Author: Alex Gaynor <alex.gay...@gmail.com> Branch: extradoc Changeset: r3887:0839f40cc0e0 Date: 2011-08-23 12:21 -0500 http://bitbucket.org/pypy/extradoc/changeset/0839f40cc0e0/
Log: added a ask for branchless INT_ABS resop diff --git a/planning/jit.txt b/planning/jit.txt --- a/planning/jit.txt +++ b/planning/jit.txt @@ -11,7 +11,7 @@ --------- - have benchmarks for jit compile time and jit memory usage - + - maybe refactor a bit the x86 backend, particularly the register allocation @@ -36,7 +36,18 @@ - implement small tuples, there are a lot of places where they are hashed and compared -- support single floats in the JIT +- implement INT_ABS in the JIT, currently jtransform generates a call to an + inlined function, which does the trivial branching. However, GCC shows that + it can be done branchless in the ASM (as we do for FLOAT_ABS, which is easier + because it has an explicit sign bit). GCC generates: + + movq %rdi, %rdx + sarq $63, %rdx + movq %rdx, %rax + xorq %rdi, %rax + subq %rdx, %rax + ret + OPTIMIZATIONS ------------- _______________________________________________ pypy-commit mailing list pypy-commit@python.org http://mail.python.org/mailman/listinfo/pypy-commit
