[PATCH] D123928: [NVPTX][tests] Do not run the tests which are not supported by nvptx

[Igor Chebykin via Phabricator via llvm-commits]

i-chebykin added a comment.

Updated the patch



================
Comment at: llvm/test/CodeGen/Generic/2007-12-17-InvokeAsm.ll:4
+; NVPTX does not support memory address operands
+; UNSUPPORTED: nvptx
+
----------------
tra wrote:
> This should probably be "REQUIRE: x86" instead, as it uses inline assembly with x86-specific constraints.
REQUIRES is matched against features only, 
so we need to use `REQUIRES: x86{{.*}}` or maybe `REQUIRES: {{.*}}x86{{.*}}`
Is it good?



================
Comment at: llvm/test/CodeGen/Generic/2010-ZeroSizedArg.ll:4
 
+; UNSUPPORTED: nvptx
+
----------------
tra wrote:
> What exactly is the problem with this case -- the datatype with 0-element fields or vararg function printf?
> If that's the latter, I think  `printf` could be removed as it's unused by the test.
NVPTX does not support empty type


================
Comment at: llvm/test/CodeGen/Generic/empty-load-store.ll:4
 
+; UNSUPPORTED: nvptx
+
----------------
tra wrote:
> What happens in this case? AFAICT, llc appears to work with this IR: https://godbolt.org/z/n1eK193br
> 
llc from assertions trunk crashed
my build with -DLLVM_ENABLE_ASSERTIONS=On crashed too


================
Comment at: llvm/test/CodeGen/Generic/zero-sized-array.ll:4
 
+; UNSUPPORTED: nvptx
+
----------------
tra wrote:
> This one compiles to PTX: https://godbolt.org/z/c3sE14j46
> What's the reason for disabling it? Something to do with `[0 x i8]` ?
> 
> I suspect ptxas would be unhappy about `.param .align 1 .b8 f4_param_0[0]`.
Again, there is a crash if assets are enabled:
`empty aggregate type not expected`


================
Comment at: llvm/test/CodeGen/MLRegalloc/default-eviction-advisor.ll:11-15
 define void @f2(i64 %lhs, i64 %rhs, i64* %addr) {
   %sum = add i64 %lhs, %rhs
   store i64 %sum, i64* %addr
   ret void
 }
----------------
tra wrote:
> Again, it would be good to explain what's going on here. Considering that we don't see the message regardless of what `-regalloc-enable-advisor` is set to, I guess it's just not enalbed for NVPTX.
Right,
regalloc-enable-advisor is not used for NVPTX


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D123928/new/

https://reviews.llvm.org/D123928