what's the theory of convertion between different forms of fixed point, like converting a[40 37] to a[32 29]?

in the verilog code generated by fdatool, it is like

assign result = ((original[33] == 1'b0 & original[32:29] != 4'b0000) || (original[33] == 1'b0 && original[29:6] == 24'b011111111111111111111111) ) ? 24'b011111111111111111111111 : 
(original[33] == 1'b1 && original[32:29] != 4'b1111) ? 24'b100000000000000000000000 : ({original[33],original[29:0] + {original[6], {5{~original[6]}}}})>>>6;

convert original[40 37] to result[32 29],and they represent the same value. I wonder that how does the core of this arithmetic, result = ({original[33],original[29:0] + {original[6], {5{~original[6]}}}})>>>6 , work? thanks for your attention.