[llvm] r249073 - AMDGPU/SI: Remove assert from AMDGPUOpenCLImageTypeLowering pass

Tom Stellard via llvm-commits llvm-commits at lists.llvm.org
Thu Oct 1 14:16:06 PDT 2015


Author: tstellar
Date: Thu Oct  1 16:16:05 2015
New Revision: 249073

URL: http://llvm.org/viewvc/llvm-project?rev=249073&view=rev
Log:
AMDGPU/SI: Remove assert from AMDGPUOpenCLImageTypeLowering pass

Summary:
Instead of asserting when the kernel metadata is different than we expect,
we should just skip lowering that function.  This fixes assertion
failures with OpenCL argument metadata from older LLVM releases.

Reviewers: arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D13356

Added:
    llvm/trunk/test/CodeGen/AMDGPU/opencl-image-metadata.ll
Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPUOpenCLImageTypeLoweringPass.cpp

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUOpenCLImageTypeLoweringPass.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUOpenCLImageTypeLoweringPass.cpp?rev=249073&r1=249072&r2=249073&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUOpenCLImageTypeLoweringPass.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUOpenCLImageTypeLoweringPass.cpp Thu Oct  1 16:16:05 2015
@@ -94,8 +94,12 @@ GetFunctionFromMDNode(MDNode *Node) {
       return nullptr;
     if (!ArgNode->getOperand(0))
       return nullptr;
-    assert(cast<MDString>(ArgNode->getOperand(0))->getString() ==
-           KernelArgMDNodeNames[i] && "Wrong kernel arg metadata name");
+
+    // FIXME: It should be possible to do image lowering when some metadata
+    // args missing or not in the expected order.
+    MDString *StringNode = dyn_cast<MDString>(ArgNode->getOperand(0));
+    if (!StringNode || StringNode->getString() != KernelArgMDNodeNames[i])
+      return nullptr;
   }
 
   return F;

Added: llvm/trunk/test/CodeGen/AMDGPU/opencl-image-metadata.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/opencl-image-metadata.ll?rev=249073&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/opencl-image-metadata.ll (added)
+++ llvm/trunk/test/CodeGen/AMDGPU/opencl-image-metadata.ll Thu Oct  1 16:16:05 2015
@@ -0,0 +1,24 @@
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG --check-prefix=FUNC %s
+; RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=FUNC %s
+
+; Make sure the OpenCL Image lowering pass doesn't crash when argument metadata
+; is not in expected order.
+
+; EG: CF_END
+; SI: s_endpgm
+define void @kernel(i32 addrspace(1)* %out) {
+entry:
+  store i32 0, i32 addrspace(1)* %out
+  ret void
+}
+
+attributes #3 = { nounwind }
+
+!opencl.kernels = !{!0}
+
+!0 = !{void (i32 addrspace(1)*)* @kernel, !1, !2, !3, !4, !5}
+!1 = !{!"kernel_arg_addr_space", i32 0}
+!2 = !{!"kernel_arg_access_qual", !"none"}
+!3 = !{!"kernel_arg_type", !"int*"}
+!4 = !{!"kernel_arg_type_qual", !""}
+!5 = !{!"kernel_arg_name", !""}




More information about the llvm-commits mailing list