AMDGPU: Force skip branches over calls

[arsenm]

Summary

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.

[rampitec]

LGTM

[arsenm]

r361202

lib/Target/AMDGPU/SIInstrInfo.cpp

bool SIInstrInfo::hasUnwantedEffectsWhenEXECEmpty(const MachineInstr &MI) const {
   // 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 (Opcode == AMDGPU::S_SENDMSG || Opcode == AMDGPU::S_SENDMSGHALT ||
       Opcode == AMDGPU::EXP || Opcode == AMDGPU::EXP_DONE ||
       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
   // whether we should return true for those.
   //
   // However, executing them with EXEC = 0 causes them to operate on undefined
   // data, which we avoid by returning true here.
   if (Opcode == AMDGPU::V_READFIRSTLANE_B32 || Opcode == AMDGPU::V_READLANE_B32)

test/CodeGen/AMDGPU/call-skip.ll

+; 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 }