[Exegesis] Implemented strategy for load operation (#113458)

This fix helps to map operand memory to destination registers. If
instruction is load, we can self-alias it in case when instruction
overrides whole address register. For that we use provided scratch
memory.

Author: AnastasiyaChernikova

llvm/test/tools/llvm-exegesis/RISCV/latency-by-load.s

@@ -0,0 +1,60 @@
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -opcode-name=LD 2>&1 | FileCheck --check-prefix=TEST1 %s
+
+TEST1:      ---
+TEST1-NEXT: mode: latency
+TEST1-NEXT: key:
+TEST1-NEXT:   instructions:
+TEST1-NEXT:     - 'LD X10 X10 i_0x0'
+
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -opcode-name=LW 2>&1 | FileCheck --check-prefix=TEST2 %s
+
+TEST2:      ---
+TEST2-NEXT: mode: latency
+TEST2-NEXT: key:
+TEST2-NEXT:   instructions:
+TEST2-NEXT:     - 'LW X10 X10 i_0x0'
+
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -opcode-name=LH 2>&1 | FileCheck --check-prefix=TEST3 %s
+
+TEST3:      ---
+TEST3-NEXT: mode: latency
+TEST3-NEXT: key:
+TEST3-NEXT:   instructions:
+TEST3-NEXT:     - 'LH X10 X10 i_0x0'
+
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -opcode-name=LWU 2>&1 | FileCheck --check-prefix=TEST4 %s
+
+TEST4:      ---
+TEST4-NEXT: mode: latency
+TEST4-NEXT: key:
+TEST4-NEXT:   instructions:
+TEST4-NEXT:     - 'LWU X10 X10 i_0x0'
+
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -opcode-name=LBU 2>&1 | FileCheck --check-prefix=TEST5 %s
+
+TEST5:      ---
+TEST5-NEXT: mode: latency
+TEST5-NEXT: key:
+TEST5-NEXT:   instructions:
+TEST5-NEXT:     - 'LBU X10 X10 i_0x0'
+
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -opcode-name=LUI 2>&1 | FileCheck --check-prefix=TEST6 %s
+
+TEST6: LUI: No strategy found to make the execution serial
+
+
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -opcode-name=LB 2>&1 | FileCheck --check-prefix=TEST7 %s
+
+TEST7:      ---
+TEST7-NEXT: mode: latency
+TEST7-NEXT: key:
+TEST7-NEXT:   instructions:
+TEST7-NEXT:     - 'LB X10 X10 i_0x0'
+
+# RUN: llvm-exegesis -mode=latency --benchmark-phase=assemble-measured-code -mtriple=riscv64-unknown-linux-gnu --mcpu=generic -mattr=+a -opcode-name=LR_W_RL 2>&1 | FileCheck --check-prefix=TEST8 %s
+
+TEST8:      ---
+TEST8-NEXT: mode: latency
+TEST8-NEXT: key:
+TEST8-NEXT:   instructions:
+TEST8-NEXT:     - 'LR_W_RL X10 X10'

llvm/tools/llvm-exegesis/lib/SerialSnippetGenerator.cpp

@@ -106,7 +106,48 @@ static void appendCodeTemplates(const LLVMState &State,
   }
   case ExecutionMode::SERIAL_VIA_MEMORY_INSTR: {
     // Select back-to-back memory instruction.
-    // TODO: Implement me.
+
+    auto &I = Variant.getInstr();
+    if (I.Description.mayLoad()) {
+      // If instruction is load, we can self-alias it in case when instruction
+      // overrides whole address register. For that we use provided scratch
+      // memory.
+
+      // TODO: now it is not checked if load writes the whole register.
+
+      auto DefOpIt = find_if(I.Operands, [](Operand const &Op) {
+        return Op.isDef() && Op.isReg();
+      });
+
+      if (DefOpIt == I.Operands.end())
+        return;
+
+      const Operand &DefOp = *DefOpIt;
+      const ExegesisTarget &ET = State.getExegesisTarget();
+      unsigned ScratchMemoryRegister = ET.getScratchMemoryRegister(
+          State.getTargetMachine().getTargetTriple());
+      const llvm::MCRegisterClass &RegClass =
+          State.getTargetMachine().getMCRegisterInfo()->getRegClass(
+              DefOp.getExplicitOperandInfo().RegClass);
+
+      // Register classes of def operand and memory operand must be the same
+      // to perform aliasing.
+      if (!RegClass.contains(ScratchMemoryRegister))
+        return;
+
+      ET.fillMemoryOperands(Variant, ScratchMemoryRegister, 0);
+      Variant.getValueFor(DefOp) = MCOperand::createReg(ScratchMemoryRegister);
+
+      CodeTemplate CT;
+      CT.Execution = ExecutionModeBit;
+      CT.ScratchSpacePointerInReg = ScratchMemoryRegister;
+
+      CT.Info = std::string(ExecutionClassDescription);
+      CT.Instructions.push_back(std::move(Variant));
+      CodeTemplates.push_back(std::move(CT));
+    }
+
+    // TODO: implement more cases
     return;
   }
   case ExecutionMode::SERIAL_VIA_EXPLICIT_REGS: {