Definining a C macro that expands to a variable number of elements - c

I'm writing USB report descriptors, which are a sequence of bytes: a tag byte (in which the lower bits tell how many data bytes follow) followed by 0, 1, 2 or 4 data bytes. e.g. to define the logical ranges of an input:
uint8_t report_descriptor[] = {
...
0x15, 0x00, // Logical Minimum (0)
0x26, 0xFF, 0x03, // Logical Maximum (1023)
...
};
Since 0 fits into one byte, we use tag type 0x15 (Logical Minimum with one data byte). But 1023 requires two bytes, so tag type 0x26 (Logical Maximum with two data bytes).
I had hoped to define some macros to make this more readable (and avoid having to comment every line):
uint8_t report_descriptor[] = {
...
LOGICAL_MINIMUM(0),
LOGICAL_MAXIMUM(1023),
...
};
However, I've hit a snag: that macro needs to expand to a different number of elements depending on the value. I don't see any easy way to achieve this. I've tried tricks like value > 255 ? (value & 0xFF, value >> 8) : value, but it always gets expanded to just one byte.
I think the spec allows to just always use the 4-byte tags, but that would be wasteful, so I'd rather not do that.
Is what I'm after possible with the preprocessor?

There is a dirty hack that will achieve the asked functionality. But being a dirty hack, it's unlikely to improve the readability. But it works. First lets define an include file helper.h like this:
#if PARAM > 255
0x26, (PARAM & 0xFF), (PARAM >> 8),
#else
0x15, (PARAM),
#endif
Then in our main we will do:
uint8_t report_descriptor[] = {
#define PARAM 0
#include "helper.h"
#undef PARAM
#define PARAM 1023
#include "helper.h"
#undef PARAM
};
To see it is working here is the test code:
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
uint8_t report_descriptor[] = {
#define PARAM 0
#include "helper.h"
#undef PARAM
#define PARAM 1023
#include "helper.h"
#undef PARAM
};
int main(int argc, char** args) {
int i;
for (i=0; i < sizeof(report_descriptor); i++ )
printf("%x\n", report_descriptor[i]);
return 0;
}
and the output is:
15
0
26
ff
3

I don't think that the C preprocessor is powerful enough to do this in a clean way. If you are willing to resort to the M4 macro processor, it can be done fairly easily. M4 should be readily available on the vast majority of GNU/Linux systems and portable implementations should be available for most platforms.
Let's define the M4 macros in a separate file and name it macros.m4.
define(`EXTRACT_BYTE', `(($1 >> (8 * $2)) & 0xFF)')
dnl You probably don't want to define these as M4 macros but as C preprocessor
dnl macros in your header files.
define(`TAG_1_BYTES', `0x15')
define(`TAG_2_BYTES', `0x26')
define(`TAG_3_BYTES', `0x37')
define(`TAG_4_BYTES', `0x48')
define(`EXPAND_1_BYTES', `TAG_1_BYTES, EXTRACT_BYTE($1, 0)')
define(`EXPAND_2_BYTES', `TAG_2_BYTES, EXTRACT_BYTE($1, 1), EXTRACT_BYTE($1, 0)')
define(`EXPAND_3_BYTES', `TAG_3_BYTES, EXTRACT_BYTE($1, 2), EXTRACT_BYTE($1, 1), EXTRACT_BYTE($1, 0)')
define(`EXPAND_4_BYTES', `TAG_4_BYTES, EXTRACT_BYTE($1, 3), EXTRACT_BYTE($1, 2), EXTRACT_BYTE($1, 1), EXTRACT_BYTE($1, 0)')
define(`ENCODE',
`ifelse(eval($1 < 256), `1', `EXPAND_1_BYTES($1)',
`ifelse(eval($1 < 65536), `1', `EXPAND_2_BYTES($1)',
`ifelse(eval($1 < 16777216), `1', `EXPAND_3_BYTES($1)',
`EXPAND_4_BYTES($1)')')')')
Now, writing your C files is straight forward. Put the following code in a file test.c.m4:
include(`macros.m4')
`static unint8_t report_descriptor[] = {'
ENCODE(50),
ENCODE(5000),
ENCODE(500000),
ENCODE(50000000),
`};'
In your Makefile, add the following rule
test.c: test.c.m4 macros.m4
${M4} $< > $#
where M4 is set to the M4 processor (usually m4).
If M4 is run on test.c.m4, it will – omitting some excess white space – produce the following test.c file:
static unint8_t report_descriptor[] = {
0x15, ((50 >> (8 * 0)) & 0xFF),
0x26, ((5000 >> (8 * 1)) & 0xFF), ((5000 >> (8 * 0)) & 0xFF),
0x37, ((500000 >> (8 * 2)) & 0xFF), ((500000 >> (8 * 1)) & 0xFF), ((500000 >> (8 * 0)) & 0xFF),
0x48, ((50000000 >> (8 * 3)) & 0xFF), ((50000000 >> (8 * 2)) & 0xFF), ((50000000 >> (8 * 1)) & 0xFF), ((50000000 >> (8 * 0)) & 0xFF),
};
You'll probably find it more convenient to keep the test.c.m4 file as minimal as possible and #include it in an ordinary C file.
If you don't know M4, you can learn the basics rather quickly. If already using GNU Autoconf, you might find it convenient to use their M4sugar M4 macro library instead of the plain M4 I've used above.

Related

Purpose of double underscore pointer operator for C functions

I am writing some C code that is for a microcontroller and have come across a curious couple of statements in some generated drivers for a peripheral I am using. Seemingly, a function uint8_t gapm_reset_req_handler (void) is supposed to reset a handler and return a status. The function is seemingly failing in its purpose, which surprises me as it seems simple enough. The relevant code I would like to ask about is this function and that INTERFACE_UNPACK_UINT8 line.
uint8_t gapm_reset_req_handler (void) {
uint8_t u8Operation, u8Status;
INTERFACE_MSG_INIT(GAPM_RESET_CMD, TASK_GAPM);
INTERFACE_PACK_ARG_UINT8(GAPM_RESET);
INTERFACE_SEND_WAIT(GAPM_CMP_EVT, TASK_GAPM);
INTERFACE_UNPACK_UINT8(&u8Operation);
INTERFACE_UNPACK_UINT8(&u8Status);
INTERFACE_MSG_DONE();
if(u8Operation!=GAPM_RESET)
return AT_BLE_FAILURE;
return u8Status;}
These INTERFACE messages are defined in another file, and I am a bit lost at what exactly is supposed to be accomplished by the generated code regarding the use of the double underscore on the ptr variable. Does anyone have any intuition as to what is going on? To me, it looks like some operation on the value that is passed to it but the use of the double underscore confuses me as I thought that was just for macros. Any thoughts are greatly appreciated!
Specific line
#define INTERFACE_UNPACK_UINT8(ptr)\
*ptr = *__ptr++
Full Definition of INTERFACE Code:
#ifndef __INTERFACE_H__
#define __INTERFACE_H__
#include "event.h"
#define INTERFACE_HDR_LENGTH 9
#define INTERFACE_API_PKT_ID 0x05
#define INTERFACE_SEND_BUF_MAX 600
#define INTERFACE_RCV_BUFF_LEN 500
extern uint8_t interface_send_msg[INTERFACE_SEND_BUF_MAX];
void platform_send_lock_aquire(void);
void platform_send_lock_release(void);
#define INTERFACE_MSG_INIT(msg_id, dest_id) \
do{\
uint8_t* __ptr = interface_send_msg;\
uint16_t __len;\
platform_send_lock_aquire();\
*__ptr++ = (INTERFACE_API_PKT_ID);\
*__ptr++ = ((msg_id) & 0x00FF );\
*__ptr++ = (((msg_id)>>8) & 0x00FF );\
*__ptr++ = ((dest_id) & 0x00FF );\
*__ptr++ = (((dest_id)>>8) & 0x00FF );\
*__ptr++ = ((TASK_EXTERN) & 0x00FF );\
*__ptr++ = (((TASK_EXTERN)>>8) & 0x00FF );\
__ptr += 2
#define INTERFACE_PACK_ARG_UINT8(arg)\
*__ptr++ = (arg)
#define INTERFACE_PACK_ARG_UINT16(arg)\
*__ptr++ = ((arg) & 0x00FF);\
*__ptr++ = (((arg) >> 8) & 0x00FF)
#define INTERFACE_PACK_ARG_UINT32(arg) \
*__ptr++ = (uint8_t)((arg) & 0x00FF );\
*__ptr++ = (uint8_t)(( (arg) >> 8) & 0x00FF) ;\
*__ptr++ = (uint8_t)(( (arg) >> 16) & 0x00FF);\
*__ptr++ = (uint8_t)(( (arg) >> 24) & 0x00FF)
#define INTERFACE_PACK_ARG_BLOCK(ptr,len)\
memcpy(__ptr, ptr, len);\
__ptr += len
#define INTERFACE_PACK_ARG_DUMMY(len)\
__ptr += len
#define INTERFACE_PACK_LEN()\
__len = __ptr - &interface_send_msg[INTERFACE_HDR_LENGTH];\
interface_send_msg[7] = ((__len) & 0x00FF );\
interface_send_msg[8] = (((__len)>>8) & 0x00FF);\
__len += INTERFACE_HDR_LENGTH;
#define INTERFACE_SEND_NO_WAIT()\
INTERFACE_PACK_LEN();\
interface_send(interface_send_msg, __len)
#define INTERFACE_SEND_WAIT(msg, src)\
watched_event.msg_id = msg;\
watched_event.src_id = src;\
INTERFACE_PACK_LEN();\
interface_send(interface_send_msg, __len);\
if(platform_cmd_cmpl_wait()){return AT_BLE_FAILURE;}\
__ptr = watched_event.params;\
#define INTERFACE_MSG_DONE()\
platform_send_lock_release();\
}while(0)
#define INTERFACE_UNPACK_INIT(ptr)\
do{\
uint8_t* __ptr = (uint8_t*)(ptr);\
#define INTERFACE_UNPACK_UINT8(ptr)\
*ptr = *__ptr++
#define INTERFACE_UNPACK_UINT16(ptr)\
*ptr = (uint16_t)__ptr[0]\
| ((uint16_t)__ptr[1] << 8);\
__ptr += 2
#define INTERFACE_UNPACK_UINT32(ptr)\
*ptr = (uint32_t)__ptr[0] \
| ((uint32_t)__ptr[1] << 8) \
| ((uint32_t)__ptr[2] << 16)\
| ((uint32_t)__ptr[3] << 24);\
__ptr += 4
#define INTERFACE_UNPACK_BLOCK(ptr, len)\
memcpy(ptr, __ptr, len);\
__ptr += len
#define INTERFACE_UNPACK_SKIP(len)\
__ptr += (len)
#define INTERFACE_UNPACK_DONE()\
}while(0)
void interface_send(uint8_t* msg, uint16_t u16TxLen);
#endif /* HCI_H_ */
*ptr = *__ptr++ is simply a byte copy followed by increasing the source pointer by one. __ptr is a local variable declared inside one of the macros then re-used in the other macros.
Notably, it is bad practice to use identifiers starting with underscore and particularly with two underscore or one underscore + an upper case letter. These are reserved for the compiler and standard lib, and the lib you post does not appear to belong to either. So there is reason to believe it was badly designed.
The following function-like macro nightmare confirms this - this is some horrible code with non-existent type safety and massive potential for undefined behavior upon bitwise arithmetic with signed numbers. People used to write macro crap like this before function inlining became industry standard back in the 1980s-1990s. Although stdint.h was introduced in 1999 so more likely they were just incompetent.
As for what the code does, it is much simpler than it looks. There's just various macros for shoveling data from one data type to another, apparently part of some protocol encoding/decoding. They also seem to make various assumptions about endianess that aren't portable.
Please never use or trust code provided to you by some silicon vendor. They have a very long tradition of employing the absolutely worst programmers in the world. If someone wrote microcontroller code like this in a normal company, they would get fired immediately. Similarly, don't trust the average open source barf posted on Github either.

define a macro that converts values between byte order

I want to create a macro that converts unsigned value conv to the opposite byte order of the current CPU. When not in a macro, it works well but let's say I want to do that as a macro. The compiler throws me an implict decleration when I try to use the macro. Take in mind that cpu_to_be32 and friends is a kernel-space functions, afaik at least.
#define be32_or_le32(conv) do { #ifdef __LITTLE_ENDIAN \
conv = cpu_to_be32(conv); \
#elif __BIG_ENDIAN \
conv = cpu_to_le32(conv); \
#endif } while (0)
u32 x = 0x12345678;
/* Convert the CPU's default byteorder to the opposite one */
be32_or_le32(x); /* Implict decleration */
Update: The answer below works really well but only without the do {} while (0), why when do.. added an error is thrown?
#ifdef __LITTLE_ENDIAN
#define be32_or_le32(conv) do { conv = cpu_to_be32(conv); } while (0)
#elif __BIG_ENDIAN
#define be32_or_le32(conv) do { conv = cpu_to_le32(conv); } while (0)
#endif
int __init sys_kernelmod_init(void)
{
u32 conv;
u32 x = 0x12345678;
/* Convert the CPU's default byteorder to the opposite one */
conv = be32_or_le32(x);
...
}
Don't use macros for things like this - use "static inline" functions. And try to give more sensible names - you are swapping the bytes in the 32-bit value, so it is a "byte swap" function. It is completely independent of the endianness of the system as you are always reversing the endianness, so there is no need to do any kind of conditional compilation. And stick to the standard type names (such as "uint32_t"), not home-made names (like "u32"), unless you have very good reason.
uint32_t bswap32(uint32_t x) {
return ((x >> 24) | ((x & 0x00FF0000) >> 8) | ((x & 0x0000FF00) << 8) | (x << 24));
}
Any decent compiler will optimise this to a single instruction if the target processor has a byte-swap instruction.
You cannot have preprocessor conditionals inside the macro expansion text.
Switch the structure to:
#ifdef __LITTLE_ENDIAN
#define be32_or_le32(conv) do { conv = cpu_to_be32(conv); } while (0)
#elif __BIG_ENDIAN
#define be32_or_le32(conv) do { conv = cpu_to_le32(conv); } while (0)
#endif

Extracting bytes from a 32 bits number

This is not important and should be quite simple, I just don't understand what I'm doing wrong.
The story behind this is that I'm playing with tinyNeoPixel lib on the attiny85, and I'm trying to dive a bit deeper than I need.
This is traditional ANSI C and I'm using a Raspberry Pi3 for this test, but for this case this should be irrelevant. The sizeof(c) on the printf just shows that 'c' is 4 bytes, as expected.
I'm trying to extract the Red, Green, and Blue part of a color that's stored as a 32 bits number
Obviously I'm failing to return the result as a 1 byte value, can same one please tell me how do I do that ? Just casting to (uint8_t) just produces zero.
Thank you.
pi3:~/src$ cat a.c
#include <stdio.h>
typedef unsigned char uint8_t;
typedef unsigned long int uint32_t;
#define Red(x) (x & 0xff000000)
#define Green(x) (x & 0x00ff0000)
#define Blue(x) (x & 0x0000ff00)
void main()
{
uint32_t c;
uint8_t r,g,b;
c=0xffeecc00;
r=Red(c);
g=Green(c);
b=Blue(c);
printf("%d - %08x - %02x %02x %02x\n", sizeof(c), c, r, g, b);
printf("%d - %08x - %02x %02x %02x\n", sizeof(c), c, Red(c), Green(c), Blue(c));
}
pi3:~/src$ gcc a.c -o a
pi3:~/src$ ./a
4 - ffeecc00 - 00 00 00
4 - ffeecc00 - ff000000 ee0000 cc00
The solution is:
#define Red(x) (((x) & 0xff000000) >> 24)
#define Green(x) (((x) & 0x00ff0000) >> 16)
#define Blue(x) (((x) & 0x0000ff00) >> 8)
With this macros this produces
pi3:~/src$ ./a
4 - ffeecc00 - ff ee cc
4 - ffeecc00 - ff ee cc
as it should.
Thank you guys.
You need to shift as well as mask. That is, try something like
#define Red(x) (((x) & 0xff000000) >> 24)
and similarly for your Green() and Blue() macros.
(Also note that I've added two extra pairs of parentheses to the macro definition, for safety in expansion.)

Is there any way to speedup be32 encoding in C?

Is there any way to speedup be32enc in C? Here's an example of what I do for uint32_t:
for (int i=0; i < 19; i++) {
be32enc(&endiandata[i], pdata[i]);
}
And the function itself:
static inline void be32enc(void *pp, uint32_t x)
{
uint8_t *p = (uint8_t *)pp;
p[3] = x & 0xff;
p[2] = (x >> 8) & 0xff;
p[1] = (x >> 16) & 0xff;
p[0] = (x >> 24) & 0xff;
}
I've googled hard, but haven't found anything - this topic is not so popular. Target CPU for this would be i3-7350k and I use msvc2017. May use MIT/GPL libs as well.
There are two modifications that are likely to improve the performance of your be32inc function. First get rid of the pointer magic and make it a function from uint32_t to uint32_t. Second, if you don't need to be portable to other architectures than x86, implement it using the _bswap-intrinsic.
If you have a decent compiler, you should be able to use builtins (btw there is a BSD standard function that does what you want, htobe32()):
#ifndef I_HAVE_A_CRAP_COMPILER
#define bswap32(x) __builtin_bswap32(x)
#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
#define htobe32(x) bswap32(x)
#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
#define htobe32(x) (x)
#else
#error Must be little or big endian
#endif
#else
/*your implementation here*/
#endif
Edit: if you want to try your C library's builtin htobe32() function you can:
#define _BSD_SOURCE
#include <endian.h>
Though the compiler builtin will likely be faster, since it will avoid a function call altogether and inline efficient assembly (a single bswap instruction on x86 and x86_64)

Is there a C preprocessor that eliminates #ifdefs but also evaluates preprocessor macros?

I'm having a project that does lots of this
#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37)
// do some legacy stuff
#else
// do current stuff
#endif
where KERNEL_VERSION is defined as
#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
I'd like to eliminate the defines that are not relevant for the current version, but tools like sunifdef don't evaluate the KERNEL_VERSION macro, so something like
sunifdef --replace -DKERNEL_VERSION\(a,b,c\)=\(\(\(a\)\<\<16\)+\(\(b\)\<\<8\)+\(c\)\) -DLINUX_VERSION_CODE=3.13.1 *
fails with the message
sunifdef: error 0x04200: Garbage in argument "-DKERNEL_VERSION(a,b,c)=(((a)<<16)+((b)<<8)+(c))"
How do I get around this?
With sunifdef 3.1.3, you can't do it, as you demonstrated. Nor can you do it with earlier versions of coan such as 4.2.2.
However, with coan 5.2 (the current version), you can almost do what you are after.
$ cat legacy.c
#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37)
do(some,legacy,stuff)
#else
do(current,stuff)
#endif
$ coan source -DLINUX_VERSION_CODE=0x020635 legacy.c
coan: /Users/jleffler/soq/legacy.c: line 1: warning 0x0041c: "-DKERNEL_VERSION(a,b,c)=(((a) << 16) + ((b) << 8) + (c))" has been assumed for the current file
#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
do(current,stuff)
$ coan source -DLINUX_VERSION_CODE=0x020624 legacy.c
coan: /Users/jleffler/soq/legacy.c: line 1: warning 0x0041c: "-DKERNEL_VERSION(a,b,c)=(((a) << 16) + ((b) << 8) + (c))" has been assumed for the current file
#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
do(some,legacy,stuff)
$
This is close to what you want, but not quite. It gives 'correct' output, but maybe not 'helpful' output. It gives you the code that would be compiled for the LINUX_VERSION_CODE specified on the command line, whereas you'd probably like the conditionals based on LINUX_VERSION_CODE and KERNEL_VERSION that are not false to survive into the output.
The successor to sunifdef seems to be coan, and the following command seemed to work (on one simple file):
coan source "-DLINUX_VERSION=KERNEL_VERSION(2,18,1)" \
"-DKERNEL_VERSION(a,b,c)=((a)*0x10000 + (b)*0x100 + (c))" \
testfile.c
I think using the KERNEL_VERSION macro to define LINUX_VERSION is prettier, but you might prefer Chris Dodd's version in hex. Two dots in a number is definitely not going to work.
The error is coming from the fact that you can't define macros with arguments on the command line with -D -- you can only define simple macros. However, you shouldn't NEED to define KERNEL_VERSION on the command line as the #define in the source should be fine. You should only need -DLINUX_VERSION_CODE=0x30d01 -- you need to define it as a single integer constant (hex is easiest) rather than withs dots.

Resources