1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-11-21 10:55:51 +02:00
FFmpeg/libavutil/x86/lls.asm
Lynne bbe95f7353
x86: replace explicit REP_RETs with RETs
From x86inc:
> On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either
> a branch or a branch target. So switch to a 2-byte form of ret in that case.
> We can automatically detect "follows a branch", but not a branch target.
> (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.)

x86inc can automatically determine whether to use REP_RET rather than
REP in most of these cases, so impact is minimal. Additionally, a few
REP_RETs were used unnecessary, despite the return being nowhere near a
branch.

The only CPUs affected were AMD K10s, made between 2007 and 2011, 16
years ago and 12 years ago, respectively.

In the future, everyone involved with x86inc should consider dropping
REP_RETs altogether.
2023-02-01 04:23:55 +01:00

291 lines
7.6 KiB
NASM

;******************************************************************************
;* linear least squares model
;*
;* Copyright (c) 2013 Loren Merritt
;*
;* This file is part of FFmpeg.
;*
;* FFmpeg is free software; you can redistribute it and/or
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
;* FFmpeg is distributed in the hope that it will be useful,
;* but WITHOUT ANY WARRANTY; without even the implied warranty of
;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
;* Lesser General Public License for more details.
;*
;* You should have received a copy of the GNU Lesser General Public
;* License along with FFmpeg; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
%include "libavutil/x86/x86util.asm"
SECTION .text
%define MAX_VARS 32
%define MAX_VARS_ALIGN (MAX_VARS+4)
%define COVAR_STRIDE MAX_VARS_ALIGN*8
%define COVAR(x,y) [covarq + (x)*8 + (y)*COVAR_STRIDE]
struc LLSModel
.covariance: resq MAX_VARS_ALIGN*MAX_VARS_ALIGN
.coeff: resq MAX_VARS*MAX_VARS
.variance: resq MAX_VARS
.indep_count: resd 1
endstruc
%macro ADDPD_MEM 2
%if cpuflag(avx)
vaddpd %2, %2, %1
%else
addpd %2, %1
%endif
mova %1, %2
%endmacro
INIT_XMM sse2
%define movdqa movaps
cglobal update_lls, 2,5,8, ctx, var, i, j, covar2
%define covarq ctxq
mov id, [ctxq + LLSModel.indep_count]
lea varq, [varq + iq*8]
neg iq
mov covar2q, covarq
.loopi:
; Compute all 3 pairwise products of a 2x2 block that lies on the diagonal
mova m1, [varq + iq*8]
mova m3, [varq + iq*8 + 16]
pshufd m4, m1, q1010
pshufd m5, m1, q3232
pshufd m6, m3, q1010
pshufd m7, m3, q3232
mulpd m0, m1, m4
mulpd m1, m1, m5
lea covarq, [covar2q + 16]
ADDPD_MEM COVAR(-2,0), m0
ADDPD_MEM COVAR(-2,1), m1
lea jq, [iq + 2]
cmp jd, -2
jg .skip4x4
.loop4x4:
; Compute all 16 pairwise products of a 4x4 block
mulpd m0, m4, m3
mulpd m1, m5, m3
mulpd m2, m6, m3
mulpd m3, m3, m7
ADDPD_MEM COVAR(0,0), m0
ADDPD_MEM COVAR(0,1), m1
ADDPD_MEM COVAR(0,2), m2
ADDPD_MEM COVAR(0,3), m3
mova m3, [varq + jq*8 + 16]
mulpd m0, m4, m3
mulpd m1, m5, m3
mulpd m2, m6, m3
mulpd m3, m3, m7
ADDPD_MEM COVAR(2,0), m0
ADDPD_MEM COVAR(2,1), m1
ADDPD_MEM COVAR(2,2), m2
ADDPD_MEM COVAR(2,3), m3
mova m3, [varq + jq*8 + 32]
add covarq, 32
add jq, 4
cmp jd, -2
jle .loop4x4
.skip4x4:
test jd, jd
jg .skip2x4
mulpd m4, m3
mulpd m5, m3
mulpd m6, m3
mulpd m7, m3
ADDPD_MEM COVAR(0,0), m4
ADDPD_MEM COVAR(0,1), m5
ADDPD_MEM COVAR(0,2), m6
ADDPD_MEM COVAR(0,3), m7
.skip2x4:
add iq, 4
add covar2q, 4*COVAR_STRIDE+32
cmp id, -2
jle .loopi
test id, id
jg .ret
mov jq, iq
%define covarq covar2q
.loop2x1:
movsd m0, [varq + iq*8]
movlhps m0, m0
mulpd m0, [varq + jq*8]
ADDPD_MEM COVAR(0,0), m0
inc iq
add covarq, COVAR_STRIDE
test id, id
jle .loop2x1
.ret:
RET
%macro UPDATE_LLS 0
cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2
%define covarq ctxq
mov countd, [ctxq + LLSModel.indep_count]
lea count2d, [countq-2]
xor id, id
.loopi:
; Compute all 10 pairwise products of a 4x4 block that lies on the diagonal
mova ymm1, [varq + iq*8]
vbroadcastsd ymm4, [varq + iq*8]
vbroadcastsd ymm5, [varq + iq*8 + 8]
vbroadcastsd ymm6, [varq + iq*8 + 16]
vbroadcastsd ymm7, [varq + iq*8 + 24]
vextractf128 xmm3, ymm1, 1
%if cpuflag(fma3)
mova ymm0, COVAR(iq ,0)
mova xmm2, COVAR(iq+2,2)
fmaddpd ymm0, ymm1, ymm4, ymm0
fmaddpd xmm2, xmm3, xmm6, xmm2
fmaddpd ymm1, ymm5, ymm1, COVAR(iq ,1)
fmaddpd xmm3, xmm7, xmm3, COVAR(iq+2,3)
mova COVAR(iq ,0), ymm0
mova COVAR(iq ,1), ymm1
mova COVAR(iq+2,2), xmm2
mova COVAR(iq+2,3), xmm3
%else
vmulpd ymm0, ymm1, ymm4
vmulpd ymm1, ymm1, ymm5
vmulpd xmm2, xmm3, xmm6
vmulpd xmm3, xmm3, xmm7
ADDPD_MEM COVAR(iq ,0), ymm0
ADDPD_MEM COVAR(iq ,1), ymm1
ADDPD_MEM COVAR(iq+2,2), xmm2
ADDPD_MEM COVAR(iq+2,3), xmm3
%endif ; cpuflag(fma3)
lea jd, [iq + 4]
cmp jd, count2d
jg .skip4x4
.loop4x4:
; Compute all 16 pairwise products of a 4x4 block
mova ymm3, [varq + jq*8]
%if cpuflag(fma3)
mova ymm0, COVAR(jq, 0)
mova ymm1, COVAR(jq, 1)
mova ymm2, COVAR(jq, 2)
fmaddpd ymm0, ymm3, ymm4, ymm0
fmaddpd ymm1, ymm3, ymm5, ymm1
fmaddpd ymm2, ymm3, ymm6, ymm2
fmaddpd ymm3, ymm7, ymm3, COVAR(jq,3)
mova COVAR(jq, 0), ymm0
mova COVAR(jq, 1), ymm1
mova COVAR(jq, 2), ymm2
mova COVAR(jq, 3), ymm3
%else
vmulpd ymm0, ymm3, ymm4
vmulpd ymm1, ymm3, ymm5
vmulpd ymm2, ymm3, ymm6
vmulpd ymm3, ymm3, ymm7
ADDPD_MEM COVAR(jq,0), ymm0
ADDPD_MEM COVAR(jq,1), ymm1
ADDPD_MEM COVAR(jq,2), ymm2
ADDPD_MEM COVAR(jq,3), ymm3
%endif ; cpuflag(fma3)
add jd, 4
cmp jd, count2d
jle .loop4x4
.skip4x4:
cmp jd, countd
jg .skip2x4
mova xmm3, [varq + jq*8]
%if cpuflag(fma3)
mova xmm0, COVAR(jq, 0)
mova xmm1, COVAR(jq, 1)
mova xmm2, COVAR(jq, 2)
fmaddpd xmm0, xmm3, xmm4, xmm0
fmaddpd xmm1, xmm3, xmm5, xmm1
fmaddpd xmm2, xmm3, xmm6, xmm2
fmaddpd xmm3, xmm7, xmm3, COVAR(jq,3)
mova COVAR(jq, 0), xmm0
mova COVAR(jq, 1), xmm1
mova COVAR(jq, 2), xmm2
mova COVAR(jq, 3), xmm3
%else
vmulpd xmm0, xmm3, xmm4
vmulpd xmm1, xmm3, xmm5
vmulpd xmm2, xmm3, xmm6
vmulpd xmm3, xmm3, xmm7
ADDPD_MEM COVAR(jq,0), xmm0
ADDPD_MEM COVAR(jq,1), xmm1
ADDPD_MEM COVAR(jq,2), xmm2
ADDPD_MEM COVAR(jq,3), xmm3
%endif ; cpuflag(fma3)
.skip2x4:
add id, 4
add covarq, 4*COVAR_STRIDE
cmp id, count2d
jle .loopi
cmp id, countd
jg .ret
mov jd, id
.loop2x1:
vmovddup xmm0, [varq + iq*8]
%if cpuflag(fma3)
mova xmm1, [varq + jq*8]
fmaddpd xmm0, xmm1, xmm0, COVAR(jq,0)
mova COVAR(jq,0), xmm0
%else
vmulpd xmm0, [varq + jq*8]
ADDPD_MEM COVAR(jq,0), xmm0
%endif ; cpuflag(fma3)
inc id
add covarq, COVAR_STRIDE
cmp id, countd
jle .loop2x1
.ret:
RET
%endmacro ; UPDATE_LLS
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
UPDATE_LLS
%endif
%if HAVE_FMA3_EXTERNAL
INIT_YMM fma3
UPDATE_LLS
%endif
INIT_XMM sse2
cglobal evaluate_lls, 3,4,2, ctx, var, order, i
; This function is often called on the same buffer as update_lls, but with
; an offset. They can't both be aligned.
; Load halves rather than movu to avoid store-forwarding stalls, since the
; input was initialized immediately prior to this function using scalar math.
%define coefsq ctxq
mov id, orderd
imul orderd, MAX_VARS
lea coefsq, [ctxq + LLSModel.coeff + orderq*8]
movsd m0, [varq]
movhpd m0, [varq + 8]
mulpd m0, [coefsq]
lea coefsq, [coefsq + iq*8]
lea varq, [varq + iq*8]
neg iq
add iq, 2
.loop:
movsd m1, [varq + iq*8]
movhpd m1, [varq + iq*8 + 8]
mulpd m1, [coefsq + iq*8]
addpd m0, m1
add iq, 2
jl .loop
jg .skip1
movsd m1, [varq + iq*8]
mulsd m1, [coefsq + iq*8]
addpd m0, m1
.skip1:
movhlps m1, m0
addsd m0, m1
%if ARCH_X86_32
movsd r0m, m0
fld qword r0m
%endif
RET