Treating empty cycles in performance simulator

[graudtv]

I was running some programs on the simulator and found a bit strange behaviour with empty loops.
The simulator executes only a fixed number of instructions, if there is an empty loop, and then stops without any messages. '-n' parameter in the command line call is ignored if it is greater than this fixed number.

Assembler source file (empty_cycle.s):

.globl _start
.section .text

_start:
    j _start

Building and running options:

riscv64-unknown-elf-as -march=rv32g empty_cycle.s -o empty_cycle.o
riscv64-unknown-elf-ld -m elf32lriscv empty_cycle.o -o empty_cycle
../mipt-v -n 1000000 -I riscv32 -b empty_cycle --bp-mode always_not_taken 

Output:

instrs:     1
cycles:     38

Expected results (such results are obtained, if at least one instruction is added into the cycle, for example nop)

instrs:     1000000
cycles:     2500032

[pavelkryukov]

Empty loops are often used as an indication to halt.
Here is where it is handled:

mipt-mips/simulator/func_sim/alu.h

Lines 49 to 53 in ba87224

	static void check_halt_trap( Instr* instr) {
	// Handles 'goto nullptr;' and 'while (1);' cases
	if ( instr->new_PC == 0 || instr->new_PC == instr->PC)
	instr->trap = Trap::HALT;
	}

[graudtv]

Does it have any practical usage?

[pavelkryukov]

No, that's why they are used to indicate simulation halt
For instance, RISC-V community tests use that convention.

[graudtv]

Ok, thanks for answer

[pavelkryukov]

You are welcome!