virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00
Code Issues Releases Activity

swresample/x86: add support for exact_rational

Browse Source 
phase_shift and phase_mask is removed
generally exact_rational=on is faster than exact_rational=off

Signed-off-by: Muhammad Faiz <mfcc64@gmail.com>
This commit is contained in:
Muhammad Faiz 2016-06-16 00:17:35 +07:00
parent d0bde818ac
commit 6031e5d1af
4 changed files with 46 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
libswresample/resample.c
View File

@ -349,8 +349,6 @@ static ResampleContext *resample_init(ResampleContext *c, int out_rate, int in_r
goto error; goto error;
} }
c->phase_shift = phase_shift;
c->phase_mask = phase_count - 1;
c->phase_count = phase_count; c->phase_count = phase_count;
c->linear = linear; c->linear = linear;
c->factor = factor; c->factor = factor;

3
libswresample/resample.h
View File

@ -40,9 +40,6 @@ typedef struct ResampleContext {
int frac; int frac;
int src_incr; int src_incr;
int compensation_distance; int compensation_distance;
/* TODO remove phase_shift and phase_mask */
attribute_deprecated int phase_shift;
attribute_deprecated int phase_mask;
int phase_count; int phase_count;
int linear; int linear;
enum SwrFilterType filter_type; enum SwrFilterType filter_type;

87
libswresample/x86/resample.asm
View File

@ -41,8 +41,7 @@ struc ResampleContext
.frac: resd 1 .frac: resd 1
.src_incr: resd 1 .src_incr: resd 1
.compensation_distance: resd 1 .compensation_distance: resd 1
.phase_shift: resd 1 .phase_count: resd 1
.phase_mask: resd 1
; there's a few more here but we only care about the first few ; there's a few more here but we only care about the first few
endstruc endstruc
@ -55,11 +54,12 @@ pd_0x4000: dd 0x4000
SECTION .text SECTION .text
; FIXME remove unneeded variables (index_incr, phase_mask)
%macro RESAMPLE_FNS 3-5 ; format [float or int16], bps, log2_bps, float op suffix [s or d], 1.0 constant %macro RESAMPLE_FNS 3-5 ; format [float or int16], bps, log2_bps, float op suffix [s or d], 1.0 constant
; int resample_common_$format(ResampleContext *ctx, $format *dst, ; int resample_common_$format(ResampleContext *ctx, $format *dst,
; const $format *src, int size, int update_ctx) ; const $format *src, int size, int update_ctx)
%if ARCH_X86_64 ; unix64 and win64 %if ARCH_X86_64 ; unix64 and win64
cglobal resample_common_%1, 0, 15, 2, ctx, dst, src, phase_shift, index, frac, \ cglobal resample_common_%1, 0, 15, 2, ctx, dst, src, phase_count, index, frac, \
dst_incr_mod, size, min_filter_count_x4, \ dst_incr_mod, size, min_filter_count_x4, \
min_filter_len_x4, dst_incr_div, src_incr, \ min_filter_len_x4, dst_incr_div, src_incr, \
phase_mask, dst_end, filter_bank phase_mask, dst_end, filter_bank
@ -76,7 +76,6 @@ cglobal resample_common_%1, 0, 15, 2, ctx, dst, src, phase_shift, index, frac, \
; load as many variables in registers as possible; for the rest, store ; load as many variables in registers as possible; for the rest, store
; on stack so that we have 'ctx' available as one extra register ; on stack so that we have 'ctx' available as one extra register
mov sized, r3d mov sized, r3d
mov phase_maskd, [ctxq+ResampleContext.phase_mask]
%if UNIX64 %if UNIX64
mov update_context_stackd, r4d mov update_context_stackd, r4d
%endif %endif
@ -92,17 +91,17 @@ cglobal resample_common_%1, 0, 15, 2, ctx, dst, src, phase_shift, index, frac, \
lea dst_endq, [dstq+sizeq*%2] lea dst_endq, [dstq+sizeq*%2]
%if UNIX64 %if UNIX64
mov ecx, [ctxq+ResampleContext.phase_shift] mov ecx, [ctxq+ResampleContext.phase_count]
mov edi, [ctxq+ResampleContext.filter_alloc] mov edi, [ctxq+ResampleContext.filter_alloc]
DEFINE_ARGS filter_alloc, dst, src, phase_shift, index, frac, dst_incr_mod, \ DEFINE_ARGS filter_alloc, dst, src, phase_count, index, frac, dst_incr_mod, \
filter, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \ filter, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank src_incr, phase_mask, dst_end, filter_bank
%elif WIN64 %elif WIN64
mov R9d, [ctxq+ResampleContext.filter_alloc] mov R9d, [ctxq+ResampleContext.filter_alloc]
mov ecx, [ctxq+ResampleContext.phase_shift] mov ecx, [ctxq+ResampleContext.phase_count]
DEFINE_ARGS phase_shift, dst, src, filter_alloc, index, frac, dst_incr_mod, \ DEFINE_ARGS phase_count, dst, src, filter_alloc, index, frac, dst_incr_mod, \
filter, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \ filter, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank src_incr, phase_mask, dst_end, filter_bank
%endif %endif
@ -112,7 +111,7 @@ cglobal resample_common_%1, 0, 15, 2, ctx, dst, src, phase_shift, index, frac, \
sub srcq, min_filter_len_x4q sub srcq, min_filter_len_x4q
mov src_stackq, srcq mov src_stackq, srcq
%else ; x86-32 %else ; x86-32
cglobal resample_common_%1, 1, 7, 2, ctx, phase_shift, dst, frac, \ cglobal resample_common_%1, 1, 7, 2, ctx, phase_count, dst, frac, \
index, min_filter_length_x4, filter_bank index, min_filter_length_x4, filter_bank
; push temp variables to stack ; push temp variables to stack
@ -127,7 +126,7 @@ cglobal resample_common_%1, 1, 7, 2, ctx, phase_shift, dst, frac, \
PUSH dword [ctxq+ResampleContext.dst_incr_mod] PUSH dword [ctxq+ResampleContext.dst_incr_mod]
PUSH dword [ctxq+ResampleContext.filter_alloc] PUSH dword [ctxq+ResampleContext.filter_alloc]
PUSH r3 PUSH r3
PUSH dword [ctxq+ResampleContext.phase_mask] PUSH dword [ctxq+ResampleContext.phase_count] ; unneeded replacement for phase_mask
PUSH dword [ctxq+ResampleContext.src_incr] PUSH dword [ctxq+ResampleContext.src_incr]
mov min_filter_length_x4d, [ctxq+ResampleContext.filter_length] mov min_filter_length_x4d, [ctxq+ResampleContext.filter_length]
mov indexd, [ctxq+ResampleContext.index] mov indexd, [ctxq+ResampleContext.index]
@ -139,9 +138,9 @@ cglobal resample_common_%1, 1, 7, 2, ctx, phase_shift, dst, frac, \
sub filter_bankq, min_filter_length_x4q sub filter_bankq, min_filter_length_x4q
PUSH min_filter_length_x4q PUSH min_filter_length_x4q
PUSH filter_bankq PUSH filter_bankq
mov phase_shiftd, [ctxq+ResampleContext.phase_shift] mov phase_countd, [ctxq+ResampleContext.phase_count]
DEFINE_ARGS src, phase_shift, dst, frac, index, min_filter_count_x4, filter DEFINE_ARGS src, phase_count, dst, frac, index, min_filter_count_x4, filter
%define filter_bankq dword [rsp+0x0] %define filter_bankq dword [rsp+0x0]
%define min_filter_length_x4q dword [rsp+0x4] %define min_filter_length_x4q dword [rsp+0x4]
@ -222,28 +221,32 @@ cglobal resample_common_%1, 1, 7, 2, ctx, phase_shift, dst, frac, \
inc indexd inc indexd
%if UNIX64 %if UNIX64
DEFINE_ARGS filter_alloc, dst, src, phase_shift, index, frac, dst_incr_mod, \ DEFINE_ARGS filter_alloc, dst, src, phase_count, index, frac, dst_incr_mod, \
index_incr, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \ index_incr, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank src_incr, phase_mask, dst_end, filter_bank
%elif WIN64 %elif WIN64
DEFINE_ARGS phase_shift, dst, src, filter_alloc, index, frac, dst_incr_mod, \ DEFINE_ARGS phase_count, dst, src, filter_alloc, index, frac, dst_incr_mod, \
index_incr, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \ index_incr, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank src_incr, phase_mask, dst_end, filter_bank
%else ; x86-32 %else ; x86-32
DEFINE_ARGS src, phase_shift, dst, frac, index, index_incr DEFINE_ARGS src, phase_count, dst, frac, index, index_incr
%endif %endif
.skip: .skip:
mov index_incrd, indexd
add dstq, %2 add dstq, %2
and indexd, phase_maskd cmp indexd, phase_countd
sar index_incrd, phase_shiftb jb .index_skip
lea srcq, [srcq+index_incrq*%2] .index_while:
sub indexd, phase_countd
lea srcq, [srcq+%2]
cmp indexd, phase_countd
jnb .index_while
.index_skip:
cmp dstq, dst_endq cmp dstq, dst_endq
jne .loop jne .loop
%if ARCH_X86_64 %if ARCH_X86_64
DEFINE_ARGS ctx, dst, src, phase_shift, index, frac DEFINE_ARGS ctx, dst, src, phase_count, index, frac
%else ; x86-32 %else ; x86-32
DEFINE_ARGS src, ctx, update_context, frac, index DEFINE_ARGS src, ctx, update_context, frac, index
%endif %endif
@ -270,14 +273,14 @@ cglobal resample_common_%1, 1, 7, 2, ctx, phase_shift, dst, frac, \
; const float *src, int size, int update_ctx) ; const float *src, int size, int update_ctx)
%if ARCH_X86_64 ; unix64 and win64 %if ARCH_X86_64 ; unix64 and win64
%if UNIX64 %if UNIX64
cglobal resample_linear_%1, 0, 15, 5, ctx, dst, phase_mask, phase_shift, index, frac, \ cglobal resample_linear_%1, 0, 15, 5, ctx, dst, phase_mask, phase_count, index, frac, \
size, dst_incr_mod, min_filter_count_x4, \ size, dst_incr_mod, min_filter_count_x4, \
min_filter_len_x4, dst_incr_div, src_incr, \ min_filter_len_x4, dst_incr_div, src_incr, \
src, dst_end, filter_bank src, dst_end, filter_bank
mov srcq, r2mp mov srcq, r2mp
%else ; win64 %else ; win64
cglobal resample_linear_%1, 0, 15, 5, ctx, phase_mask, src, phase_shift, index, frac, \ cglobal resample_linear_%1, 0, 15, 5, ctx, phase_mask, src, phase_count, index, frac, \
size, dst_incr_mod, min_filter_count_x4, \ size, dst_incr_mod, min_filter_count_x4, \
min_filter_len_x4, dst_incr_div, src_incr, \ min_filter_len_x4, dst_incr_div, src_incr, \
dst, dst_end, filter_bank dst, dst_end, filter_bank
@ -298,7 +301,6 @@ cglobal resample_linear_%1, 0, 15, 5, ctx, phase_mask, src, phase_shift, index,
; load as many variables in registers as possible; for the rest, store ; load as many variables in registers as possible; for the rest, store
; on stack so that we have 'ctx' available as one extra register ; on stack so that we have 'ctx' available as one extra register
mov sized, r3d mov sized, r3d
mov phase_maskd, [ctxq+ResampleContext.phase_mask]
%if UNIX64 %if UNIX64
mov update_context_stackd, r4d mov update_context_stackd, r4d
%endif %endif
@ -308,7 +310,6 @@ cglobal resample_linear_%1, 0, 15, 5, ctx, phase_mask, src, phase_shift, index,
mov filter_bankq, [ctxq+ResampleContext.filter_bank] mov filter_bankq, [ctxq+ResampleContext.filter_bank]
mov src_incrd, [ctxq+ResampleContext.src_incr] mov src_incrd, [ctxq+ResampleContext.src_incr]
mov ctx_stackq, ctxq mov ctx_stackq, ctxq
mov phase_mask_stackd, phase_maskd
mov min_filter_len_x4d, [ctxq+ResampleContext.filter_length] mov min_filter_len_x4d, [ctxq+ResampleContext.filter_length]
%ifidn %1, int16 %ifidn %1, int16
movd m4, [pd_0x4000] movd m4, [pd_0x4000]
@ -322,17 +323,17 @@ cglobal resample_linear_%1, 0, 15, 5, ctx, phase_mask, src, phase_shift, index,
lea dst_endq, [dstq+sizeq*%2] lea dst_endq, [dstq+sizeq*%2]
%if UNIX64 %if UNIX64
mov ecx, [ctxq+ResampleContext.phase_shift] mov ecx, [ctxq+ResampleContext.phase_count]
mov edi, [ctxq+ResampleContext.filter_alloc] mov edi, [ctxq+ResampleContext.filter_alloc]
DEFINE_ARGS filter_alloc, dst, filter2, phase_shift, index, frac, filter1, \ DEFINE_ARGS filter_alloc, dst, filter2, phase_count, index, frac, filter1, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \ dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, src, dst_end, filter_bank dst_incr_div, src_incr, src, dst_end, filter_bank
%elif WIN64 %elif WIN64
mov R9d, [ctxq+ResampleContext.filter_alloc] mov R9d, [ctxq+ResampleContext.filter_alloc]
mov ecx, [ctxq+ResampleContext.phase_shift] mov ecx, [ctxq+ResampleContext.phase_count]
DEFINE_ARGS phase_shift, filter2, src, filter_alloc, index, frac, filter1, \ DEFINE_ARGS phase_count, filter2, src, filter_alloc, index, frac, filter1, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \ dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, dst, dst_end, filter_bank dst_incr_div, src_incr, dst, dst_end, filter_bank
%endif %endif
@ -361,7 +362,7 @@ cglobal resample_linear_%1, 1, 7, 5, ctx, min_filter_length_x4, filter2, \
shl r3, %3 shl r3, %3
PUSH r3 PUSH r3
mov r3, dword [ctxq+ResampleContext.src_incr] mov r3, dword [ctxq+ResampleContext.src_incr]
PUSH dword [ctxq+ResampleContext.phase_mask] PUSH dword [ctxq+ResampleContext.phase_count] ; unneeded replacement of phase_mask
PUSH r3d PUSH r3d
%ifidn %1, int16 %ifidn %1, int16
movd m4, [pd_0x4000] movd m4, [pd_0x4000]
@ -380,11 +381,11 @@ cglobal resample_linear_%1, 1, 7, 5, ctx, min_filter_length_x4, filter2, \
sub filter_bankq, min_filter_length_x4q sub filter_bankq, min_filter_length_x4q
PUSH min_filter_length_x4q PUSH min_filter_length_x4q
PUSH filter_bankq PUSH filter_bankq
PUSH dword [ctxq+ResampleContext.phase_shift] PUSH dword [ctxq+ResampleContext.phase_count]
DEFINE_ARGS filter1, min_filter_count_x4, filter2, frac, index, dst, src DEFINE_ARGS filter1, min_filter_count_x4, filter2, frac, index, dst, src
%define phase_shift_stackd dword [rsp+0x0] %define phase_count_stackd dword [rsp+0x0]
%define filter_bankq dword [rsp+0x4] %define filter_bankq dword [rsp+0x4]
%define min_filter_length_x4q dword [rsp+0x8] %define min_filter_length_x4q dword [rsp+0x8]
%define src_incrd dword [rsp+0xc] %define src_incrd dword [rsp+0xc]
@ -519,35 +520,39 @@ cglobal resample_linear_%1, 1, 7, 5, ctx, min_filter_length_x4, filter2, \
inc indexd inc indexd
%if UNIX64 %if UNIX64
DEFINE_ARGS filter_alloc, dst, filter2, phase_shift, index, frac, index_incr, \ DEFINE_ARGS filter_alloc, dst, filter2, phase_count, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \ dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, src, dst_end, filter_bank dst_incr_div, src_incr, src, dst_end, filter_bank
%elif WIN64 %elif WIN64
DEFINE_ARGS phase_shift, filter2, src, filter_alloc, index, frac, index_incr, \ DEFINE_ARGS phase_count, filter2, src, filter_alloc, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \ dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, dst, dst_end, filter_bank dst_incr_div, src_incr, dst, dst_end, filter_bank
%else ; x86-32 %else ; x86-32
DEFINE_ARGS filter1, phase_shift, index_incr, frac, index, dst, src DEFINE_ARGS filter1, phase_count, index_incr, frac, index, dst, src
%endif %endif
.skip: .skip:
%if ARCH_X86_32 %if ARCH_X86_32
mov phase_shiftd, phase_shift_stackd mov phase_countd, phase_count_stackd
%endif %endif
mov index_incrd, indexd
add dstq, %2 add dstq, %2
and indexd, phase_mask_stackd cmp indexd, phase_countd
sar index_incrd, phase_shiftb jb .index_skip
lea srcq, [srcq+index_incrq*%2] .index_while:
sub indexd, phase_countd
lea srcq, [srcq+%2]
cmp indexd, phase_countd
jnb .index_while
.index_skip:
cmp dstq, dst_endq cmp dstq, dst_endq
jne .loop jne .loop
%if UNIX64 %if UNIX64
DEFINE_ARGS ctx, dst, filter2, phase_shift, index, frac, index_incr, \ DEFINE_ARGS ctx, dst, filter2, phase_count, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \ dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, src, dst_end, filter_bank dst_incr_div, src_incr, src, dst_end, filter_bank
%elif WIN64 %elif WIN64
DEFINE_ARGS ctx, filter2, src, phase_shift, index, frac, index_incr, \ DEFINE_ARGS ctx, filter2, src, phase_count, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \ dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, dst, dst_end, filter_bank dst_incr_div, src_incr, dst, dst_end, filter_bank
%else ; x86-32 %else ; x86-32

4
libswresample/x86/resample_init.c
View File

@ -47,10 +47,6 @@ av_cold void swri_resample_dsp_x86_init(ResampleContext *c)
{ {
int av_unused mm_flags = av_get_cpu_flags(); int av_unused mm_flags = av_get_cpu_flags();
/* FIXME use phase_count on asm */
if (c->phase_count != 1 << c->phase_shift)
return;
switch(c->format){ switch(c->format){
case AV_SAMPLE_FMT_S16P: case AV_SAMPLE_FMT_S16P:
if (ARCH_X86_32 && EXTERNAL_MMXEXT(mm_flags)) { if (ARCH_X86_32 && EXTERNAL_MMXEXT(mm_flags)) {