1cabdff1aSopenharmony_ci;******************************************************************************
2cabdff1aSopenharmony_ci;* x86 optimized Format Conversion Utils
3cabdff1aSopenharmony_ci;* Copyright (c) 2008 Loren Merritt
4cabdff1aSopenharmony_ci;*
5cabdff1aSopenharmony_ci;* This file is part of FFmpeg.
6cabdff1aSopenharmony_ci;*
7cabdff1aSopenharmony_ci;* FFmpeg is free software; you can redistribute it and/or
8cabdff1aSopenharmony_ci;* modify it under the terms of the GNU Lesser General Public
9cabdff1aSopenharmony_ci;* License as published by the Free Software Foundation; either
10cabdff1aSopenharmony_ci;* version 2.1 of the License, or (at your option) any later version.
11cabdff1aSopenharmony_ci;*
12cabdff1aSopenharmony_ci;* FFmpeg is distributed in the hope that it will be useful,
13cabdff1aSopenharmony_ci;* but WITHOUT ANY WARRANTY; without even the implied warranty of
14cabdff1aSopenharmony_ci;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15cabdff1aSopenharmony_ci;* Lesser General Public License for more details.
16cabdff1aSopenharmony_ci;*
17cabdff1aSopenharmony_ci;* You should have received a copy of the GNU Lesser General Public
18cabdff1aSopenharmony_ci;* License along with FFmpeg; if not, write to the Free Software
19cabdff1aSopenharmony_ci;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20cabdff1aSopenharmony_ci;******************************************************************************
21cabdff1aSopenharmony_ci
22cabdff1aSopenharmony_ci%include "libavutil/x86/x86util.asm"
23cabdff1aSopenharmony_ci
24cabdff1aSopenharmony_ciSECTION .text
25cabdff1aSopenharmony_ci
26cabdff1aSopenharmony_ci;------------------------------------------------------------------------------
27cabdff1aSopenharmony_ci; void ff_int32_to_float_fmul_scalar(float *dst, const int32_t *src, float mul,
28cabdff1aSopenharmony_ci;                                    int len);
29cabdff1aSopenharmony_ci;------------------------------------------------------------------------------
30cabdff1aSopenharmony_ci%macro INT32_TO_FLOAT_FMUL_SCALAR 1
31cabdff1aSopenharmony_ci%if UNIX64
32cabdff1aSopenharmony_cicglobal int32_to_float_fmul_scalar, 3, 3, %1, dst, src, len
33cabdff1aSopenharmony_ci%else
34cabdff1aSopenharmony_cicglobal int32_to_float_fmul_scalar, 4, 4, %1, dst, src, mul, len
35cabdff1aSopenharmony_ci%endif
36cabdff1aSopenharmony_ci%if WIN64
37cabdff1aSopenharmony_ci    SWAP 0, 2
38cabdff1aSopenharmony_ci%elif ARCH_X86_32
39cabdff1aSopenharmony_ci    movss   m0, mulm
40cabdff1aSopenharmony_ci%endif
41cabdff1aSopenharmony_ci    SPLATD  m0
42cabdff1aSopenharmony_ci    shl     lend, 2
43cabdff1aSopenharmony_ci    add     srcq, lenq
44cabdff1aSopenharmony_ci    add     dstq, lenq
45cabdff1aSopenharmony_ci    neg     lenq
46cabdff1aSopenharmony_ci.loop:
47cabdff1aSopenharmony_ci    cvtdq2ps  m1, [srcq+lenq   ]
48cabdff1aSopenharmony_ci    cvtdq2ps  m2, [srcq+lenq+16]
49cabdff1aSopenharmony_ci    mulps     m1, m0
50cabdff1aSopenharmony_ci    mulps     m2, m0
51cabdff1aSopenharmony_ci    mova  [dstq+lenq   ], m1
52cabdff1aSopenharmony_ci    mova  [dstq+lenq+16], m2
53cabdff1aSopenharmony_ci    add     lenq, 32
54cabdff1aSopenharmony_ci    jl .loop
55cabdff1aSopenharmony_ci    RET
56cabdff1aSopenharmony_ci%endmacro
57cabdff1aSopenharmony_ci
58cabdff1aSopenharmony_ciINIT_XMM sse2
59cabdff1aSopenharmony_ciINT32_TO_FLOAT_FMUL_SCALAR 3
60cabdff1aSopenharmony_ci
61cabdff1aSopenharmony_ci;------------------------------------------------------------------------------
62cabdff1aSopenharmony_ci; void ff_int32_to_float_fmul_array8(FmtConvertContext *c, float *dst, const int32_t *src,
63cabdff1aSopenharmony_ci;                                    const float *mul, int len);
64cabdff1aSopenharmony_ci;------------------------------------------------------------------------------
65cabdff1aSopenharmony_ci%macro INT32_TO_FLOAT_FMUL_ARRAY8 0
66cabdff1aSopenharmony_cicglobal int32_to_float_fmul_array8, 5, 5, 5, c, dst, src, mul, len
67cabdff1aSopenharmony_ci    shl     lend, 2
68cabdff1aSopenharmony_ci    add     srcq, lenq
69cabdff1aSopenharmony_ci    add     dstq, lenq
70cabdff1aSopenharmony_ci    neg     lenq
71cabdff1aSopenharmony_ci.loop:
72cabdff1aSopenharmony_ci    movss     m0, [mulq]
73cabdff1aSopenharmony_ci    SPLATD    m0
74cabdff1aSopenharmony_ci    cvtdq2ps  m1, [srcq+lenq   ]
75cabdff1aSopenharmony_ci    cvtdq2ps  m2, [srcq+lenq+16]
76cabdff1aSopenharmony_ci    mulps     m1, m0
77cabdff1aSopenharmony_ci    mulps     m2, m0
78cabdff1aSopenharmony_ci    mova  [dstq+lenq   ], m1
79cabdff1aSopenharmony_ci    mova  [dstq+lenq+16], m2
80cabdff1aSopenharmony_ci    add     mulq, 4
81cabdff1aSopenharmony_ci    add     lenq, 32
82cabdff1aSopenharmony_ci    jl .loop
83cabdff1aSopenharmony_ci    RET
84cabdff1aSopenharmony_ci%endmacro
85cabdff1aSopenharmony_ci
86cabdff1aSopenharmony_ciINIT_XMM sse2
87cabdff1aSopenharmony_ciINT32_TO_FLOAT_FMUL_ARRAY8
88cabdff1aSopenharmony_ci
89