changing reject to accept introduce stack increaseing about p4c-xdp

[eggqq007]

Upload the llvm-objdump -S -no-show-raw-insn xdp_test.o files:

xdp_test_smallstack.txt
xdp_test_bigstack.txt

[eggqq007]

ping......

[mihaibudiu]

I don't know if there is much we can say about this; we do not have any control on how clang generates code.

[eggqq007]

OK....I found a workaround.
I rewrite some logical in ControlBodyTranslator::compileEmitField to consume memcpy, and eliminate the write_byte. It helps reduce the stack from about 350 to 128byte. And performance is better also.

I also rewrite some logical in ControlBodyTranslator::preorder to help to judge if we should rewrite re-populate a header.

And also fix the issue that using if-else in control.("if" emit a useless ";" in if-else statement which cause compile error)

I found the counter only support 32bit which is too small, so increase it from 32bit to 64bit if using counterArray which type is hash.

[eggqq007]

introducing 64bit counter: 2.7Mpps -->2.5Mpps
using memcpy and eliminating write_byte & avoid re-populate packet's header: 2.5Mpps --> 2.8Mpps

[mihaibudiu]

Do you plan to contribute these changes to our code?

[eggqq007]

It is draft code and I am afraid I don't have enough time to make it become a accepted patch recently.
I will upload my draft code diff first in case someone need it.

[williamtu]

if you're allowed to change kernel, then you can increase MAX_BPF_STACK to a little larger value.

[eggqq007]

Not only the linux stack size, the llvm-6 and llvm-7 would hit error in compiling if the stack size above 512bytes...