Hi all


I am trying to FPU simulation by using the mor1kx-4.1 in fusesoc.

I am using a variety of numbers in simulation

The use of a number such as 0.5 or 0.0, simulation is correctly.

However, in other numbers will be simulation suddenly end with a message that Closing RSP

server.

Please tell me about the causes and solutions of this problem.

Thanks

I'm not really sure what you mean here. Could you provide some more context? Where does your numbers come from? Are you sending a program with these numbers to FuseSoC? Which simulator?

i'm using ELF loader in fusesoc and icarus simulator i have make a C program #include #include int main() { float a,b,c,d,e,f; a = 4.5; b = 10.3; printf("start\n"); c = a + b; printf(" c = a + b = %f\n",c); d = a - b; printf(" d = a - b = %f\n",d); e = a / b; printf(" e = a / b = %f\n",e); f = a * b; printf(" f = a * b = %f\n",f); exit(0); } and i was compile with or1k-elf-gcc i ran the simulation in this command fusesoc sim --force mor1kx-generic --elf-load /root/file name as a result, Simulation is terminated with a masage Closing RSP server

Hi,

The mor1kx-generic system will always print out Closing RSP server before exiting. I thought I had fixed that, but apparently not. Anyway, it's not an error message, so you can ignore it.

I compiled and ran your program, and I get the following output (just before Closing RSP server)

start
c = a + b = 14.800000
d = a - b = -5.800000
e = a / b = 0.436893
f = a * b = 46.350002

Is this what you see as well?

thanks olof.

your answer is a big help for me

Hi, my app crashes at runtime when it execute FPU instruction. Can anybody help me?
My src very simple:

int main(){
float c = 0.0; //
c = c + 1.2; // (1) OK
c = c + 2.2; // (2) OK
c = c + 3.2; // (3) CRASH
return 0;
}

I use -lsingle-precision-constant and -mhart-float opts at compilation stage:

> or1k-elf-gcc -lsingle-precision-constant -mhard-float a.c -o a.elf

and have nice assembler code with FPU instructions (gcc -S), like that:

l.movhi r3,hi(.LC1) # movsi_high
l.ori r3,r3,lo(.LC1) # movsi_lo_sum
l.lwz r3,0(r3) # movsf
lf.add.s r3,r4,r3 # addsf3 (1)
l.sw -8(r2),r3 # movsf
l.lwz r4,-8(r2) # movsf
l.movhi r3,hi(.LC2) # movsi_high
l.ori r3,r3,lo(.LC2) # movsi_lo_sum
l.lwz r3,0(r3) # movsf
lf.add.s r3,r4,r3 # addsf3 (2)
l.sw -8(r2),r3 # movsf
l.lwz r4,-8(r2) # movsf
l.movhi r3,hi(.LC3) # movsi_high
l.ori r3,r3,lo(.LC3) # movsi_lo_sum
l.lwz r3,0(r3) # movsf
lf.add.s r3,r4,r3 # addsf3 (3)
l.sw -8(r2),r3 # movsf

I use fusesoc to simulate modificated mor1kx-generic (set mor1kx-4.1 instead mor1kx and enabled FEATURE_FPU) + iverilog

> fusesoc sim mor1kx-generic --elf-load=a.elf --vcd --trace_enable

Trace log:

S 000023b4: 8482fff8 l.lwz r4,0xfff8(r2) r4 = 007fdff8 flag: 0
S 000023b8: 18600000 l.movhi r3,0x0000 r3 = 00000000 flag: 0
S 000023bc: a8633a8c l.ori r3,r3,0x3a8c r3 = 00003a8c flag: 0
S 000023c0: 84630000 l.lwz r3,0x0000(r3) r3 = 00003a8c flag: 0
S 000023c4: c8641800 0.00 ns: Unknown opcode 0x32 flag: 0
S 000023c4: c8641800 0.00 ns: Unknown opcode 0x32 flag: 0
S 000023c8: d7e21ff8 l.sw 0xfff8(r2),r3 [007fdff8] = 3f99999a flag: 0 (1)
S 000023cc: 8482fff8 l.lwz r4,0xfff8(r2) r4 = 007fdff8 flag: 0
S 000023d0: 18600000 l.movhi r3,0x0000 r3 = 00000000 flag: 0
S 000023d4: a8633a90 l.ori r3,r3,0x3a90 r3 = 00003a90 flag: 0
S 000023d8: 84630000 l.lwz r3,0x0000(r3) r3 = 00003a90 flag: 0
S 000023dc: c8641800 0.00 ns: Unknown opcode 0x32 flag: 0
S 000023dc: c8641800 0.00 ns: Unknown opcode 0x32 flag: 0
S 000023e0: d7e21ff8 l.sw 0xfff8(r2),r3 [007fdff8] = 4059999a flag: 0 (2)
S 000023e4: 8482fff8 l.lwz r4,0xfff8(r2) r4 = 007fdff8 flag: 0
S 000023e8: 18600000 l.movhi r3,0x0000 r3 = 00000000 flag: 0
S 000023ec: a8633a94 l.ori r3,r3,0x3a94 r3 = 00003a94 flag: 0
S 000023f0: 84630000 l.lwz r3,0x0000(r3) r3 = 00003a94 flag: 0
S 000023f4: c8641800 0.00 ns: Unknown opcode 0x32 flag: 0
S 000023f4: c8641800 0.00 ns: Unknown opcode 0x32 flag: 0
S 00000d00: d4000804 l.sw 0x0004(r0),r1 [00000004] = 007fdff8 flag: 0

As you can see, expressions (1) and (2) are calculated correctly, but (3) is falling down.
I wonder why this is happening. How do I find out what happened?

Hi,

the cpu gets into the floating point exception. Can you please provide the generate vcd file?

Cheers,
Stefan

We got this worked out now. It turns out that there is an exception when the does a calculation where the result is not exact. We will change mor1kx to disable the FP exceptions by default, as we usually aren't interested in them and we have no software to take care of them yet.

In the meantime you can set iverilog_options = -DSIM -DOR1K_FPCSR_MASK_FLAGS in mor1kx-generic.core to disable this when running simulations with Icarus

//Olof