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
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
|
05/23/2001
Bit file header specification
A bit file has two sections. The first section is a variable length header.
The second section is a fixed-length bitstream which is downloaded into the
FPGA. The format of the bitstream can be found in the data book for the
appropriate FPGA, but the format of the header is not anywhere that I have
found in the Xilinx literature. This document describes the structure the
bit file header for the XC4005XL and XC4005E FPGAs. This structure may or
may not be appropriate for headers of bit files for other Xilinx FPGAs.
Preamble: unknown values
The bit files generated with xmake for the XC4005XL and XC4005E FPGAs all
begin with the following sequence of 13 bytes (shown here in hexadecimal):
00 09 0F F0 0F F0 0F F0 0F F0 00 00 01
Part "a": source file
Parts a through e follow the same general format. First the part begins
with the ASCII letter of the section followed by a NULL (00h). For example,
part c begins with hexadecimal 63 00, since 63 is ASCII for "c". The next
byte is a length byte, giving the length of the string to follow plus 1.
The length is followed by an ASCII string, and that string is followed by a
NULL (00h). If there is no string at all in the section, then the length
byte is 0 and there is no NULL termination.
Part a of the header tells the filename used to generate the bit file. For
example, if the file "xc4005.ncd" was used to generate the bit file then the
header will be:
"a" 00 0B "xc4005.ncd" 00
Convert the quoted letters into their ASCII representation.
Part "b": part name
Part b follows the same format as part a. The string identifies the part
and package of the target device. In my limited tests this string has
always been "4005xlpc84" or "4005epc84".
Part "c": date
Part c follows the same format as part a. The string gives the date the bit
file was generated, for example "2001/03/12".
Part "d": time
Part d follows the same format as part a. The string gives the time the bit
file was generated, for example "20:43:04".
Part "e": unknown
Part e seems to follow the same format as part a. However, in my tests this
part has always been empty. In other words, the entire part is hexadecimal
65 00 00.
Final part: bitstream length
The final part of the header gives the length of the bitstream to follow, in
binary. This is two bytes long for the XC4005XL and XC4005E parts, but it
is probable that this length will be more than two bytes long for larger
Xilinx parts.
For example, the XC4005XL part has a 18995 byte long bitstream. The final
two bytes of the header are hexadecimal 4A33, which is 18995 when converted
to decimal.
Immediately following the final part of the header, the bitstream begins.
The file ends immediately after the bitstream.
05/24/2001
UPDATE:
I examined the bit file headers for two more Xilinx parts, a Spartan part
and 1000 gate Virtex part. From this I determined my understanding of the
sections was a little off. The section letters FOLLOW the sections. So the
unknown garbage at the beginning (which was the same for all parts) is
section a, the part name was section b, and so on. The time was section e,
and the 00 following the 65 00 was the beginning of the length. So the
length is actually three bytes long. This gives a maximum bitstream length
of FFFFFFh, or 16,777,215 bytes.
09/10/2002
UPDATE:
I got an email message from "Stephen from Australia" who set me straight on
the format of the bit file header. He pointed me to the online FPGA FAQ,
which gives an alternative view of how the header is laid out. Each section
letter is a single byte. The 00 which follows the section letter is
actually the first byte of a two-byte size. The final four bytes of the
header make up a four-byte size. After considering the difference between
the two specifications, I realized that even if the FAQ is wrong it wouldn't
hurt to parse it that way since the first byte of the length would always be
00. And it seems more aesthetically pleasing to have a four-byte bitstream
length than a three-byte bitstream length. I have altered bitfile.c to
reflect the change in my understanding of the format.
Thanks for setting me straight, Steve!
|