[PATCH] R600/SI: Fix crash in SIAnnotateControlFlow

Christian König christian.koenig at amd.com
Mon Feb 9 02:11:14 PST 2015


Am 06.02.2015 um 20:22 schrieb Tom Stellard:
> On Fri, Feb 06, 2015 at 11:06:59AM -0800, Matt Arsenault wrote:
>> On 02/06/2015 07:41 AM, Tom Stellard wrote:
>>> When analyzing PHI loop conditions, we should handle undef values the same
>>> as constants.
>>> ---
>>>   lib/Target/R600/SIAnnotateControlFlow.cpp |  2 +-
>>>   test/CodeGen/R600/si-annotate-cf-crash.ll | 62 +++++++++++++++++++++++++++++++
>>>   2 files changed, 63 insertions(+), 1 deletion(-)
>>>   create mode 100644 test/CodeGen/R600/si-annotate-cf-crash.ll
>> I don't see where the undef comes from in the test? I think the test
>> name should also include undef
> The undef is added by the structurizecfg pass.

Which sounds like there is a case where the control flow isn't determined.

Haven't had a deeper look, so is that correct? Or is this just another 
symptom of a bug in the structurizer?

Regards,
Christian.

>
> -Tom
>
>>> diff --git a/lib/Target/R600/SIAnnotateControlFlow.cpp b/lib/Target/R600/SIAnnotateControlFlow.cpp
>>> index 79f6532..a780171 100644
>>> --- a/lib/Target/R600/SIAnnotateControlFlow.cpp
>>> +++ b/lib/Target/R600/SIAnnotateControlFlow.cpp
>>> @@ -217,7 +217,7 @@ Value *SIAnnotateControlFlow::handleLoopCondition(Value *Cond, PHINode *Broken)
>>>       for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
>>>         Value *Incoming = Phi->getIncomingValue(i);
>>>         BasicBlock *From = Phi->getIncomingBlock(i);
>>> -      if (isa<ConstantInt>(Incoming)) {
>>> +      if (isa<ConstantInt>(Incoming) || isa<UndefValue>(Incoming)) {
>>>           NewPhi->addIncoming(Broken, From);
>>>           continue;
>>>         }
>>> diff --git a/test/CodeGen/R600/si-annotate-cf-crash.ll b/test/CodeGen/R600/si-annotate-cf-crash.ll
>>> new file mode 100644
>>> index 0000000..a9f1835
>>> --- /dev/null
>>> +++ b/test/CodeGen/R600/si-annotate-cf-crash.ll
>>> @@ -0,0 +1,62 @@
>>> +; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs-asm < %s | FileCheck %s
>>> +; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck %s
>>> +
>>> +; CHECK-LABEL: {{^}}test:
>>> +; CHECK: s_endpgm
>>> +define void @test(float addrspace(1)* nocapture %arg, i32 %arg1, float addrspace(1)* nocapture readonly %arg2, float %arg3) {
>>> +bb:
>>> +  br label %bb3
>>> +
>>> +bb3:                                              ; preds = %bb36, %bb
>>> +  %tmp = phi i64 [ %tmp39, %bb36 ], [ 1, %bb ]
>>> +  %tmp4 = phi float [ %tmp38, %bb36 ], [ 5.000000e+01, %bb ]
>>> +  %tmp5 = phi i32 [ %tmp28, %bb36 ], [ 0, %bb ]
>>> +  %tmp6 = phi i32 [ %tmp37, %bb36 ], [ 0, %bb ]
>>> +  %tmp7 = trunc i64 %tmp to i32
>>> +  %tmp8 = icmp eq i32 %tmp7, %arg1
>>> +  br i1 %tmp8, label %bb26, label %bb9
>>> +
>>> +bb9:                                              ; preds = %bb18, %bb3
>>> +  %tmp10 = phi i32 [ %tmp20, %bb18 ], [ %tmp7, %bb3 ]
>>> +  %tmp11 = phi i64 [ %tmp19, %bb18 ], [ %tmp, %bb3 ]
>>> +  %tmp12 = getelementptr inbounds float addrspace(1)* %arg2, i64 %tmp11
>>> +  %tmp13 = load float addrspace(1)* %tmp12, align 4
>>> +  %tmp17 = fcmp une float %tmp13, %arg3
>>> +  br i1 %tmp17, label %bb22, label %bb18
>>> +
>>> +bb18:                                             ; preds = %bb9
>>> +  %tmp19 = add nuw nsw i64 %tmp11, 1
>>> +  %tmp20 = trunc i64 %tmp19 to i32
>>> +  %tmp21 = icmp eq i32 %tmp20, %arg1
>>> +  br i1 %tmp21, label %bb22, label %bb9
>>> +
>>> +bb22:                                             ; preds = %bb18, %bb9
>>> +  %tmp23 = phi i1 [ false, %bb9 ], [ true, %bb18 ]
>>> +  %tmp24 = phi i32 [ %tmp10, %bb9 ], [ %arg1, %bb18 ]
>>> +  %tmp25 = phi i64 [ %tmp11, %bb9 ], [ %tmp19, %bb18 ]
>>> +  br label %bb26
>>> +
>>> +bb26:                                             ; preds = %bb22, %bb3
>>> +  %tmp27 = phi i1 [ true, %bb3 ], [ %tmp23, %bb22 ]
>>> +  %tmp28 = phi i32 [ %arg1, %bb3 ], [ %tmp24, %bb22 ]
>>> +  %tmp29 = phi i64 [ %tmp, %bb3 ], [ %tmp25, %bb22 ]
>>> +  %tmp30 = sub nsw i32 %tmp28, %tmp5
>>> +  %tmp31 = icmp slt i32 %tmp6, %tmp30
>>> +  br i1 %tmp31, label %bb32, label %bb36
>>> +
>>> +bb32:                                             ; preds = %bb26
>>> +  %tmp33 = add nsw i64 %tmp29, -1
>>> +  %tmp34 = getelementptr inbounds float addrspace(1)* %arg2, i64 %tmp33
>>> +  %tmp35 = load float addrspace(1)* %tmp34, align 4
>>> +  br label %bb36
>>> +
>>> +bb36:                                             ; preds = %bb32, %bb26
>>> +  %tmp37 = phi i32 [ %tmp30, %bb32 ], [ %tmp6, %bb26 ]
>>> +  %tmp38 = phi float [ %tmp35, %bb32 ], [ %tmp4, %bb26 ]
>>> +  %tmp39 = add nuw nsw i64 %tmp29, 1
>>> +  br i1 %tmp27, label %bb40, label %bb3
>>> +
>>> +bb40:                                             ; preds = %bb36
>>> +  store float %tmp38, float addrspace(1)* %arg, align 4
>>> +  ret void
>>> +}
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits




More information about the llvm-commits mailing list