I’m sorry for late response. I couldn’t reply for this topic. Then, I started the following new topic.
Zynq Ultrascale MPSoC task floating point corruption 2
I’m sorry for separated topics. Thank you for your support.
[RTEL EDIT: Pasting the post from the other thread below to combine threads. I also upped your privileges so you can continue to post in this thread:]
The overhead is both in time and stack usage - whether that is an issue for you is very much dependent on the application you are writing. These devices can have a lot of RAM.
Yes, my device has enough RAM. It means stack usage is not problem. I will estimate time overhead.
It is unexpected that you need to save all the floating point registers on interrupt entry
I analyzed disassembled code for my floating point operations in interrupt. I found that s0, s1 and s2 registers are used in those operations. Does it mean v0-v2 require to be saved on IRQ entry?
I need to compile some code to see what it is doing. We have a Zynq project that uses Vitis and runs in QEMU now too
Thank you very much for an investigation on your side. I am using Xilinx SDK 2019.1 instead of Vitis. Compiler is GCC which is built in Xilinx SDK. Maybe it is almost same as compiler in Vitis. Thank you for your support!