbw m0, m5
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m0, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m0 ; p6+p5+p4+p3+p2*2+p1*2+p0*2+q0*2+q1+q2+q3+q4
+ punpcklbw m0, m13, m9
+ punpckhbw m9, m13, m9
+ mova m13, [rsp+6*32]
+ pmaddubsw m0, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ mova [rsp+ 9*32], m0
+ mova [rsp+10*32], m9
+ paddw m10, m0
+ paddw m11, m9 ; p6+p5+p4+p3+p2+p1*2+p0*2+q0*2+q1+q2+q3+q4+q5
+ pmulhrsw m0, m10, [pw_2048]
+ pmulhrsw m8, m11, [pw_2048]
+ packuswb m0, m8
+ pand m0, m1
+ pandn m8, m1, m4
+ por m0, m8
+ mova [rsp+6*32], m0 ; don't clobber p0/m4 since we need it in H
+
+ ; sub p6/p1, add q1/q6 [reuse -p6,+q1 from B][-p1,+q6|save] G
+ ; write +0
+ mova m0, [tmpq+strideq*2] ; q6
+ paddw m10, [rsp+3*32]
+ paddw m11, [rsp+4*32] ; p5+p4+p3+p2+p1*2+p0*2+q0*2+q1*2+q2+q3+q4+q5
+ punpcklbw m8, m3, m0
+ punpckhbw m9, m3, m0
+ mova m3, [rsp+8*32]
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ mova [rsp+3*32], m8
+ mova [rsp+4*32], m9
+ paddw m10, m8
+ paddw m11, m9 ; p5+p4+p3+p2+p1+p0*2+q0*2+q1*2+q2+q3+q4+q5+q6
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, m5
+ por m8, m9
+ mova [rsp+8*32], m8 ; don't clobber q0/m5 since we need it in I
+
+ ; sub p5/p0, add q2/q6 [reuse -p5,+q2 from C][-p0,+q6] H
+ ; write +1
+ paddw m10, [rsp+1*32]
+ paddw m11, m2 ; p4+p3+p2+p1+p0*2+q0*2+q1*2+q2*2+q3+q4+q5+q6
+ punpcklbw m8, m4, m0
+ punpckhbw m2, m4, m0
+ mova m4, [rsp+6*32]
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m2, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m2 ; p4+p3+p2+p1+p0+q0*2+q1*2+q2*2+q3+q4+q5+q6*2
+ pmulhrsw m2, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m2, m9
+ pand m2, m1
+ pandn m9, m1, m6
+ por m2, m9 ; don't clobber q1/m6 since we need it in K
+
+ ; sub p4/q0, add q3/q6 [reuse -p4,+q3 from D][-q0,+q6] I
+ ; write +2
+ paddw m10, [rsp+2*32]
+ paddw m11, m7 ; p3+p2+p1+p0+q0*2+q1*2+q2*2+q3*2+q4+q5+q6*2
+ punpcklbw m8, m5, m0
+ punpckhbw m9, m5, m0
+ mova m5, [rsp+8*32]
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m9 ; p3+p2+p1+p0+q0+q1*2+q2*2+q3*2+q4+q5+q6*3
+ pmulhrsw m7, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m7, m9
+ pand m7, m1
+ pandn m9, m1, m14
+ por m7, m9 ; don't clobber q2/m14 since we need it in K
+
+ ; sub p3/q1, add q4/q6 [reuse -p3,+q4 from E][-q1,+q6] J
+ ; write +3
+ paddw m10, [rsp+7*32]
+ paddw m11, [rsp+5*32] ; p2+p1+p0+q0+q1*2+q2*2+q3*2+q4*2+q5+q6*3
+ punpcklbw m8, m6, m0
+ punpckhbw m9, m6, m0
+ SWAP 2, 6
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m9 ; p2+p1+p0+q0+q1+q2*2+q3*2+q4*2+q5+q6*4
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, m15
+ por m8, m9
+ mova [tmpq+mstrideq], m8 ; q3
+
+ ; sub p2/q2, add q5/q6 [reuse -p2,+q5 from F][-q2,+q6] K
+ ; write +4
+ paddw m10, [rsp+ 9*32]
+ paddw m11, [rsp+10*32] ; p1+p0+q0+q1+q2*2+q3*2+q4*2+q5*2+q6*4
+ punpcklbw m8, m14, m0
+ punpckhbw m9, m14, m0
+ SWAP 14, 7
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m9 ; p1+p0+q0+q1+q2+q3*2+q4*2+q5*2+q6*5
+ pmulhrsw m8, m10, [pw_2048]
+ pmulhrsw m9, m11, [pw_2048]
+ packuswb m8, m9
+ pand m8, m1
+ pandn m9, m1, [tmpq+strideq*0]
+ por m8, m9
+ mova [tmpq+strideq*0], m8 ; q4
+
+ ; sub p1/q3, add q6*2 [reuse -p1,+q6 from G][-q3,+q6] L
+ ; write +5
+ paddw m10, [rsp+3*32]
+ paddw m11, [rsp+4*32] ; p1+p0+q0+q1+q2*2+q3*2+q4*2+q5*2+q6*4
+ punpcklbw m8, m15, m0
+ punpckhbw m9, m15, m0
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m9, [pb_m1_1]
+ paddw m10, m8
+ paddw m11, m9 ; p1+p0+q0+q1+q2+q3*2+q4*2+q5*2+q6*5
+ pmulhrsw m10, [pw_2048]
+ pmulhrsw m11, [pw_2048]
+ packuswb m10, m11
+ pand m10, m1
+ pandn m11, m1, [tmpq+strideq*1]
+ por m10, m11
+ mova [tmpq+strideq*1], m10 ; q5
+
+ mova m9, [rsp+0*32]
+ lea tmpq, [dstq+mstrideq*4]
+%endif
+%if %1 >= 8
+ ; flat8 filter
+
+ punpcklbw m0, m12, m3
+ punpckhbw m1, m12, m3
+ pmaddubsw m2, m0, [pb_3_1]
+ pmaddubsw m7, m1, [pb_3_1] ; 3 * p3 + p1
+ punpcklbw m8, m13, m4
+ punpckhbw m11, m13, m4
+ pmaddubsw m8, [pb_2_1]
+ pmaddubsw m11, [pb_2_1]
+ paddw m2, m8
+ paddw m7, m11 ; 3 * p3 + 2 * p2 + p1 + p0
+ punpcklbw m8, m5, [pb_4]
+ punpckhbw m11, m5, [pb_4]
+ pmaddubsw m8, [pb_1]
+ pmaddubsw m11, [pb_1]
+ paddw m2, m8
+ paddw m7, m11 ; 3 * p3 + 2 * p2 + p1 + p0 + q0 + 4
+ psrlw m8, m2, 3
+ psrlw m11, m7, 3
+ packuswb m8, m11
+ pand m8, m9
+ pandn m11, m9, m13
+ por m8, m11 ; p2
+ mova [tmpq+strideq*1], m8 ; p2
+
+ pmaddubsw m8, m0, [pb_m1_1]
+ pmaddubsw m11, m1, [pb_m1_1]
+ paddw m2, m8
+ paddw m7, m11
+ punpcklbw m8, m13, m6
+ punpckhbw m11, m13, m6
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m11, [pb_m1_1]
+ paddw m2, m8
+ paddw m7, m11 ; 2 * p3 + p2 + 2 * p1 + p0 + q0 + q1 + 4
+ psrlw m8, m2, 3
+ psrlw m11, m7, 3
+ packuswb m8, m11
+ pand m8, m9
+ pandn m11, m9, m3
+ por m8, m11 ; p1
+ mova [tmpq+strideq*2], m8 ; p1
+
+ pmaddubsw m0, [pb_1]
+ pmaddubsw m1, [pb_1]
+ psubw m2, m0
+ psubw m7, m1
+ punpcklbw m8, m4, m14
+ punpckhbw m11, m4, m14
+ pmaddubsw m8, [pb_1]
+ pmaddubsw m11, [pb_1]
+ paddw m2, m8
+ paddw m7, m11 ; p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2 + 4
+ psrlw m8, m2, 3
+ psrlw m11, m7, 3
+ packuswb m8, m11
+ pand m8, m9
+ pandn m11, m9, m4
+ por m8, m11 ; p0
+ mova [tmpq+stride3q ], m8 ; p0
+
+ punpcklbw m0, m5, m15
+ punpckhbw m1, m5, m15
+ pmaddubsw m8, m0, [pb_1]
+ pmaddubsw m11, m1, [pb_1]
+ paddw m2, m8
+ paddw m7, m11
+ punpcklbw m8, m4, m12
+ punpckhbw m11, m4, m12
+ pmaddubsw m8, [pb_1]
+ pmaddubsw m11, [pb_1]
+ psubw m2, m8
+ psubw m7, m11 ; p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3 + 4
+ psrlw m8, m2, 3
+ psrlw m11, m7, 3
+ packuswb m8, m11
+ pand m8, m9
+ pandn m11, m9, m5
+ por m8, m11 ; q0
+ mova [dstq+strideq*0], m8 ; q0
+
+ pmaddubsw m0, [pb_m1_1]
+ pmaddubsw m1, [pb_m1_1]
+ paddw m2, m0
+ paddw m7, m1
+ punpcklbw m8, m13, m6
+ punpckhbw m11, m13, m6
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m11, [pb_m1_1]
+ paddw m2, m8
+ paddw m7, m11 ; p1 + p0 + q0 + 2 * q1 + q2 + 2 * q3 + 4
+ psrlw m8, m2, 3
+ psrlw m11, m7, 3
+ packuswb m8, m11
+ pand m8, m9
+ pandn m11, m9, m6
+ por m8, m11 ; q1
+ mova [dstq+strideq*1], m8 ; q1
+
+ punpcklbw m0, m3, m6
+ punpckhbw m1, m3, m6
+ pmaddubsw m0, [pb_1]
+ pmaddubsw m1, [pb_1]
+ psubw m2, m0
+ psubw m7, m1
+ punpcklbw m0, m14, m15
+ punpckhbw m1, m14, m15
+ pmaddubsw m0, [pb_1]
+ pmaddubsw m1, [pb_1]
+ paddw m2, m0
+ paddw m7, m1 ; p0 + q0 + q1 + q2 + 2 * q2 + 3 * q3 + 4
+ psrlw m2, 3
+ psrlw m7, 3
+ packuswb m2, m7
+ pand m2, m9
+ pandn m11, m9, m14
+ por m2, m11 ; q2
+ mova [dstq+strideq*2], m2 ; q2
+%elif %1 == 6
+ ; flat6 filter
+
+ punpcklbw m8, m13, m5
+ punpckhbw m11, m13, m5
+ pmaddubsw m0, m8, [pb_3_1]
+ pmaddubsw m1, m11, [pb_3_1]
+ punpcklbw m7, m4, m3
+ punpckhbw m10, m4, m3
+ pmaddubsw m2, m7, [pb_2]
+ pmaddubsw m12, m10, [pb_2]
+ paddw m0, m2
+ paddw m1, m12
+ pmulhrsw m2, m0, [pw_4096]
+ pmulhrsw m12, m1, [pw_4096]
+ packuswb m2, m12
+ pand m2, m9
+ pandn m12, m9, m3
+ por m2, m12
+ mova [tmpq+strideq*2], m2 ; p1
+
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m11, [pb_m1_1]
+ paddw m0, m8
+ paddw m1, m11
+ punpcklbw m8, m13, m6
+ punpckhbw m11, m13, m6
+ pmaddubsw m8, [pb_m1_1]
+ pmaddubsw m11, [pb_m1_1]
+ paddw m0, m8
+ paddw m1, m11
+ pmulhrsw m12, m0, [pw_4096]
+ pmulhrsw m13, m1, [pw_4096]
+ packuswb m12, m13
+ pand m12, m9
+ pandn m13, m9, m4
+ por m12, m13
+ mova [tmpq+stride3q], m12 ; p0
+
+ paddw m0, m8
+ paddw m1, m11
+ punpcklbw m8, m3, m14
+ punpckhbw m11, m3, m14
+ pmaddubsw m12, m8, [pb_m1_1]
+ pmaddubsw m13, m11, [pb_m1_1]
+ paddw m0, m12
+ paddw m1, m13
+ pmulhrsw m12, m0, [pw_4096]
+ pmulhrsw m13, m1, [pw_4096]
+ packuswb m12, m13
+ pand m12, m9
+ pandn m13, m9, m5
+ por m12, m13
+ mova [dstq+strideq*0], m12 ; q0
+
+ pmaddubsw m8, [pb_m1_2]
+ pmaddubsw m11, [pb_m1_2]
+ paddw m0, m8
+ paddw m1, m11
+ pmaddubsw m7, [pb_m1_0]
+ pmaddubsw m10, [pb_m1_0]
+ paddw m0, m7
+ paddw m1, m10
+ pmulhrsw m0, [pw_4096]
+ pmulhrsw m1, [pw_4096]
+ packuswb m0, m1
+ pand m0, m9
+ pandn m9, m6
+ por m0, m9
+ mova [dstq+strideq*1], m0 ; q1
+%else
+ mova [tmpq+strideq*0], m3 ; p1
+ mova [tmpq+strideq*1], m4 ; p0
+ mova [tmpq+strideq*2], m5 ; q0
+ mova [tmpq+stride3q ], m6 ; q1
+%endif
+%endmacro
+
+INIT_YMM avx2
+cglobal lpf_v_sb128y, 7, 10, 16, 32 * 11, \
+ dst, stride, mask, l, l_stride, lut, \
+ w, stride3, mstride, tmp
+ shl l_strideq, 2
+ sub lq, l_strideq
+ mov mstrideq, strideq
+ neg mstrideq
+ lea stride3q, [strideq*3]
+
+.loop:
+ cmp byte [maskq+8], 0 ; vmask[2]
+ je .no_flat16
+
+ FILTER 16
+ jmp .end
+
+.no_flat16:
+ cmp byte [maskq+4], 0 ; vmask[1]
+ je .no_flat
+
+ FILTER 8
+ jmp .end
+
+.no_flat:
+ cmp byte [maskq+0], 0 ; vmask[0]
+ je .end
+
+ FILTER 4
+
+.end:
+ add lq, 32
+ add dstq, 32
+ add maskq, 1
+ sub wd, 8
+ jg .loop
+ RET
+
+INIT_YMM avx2
+cglobal lpf_v_sb128uv, 7, 10, 16, \
+ dst, stride, mask, l, l_stride, lut, \
+ w, stride3, mstride, tmp
+ shl l_strideq, 2
+ sub lq, l_strideq
+ mov mstrideq, strideq
+ neg mstrideq
+ lea stride3q, [strideq*3]
+
+.loop:
+ cmp byte [maskq+4], 0 ; vmask[1]
+ je .no_flat
+
+ FILTER 6
+ jmp .end
+
+.no_flat:
+ cmp byte [maskq+0], 0 ; vmask[0]
+ je .end
+
+ FILTER 4
+
+.end:
+ add lq, 32
+ add dstq, 32
+ add maskq, 1
+ sub wd, 8
+ jg .loop
+ RET
+
+%endif ; ARCH_X86_64
--- /dev/null
+++ b/src/x86/loopfilter_init.c
@@ -1,0 +1,43 @@
+/*
+ * Copyright © 2018, VideoLAN and dav1d authors
+ * Copyright © 2018, Two Orioles, LLC
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "src/cpu.h"
+#include "src/loopfilter.h"
+
+decl_loopfilter_sb_fn(dav1d_lpf_v_sb128y_avx2);
+decl_loopfilter_sb_fn(dav1d_lpf_v_sb128uv_avx2);
+
+void bitfn(dav1d_loop_filter_dsp_init_x86)(Dav1dLoopFilterDSPContext *const c) {+ const unsigned flags = dav1d_get_cpu_flags();
+
+ if (!(flags & DAV1D_X86_CPU_FLAG_AVX2)) return;
+
+#if BITDEPTH == 8 && ARCH_X86_64
+ c->loop_filter_sb128y = dav1d_lpf_v_sb128y_avx2;
+ c->loop_filter_sb128uv = dav1d_lpf_v_sb128uv_avx2;
+#endif
+}
--- a/tests/checkasm/loopfilter.c
+++ b/tests/checkasm/loopfilter.c
@@ -115,12 +115,14 @@
lut.i[level] = limit;
lut.e[level] = 2 * (level + 2) + limit;
}
+ lut.sharp[0] = (sharp + 3) >> 2;
+ lut.sharp[1] = sharp ? 9 - sharp : 0xff;
for (int i = 0; i < n_strengths; i++) {if (check_func(fn, "%s_w%d_%dbpc", name,
n_strengths == 3 ? 4 << i : 4 * 2 * i, BITDEPTH))
{- uint32_t vmask[3] = { 0 };+ uint32_t vmask[4] = { 0 };uint8_t l[32 * 2][4];
for (int j = 0; j < n_blks; j++) {