ref: 6fa9df9c6f6c620ea6f3d9848465adbdf958648f
dir: /src/x86/looprestoration.asm/
; 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_right_ext_mask: times 32 db 0xff
times 32 db 0
pb_14x0_1_2: times 14 db 0
db 1, 2
pb_0_to_15_min_n: db 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 13, 13
db 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 14
pb_15: times 16 db 15
pw_2048: times 2 dw 2048
pw_16380: times 2 dw 16380
pw_0_128: dw 0, 128
pd_1024: dd 1024
SECTION .text
INIT_YMM avx2
cglobal wiener_filter_h, 8, 12, 16, dst, left, src, stride, fh, w, h, edge
vpbroadcastb m15, [fhq+0]
vpbroadcastb m14, [fhq+2]
vpbroadcastb m13, [fhq+4]
vpbroadcastw m12, [fhq+6]
vpbroadcastd m11, [pw_2048]
vpbroadcastd m10, [pw_16380]
lea r11, [pb_right_ext_mask]
DEFINE_ARGS dst, left, src, stride, x, w, h, edge, srcptr, dstptr, xlim
; if (edge & has_right) align_w_to_32
; else w -= 32, and use that as limit in x loop
test edged, 2 ; has_right
jnz .align
mov xlimq, -3
jmp .loop
.align:
add wd, 31
and wd, ~31
xor xlimd, xlimd
; main y loop for vertical filter
.loop:
mov srcptrq, srcq
mov dstptrq, dstq
lea xq, [wq+xlimq]
; load left edge pixels
test edged, 1 ; have_left
jz .emu_left
test leftq, leftq ; left == NULL for the edge-extended bottom/top
jz .load_left_combined
movd xm0, [leftq]
add leftq, 4
pinsrd xm0, [srcq], 1
pslldq xm0, 9
jmp .left_load_done
.load_left_combined:
movq xm0, [srcq-3]
pslldq xm0, 10
jmp .left_load_done
.emu_left:
movd xm0, [srcq]
pshufb xm0, [pb_14x0_1_2]
; load right edge pixels
.left_load_done:
cmp xd, 32
jg .main_load
test xd, xd
jg .load_and_splat
je .splat_right
; for very small images (w=[1-2]), edge-extend the original cache,
; ugly, but only runs in very odd cases
add wd, wd
pshufb xm0, [r11-pb_right_ext_mask+pb_0_to_15_min_n+wq*8-16]
shr wd, 1
; main x loop, mostly this starts in .main_load
.splat_right:
; no need to load new pixels, just extend them from the (possibly previously
; extended) previous load into m0
pshufb xm1, xm0, [pb_15]
jmp .main_loop
.load_and_splat:
; load new pixels and extend edge for right-most
movu m1, [srcptrq+3]
sub r11, xq
movu m2, [r11-pb_right_ext_mask+pb_right_ext_mask+32]
add r11, xq
vpbroadcastb m3, [srcptrq+2+xq]
pand m1, m2
pandn m3, m2, m3
por m1, m3
jmp .main_loop
.main_load:
; load subsequent line
movu m1, [srcptrq+3]
.main_loop:
vinserti128 m0, xm1, 1
palignr m2, m1, m0, 10
palignr m3, m1, m0, 11
palignr m4, m1, m0, 12
palignr m5, m1, m0, 13
palignr m6, m1, m0, 14
palignr m7, m1, m0, 15
punpcklbw m0, m2, m1
punpckhbw m2, m1
punpcklbw m8, m3, m7
punpckhbw m3, m7
punpcklbw m7, m4, m6
punpckhbw m4, m6
pxor m9, m9
punpcklbw m6, m5, m9
punpckhbw m5, m9
pmaddubsw m0, m15
pmaddubsw m2, m15
pmaddubsw m8, m14
pmaddubsw m3, m14
pmaddubsw m7, m13
pmaddubsw m4, m13
paddw m0, m8
paddw m2, m3
psllw m8, m6, 7
psllw m3, m5, 7
psubw m8, m10
psubw m3, m10
pmullw m6, m12
pmullw m5, m12
paddw m0, m7
paddw m2, m4
paddw m0, m6
paddw m2, m5
paddsw m0, m8
paddsw m2, m3
psraw m0, 3
psraw m2, 3
paddw m0, m11
paddw m2, m11
mova [dstptrq], xm0
mova [dstptrq+16], xm2
vextracti128 [dstptrq+32], m0, 1
vextracti128 [dstptrq+48], m2, 1
vextracti128 xm0, m1, 1
add srcptrq, 32
add dstptrq, 64
sub xq, 32
cmp xd, 32
jg .main_load
test xd, xd
jg .load_and_splat
cmp xd, xlimd
jg .splat_right
add srcq, strideq
add dstq, 384*2
dec hd
jg .loop
RET
cglobal wiener_filter_v, 7, 10, 16, dst, stride, mid, w, h, fv, edge
vpbroadcastd m14, [fvq+4]
vpbroadcastd m15, [fvq]
vpbroadcastd m13, [pw_0_128]
paddw m14, m13
vpbroadcastd m12, [pd_1024]
DEFINE_ARGS dst, stride, mid, w, h, ylim, edge, y, mptr, dstptr
mov ylimd, edged
and ylimd, 8 ; have_bottom
shr ylimd, 2
sub ylimd, 3
; main x loop for vertical filter, does one column of 16 pixels
.loop_x:
mova m3, [midq] ; middle line
; load top pixels
test edged, 4 ; have_top
jz .emu_top
mova m0, [midq-384*4]
mova m2, [midq-384*2]
mova m1, m0
jmp .load_bottom_pixels
.emu_top:
mova m0, m3
mova m1, m3
mova m2, m3
; load bottom pixels
.load_bottom_pixels:
mov yd, hd
mov mptrq, midq
mov dstptrq, dstq
add yd, ylimd
jg .load_threelines
; the remainder here is somewhat messy but only runs in very weird
; circumstances at the bottom of the image in very small blocks (h=[1-3]),
; so performance is not terribly important here...
je .load_twolines
cmp yd, -1
je .load_oneline
; h == 1 case
mova m5, m3
mova m4, m3
mova m6, m3
jmp .loop
.load_oneline:
; h == 2 case
mova m4, [midq+384*2]
mova m5, m4
mova m6, m4
jmp .loop
.load_twolines:
; h == 3 case
mova m4, [midq+384*2]
mova m5, [midq+384*4]
mova m6, m5
jmp .loop
.load_threelines:
; h > 3 case
mova m4, [midq+384*2]
mova m5, [midq+384*4]
; third line loaded in main loop below
; main y loop for vertical filter
.loop_load:
; load one line into m6. if that pixel is no longer available, do
; nothing, since m6 still has the data from the previous line in it. We
; try to structure the loop so that the common case is evaluated fastest
mova m6, [mptrq+384*6]
.loop:
paddw m7, m0, m6
paddw m8, m1, m5
paddw m9, m2, m4
punpcklwd m10, m7, m8
punpckhwd m7, m8
punpcklwd m11, m9, m3
punpckhwd m9, m3
pmaddwd m10, m15
pmaddwd m7, m15
pmaddwd m11, m14
pmaddwd m9, m14
paddd m10, m11
paddd m7, m9
paddd m10, m12
paddd m7, m12
psrad m10, 11
psrad m7, 11
packssdw m10, m7
packuswb m10, m10
vpermq m10, m10, q3120
mova [dstptrq], xm10
; shift pixels one position
mova m0, m1
mova m1, m2
mova m2, m3
mova m3, m4
mova m4, m5
mova m5, m6
add dstptrq, strideq
add mptrq, 384*2
dec yd
jg .loop_load
; for the bottom pixels, continue using m6 (as extended edge)
cmp yd, ylimd
jg .loop
add dstq, 16
add midq, 32
sub wd, 16
jg .loop_x
RET
%endif ; ARCH_X86_64