Oops, I forgot to test that: I had both implementations.
Doing a million loop (so there's loop overhead):
350MHz dual Pentium II, atomic is almost twice as fast
PPC 500MHz G4: atomic is 3.5 times as fast.
Power3 4-way 375MHz machine: atomic is 10% faster.
Power4 4-way 1.3GHz machine: atomic was 20% slower (depending
on version if irq_restore).
I suspect a P4 might get similar pro-irq results (code below,
anyone?)
Ideally we'd have a "local_inc()" and "local_dec()". Architectures
which do soft interrupts enabling should see a real win from the
save/restore version.
Meanwhile, I'll revert to atomics, since it's sometimes dramatically
faster.
Thanks!
Rusty.
-- Anyone who quotes me in their sig is an idiot. -- Rusty Russell.static void test(void) { struct timeval start, end; atomic_t x; unsigned int tmp, i, diff; unsigned long flags;
/* Atomic test. */ atomic_set(&x, 0); do_gettimeofday(&start); for (i = 0; i < 1000000; i++) atomic_dec(&x); do_gettimeofday(&end);
diff = (end.tv_sec - start.tv_sec) * 1000000 + (end.tv_usec - start.tv_usec); printk("Atomic test: %u usec\n", diff);
/* Interrupt test (interrupts enabled) */ tmp = 0; do_gettimeofday(&start); for (i = 0; i < 1000000; i++) { local_irq_save(flags); tmp++; local_irq_restore(flags); } do_gettimeofday(&end);
diff = (end.tv_sec - start.tv_sec) * 1000000 + (end.tv_usec - start.tv_usec); printk("Interrupt test: %u usec\n", diff);
/* Interrupt test (interrupts disabled) */ tmp = 0; local_irq_disable(); do_gettimeofday(&start); for (i = 0; i < 1000000; i++) { local_irq_save(flags); tmp++; local_irq_restore(flags); } do_gettimeofday(&end); local_irq_enable();
diff = (end.tv_sec - start.tv_sec) * 1000000 + (end.tv_usec - start.tv_usec); printk("Interrupt test (interrupts disabled): %u usec\n", diff); } - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/