Reducing memory access and Float <-> f64 conversion in hot loops is especially important to improve performance in float-heavy benchmark such as so_mandelbrot or so_nbody.
To achieved this, precise liveness analysis for register slots is needed, thus we introduced basic block and control flow graph in this PR.
Reducing memory access and Float <-> f64 conversion in hot loops is especially important to improve performance in float-heavy benchmark such as so_mandelbrot or so_nbody. To achieved this, precise liveness analysis for register slots is needed, thus we introduced basic block and control flow graph in this PR.
jitgen::analysis.rs jitgen::basic_block.rs