I'm using nvcc to compile a CUDA kernel. Unfortunately, nvcc doesn't seem to support uint8_t
, although it does support int8_t
(!). I'd just as soon not use unsigned char
, for portability, readability, and sanity reasons. Is there another good alternative?
Just to forestall any possible misunderstanding, here are some details.
$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2010 NVIDIA Corporation
Built on Mon_Jun__7_18:56:31_PDT_2010
Cuda compilat开发者_StackOverflowion tools, release 3.1, V0.2.1221
Code containing
int8_t test = 0;
is fine, but code containing
uint8_t test = 0;
throws an error message like
test.cu(8): error: identifier "uint8_t" is undefined
C99 integer types are not "defined by the compiler" - they are defined in <stdint.h>
.
Try:
#include <stdint.h>
typedef unsigned char uint8_t;
This is no different from what Mac OS X uses:
typedef unsigned char uint8_t;
What is your concern about the portability of unsigned char
? If the concern is that a char
might not represent 8 bits of storage, then you can include a static assertion along the lines of:
typedef int Assert8BitChar[(CHAR_BIT == 8)? 0 : -1];
This will cause compilation to error out when the assumption is violated.
This seems to compile just fine with nvcc
:
#include <stdint.h>
int main() {
uint8_t x = 0;
return (int) x;
}
精彩评论