[PATCH] D130488: AMDGPU: Fix assertion when printing unreachable functions

[Matt Arsenault via Phabricator via llvm-commits]

arsenm created this revision.
arsenm added reviewers: jweightman, rampitec, AMDGPU.
Herald added subscribers: kosarev, jsilvanus, foad, kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl.
Herald added a project: All.
arsenm requested review of this revision.
Herald added a subscriber: wdng.
Herald added a project: LLVM.

Since 814a0abccefdd2e52b1b507f21ce842b689dbedd <https://reviews.llvm.org/rG814a0abccefdd2e52b1b507f21ce842b689dbedd>, this would break if we
had a function in the module that becomes dead in any codegen IR
pass. The function wasn't deleted since it was initially used in dead
code, but is detached from the call graph and doesn't appear in the PO
traversal. Do a second walk over the module to populate the resources
of any functions which weren't already processed.


https://reviews.llvm.org/D130488

Files:
  llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
  llvm/test/CodeGen/AMDGPU/resource-usage-dead-function.ll


Index: llvm/test/CodeGen/AMDGPU/resource-usage-dead-function.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/AMDGPU/resource-usage-dead-function.ll
@@ -0,0 +1,34 @@
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefix=GCN %s
+
+; Make sure there's no assertion when trying to report the resource
+; usage for a function which becomes dead during codegen.
+
+ at gv.fptr0 = external hidden unnamed_addr addrspace(4) constant void()*, align 4
+
+; GCN-LABEL: unreachable:
+; Function info:
+; codeLenInByte = 4
+define internal fastcc void @unreachable() {
+  %fptr = load void()*, void()* addrspace(4)* @gv.fptr0
+  call void %fptr()
+  unreachable
+}
+
+
+; GCN-LABEL: entry:
+; GCN-NOT: s_swappc_b64
+; GCN: s_endpgm
+
+; GCN: .amdhsa_private_segment_fixed_size 0
+; GCN: .amdhsa_uses_dynamic_stack 0
+define amdgpu_kernel void @entry() {
+bb0:
+  br i1 false, label %bb1, label %bb2
+
+bb1:
+  tail call fastcc void @unreachable()
+  unreachable
+
+bb2:
+  ret void
+}
Index: llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUResourceUsageAnalysis.cpp
@@ -125,6 +125,26 @@
     HasIndirectCall |= Info.HasIndirectCall;
   }
 
+  // It's possible we have unreachable functions in the module which weren't
+  // visited by the PO traversal. Make sure we have some resource counts to
+  // report.
+  for (const auto &IT : CG) {
+    const Function *F = IT.first;
+    if (!F || F->isDeclaration())
+      continue;
+
+    auto CI = CallGraphResourceInfo.insert(
+      std::make_pair(F, SIFunctionResourceInfo()));
+    if (CI.second) // Skip already visited functions
+      continue;
+
+    SIFunctionResourceInfo &Info = CI.first->second;
+    MachineFunction *MF = MMI.getMachineFunction(*F);
+    assert(MF && "function must have been generated already");
+    Info = analyzeResourceUsage(*MF, TM);
+    HasIndirectCall |= Info.HasIndirectCall;
+  }
+
   if (HasIndirectCall)
     propagateIndirectCallRegisterUsage();
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D130488.447335.patch
Type: text/x-patch
Size: 2180 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220725/33c921a1/attachment-0001.bin>