xorR is broken in chisel-tester?

[zinechant]

Type of issue: bug report

From: chipsalliance/firrtl#993

I am new to chisel world. While I am exploring, I noticed that the xorR is broken in firrtl backend for unit tests.

There were discussions about broken of reduction operators. chipsalliance/firrtl#356
But it seems xorR is still tied to the lowest bit.

A simple repo testing xorR.

Regards,
Chen

[edwardcwang]

Hello Chen,

Thank you for your interest in Chisel/FIRRTL! Does it work correctly with our newest FIRRTL simulator, treadle? https://github.com/freechipsproject/treadle

[zinechant]

Hello

I am using the chisel-template. And if I specify treadle as the backend, xorR is still tied to the lowest bit.

[info] Running xor_reduce.XorReduceMain --backend-name treadle
[info] [0.002] Elaborating design...
[info] [0.096] Done elaborating.
Total FIRRTL Compile Time: 189.9 ms
Total FIRRTL Compile Time: 52.1 ms
file loaded in 0.092280187 seconds, 8 symbols, 2 statements
[info] [0.001] SEED 1547677576725
xor.reduce.io.in1: Vector(0, 0, 0, 0)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 0, 0, 0)
xor.reduce.io.out1: 1
xor.reduce.io.in1: Vector(0, 1, 0, 0)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 1, 0, 0)
xor.reduce.io.out1: 1
xor.reduce.io.in1: Vector(0, 0, 1, 0)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 0, 1, 0)
xor.reduce.io.out1: 1
xor.reduce.io.in1: Vector(0, 1, 1, 0)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 1, 1, 0)
xor.reduce.io.out1: 1
xor.reduce.io.in1: Vector(0, 0, 0, 1)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 0, 0, 1)
xor.reduce.io.out1: 1
xor.reduce.io.in1: Vector(0, 1, 0, 1)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 1, 0, 1)
xor.reduce.io.out1: 1
xor.reduce.io.in1: Vector(0, 0, 1, 1)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 0, 1, 1)
xor.reduce.io.out1: 1
xor.reduce.io.in1: Vector(0, 1, 1, 1)
xor.reduce.io.out1: 0
xor.reduce.io.in1: Vector(1, 1, 1, 1)
xor.reduce.io.out1: 1
test XorReduce Success: 0 tests passed in 5 cycles in 0.018399 seconds 271.76 Hz
[info] [0.008] RAN 0 CYCLES PASSED
[success] Total time: 1 s, completed Jan 16, 2019 4:26:17 PM

I am not sure if this is the correct way to use treadle. But consider that specifying a random backend name would result in an error, I am assuming treadle support is already built into the iotesters.Driver.

Oh, BTW, by using the verilator backend, I get correct xorR behavior.

Regards,
Chen

[chick]

This is definitely a bug, interpreter is using the width of the output to construct and input mask. This is a bad idea.

[chick]

Thanks for the report, this seems to be a treadle only bug, I have created PR #79 to fix it.

[zinechant]

Thanks.
It seems the default backend for iotester.Driver is already treadle.
I confirm that firrtl backend actually delivers the correct behavior of xorR.