[CODEGEN] LLVM PReLU Error #2364

Activity

1. tqchen

   changed the title [-]Failed to export mxnet models with PRelu operator on arm devices.[/-] [+][CODEGEN] LLVM PReLU Error[/+] on Jan 4, 2019
2. tqchen

   added

   status: help wanted

   on Jan 4, 2019
3. tqchen commented on Jan 4, 2019

   

   tqchen

   on Jan 4, 2019

   Last edited by tqchen

   Member

   More actions

   Thanks for reporting this issue. It would be helpful to narrow the case down, for example, try to compose a single layer Prelu and see if the error still occurs, this will help to give evidence to other developers.

   Likely this problem is due to LLVM or low-level lowering type check issues. cc @masahi @grwlf @ZihengJiang @denis0x0Dwho might be familiar with this part of the code
4. denis0x0D commented on Jan 6, 2019

   

   denis0x0D

   on Jan 6, 2019

   Last edited by denis0x0D

   Contributor

   More actions

   @nttstar can you please add simple test case, which fails. In this case it would be easy to reduce problem in llvm codegen part.
   Thanks.
5. FrozenGene commented on Jan 7, 2019

   

   FrozenGene

   on Jan 7, 2019

   Last edited by FrozenGene

   Member

   More actions

   I can reproduce this issue after 100 trials auto tvm tunning on ARM CPU, but can not reproduce without auto tvm tunning.

   input_sym = nnvm.sym.Variable('data')
   alpha = nnvm.sym.Variable('alpha')
   net = nnvm.sym.conv2d(input_sym, channels=3, kernel_size=(1,1), padding=(0,0), strides=(1, 1), use_bias=False)
   net = nnvm.sym.prelu(data=net, alpha=alpha, axis=1)
   params = {'alpha': tvm.nd.array(np.ones((3,), dtype=dtype))}

   The issue is that when we apply schedule of conv2d combined prelu we emit wrong LLVM IR. For example, we would emit this kind of LLVM IR:

   %72 = fcmp ogt <16 x float> %68, zeroinitializer
   br <16 x i1> %72, label %if_end, label %if_else

   However, br instruction can not accept <N x i1> type, which only accept i1 type. Please see this link: https://llvm.org/docs/LangRef.html#br-instruction

   So, we can add condition type check of UnsafeSelectRewriter to avoid this issue. i.e. to make sure condition type is just i1, not other things else.

   I can help make a PR to resolve this issue soon.
6. denis0x0D commented on Jan 7, 2019

   

   denis0x0D

   on Jan 7, 2019

   Contributor

   More actions

   @FrozenGene, thanks for the answer, good to know you've resolved the issue.
7. merrymercy

   closed this as completedon Jan 8, 2019