FFmpeg
mpegvideoencdsp_init.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 #include <stdint.h>
20 
21 #include "libavutil/attributes.h"
22 #include "libavutil/avassert.h"
23 #include "libavutil/common.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/x86/asm.h"
26 #include "libavutil/x86/cpu.h"
27 #include "libavcodec/avcodec.h"
28 #include "libavcodec/mpegvideoencdsp.h"
29 
30 void ff_mpv_denoise_dct_sse2(int16_t block[64], int dct_error_sum[64],
31  const uint16_t dct_offset[64]);
32 int ff_pix_sum16_sse2(const uint8_t *pix, ptrdiff_t line_size);
33 int ff_pix_sum16_xop(const uint8_t *pix, ptrdiff_t line_size);
34 int ff_pix_norm1_sse2(const uint8_t *pix, ptrdiff_t line_size);
35 void ff_add_8x8basis_ssse3(int16_t rem[64], const int16_t basis[64], int scale);
36 
37 #if HAVE_INLINE_ASM
38 #if HAVE_SSSE3_INLINE
39 #define SCALE_OFFSET -1
40 
41 #define MAX_ABS 512
42 
43 static int try_8x8basis_ssse3(const int16_t rem[64], const int16_t weight[64], const int16_t basis[64], int scale)
44 {
45  x86_reg i=0;
46 
47  av_assert2(FFABS(scale) < MAX_ABS);
48  scale *= 1 << (16 + SCALE_OFFSET - BASIS_SHIFT + RECON_SHIFT);
49 
50  __asm__ volatile(
51  "pxor %%xmm2, %%xmm2 \n\t"
52  "movd %4, %%xmm3 \n\t"
53  "punpcklwd %%xmm3, %%xmm3 \n\t"
54  "pshufd $0, %%xmm3, %%xmm3 \n\t"
55  ".p2align 4 \n\t"
56  "1: \n\t"
57  "movdqa (%1, %0), %%xmm0 \n\t"
58  "movdqa 16(%1, %0), %%xmm1 \n\t"
59  "pmulhrsw %%xmm3, %%xmm0 \n\t"
60  "pmulhrsw %%xmm3, %%xmm1 \n\t"
61  "paddw (%2, %0), %%xmm0 \n\t"
62  "paddw 16(%2, %0), %%xmm1 \n\t"
63  "psraw $6, %%xmm0 \n\t"
64  "psraw $6, %%xmm1 \n\t"
65  "pmullw (%3, %0), %%xmm0 \n\t"
66  "pmullw 16(%3, %0), %%xmm1 \n\t"
67  "pmaddwd %%xmm0, %%xmm0 \n\t"
68  "pmaddwd %%xmm1, %%xmm1 \n\t"
69  "paddd %%xmm1, %%xmm0 \n\t"
70  "psrld $4, %%xmm0 \n\t"
71  "paddd %%xmm0, %%xmm2 \n\t"
72  "add $32, %0 \n\t"
73  "cmp $128, %0 \n\t" //FIXME optimize & bench
74  " jb 1b \n\t"
75  "pshufd $0x0E, %%xmm2, %%xmm0 \n\t"
76  "paddd %%xmm0, %%xmm2 \n\t"
77  "pshufd $0x01, %%xmm2, %%xmm0 \n\t"
78  "paddd %%xmm0, %%xmm2 \n\t"
79  "psrld $2, %%xmm2 \n\t"
80  "movd %%xmm2, %0 \n\t"
81  : "+r" (i)
82  : "r"(basis), "r"(rem), "r"(weight), "g"(scale)
83  XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3")
84  );
85  return i;
86 }
87 #endif /* HAVE_SSSE3_INLINE */
88 
89 /* Draw the edges of width 'w' of an image of size width, height */
90 static void draw_edges_mmx(uint8_t *buf, ptrdiff_t wrap, int width, int height,
91  int w, int h, int sides)
92 {
93  uint8_t *ptr, *last_line;
94  int i;
95 
96  /* left and right */
97  ptr = buf;
98  if (w == 8) {
99  __asm__ volatile (
100  "1: \n\t"
101  "movd (%0), %%mm0 \n\t"
102  "punpcklbw %%mm0, %%mm0 \n\t"
103  "punpcklwd %%mm0, %%mm0 \n\t"
104  "punpckldq %%mm0, %%mm0 \n\t"
105  "movq %%mm0, -8(%0) \n\t"
106  "movq -8(%0, %2), %%mm1 \n\t"
107  "punpckhbw %%mm1, %%mm1 \n\t"
108  "punpckhwd %%mm1, %%mm1 \n\t"
109  "punpckhdq %%mm1, %%mm1 \n\t"
110  "movq %%mm1, (%0, %2) \n\t"
111  "add %1, %0 \n\t"
112  "cmp %3, %0 \n\t"
113  "jnz 1b \n\t"
114  : "+r" (ptr)
115  : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
116  "r" (ptr + wrap * height));
117  } else if (w == 16) {
118  __asm__ volatile (
119  "1: \n\t"
120  "movd (%0), %%mm0 \n\t"
121  "punpcklbw %%mm0, %%mm0 \n\t"
122  "punpcklwd %%mm0, %%mm0 \n\t"
123  "punpckldq %%mm0, %%mm0 \n\t"
124  "movq %%mm0, -8(%0) \n\t"
125  "movq %%mm0, -16(%0) \n\t"
126  "movq -8(%0, %2), %%mm1 \n\t"
127  "punpckhbw %%mm1, %%mm1 \n\t"
128  "punpckhwd %%mm1, %%mm1 \n\t"
129  "punpckhdq %%mm1, %%mm1 \n\t"
130  "movq %%mm1, (%0, %2) \n\t"
131  "movq %%mm1, 8(%0, %2) \n\t"
132  "add %1, %0 \n\t"
133  "cmp %3, %0 \n\t"
134  "jnz 1b \n\t"
135  : "+r"(ptr)
136  : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
137  );
138  } else {
139  av_assert1(w == 4);
140  __asm__ volatile (
141  "1: \n\t"
142  "movd (%0), %%mm0 \n\t"
143  "punpcklbw %%mm0, %%mm0 \n\t"
144  "punpcklwd %%mm0, %%mm0 \n\t"
145  "movd %%mm0, -4(%0) \n\t"
146  "movd -4(%0, %2), %%mm1 \n\t"
147  "punpcklbw %%mm1, %%mm1 \n\t"
148  "punpckhwd %%mm1, %%mm1 \n\t"
149  "punpckhdq %%mm1, %%mm1 \n\t"
150  "movd %%mm1, (%0, %2) \n\t"
151  "add %1, %0 \n\t"
152  "cmp %3, %0 \n\t"
153  "jnz 1b \n\t"
154  : "+r" (ptr)
155  : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
156  "r" (ptr + wrap * height));
157  }
158 
159  /* top and bottom + corners */
160  buf -= w;
161  last_line = buf + (height - 1) * wrap;
162  if (sides & EDGE_TOP)
163  for (i = 0; i < h; i++)
164  // top
165  memcpy(buf - (i + 1) * wrap, buf, width + w + w);
166  if (sides & EDGE_BOTTOM)
167  for (i = 0; i < h; i++)
168  // bottom
169  memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);
170 }
171 
172 #endif /* HAVE_INLINE_ASM */
173 
174 av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
175  AVCodecContext *avctx)
176 {
177  int cpu_flags = av_get_cpu_flags();
178 
179  if (EXTERNAL_SSE2(cpu_flags)) {
180  c->denoise_dct = ff_mpv_denoise_dct_sse2;
181  c->pix_sum = ff_pix_sum16_sse2;
182  c->pix_norm1 = ff_pix_norm1_sse2;
183  }
184 
185  if (EXTERNAL_XOP(cpu_flags)) {
186  c->pix_sum = ff_pix_sum16_xop;
187  }
188 
189 #if HAVE_INLINE_ASM
190 
191  if (INLINE_MMX(cpu_flags)) {
192  if (avctx->bits_per_raw_sample <= 8) {
193  c->draw_edges = draw_edges_mmx;
194  }
195  }
196 #endif /* HAVE_INLINE_ASM */
197 
198  if (X86_SSSE3(cpu_flags)) {
199 #if HAVE_SSSE3_INLINE
200  if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
201  c->try_8x8basis = try_8x8basis_ssse3;
202  }
203 #endif /* HAVE_SSSE3_INLINE */
204 #if HAVE_SSSE3_EXTERNAL
205  c->add_8x8basis = ff_add_8x8basis_ssse3;
206 #endif
207  }
208 
209 }
INLINE_MMX
#define INLINE_MMX(flags)
Definition: cpu.h:78
cpu.h
EDGE_BOTTOM
#define EDGE_BOTTOM
Definition: mpegvideoencdsp.h:30
x86_reg
int x86_reg
Definition: asm.h:71
basis
static int16_t basis[64][64]
Definition: mpegvideo_enc.c:4275
ff_mpegvideoencdsp_init_x86
av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c, AVCodecContext *avctx)
Definition: mpegvideoencdsp_init.c:174
w
uint8_t w
Definition: llviddspenc.c:38
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:109
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:56
pix
enum AVPixelFormat pix
Definition: ohcodec.c:55
wrap
#define wrap(func)
Definition: neontest.h:65
AVCodecContext::flags
int flags
AV_CODEC_FLAG_*.
Definition: avcodec.h:488
weight
const h264_weight_func weight
Definition: h264dsp_init.c:33
avassert.h
av_cold
#define av_cold
Definition: attributes.h:106
ff_pix_norm1_sse2
int ff_pix_norm1_sse2(const uint8_t *pix, ptrdiff_t line_size)
BASIS_SHIFT
#define BASIS_SHIFT
Definition: mpegvideoencdsp.h:26
AVCodecContext::bits_per_raw_sample
int bits_per_raw_sample
Bits per sample/pixel of internal libavcodec pixel/sample format.
Definition: avcodec.h:1553
FFABS
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:74
EXTERNAL_XOP
#define EXTERNAL_XOP(flags)
Definition: cpu.h:66
asm.h
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
height
#define height
Definition: dsp.h:89
cpu.h
RECON_SHIFT
#define RECON_SHIFT
Definition: mpegvideoencdsp.h:27
attributes.h
ff_pix_sum16_xop
int ff_pix_sum16_xop(const uint8_t *pix, ptrdiff_t line_size)
EXTERNAL_SSE2
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:52
MpegvideoEncDSPContext
Definition: mpegvideoencdsp.h:32
av_assert2
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
Definition: avassert.h:68
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:256
common.h
av_assert1
#define av_assert1(cond)
assert() equivalent, that does not lie in speed critical code.
Definition: avassert.h:57
avcodec.h
ff_pix_sum16_sse2
int ff_pix_sum16_sse2(const uint8_t *pix, ptrdiff_t line_size)
__asm__
__asm__(".macro parse_r var r\n\t" "\\var = -1\n\t" _IFC_REG(0) _IFC_REG(1) _IFC_REG(2) _IFC_REG(3) _IFC_REG(4) _IFC_REG(5) _IFC_REG(6) _IFC_REG(7) _IFC_REG(8) _IFC_REG(9) _IFC_REG(10) _IFC_REG(11) _IFC_REG(12) _IFC_REG(13) _IFC_REG(14) _IFC_REG(15) _IFC_REG(16) _IFC_REG(17) _IFC_REG(18) _IFC_REG(19) _IFC_REG(20) _IFC_REG(21) _IFC_REG(22) _IFC_REG(23) _IFC_REG(24) _IFC_REG(25) _IFC_REG(26) _IFC_REG(27) _IFC_REG(28) _IFC_REG(29) _IFC_REG(30) _IFC_REG(31) ".iflt \\var\n\t" ".error \"Unable to parse register name \\r\"\n\t" ".endif\n\t" ".endm")
AVCodecContext
main external API structure.
Definition: avcodec.h:431
EDGE_TOP
#define EDGE_TOP
Definition: mpegvideoencdsp.h:29
ff_add_8x8basis_ssse3
void ff_add_8x8basis_ssse3(int16_t rem[64], const int16_t basis[64], int scale)
XMM_CLOBBERS_ONLY
#define XMM_CLOBBERS_ONLY(...)
Definition: asm.h:98
mpegvideoencdsp.h
X86_SSSE3
#define X86_SSSE3(flags)
Definition: cpu.h:34
ff_mpv_denoise_dct_sse2
void ff_mpv_denoise_dct_sse2(int16_t block[64], int dct_error_sum[64], const uint16_t dct_offset[64])
AV_CODEC_FLAG_BITEXACT
#define AV_CODEC_FLAG_BITEXACT
Use only bitexact stuff (except (I)DCT).
Definition: avcodec.h:322
scale
static void scale(int *out, const int *in, const int w, const int h, const int shift)
Definition: intra.c:273
block
The exact code depends on how similar the blocks are and how related they are to the block
Definition: filter_design.txt:207
h
h
Definition: vp9dsp_template.c:2070
width
#define width
Definition: dsp.h:89
Generated on Sat Nov 22 2025 19:22:03 for FFmpeg by   doxygen 1.8.17