This method is also essentially a loop with a big switch table inside. Each iteration of the loop processes one single bytecode instruction of the method being compiled. The switch table identifies which bytecode is being currently processed and invoke a method to handle that specific bytecode. Each "block" is a list of instructions connected by pointers. During the IR construction a stack is used to push and pop constant values that are instruction operands. When the instruction that uses those constants is being created the operands are popped from the stack and used to construct a node representing the instruction.

C1 HIR nodes are described in classes defined in the c1_Instruction.hpp/cpp file. Some simple optimizations, like simplifying shift expressions and local value numbering are performed while constructing the IR. Also, profiling instructions are inserted into the IR when the compilation is in a level that requires profiling.

Interestingly, even C1 bails out compilation if a method body is too large!

Note that this method is in c1_LIRGenerator_x86.cpp instead of c1_LIRGenerator.cpp. During the JVM build time, given the values using the configure the project, the build infrastructure will use preprocessor macros to include the correct .cpp file for the target architecture of the JVM - in this case x86_64.

This method basically extract the information from the HL IR nodes and create LL IR operand nodes, while doing that the method allocate virtual registers for loading operands and storing results of operations. The call then is forwarded to logic_op which is also part of LIRGenerator.