(gdb) bt
#0  0x000000000056edf7 in fast_memcpy ()
#1  0x0000000000498119 in ?? ()
#2  0x0000000000571c7c in ?? ()
#3  0x00007ffff3c9f9b3 in ?? () from /usr/lib/x86_64-linux-gnu/libavcodec.so.54
#4  0x00007ffff3c9fc52 in ?? () from /usr/lib/x86_64-linux-gnu/libavcodec.so.54
#5  0x00007ffff3c6cf13 in ?? () from /usr/lib/x86_64-linux-gnu/libavcodec.so.54
#6  0x00007ffff3c7013d in ?? () from /usr/lib/x86_64-linux-gnu/libavcodec.so.54
#7  0x00007ffff3c70a15 in ?? () from /usr/lib/x86_64-linux-gnu/libavcodec.so.54
#8  0x00007ffff3d46363 in avcodec_decode_video2 ()
   from /usr/lib/x86_64-linux-gnu/libavcodec.so.54
#9  0x0000000000572342 in ?? ()
#10 0x00000000004c35a1 in decode_video ()
#11 0x000000000043f5bf in ?? ()
#12 0x00000000004337d6 in main ()
(gdb) disass $pc-32,$pc+32
Dump of assembler code from 0x56edd7 to 0x56ee17:
   0x000000000056edd7 <fast_memcpy+151>:	popfq  
   0x000000000056edd8 <fast_memcpy+152>:	add    (%rax),%eax
   0x000000000056edda <fast_memcpy+154>:	add    %cl,-0x7b(%rbp)
   0x000000000056eddd <fast_memcpy+157>:	rorb   %cl,-0x77(%rbp)
   0x000000000056ede0 <fast_memcpy+160>:	rorl   -0x77(%rax)
   0x000000000056ede3 <fast_memcpy+163>:	stc    
   0x000000000056ede4 <fast_memcpy+164>:	mov    %rsi,%rax
   0x000000000056ede7 <fast_memcpy+167>:	je     0x56ee2d <fast_memcpy+237>
   0x000000000056ede9 <fast_memcpy+169>:	nopl   0x0(%rax)
   0x000000000056edf0 <fast_memcpy+176>:	prefetchnta 0x140(%rax)
=> 0x000000000056edf7 <fast_memcpy+183>:	movaps (%rax),%xmm0
   0x000000000056edfa <fast_memcpy+186>:	movaps 0x10(%rax),%xmm1
   0x000000000056edfe <fast_memcpy+190>:	movaps 0x20(%rax),%xmm2
   0x000000000056ee02 <fast_memcpy+194>:	movaps 0x30(%rax),%xmm3
   0x000000000056ee06 <fast_memcpy+198>:	movntps %xmm0,(%rcx)
   0x000000000056ee09 <fast_memcpy+201>:	movntps %xmm1,0x10(%rcx)
   0x000000000056ee0d <fast_memcpy+205>:	movntps %xmm2,0x20(%rcx)
   0x000000000056ee11 <fast_memcpy+209>:	movntps %xmm3,0x30(%rcx)
   0x000000000056ee15 <fast_memcpy+213>:	add    $0x40,%rax
End of assembler dump.
(gdb) info all-registers
rax            0x0	0
rbx            0x10	16
rcx            0x7ffff7f12000	140737353162752
rdx            0x10	16
rsi            0x0	0
rdi            0x7ffff7f12000	140737353162752
rbp            0x2d0	0x2d0
rsp            0x7fffffffc6f8	0x7fffffffc6f8
r8             0x7ffff7f12000	140737353162752
r9             0xb	11
r10            0xb	11
r11            0x7fffffffc780	140737488340864
r12            0x0	0
r13            0x0	0
r14            0x1	1
r15            0x7ffff7f122d0	140737353163472
rip            0x56edf7	0x56edf7 <fast_memcpy+183>
eflags         0x10202	[ IF RF ]
cs             0x33	51
ss             0x2b	43
ds             0x0	0
es             0x0	0
fs             0x0	0
---Type <return> to continue, or q <return> to quit---
gs             0x0	0
st0            -nan(0x8080808080808080)	(raw 0xffff8080808080808080)
st1            -nan(0x80008000800080)	(raw 0xffff0080008000800080)
st2            -nan(0x8080808080808080)	(raw 0xffff8080808080808080)
st3            -nan(0x80008000800080)	(raw 0xffff0080008000800080)
st4            -nan(0x8080808080808080)	(raw 0xffff8080808080808080)
st5            -nan(0x80008000800080)	(raw 0xffff0080008000800080)
st6            -nan(0x8080808080808080)	(raw 0xffff8080808080808080)
st7            -nan(0x80008000800080)	(raw 0xffff0080008000800080)
fctrl          0x37f	895
fstat          0x0	0
ftag           0xffff	65535
fiseg          0x0	0
fioff          0x0	0
foseg          0x0	0
fooff          0x0	0
fop            0x0	0
mxcsr          0x1fa0	[ PE IM DM ZM OM UM PM ]
ymm0           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, 
  v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 
    0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {
    0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
---Type <return> to continue, or q <return> to quit---
ymm1           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, 
  v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 
    0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {
    0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm2           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x69, 0x6d, 0x61, 0x67, 0x65, 
    0x5f, 0x74, 0x2c, 0x20, 0x37, 0x33, 0x36, 0x78, 0x35, 0x37, 0x36, 
    0x0 <repeats 16 times>}, v16_int16 = {0x6d69, 0x6761, 0x5f65, 0x2c74, 
    0x3720, 0x3633, 0x3578, 0x3637, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v8_int32 = {0x67616d69, 0x2c745f65, 0x36333720, 0x36373578, 0x0, 0x0, 0x0, 
    0x0}, v4_int64 = {0x2c745f6567616d69, 0x3637357836333720, 0x0, 0x0}, 
  v2_int128 = {0x36373578363337202c745f6567616d69, 
    0x00000000000000000000000000000000}}
ymm3           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x8000000000000000, 0x8000000000000000, 0x0, 0x0}, v32_int8 = {
    0x78, 0x31, 0x32, 0x62, 0x70, 0x70, 0x20, 0x59, 0x55, 0x56, 0x20, 0x70, 
    0x6c, 0x61, 0x6e, 0x61, 0x0 <repeats 16 times>}, v16_int16 = {0x3178, 
    0x6232, 0x7070, 0x5920, 0x5655, 0x7020, 0x616c, 0x616e, 0x0, 0x0, 0x0, 
    0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0x62323178, 0x59207070, 0x70205655, 
    0x616e616c, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x5920707062323178, 
    0x616e616c70205655, 0x0, 0x0}, v2_int128 = {
    0x616e616c702056555920707062323178, 0x00000000000000000000000000000000}}
---Type <return> to continue, or q <return> to quit---
ymm4           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x72, 0x2c, 0x20, 0x36, 0x33, 
    0x35, 0x39, 0x30, 0x34, 0x20, 0x62, 0x79, 0x74, 0x65, 0x73, 0xa, 
    0x0 <repeats 16 times>}, v16_int16 = {0x2c72, 0x3620, 0x3533, 0x3039, 
    0x2034, 0x7962, 0x6574, 0xa73, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v8_int32 = {0x36202c72, 0x30393533, 0x79622034, 0xa736574, 0x0, 0x0, 0x0, 
    0x0}, v4_int64 = {0x3039353336202c72, 0xa73657479622034, 0x0, 0x0}, 
  v2_int128 = {0x0a736574796220343039353336202c72, 
    0x00000000000000000000000000000000}}
ymm5           {v8_float = {0x80000000, 0x80000000, 0x80000000, 0x80000000, 
    0x0, 0x0, 0x0, 0x0}, v4_double = {0x8000000000000000, 0x8000000000000000, 
    0x0, 0x0}, v32_int8 = {0x5a <repeats 16 times>, 0x0 <repeats 16 times>}, 
  v16_int16 = {0x5a5a, 0x5a5a, 0x5a5a, 0x5a5a, 0x5a5a, 0x5a5a, 0x5a5a, 0x5a5a, 
    0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0x5a5a5a5a, 
    0x5a5a5a5a, 0x5a5a5a5a, 0x5a5a5a5a, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {
    0x5a5a5a5a5a5a5a5a, 0x5a5a5a5a5a5a5a5a, 0x0, 0x0}, v2_int128 = {
    0x5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a, 0x00000000000000000000000000000000}}
ymm6           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x20 <repeats 16 times>, 
    0x0 <repeats 16 times>}, v16_int16 = {0x2020, 0x2020, 0x2020, 0x2020, 
    0x2020, 0x2020, 0x2020, 0x2020, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v8_int32 = {0x20202020, 0x20202020, 0x20202020, 0x20202020, 0x0, 0x0, 0x0, 
    0x0}, v4_int64 = {0x2020202020202020, 0x2020202020202020, 0x0, 0x0}, 
---Type <return> to continue, or q <return> to quit---
  v2_int128 = {0x20202020202020202020202020202020, 
    0x00000000000000000000000000000000}}
ymm7           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x8000000000000000, 0x8000000000000000, 0x0, 0x0}, v32_int8 = {
    0xff, 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 
    0x0, 0xff, 0xff, 0x0 <repeats 16 times>}, v16_int16 = {0xff, 0x0, 0xff, 
    0xff00, 0xffff, 0xffff, 0xff, 0xffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 
    0x0}, v8_int32 = {0xff, 0xff0000ff, 0xffffffff, 0xffff00ff, 0x0, 0x0, 0x0, 
    0x0}, v4_int64 = {0xff0000ff000000ff, 0xffff00ffffffffff, 0x0, 0x0}, 
  v2_int128 = {0xffff00ffffffffffff0000ff000000ff, 
    0x00000000000000000000000000000000}}
ymm8           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x8000000000000000, 0x8000000000000000, 0x0, 0x0}, v32_int8 = {
    0x7f <repeats 16 times>, 0x0 <repeats 16 times>}, v16_int16 = {0x7f7f, 
    0x7f7f, 0x7f7f, 0x7f7f, 0x7f7f, 0x7f7f, 0x7f7f, 0x7f7f, 0x0, 0x0, 0x0, 
    0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0x7f7f7f7f, 0x7f7f7f7f, 0x7f7f7f7f, 
    0x7f7f7f7f, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x7f7f7f7f7f7f7f7f, 
    0x7f7f7f7f7f7f7f7f, 0x0, 0x0}, v2_int128 = {
    0x7f7f7f7f7f7f7f7f7f7f7f7f7f7f7f7f, 0x00000000000000000000000000000000}}
ymm9           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x8000000000000000, 0x8000000000000000, 0x0, 0x0}, v32_int8 = {
    0xff <repeats 16 times>, 0x0 <repeats 16 times>}, v16_int16 = {0xffff, 
    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0, 0x0, 0x0, 
---Type <return> to continue, or q <return> to quit---
    0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0xffffffff, 0xffffffff, 0xffffffff, 
    0xffffffff, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xffffffffffffffff, 
    0xffffffffffffffff, 0x0, 0x0}, v2_int128 = {
    0xffffffffffffffffffffffffffffffff, 0x00000000000000000000000000000000}}
ymm10          {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x8000000000000000, 0x8000000000000000, 0x0, 0x0}, v32_int8 = {
    0xff <repeats 16 times>, 0x0 <repeats 16 times>}, v16_int16 = {0xffff, 
    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0, 0x0, 0x0, 
    0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0xffffffff, 0xffffffff, 0xffffffff, 
    0xffffffff, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xffffffffffffffff, 
    0xffffffffffffffff, 0x0, 0x0}, v2_int128 = {
    0xffffffffffffffffffffffffffffffff, 0x00000000000000000000000000000000}}
ymm11          {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x8000000000000000, 0x8000000000000000, 0x0, 0x0}, v32_int8 = {
    0xff <repeats 16 times>, 0x0 <repeats 16 times>}, v16_int16 = {0xffff, 
    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0, 0x0, 0x0, 
    0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0xffffffff, 0xffffffff, 0xffffffff, 
    0xffffffff, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xffffffffffffffff, 
    0xffffffffffffffff, 0x0, 0x0}, v2_int128 = {
    0xffffffffffffffffffffffffffffffff, 0x00000000000000000000000000000000}}
ymm12          {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, 
  v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 
---Type <return> to continue, or q <return> to quit---
    0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {
    0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm13          {v8_float = {0x33ad0000, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x68, 0x9d, 0x99, 0x51, 0x66, 
    0xf7, 0x87, 0x3e, 0x0 <repeats 24 times>}, v16_int16 = {0x9d68, 0x5199, 
    0xf766, 0x3e87, 0x0 <repeats 12 times>}, v8_int32 = {0x51999d68, 
    0x3e87f766, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x3e87f76651999d68, 
    0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000003e87f76651999d68, 
    0x00000000000000000000000000000000}}
ymm14          {v8_float = {0x0, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x85, 0xce, 0x35, 0xa4, 0xc6, 
    0x97, 0xe7, 0x3f, 0x0 <repeats 24 times>}, v16_int16 = {0xce85, 0xa435, 
    0x97c6, 0x3fe7, 0x0 <repeats 12 times>}, v8_int32 = {0xa435ce85, 
    0x3fe797c6, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x3fe797c6a435ce85, 
    0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000003fe797c6a435ce85, 
    0x00000000000000000000000000000000}}
ymm15          {v8_float = {0x62000000, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, 
  v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0, 0x88, 0x39, 0x52, 0x83, 
    0xb1, 0x3b, 0xbf, 0x0 <repeats 24 times>}, v16_int16 = {0x8800, 0x5239, 
    0xb183, 0xbf3b, 0x0 <repeats 12 times>}, v8_int32 = {0x52398800, 
    0xbf3bb183, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xbf3bb18352398800, 
    0x0, 0x0, 0x0}, v2_int128 = {0x0000000000000000bf3bb18352398800, 
    0x00000000000000000000000000000000}}
(gdb)