You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Right now compilation is controlled in one of two ways: globally with GxB_JIT_C_CONTROL and quasi-local control in the form of type / operator definitions (providing a definition enables JIT, not providing one disables it. This is a one time switch though).
I would like to allow users more manual control, and to allow me to JIT only when I deem it worth the overhead. I propose allowing GxB_JIT_C_CONTROL to be set on individual operators and types as well. I believe I understand the compilation pipeline relatively well now, and I don't think the overhead to do this would be too extreme (as the JIT works currently).
The default would be the same as the global.
The text was updated successfully, but these errors were encountered:
That wouldn't be too hard to do, like GrB_set (operator, GxB_JIT_C_CONTROL, GxB_JIT_OFF) or something. It could be done for types, monoids, ops, semirings, etc without much difficulty.
Right now compilation is controlled in one of two ways: globally with
GxB_JIT_C_CONTROL
and quasi-local control in the form of type / operator definitions (providing a definition enables JIT, not providing one disables it. This is a one time switch though).I would like to allow users more manual control, and to allow me to JIT only when I deem it worth the overhead. I propose allowing
GxB_JIT_C_CONTROL
to be set on individual operators and types as well. I believe I understand the compilation pipeline relatively well now, and I don't think the overhead to do this would be too extreme (as the JIT works currently).The default would be the same as the global.
The text was updated successfully, but these errors were encountered: