- Jun 09, 1990
-
-
wlott authored
accessors (still in cell.lisp), the allocation stuff (in alloc.lisp) and the defknowns (in vm-fndb.lisp). This was prompted by the the old loader barfing on the huge form generated when they were all in cell.lisp.
-
wlott authored
the error system so that the name comes out right. Also, rewrote the stuff to calc the pc of the break instruction so that doesn't hit the load-tn bug.
-
wlott authored
which included commits to RCS files with non-trunk default branches.
-
wlott authored
address.
-
wlott authored
compile codegen.lisp (which is where they are defined).
-
wlott authored
-
- Jun 08, 1990
- Jun 07, 1990
- Jun 06, 1990
-
-
wlott authored
getting a segv when at the end of a page.
-
wlott authored
-
wlott authored
need shifting.)
-
wlott authored
-
wlott authored
-
wlott authored
-
ram authored
Added new type-checking COERCE-TO-FUNCTION.
-
ram authored
-
ram authored
can work.
-
ram authored
FIXNUM restrictions.
-
ram authored
disallowed by the primitive type.
-
ram authored
and added appropriate new error routines.
-
ram authored
and changes to preserve enough type info in IR2 so that representation selection and load-TN packing can tell what representations are legal for a value. It is now guarantees that the TN passed as a VOP operand (via :TRANSLATE) will be of one of the primitive types allowed in the operand primitive-type restriction. Took advantage of the new delayed type checking to somewhat improve the code we get for FUNCALL when the operand type is unknown: we now call COERCE-TO-FUNCTION when we don't know whether the arg is legal, and call FAST-SAFE-COERCE-TO-FUNCTION when we know it is a function or a symbol, but not which. This eliminates the duplication of the function-p test in the normal (function) case where we don't know anything about the value at compile-time. The new CONTINUATION-TNS function is used to get and check the values of fixed multiple value continuations.
-
ram authored
there is a type check and the policy is safe so that we can represent whetever type the value might be. This allows checking to be done at the DEST, rather than at uses (preventing spurious type errors.) Allow a T operand restriction to encompass any type. Change the test for determining whether template results are o.k. to be more easily replicated in IR2 conversion (when necessary.)
-
ram authored
stuff. load-function => move-function Add meta-compile-time primitive-type info so that we can do more consistency checking. Put back primitive-type args to TN constructors. Changed load-scs vector to hold a list of all SCs that we can load into, ot T if the SC is directly allowed. Added consistency checking to compare the operand primitive-type restrictions with the SC restrictions.
-
ram authored
Removed the #-new-compiler on the COMPILE in some EVAL-WHEN's. sc-load-functions => sc-move-functions.
-
ram authored
it seems we do need that info after all.
-
ram authored
coming from explicit conditionals.
-
ram authored
-
ram authored
-
ram authored
types from being checked, even when they are buried in other types (like OR.)
-
wlott authored
Fixed internal-error to check to see if the trap was hit while in a branch-delay slot, and deal with it correctly if so.
-
wlott authored
-
wlott authored
-
wlott authored
-
- Jun 05, 1990
-
-
wlott authored
Added object-not-weak-pointer.
-
- Jun 04, 1990