The optimizations added are as follows:

1. remove_useless_mov which treats the case:

    mov r1, r2
    mov r2, r1
   
    In this case the second move can be deleted
   

2. fold_intop_imm which treats the case:

    op1 r, imm1
    op2 r, imm2
   
    We can rewrite this as:
   
    op1 r, (imm1 op2 imm2)
   
    Iff op1 and op2 are associative and the result (imm1 op2 imm2) does not overflow.
   

Added the following extra functionality for the Flambda_backend.Doubly_linked_list data structure:

• next cell -> get the next cell
• hd_cell t -> get a cell to the first element in the list
• last_cell t -> get a cell to the last element of the list
• delete_curr/delete_after/delete_before cell -> delete the current element, the one after or before a cell. asserts if there is no cell to delete (i.e. we try to delete before the first element or after the last)
• extra functionlity for inserting and getting back the newly inserted cell in a Doubly_linked_list

Added:

• Flags for enabling peephole optimizations (-cfg-peephole-optimize)
• A generic CSV data structure

This was benchmarked on the following libraries with the following results:

• Base:

  • remove_useless_mov hits: 4145
  • fold_intop_arith hits: 4
  • fold_intop_bitwise hits: 8
  • total build time without optimizations (avg of 10 runs): 171.54 sec
  • total build time with optimizations (avg of 10 runs): 171.84 sec
  • total .o file size reduction from optimizations: 2288 bytes

• Core:

  • remove_useless_mov hits: 4473
  • fold_intop_arith hits: 9
  • fold_intop_bitwise hits: 51
  • total build time without optimizations (avg of 10 runs): 177.83 sec
  • total build time with optimizations (avg of 10 runs): 177.43 sec
  • total .o file size reduction from optimizations: 5968 bytes