FFmpeg4/libavcodec/x86/ac3dsp.asm

553 lines
15 KiB
NASM

;*****************************************************************************
;* x86-optimized AC-3 DSP functions
;* Copyright (c) 2011 Justin Ruggles
;*
;* 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
; 16777216.0f - used in ff_float_to_fixed24()
pf_1_24: times 4 dd 0x4B800000
; used in ff_ac3_compute_mantissa_size()
cextern ac3_bap_bits
pw_bap_mul1: dw 21846, 21846, 0, 32768, 21846, 21846, 0, 32768
pw_bap_mul2: dw 5, 7, 0, 7, 5, 7, 0, 7
; used in ff_ac3_extract_exponents()
cextern pd_1
pd_151: times 4 dd 151
; used in ff_apply_window_int16()
pb_revwords: SHUFFLE_MASK_W 7, 6, 5, 4, 3, 2, 1, 0
pd_16384: times 4 dd 16384
SECTION .text
;-----------------------------------------------------------------------------
; void ff_ac3_exponent_min(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
;-----------------------------------------------------------------------------
%macro AC3_EXPONENT_MIN 0
cglobal ac3_exponent_min, 3, 4, 2, exp, reuse_blks, expn, offset
shl reuse_blksq, 8
jz .end
LOOP_ALIGN
.nextexp:
mov offsetq, reuse_blksq
mova m0, [expq+offsetq]
sub offsetq, 256
LOOP_ALIGN
.nextblk:
PMINUB m0, [expq+offsetq], m1
sub offsetq, 256
jae .nextblk
mova [expq], m0
add expq, mmsize
sub expnq, mmsize
jg .nextexp
.end:
REP_RET
%endmacro
%define LOOP_ALIGN
INIT_MMX mmx
AC3_EXPONENT_MIN
%if HAVE_MMXEXT_EXTERNAL
%define LOOP_ALIGN ALIGN 16
INIT_MMX mmxext
AC3_EXPONENT_MIN
%endif
%if HAVE_SSE2_EXTERNAL
INIT_XMM sse2
AC3_EXPONENT_MIN
%endif
%undef LOOP_ALIGN
;-----------------------------------------------------------------------------
; int ff_ac3_max_msb_abs_int16(const int16_t *src, int len)
;
; This function uses 2 different methods to calculate a valid result.
; 1) logical 'or' of abs of each element
; This is used for ssse3 because of the pabsw instruction.
; It is also used for mmx because of the lack of min/max instructions.
; 2) calculate min/max for the array, then or(abs(min),abs(max))
; This is used for mmxext and sse2 because they have pminsw/pmaxsw.
;-----------------------------------------------------------------------------
; logical 'or' of 4 or 8 words in an mmx or xmm register into the low word
%macro OR_WORDS_HORIZ 2 ; src, tmp
%if cpuflag(sse2)
movhlps %2, %1
por %1, %2
pshuflw %2, %1, q0032
por %1, %2
pshuflw %2, %1, q0001
por %1, %2
%elif cpuflag(mmxext)
pshufw %2, %1, q0032
por %1, %2
pshufw %2, %1, q0001
por %1, %2
%else ; mmx
movq %2, %1
psrlq %2, 32
por %1, %2
movq %2, %1
psrlq %2, 16
por %1, %2
%endif
%endmacro
%macro AC3_MAX_MSB_ABS_INT16 1
cglobal ac3_max_msb_abs_int16, 2,2,5, src, len
pxor m2, m2
pxor m3, m3
.loop:
%ifidn %1, min_max
mova m0, [srcq]
mova m1, [srcq+mmsize]
pminsw m2, m0
pminsw m2, m1
pmaxsw m3, m0
pmaxsw m3, m1
%else ; or_abs
%if notcpuflag(ssse3)
mova m0, [srcq]
mova m1, [srcq+mmsize]
ABS2 m0, m1, m3, m4
%else ; ssse3
; using memory args is faster for ssse3
pabsw m0, [srcq]
pabsw m1, [srcq+mmsize]
%endif
por m2, m0
por m2, m1
%endif
add srcq, mmsize*2
sub lend, mmsize
ja .loop
%ifidn %1, min_max
ABS2 m2, m3, m0, m1
por m2, m3
%endif
OR_WORDS_HORIZ m2, m0
movd eax, m2
and eax, 0xFFFF
RET
%endmacro
INIT_MMX mmx
AC3_MAX_MSB_ABS_INT16 or_abs
INIT_MMX mmxext
AC3_MAX_MSB_ABS_INT16 min_max
INIT_XMM sse2
AC3_MAX_MSB_ABS_INT16 min_max
INIT_XMM ssse3
AC3_MAX_MSB_ABS_INT16 or_abs
;-----------------------------------------------------------------------------
; macro used for ff_ac3_lshift_int16() and ff_ac3_rshift_int32()
;-----------------------------------------------------------------------------
%macro AC3_SHIFT 3 ; l/r, 16/32, shift instruction, instruction set
cglobal ac3_%1shift_int%2, 3, 3, 5, src, len, shift
movd m0, shiftd
.loop:
mova m1, [srcq ]
mova m2, [srcq+mmsize ]
mova m3, [srcq+mmsize*2]
mova m4, [srcq+mmsize*3]
%3 m1, m0
%3 m2, m0
%3 m3, m0
%3 m4, m0
mova [srcq ], m1
mova [srcq+mmsize ], m2
mova [srcq+mmsize*2], m3
mova [srcq+mmsize*3], m4
add srcq, mmsize*4
sub lend, mmsize*32/%2
ja .loop
.end:
REP_RET
%endmacro
;-----------------------------------------------------------------------------
; void ff_ac3_lshift_int16(int16_t *src, unsigned int len, unsigned int shift)
;-----------------------------------------------------------------------------
INIT_MMX mmx
AC3_SHIFT l, 16, psllw
INIT_XMM sse2
AC3_SHIFT l, 16, psllw
;-----------------------------------------------------------------------------
; void ff_ac3_rshift_int32(int32_t *src, unsigned int len, unsigned int shift)
;-----------------------------------------------------------------------------
INIT_MMX mmx
AC3_SHIFT r, 32, psrad
INIT_XMM sse2
AC3_SHIFT r, 32, psrad
;-----------------------------------------------------------------------------
; void ff_float_to_fixed24(int32_t *dst, const float *src, unsigned int len)
;-----------------------------------------------------------------------------
; The 3DNow! version is not bit-identical because pf2id uses truncation rather
; than round-to-nearest.
INIT_MMX 3dnow
cglobal float_to_fixed24, 3, 3, 0, dst, src, len
movq m0, [pf_1_24]
.loop:
movq m1, [srcq ]
movq m2, [srcq+8 ]
movq m3, [srcq+16]
movq m4, [srcq+24]
pfmul m1, m0
pfmul m2, m0
pfmul m3, m0
pfmul m4, m0
pf2id m1, m1
pf2id m2, m2
pf2id m3, m3
pf2id m4, m4
movq [dstq ], m1
movq [dstq+8 ], m2
movq [dstq+16], m3
movq [dstq+24], m4
add srcq, 32
add dstq, 32
sub lend, 8
ja .loop
femms
RET
INIT_XMM sse
cglobal float_to_fixed24, 3, 3, 3, dst, src, len
movaps m0, [pf_1_24]
.loop:
movaps m1, [srcq ]
movaps m2, [srcq+16]
mulps m1, m0
mulps m2, m0
cvtps2pi mm0, m1
movhlps m1, m1
cvtps2pi mm1, m1
cvtps2pi mm2, m2
movhlps m2, m2
cvtps2pi mm3, m2
movq [dstq ], mm0
movq [dstq+ 8], mm1
movq [dstq+16], mm2
movq [dstq+24], mm3
add srcq, 32
add dstq, 32
sub lend, 8
ja .loop
emms
RET
INIT_XMM sse2
cglobal float_to_fixed24, 3, 3, 9, dst, src, len
movaps m0, [pf_1_24]
.loop:
movaps m1, [srcq ]
movaps m2, [srcq+16 ]
movaps m3, [srcq+32 ]
movaps m4, [srcq+48 ]
%ifdef m8
movaps m5, [srcq+64 ]
movaps m6, [srcq+80 ]
movaps m7, [srcq+96 ]
movaps m8, [srcq+112]
%endif
mulps m1, m0
mulps m2, m0
mulps m3, m0
mulps m4, m0
%ifdef m8
mulps m5, m0
mulps m6, m0
mulps m7, m0
mulps m8, m0
%endif
cvtps2dq m1, m1
cvtps2dq m2, m2
cvtps2dq m3, m3
cvtps2dq m4, m4
%ifdef m8
cvtps2dq m5, m5
cvtps2dq m6, m6
cvtps2dq m7, m7
cvtps2dq m8, m8
%endif
movdqa [dstq ], m1
movdqa [dstq+16 ], m2
movdqa [dstq+32 ], m3
movdqa [dstq+48 ], m4
%ifdef m8
movdqa [dstq+64 ], m5
movdqa [dstq+80 ], m6
movdqa [dstq+96 ], m7
movdqa [dstq+112], m8
add srcq, 128
add dstq, 128
sub lenq, 32
%else
add srcq, 64
add dstq, 64
sub lenq, 16
%endif
ja .loop
REP_RET
;------------------------------------------------------------------------------
; int ff_ac3_compute_mantissa_size(uint16_t mant_cnt[6][16])
;------------------------------------------------------------------------------
%macro PHADDD4 2 ; xmm src, xmm tmp
movhlps %2, %1
paddd %1, %2
pshufd %2, %1, 0x1
paddd %1, %2
%endmacro
INIT_XMM sse2
cglobal ac3_compute_mantissa_size, 1, 2, 4, mant_cnt, sum
movdqa m0, [mant_cntq ]
movdqa m1, [mant_cntq+ 1*16]
paddw m0, [mant_cntq+ 2*16]
paddw m1, [mant_cntq+ 3*16]
paddw m0, [mant_cntq+ 4*16]
paddw m1, [mant_cntq+ 5*16]
paddw m0, [mant_cntq+ 6*16]
paddw m1, [mant_cntq+ 7*16]
paddw m0, [mant_cntq+ 8*16]
paddw m1, [mant_cntq+ 9*16]
paddw m0, [mant_cntq+10*16]
paddw m1, [mant_cntq+11*16]
pmaddwd m0, [ac3_bap_bits ]
pmaddwd m1, [ac3_bap_bits+16]
paddd m0, m1
PHADDD4 m0, m1
movd sumd, m0
movdqa m3, [pw_bap_mul1]
movhpd m0, [mant_cntq +2]
movlpd m0, [mant_cntq+1*32+2]
movhpd m1, [mant_cntq+2*32+2]
movlpd m1, [mant_cntq+3*32+2]
movhpd m2, [mant_cntq+4*32+2]
movlpd m2, [mant_cntq+5*32+2]
pmulhuw m0, m3
pmulhuw m1, m3
pmulhuw m2, m3
paddusw m0, m1
paddusw m0, m2
pmaddwd m0, [pw_bap_mul2]
PHADDD4 m0, m1
movd eax, m0
add eax, sumd
RET
;------------------------------------------------------------------------------
; void ff_ac3_extract_exponents(uint8_t *exp, int32_t *coef, int nb_coefs)
;------------------------------------------------------------------------------
%macro PABSD 1-2 ; src/dst, unused
%if cpuflag(ssse3)
pabsd %1, %1
%else ; src/dst, tmp
pxor %2, %2
pcmpgtd %2, %1
pxor %1, %2
psubd %1, %2
%endif
%endmacro
%macro AC3_EXTRACT_EXPONENTS 0
cglobal ac3_extract_exponents, 3, 3, 4, exp, coef, len
add expq, lenq
lea coefq, [coefq+4*lenq]
neg lenq
mova m2, [pd_1]
mova m3, [pd_151]
.loop:
; move 4 32-bit coefs to xmm0
mova m0, [coefq+4*lenq]
; absolute value
PABSD m0, m1
; convert to float and extract exponents
pslld m0, 1
por m0, m2
cvtdq2ps m1, m0
psrld m1, 23
mova m0, m3
psubd m0, m1
; move the lowest byte in each of 4 dwords to the low dword
; NOTE: We cannot just extract the low bytes with pshufb because the dword
; result for 16777215 is -1 due to float inaccuracy. Using packuswb
; clips this to 0, which is the correct exponent.
packssdw m0, m0
packuswb m0, m0
movd [expq+lenq], m0
add lenq, 4
jl .loop
REP_RET
%endmacro
%if HAVE_SSE2_EXTERNAL
INIT_XMM sse2
AC3_EXTRACT_EXPONENTS
%endif
%if HAVE_SSSE3_EXTERNAL
INIT_XMM ssse3
AC3_EXTRACT_EXPONENTS
%endif
;-----------------------------------------------------------------------------
; void ff_apply_window_int16(int16_t *output, const int16_t *input,
; const int16_t *window, unsigned int len)
;-----------------------------------------------------------------------------
%macro REVERSE_WORDS 1-2
%if cpuflag(ssse3) && notcpuflag(atom)
pshufb %1, %2
%elif cpuflag(sse2)
pshuflw %1, %1, 0x1B
pshufhw %1, %1, 0x1B
pshufd %1, %1, 0x4E
%elif cpuflag(mmxext)
pshufw %1, %1, 0x1B
%endif
%endmacro
%macro MUL16FIXED 3
%if cpuflag(ssse3) ; dst, src, unused
; dst = ((dst * src) + (1<<14)) >> 15
pmulhrsw %1, %2
%elif cpuflag(mmxext) ; dst, src, temp
; dst = (dst * src) >> 15
; pmulhw cuts off the bottom bit, so we have to lshift by 1 and add it back
; in from the pmullw result.
mova %3, %1
pmulhw %1, %2
pmullw %3, %2
psrlw %3, 15
psllw %1, 1
por %1, %3
%endif
%endmacro
%macro APPLY_WINDOW_INT16 1 ; %1 bitexact version
%if %1
cglobal apply_window_int16, 4,5,6, output, input, window, offset, offset2
%else
cglobal apply_window_int16_round, 4,5,6, output, input, window, offset, offset2
%endif
lea offset2q, [offsetq-mmsize]
%if cpuflag(ssse3) && notcpuflag(atom)
mova m5, [pb_revwords]
ALIGN 16
%elif %1
mova m5, [pd_16384]
%endif
.loop:
%if cpuflag(ssse3)
; This version does the 16x16->16 multiplication in-place without expanding
; to 32-bit. The ssse3 version is bit-identical.
mova m0, [windowq+offset2q]
mova m1, [ inputq+offset2q]
pmulhrsw m1, m0
REVERSE_WORDS m0, m5
pmulhrsw m0, [ inputq+offsetq ]
mova [outputq+offset2q], m1
mova [outputq+offsetq ], m0
%elif %1
; This version expands 16-bit to 32-bit, multiplies by the window,
; adds 16384 for rounding, right shifts 15, then repacks back to words to
; save to the output. The window is reversed for the second half.
mova m3, [windowq+offset2q]
mova m4, [ inputq+offset2q]
pxor m0, m0
punpcklwd m0, m3
punpcklwd m1, m4
pmaddwd m0, m1
paddd m0, m5
psrad m0, 15
pxor m2, m2
punpckhwd m2, m3
punpckhwd m1, m4
pmaddwd m2, m1
paddd m2, m5
psrad m2, 15
packssdw m0, m2
mova [outputq+offset2q], m0
REVERSE_WORDS m3
mova m4, [ inputq+offsetq]
pxor m0, m0
punpcklwd m0, m3
punpcklwd m1, m4
pmaddwd m0, m1
paddd m0, m5
psrad m0, 15
pxor m2, m2
punpckhwd m2, m3
punpckhwd m1, m4
pmaddwd m2, m1
paddd m2, m5
psrad m2, 15
packssdw m0, m2
mova [outputq+offsetq], m0
%else
; This version does the 16x16->16 multiplication in-place without expanding
; to 32-bit. The mmxext and sse2 versions do not use rounding, and
; therefore are not bit-identical to the C version.
mova m0, [windowq+offset2q]
mova m1, [ inputq+offset2q]
mova m2, [ inputq+offsetq ]
MUL16FIXED m1, m0, m3
REVERSE_WORDS m0
MUL16FIXED m2, m0, m3
mova [outputq+offset2q], m1
mova [outputq+offsetq ], m2
%endif
add offsetd, mmsize
sub offset2d, mmsize
jae .loop
REP_RET
%endmacro
INIT_MMX mmxext
APPLY_WINDOW_INT16 0
INIT_XMM sse2
APPLY_WINDOW_INT16 0
INIT_MMX mmxext
APPLY_WINDOW_INT16 1
INIT_XMM sse2
APPLY_WINDOW_INT16 1
INIT_XMM ssse3
APPLY_WINDOW_INT16 1
INIT_XMM ssse3, atom
APPLY_WINDOW_INT16 1