Index: lib/Target/AMDGPU/SIInsertSkips.cpp
===================================================================
--- lib/Target/AMDGPU/SIInsertSkips.cpp
+++ lib/Target/AMDGPU/SIInsertSkips.cpp
@@ -143,6 +143,18 @@
         return true;
       }
 
+      // These instructions cause shader I/O that may cause hardware lockups
+      // when executed with an empty EXEC mask.
+      //
+      // Note: exp with VM = DONE = 0 is automatically skipped by hardware when
+      //       EXEC = 0, but checking for that case here seems not worth it
+      //       given the typical code patterns.
+      if ((I->getOpcode() == AMDGPU::S_SENDMSG) ||
+          (I->getOpcode() == AMDGPU::S_SENDMSGHALT) ||
+          (I->getOpcode() == AMDGPU::EXP) ||
+          (I->getOpcode() == AMDGPU::EXP_DONE))
+        return true;
+
       if (I->isInlineAsm()) {
         const MCAsmInfo *MAI = MF->getTarget().getMCAsmInfo();
         const char *AsmStr = I->getOperand(0).getSymbolName();
Index: test/CodeGen/AMDGPU/llvm.amdgcn.exp.ll
===================================================================
--- test/CodeGen/AMDGPU/llvm.amdgcn.exp.ll
+++ test/CodeGen/AMDGPU/llvm.amdgcn.exp.ll
@@ -480,5 +480,65 @@
   ret void
 }
 
+; GCN-LABEL: {{^}}test_if_export_f32:
+; GCN: s_cbranch_execz
+; GCN: exp
+define amdgpu_ps void @test_if_export_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+  %cc = icmp eq i32 %flag, 0
+  br i1 %cc, label %end, label %exp
+
+exp:
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 false, i1 false)
+  br label %end
+
+end:
+  ret void
+}
+
+; GCN-LABEL: {{^}}test_if_export_vm_f32:
+; GCN: s_cbranch_execz
+; GCN: exp
+define amdgpu_ps void @test_if_export_vm_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+  %cc = icmp eq i32 %flag, 0
+  br i1 %cc, label %end, label %exp
+
+exp:
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 false, i1 true)
+  br label %end
+
+end:
+  ret void
+}
+
+; GCN-LABEL: {{^}}test_if_export_done_f32:
+; GCN: s_cbranch_execz
+; GCN: exp
+define amdgpu_ps void @test_if_export_done_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+  %cc = icmp eq i32 %flag, 0
+  br i1 %cc, label %end, label %exp
+
+exp:
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 true, i1 false)
+  br label %end
+
+end:
+  ret void
+}
+
+; GCN-LABEL: {{^}}test_if_export_vm_done_f32:
+; GCN: s_cbranch_execz
+; GCN: exp
+define amdgpu_ps void @test_if_export_vm_done_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+  %cc = icmp eq i32 %flag, 0
+  br i1 %cc, label %end, label %exp
+
+exp:
+  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 true, i1 true)
+  br label %end
+
+end:
+  ret void
+}
+
 attributes #0 = { nounwind }
 attributes #1 = { nounwind inaccessiblememonly }
Index: test/CodeGen/AMDGPU/llvm.amdgcn.sendmsg.ll
===================================================================
--- test/CodeGen/AMDGPU/llvm.amdgcn.sendmsg.ll
+++ test/CodeGen/AMDGPU/llvm.amdgcn.sendmsg.ll
@@ -136,6 +136,21 @@
   ret void
 }
 
+; GCN-LABEL: {{^}}if_sendmsg:
+; GCN: s_cbranch_execz
+; GCN: s_sendmsg sendmsg(MSG_GS_DONE, GS_OP_NOP)
+define amdgpu_gs void @if_sendmsg(i32 %flag) #0 {
+  %cc = icmp eq i32 %flag, 0
+  br i1 %cc, label %sendmsg, label %end
+
+sendmsg:
+  call void @llvm.amdgcn.s.sendmsg(i32 3, i32 0)
+  br label %end
+
+end:
+  ret void
+}
+
 declare void @llvm.amdgcn.s.sendmsg(i32, i32) #0
 declare void @llvm.amdgcn.s.sendmsghalt(i32, i32) #0