clu2's notes: Notes on data types & floating-point arithmetic in C

Notes on data types & floating-point arithmetic in C

Common data types in C

Data Type	as in stdint.h	Suffix for literals	printf format	printf format as in inttypes.h	Range as in limits.h or float.h
double		No suffix	%e, %E, %f, %g, %G		DBL_MAX, DBL_MIN, DBL_EPSILON
float		f or F	%e, %E, %f, %g, %G		FLT_MAX, FLT_MIN, FLT_EPSILON
long double (could be either 80 or 128 bits)		l or L	%Le, %LE, %Lf, %Lg, %LG		LDBL_MAX, LDBL_MIN, LDBL_EPSILON
__float128
int	int32_t	No suffix	%d	PRId32	INT_MAX, INT_MIN
unsigned int	uint32_t	u or U	%u	PRIu32	UINT_MAX, UINT_MIN
long	int64_t	l or L	%ld	PRId64	LONG_MAX, LONG_MIN
unsigned long	uint64_t	ul or UL	%lu	PRIu64	ULONG_MAX, ULONG_MIN
long long int	int64_t	ll or LL	%lld	PRId64	LLONG_MAX, LLONG_MIN
unsigned long long int	uint64_t	ull or ULL	%llu	PRIu64	ULLONG_MAX, ULLONG_MIN
__int128_t __uint128_t
intmax_t uintmax_t	intmax_t uintmax_t		%jd %ju	PRIdMAX PRIuMAX	INTMAX_MAX, INTMAX_MIN, UINTMAX_MAX
ssize_t size_t	(ssize_t is defined in unistd.h and size_t in stddef.h)		%zd %zu		SIZE_MAX
intptr_t uintptr_t ptrdiff_t	(ptrdiff_t is defined in stddef.h and is for result of pointer subtraction)		%p	__PRIPTR_PREFIX	PTRDIFF_MAX, PTRDIFF_MIN

LP64 vs LLP64

I = integer, L = long, LL = long long, P = pointer

Data Type	LP64 (64-bit Linux)	ILP64	LLP64 (64-bit Windows)	ILP32	LP32
char	8	8	8	8	8
short	16	16	16	16	16
int	32	64	32	32	16
long	64	64	32	32	32
long long	64	64	64
pointer	64	64	64	32	32

See here and here for details.

Floating-point handling in optimizing compilers

Compilers have fine-grained control over optimizations they perform on numerical computation code. Here are some common compiler command-line options:

Optimization	GCC	Intel	PGI	PathScale Open64	Sun Studio
Denormals Are Zero (DAZ) Flush To Zero (FTZ)	Enabled if either of the following are used: `-ffast-math` `-funsafe-math-optimizations`	Enabled at the default level of optimization. If `-O0` is used, use `-ftz` to enable.	`-Mdaz` and `-Mflushz`	Enabled if either of the following are used or implied: `-OPT:IEEE_arith=2` `-OPT:IEEE_arith=3`	`-fns`
Legacy x87 precision control. The significand is rounded to the specified bits. Default is 80 (64-bit significand)	`-mpc32` `-mpc64` `-mpc80`	`-pc32` `-pc64` `-pc80`	`-pc 32` `-pc 64` `-pc 80`	`-mx87-precision=32` `-mx87-precision=64` `-mx87-precision=80`	`-fprecision=single` `-fprecision=double` `-fprecision=extended`