AMDGPU: Fix converting unanalyzable global loads to SMRD

Not all memory dependence queries succeed, so this needs to
be conservative if it fails.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307861 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/AMDGPUAnnotateUniformValues.cpp b/lib/Target/AMDGPU/AMDGPUAnnotateUniformValues.cpp
index 6f00286..ed53708 100644 (file)
--- a/lib/Target/AMDGPU/AMDGPUAnnotateUniformValues.cpp
+++ b/lib/Target/AMDGPU/AMDGPUAnnotateUniformValues.cpp
@@ -108,10 +108,11 @@ bool AMDGPUAnnotateUniformValues::isClobberedInFunction(LoadInst * Load) {
   DFS(Start, Checklist);
   for (auto &BB : Checklist) {
     BasicBlock::iterator StartIt = (!L && (BB == Load->getParent())) ?
-     BasicBlock::iterator(Load) : BB->end();
-     if (MDR->getPointerDependencyFrom(MemoryLocation(Ptr),
-       true, StartIt, BB, Load).isClobber())
-       return true;
+      BasicBlock::iterator(Load) : BB->end();
+    auto Q = MDR->getPointerDependencyFrom(MemoryLocation(Ptr), true,
+                                           StartIt, BB, Load);
+    if (Q.isClobber() || Q.isUnknown())
+      return true;
   }
   return false;
 }

diff --git a/test/CodeGen/AMDGPU/global-smrd-unknown.ll b/test/CodeGen/AMDGPU/global-smrd-unknown.ll
new file mode 100644 (file)
index 0000000..8a576e6
--- /dev/null
+++ b/test/CodeGen/AMDGPU/global-smrd-unknown.ll
@@ -0,0 +1,20 @@
+; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji  -memdep-block-scan-limit=1 -amdgpu-scalarize-global-loads -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
+
+; GCN-LABEL: {{^}}unknown_memdep_analysis:
+; GCN: flat_load_dword
+; GCN: flat_load_dword
+; GCN: flat_store_dword
+define amdgpu_kernel void @unknown_memdep_analysis(float addrspace(1)* nocapture readonly %arg) #0 {
+bb:
+  %tmp53 = load float, float addrspace(1)* undef, align 4
+  %tmp54 = getelementptr inbounds float, float addrspace(1)* %arg, i32 31
+  %tmp55 = load float, float addrspace(1)* %tmp54, align 4
+  %tmp56 = tail call float @llvm.fmuladd.f32(float undef, float %tmp53, float %tmp55)
+  store float %tmp56, float addrspace(1)* undef, align 4
+  ret void
+}
+
+declare float @llvm.fmuladd.f32(float, float, float) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readnone speculatable }