[core] Fix delay_ns cycle calculation

examples/nucleo_f303k8/blink/main.cpp

@@ -17,23 +17,12 @@ int
 main()
 {
 	Board::initialize();
-	LedD13::setOutput();
-
-	// Use the logging streams to print some messages.
-	// Change MODM_LOG_LEVEL above to enable or disable these messages
-	MODM_LOG_DEBUG   << "debug"   << modm::endl;
-	MODM_LOG_INFO    << "info"    << modm::endl;
-	MODM_LOG_WARNING << "warning" << modm::endl;
-	MODM_LOG_ERROR   << "error"   << modm::endl;
-
-	uint32_t counter(0);
+	GpioA4::setOutput();
 
 	while (true)
 	{
-		LedD13::toggle();
-		modm::delay(Button::read() ? 100ms : 500ms);
-
-		MODM_LOG_INFO << "loop: " << counter++ << modm::endl;
+		GpioA4::toggle();
+		modm::delay_ns(100000);
 	}
 
 	return 0;

src/modm/platform/core/cortex/delay_impl.hpp.in

@@ -31,15 +31,15 @@ extern uint16_t delay_ns_per_loop;
 extern uint16_t delay_fcpu_MHz;
 }
 
-inline void modm_fastcode delay_ns(uint32_t ns)
+inline void modm_fastcode __attribute__((noinline)) delay_ns(uint32_t ns)
 {
 	volatile uint32_t cycles;
 	// ns_per_loop = nanoseconds per cycle times cycles per loop ({{loop}} cycles)
 	asm volatile (
 		".syntax unified \n\t"
 		".align 4 \n\t"
-		"muls.n	%[cyc], %[cyc], %[dnpl] \n\t"	// multiply the overhead cycles with the ns per cycle:  1-2 cycles on cm3, up to 32 cycles on cm0
-		"subs.n	%[cyc], %[cyc], %[ovhd] \n\t"	// subtract the overhead in ns from the input:          1 cycle
+		// "muls.n	%[ovhd], %[ovhd], %[dnpl] \n\t"	// multiply the overhead cycles with the ns per cycle:  1-2 cycles on cm3, up to 32 cycles on cm0
+		// "subs.n	%[cyc], %[cyc], %[ovhd] \n\t"	// subtract the overhead in ns from the input:          1 cycle
 	"1:  subs.n	%[cyc], %[cyc], %[dnpl] \n\t"	// subtract the ns per loop from the input:             1 cycle
 		"bpl.n	1b"								// keep doing that while result is still positive:      2 cycles (when taken)
 	: [cyc] "=l" (cycles) : "0" (ns), [dnpl] "l" (platform::delay_ns_per_loop), [ovhd] "l" ({{(overhead / loop) | int}}));