[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbt asm instructions

lewis-revill · zmodem · commit 1daf1144ad1e · 2020-07-27T13:07:37.000+02:00
This patch provides optimization of bit manipulation operations by enabling the +experimental-b target feature. It adds matching of single block patterns of instructions to specific bit-manip instructions from the ternary subset (zbt subextension) of the experimental B extension of RISC-V. It adds also the correspondent codegen tests. This patch is based on Claire Wolf's proposal for the bit manipulation extension of RISCV: https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf Differential Revision: https://reviews.llvm.org/D79875 (cherry picked from commit c9c955a)
diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
@@ -459,6 +459,55 @@ bool RISCVDAGToDAGISel::SelectRORIW(SDValue N, SDValue &RS1, SDValue &Shamt) {
   return false;
 }
 
+// Check that it is a FSRIW (i32 Funnel Shift Right Immediate on RV64).
+// We first check that it is the right node tree:
+//
+//  (SIGN_EXTEND_INREG (OR (SHL (AsserSext RS1, i32), VC2),
+//                         (SRL (AND (AssertSext RS2, i32), VC3), VC1)))
+//
+// Then we check that the constant operands respect these constraints:
+//
+// VC2 == 32 - VC1
+// VC3 == maskLeadingOnes<uint32_t>(VC2)
+//
+// being VC1 the Shamt we need, VC2 the complementary of Shamt over 32
+// and VC3 a 32 bit mask of (32 - VC1) leading ones.
+
+bool RISCVDAGToDAGISel::SelectFSRIW(SDValue N, SDValue &RS1, SDValue &RS2,
+                                    SDValue &Shamt) {
+  if (N.getOpcode() == ISD::SIGN_EXTEND_INREG &&
+      Subtarget->getXLenVT() == MVT::i64 &&
+      cast<VTSDNode>(N.getOperand(1))->getVT() == MVT::i32) {
+    if (N.getOperand(0).getOpcode() == ISD::OR) {
+      SDValue Or = N.getOperand(0);
+      if (Or.getOperand(0).getOpcode() == ISD::SHL &&
+          Or.getOperand(1).getOpcode() == ISD::SRL) {
+        SDValue Shl = Or.getOperand(0);
+        SDValue Srl = Or.getOperand(1);
+        if (Srl.getOperand(0).getOpcode() == ISD::AND) {
+          SDValue And = Srl.getOperand(0);
+          if (isa<ConstantSDNode>(Srl.getOperand(1)) &&
+              isa<ConstantSDNode>(Shl.getOperand(1)) &&
+              isa<ConstantSDNode>(And.getOperand(1))) {
+            uint32_t VC1 = Srl.getConstantOperandVal(1);
+            uint32_t VC2 = Shl.getConstantOperandVal(1);
+            uint32_t VC3 = And.getConstantOperandVal(1);
+            if (VC2 == (32 - VC1) &&
+                VC3 == maskLeadingOnes<uint32_t>(VC2)) {
+              RS1 = Shl.getOperand(0);
+              RS2 = And.getOperand(0);
+              Shamt = CurDAG->getTargetConstant(VC1, SDLoc(N),
+                                              Srl.getOperand(1).getValueType());
+              return true;
+            }
+          }
+        }
+      }
+    }
+  }
+  return false;
+}
+
 // Merge an ADDI into the offset of a load/store instruction where possible.
 // (load (addi base, off1), off2) -> (load base, off1+off2)
 // (store val, (addi base, off1), off2) -> (store val, base, off1+off2)
diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.h b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.h
@@ -52,6 +52,7 @@ class RISCVDAGToDAGISel : public SelectionDAGISel {
   bool SelectSLOIW(SDValue N, SDValue &RS1, SDValue &Shamt);
   bool SelectSROIW(SDValue N, SDValue &RS1, SDValue &Shamt);
   bool SelectRORIW(SDValue N, SDValue &RS1, SDValue &Shamt);
+  bool SelectFSRIW(SDValue N, SDValue &RS1, SDValue &RS2, SDValue &Shamt);
 
 // Include the pieces autogenerated from the target description.
 #include "RISCVGenDAGISel.inc"
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -166,6 +166,11 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
   if (Subtarget.hasStdExtZbp())
     setOperationAction(ISD::BITREVERSE, XLenVT, Legal);
 
+  if (Subtarget.hasStdExtZbt()) {
+    setOperationAction(ISD::FSHL, XLenVT, Legal);
+    setOperationAction(ISD::FSHR, XLenVT, Legal);
+  }
+
   ISD::CondCode FPCCToExtend[] = {
       ISD::SETOGT, ISD::SETOGE, ISD::SETONE, ISD::SETUEQ, ISD::SETUGT,
       ISD::SETUGE, ISD::SETULT, ISD::SETULE, ISD::SETUNE, ISD::SETGT,
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoB.td b/llvm/lib/Target/RISCV/RISCVInstrInfoB.td
@@ -643,6 +643,7 @@ def SLLIUWPat : ComplexPattern<i64, 2, "SelectSLLIUW", [and]>;
 def SLOIWPat  : ComplexPattern<i64, 2, "SelectSLOIW", [sext_inreg]>;
 def SROIWPat  : ComplexPattern<i64, 2, "SelectSROIW", [or]>;
 def RORIWPat  : ComplexPattern<i64, 2, "SelectRORIW", [sext_inreg]>;
+def FSRIWPat  : ComplexPattern<i64, 3, "SelectFSRIW", [sext_inreg]>;
 
 let Predicates = [HasStdExtZbbOrZbp] in {
 def : Pat<(and GPR:$rs1, (not GPR:$rs2)), (ANDN GPR:$rs1, GPR:$rs2)>;
@@ -804,6 +805,19 @@ def : Pat<(bswap GPR:$rs1), (GREVI GPR:$rs1, (i64 56))>;
 def : Pat<(bitreverse GPR:$rs1), (GREVI GPR:$rs1, (i64 63))>;
 } // Predicates = [HasStdExtZbp, IsRV64]
 
+let Predicates = [HasStdExtZbt] in {
+def : Pat<(or (and (xor GPR:$rs2, -1), GPR:$rs3), (and GPR:$rs2, GPR:$rs1)),
+          (CMIX GPR:$rs1, GPR:$rs2, GPR:$rs3)>;
+def : Pat<(riscv_selectcc GPR:$rs2, (XLenVT 0), (XLenVT 17), GPR:$rs3, GPR:$rs1),
+          (CMOV GPR:$rs1, GPR:$rs2, GPR:$rs3)>;
+def : Pat<(fshl GPR:$rs1, GPR:$rs2, GPR:$rs3),
+          (FSL GPR:$rs1, GPR:$rs2, GPR:$rs3)>;
+def : Pat<(fshr GPR:$rs1, GPR:$rs2, GPR:$rs3),
+          (FSR GPR:$rs1, GPR:$rs2, GPR:$rs3)>;
+def : Pat<(fshr GPR:$rs1, GPR:$rs2, uimmlog2xlen:$shamt),
+          (FSRI GPR:$rs1, GPR:$rs2, uimmlog2xlen:$shamt)>;
+} // Predicates = [HasStdExtZbt]
+
 let Predicates = [HasStdExtZbb] in {
 def : Pat<(ctlz GPR:$rs1), (CLZ GPR:$rs1)>;
 def : Pat<(cttz GPR:$rs1), (CTZ GPR:$rs1)>;
@@ -1004,6 +1018,31 @@ def : Pat<(sra (bswap GPR:$rs1), (i64 32)), (GREVIW GPR:$rs1, (i64 24))>;
 def : Pat<(sra (bitreverse GPR:$rs1), (i64 32)), (GREVIW GPR:$rs1, (i64 31))>;
 } // Predicates = [HasStdExtZbp, IsRV64]
 
+let Predicates = [HasStdExtZbt, IsRV64] in {
+def : Pat<(riscv_selectcc (and (assertsexti32 GPR:$rs3), 31),
+                          (i64 0),
+                          (i64 17),
+                          (assertsexti32 GPR:$rs1),
+                          (or (riscv_sllw (assertsexti32 GPR:$rs1),
+                                          (and (assertsexti32 GPR:$rs3), 31)),
+                              (riscv_srlw (assertsexti32 GPR:$rs2),
+                                          (sub (i64 32),
+                                               (assertsexti32 GPR:$rs3))))),
+          (FSLW GPR:$rs1, GPR:$rs2, GPR:$rs3)>;
+def : Pat<(riscv_selectcc (and (assertsexti32 GPR:$rs3), 31),
+                          (i64 0),
+                          (i64 17),
+                          (assertsexti32 GPR:$rs2),
+                          (or (riscv_sllw (assertsexti32 GPR:$rs1),
+                                          (sub (i64 32),
+                                               (assertsexti32 GPR:$rs3))),
+                              (riscv_srlw (assertsexti32 GPR:$rs2),
+                                          (and (assertsexti32 GPR:$rs3), 31)))),
+          (FSRW GPR:$rs1, GPR:$rs2, GPR:$rs3)>;
+def : Pat<(FSRIWPat GPR:$rs1, GPR:$rs2, uimmlog2xlen:$shamt),
+          (FSRIW GPR:$rs1, GPR:$rs2, uimmlog2xlen:$shamt)>;
+} // Predicates = [HasStdExtZbt, IsRV64]
+
 let Predicates = [HasStdExtZbb, IsRV64] in {
 def : Pat<(add (ctlz (and GPR:$rs1, (i64 0xFFFFFFFF))), (i64 -32)),
           (CLZW GPR:$rs1)>;
diff --git a/llvm/test/CodeGen/RISCV/rv32Zbt.ll b/llvm/test/CodeGen/RISCV/rv32Zbt.ll
diff --git a/llvm/test/CodeGen/RISCV/rv64Zbt.ll b/llvm/test/CodeGen/RISCV/rv64Zbt.ll
