7.6.3 Fast Integer Division Intrinsics

The intrinsics listed in Table 7-9 perform faster division using hardware fast integer division support. These intrinsics are enabled if the --idiv_support=idiv0 compiler option is used.

In order to use these intrinsics, your code must include the stdlib.h header file and the --float_support option must be set to fpu32 or fpu64. Fast integer division support is available for EABI only.

These intrinsics follow the format of the ldiv and lldiv standard library functions. They take as input the dividend and divisor, and return a structure containing both the quotient and remainder in the fields quot and rem, respectively. The __uldiv_t and __ulldiv_t types are unsigned equivalents to ldiv and lldiv provided by stdlib.h. In addition, the following additional structure types are provided to return long long quotients with a long remainder:

typedef struct { long long quot; long rem; } __llldiv_t; typedef struct { unsigned long long quot; unsigned long rem; } __ullldiv_t;

These intrinsics support three types of integer division:

• Traditional division. Also called truncated division or C99 division. This type of division rounds toward zero. As a consequence, the remainder sign is the same as the sign of the dividend.
• Modulo division. Also called floored division. This type of division rounds toward negative infinity. As a consequence, the remainder sign is the same as the sign of the divisor.
• Euclidean division. This type of division rounds toward infinite magnitude. As a consequence, the remainder is always non-negative.

For all three types of division, the dividend is exactly equal to quotient * divisor + remainder. All three types of division compute the same quotient and remainder if the remainder is zero or if the dividend and divisor are positive.

Since the results of division between two unsigned values do not vary for the traditional, Euclidean, and modulo versions, only the traditional versions are provided for unsigned inputs.

In general, if the dividend and divisor are of different types, the divisor is converted to the type of the dividend before division. However, in cases where the dividend type is signed, the divisor type is unsigned, and the dividend type is not larger than the divisor type—that is, for the *_div_i32byu32 and *_div_i64byu64 intrinsics—division is carried out using a signed type larger than both the dividend and divisor, and then converted to the type of the dividend.

Division by 0 is always undefined.

In addition to these intrinsics, when the --idiv_support=idiv0 compiler option is used, the built-in integer division and modulo operators ("/" and "%") use the appropriate faster instructions as described in Section 7.8.2. The faster versions of these built-in operators are used whether or not the stdlib.h header file is included.