[llvm] [AMDGPU] Implement vop3p complex pattern optmization for gisel (PR #130234)

[Juan Manuel Martinez Caamaño via llvm-commits]

================
@@ -4282,44 +4282,362 @@ AMDGPUInstructionSelector::selectVOP3NoMods(MachineOperand &Root) const {
   }};
 }
 
-std::pair<Register, unsigned>
-AMDGPUInstructionSelector::selectVOP3PModsImpl(
-  Register Src, const MachineRegisterInfo &MRI, bool IsDOT) const {
-  unsigned Mods = 0;
-  MachineInstr *MI = MRI.getVRegDef(Src);
+enum srcStatus {
+  IS_SAME,
+  IS_UPPER_HALF,
+  IS_LOWER_HALF,
+  IS_NEG,
+  IS_UPPER_HALF_NEG,
+  IS_LOWER_HALF_NEG,
+  LAST_STAT = IS_LOWER_HALF_NEG
+};
+
+static bool isTruncHalf(const MachineInstr *MI,
+                        const MachineRegisterInfo &MRI) {
+  assert(MI->getOpcode() == AMDGPU::G_TRUNC);
+  unsigned dstSize = MRI.getType(MI->getOperand(0).getReg()).getSizeInBits();
+  unsigned srcSize = MRI.getType(MI->getOperand(1).getReg()).getSizeInBits();
+  return dstSize * 2 == srcSize;
+}
+
+static bool isLshrHalf(const MachineInstr *MI, const MachineRegisterInfo &MRI) {
+  assert(MI->getOpcode() == AMDGPU::G_LSHR);
+  Register ShiftSrc;
+  std::optional<ValueAndVReg> ShiftAmt;
+  if (mi_match(MI->getOperand(0).getReg(), MRI,
+               m_GLShr(m_Reg(ShiftSrc), m_GCst(ShiftAmt)))) {
+    unsigned srcSize = MRI.getType(MI->getOperand(1).getReg()).getSizeInBits();
+    unsigned shift = ShiftAmt->Value.getZExtValue();
+    return shift * 2 == srcSize;
+  }
+  return false;
+}
 
-  if (MI->getOpcode() == AMDGPU::G_FNEG &&
-      // It's possible to see an f32 fneg here, but unlikely.
-      // TODO: Treat f32 fneg as only high bit.
-      MRI.getType(Src) == LLT::fixed_vector(2, 16)) {
-    Mods ^= (SISrcMods::NEG | SISrcMods::NEG_HI);
-    Src = MI->getOperand(1).getReg();
-    MI = MRI.getVRegDef(Src);
+static bool isShlHalf(const MachineInstr *MI, const MachineRegisterInfo &MRI) {
+  assert(MI->getOpcode() == AMDGPU::G_SHL);
+  Register ShiftSrc;
+  std::optional<ValueAndVReg> ShiftAmt;
+  if (mi_match(MI->getOperand(0).getReg(), MRI,
+               m_GShl(m_Reg(ShiftSrc), m_GCst(ShiftAmt)))) {
+    unsigned srcSize = MRI.getType(MI->getOperand(1).getReg()).getSizeInBits();
+    unsigned shift = ShiftAmt->Value.getZExtValue();
+    return shift * 2 == srcSize;
+  }
+  return false;
+}
+
+static bool retOpStat(const MachineOperand *Op, int stat,
+                      std::pair<const MachineOperand *, int> &curr) {
+  if ((Op->isReg() && !(Op->getReg().isPhysical())) || Op->isImm() ||
+      Op->isCImm() || Op->isFPImm()) {
+    curr = {Op, stat};
+    return true;
+  }
+  return false;
+}
+
+static bool calcNextStatus(std::pair<const MachineOperand *, int> &curr,
+                           const MachineRegisterInfo &MRI) {
+  if (!curr.first->isReg()) {
+    return false;
+  }
+  const MachineInstr *MI = nullptr;
+
+  if (!curr.first->isDef()) {
+    MI = MRI.getVRegDef(curr.first->getReg());
+  } else {
+    MI = curr.first->getParent();
+  }
+  if (!MI) {
+    return false;
+  }
+
+  unsigned Opc = MI->getOpcode();
+
+  // Handle general Opc cases
+  switch (Opc) {
+  case AMDGPU::G_BITCAST:
+  case AMDGPU::G_CONSTANT:
+  case AMDGPU::G_FCONSTANT:
+  case AMDGPU::COPY:
+    return retOpStat(&MI->getOperand(1), curr.second, curr);
+  case AMDGPU::G_FNEG:
+    // XXXX + 3 = XXXX_NEG, (XXXX_NEG + 3) mod 3 = XXXX
+    return retOpStat(&MI->getOperand(1),
+                     (curr.second + ((LAST_STAT + 1) / 2)) % (LAST_STAT + 1),
----------------
jmmartinez wrote:

Using the `+3 mod 6` was clever; but to be honest, I'd rather read:

```cpp
srcStatus getNegStatus(srcStatus S) {
 switch(S) {
      case IS_SAME: return IS_NEG;
      case IS_UPPER_HALF: return IS_UPPER_HALF_NEG;
      case IS_LOWER_HALF: return IS_LOWER_HALF_NEG;
      case IS_NEG: return IS_SAME;
      case IS_UPPER_HALF_NEG: return IS_UPPER_HALF;
      case IS_LOWER_HALF_NEG: return IS_LOWER_HALF;
 }
}
```

https://github.com/llvm/llvm-project/pull/130234