1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-11-26 19:01:44 +02:00
FFmpeg/libavcodec/x86/flacdsp.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

327 lines
7.9 KiB
NASM

;******************************************************************************
;* FLAC DSP SIMD optimizations
;*
;* Copyright (C) 2014 Loren Merritt
;* Copyright (C) 2014 James Almer
;*
;* 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_RODATA
vector: db 0,1,4,5,8,9,12,13,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,0,1,4,5,8,9,12,13,
SECTION .text
%macro PMACSDQL 5
%if cpuflag(xop)
pmacsdql %1, %2, %3, %1
%else
pmuldq %2, %3
paddq %1, %2
%endif
%endmacro
%macro LPC_32 1
INIT_XMM %1
cglobal flac_lpc_32, 5,6,5, decoded, coeffs, pred_order, qlevel, len, j
sub lend, pred_orderd
jle .ret
lea decodedq, [decodedq+pred_orderq*4-8]
lea coeffsq, [coeffsq+pred_orderq*4]
neg pred_orderq
movd m4, qlevelm
ALIGN 16
.loop_sample:
movd m0, [decodedq+pred_orderq*4+8]
add decodedq, 8
movd m1, [coeffsq+pred_orderq*4]
pxor m2, m2
pxor m3, m3
lea jq, [pred_orderq+1]
test jq, jq
jz .end_order
.loop_order:
PMACSDQL m2, m0, m1, m2, m0
movd m0, [decodedq+jq*4]
PMACSDQL m3, m1, m0, m3, m1
movd m1, [coeffsq+jq*4]
inc jq
jl .loop_order
.end_order:
PMACSDQL m2, m0, m1, m2, m0
psrlq m2, m4
movd m0, [decodedq]
paddd m0, m2
movd [decodedq], m0
sub lend, 2
jl .ret
PMACSDQL m3, m1, m0, m3, m1
psrlq m3, m4
movd m1, [decodedq+4]
paddd m1, m3
movd [decodedq+4], m1
jg .loop_sample
.ret:
RET
%endmacro
%if HAVE_XOP_EXTERNAL
LPC_32 xop
%endif
LPC_32 sse4
;----------------------------------------------------------------------------------
;void ff_flac_decorrelate_[lrm]s_16_sse2(uint8_t **out, int32_t **in, int channels,
; int len, int shift);
;----------------------------------------------------------------------------------
%macro FLAC_DECORRELATE_16 3-4
cglobal flac_decorrelate_%1_16, 2, 4, 4, out, in0, in1, len
%ifidn %1, indep2
VBROADCASTI128 m2, [vector]
%endif
%if ARCH_X86_32
mov lend, lenm
%endif
movd m3, r4m
shl lend, 2
mov in1q, [in0q + gprsize]
mov in0q, [in0q]
mov outq, [outq]
add in1q, lenq
add in0q, lenq
add outq, lenq
neg lenq
align 16
.loop:
mova m0, [in0q + lenq]
mova m1, [in1q + lenq]
%ifidn %1, ms
psrad m2, m1, 1
psubd m0, m2
%endif
%ifnidn %1, indep2
p%4d m2, m0, m1
packssdw m%2, m%2
packssdw m%3, m%3
punpcklwd m%2, m%3
psllw m%2, m3
%else
pslld m%2, m3
pslld m%3, m3
pshufb m%2, m%2, m2
pshufb m%3, m%3, m2
punpcklwd m%2, m%3
%endif
mova [outq + lenq], m%2
add lenq, 16
jl .loop
RET
%endmacro
INIT_XMM sse2
FLAC_DECORRELATE_16 ls, 0, 2, sub
FLAC_DECORRELATE_16 rs, 2, 1, add
FLAC_DECORRELATE_16 ms, 2, 0, add
;----------------------------------------------------------------------------------
;void ff_flac_decorrelate_[lrm]s_32_sse2(uint8_t **out, int32_t **in, int channels,
; int len, int shift);
;----------------------------------------------------------------------------------
%macro FLAC_DECORRELATE_32 5
cglobal flac_decorrelate_%1_32, 2, 4, 4, out, in0, in1, len
%if ARCH_X86_32
mov lend, lenm
%endif
movd m3, r4m
mov in1q, [in0q + gprsize]
mov in0q, [in0q]
mov outq, [outq]
sub in1q, in0q
align 16
.loop:
mova m0, [in0q]
mova m1, [in0q + in1q]
%ifidn %1, ms
psrad m2, m1, 1
psubd m0, m2
%endif
p%5d m2, m0, m1
pslld m%2, m3
pslld m%3, m3
SBUTTERFLY dq, %2, %3, %4
mova [outq ], m%2
mova [outq + mmsize], m%3
add in0q, mmsize
add outq, mmsize*2
sub lend, mmsize/4
jg .loop
RET
%endmacro
INIT_XMM sse2
FLAC_DECORRELATE_32 ls, 0, 2, 1, sub
FLAC_DECORRELATE_32 rs, 2, 1, 0, add
FLAC_DECORRELATE_32 ms, 2, 0, 1, add
;-----------------------------------------------------------------------------------------
;void ff_flac_decorrelate_indep<ch>_<bps>_<opt>(uint8_t **out, int32_t **in, int channels,
; int len, int shift);
;-----------------------------------------------------------------------------------------
;%1 = bps
;%2 = channels
;%3 = last xmm reg used
;%4 = word/dword (shift instruction)
%macro FLAC_DECORRELATE_INDEP 4
%define REPCOUNT %2/(32/%1) ; 16bits = channels / 2; 32bits = channels
cglobal flac_decorrelate_indep%2_%1, 2, %2+2, %3+1, out, in0, in1, len, in2, in3, in4, in5, in6, in7
%if ARCH_X86_32
%if %2 == 6
DEFINE_ARGS out, in0, in1, in2, in3, in4, in5
%define lend dword r3m
%else
mov lend, lenm
%endif
%endif
movd m%3, r4m
%assign %%i 1
%rep %2-1
mov in %+ %%i %+ q, [in0q+%%i*gprsize]
%assign %%i %%i+1
%endrep
mov in0q, [in0q]
mov outq, [outq]
%assign %%i 1
%rep %2-1
sub in %+ %%i %+ q, in0q
%assign %%i %%i+1
%endrep
align 16
.loop:
mova m0, [in0q]
%assign %%i 1
%rep REPCOUNT-1
mova m %+ %%i, [in0q + in %+ %%i %+ q]
%assign %%i %%i+1
%endrep
%if %1 == 32
%if %2 == 8
TRANSPOSE8x4D 0, 1, 2, 3, 4, 5, 6, 7, 8
%elif %2 == 6
SBUTTERFLY dq, 0, 1, 6
SBUTTERFLY dq, 2, 3, 6
SBUTTERFLY dq, 4, 5, 6
punpcklqdq m6, m0, m2
punpckhqdq m2, m4
shufps m4, m0, 0xe4
punpcklqdq m0, m1, m3
punpckhqdq m3, m5
shufps m5, m1, 0xe4
SWAP 0,6,1,4,5,3
%elif %2 == 4
TRANSPOSE4x4D 0, 1, 2, 3, 4
%else ; %2 == 2
SBUTTERFLY dq, 0, 1, 2
%endif
%else ; %1 == 16
%if %2 == 8
packssdw m0, [in0q + in4q]
packssdw m1, [in0q + in5q]
packssdw m2, [in0q + in6q]
packssdw m3, [in0q + in7q]
TRANSPOSE2x4x4W 0, 1, 2, 3, 4
%elif %2 == 6
packssdw m0, [in0q + in3q]
packssdw m1, [in0q + in4q]
packssdw m2, [in0q + in5q]
pshufd m3, m0, q1032
punpcklwd m0, m1
punpckhwd m1, m2
punpcklwd m2, m3
shufps m3, m0, m2, q2020
shufps m0, m1, q2031
shufps m2, m1, q3131
shufps m1, m2, m3, q3120
shufps m3, m0, q0220
shufps m0, m2, q3113
SWAP 2, 0, 3
%else ; %2 == 4
packssdw m0, [in0q + in2q]
packssdw m1, [in0q + in3q]
SBUTTERFLY wd, 0, 1, 2
SBUTTERFLY dq, 0, 1, 2
%endif
%endif
%assign %%i 0
%rep REPCOUNT
psll%4 m %+ %%i, m%3
%assign %%i %%i+1
%endrep
%assign %%i 0
%rep REPCOUNT
mova [outq + %%i*mmsize], m %+ %%i
%assign %%i %%i+1
%endrep
add in0q, mmsize
add outq, mmsize*REPCOUNT
sub lend, mmsize/4
jg .loop
RET
%endmacro
INIT_XMM ssse3
FLAC_DECORRELATE_16 indep2, 0, 1 ; Reuse stereo 16bits macro
FLAC_DECORRELATE_INDEP 32, 2, 3, d
FLAC_DECORRELATE_INDEP 16, 4, 3, w
FLAC_DECORRELATE_INDEP 32, 4, 5, d
FLAC_DECORRELATE_INDEP 16, 6, 4, w
FLAC_DECORRELATE_INDEP 32, 6, 7, d
%if ARCH_X86_64
FLAC_DECORRELATE_INDEP 16, 8, 5, w
FLAC_DECORRELATE_INDEP 32, 8, 9, d
%endif
INIT_XMM avx
FLAC_DECORRELATE_INDEP 32, 4, 5, d
FLAC_DECORRELATE_INDEP 32, 6, 7, d
%if ARCH_X86_64
FLAC_DECORRELATE_INDEP 16, 8, 5, w
FLAC_DECORRELATE_INDEP 32, 8, 9, d
%endif