wrapper/FFmpeg
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

x86/float_dsp: remove usage of integer instructions

Browse source
This commit is contained in:
James Almer 2017-05-12 23:31:30 -03:00
parent 96cbaaa548
commit 0fbc7a2169
1 changed files with 7 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

14
libavutil/x86/float_dsp.asm
View file

@ -243,8 +243,8 @@ cglobal vector_dmul_scalar, 4,4,3, dst, src, mul, len
.loop: .loop:
mulpd m1, m0, [srcq+lenq ] mulpd m1, m0, [srcq+lenq ]
mulpd m2, m0, [srcq+lenq+mmsize] mulpd m2, m0, [srcq+lenq+mmsize]
mova [dstq+lenq ], m1 movaps [dstq+lenq ], m1
mova [dstq+lenq+mmsize], m2 movaps [dstq+lenq+mmsize], m2
sub lenq, 2*mmsize sub lenq, 2*mmsize
jge .loop jge .loop
REP_RET REP_RET
@ -363,14 +363,14 @@ VECTOR_FMUL_ADD
%macro VECTOR_FMUL_REVERSE 0 %macro VECTOR_FMUL_REVERSE 0
cglobal vector_fmul_reverse, 4,4,2, dst, src0, src1, len cglobal vector_fmul_reverse, 4,4,2, dst, src0, src1, len
%if cpuflag(avx2) %if cpuflag(avx2)
mova m2, [pd_reverse] movaps m2, [pd_reverse]
%endif %endif
lea lenq, [lend*4 - 2*mmsize] lea lenq, [lend*4 - 2*mmsize]
ALIGN 16 ALIGN 16
.loop: .loop:
%if cpuflag(avx2) %if cpuflag(avx2)
vpermd m0, m2, [src1q] vpermps m0, m2, [src1q]
vpermd m1, m2, [src1q+mmsize] vpermps m1, m2, [src1q+mmsize]
%elif cpuflag(avx) %elif cpuflag(avx)
vmovaps xmm0, [src1q + 16] vmovaps xmm0, [src1q + 16]
vinsertf128 m0, m0, [src1q], 1 vinsertf128 m0, m0, [src1q], 1
@ -386,8 +386,8 @@ ALIGN 16
%endif %endif
mulps m0, m0, [src0q + lenq + mmsize] mulps m0, m0, [src0q + lenq + mmsize]
mulps m1, m1, [src0q + lenq] mulps m1, m1, [src0q + lenq]
mova [dstq + lenq + mmsize], m0 movaps [dstq + lenq + mmsize], m0
mova [dstq + lenq], m1 movaps [dstq + lenq], m1
add src1q, 2*mmsize add src1q, 2*mmsize
sub lenq, 2*mmsize sub lenq, 2*mmsize
jge .loop jge .loop