[stm32] Interrupt handler indirection causes unnecessary latency

[strongly-typed]

MODM_ISR(DMA1_Channel{{N}}) calls nothing but Dma1::Channel<DmaBase::Channel::Channel{{N}}>::interruptHandler(); in dma.cpp.in

Unfortunately, these calls are not inlined. Adding an inline did not change anything for me. These calls are real calls, increasing interrupt latency unnecessarily.

InterruptHandler is a jinja-generated C++ template ...

Why not jinja-generate this code directly into dma.cpp without the C++ template?

(Yes, I have to shave off some nanoseconds IRQ latency and I try to stay with modm for convenience.)

[chris-durand]

You could try adding [[gnu::always_inline]] at the definition of interruptHandler(). That should force inlining.