On Tue, 16 Sep 2008 15:35:42 +0200
Sebastian Siewior <bigeasy at linutronix.de> wrote:
> + return sprintf(buf, "%d\n", TEMP_FROM_REG((data->kind == max1618) ?
> + data->remote : data->local));
oh dear.
#define TEMP_FROM_REG(val) ((val & 0x80 ? val-0x100 : val) * 1000)
Imagine how truly awful the code generation must be here.
Look:
--- a/drivers/hwmon/max1619.c~a
+++ a/drivers/hwmon/max1619.c
@@ -75,8 +75,15 @@ I2C_CLIENT_INSMOD_2(max1619, max1618);
* Conversions and various macros
*/
-#define TEMP_FROM_REG(val) ((val & 0x80 ? val-0x100 : val) * 1000)
-#define TEMP_TO_REG(val) ((val < 0 ? val+0x100*1000 : val) / 1000)
+static int TEMP_FROM_REG(int val)
+{
+ return ((val & 0x80 ? val-0x100 : val) * 1000);
+}
+
+static int TEMP_TO_REG(int val)
+{
+ return (val < 0 ? val+0x100*1000 : val) / 1000;
+}
/*
* Functions declaration
_
text data bss dec hex filename
before: 3927 1148 28 5103 13ef drivers/hwmon/max1619.o
after: 3743 1148 28 4919 1337 drivers/hwmon/max1619.o
That's a 6% reduction in the number of instructions in the whole driver!
Not only that, it generates nicer-to-read code and it fixes the bugs
which will occur if someone calls one of these macros with an
expression which has side-effects.
Macros suck, suck, suck, suck and the kernel is just littered with the
stupid things in places where they were completely unnecessary.
argh.
