TDA2P-ACD: IPU1 GPIO interrupt response time too high

Part Number:TDA2P-ACD

We are using a TDA2P processor and have to react to GPIO interrupts in GPIO module 6 pin 19.
The interrupt is sent to IPU1 M4 which is running at 212 MHz.

Time between interrupts:
First to second interrupt: 46.08 µs
Second to third interrupt: 9.6 µs
Third to fourth interrupt: 46.08 µs
and so on...

Our software is a bare-metal system using CSL functions to configure interrupts and the GPIO module.

For test purposes I created an interrupt handler that sets, clears and then again sets a second GPIO pin. In the main loop I am simply clearing this GPIO pin all the time in an endless loop.

void gpioIsr(void * p)
{
   GPIOPinWrite(
      SOC_GPIO4_BASE,
      0,
      1);
   GPIOPinWrite(
      SOC_GPIO4_BASE,
      0,
      0);
   GPIOPinWrite(
      SOC_GPIO4_BASE,
      0,
      1);
   GPIOPinIntClear(SOC_GPIO6_BASE, GPIO_INT_LINE_1, 19);
}

This is the code to configure the GPIO interrupt:

GPIOModuleEnable(SOC_GPIO6_BASE);
GPIOPinIntDisable(SOC_GPIO6_BASE, GPIO_INT_LINE_1, 19);
CSL_xbarIrqConfigure(IRQ_XBAR_CPU_ID, IRQ_XBAR_INST_NO, IRQ_XBAR_INDEX);
Intc_Init();
Intc_IntEnable(51);
Intc_IntRegister(51, (IntrFuncPtr)gpioIsr, (void *)0);
Intc_IntPrioritySet(51, 1);
Intc_SystemEnable(51);
GPIOIntTypeSet(SOC_GPIO6_BASE, 19, GPIO_INT_TYPE_RISE_EDGE);
GPIOPinIntClear(SOC_GPIO6_BASE, GPIO_INT_LINE_1, 19);
GPIOPinIntEnable(SOC_GPIO6_BASE, GPIO_INT_LINE_1, 19);

Setting and clearing a GPIO pin both take about 5 µs.

During time measurements I found out that the time to enter interrupts is quite long. The time from the incoming GPIO interrupt to the rising edge on the GPIO signal I set is about 10 µs. Even if I subtract 5 µs for setting the GPIO it still takes the CPU ~1000 clock cycles to reach the interrupt code.
After the code in the ISR has been executed it also takes about 6 to 10 µs to go back to the main loop. The first interrupt even takes ~35 µs to return to the main loop (maybe caused by caching?).
This timing means that I am missing a lot of interrupts.

This is a trace of the two GPIO signals used for the measurements:

Are these timings reasonable for the used CPU or is there a way to speed-up the interrupt response time?
Are 5 µs a reasonable time to toggle a GPIO pin? Maybe something in the crossbars is not configured correctly yet.

Our required actions in the ISR would be to:

1. Set a GPIO pin
2. Start a DMA transfer to one SPI channel

Can this be achieved in 9.6 µs? If not could using the PRUs be a solution for this timing requirement?

Many thanks

Sascha Bannier