AMDGPU: readlane/writelane do not read exec

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@273525 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/SIInstrInfo.cpp b/lib/Target/AMDGPU/SIInstrInfo.cpp
index 05c73f9..d23b14c 100644 (file)
--- a/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -1631,6 +1631,29 @@ static unsigned findImplicitSGPRRead(const MachineInstr &MI) {
   return AMDGPU::NoRegister;
 }
 
+static bool shouldReadExec(const MachineInstr &MI) {
+  if (SIInstrInfo::isVALU(MI)) {
+    switch (MI.getOpcode()) {
+    case AMDGPU::V_READLANE_B32:
+    case AMDGPU::V_READLANE_B32_si:
+    case AMDGPU::V_READLANE_B32_vi:
+    case AMDGPU::V_WRITELANE_B32:
+    case AMDGPU::V_WRITELANE_B32_si:
+    case AMDGPU::V_WRITELANE_B32_vi:
+      return false;
+    }
+
+    return true;
+  }
+
+  if (SIInstrInfo::isGenericOpcode(MI.getOpcode()) ||
+      SIInstrInfo::isSALU(MI) ||
+      SIInstrInfo::isSMRD(MI))
+    return false;
+
+  return true;
+}
+
 bool SIInstrInfo::verifyInstruction(const MachineInstr *MI,
                                     StringRef &ErrInfo) const {
   uint16_t Opcode = MI->getOpcode();
@@ -1751,7 +1774,7 @@ bool SIInstrInfo::verifyInstruction(const MachineInstr *MI,
 
   // Make sure we aren't losing exec uses in the td files. This mostly requires
   // being careful when using let Uses to try to add other use registers.
-  if (!isGenericOpcode(Opcode) && !isSALU(Opcode) && !isSMRD(Opcode)) {
+  if (shouldReadExec(*MI)) {
     if (!MI->hasRegisterImplicitUseOperand(AMDGPU::EXEC)) {
       ErrInfo = "VALU instruction does not implicitly read exec mask";
       return false;

diff --git a/lib/Target/AMDGPU/SIInstructions.td b/lib/Target/AMDGPU/SIInstructions.td
index 23717da..db1038d 100644 (file)
--- a/lib/Target/AMDGPU/SIInstructions.td
+++ b/lib/Target/AMDGPU/SIInstructions.td
@@ -1586,7 +1586,8 @@ defm V_SUBBREV_U32 : VOP2bInst <vop2<0x2a, 0x1e>, "v_subbrev_u32",
 
 } // End isCommutable = 1
 
-let isConvergent = 1 in {
+// These are special and do not read the exec mask.
+let isConvergent = 1, Uses = []<Register> in {
 
 defm V_READLANE_B32 : VOP2SI_3VI_m <
   vop3 <0x001, 0x289>,