1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68

Hi all!


For some time I have been using tremor with fastermdct patch found on

http://www.rockbox.org/wiki/FasterMDCT. This patch gives huge speedup on

i686/x86_64 machines and smaller, but still significant, speedup on ARM.


Everything works great until I compile tremor on ARM with -fPIC flag (I link

tremor into a shared object). In such case it segfaults in macro TRANSFORM_EQUAL

in file fft-ffmpeg_arm.h at line 158:


asm volatile( "ldmia %[my_z], {%[t5],%[t6]}\n\t":[t5] "=r" (t5), [t6] "=r"

                (t6):[my_z] "r" (z));\


When I compile tremor without -fPIC or with -U_ARM_ASSEM_, there is no segfault.

Tremor without fastermdct.patch also doesn't have this problem.


I tried debugging the code for a moment, but this code (MDCT +

#defines + lots of

assmebly) is completely incomprehensible to me.


Unfortunately, I don't have gdb built for this platform (I managed to find the

problematic line with `fprintf(stderr, ...)`, which itself could change the way

it segfaults). At first sight all variables are initialized to proper values.


The ARM processor I'm using is (according to /proc/cpuinfo):


        Processor       : ARMv6-compatible processor rev 4 (v6l)

        CPU implementer : 0x41

        CPU architecture: 6TEJ

        CPU variant     : 0x0

        CPU part        : 0xb36

        CPU revision    : 4

        Hardware        : Freescale MX31/MX32 3-Stack Board



Can you help me handle this issue?


--

regards,

Slawomir Testowy