[GitHub] [incubator-tvm] abergeron opened a new issue #4997: [relay][VM] Miscompile with double-if

[GitBox]

abergeron opened a new issue #4997: [relay][VM] Miscompile with double-if
URL: https://github.com/apache/incubator-tvm/issues/4997
 
 
   This relay code:
   
   ```
   def @main(%b: bool) -> ((int32,), (int32,)) {
     let %v0: (int32,) = (0 /* ty=int32 */,);
     let %v1: (int32,) = (1 /* ty=int32 */,);
     let %v2: (int32,) = if (%b) {
       %v0
     } else {
       %v1
     };
     let %v3: (int32,) = if (%b) {
       %v1
     } else {
       %v0
     };
     (%v2, %v3)
   }
   ```
   
   Compiles to this in VM bytcode:
   
   ```
   main: 
   0: load_const $1 Const[0];
   1: alloc_data $2 tag(0) [$1];
   2: load_const $3 Const[1];
   3: alloc_data $4 tag(0) [$3];
   4: load_consti $5 1;
   5: if $0 $5 1 2;
   6: goto 2;
   7: move $2 $4;
   8: load_consti $6 1;
   9: if $0 $6 1 2;
   10: goto 2;
   11: move $4 $2;
   12: alloc_data $7 tag(0) [$2,$4];
   13: ret $7;
   ```
   
   If you pass `False` to the original function this makes the VM run both `7: 
move $2 $4` and `11: move $4 $2` which sets both of them to (1,) thus returning 
the wrong result.
   
   The debug interpreter doesn't have this problem.
   
   This script reproduces the problem:
   ```
   import tvm
   from tvm import relay
   
   
   def f(b):
       v0 = (0,)
       v1 = (1,)
       v2 = v0 if (b) else v1
       v3 = v1 if (b) else v0
       return (v2, v3)
   
   mod = tvm.IRModule({})
   
   b = relay.var('b')
   
   v0 = relay.var('v0')
   v1 = relay.var('v1')
   v2 = relay.var('v2')
   v3 = relay.var('v3')
   
   out = relay.Tuple([v2, v3])
   out = relay.Let(v3, relay.If(b, v1, v0), out)
   out = relay.Let(v2, relay.If(b, v0, v1), out)
   out = relay.Let(v1, relay.Tuple([relay.const(1)]), out)
   out = relay.Let(v0, relay.Tuple([relay.const(0)]), out)
   
   fn = relay.Function([b], out)
   
   mod['main'] = fn
   
   print(str(mod))
   
   ctx = tvm.runtime.ndarray.context('llvm', 0)
   
   debug = relay.create_executor(ctx=ctx, mod=mod, kind='debug')
   res = debug.evaluate()(False)
   res = ((int(res[0][0].asnumpy()),), (int(res[1][0].asnumpy()),))
   print("debug:", res)
   
   vm = relay.create_executor(ctx=ctx, mod=mod, kind='vm')
   res = vm.evaluate()(False)
   res = ((int(res[0][0].asnumpy()),), (int(res[1][0].asnumpy()),))
   print("vm:", res)
   ```
   
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


With regards,
Apache Git Services