yading@11: ;****************************************************************************** yading@11: ;* x86 optimized Format Conversion Utils yading@11: ;* Copyright (c) 2008 Loren Merritt yading@11: ;* Copyright (c) 2012 Justin Ruggles yading@11: ;* yading@11: ;* This file is part of Libav. yading@11: ;* yading@11: ;* Libav is free software; you can redistribute it and/or yading@11: ;* modify it under the terms of the GNU Lesser General Public yading@11: ;* License as published by the Free Software Foundation; either yading@11: ;* version 2.1 of the License, or (at your option) any later version. yading@11: ;* yading@11: ;* Libav is distributed in the hope that it will be useful, yading@11: ;* but WITHOUT ANY WARRANTY; without even the implied warranty of yading@11: ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@11: ;* Lesser General Public License for more details. yading@11: ;* yading@11: ;* You should have received a copy of the GNU Lesser General Public yading@11: ;* License along with Libav; if not, write to the Free Software yading@11: ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@11: ;****************************************************************************** yading@11: yading@11: %include "libavutil/x86/x86util.asm" yading@11: %include "util.asm" yading@11: yading@11: SECTION_RODATA 32 yading@11: yading@11: pf_s32_inv_scale: times 8 dd 0x30000000 yading@11: pf_s32_scale: times 8 dd 0x4f000000 yading@11: pf_s32_clip: times 8 dd 0x4effffff yading@11: pf_s16_inv_scale: times 4 dd 0x38000000 yading@11: pf_s16_scale: times 4 dd 0x47000000 yading@11: pb_shuf_unpack_even: db -1, -1, 0, 1, -1, -1, 2, 3, -1, -1, 8, 9, -1, -1, 10, 11 yading@11: pb_shuf_unpack_odd: db -1, -1, 4, 5, -1, -1, 6, 7, -1, -1, 12, 13, -1, -1, 14, 15 yading@11: pb_interleave_words: SHUFFLE_MASK_W 0, 4, 1, 5, 2, 6, 3, 7 yading@11: pb_deinterleave_words: SHUFFLE_MASK_W 0, 2, 4, 6, 1, 3, 5, 7 yading@11: pw_zero_even: times 4 dw 0x0000, 0xffff yading@11: yading@11: SECTION_TEXT yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16_to_s32(int32_t *dst, const int16_t *src, int len); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: INIT_XMM sse2 yading@11: cglobal conv_s16_to_s32, 3,3,3, dst, src, len yading@11: lea lenq, [2*lend] yading@11: lea dstq, [dstq+2*lenq] yading@11: add srcq, lenq yading@11: neg lenq yading@11: .loop: yading@11: mova m2, [srcq+lenq] yading@11: pxor m0, m0 yading@11: pxor m1, m1 yading@11: punpcklwd m0, m2 yading@11: punpckhwd m1, m2 yading@11: mova [dstq+2*lenq ], m0 yading@11: mova [dstq+2*lenq+mmsize], m1 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16_to_flt(float *dst, const int16_t *src, int len); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16_TO_FLT 0 yading@11: cglobal conv_s16_to_flt, 3,3,3, dst, src, len yading@11: lea lenq, [2*lend] yading@11: add srcq, lenq yading@11: lea dstq, [dstq + 2*lenq] yading@11: neg lenq yading@11: mova m2, [pf_s16_inv_scale] yading@11: ALIGN 16 yading@11: .loop: yading@11: mova m0, [srcq+lenq] yading@11: S16_TO_S32_SX 0, 1 yading@11: cvtdq2ps m0, m0 yading@11: cvtdq2ps m1, m1 yading@11: mulps m0, m2 yading@11: mulps m1, m2 yading@11: mova [dstq+2*lenq ], m0 yading@11: mova [dstq+2*lenq+mmsize], m1 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16_TO_FLT yading@11: INIT_XMM sse4 yading@11: CONV_S16_TO_FLT yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s32_to_s16(int16_t *dst, const int32_t *src, int len); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S32_TO_S16 0 yading@11: cglobal conv_s32_to_s16, 3,3,4, dst, src, len yading@11: lea lenq, [2*lend] yading@11: lea srcq, [srcq+2*lenq] yading@11: add dstq, lenq yading@11: neg lenq yading@11: .loop: yading@11: mova m0, [srcq+2*lenq ] yading@11: mova m1, [srcq+2*lenq+ mmsize] yading@11: mova m2, [srcq+2*lenq+2*mmsize] yading@11: mova m3, [srcq+2*lenq+3*mmsize] yading@11: psrad m0, 16 yading@11: psrad m1, 16 yading@11: psrad m2, 16 yading@11: psrad m3, 16 yading@11: packssdw m0, m1 yading@11: packssdw m2, m3 yading@11: mova [dstq+lenq ], m0 yading@11: mova [dstq+lenq+mmsize], m2 yading@11: add lenq, mmsize*2 yading@11: jl .loop yading@11: %if mmsize == 8 yading@11: emms yading@11: RET yading@11: %else yading@11: REP_RET yading@11: %endif yading@11: %endmacro yading@11: yading@11: INIT_MMX mmx yading@11: CONV_S32_TO_S16 yading@11: INIT_XMM sse2 yading@11: CONV_S32_TO_S16 yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s32_to_flt(float *dst, const int32_t *src, int len); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S32_TO_FLT 0 yading@11: cglobal conv_s32_to_flt, 3,3,3, dst, src, len yading@11: lea lenq, [4*lend] yading@11: add srcq, lenq yading@11: add dstq, lenq yading@11: neg lenq yading@11: mova m0, [pf_s32_inv_scale] yading@11: ALIGN 16 yading@11: .loop: yading@11: cvtdq2ps m1, [srcq+lenq ] yading@11: cvtdq2ps m2, [srcq+lenq+mmsize] yading@11: mulps m1, m1, m0 yading@11: mulps m2, m2, m0 yading@11: mova [dstq+lenq ], m1 yading@11: mova [dstq+lenq+mmsize], m2 yading@11: add lenq, mmsize*2 yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S32_TO_FLT yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_YMM avx yading@11: CONV_S32_TO_FLT yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_flt_to_s16(int16_t *dst, const float *src, int len); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: INIT_XMM sse2 yading@11: cglobal conv_flt_to_s16, 3,3,5, dst, src, len yading@11: lea lenq, [2*lend] yading@11: lea srcq, [srcq+2*lenq] yading@11: add dstq, lenq yading@11: neg lenq yading@11: mova m4, [pf_s16_scale] yading@11: .loop: yading@11: mova m0, [srcq+2*lenq ] yading@11: mova m1, [srcq+2*lenq+1*mmsize] yading@11: mova m2, [srcq+2*lenq+2*mmsize] yading@11: mova m3, [srcq+2*lenq+3*mmsize] yading@11: mulps m0, m4 yading@11: mulps m1, m4 yading@11: mulps m2, m4 yading@11: mulps m3, m4 yading@11: cvtps2dq m0, m0 yading@11: cvtps2dq m1, m1 yading@11: cvtps2dq m2, m2 yading@11: cvtps2dq m3, m3 yading@11: packssdw m0, m1 yading@11: packssdw m2, m3 yading@11: mova [dstq+lenq ], m0 yading@11: mova [dstq+lenq+mmsize], m2 yading@11: add lenq, mmsize*2 yading@11: jl .loop yading@11: REP_RET yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_flt_to_s32(int32_t *dst, const float *src, int len); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLT_TO_S32 0 yading@11: cglobal conv_flt_to_s32, 3,3,6, dst, src, len yading@11: lea lenq, [lend*4] yading@11: add srcq, lenq yading@11: add dstq, lenq yading@11: neg lenq yading@11: mova m4, [pf_s32_scale] yading@11: mova m5, [pf_s32_clip] yading@11: .loop: yading@11: mulps m0, m4, [srcq+lenq ] yading@11: mulps m1, m4, [srcq+lenq+1*mmsize] yading@11: mulps m2, m4, [srcq+lenq+2*mmsize] yading@11: mulps m3, m4, [srcq+lenq+3*mmsize] yading@11: minps m0, m0, m5 yading@11: minps m1, m1, m5 yading@11: minps m2, m2, m5 yading@11: minps m3, m3, m5 yading@11: cvtps2dq m0, m0 yading@11: cvtps2dq m1, m1 yading@11: cvtps2dq m2, m2 yading@11: cvtps2dq m3, m3 yading@11: mova [dstq+lenq ], m0 yading@11: mova [dstq+lenq+1*mmsize], m1 yading@11: mova [dstq+lenq+2*mmsize], m2 yading@11: mova [dstq+lenq+3*mmsize], m3 yading@11: add lenq, mmsize*4 yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_FLT_TO_S32 yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_YMM avx yading@11: CONV_FLT_TO_S32 yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16p_to_s16_2ch(int16_t *dst, int16_t *const *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16P_TO_S16_2CH 0 yading@11: cglobal conv_s16p_to_s16_2ch, 3,4,5, dst, src0, len, src1 yading@11: mov src1q, [src0q+gprsize] yading@11: mov src0q, [src0q ] yading@11: lea lenq, [2*lend] yading@11: add src0q, lenq yading@11: add src1q, lenq yading@11: lea dstq, [dstq+2*lenq] yading@11: neg lenq yading@11: .loop: yading@11: mova m0, [src0q+lenq ] yading@11: mova m1, [src1q+lenq ] yading@11: mova m2, [src0q+lenq+mmsize] yading@11: mova m3, [src1q+lenq+mmsize] yading@11: SBUTTERFLY2 wd, 0, 1, 4 yading@11: SBUTTERFLY2 wd, 2, 3, 4 yading@11: mova [dstq+2*lenq+0*mmsize], m0 yading@11: mova [dstq+2*lenq+1*mmsize], m1 yading@11: mova [dstq+2*lenq+2*mmsize], m2 yading@11: mova [dstq+2*lenq+3*mmsize], m3 yading@11: add lenq, 2*mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16P_TO_S16_2CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16P_TO_S16_2CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16p_to_s16_6ch(int16_t *dst, int16_t *const *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; NOTE: In the 6-channel functions, len could be used as an index on x86-64 yading@11: ; instead of just a counter, which would avoid incrementing the yading@11: ; pointers, but the extra complexity and amount of code is not worth yading@11: ; the small gain. On x86-32 there are not enough registers to use len yading@11: ; as an index without keeping two of the pointers on the stack and yading@11: ; loading them in each iteration. yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16P_TO_S16_6CH 0 yading@11: %if ARCH_X86_64 yading@11: cglobal conv_s16p_to_s16_6ch, 3,8,7, dst, src0, len, src1, src2, src3, src4, src5 yading@11: %else yading@11: cglobal conv_s16p_to_s16_6ch, 2,7,7, dst, src0, src1, src2, src3, src4, src5 yading@11: %define lend dword r2m yading@11: %endif yading@11: mov src1q, [src0q+1*gprsize] yading@11: mov src2q, [src0q+2*gprsize] yading@11: mov src3q, [src0q+3*gprsize] yading@11: mov src4q, [src0q+4*gprsize] yading@11: mov src5q, [src0q+5*gprsize] yading@11: mov src0q, [src0q] yading@11: sub src1q, src0q yading@11: sub src2q, src0q yading@11: sub src3q, src0q yading@11: sub src4q, src0q yading@11: sub src5q, src0q yading@11: .loop: yading@11: %if cpuflag(sse2slow) yading@11: movq m0, [src0q ] ; m0 = 0, 6, 12, 18, x, x, x, x yading@11: movq m1, [src0q+src1q] ; m1 = 1, 7, 13, 19, x, x, x, x yading@11: movq m2, [src0q+src2q] ; m2 = 2, 8, 14, 20, x, x, x, x yading@11: movq m3, [src0q+src3q] ; m3 = 3, 9, 15, 21, x, x, x, x yading@11: movq m4, [src0q+src4q] ; m4 = 4, 10, 16, 22, x, x, x, x yading@11: movq m5, [src0q+src5q] ; m5 = 5, 11, 17, 23, x, x, x, x yading@11: ; unpack words: yading@11: punpcklwd m0, m1 ; m0 = 0, 1, 6, 7, 12, 13, 18, 19 yading@11: punpcklwd m2, m3 ; m2 = 4, 5, 10, 11, 16, 17, 22, 23 yading@11: punpcklwd m4, m5 ; m4 = 2, 3, 8, 9, 14, 15, 20, 21 yading@11: ; blend dwords yading@11: shufps m1, m0, m2, q2020 ; m1 = 0, 1, 12, 13, 2, 3, 14, 15 yading@11: shufps m0, m4, q2031 ; m0 = 6, 7, 18, 19, 4, 5, 16, 17 yading@11: shufps m2, m4, q3131 ; m2 = 8, 9, 20, 21, 10, 11, 22, 23 yading@11: ; shuffle dwords yading@11: pshufd m0, m0, q1302 ; m0 = 4, 5, 6, 7, 16, 17, 18, 19 yading@11: pshufd m1, m1, q3120 ; m1 = 0, 1, 2, 3, 12, 13, 14, 15 yading@11: pshufd m2, m2, q3120 ; m2 = 8, 9, 10, 11, 20, 21, 22, 23 yading@11: movq [dstq+0*mmsize/2], m1 yading@11: movq [dstq+1*mmsize/2], m0 yading@11: movq [dstq+2*mmsize/2], m2 yading@11: movhps [dstq+3*mmsize/2], m1 yading@11: movhps [dstq+4*mmsize/2], m0 yading@11: movhps [dstq+5*mmsize/2], m2 yading@11: add src0q, mmsize/2 yading@11: add dstq, mmsize*3 yading@11: sub lend, mmsize/4 yading@11: %else yading@11: mova m0, [src0q ] ; m0 = 0, 6, 12, 18, 24, 30, 36, 42 yading@11: mova m1, [src0q+src1q] ; m1 = 1, 7, 13, 19, 25, 31, 37, 43 yading@11: mova m2, [src0q+src2q] ; m2 = 2, 8, 14, 20, 26, 32, 38, 44 yading@11: mova m3, [src0q+src3q] ; m3 = 3, 9, 15, 21, 27, 33, 39, 45 yading@11: mova m4, [src0q+src4q] ; m4 = 4, 10, 16, 22, 28, 34, 40, 46 yading@11: mova m5, [src0q+src5q] ; m5 = 5, 11, 17, 23, 29, 35, 41, 47 yading@11: ; unpack words: yading@11: SBUTTERFLY2 wd, 0, 1, 6 ; m0 = 0, 1, 6, 7, 12, 13, 18, 19 yading@11: ; m1 = 24, 25, 30, 31, 36, 37, 42, 43 yading@11: SBUTTERFLY2 wd, 2, 3, 6 ; m2 = 2, 3, 8, 9, 14, 15, 20, 21 yading@11: ; m3 = 26, 27, 32, 33, 38, 39, 44, 45 yading@11: SBUTTERFLY2 wd, 4, 5, 6 ; m4 = 4, 5, 10, 11, 16, 17, 22, 23 yading@11: ; m5 = 28, 29, 34, 35, 40, 41, 46, 47 yading@11: ; blend dwords yading@11: shufps m6, m0, m2, q2020 ; m6 = 0, 1, 12, 13, 2, 3, 14, 15 yading@11: shufps m0, m4, q2031 ; m0 = 6, 7, 18, 19, 4, 5, 16, 17 yading@11: shufps m2, m4, q3131 ; m2 = 8, 9, 20, 21, 10, 11, 22, 23 yading@11: SWAP 4,6 ; m4 = 0, 1, 12, 13, 2, 3, 14, 15 yading@11: shufps m6, m1, m3, q2020 ; m6 = 24, 25, 36, 37, 26, 27, 38, 39 yading@11: shufps m1, m5, q2031 ; m1 = 30, 31, 42, 43, 28, 29, 40, 41 yading@11: shufps m3, m5, q3131 ; m3 = 32, 33, 44, 45, 34, 35, 46, 47 yading@11: SWAP 5,6 ; m5 = 24, 25, 36, 37, 26, 27, 38, 39 yading@11: ; shuffle dwords yading@11: pshufd m0, m0, q1302 ; m0 = 4, 5, 6, 7, 16, 17, 18, 19 yading@11: pshufd m2, m2, q3120 ; m2 = 8, 9, 10, 11, 20, 21, 22, 23 yading@11: pshufd m4, m4, q3120 ; m4 = 0, 1, 2, 3, 12, 13, 14, 15 yading@11: pshufd m1, m1, q1302 ; m1 = 28, 29, 30, 31, 40, 41, 42, 43 yading@11: pshufd m3, m3, q3120 ; m3 = 32, 33, 34, 35, 44, 45, 46, 47 yading@11: pshufd m5, m5, q3120 ; m5 = 24, 25, 26, 27, 36, 37, 38, 39 yading@11: ; shuffle qwords yading@11: punpcklqdq m6, m4, m0 ; m6 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: punpckhqdq m0, m2 ; m0 = 16, 17, 18, 19, 20, 21, 22, 23 yading@11: shufps m2, m4, q3210 ; m2 = 8, 9, 10, 11, 12, 13, 14, 15 yading@11: SWAP 4,6 ; m4 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: punpcklqdq m6, m5, m1 ; m6 = 24, 25, 26, 27, 28, 29, 30, 31 yading@11: punpckhqdq m1, m3 ; m1 = 40, 41, 42, 43, 44, 45, 46, 47 yading@11: shufps m3, m5, q3210 ; m3 = 32, 33, 34, 35, 36, 37, 38, 39 yading@11: SWAP 5,6 ; m5 = 24, 25, 26, 27, 28, 29, 30, 31 yading@11: mova [dstq+0*mmsize], m4 yading@11: mova [dstq+1*mmsize], m2 yading@11: mova [dstq+2*mmsize], m0 yading@11: mova [dstq+3*mmsize], m5 yading@11: mova [dstq+4*mmsize], m3 yading@11: mova [dstq+5*mmsize], m1 yading@11: add src0q, mmsize yading@11: add dstq, mmsize*6 yading@11: sub lend, mmsize/2 yading@11: %endif yading@11: jg .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16P_TO_S16_6CH yading@11: INIT_XMM sse2slow yading@11: CONV_S16P_TO_S16_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16P_TO_S16_6CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16p_to_flt_2ch(float *dst, int16_t *const *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16P_TO_FLT_2CH 0 yading@11: cglobal conv_s16p_to_flt_2ch, 3,4,6, dst, src0, len, src1 yading@11: lea lenq, [2*lend] yading@11: mov src1q, [src0q+gprsize] yading@11: mov src0q, [src0q ] yading@11: lea dstq, [dstq+4*lenq] yading@11: add src0q, lenq yading@11: add src1q, lenq yading@11: neg lenq yading@11: mova m5, [pf_s32_inv_scale] yading@11: .loop: yading@11: mova m2, [src0q+lenq] ; m2 = 0, 2, 4, 6, 8, 10, 12, 14 yading@11: mova m4, [src1q+lenq] ; m4 = 1, 3, 5, 7, 9, 11, 13, 15 yading@11: SBUTTERFLY2 wd, 2, 4, 3 ; m2 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: ; m4 = 8, 9, 10, 11, 12, 13, 14, 15 yading@11: pxor m3, m3 yading@11: punpcklwd m0, m3, m2 ; m0 = 0, 1, 2, 3 yading@11: punpckhwd m1, m3, m2 ; m1 = 4, 5, 6, 7 yading@11: punpcklwd m2, m3, m4 ; m2 = 8, 9, 10, 11 yading@11: punpckhwd m3, m4 ; m3 = 12, 13, 14, 15 yading@11: cvtdq2ps m0, m0 yading@11: cvtdq2ps m1, m1 yading@11: cvtdq2ps m2, m2 yading@11: cvtdq2ps m3, m3 yading@11: mulps m0, m5 yading@11: mulps m1, m5 yading@11: mulps m2, m5 yading@11: mulps m3, m5 yading@11: mova [dstq+4*lenq ], m0 yading@11: mova [dstq+4*lenq+ mmsize], m1 yading@11: mova [dstq+4*lenq+2*mmsize], m2 yading@11: mova [dstq+4*lenq+3*mmsize], m3 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16P_TO_FLT_2CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16P_TO_FLT_2CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16p_to_flt_6ch(float *dst, int16_t *const *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16P_TO_FLT_6CH 0 yading@11: %if ARCH_X86_64 yading@11: cglobal conv_s16p_to_flt_6ch, 3,8,8, dst, src, len, src1, src2, src3, src4, src5 yading@11: %else yading@11: cglobal conv_s16p_to_flt_6ch, 2,7,8, dst, src, src1, src2, src3, src4, src5 yading@11: %define lend dword r2m yading@11: %endif yading@11: mov src1q, [srcq+1*gprsize] yading@11: mov src2q, [srcq+2*gprsize] yading@11: mov src3q, [srcq+3*gprsize] yading@11: mov src4q, [srcq+4*gprsize] yading@11: mov src5q, [srcq+5*gprsize] yading@11: mov srcq, [srcq] yading@11: sub src1q, srcq yading@11: sub src2q, srcq yading@11: sub src3q, srcq yading@11: sub src4q, srcq yading@11: sub src5q, srcq yading@11: mova m7, [pf_s32_inv_scale] yading@11: %if cpuflag(ssse3) yading@11: %define unpack_even m6 yading@11: mova m6, [pb_shuf_unpack_even] yading@11: %if ARCH_X86_64 yading@11: %define unpack_odd m8 yading@11: mova m8, [pb_shuf_unpack_odd] yading@11: %else yading@11: %define unpack_odd [pb_shuf_unpack_odd] yading@11: %endif yading@11: %endif yading@11: .loop: yading@11: movq m0, [srcq ] ; m0 = 0, 6, 12, 18, x, x, x, x yading@11: movq m1, [srcq+src1q] ; m1 = 1, 7, 13, 19, x, x, x, x yading@11: movq m2, [srcq+src2q] ; m2 = 2, 8, 14, 20, x, x, x, x yading@11: movq m3, [srcq+src3q] ; m3 = 3, 9, 15, 21, x, x, x, x yading@11: movq m4, [srcq+src4q] ; m4 = 4, 10, 16, 22, x, x, x, x yading@11: movq m5, [srcq+src5q] ; m5 = 5, 11, 17, 23, x, x, x, x yading@11: ; unpack words: yading@11: punpcklwd m0, m1 ; m0 = 0, 1, 6, 7, 12, 13, 18, 19 yading@11: punpcklwd m2, m3 ; m2 = 2, 3, 8, 9, 14, 15, 20, 21 yading@11: punpcklwd m4, m5 ; m4 = 4, 5, 10, 11, 16, 17, 22, 23 yading@11: ; blend dwords yading@11: shufps m1, m4, m0, q3120 ; m1 = 4, 5, 16, 17, 6, 7, 18, 19 yading@11: shufps m0, m2, q2020 ; m0 = 0, 1, 12, 13, 2, 3, 14, 15 yading@11: shufps m2, m4, q3131 ; m2 = 8, 9, 20, 21, 10, 11, 22, 23 yading@11: %if cpuflag(ssse3) yading@11: pshufb m3, m0, unpack_odd ; m3 = 12, 13, 14, 15 yading@11: pshufb m0, unpack_even ; m0 = 0, 1, 2, 3 yading@11: pshufb m4, m1, unpack_odd ; m4 = 16, 17, 18, 19 yading@11: pshufb m1, unpack_even ; m1 = 4, 5, 6, 7 yading@11: pshufb m5, m2, unpack_odd ; m5 = 20, 21, 22, 23 yading@11: pshufb m2, unpack_even ; m2 = 8, 9, 10, 11 yading@11: %else yading@11: ; shuffle dwords yading@11: pshufd m0, m0, q3120 ; m0 = 0, 1, 2, 3, 12, 13, 14, 15 yading@11: pshufd m1, m1, q3120 ; m1 = 4, 5, 6, 7, 16, 17, 18, 19 yading@11: pshufd m2, m2, q3120 ; m2 = 8, 9, 10, 11, 20, 21, 22, 23 yading@11: pxor m6, m6 ; convert s16 in m0-m2 to s32 in m0-m5 yading@11: punpcklwd m3, m6, m0 ; m3 = 0, 1, 2, 3 yading@11: punpckhwd m4, m6, m0 ; m4 = 12, 13, 14, 15 yading@11: punpcklwd m0, m6, m1 ; m0 = 4, 5, 6, 7 yading@11: punpckhwd m5, m6, m1 ; m5 = 16, 17, 18, 19 yading@11: punpcklwd m1, m6, m2 ; m1 = 8, 9, 10, 11 yading@11: punpckhwd m6, m2 ; m6 = 20, 21, 22, 23 yading@11: SWAP 6,2,1,0,3,4,5 ; swap registers 3,0,1,4,5,6 to 0,1,2,3,4,5 yading@11: %endif yading@11: cvtdq2ps m0, m0 ; convert s32 to float yading@11: cvtdq2ps m1, m1 yading@11: cvtdq2ps m2, m2 yading@11: cvtdq2ps m3, m3 yading@11: cvtdq2ps m4, m4 yading@11: cvtdq2ps m5, m5 yading@11: mulps m0, m7 ; scale float from s32 range to [-1.0,1.0] yading@11: mulps m1, m7 yading@11: mulps m2, m7 yading@11: mulps m3, m7 yading@11: mulps m4, m7 yading@11: mulps m5, m7 yading@11: mova [dstq ], m0 yading@11: mova [dstq+ mmsize], m1 yading@11: mova [dstq+2*mmsize], m2 yading@11: mova [dstq+3*mmsize], m3 yading@11: mova [dstq+4*mmsize], m4 yading@11: mova [dstq+5*mmsize], m5 yading@11: add srcq, mmsize/2 yading@11: add dstq, mmsize*6 yading@11: sub lend, mmsize/4 yading@11: jg .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16P_TO_FLT_6CH yading@11: INIT_XMM ssse3 yading@11: CONV_S16P_TO_FLT_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16P_TO_FLT_6CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_fltp_to_s16_2ch(int16_t *dst, float *const *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLTP_TO_S16_2CH 0 yading@11: cglobal conv_fltp_to_s16_2ch, 3,4,3, dst, src0, len, src1 yading@11: lea lenq, [4*lend] yading@11: mov src1q, [src0q+gprsize] yading@11: mov src0q, [src0q ] yading@11: add dstq, lenq yading@11: add src0q, lenq yading@11: add src1q, lenq yading@11: neg lenq yading@11: mova m2, [pf_s16_scale] yading@11: %if cpuflag(ssse3) yading@11: mova m3, [pb_interleave_words] yading@11: %endif yading@11: .loop: yading@11: mulps m0, m2, [src0q+lenq] ; m0 = 0, 2, 4, 6 yading@11: mulps m1, m2, [src1q+lenq] ; m1 = 1, 3, 5, 7 yading@11: cvtps2dq m0, m0 yading@11: cvtps2dq m1, m1 yading@11: %if cpuflag(ssse3) yading@11: packssdw m0, m1 ; m0 = 0, 2, 4, 6, 1, 3, 5, 7 yading@11: pshufb m0, m3 ; m0 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: %else yading@11: packssdw m0, m0 ; m0 = 0, 2, 4, 6, x, x, x, x yading@11: packssdw m1, m1 ; m1 = 1, 3, 5, 7, x, x, x, x yading@11: punpcklwd m0, m1 ; m0 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: %endif yading@11: mova [dstq+lenq], m0 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_FLTP_TO_S16_2CH yading@11: INIT_XMM ssse3 yading@11: CONV_FLTP_TO_S16_2CH yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_fltp_to_s16_6ch(int16_t *dst, float *const *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLTP_TO_S16_6CH 0 yading@11: %if ARCH_X86_64 yading@11: cglobal conv_fltp_to_s16_6ch, 3,8,7, dst, src, len, src1, src2, src3, src4, src5 yading@11: %else yading@11: cglobal conv_fltp_to_s16_6ch, 2,7,7, dst, src, src1, src2, src3, src4, src5 yading@11: %define lend dword r2m yading@11: %endif yading@11: mov src1q, [srcq+1*gprsize] yading@11: mov src2q, [srcq+2*gprsize] yading@11: mov src3q, [srcq+3*gprsize] yading@11: mov src4q, [srcq+4*gprsize] yading@11: mov src5q, [srcq+5*gprsize] yading@11: mov srcq, [srcq] yading@11: sub src1q, srcq yading@11: sub src2q, srcq yading@11: sub src3q, srcq yading@11: sub src4q, srcq yading@11: sub src5q, srcq yading@11: movaps xmm6, [pf_s16_scale] yading@11: .loop: yading@11: %if cpuflag(sse2) yading@11: mulps m0, m6, [srcq ] yading@11: mulps m1, m6, [srcq+src1q] yading@11: mulps m2, m6, [srcq+src2q] yading@11: mulps m3, m6, [srcq+src3q] yading@11: mulps m4, m6, [srcq+src4q] yading@11: mulps m5, m6, [srcq+src5q] yading@11: cvtps2dq m0, m0 yading@11: cvtps2dq m1, m1 yading@11: cvtps2dq m2, m2 yading@11: cvtps2dq m3, m3 yading@11: cvtps2dq m4, m4 yading@11: cvtps2dq m5, m5 yading@11: packssdw m0, m3 ; m0 = 0, 6, 12, 18, 3, 9, 15, 21 yading@11: packssdw m1, m4 ; m1 = 1, 7, 13, 19, 4, 10, 16, 22 yading@11: packssdw m2, m5 ; m2 = 2, 8, 14, 20, 5, 11, 17, 23 yading@11: ; unpack words: yading@11: movhlps m3, m0 ; m3 = 3, 9, 15, 21, x, x, x, x yading@11: punpcklwd m0, m1 ; m0 = 0, 1, 6, 7, 12, 13, 18, 19 yading@11: punpckhwd m1, m2 ; m1 = 4, 5, 10, 11, 16, 17, 22, 23 yading@11: punpcklwd m2, m3 ; m2 = 2, 3, 8, 9, 14, 15, 20, 21 yading@11: ; blend dwords: yading@11: shufps m3, m0, m2, q2020 ; m3 = 0, 1, 12, 13, 2, 3, 14, 15 yading@11: shufps m0, m1, q2031 ; m0 = 6, 7, 18, 19, 4, 5, 16, 17 yading@11: shufps m2, m1, q3131 ; m2 = 8, 9, 20, 21, 10, 11, 22, 23 yading@11: ; shuffle dwords: yading@11: shufps m1, m2, m3, q3120 ; m1 = 8, 9, 10, 11, 12, 13, 14, 15 yading@11: shufps m3, m0, q0220 ; m3 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: shufps m0, m2, q3113 ; m0 = 16, 17, 18, 19, 20, 21, 22, 23 yading@11: mova [dstq+0*mmsize], m3 yading@11: mova [dstq+1*mmsize], m1 yading@11: mova [dstq+2*mmsize], m0 yading@11: %else ; sse yading@11: movlps xmm0, [srcq ] yading@11: movlps xmm1, [srcq+src1q] yading@11: movlps xmm2, [srcq+src2q] yading@11: movlps xmm3, [srcq+src3q] yading@11: movlps xmm4, [srcq+src4q] yading@11: movlps xmm5, [srcq+src5q] yading@11: mulps xmm0, xmm6 yading@11: mulps xmm1, xmm6 yading@11: mulps xmm2, xmm6 yading@11: mulps xmm3, xmm6 yading@11: mulps xmm4, xmm6 yading@11: mulps xmm5, xmm6 yading@11: cvtps2pi mm0, xmm0 yading@11: cvtps2pi mm1, xmm1 yading@11: cvtps2pi mm2, xmm2 yading@11: cvtps2pi mm3, xmm3 yading@11: cvtps2pi mm4, xmm4 yading@11: cvtps2pi mm5, xmm5 yading@11: packssdw mm0, mm3 ; m0 = 0, 6, 3, 9 yading@11: packssdw mm1, mm4 ; m1 = 1, 7, 4, 10 yading@11: packssdw mm2, mm5 ; m2 = 2, 8, 5, 11 yading@11: ; unpack words yading@11: pshufw mm3, mm0, q1032 ; m3 = 3, 9, 0, 6 yading@11: punpcklwd mm0, mm1 ; m0 = 0, 1, 6, 7 yading@11: punpckhwd mm1, mm2 ; m1 = 4, 5, 10, 11 yading@11: punpcklwd mm2, mm3 ; m2 = 2, 3, 8, 9 yading@11: ; unpack dwords yading@11: pshufw mm3, mm0, q1032 ; m3 = 6, 7, 0, 1 yading@11: punpckldq mm0, mm2 ; m0 = 0, 1, 2, 3 (final) yading@11: punpckhdq mm2, mm1 ; m2 = 8, 9, 10, 11 (final) yading@11: punpckldq mm1, mm3 ; m1 = 4, 5, 6, 7 (final) yading@11: mova [dstq+0*mmsize], mm0 yading@11: mova [dstq+1*mmsize], mm1 yading@11: mova [dstq+2*mmsize], mm2 yading@11: %endif yading@11: add srcq, mmsize yading@11: add dstq, mmsize*3 yading@11: sub lend, mmsize/4 yading@11: jg .loop yading@11: %if mmsize == 8 yading@11: emms yading@11: RET yading@11: %else yading@11: REP_RET yading@11: %endif yading@11: %endmacro yading@11: yading@11: INIT_MMX sse yading@11: CONV_FLTP_TO_S16_6CH yading@11: INIT_XMM sse2 yading@11: CONV_FLTP_TO_S16_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_FLTP_TO_S16_6CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_fltp_to_flt_2ch(float *dst, float *const *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLTP_TO_FLT_2CH 0 yading@11: cglobal conv_fltp_to_flt_2ch, 3,4,5, dst, src0, len, src1 yading@11: mov src1q, [src0q+gprsize] yading@11: mov src0q, [src0q] yading@11: lea lenq, [4*lend] yading@11: add src0q, lenq yading@11: add src1q, lenq yading@11: lea dstq, [dstq+2*lenq] yading@11: neg lenq yading@11: .loop: yading@11: mova m0, [src0q+lenq ] yading@11: mova m1, [src1q+lenq ] yading@11: mova m2, [src0q+lenq+mmsize] yading@11: mova m3, [src1q+lenq+mmsize] yading@11: SBUTTERFLYPS 0, 1, 4 yading@11: SBUTTERFLYPS 2, 3, 4 yading@11: mova [dstq+2*lenq+0*mmsize], m0 yading@11: mova [dstq+2*lenq+1*mmsize], m1 yading@11: mova [dstq+2*lenq+2*mmsize], m2 yading@11: mova [dstq+2*lenq+3*mmsize], m3 yading@11: add lenq, 2*mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse yading@11: CONV_FLTP_TO_FLT_2CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_FLTP_TO_FLT_2CH yading@11: %endif yading@11: yading@11: ;----------------------------------------------------------------------------- yading@11: ; void ff_conv_fltp_to_flt_6ch(float *dst, float *const *src, int len, yading@11: ; int channels); yading@11: ;----------------------------------------------------------------------------- yading@11: yading@11: %macro CONV_FLTP_TO_FLT_6CH 0 yading@11: cglobal conv_fltp_to_flt_6ch, 2,8,7, dst, src, src1, src2, src3, src4, src5, len yading@11: %if ARCH_X86_64 yading@11: mov lend, r2d yading@11: %else yading@11: %define lend dword r2m yading@11: %endif yading@11: mov src1q, [srcq+1*gprsize] yading@11: mov src2q, [srcq+2*gprsize] yading@11: mov src3q, [srcq+3*gprsize] yading@11: mov src4q, [srcq+4*gprsize] yading@11: mov src5q, [srcq+5*gprsize] yading@11: mov srcq, [srcq] yading@11: sub src1q, srcq yading@11: sub src2q, srcq yading@11: sub src3q, srcq yading@11: sub src4q, srcq yading@11: sub src5q, srcq yading@11: .loop: yading@11: mova m0, [srcq ] yading@11: mova m1, [srcq+src1q] yading@11: mova m2, [srcq+src2q] yading@11: mova m3, [srcq+src3q] yading@11: mova m4, [srcq+src4q] yading@11: mova m5, [srcq+src5q] yading@11: %if cpuflag(sse4) yading@11: SBUTTERFLYPS 0, 1, 6 yading@11: SBUTTERFLYPS 2, 3, 6 yading@11: SBUTTERFLYPS 4, 5, 6 yading@11: yading@11: blendps m6, m4, m0, 1100b yading@11: movlhps m0, m2 yading@11: movhlps m4, m2 yading@11: blendps m2, m5, m1, 1100b yading@11: movlhps m1, m3 yading@11: movhlps m5, m3 yading@11: yading@11: movaps [dstq ], m0 yading@11: movaps [dstq+16], m6 yading@11: movaps [dstq+32], m4 yading@11: movaps [dstq+48], m1 yading@11: movaps [dstq+64], m2 yading@11: movaps [dstq+80], m5 yading@11: %else ; mmx yading@11: SBUTTERFLY dq, 0, 1, 6 yading@11: SBUTTERFLY dq, 2, 3, 6 yading@11: SBUTTERFLY dq, 4, 5, 6 yading@11: yading@11: movq [dstq ], m0 yading@11: movq [dstq+ 8], m2 yading@11: movq [dstq+16], m4 yading@11: movq [dstq+24], m1 yading@11: movq [dstq+32], m3 yading@11: movq [dstq+40], m5 yading@11: %endif yading@11: add srcq, mmsize yading@11: add dstq, mmsize*6 yading@11: sub lend, mmsize/4 yading@11: jg .loop yading@11: %if mmsize == 8 yading@11: emms yading@11: RET yading@11: %else yading@11: REP_RET yading@11: %endif yading@11: %endmacro yading@11: yading@11: INIT_MMX mmx yading@11: CONV_FLTP_TO_FLT_6CH yading@11: INIT_XMM sse4 yading@11: CONV_FLTP_TO_FLT_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_FLTP_TO_FLT_6CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16_to_s16p_2ch(int16_t *const *dst, int16_t *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16_TO_S16P_2CH 0 yading@11: cglobal conv_s16_to_s16p_2ch, 3,4,4, dst0, src, len, dst1 yading@11: lea lenq, [2*lend] yading@11: mov dst1q, [dst0q+gprsize] yading@11: mov dst0q, [dst0q ] yading@11: lea srcq, [srcq+2*lenq] yading@11: add dst0q, lenq yading@11: add dst1q, lenq yading@11: neg lenq yading@11: %if cpuflag(ssse3) yading@11: mova m3, [pb_deinterleave_words] yading@11: %endif yading@11: .loop: yading@11: mova m0, [srcq+2*lenq ] ; m0 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: mova m1, [srcq+2*lenq+mmsize] ; m1 = 8, 9, 10, 11, 12, 13, 14, 15 yading@11: %if cpuflag(ssse3) yading@11: pshufb m0, m3 ; m0 = 0, 2, 4, 6, 1, 3, 5, 7 yading@11: pshufb m1, m3 ; m1 = 8, 10, 12, 14, 9, 11, 13, 15 yading@11: SBUTTERFLY2 qdq, 0, 1, 2 ; m0 = 0, 2, 4, 6, 8, 10, 12, 14 yading@11: ; m1 = 1, 3, 5, 7, 9, 11, 13, 15 yading@11: %else ; sse2 yading@11: pshuflw m0, m0, q3120 ; m0 = 0, 2, 1, 3, 4, 5, 6, 7 yading@11: pshufhw m0, m0, q3120 ; m0 = 0, 2, 1, 3, 4, 6, 5, 7 yading@11: pshuflw m1, m1, q3120 ; m1 = 8, 10, 9, 11, 12, 13, 14, 15 yading@11: pshufhw m1, m1, q3120 ; m1 = 8, 10, 9, 11, 12, 14, 13, 15 yading@11: DEINT2_PS 0, 1, 2 ; m0 = 0, 2, 4, 6, 8, 10, 12, 14 yading@11: ; m1 = 1, 3, 5, 7, 9, 11, 13, 15 yading@11: %endif yading@11: mova [dst0q+lenq], m0 yading@11: mova [dst1q+lenq], m1 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16_TO_S16P_2CH yading@11: INIT_XMM ssse3 yading@11: CONV_S16_TO_S16P_2CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16_TO_S16P_2CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16_to_s16p_6ch(int16_t *const *dst, int16_t *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16_TO_S16P_6CH 0 yading@11: %if ARCH_X86_64 yading@11: cglobal conv_s16_to_s16p_6ch, 3,8,5, dst, src, len, dst1, dst2, dst3, dst4, dst5 yading@11: %else yading@11: cglobal conv_s16_to_s16p_6ch, 2,7,5, dst, src, dst1, dst2, dst3, dst4, dst5 yading@11: %define lend dword r2m yading@11: %endif yading@11: mov dst1q, [dstq+ gprsize] yading@11: mov dst2q, [dstq+2*gprsize] yading@11: mov dst3q, [dstq+3*gprsize] yading@11: mov dst4q, [dstq+4*gprsize] yading@11: mov dst5q, [dstq+5*gprsize] yading@11: mov dstq, [dstq ] yading@11: sub dst1q, dstq yading@11: sub dst2q, dstq yading@11: sub dst3q, dstq yading@11: sub dst4q, dstq yading@11: sub dst5q, dstq yading@11: .loop: yading@11: mova m0, [srcq+0*mmsize] ; m0 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: mova m3, [srcq+1*mmsize] ; m3 = 8, 9, 10, 11, 12, 13, 14, 15 yading@11: mova m2, [srcq+2*mmsize] ; m2 = 16, 17, 18, 19, 20, 21, 22, 23 yading@11: PALIGNR m1, m3, m0, 12, m4 ; m1 = 6, 7, 8, 9, 10, 11, x, x yading@11: shufps m3, m2, q1032 ; m3 = 12, 13, 14, 15, 16, 17, 18, 19 yading@11: psrldq m2, 4 ; m2 = 18, 19, 20, 21, 22, 23, x, x yading@11: SBUTTERFLY2 wd, 0, 1, 4 ; m0 = 0, 6, 1, 7, 2, 8, 3, 9 yading@11: ; m1 = 4, 10, 5, 11, x, x, x, x yading@11: SBUTTERFLY2 wd, 3, 2, 4 ; m3 = 12, 18, 13, 19, 14, 20, 15, 21 yading@11: ; m2 = 16, 22, 17, 23, x, x, x, x yading@11: SBUTTERFLY2 dq, 0, 3, 4 ; m0 = 0, 6, 12, 18, 1, 7, 13, 19 yading@11: ; m3 = 2, 8, 14, 20, 3, 9, 15, 21 yading@11: punpckldq m1, m2 ; m1 = 4, 10, 16, 22, 5, 11, 17, 23 yading@11: movq [dstq ], m0 yading@11: movhps [dstq+dst1q], m0 yading@11: movq [dstq+dst2q], m3 yading@11: movhps [dstq+dst3q], m3 yading@11: movq [dstq+dst4q], m1 yading@11: movhps [dstq+dst5q], m1 yading@11: add srcq, mmsize*3 yading@11: add dstq, mmsize/2 yading@11: sub lend, mmsize/4 yading@11: jg .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16_TO_S16P_6CH yading@11: INIT_XMM ssse3 yading@11: CONV_S16_TO_S16P_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16_TO_S16P_6CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16_to_fltp_2ch(float *const *dst, int16_t *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16_TO_FLTP_2CH 0 yading@11: cglobal conv_s16_to_fltp_2ch, 3,4,5, dst0, src, len, dst1 yading@11: lea lenq, [4*lend] yading@11: mov dst1q, [dst0q+gprsize] yading@11: mov dst0q, [dst0q ] yading@11: add srcq, lenq yading@11: add dst0q, lenq yading@11: add dst1q, lenq yading@11: neg lenq yading@11: mova m3, [pf_s32_inv_scale] yading@11: mova m4, [pw_zero_even] yading@11: .loop: yading@11: mova m1, [srcq+lenq] yading@11: pslld m0, m1, 16 yading@11: pand m1, m4 yading@11: cvtdq2ps m0, m0 yading@11: cvtdq2ps m1, m1 yading@11: mulps m0, m0, m3 yading@11: mulps m1, m1, m3 yading@11: mova [dst0q+lenq], m0 yading@11: mova [dst1q+lenq], m1 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16_TO_FLTP_2CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16_TO_FLTP_2CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_s16_to_fltp_6ch(float *const *dst, int16_t *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_S16_TO_FLTP_6CH 0 yading@11: %if ARCH_X86_64 yading@11: cglobal conv_s16_to_fltp_6ch, 3,8,7, dst, src, len, dst1, dst2, dst3, dst4, dst5 yading@11: %else yading@11: cglobal conv_s16_to_fltp_6ch, 2,7,7, dst, src, dst1, dst2, dst3, dst4, dst5 yading@11: %define lend dword r2m yading@11: %endif yading@11: mov dst1q, [dstq+ gprsize] yading@11: mov dst2q, [dstq+2*gprsize] yading@11: mov dst3q, [dstq+3*gprsize] yading@11: mov dst4q, [dstq+4*gprsize] yading@11: mov dst5q, [dstq+5*gprsize] yading@11: mov dstq, [dstq ] yading@11: sub dst1q, dstq yading@11: sub dst2q, dstq yading@11: sub dst3q, dstq yading@11: sub dst4q, dstq yading@11: sub dst5q, dstq yading@11: mova m6, [pf_s16_inv_scale] yading@11: .loop: yading@11: mova m0, [srcq+0*mmsize] ; m0 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: mova m3, [srcq+1*mmsize] ; m3 = 8, 9, 10, 11, 12, 13, 14, 15 yading@11: mova m2, [srcq+2*mmsize] ; m2 = 16, 17, 18, 19, 20, 21, 22, 23 yading@11: PALIGNR m1, m3, m0, 12, m4 ; m1 = 6, 7, 8, 9, 10, 11, x, x yading@11: shufps m3, m2, q1032 ; m3 = 12, 13, 14, 15, 16, 17, 18, 19 yading@11: psrldq m2, 4 ; m2 = 18, 19, 20, 21, 22, 23, x, x yading@11: SBUTTERFLY2 wd, 0, 1, 4 ; m0 = 0, 6, 1, 7, 2, 8, 3, 9 yading@11: ; m1 = 4, 10, 5, 11, x, x, x, x yading@11: SBUTTERFLY2 wd, 3, 2, 4 ; m3 = 12, 18, 13, 19, 14, 20, 15, 21 yading@11: ; m2 = 16, 22, 17, 23, x, x, x, x yading@11: SBUTTERFLY2 dq, 0, 3, 4 ; m0 = 0, 6, 12, 18, 1, 7, 13, 19 yading@11: ; m3 = 2, 8, 14, 20, 3, 9, 15, 21 yading@11: punpckldq m1, m2 ; m1 = 4, 10, 16, 22, 5, 11, 17, 23 yading@11: S16_TO_S32_SX 0, 2 ; m0 = 0, 6, 12, 18 yading@11: ; m2 = 1, 7, 13, 19 yading@11: S16_TO_S32_SX 3, 4 ; m3 = 2, 8, 14, 20 yading@11: ; m4 = 3, 9, 15, 21 yading@11: S16_TO_S32_SX 1, 5 ; m1 = 4, 10, 16, 22 yading@11: ; m5 = 5, 11, 17, 23 yading@11: SWAP 1,2,3,4 yading@11: cvtdq2ps m0, m0 yading@11: cvtdq2ps m1, m1 yading@11: cvtdq2ps m2, m2 yading@11: cvtdq2ps m3, m3 yading@11: cvtdq2ps m4, m4 yading@11: cvtdq2ps m5, m5 yading@11: mulps m0, m6 yading@11: mulps m1, m6 yading@11: mulps m2, m6 yading@11: mulps m3, m6 yading@11: mulps m4, m6 yading@11: mulps m5, m6 yading@11: mova [dstq ], m0 yading@11: mova [dstq+dst1q], m1 yading@11: mova [dstq+dst2q], m2 yading@11: mova [dstq+dst3q], m3 yading@11: mova [dstq+dst4q], m4 yading@11: mova [dstq+dst5q], m5 yading@11: add srcq, mmsize*3 yading@11: add dstq, mmsize yading@11: sub lend, mmsize/4 yading@11: jg .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_S16_TO_FLTP_6CH yading@11: INIT_XMM ssse3 yading@11: CONV_S16_TO_FLTP_6CH yading@11: INIT_XMM sse4 yading@11: CONV_S16_TO_FLTP_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_S16_TO_FLTP_6CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_flt_to_s16p_2ch(int16_t *const *dst, float *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLT_TO_S16P_2CH 0 yading@11: cglobal conv_flt_to_s16p_2ch, 3,4,6, dst0, src, len, dst1 yading@11: lea lenq, [2*lend] yading@11: mov dst1q, [dst0q+gprsize] yading@11: mov dst0q, [dst0q ] yading@11: lea srcq, [srcq+4*lenq] yading@11: add dst0q, lenq yading@11: add dst1q, lenq yading@11: neg lenq yading@11: mova m5, [pf_s16_scale] yading@11: .loop: yading@11: mova m0, [srcq+4*lenq ] yading@11: mova m1, [srcq+4*lenq+ mmsize] yading@11: mova m2, [srcq+4*lenq+2*mmsize] yading@11: mova m3, [srcq+4*lenq+3*mmsize] yading@11: DEINT2_PS 0, 1, 4 yading@11: DEINT2_PS 2, 3, 4 yading@11: mulps m0, m0, m5 yading@11: mulps m1, m1, m5 yading@11: mulps m2, m2, m5 yading@11: mulps m3, m3, m5 yading@11: cvtps2dq m0, m0 yading@11: cvtps2dq m1, m1 yading@11: cvtps2dq m2, m2 yading@11: cvtps2dq m3, m3 yading@11: packssdw m0, m2 yading@11: packssdw m1, m3 yading@11: mova [dst0q+lenq], m0 yading@11: mova [dst1q+lenq], m1 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_FLT_TO_S16P_2CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_FLT_TO_S16P_2CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_flt_to_s16p_6ch(int16_t *const *dst, float *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLT_TO_S16P_6CH 0 yading@11: %if ARCH_X86_64 yading@11: cglobal conv_flt_to_s16p_6ch, 3,8,7, dst, src, len, dst1, dst2, dst3, dst4, dst5 yading@11: %else yading@11: cglobal conv_flt_to_s16p_6ch, 2,7,7, dst, src, dst1, dst2, dst3, dst4, dst5 yading@11: %define lend dword r2m yading@11: %endif yading@11: mov dst1q, [dstq+ gprsize] yading@11: mov dst2q, [dstq+2*gprsize] yading@11: mov dst3q, [dstq+3*gprsize] yading@11: mov dst4q, [dstq+4*gprsize] yading@11: mov dst5q, [dstq+5*gprsize] yading@11: mov dstq, [dstq ] yading@11: sub dst1q, dstq yading@11: sub dst2q, dstq yading@11: sub dst3q, dstq yading@11: sub dst4q, dstq yading@11: sub dst5q, dstq yading@11: mova m6, [pf_s16_scale] yading@11: .loop: yading@11: mulps m0, m6, [srcq+0*mmsize] yading@11: mulps m3, m6, [srcq+1*mmsize] yading@11: mulps m1, m6, [srcq+2*mmsize] yading@11: mulps m4, m6, [srcq+3*mmsize] yading@11: mulps m2, m6, [srcq+4*mmsize] yading@11: mulps m5, m6, [srcq+5*mmsize] yading@11: cvtps2dq m0, m0 yading@11: cvtps2dq m1, m1 yading@11: cvtps2dq m2, m2 yading@11: cvtps2dq m3, m3 yading@11: cvtps2dq m4, m4 yading@11: cvtps2dq m5, m5 yading@11: packssdw m0, m3 ; m0 = 0, 1, 2, 3, 4, 5, 6, 7 yading@11: packssdw m1, m4 ; m1 = 8, 9, 10, 11, 12, 13, 14, 15 yading@11: packssdw m2, m5 ; m2 = 16, 17, 18, 19, 20, 21, 22, 23 yading@11: PALIGNR m3, m1, m0, 12, m4 ; m3 = 6, 7, 8, 9, 10, 11, x, x yading@11: shufps m1, m2, q1032 ; m1 = 12, 13, 14, 15, 16, 17, 18, 19 yading@11: psrldq m2, 4 ; m2 = 18, 19, 20, 21, 22, 23, x, x yading@11: SBUTTERFLY2 wd, 0, 3, 4 ; m0 = 0, 6, 1, 7, 2, 8, 3, 9 yading@11: ; m3 = 4, 10, 5, 11, x, x, x, x yading@11: SBUTTERFLY2 wd, 1, 2, 4 ; m1 = 12, 18, 13, 19, 14, 20, 15, 21 yading@11: ; m2 = 16, 22, 17, 23, x, x, x, x yading@11: SBUTTERFLY2 dq, 0, 1, 4 ; m0 = 0, 6, 12, 18, 1, 7, 13, 19 yading@11: ; m1 = 2, 8, 14, 20, 3, 9, 15, 21 yading@11: punpckldq m3, m2 ; m3 = 4, 10, 16, 22, 5, 11, 17, 23 yading@11: movq [dstq ], m0 yading@11: movhps [dstq+dst1q], m0 yading@11: movq [dstq+dst2q], m1 yading@11: movhps [dstq+dst3q], m1 yading@11: movq [dstq+dst4q], m3 yading@11: movhps [dstq+dst5q], m3 yading@11: add srcq, mmsize*6 yading@11: add dstq, mmsize/2 yading@11: sub lend, mmsize/4 yading@11: jg .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_FLT_TO_S16P_6CH yading@11: INIT_XMM ssse3 yading@11: CONV_FLT_TO_S16P_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_FLT_TO_S16P_6CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_flt_to_fltp_2ch(float *const *dst, float *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLT_TO_FLTP_2CH 0 yading@11: cglobal conv_flt_to_fltp_2ch, 3,4,3, dst0, src, len, dst1 yading@11: lea lenq, [4*lend] yading@11: mov dst1q, [dst0q+gprsize] yading@11: mov dst0q, [dst0q ] yading@11: lea srcq, [srcq+2*lenq] yading@11: add dst0q, lenq yading@11: add dst1q, lenq yading@11: neg lenq yading@11: .loop: yading@11: mova m0, [srcq+2*lenq ] yading@11: mova m1, [srcq+2*lenq+mmsize] yading@11: DEINT2_PS 0, 1, 2 yading@11: mova [dst0q+lenq], m0 yading@11: mova [dst1q+lenq], m1 yading@11: add lenq, mmsize yading@11: jl .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse yading@11: CONV_FLT_TO_FLTP_2CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_FLT_TO_FLTP_2CH yading@11: %endif yading@11: yading@11: ;------------------------------------------------------------------------------ yading@11: ; void ff_conv_flt_to_fltp_6ch(float *const *dst, float *src, int len, yading@11: ; int channels); yading@11: ;------------------------------------------------------------------------------ yading@11: yading@11: %macro CONV_FLT_TO_FLTP_6CH 0 yading@11: %if ARCH_X86_64 yading@11: cglobal conv_flt_to_fltp_6ch, 3,8,7, dst, src, len, dst1, dst2, dst3, dst4, dst5 yading@11: %else yading@11: cglobal conv_flt_to_fltp_6ch, 2,7,7, dst, src, dst1, dst2, dst3, dst4, dst5 yading@11: %define lend dword r2m yading@11: %endif yading@11: mov dst1q, [dstq+ gprsize] yading@11: mov dst2q, [dstq+2*gprsize] yading@11: mov dst3q, [dstq+3*gprsize] yading@11: mov dst4q, [dstq+4*gprsize] yading@11: mov dst5q, [dstq+5*gprsize] yading@11: mov dstq, [dstq ] yading@11: sub dst1q, dstq yading@11: sub dst2q, dstq yading@11: sub dst3q, dstq yading@11: sub dst4q, dstq yading@11: sub dst5q, dstq yading@11: .loop: yading@11: mova m0, [srcq+0*mmsize] ; m0 = 0, 1, 2, 3 yading@11: mova m1, [srcq+1*mmsize] ; m1 = 4, 5, 6, 7 yading@11: mova m2, [srcq+2*mmsize] ; m2 = 8, 9, 10, 11 yading@11: mova m3, [srcq+3*mmsize] ; m3 = 12, 13, 14, 15 yading@11: mova m4, [srcq+4*mmsize] ; m4 = 16, 17, 18, 19 yading@11: mova m5, [srcq+5*mmsize] ; m5 = 20, 21, 22, 23 yading@11: yading@11: SBUTTERFLY2 dq, 0, 3, 6 ; m0 = 0, 12, 1, 13 yading@11: ; m3 = 2, 14, 3, 15 yading@11: SBUTTERFLY2 dq, 1, 4, 6 ; m1 = 4, 16, 5, 17 yading@11: ; m4 = 6, 18, 7, 19 yading@11: SBUTTERFLY2 dq, 2, 5, 6 ; m2 = 8, 20, 9, 21 yading@11: ; m5 = 10, 22, 11, 23 yading@11: SBUTTERFLY2 dq, 0, 4, 6 ; m0 = 0, 6, 12, 18 yading@11: ; m4 = 1, 7, 13, 19 yading@11: SBUTTERFLY2 dq, 3, 2, 6 ; m3 = 2, 8, 14, 20 yading@11: ; m2 = 3, 9, 15, 21 yading@11: SBUTTERFLY2 dq, 1, 5, 6 ; m1 = 4, 10, 16, 22 yading@11: ; m5 = 5, 11, 17, 23 yading@11: mova [dstq ], m0 yading@11: mova [dstq+dst1q], m4 yading@11: mova [dstq+dst2q], m3 yading@11: mova [dstq+dst3q], m2 yading@11: mova [dstq+dst4q], m1 yading@11: mova [dstq+dst5q], m5 yading@11: add srcq, mmsize*6 yading@11: add dstq, mmsize yading@11: sub lend, mmsize/4 yading@11: jg .loop yading@11: REP_RET yading@11: %endmacro yading@11: yading@11: INIT_XMM sse2 yading@11: CONV_FLT_TO_FLTP_6CH yading@11: %if HAVE_AVX_EXTERNAL yading@11: INIT_XMM avx yading@11: CONV_FLT_TO_FLTP_6CH yading@11: %endif