On Mon, 11 Apr 2011 12:59:12 -0700, Kenneth Graunke <kenn...@whitecape.org> wrote: > On 04/09/2011 10:17 PM, Eric Anholt wrote: > > The specs only say that they have to be a constant expression, which > > would allow things like builtin function calls that generate > > instructions as a side effect of their parsing. > > > > Fixes the fixed array-size-constant-relational.vert. > > --- > > src/glsl/ast.h | 2 +- > > src/glsl/ast_function.cpp | 3 ++- > > src/glsl/ast_to_hir.cpp | 34 ++++++++++++++++++---------------- > > 3 files changed, 21 insertions(+), 18 deletions(-) > > NAK. It seems utterly wrong that constant expressions generate > instructions. I think we should fix this, rather than allow it. > > Apparently what's being emitted is: > 1. The function prototype > 2. A global variable declaration (i.e. lessThan_retval) > 3. A global call and assignment (initializer) > > Of course, these will likely not be used. ast_function could probably > detect if a function call is a C.E. and avoid emitting them.
I kind of like the sound of up-front constant expression detection for function calls. The thing I want to avoid is function calls appearing in expression trees, which seems like the other way this would work.
pgpY6dC2aZBPq.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
