From: Pali Rohár

Sent: 22 July 2021 22:58

On Monday 19 July 2021 15:47:07 Andy Shevchenko wrote:

quoted

On Sat, Jul 17, 2021 at 3:39 PM Pali Rohár [off-list ref] wrote:

quoted

Provide DIV_U64_ROUND_CLOSEST helper which uses div_u64 to perform
division rounded to the closest integer using unsigned 64bit
dividend and unsigned 32bit divisor.

...

quoted

+/*
+ * DIV_U64_ROUND_CLOSEST - unsigned 64bit divide with 32bit divisor rounded to nearest integer

quoted

+ * @dividend: unsigned 64bit dividend

Here you insist users to provide a u64 (or compatible) type.

quoted

+ * @divisor: unsigned 32bit divisor
+ *
+ * Divide unsigned 64bit dividend by unsigned 32bit divisor
+ * and round to closest integer.
+ *
+ * Return: dividend / divisor rounded to nearest integer
+ */
+#define DIV_U64_ROUND_CLOSEST(dividend, divisor)       \

quoted

+       ({ u32 _tmp = (divisor); div_u64((u64)(dividend) + _tmp / 2, _tmp); })

Here is the casting to u64. Why? (Yes, I have read v1 discussion and I
just want to continue it here).

See also Willy's response: https://lore.kernel.org/lkml/20210625155008.GB16901@1wt.eu/ (local)

Macro does not enforce type as opposite to function.

There is no compile time check for correct type and neither compile time
warning if smaller typed value is passed.

And e.g. passing constant with explicit ULL suffix or casting external
constant to 64bit type is impractical.

What is wrong with:
	({ u32 _tmp = (divisor); div_u64((dividend) + (u64)0 + _tmp / 2, _tmp); })
Or:
	({ u32 _tmp = (divisor); div_u64((dividend) + (u64)(_tmp / 2), _tmp); })
Both will ensure a 32bit 'dividend' is promoted to 64 bits before the '+'.
Both will fail if 'dividend' isn't an integer type.
On 32bit systems the compiler will also know when the high bits van be ignored.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)