Unexpected output when running rv32emu about rv32emu

[jserv]

Cause I had found that the unexpected output only occur when the size of the previous array is not equal to multiples of four byte, so I suspected that maybe their have some bug in lw or la.

The concept of alignment is dependent upon the size of data. For word data the address is either multiple of 4 (aligned), or it is misaligned. @maromaSamsa, you can now prepare a revised MRE which reflects the above findings.

[qwe661234]

Thanks for your reply, I think this explanation convince me why the unexpected output doesn't print any character when misaligned occur in rv32emu, and I am curious about the source code.

You can check the op_load function in file emulate.c.

[jserv]

@maromaSamsa, you shall provide a minimal reproducible example (MRE) instead of hyperlinks. Edit the description of this issue to inline MRE.

[Risheng1128]

@maromaSamsa I want to check your problem. Does your problem is that the emulator executes wrong when instruction la is changed to instruction lw ?

Example:

la     s1, arr1    ->    lw    s1, arr1

[maromaSamsa]

@maromaSamsa I want to check your problem. Does your problem is that the emulator executes wrong when instruction la is changed to instruction lw ?

Example:

la     s1, arr1    ->    lw    s1, arr1

Actually I am not sure, in this RISC-V source code I just use la to get address of the byte array, and use lw to load 4 byte value from the array.

Cause I had found that the unexpected output only occur when the size of the previous array is not equal to multiples of four byte, so I suspected that maybe their have some bug in lw or la.

[maromaSamsa]

Cause I had found that the unexpected output only occur when the size of the previous array is not equal to multiples of four byte, so I suspected that maybe their have some bug in lw or la.

The concept of alignment is dependent upon the size of data. For word data the address is either multiple of 4 (aligned), or it is misaligned. @maromaSamsa, you can now prepare a revised MRE which reflects the above findings.

Please check this commit

the output is normal when you change the test case

.data
arr1:
	.byte 48, 49, 50 # this will cause unexpected output
	#.byte 48, 49, 50, 51 # check this

PS: sorry for the wrong term in this commit, is MRE(minimal reproducible example) not MSE(Mean squared error).

[qwe661234]

@maromaSamsa, I already checked this commit. The reason why test case .byte 48, 49, 50 cause unexpected output is misalignment.

In the rv32emu, arr1 and arr2 are stored in contiguous memory address. First lw instruction checks whether the address of arr1 + 4 is misaligned. Because the start address of arr1 is aligned, the address of arr1 + 4 is also aligned. Therefore, first lw instruction loads 3 bytes of data of arr1 and 1 byte of data of arr2. However, the start address of arr2 is the start address of arr1 + 3, so second lw instruction finds out arr2 + 4 is misaligned and calls exeception handler without printing any data.

As alternative, you can use lb instruction to load a byte of data.

[maromaSamsa]

However, the start address of arr2 is the start address of arr1 + 3, so second lw instruction finds out arr2 + 4 is misaligned and calls exeception handler without printing any data.

Thanks for your reply, I think this explanation convince me why the unexpected output doesn't print any character when misaligned occur in rv32emu, and I am curious about the source code.

As alternative, you can use lb instruction to load a byte of data.

Since I hope I can save the memory by constrain the test data into byte, and also lw is more efficiency than lb cause I can spent only one instruction time to load 4 byte, I choose to use lw if it is possible.

[jserv]

I probably close this as the issue is not one of emulator flaws.