- VC++ generated horribly stupid code for x64 when incrementing pc at the
beginning of each instruction by storing hundreds of copies of it for
every opcode executed. Incrementing pc at the end avoids this madness.
- It is possible I messed something up with this change. Hopefully not.
- Every update rolled into one, because I'm pretty sure I missed some while
updating lemon.c (not counting today's commits), since it wasn't always
updated at the same time as lemon.c.
- In particular, I think this check-in from 2016-06-06 was very important to
us after commit 3d5867d29e (For the
Lemon-generated parser, add a new action type SHIFTREDUCE and use it to
further compress the parser tables and improve parser performance.):
* Fix lempar.c so that the shift-reduce optimization works for error
processing.
- DECORATE now has atan2(y,x) and VectorAngle(x,y) functions. They are
identical except for the order of their parameters. The returned angle
is in degrees (not radians).
- Enhancements to lemon to generate more compact action tables and to avoid making array bounds tests that can never fail on action table calculations. (user: drh)
- Update zcc-parse.lemon: YY_SZ_ACTTAB is now YY_ACTTAB_COUNT
- Don't bother keeping track of uncompiled nodes in a special table. Use
the regular symbol table instead. This should in the future make
compiling nodes referenced deeper than (and before) their definitions
fairly straightforward.
- Also, break up the compiler's Message() function into Warn() and Error()
and get rid of zcc_errors.h. I can't really see having a set of error
numbers being useful.
Patched up everything so that it compiles without errors again. This only addresses code related to some compile error. A large portion of the angle code still uses angle_t and converts back and forth.
- Added new state options that DECORATE got to the lemon parser.
- Enable token generation for state options. They were previously not
generated, so the grammar treated them as function calls instead.
- Now that state jumps are handled by returning a state, we still need a
way for them to jump to a NULL state. If the parameter processed by this
macro turns out to be NULL, Actor's 'Null' state will be substituted
instead, since that's something that can be jumped to.
- The A_Jump family of action functions now return the state to jump
to (NULL if no jump is to be taken) instead of jumping directly.
It is the caller's responsibility to handle the jump. This will
make it possible to use their results in if statements and
do something other than jump.
- DECORATE return statements can now return the result of a function
(but not any random expression--it must be a function call). To
make a jump happen from inside a multi-action block, you must
return the value of an A_Jump function. e.g.:
{ return A_Jump(128, "SomeState"); }
- The VMFunction class now contains its prototype instead of storing
it at a higher level in PFunction. This is so that
FState::CallAction can easily tell if a function returns a state.
- Removed the FxTailable class because with explicit return
statements, it's not useful anymore.
Converting a floating point value that is out of range for a signed integer will result in 0x80000000 with SSE math, which is used exclusively for this purpose on modern Visual C++ compilers, so this cannot be used anywhere.
On ARM there's problems with float to unsigned int conversions.
xs_Float does not depend on these
- I kept getting confused trying to read these instructions, so now their
disassembly looks more MIPS-like:
* All mnemonics have had 'b' prepended to them for "branch".
* The CMP_CHECK bit alters the displayed mnemonic for the inverted
versions. e.g. BEQ can be displayed as BNE.
* The following JMP instruction that encodes the branch destination has
been folded into the disassembly of the branch instruction. Unlike
MIPS, I chose to display it offset from the branch check with =>
instead of another comma.
- Fixed: Don't assume operator new will return a pointer with 16-byte
alignment when allocating a block for the VMFrameStack. Because it seems
it's actually guaranteed to be 8-byte aligned. Don't know where I got
the idea it would always be 16-byte aligned.
