mirror of https://git.ffmpeg.org/ffmpeg.git
x86: lavr: add SSE2/AVX dither_int_to_float()
This commit is contained in:
parent
1fb8f6a44f
commit
a6a3164b13
|
@ -23,6 +23,9 @@
|
|||
|
||||
SECTION_RODATA 32
|
||||
|
||||
; 1.0f / (2.0f * INT32_MAX)
|
||||
pf_dither_scale: times 8 dd 2.32830643762e-10
|
||||
|
||||
pf_s16_scale: times 4 dd 32753.0
|
||||
|
||||
SECTION_TEXT
|
||||
|
@ -51,3 +54,64 @@ cglobal quantize, 4,4,3, dst, src, dither, len
|
|||
add lenq, mmsize
|
||||
jl .loop
|
||||
REP_RET
|
||||
|
||||
;------------------------------------------------------------------------------
|
||||
; void ff_dither_int_to_float_rectangular(float *dst, int *src, int len)
|
||||
;------------------------------------------------------------------------------
|
||||
|
||||
%macro DITHER_INT_TO_FLOAT_RECTANGULAR 0
|
||||
cglobal dither_int_to_float_rectangular, 3,3,3, dst, src, len
|
||||
lea lenq, [4*lend]
|
||||
add srcq, lenq
|
||||
add dstq, lenq
|
||||
neg lenq
|
||||
mova m0, [pf_dither_scale]
|
||||
.loop:
|
||||
cvtdq2ps m1, [srcq+lenq]
|
||||
cvtdq2ps m2, [srcq+lenq+mmsize]
|
||||
mulps m1, m1, m0
|
||||
mulps m2, m2, m0
|
||||
mova [dstq+lenq], m1
|
||||
mova [dstq+lenq+mmsize], m2
|
||||
add lenq, 2*mmsize
|
||||
jl .loop
|
||||
REP_RET
|
||||
%endmacro
|
||||
|
||||
INIT_XMM sse2
|
||||
DITHER_INT_TO_FLOAT_RECTANGULAR
|
||||
INIT_YMM avx
|
||||
DITHER_INT_TO_FLOAT_RECTANGULAR
|
||||
|
||||
;------------------------------------------------------------------------------
|
||||
; void ff_dither_int_to_float_triangular(float *dst, int *src0, int len)
|
||||
;------------------------------------------------------------------------------
|
||||
|
||||
%macro DITHER_INT_TO_FLOAT_TRIANGULAR 0
|
||||
cglobal dither_int_to_float_triangular, 3,4,5, dst, src0, len, src1
|
||||
lea lenq, [4*lend]
|
||||
lea src1q, [src0q+2*lenq]
|
||||
add src0q, lenq
|
||||
add dstq, lenq
|
||||
neg lenq
|
||||
mova m0, [pf_dither_scale]
|
||||
.loop:
|
||||
cvtdq2ps m1, [src0q+lenq]
|
||||
cvtdq2ps m2, [src0q+lenq+mmsize]
|
||||
cvtdq2ps m3, [src1q+lenq]
|
||||
cvtdq2ps m4, [src1q+lenq+mmsize]
|
||||
addps m1, m1, m3
|
||||
addps m2, m2, m4
|
||||
mulps m1, m1, m0
|
||||
mulps m2, m2, m0
|
||||
mova [dstq+lenq], m1
|
||||
mova [dstq+lenq+mmsize], m2
|
||||
add lenq, 2*mmsize
|
||||
jl .loop
|
||||
REP_RET
|
||||
%endmacro
|
||||
|
||||
INIT_XMM sse2
|
||||
DITHER_INT_TO_FLOAT_TRIANGULAR
|
||||
INIT_YMM avx
|
||||
DITHER_INT_TO_FLOAT_TRIANGULAR
|
||||
|
|
|
@ -26,6 +26,12 @@
|
|||
extern void ff_quantize_sse2(int16_t *dst, const float *src, float *dither,
|
||||
int len);
|
||||
|
||||
extern void ff_dither_int_to_float_rectangular_sse2(float *dst, int *src, int len);
|
||||
extern void ff_dither_int_to_float_rectangular_avx(float *dst, int *src, int len);
|
||||
|
||||
extern void ff_dither_int_to_float_triangular_sse2(float *dst, int *src0, int len);
|
||||
extern void ff_dither_int_to_float_triangular_avx(float *dst, int *src0, int len);
|
||||
|
||||
av_cold void ff_dither_init_x86(DitherDSPContext *ddsp,
|
||||
enum AVResampleDitherMethod method)
|
||||
{
|
||||
|
@ -36,4 +42,20 @@ av_cold void ff_dither_init_x86(DitherDSPContext *ddsp,
|
|||
ddsp->ptr_align = 16;
|
||||
ddsp->samples_align = 8;
|
||||
}
|
||||
|
||||
if (method == AV_RESAMPLE_DITHER_RECTANGULAR) {
|
||||
if (EXTERNAL_SSE2(mm_flags)) {
|
||||
ddsp->dither_int_to_float = ff_dither_int_to_float_rectangular_sse2;
|
||||
}
|
||||
if (EXTERNAL_AVX(mm_flags)) {
|
||||
ddsp->dither_int_to_float = ff_dither_int_to_float_rectangular_avx;
|
||||
}
|
||||
} else {
|
||||
if (EXTERNAL_SSE2(mm_flags)) {
|
||||
ddsp->dither_int_to_float = ff_dither_int_to_float_triangular_sse2;
|
||||
}
|
||||
if (EXTERNAL_AVX(mm_flags)) {
|
||||
ddsp->dither_int_to_float = ff_dither_int_to_float_triangular_avx;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue