AMDGPU: Force skip branches over calls

Unfortunately the way SIInsertSkips works is backwards, and is
required for correctness. r338235 added handling of some special cases
where skipping is mandatory to avoid side effects if no lanes are
active. It conservatively handled asm correctly, but the same logic
needs to apply to calls.

Usually the call sequence code is larger than the skip threshold,
although the way the count is computed is really broken, so I'm not
sure if anything was likely to really hit this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@361202 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/SIInstrInfo.cpp b/lib/Target/AMDGPU/SIInstrInfo.cpp
index 0e87b4354f2bbcb88c6367421e0bb276e4dc3bc3..dd7f173e44d9a73daba5c76e92e742b7c22518ab 100644 (file)
--- a/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -2486,7 +2486,7 @@ bool SIInstrInfo::hasUnwantedEffectsWhenEXECEmpty(const MachineInstr &MI) const
       Opcode == AMDGPU::DS_ORDERED_COUNT)
     return true;
 
-  if (MI.isInlineAsm())
+  if (MI.isCall() || MI.isInlineAsm())
     return true; // conservative assumption
 
   // These are like SALU instructions in terms of effects, so it's questionable

diff --git a/test/CodeGen/AMDGPU/call-skip.ll b/test/CodeGen/AMDGPU/call-skip.ll
new file mode 100644 (file)
index 0000000..cd963df
--- /dev/null
+++ b/test/CodeGen/AMDGPU/call-skip.ll
@@ -0,0 +1,67 @@
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
+
+; A call should be skipped if all lanes are zero, since we don't know
+; what side effects should be avoided inside the call.
+define hidden void @func() #1 {
+  ret void
+}
+
+; GCN-LABEL: {{^}}if_call:
+; GCN: s_and_saveexec_b64
+; GCN-NEXT: ; mask branch [[END:BB[0-9]+_[0-9]+]]
+; GCN-NEXT: s_cbranch_execz [[END]]
+; GCN: s_swappc_b64
+; GCN: [[END]]:
+define void @if_call(i32 %flag) #0 {
+  %cc = icmp eq i32 %flag, 0
+  br i1 %cc, label %call, label %end
+
+call:
+  call void @func()
+  br label %end
+
+end:
+  ret void
+}
+
+; GCN-LABEL: {{^}}if_asm:
+; GCN: s_and_saveexec_b64
+; GCN-NEXT: ; mask branch [[END:BB[0-9]+_[0-9]+]]
+; GCN-NEXT: s_cbranch_execz [[END]]
+; GCN: ; sample asm
+; GCN: [[END]]:
+define void @if_asm(i32 %flag) #0 {
+  %cc = icmp eq i32 %flag, 0
+  br i1 %cc, label %call, label %end
+
+call:
+  call void asm sideeffect "; sample asm", ""()
+  br label %end
+
+end:
+  ret void
+}
+
+; GCN-LABEL: {{^}}if_call_kernel:
+; GCN: s_and_saveexec_b64
+; GCN-NEXT: ; mask branch [[END:BB[0-9]+_[0-9]+]]
+; GCN-NEXT: s_cbranch_execz [[END]]
+; GCN: s_swappc_b64
+define amdgpu_kernel void @if_call_kernel() #0 {
+  %id = call i32 @llvm.amdgcn.workitem.id.x()
+  %cc = icmp eq i32 %id, 0
+  br i1 %cc, label %call, label %end
+
+call:
+  call void @func()
+  br label %end
+
+end:
+  ret void
+}
+
+declare i32 @llvm.amdgcn.workitem.id.x() #2
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind noinline }
+attributes #2 = { nounwind readnone speculatable }