PROPOSAL: targetm.expand_tree_builtin hook

Andrew Pinski pinskia@physics.uc.edu
Mon May 10 19:27:00 GMT 2004


On May 10, 2004, at 15:18, Ziemowit Laski wrote:

It already lives in the apple-ppc-branch, but I thought it might be of 
more general interest.
The hook would be invoked from c-common.c:expand_tree_builtin, and 
would facilitate
target-specific, eager (i.e., the moment you see the function call) 
TREE-to-TREE transformations.
On Darwin, we use it construct so-called "constant CFStrings", and I'm 
about to hook up
__builtin_altivec_compiletime_error to it as well (since using the RTL 
targetm.expand_builtin
hook is too little, too late).

Would there be interest (on behalf of the powers that be) in me 
whipping up such a patch for the mainline?
I think this should be called not be called expand_tree_builtin as that
saya TREE to RTL expansion but rather fold_builtin as that is more
consistent with the rest of the functions in builtins.  Also this should
be called from fold_builtin in builtins.c rather than 
expand_tree_builtin.
This will allow the builtin to be "folded" into trees so that the 
tree-ssa
to work on the "folded" trees.

Thanks,
Andrew Pinski



More information about the Gcc mailing list