R600/SI: Fix dependency between instruction writing M0 and S_SENDMSG on VI (v2)

This fixes a hang when using an empty geometry shader.

v2: - don't add s_nop when followed by s_waitcnt
    - comestic changes

Tested-by: Michel Dänzer <michel.daenzer@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@227986 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/SIInsertWaits.cpp b/lib/Target/R600/SIInsertWaits.cpp
index 181b11643bf3e87cdf5028a49ffc6fe42583bd58..50f20ac3619e5e6cdd34b2c27ce89e8372ce7e74 100644 (file)
--- a/lib/Target/R600/SIInsertWaits.cpp
+++ b/lib/Target/R600/SIInsertWaits.cpp
@@ -82,6 +82,8 @@ private:
   /// \brief Type of the last opcode.
   InstType LastOpcodeType;
 
+  bool LastInstWritesM0;
+
   /// \brief Get increment/decrement amount for this instruction.
   Counters getHwCounts(MachineInstr &MI);
 
@@ -106,6 +108,9 @@ private:
   /// \brief Resolve all operand dependencies to counter requirements
   Counters handleOperands(MachineInstr &MI);
 
+  /// \brief Insert S_NOP between an instruction writing M0 and S_SENDMSG.
+  void handleSendMsg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I);
+
 public:
   SIInsertWaits(TargetMachine &tm) :
     MachineFunctionPass(ID),
@@ -269,6 +274,7 @@ void SIInsertWaits::pushInstruction(MachineBasicBlock &MBB,
       // Insert a NOP to break the clause.
       BuildMI(MBB, I, DebugLoc(), TII->get(AMDGPU::S_NOP))
           .addImm(0);
+      LastInstWritesM0 = false;
     }
 
     if (TII->isSMRD(I->getOpcode()))
@@ -362,6 +368,7 @@ bool SIInsertWaits::insertWait(MachineBasicBlock &MBB,
                   ((Counts.Named.LGKM & 0x7) << 8));
 
   LastOpcodeType = OTHER;
+  LastInstWritesM0 = false;
   return true;
 }
 
@@ -403,6 +410,30 @@ Counters SIInsertWaits::handleOperands(MachineInstr &MI) {
   return Result;
 }
 
+void SIInsertWaits::handleSendMsg(MachineBasicBlock &MBB,
+                                  MachineBasicBlock::iterator I) {
+  if (TRI->ST.getGeneration() < AMDGPUSubtarget::VOLCANIC_ISLANDS)
+    return;
+
+  // There must be "S_NOP 0" between an instruction writing M0 and S_SENDMSG.
+  if (LastInstWritesM0 && I->getOpcode() == AMDGPU::S_SENDMSG) {
+    BuildMI(MBB, I, DebugLoc(), TII->get(AMDGPU::S_NOP)).addImm(0);
+    LastInstWritesM0 = false;
+    return;
+  }
+
+  // Set whether this instruction sets M0
+  LastInstWritesM0 = false;
+
+  unsigned NumOperands = I->getNumOperands();
+  for (unsigned i = 0; i < NumOperands; i++) {
+    const MachineOperand &Op = I->getOperand(i);
+
+    if (Op.isReg() && Op.isDef() && Op.getReg() == AMDGPU::M0)
+      LastInstWritesM0 = true;
+  }
+}
+
 // FIXME: Insert waits listed in Table 4.2 "Required User-Inserted Wait States"
 // around other non-memory instructions.
 bool SIInsertWaits::runOnMachineFunction(MachineFunction &MF) {
@@ -417,6 +448,7 @@ bool SIInsertWaits::runOnMachineFunction(MachineFunction &MF) {
   WaitedOn = ZeroCounts;
   LastIssued = ZeroCounts;
   LastOpcodeType = OTHER;
+  LastInstWritesM0 = false;
 
   memset(&UsedRegs, 0, sizeof(UsedRegs));
   memset(&DefinedRegs, 0, sizeof(DefinedRegs));
@@ -433,7 +465,9 @@ bool SIInsertWaits::runOnMachineFunction(MachineFunction &MF) {
         Changes |= insertWait(MBB, I, LastIssued);
       else
         Changes |= insertWait(MBB, I, handleOperands(*I));
+
       pushInstruction(MBB, I);
+      handleSendMsg(MBB, I);
     }
 
     // Wait for everything at the end of the MBB

diff --git a/test/CodeGen/R600/llvm.SI.sendmsg-m0.ll b/test/CodeGen/R600/llvm.SI.sendmsg-m0.ll
new file mode 100644 (file)
index 0000000..2198590
--- /dev/null
+++ b/test/CodeGen/R600/llvm.SI.sendmsg-m0.ll
@@ -0,0 +1,20 @@
+;RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=BOTH %s
+;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck --check-prefix=VI --check-prefix=BOTH %s
+
+; BOTH-LABEL: {{^}}main:
+; BOTH: s_mov_b32 m0, s0
+; VI-NEXT: s_nop 0
+; BOTH-NEXT: s_sendmsg Gs_done(nop)
+; BOTH-NEXT: s_endpgm
+
+define void @main(i32 inreg %a) #0 {
+main_body:
+  call void @llvm.SI.sendmsg(i32 3, i32 %a)
+  ret void
+}
+
+; Function Attrs: nounwind
+declare void @llvm.SI.sendmsg(i32, i32) #1
+
+attributes #0 = { "ShaderType"="2" "unsafe-fp-math"="true" }
+attributes #1 = { nounwind }