wierd float64_t vs. float32_t

Question

Hi,
I have wierd experience about using float32_t.
This is my test: 
 When i use operations on float64_t i got: 
 0.0001s MAIN:INFO:init_finished
0.0001s MAIN:INFO:loop
0.5077s MAIN:INFO:loop //timer interrupt
0.4978s MAIN:INFO:loop //timer interrupt
0.4970s MAIN:INFO:loop //timer interrupt
 
 normal timer interrupt which happend every 500ms. 
 in main i have: 
 for (;;)
{
 NRF_LOG_INFO("loop
");
 power_manage();
}
 
 but when i use float32_t i have strange interrupts: 
 0.0001s MAIN:INFO:init_finished
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
0.0001s MAIN:INFO:loop
 
 Does this have any sense? 
 I use nrf52 with sdk 12.2.0 i have this code: 
 #define ARM_MATH_CM4
#include "arm_const_structs.h"
#include "arm_math.h"
 
 and flags 
 CFLAGS += -mfloat-abi=hard
CFLAGS += -mfpu=fpv4-sp-d16
LDFLAGS += -mfloat-abi=hard
LDFLAGS += -mfpu=fpv4-sp-d16

Sigurd · Accepted Answer

Hi, 
 You seem to be running into errata [87] CPU: Unexpected wake from System ON Idle when using FPU . Add the workaround to the power_manage() function, and the problem should be solved: 
 void power_manage(void)
{
 NRF_LOG_INFO("power_manage!
");
 #if (__FPU_USED == 1)
 __set_FPSCR(__get_FPSCR() & ~(0x0000009F)); 
 (void) __get_FPSCR();
 NVIC_ClearPendingIRQ(FPU_IRQn);
 #endif
 uint32_t err_code = sd_app_evt_wait();
 APP_ERROR_CHECK(err_code);
}

wierd float64_t vs. float32_t

Top Replies