generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
7e40a55ef9
vpx/vp9/encoder/x86/vp9_highbd_sad4d_sse2.asm
Peter de Rivaz 7eee487c00 Added highbitdepth sse2 SAD acceleration and tests 
Change-Id: I1a74a1b032b198793ef9cc526327987f7799125f
(cherry picked from commit b1a6f6b9cb)
2014-11-12 14:25:45 -08:00

285 lines
8.9 KiB
NASM
Raw Blame History

;
; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
;
; Use of this source code is governed by a BSD-style license
; that can be found in the LICENSE file in the root of the source
; tree. An additional intellectual property rights grant can be found
; in the file PATENTS. All contributing project authors may
; be found in the AUTHORS file in the root of the source tree.
;
%include "third_party/x86inc/x86inc.asm"
SECTION .text
; HIGH_PROCESS_4x2x4 first, off_{first,second}_{src,ref}, advance_at_end
%macro HIGH_PROCESS_4x2x4 5-6 0
movh m0, [srcq +%2*2]
%if %1 == 1
movu m4, [ref1q+%3*2]
movu m5, [ref2q+%3*2]
movu m6, [ref3q+%3*2]
movu m7, [ref4q+%3*2]
movhps m0, [srcq +%4*2]
movhps m4, [ref1q+%5*2]
movhps m5, [ref2q+%5*2]
movhps m6, [ref3q+%5*2]
movhps m7, [ref4q+%5*2]
mova m3, m0
mova m2, m0
psubusw m3, m4
psubusw m2, m5
psubusw m4, m0
psubusw m5, m0
por m4, m3
por m5, m2
pmaddwd m4, m1
pmaddwd m5, m1
mova m3, m0
mova m2, m0
psubusw m3, m6
psubusw m2, m7
psubusw m6, m0
psubusw m7, m0
por m6, m3
por m7, m2
pmaddwd m6, m1
pmaddwd m7, m1
%else
movu m2, [ref1q+%3*2]
movhps m0, [srcq +%4*2]
movhps m2, [ref1q+%5*2]
mova m3, m0
psubusw m3, m2
psubusw m2, m0
por m2, m3
pmaddwd m2, m1
paddd m4, m2
movu m2, [ref2q+%3*2]
mova m3, m0
movhps m2, [ref2q+%5*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
pmaddwd m2, m1
paddd m5, m2
movu m2, [ref3q+%3*2]
mova m3, m0
movhps m2, [ref3q+%5*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
pmaddwd m2, m1
paddd m6, m2
movu m2, [ref4q+%3*2]
mova m3, m0
movhps m2, [ref4q+%5*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
pmaddwd m2, m1
paddd m7, m2
%endif
%if %6 == 1
lea srcq, [srcq +src_strideq*4]
lea ref1q, [ref1q+ref_strideq*4]
lea ref2q, [ref2q+ref_strideq*4]
lea ref3q, [ref3q+ref_strideq*4]
lea ref4q, [ref4q+ref_strideq*4]
%endif
%endmacro
; PROCESS_8x2x4 first, off_{first,second}_{src,ref}, advance_at_end
%macro HIGH_PROCESS_8x2x4 5-6 0
; 1st 8 px
mova m0, [srcq +%2*2]
%if %1 == 1
movu m4, [ref1q+%3*2]
movu m5, [ref2q+%3*2]
movu m6, [ref3q+%3*2]
movu m7, [ref4q+%3*2]
mova m3, m0
mova m2, m0
psubusw m3, m4
psubusw m2, m5
psubusw m4, m0
psubusw m5, m0
por m4, m3
por m5, m2
pmaddwd m4, m1
pmaddwd m5, m1
mova m3, m0
mova m2, m0
psubusw m3, m6
psubusw m2, m7
psubusw m6, m0
psubusw m7, m0
por m6, m3
por m7, m2
pmaddwd m6, m1
pmaddwd m7, m1
%else
mova m3, m0
movu m2, [ref1q+%3*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
mova m3, m0
pmaddwd m2, m1
paddd m4, m2
movu m2, [ref2q+%3*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
mova m3, m0
pmaddwd m2, m1
paddd m5, m2
movu m2, [ref3q+%3*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
mova m3, m0
pmaddwd m2, m1
paddd m6, m2
movu m2, [ref4q+%3*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
pmaddwd m2, m1
paddd m7, m2
%endif
; 2nd 8 px
mova m0, [srcq +(%4)*2]
mova m3, m0
movu m2, [ref1q+(%5)*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
mova m3, m0
pmaddwd m2, m1
paddd m4, m2
movu m2, [ref2q+(%5)*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
mova m3, m0
pmaddwd m2, m1
paddd m5, m2
movu m2, [ref3q+(%5)*2]
psubusw m3, m2
psubusw m2, m0
por m2, m3
mova m3, m0
pmaddwd m2, m1
paddd m6, m2
movu m2, [ref4q+(%5)*2]
psubusw m3, m2
psubusw m2, m0
%if %6 == 1
lea srcq, [srcq +src_strideq*4]
lea ref1q, [ref1q+ref_strideq*4]
lea ref2q, [ref2q+ref_strideq*4]
lea ref3q, [ref3q+ref_strideq*4]
lea ref4q, [ref4q+ref_strideq*4]
%endif
por m2, m3
pmaddwd m2, m1
paddd m7, m2
%endmacro
; HIGH_PROCESS_16x2x4 first, off_{first,second}_{src,ref}, advance_at_end
%macro HIGH_PROCESS_16x2x4 5-6 0
HIGH_PROCESS_8x2x4 %1, %2, %3, (%2 + 8), (%3 + 8)
HIGH_PROCESS_8x2x4 0, %4, %5, (%4 + 8), (%5 + 8), %6
%endmacro
; HIGH_PROCESS_32x2x4 first, off_{first,second}_{src,ref}, advance_at_end
%macro HIGH_PROCESS_32x2x4 5-6 0
HIGH_PROCESS_16x2x4 %1, %2, %3, (%2 + 16), (%3 + 16)
HIGH_PROCESS_16x2x4 0, %4, %5, (%4 + 16), (%5 + 16), %6
%endmacro
; HIGH_PROCESS_64x2x4 first, off_{first,second}_{src,ref}, advance_at_end
%macro HIGH_PROCESS_64x2x4 5-6 0
HIGH_PROCESS_32x2x4 %1, %2, %3, (%2 + 32), (%3 + 32)
HIGH_PROCESS_32x2x4 0, %4, %5, (%4 + 32), (%5 + 32), %6
%endmacro
; void vp9_highbd_sadNxNx4d_sse2(uint8_t *src, int src_stride,
; uint8_t *ref[4], int ref_stride,
; unsigned int res[4]);
; where NxN = 64x64, 32x32, 16x16, 16x8, 8x16 or 8x8
%macro HIGH_SADNXN4D 2
%if UNIX64
cglobal highbd_sad%1x%2x4d, 5, 9, 8, src, src_stride, ref1, ref_stride, \
res, ref2, ref3, ref4, one
%else
cglobal highbd_sad%1x%2x4d, 4, 8, 8, src, src_stride, ref1, ref_stride, \
ref2, ref3, ref4, one
%endif
movsxdifnidn src_strideq, src_strided
movsxdifnidn ref_strideq, ref_strided
mov ref2q, [ref1q+gprsize*1]
mov ref3q, [ref1q+gprsize*2]
mov ref4q, [ref1q+gprsize*3]
mov ref1q, [ref1q+gprsize*0]
; convert byte pointers to short pointers
shl srcq, 1
shl ref2q, 1
shl ref3q, 1
shl ref4q, 1
shl ref1q, 1
mov oned, 0x00010001
movd m1, oned
pshufd m1, m1, 0x0
HIGH_PROCESS_%1x2x4 1, 0, 0, src_strideq, ref_strideq, 1
%rep (%2-4)/2
HIGH_PROCESS_%1x2x4 0, 0, 0, src_strideq, ref_strideq, 1
%endrep
HIGH_PROCESS_%1x2x4 0, 0, 0, src_strideq, ref_strideq, 0
; N.B. HIGH_PROCESS outputs dwords (32 bits)
; so in high bit depth even the smallest width (4) needs 128bits i.e. XMM
movhlps m0, m4
movhlps m1, m5
movhlps m2, m6
movhlps m3, m7
paddd m4, m0
paddd m5, m1
paddd m6, m2
paddd m7, m3
punpckldq m4, m5
punpckldq m6, m7
movhlps m0, m4
movhlps m1, m6
paddd m4, m0
paddd m6, m1
punpcklqdq m4, m6
movifnidn r4, r4mp
movu [r4], m4
RET
%endmacro
INIT_XMM sse2
HIGH_SADNXN4D 64, 64
HIGH_SADNXN4D 64, 32
HIGH_SADNXN4D 32, 64
HIGH_SADNXN4D 32, 32
HIGH_SADNXN4D 32, 16
HIGH_SADNXN4D 16, 32
HIGH_SADNXN4D 16, 16
HIGH_SADNXN4D 16, 8
HIGH_SADNXN4D 8, 16
HIGH_SADNXN4D 8, 8
HIGH_SADNXN4D 8, 4
HIGH_SADNXN4D 4, 8
HIGH_SADNXN4D 4, 4
Reference in New Issue View Git Blame Copy Permalink