Custom Flow Graph representation and data extraction

clang -g -S -emit-llvm code.c -Xclang -disable-O0-optnone -o code.ll

path/to/opt -load /path/to/lib/LLVMFG.so -enable-new-pm=0 -printFGNL < code.ll > /dev/null

path/to/opt -load /path/to/lib/LLVMFG.so -enable-new-pm=0 -basic-aa -mem2reg -simplifycfg -loop-simplify -loop-rotate -simplifycfg -instcombine -indvars -da -analyze -printFG < code.ll > /dev/null

• dot files that can be used to generate a picture of the FG of all the functions

• files that contains information about the functions in the code

[node_id, llvm_inst_opcode, num_operands, operands_size, [operand0_type, operand1_type..], is_array, array_size, in_loop]

Loopid{
[loop_stride, trip_count, depth]
innerloopid{
[loop_stride, trip_count, depth]
[node_id, llvm_inst_opcode, num_operands, operands_size, [operand0_type, operand1_type..], is_array, array_size, in_loop]
[node_id, llvm_inst_opcode, num_operands, operands_size, [operand0_type, operand1_type..], is_array, array_size, in_loop]
...
end innerloopid}
[node_id, llvm_inst_opcode, num_operands, operands_size, [operand0_type, operand1_type..], is_array, array_size, in_loop]
[node_id, llvm_inst_opcode, num_operands, operands_size, [operand0_type, operand1_type..], is_array, array_size, in_loop]
...
...
end loopid}

NOTE: loop_stride, trip_count, depth are reliable only with additional pass (-basic-aa -mem2reg -simplifycfg -loop-simplify -loop-rotate -simplifycfg -instcombine -indvars -da -analyze)

FunctionName{
BB0[#nodes_in_bb0,in_loop_bb0,[(inst1:#inst1),(inst2:#inst2),..,(otherinst:0)]]
BB1[#nodes_in_bb1,in_loop_bb1,[(inst1:#inst1),(inst2:#inst2),..,(otherinst:0)]]
...
}

dot -Tpng mainFG.dot -o mainFG.png

• Green Box: Basic Blocks in a Loop
• Blue Box: Basic Blocks outside a Loop
• Red Box: Loop (Nested loops are darker)
• Red edge: Control Flow dependence
• Black edge: Data dependence
• Gold edge: Data dependence that refers to a function argument
• Gold dashed edge: Reference to the base address of an array/structure that is a function parameter
• Darkgreen dashed edge: Reference to the base address of an local array/structure

cat function_1_loopdata

Loop0 {
[3, 3, 1]
[119, 55, 2, 64, int, int, 0, -1, 1]
[120, 56, 4, 64, ptr, 0, -1, 1]
[121, 34, 3, 64, ptr, int, int, 0, -1, 1]
[122, 33, 2, 64, int, ptr, 0, -1, 1]
[123, 13, 2, 64, int, int, 0, -1, 1]
[124, 56, 4, 64, ptr, 0, -1, 1]
[125, 53, 2, 64, int, int, 0, -1, 1]
[126, 2, 3, 1, int, 0, -1, 1]
End Loop0 }

TODO:

• organize code in functions
• fix some bug that create wrong cdfg dot file
• ignore llvm debug instruction