highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

x86/flacdsp: add a SSE4 version of lpc16

Browse Source 
flac_lpc_16_13_c: 2841.3
flac_lpc_16_13_sse4: 2151.8
flac_lpc_16_16_c: 3382.8
flac_lpc_16_16_sse4: 2228.3
flac_lpc_16_29_c: 5800.3
flac_lpc_16_29_sse4: 3727.3
flac_lpc_16_32_c: 5972.8
flac_lpc_16_32_sse4: 4052.3

Signed-off-by: James Almer <jamrial@gmail.com>
This commit is contained in:
James Almer
2024-05-11 23:57:57 -03:00
parent 479d26cea2
commit c00c8679ed
2 changed files with 10 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
libavcodec/x86/flacdsp.asm
View File

@ -38,9 +38,9 @@ SECTION .text
%endif %endif
%endmacro %endmacro
%macro LPC_32 1 %macro LPC_32 3
INIT_XMM %1 INIT_XMM %1
cglobal flac_lpc_32, 5,6,5, decoded, coeffs, pred_order, qlevel, len, j cglobal flac_lpc_%2, 5,6,5, decoded, coeffs, pred_order, qlevel, len, j
sub lend, pred_orderd sub lend, pred_orderd
jle .ret jle .ret
movsxdifnidn pred_orderq, pred_orderd movsxdifnidn pred_orderq, pred_orderd
@ -67,14 +67,14 @@ ALIGN 16
jl .loop_order jl .loop_order
.end_order: .end_order:
PMACSDQL m2, m0, m1, m2, m0 PMACSDQL m2, m0, m1, m2, m0
psrlq m2, m4 %3 m2, m4
movd m0, [decodedq] movd m0, [decodedq]
paddd m0, m2 paddd m0, m2
movd [decodedq], m0 movd [decodedq], m0
sub lend, 2 sub lend, 2
jl .ret jl .ret
PMACSDQL m3, m1, m0, m3, m1 PMACSDQL m3, m1, m0, m3, m1
psrlq m3, m4 %3 m3, m4
movd m1, [decodedq+4] movd m1, [decodedq+4]
paddd m1, m3 paddd m1, m3
movd [decodedq+4], m1 movd [decodedq+4], m1
@ -83,10 +83,11 @@ ALIGN 16
RET RET
%endmacro %endmacro
LPC_32 sse4, 16, psrad
LPC_32 sse4, 32, psrlq
%if HAVE_XOP_EXTERNAL %if HAVE_XOP_EXTERNAL
LPC_32 xop LPC_32 xop, 32, psrlq
%endif %endif
LPC_32 sse4
;---------------------------------------------------------------------------------- ;----------------------------------------------------------------------------------
;void ff_flac_decorrelate_[lrm]s_16_sse2(uint8_t **out, int32_t **in, int channels, ;void ff_flac_decorrelate_[lrm]s_16_sse2(uint8_t **out, int32_t **in, int channels,

3
libavcodec/x86/flacdsp_init.c
View File

@ -23,6 +23,8 @@
#include "libavutil/x86/cpu.h" #include "libavutil/x86/cpu.h"
#include "config.h" #include "config.h"
void ff_flac_lpc_16_sse4(int32_t *samples, const int coeffs[32], int order,
int qlevel, int len);
void ff_flac_lpc_32_sse4(int32_t *samples, const int coeffs[32], int order, void ff_flac_lpc_32_sse4(int32_t *samples, const int coeffs[32], int order,
int qlevel, int len); int qlevel, int len);
void ff_flac_lpc_32_xop(int32_t *samples, const int coeffs[32], int order, void ff_flac_lpc_32_xop(int32_t *samples, const int coeffs[32], int order,
@ -93,6 +95,7 @@ av_cold void ff_flacdsp_init_x86(FLACDSPContext *c, enum AVSampleFormat fmt, int
} }
} }
if (EXTERNAL_SSE4(cpu_flags)) { if (EXTERNAL_SSE4(cpu_flags)) {
c->lpc16 = ff_flac_lpc_16_sse4;
c->lpc32 = ff_flac_lpc_32_sse4; c->lpc32 = ff_flac_lpc_32_sse4;
} }
if (EXTERNAL_AVX(cpu_flags)) { if (EXTERNAL_AVX(cpu_flags)) {