================
@@ -8030,7 +8077,431 @@ SDValue SystemZTargetLowering::combineBSWAP(
return SDValue();
}
+// Combine IPM sequence for flag output operands.
+static bool combineSRL_IPM_CCMask(SDValue &CCReg, int &CCValid, int &CCMask) {
+ // Convert CCVal to CCMask and update it along with CCValid.
+ const auto convertCCValToCCMask = [&CCMask, &CCValid](int CCVal) {
+ bool Invert = false;
+ if (CCMask == SystemZ::CCMASK_CMP_NE)
+ Invert = !Invert;
+ if (CCMask == SystemZ::CCMASK_CMP_EQ || CCMask == SystemZ::CCMASK_CMP_NE) {
+ CCMask = CCMaskForSystemZCCVal(CCVal);
+ if (Invert)
+ CCMask ^= SystemZ::CCMASK_ANY;
+ CCValid = SystemZ::CCMASK_ANY;
+ return true;
+ } else if (CCMask == SystemZ::CCMASK_CMP_LT) {
+ // CC in range [0, CCVal).
+ CCMask = ((~0U << (4 - CCVal)) & SystemZ::CCMASK_ANY);
+ CCValid = SystemZ::CCMASK_ANY;
+ return true;
+ } else if (CCMask == SystemZ::CCMASK_CMP_GT) {
+ // CC in range (CCVal, 3].
+ CCMask = (~(~0U << (3 - CCVal))) & SystemZ::CCMASK_ANY;
+ CCValid = SystemZ::CCMASK_ANY;
+ return true;
+ }
+ return false;
+ };
+ // Check (SRL (IPM (CC))) and update CCReg to combine.
+ const auto isSRL_IPM_CCSequence = [&CCReg](SDNode *N) {
+ if (!N || N->getOpcode() != ISD::SRL)
+ return false;
+ auto *SRLCount = dyn_cast<ConstantSDNode>(N->getOperand(1));
+ if (!SRLCount || SRLCount->getZExtValue() != SystemZ::IPM_CC)
+ return false;
+ auto *IPM = N->getOperand(0).getNode();
+ if (!IPM || IPM->getOpcode() != SystemZISD::IPM)
+ return false;
+ auto *IPMOp0 = IPM->getOperand(0).getNode();
+ if (!IPMOp0 || IPMOp0->getNumOperands() < 2)
+ return false;
+ auto *RN = dyn_cast<RegisterSDNode>(IPMOp0->getOperand(1));
+ // Check if operand 1 is SystemZ::CC. Also, it avoids srl/ipm/tbegin and
+ // srl/ipm/tend kind of sequences.
+ if (!RN || !RN->getReg().isPhysical() || RN->getReg() != SystemZ::CC)
+ return false;
+ // Return the updated CCReg link.
+ CCReg = IPM->getOperand(0);
+ return true;
+ };
+ // Check if N has SystemZ::CC operand.
+ const auto isCCOperand = [](SDNode *N) {
+ if (!N || N->getNumOperands() < 2)
+ return false;
+ auto *RN = dyn_cast<RegisterSDNode>(N->getOperand(1));
+ if (!RN || !RN->getReg().isPhysical() || RN->getReg() != SystemZ::CC)
+ return false;
+ return true;
+ };
+
+ auto *CCNode = CCReg.getNode();
+ if (!CCNode)
+ return false;
+
+ int RestoreCCValid = CCValid;
+ // Optimize (TM (IPM (CC)))
+ if (CCNode->getOpcode() == SystemZISD::TM) {
+ bool Invert = false;
+ if (CCMask == SystemZ::CCMASK_TM_SOME_1)
+ Invert = !Invert;
+ auto *N = CCNode->getOperand(0).getNode();
+ auto Shift = dyn_cast<ConstantSDNode>(CCNode->getOperand(1));
+ if (!N || !Shift)
+ return false;
+ if (N->getOpcode() == SystemZISD::IPM) {
+ auto ShiftVal = Shift->getZExtValue();
+ if (ShiftVal == (1 << SystemZ::IPM_CC))
+ CCMask = SystemZ::CCMASK_CMP_GE;
+ if (Invert)
+ CCMask ^= CCValid;
+ // Return the updated CCReg link.
+ CCReg = N->getOperand(0);
+ return true;
+ } else if (N->getOpcode() == ISD::XOR) {
+ // Optimize (TM (XOR (OP1 OP2))).
+ auto *XOROp1 = N->getOperand(0).getNode();
+ auto *XOROp2 = N->getOperand(1).getNode();
+ if (!XOROp1 || !XOROp2)
+ return false;
+ // OP1. (SELECT_CCMASK (ICMP (SRL (IPM (CC))))).
+ // OP2. (SRL (IPM (CC))).
+ if (XOROp1->getOpcode() == SystemZISD::SELECT_CCMASK &&
+ isSRL_IPM_CCSequence(XOROp2)) {
+ auto *CCValid1 = dyn_cast<ConstantSDNode>(XOROp1->getOperand(2));
+ auto *CCMask1 = dyn_cast<ConstantSDNode>(XOROp1->getOperand(3));
+ SDValue XORReg = XOROp1->getOperand(4);
+ if (!CCValid1 || !CCMask1)
+ return false;
+ int CCValidVal = CCValid1->getZExtValue();
+ int CCMaskVal = CCMask1->getZExtValue();
+ if (combineSRL_IPM_CCMask(XORReg, CCValidVal, CCMaskVal)) {
----------------
uweigand wrote:
Do we really need this recursion here? It seems to me there should be some
way to successively simplify subexpressions. The common code combiner logic
should iterate already, no?
https://github.com/llvm/llvm-project/pull/125970
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
