This doesn't fix the motor bug, but it doesn't make it worse. However,
it does simplify the trees quite a bit, so it should be easier to debug.
It seems the problem has something to do with fold_constants messing up
dagged aliases: in particular, const-folding multiplication by e0123 in
3d PGA as fold_constants sees it as 1.
I'm not yet sure what went wrong, but the introduction of dags broke
something in my set_transform function (perhaps the dual?), but it's
something to do with the symbol being dagged (I guess because it's
required for everything else to dag). However, the strangest thing is
the error shows up with 155a8cbcda which
is before dags had any direct effect on the geometric algebra code. I
have a sneaking suspicion it's yet another convert_name issue.
They should be treated as such only when merging vector components. This
fixes a bug that doesn't actually exist (it's in experimental code),
where the sum of two 3-component vectors was getting lost.
For cross products: remove any a from a×(...+/-a...)
For dot products: remove any a×b from a•(...+/-a×b...) (or b×a)
This removed another 2 instructions :)
They don't have much effect that I've noticed, but the expression dags
code does check for commutative expressions. The algebra code uses the
anticommutative flag for cross, wedge and subtract (unconditional at
this stage). Integer ops that are commutative are always commutative (or
anticommutative). Floating point ops can be controlled (default to non),
but no way to set the options currently.
This takes advantage of the expression dag to detect when an expression
is on both sides of a cross product (which always results in 0). This
removes 3 instructions from my motor test (28 to go).
Especially binary expressions. That expressions can now be reused is
what caused the need to make expression lists non-invasive: the reuse
resulted in loops in the lists. This doesn't directly affect code
generation at this stage but it will help with optimizing algebraic
expressions.
The dags are per sequence point (as per my reading of the C spec).
The removal of the e tag from expr_t necessitated making convert_name
return a new expressions which resulted in get_type no longer being
enough to both convert a name expression and get the type. This was just
another missed spot. With this, all of game-source except ctf builds.
Finally, that little e. is cleaned up. convert_name was a bit of a pain
(in that it relied on modifying the expression rather than returning a
new one, or more that such behavior was relied on).
Sum expressions pull the negation through extend expressions allowing
them to switch to subtraction when appropriate, and offset_cast reaches
past negation to check for extend expressions. This has eliminated all
negation-only instructions from the motor-point, shaving off more
instructions (now 27 not including the return).
I don't know why I didn't think to do it this way before, but simply
recursing into each operand for + or - expressions makes it much easier
to generate correct code. Fixes the motor-point test.
That is, passing int constants through ... in Ruamoko progs is no longer
a warning (still is for v6p and v6 progs). I got tired of getting the
warning for sizeof expressions when int through ... hasn't been a
problem for even most v6p progs, and was intended to not be a problem
for Ruamoko progs.
But really only for memset and memmove because they need to use an int
alias of the variable and it may be only that alias that sets a much
larger variable.
This removes all the special cases and thus it should be more robust. It
did show up some out-by-one (or a factor of two!) errors elsewhere in
the group mask calculations.
I'm uncertain about the names, but this makes it much easier to get at
specific subtypes (eg, PGA.tvec as a type) rather than having to know
the group mask.
This makes working with the plethora of types a little easier still. The
check for an algebra expression in field_expr needed to be moved up
because the struct code thought the algebra type was a normal vector.
And vis-versa.
I'm not sure what I was thinking, but I've decided that not being able
to cast the pseudo-scalar from float to double (for printf etc) was a
bug.
The merge_blocks function wasn't reporting whether it had done anything
so the thread/merge/dead blocks loop was bailing early. With this,
simple functions (ie, no control flow) are fully visible to the CSE
optimizer and it can get quite aggressive (removed 3 assignments and a
cross product from my barycenter test code).
Failing to promote ints to the algebra type results in a segfault in
assignment of a multi-vector due to the symbol pointer walking off the
end of the list of symbols.
And convert addition to subtraction when extend expressions are not
involved. This has taken my little test down to 56 instructions total
(21 for `l p ~l`), down from 74 (39).
This takes care of chained sums of extend expressions. Now `l p ~l` has
only four extend instructions which is expected for the code not
detecting the cross product that always produces 0.
This goes a ways towards minimizing extend expressions, even finding
zeros and thus eliminating whole branches of expressions, but shows the
need for better handling of chained sums of extends.
Any geometric algebra product of two negatives cancels out the negative,
and if the result is negative (because only one operand was negative),
the negation is migrated to above the operation. This resulted in
removing 2 instructions from one if my mini-tests (went from 74 to 78
with the addition/subtraction change, but this takes it back to 76
instructions).
Summed extend expressions are used for merging a sub-vector with a
scalar. Putting the vector first in the sum will simplify checks later
on (it really doesn't matter which is first so long as it's consistent).
Subtraction is implemented as adding a negative (with the plan of
optimizing it later). The idea is to give tree inspection and
manipulation a more consistent view without having to worry about
addition vs subtraction.
Negation is moved as high as possible in the expression, but is always
below an extend expression. The plane here is that the manipulation code
can bypass an alias-add-extend combo and see the negation.
This doesn't affect the generated code (aliases are free), but does
simplify the dag significantly, thus optimizing the compiler somewhat,
but also makes reading dags much easier and therefore optimizing the
debugging process.
Because the aliases were treated as live, every alias of a temp resulted
in an assignment, which proved to be quite significant (4-5 assignments
in some simple GA expressions). By using an alias node in the dag, the
unaliased temp can be marked live while the alias is treated as an
operation rather than an operand. Now my GA expressions have no
superfluous assignments (generally no assignments at all).
Simple k-vectors don't use structs for their layout since they're just
an array of scalars, but having the structs for group sets or full
multi-vectors makes the system alignment agnostic.
And geometric algebra vectors. This does break things a little in GA,
but it does bring qfcc's C closer to standard C in that sizeof respects
the alignment of the type (very important for arrays).
It's implemented as the Hodge dual, which is probably reasonable until
people complain. Both ⋆ and ! are supported, though the former is a
little hard to see in Consola.
That was surprisingly harder than expected due to recursion and a
not-so-good implementation in expr_negate (it went too high-level thus
resulting in multivec expressions getting to the code generator).
But only for scalar divisors. The simple method of AB†/(BB†) works only
if B is a versor and there's also the problem of left and right
division. Thanks to sudgy for making me stop and think before I actually
implemented anything (though he mentioned only that it doesn't work for
general mutli-vector divisors).
That was tedious. I can't say I'm looking forward to writing the tests
for 3d. And even though trivector . bivector and bivector . trivector
give the same answer, they're not really commutative when it comes to
the code.
Meaning vec3 is aligned to 4 components instead of 1. 2-component ops
use vec2 in the VM thus requiring alignment to boundaries of 2, but 4
seems better as it conforms with OpenGL and Vulkan (and, I imagine,
DirectX, but I doubt QF will ever use DirectX).
The singleton alias resulted in the adjusted swizzles being corrupted
when for the same def. Other than adding properly sized swizzles
(planned), the simplest solution is to (separately) allow alias that
stick out from from the def.
While the progs engine itself implements the instructions correctly, the
opcode specs (and thus qfcc) treated the results as 32-bit (which was,
really, a hidden fixme, it seems).
I didn't particularly like that solution due to the implied extra
bandwidth (probably should profile such sometime), but I think the
extend operations could be merged into simple assignments by the
optimizer at some stage (or further cleaned up when this stuff gets
moved to actual code gen where it should be).
Currently via only the group mask (which is really horrible to work
with: requires too much knowledge of implementation details, but does
the job for testing), but it got some basics working.
It turned out they were always using floats for the source type (meaning
doubles were broken), and not shifting the component in the final sizzle
code meaning all swizzles were ?xxx (neglecting minus or 0). I'd make
tests, but I plan on modifying the instruction set a little bit.
Also, correct the handling of scalars in dot and wedge products: it
turns out s.v and s^v both scale. However, it seems the CSE code loses
things sometimes.
This has shown the need for more instructions, such as a 2d wedge
product and narrower swizzles. Also, making dot product produce a vector
instead of a scalar was a big mistake (works nicely in C, but not so
well in Ruamoko).
The current code is pretty broken when it comes to vector types (losing
the vector and bogus errors among other issues). The whole thing needs a
rework or even just to be tossed in favor of better DAG processing.
I guess Hamish's suggestion made sense at the time, but I found that
with the current instructions, the reversed bivector wasn't so nice to
implement it would need a swizzle as well as the cross-product.
By default. Conversion of quake strings needs to be requested (which is
done by nq and qw clients and servers, as well as qfprogs via an
option). I got tired of seeing mangled source code in the disassembly.
I'm not sure if that was a thinko, typo, or something else, but judging
by the relevant commit message, the use of quaternion and vector was
intended only for advanced progs (v6p).
This makes working with them much easier, and the type system reflects
what's in the multi-vector. Unfortunately, that does mean that large
algebras will wind up having a LOT of types, but it allows for efficient
storage of sparse multi-vectors:
auto v = 4*(e1 + e032 + e123);
results in:
0005 0213 1:0008<00000008>4:void 0:0000<00000000>?:invalid
0:0044<00000044>4:void assign (<void>), v
0006 0213 1:000c<0000000c>4:void 0:0000<00000000>?:invalid
0:0048<00000048>4:void assign (<void>), {v + 4}
Where the two source vectors are:
44:1 0 .imm float:18e [4, 0, 0, 0]
48:1 0 .imm float:1aa [4, 0, 0, 4]
They just happen to be adjacent, but don't need to be.
Scaling now works for multi-vector expressions, and always subtracting
even when addition is wanted doesn't work too well. However, now there's
the problem of multi-vectors very quickly becoming full algebra vectors,
which means certain things need a rethink.
This gets only some very basics working:
* Algebra (multi-vector) types: eg @algebra(float(3,0,1)).
* Algebra scopes (using either the above or @algebra(TYPE_NAME) where
the above was used in a typedef.
* Basis blades (eg, e12) done via procedural symbols that evaluate to
suitable constants based on the basis group for the blade.
* Addition and subtraction of multi-vectors (only partially tested).
* Assignment of sub-algebra multi-vectors to full-algebra multi-vectors
(missing elements zeroed).
There's still much work to be done, but I thought it time to get
something into git.
If a symbol is not found in the table and a callback is provided, the
callback will be used to check for a valid procedural symbol before
moving on to the next table in the chain. This allows for both tight
scoping of the procedural symbols and caching.
Due to joys of pointers and the like, it's a bit of a bolt-on for now,
but it works nicely for basic math ops which is what I wanted, and the
code is generated from the expression.
Only · (dot product) and × (cross product for vector, commutator product
for geometric algebra) have been tested so far, but that involved
fighting with cpp to get it to not convert the · to \U000000b7, which
was rather annoying.
Probably not a good idea for large maps, but handy for generating C
structs for small test maps. Does not include vertices or surfaces, just
the bsp tree itself for now.
Fixing a load of issues related to autoconf and some small source-level issues to re-add clang support.
autoconf feature detection probably needs some addressing - partially as -Werror is applied late.
The source tree is made read-only by `make distcheck`, so writing
temporary files to the source directory is a no-no (really, it's a bit
of a bug in qfcc, as per #51).
With the use of the full type for encoding type aliases, ptraliasenc's
simple check became invalid (it's purpose is to ensure the encoding
doesn't have "null" in it, not the exact encoding itself, but this is
good enough).
Two variables declared as arrays (same size) of different typedefs to
the same base type have their type encodings both pointing to the same
short alias.
From vkgen:
51d3 ty_array [4={int32_t>i}] 207f 0 4
51d9 ty_array [4=i] 1035 0 4
51df ty_alias {>[4=i]} 16 51d9 51e6
51e6 ty_array [4={uint32_t>i}] 2063 0 4
51ec ty_union {tag VkClearColorValue-} tag VkClearColorValue
4ca0 0 float32
51df 0 int32
51df 0 uint32
uint32 should use 51e6 and int32 should use 513d,
I never liked it, but with C2x coming out, it's best to handle bools
properly. I haven't gone through all the uses of int as bool (I'll leave
that for fixing when I encounter them), but this gets QF working with
both c2x (really, gnu2x because of raw strings).
The warning flag check worked too well: it enabled the warning and
autoconf's default main wanted the const attribute. The bug has been
floating around for a while, it seems.
This uses ud-chains for function statements (call/return) to force their
arguments to be live (in particular, indirect references via pointers)
this fixes the arraylife test.
The ud- and du-chains include known side-effects of the instructions and
thus depict a more accurate view of what operands an instruction uses or
defines. Fixes the arraylife2 test.
Like defs, a partial write should not define the whole temp. Thus, copy
the "don't visit main" behavior recently added to def_visit_all. Fixes
missing ud-chains for component-by-component assignments to temporary
vectors.
I'm not certain this is correct, but it seems to me that du-chains are
the same information as ud-chains, but from the defining statement's
point of view instead of that of the using statement.
As certain statements (in particular, function calls) can use additional
variables via pointer parameters, it's necessary to iterate ud-chain
building until the count stabilizes. This should make live variable
analysis much easier.
I think the current build_element_chain implementation does a reasonable
job, but I'm in the process of getting designated initializers working,
thus it will become important to ensure uninitialized members get
initialized.
I never liked the various hacks I had come up with for representing
resource handles in Ruamoko. Structs with an int were awkward to test,
pointers and ints could be modified, etc etc. The new @handle keyword (@
used to keep handle free for use) works just like struct, union and
enum in syntax, but creates an opaque type suitable for a 32-bit handle.
The backing type is a function so v6 progs can use it without (all the
necessary opcodes exist) and no modifications were needed for
type-checking in binary expressions, but only assignment and comparisons
are supported, and (of course) nil. Tested using cbuf_t and QFile: seems
to work as desired.
I had considered 64-bit handles, but really, if more than 4G resource
objects are needed, I'm not sure QF can handle the game. However, that
limit is per resource manager, not total.
This takes advantage of the ud-chains to follow the trail of pointer
assignments looking for an address. This gets array element assignments
surviving across blocks when the array itself is passed to a function.
It doesn't help when the address of the element is taken though. I think
that's a dags problem and probably needs du-chains. Also, the ud-chain
creation should probably be done in two passes so the newly found
information can be recorded.
Def and kill are still handled in flow_analyze_statement, but this makes
call meta data more consistent between v6 and ruamoko progs, allowing
the statement use chain to be used for call argument analysis. It even
found a bug in the extraction of param counts from the call instruction.
I had missed the flowvar clearing for auxiliary use/def/kill operands.
It's possible it wasn't necessary at the time since the operands were
added just for dealloc checking, but there's every reason it could
become necessary.
The first use will be pointer analysis for function arguments where the
argument points to an array to mark the array as live, but I'm sure
there'll be plenty of other uses.
A partial write to a def should not define the whole def, thus
def_visit_all's overlap parameter now has a flag that prevents a visit
to the main def when accessing the def from an alias def. This prevents
a lot of spurious kills and defines in flow analysis.
The array access code was loading the vector, modifying the element,
then forgetting to write the modified vector back to whence it came.
However, that would be rather sub-optimal, so now when the vector is
accessed by a pointer, the array code switches to field access to get at
the vector element thus avoiding the need to copy the whole vector.
Needed for proper analysis (ud-chains etc). Of course, it was then
necessary to remove the parameter defs from the uninitialized defs.
Also, plug a couple of memory leaks (forgot to free some temporary
sets).
That is, `array + offset`. This actually works around the bug
highlighted by arraylife.r (because the array is explicitly used), but
is not a proper solution, so that test still fails of course. However,
with this, it's no longer necessary to use `&array[index]` instead of
`array + index`.
I could never remember what any of the numbers meant. While define is
still a little fuzzy (they're (pseudo)statement numbers), at least now
I'll always know that the numbers are the define set. Also, having the
flow address of the variable helps with understanding the reaching defs
output.
It seems that the optimizer keeps array assignments live when passing
the array as a pointer, but not when passing the address of an element.
Found when testing the following code:
BasisBlade *pga_blades[16] = {
blades[1], blades[2], blades[3], blades[4],
blades[7], blades[6], blades[5], blades[0],
blades[8], blades[9], blades[10], blades[15],
blades[14], blades[13], blades[12], blades[11],
};
BasisGroup *pga_groups[4] = {
[BasisGroup new:4 basis:&pga_blades[ 0]],
[BasisGroup new:4 basis:&pga_blades[ 4]],
[BasisGroup new:4 basis:&pga_blades[ 8]],
[BasisGroup new:4 basis:&pga_blades[12]],
};
Only the first element of pga_blades is being assigned in the optimized
code, but everything is correct when not optimizing.
I had messed up the handling of declarators for combinations of pointer,
function, and array: the pointer would get lost (and presumably arrays
of functions etc). I think I had gotten confused and thought things were
a tree rather than a simple list, but Holub set me straight once again
(I've never regretted getting that book). Once I understood that, it was
just a matter of finding all the places that needed to be fixed. Nicely,
most of the duplicated code has been refactored and should be easier to
debug in the future.
It turns out I broke the type system when it comes to pointers to
functions and arrays. This test checks basic function and array pointers
and passes with qfcc from before the type system rework.
The type system rewrite had lost some of the checks for function fields.
This puts the actual code in the one place and covers parameters as well
as globals.
Internally, * is not really a valid operator for vectors since it can
have many meanings. This didn't cause trouble until trying to build
everything in game-source (since there's still a lot of legacy code in
there).
The precedence check changes done in
63795e790b seem to have been incorrect
(game-source/ctf produced many false positives), so putting that check
against '=' back into the code seems like a good idea (no more false
positives). That sounds a bit cargo-cult, but I'm really not sure what I
was thinking when I did the changes (probably just tired).
This applies only to the top-level scope of the function. I'm not sure
if it's right for traditional quakec code, but that can be adjusted
easily enough.
The symtab code itself cares only about global/not global for the size
of the hash table, but other code can use the symtab type for various
checks (eg, parameter shadowing).
Along with QuakeC's, of course. This fixes type typeredef2 test (a lot
of work for one little syntax error). Unfortunately, it came at the cost
of requiring `>>` in front of state expressions on C-style functions
(QuakeC-style functions are unaffected). Also, there are now two
shift/reduce conflicts with structs and unions (but these same conflicts
are in gcc 3.4).
This has highlighted the need for having the equivalent of the
expression tree for the declaration system as there are now several
hacks to deal with the separation of types and declarators. But that's a
job for another week.
The grammar constructs for declarations come from gcc 3.4's parser (I
think it's the last version of gcc that used bison. Also, 3.4 is still
GPL 2, so no chance of an issue there).
This simplifies type type_specifier rule significantly as now TYPE_SPEC
(was TYPE) includes all types and their basic modifiers (long, short,
signed, unsigned). This should allow me to make the type system closer
to gcc's (as of 3.4 as that seems to be the last version that used a
bison parser) and thus fix typeredef2.
typeredef1 parses properly but fails due to it erroneously complaining
that foo is redeclared as a different kind of object (it's the same
kind).
typeredef2 is the real problem in that it's a syntax error when it
should not be. This has proven to be a show-stopper for development on
my laptop as it has very recent vulkan headers which have such a
duplicate typedef.
Once a unicode char (ie, > 127) was used, any ascii chars would get the
tail of the last unicode char resulting in broken utf-8 streams. The
resulting null glyph boxes were not very appealing.
Because of the way the plane normal is used (front/on/back checks, and
midpoint calculation), other than possible precision, there is no need
to normalize the normal. Removing the square root and division resulted
in a huge boost: from 34s to 14 seconds. The average clusters visible
hasn't change much, and a quick check in-game didn't show any issues.
At least modern gcc produces nice code for ?: (cmov), and a SIMD
cross-product uses several fewer instructions. The cross-product shaved
off 0.5-1s, but the modulo -> ?: shaved off about 3-4s, for a total of
about 10% speedup (1.09 insn/cyc vs 1.01 insn/cyc, so even perf agrees).
This fixes the basic vecconst test (extending it to other types breaks
because long and ulong are not properly supported yet). The conversion
is done by the progs VM rather than writing another 256 conversions
(though loops could be used). This works nicely as a test for using the
VM to help with compiling.
Raw 'x y z' style vector constants that look like ints (no fractional
parts) used to initialize vector globals/constants don't get converted
to float vectors, resulting in nans for negative values and denormals
for positive values. This tends to make game physics... interesting.
While the option to make '*' mean dot product for vectors is important,
it breaks vector scaling in ruamoko progs as the resultant vector op
becomes a dot product instead of the indented hadamard product (ie,
component-wise).
The common idiom for self init (below) causes a double-call when
compiling with --advanced, resulting in an incorrect retain count.
if (!(self = [super init])) {
return nil;
}
The support for the new vector types broke compiling code using
--advanced. Thus it's necessary to ensure vector constants are
float-type and vec3 and vec4 are treated as vector and quaternion, which
meant resurrecting the old vector expression code for v6p progs.
Id's comments are a little inconsistent, but for the most part usable
info can be extracted. While not yet supported, Arcane Dimensions'
comments are extremely consistent (just some issues with hyphen counts
in separators), so parsing out usable info will be fairly easy. The hard
part will be presenting it.
The method is still held by known_methods, so freeing it causes grief.
However, it may cause a leak thus the free is only commented out. More
investigation is needed. I'm surprised the problem didn't show on linux,
but cygwin-native hit it and valgrind on linux found the spot :)
While it does get a bit cluttered currently, being able to see the
contents of structures makes a huge difference. Also highlights that
vector immediates do not get the correct type encodings.
This fixes the internal error generated by the likes of
`(sv_gravity * '0 0 1')` where sv_gravity is a float and `'0 0 1'` is an
ivec3: the vector is promoted to vec3 first so that expanding sv_gravity
is expanded to vec3 instead of ivec3 (which is not permitted for a
float: expansion requires the destination base type to be the same as
the source).
For now, anyway, as the generated code looks good. There might be
problems with actual pointer expressions, but it allows entity.field to
work as expected rather than generate an ICE.
The resultant unicode is encoded as utf-8, which does conflict with the
quake character map, but right now unicode is useful only with font
text, and those support only standard unicode (currently only as utf-8),
but something will need to be sorted out.
Arrays are passed as a pointer to the first element, so are always valid
parameters. Fixes a bogus "formal parameter N is too large to be passed
by value" error.
While swizzle does work, it requires the source to be properly aligned
and thus is not really the best choice. The extend instruction has no
alignment requirements (at all) and thus is much better suited to
converting a scalar to a vector type.
Fixes#30
It seems clang loses track of the usage of the referenced unions by the
time the code leaves the switch. Due to the misoptimization, "random"
values would get into the vector constants. This puts the usages in the
same blocks as the unions, causing clang to "get it right" (though I
strongly suspect I was running into UB).
While I might need to tighten up the rules later, this allows binary
operations between vector (the type) and explicitly typed vec3 constants
(and non-constants, about which I am undecided). The idea is that
explicit constants such as '1 2 3'f should be compatible with either
type.
This applies to quaternions as well.
As a class's ivars are built up by inheritance, but with only that
class's ivars in the symbol table, is is necessary to include an offset
based on the super class's ivars in order to ensure alignments are
respected. This is achieved via the new `base` parameter to
build_struct(), which is used to offset the current size while
calculating the aligned offset of the symbols. The parameter is ignored
for unions, as they always start at 0. The ivars for the current class
still have a base offset of 0 until they are actually added to the
class.
Fixes#29
The alignment is specified as a power of 2 (ie, actual alignment = 1 <<
alignment) allowing old object files to be compatible (as their
alignment is 0). This is necessary for (in part for #30) as it turned
out even global vectors were not aligned correctly.
Currently, only data spaces even vaguely respect alignment. This may
need to be fixed in the future.
Most were pretty easy and fairly logical, but gib's regex was a bit of a
pain until I figured out the real problem was the conditional
assignments.
However, libs/gamecode/test/test-conv4 fails when optimizing due to gcc
using vcvttps2dq (which is nice, actually) for vector forms, but not the
single equivalent other times. I haven't decided what to do with the
test (I might abandon it as it does seem to be UD).
At at some stage blender enforced frames being integers (In the past,
there was support for fractional, I think, but I also seem to remember
it not working) (yes, for anybody looking, this commit message is more
or less copied from io_object_mu).
Defs and symbols benefit from swizzling as that's one instruction vs 2-3
for loading a scalar into a vector component by component. Constants are
ok because the result gets converted to a vector constant.
qfcc is putting two temps in the same location due to
defspace_alloc_aligned_loc returning the same address when there was a
hole caused by an earlier aligned alloc: specifically, a size-3 hole and
a size-2 allocation with alignment-2.
The destination operand must be a full four component vector, but the
source can be smaller and small sources do not need to be aligned: the
offset of the source operand and the swizzle indices are adjusted. The
adjustments are done during final statement emission in order to avoid
confusing the data flow analyser (and that's when def offsets are known).
This reverts commit 2904c619c1.
In order to support swizzle operations, I need to be able to alias defs
to larger types (eg, float to vec4), but alias_def rightly won't allow
this. However, as the plan is to do this in the final steps before
emitting the instruction, I plan on creating an alias to a float then
adjusting the type in the alias, but to do so without extra shenanigans,
I need alias_def to allow aliases to the same type. As a fringe benefit,
it makes the code agree with the comment in def.h :P
This came up when investigating an internal error from the line above.
It turned out the error was correct (problem with converting scalars to
vectors), but the break was not.
Currently, only vector/vec3 and quaternion/vec4 can be printed anyway,
but I plan on making explicit format strings for the types, so there
should be no need to promote any vector types (and really, any hidden
promotion is a bit of a pain, but standards...).
While the code would handle int vector types, there aren't any such
instructions, and the expression code shouldn't generate them, but all
float (32 and 64 bit) vector types do have a dot product instruction, so
check width rather than just vector/quaternion.
This fixes an error that's been lurking for over two years (since I made
parameters unlimited internally). The problem was the array was being
allocated on the stack and a simple struct copy was used to store type
type, resulting in a dangling pointer onto the stack. I'm surprised it
didn't cause more problems.
This allows all the tests to build and pass. I'll need to add tests to
ensure warnings happen when they should and that all vec operations are
correct (ouch, that'll be a lot of work), but vectors and quaternions
are working again.
Vector expressions no longer auto-widen due to the new vector types (I
might add such later, but for now this lets the tests try to build
(minus actual fixes in qfcc)).
With this, all vector widths and types are supported: 2, 3, 4 and int,
uint, long, ulong, float and double, along with support for suffixes to
make the type explicit: '1 2'd specifies a dvec2 constant, while '1 2 3'u
is a uivec3 constant. Default types are double (dvec2, dvec3, dvec4) for
literals with float-type components, and int (ivec2...) for those with
integer-type components.
Having three very similar sets of code for outputting values (just for
debug purposes even) got to be a tad annoying. Now there's only one, and
in the right place, too (with the other value code).
I'd created new_value_expr some time ago, but never used it...
Also, replace convert_* with cast_expr to the appropriate type (removes
a pile of value check and create code).
Use with quaternions and vectors is a little broken in that
vec4/quaternion and vec3/vector are not the same types (by design) and
thus a cast is needed (not what I want, though). However, creating
vectors (that happen to be int due to int constants) does seem to be
working nicely otherwise.
Nicely, I was able to reuse the generated conversion code used by the
progs engine to do the work in qfcc, just needed appropriate definitions
for the operand macros, and to set up the conversion code. Helped
greatly by the new value load/store functions.
pr_type_t now contains only the one "value" field, and all the access
macros now use their PACKED variant for base access, making access to
larger types more consistent with the smaller types.
This is an extremely extensive patch as it hits every cvar, and every
usage of the cvars. Cvars no longer store the value they control,
instead, they use a cexpr value object to reference the value and
specify the value's type (currently, a null type is used for strings).
Non-string cvars are passed through cexpr, allowing expressions in the
cvars' settings. Also, cvars have returned to an enhanced version of the
original (id quake) registration scheme.
As a minor benefit, relevant code having direct access to the
cvar-controlled variables is probably a slight optimization as it
removed a pointer dereference, and the variables can be located for data
locality.
The static cvar descriptors are made private as an additional safety
layer, though there's nothing stopping external modification via
Cvar_FindVar (which is needed for adding listeners).
While not used yet (partly due to working out the design), cvars can
have a validation function.
Registering a cvar allows a primary listener (and its data) to be
specified: it will always be called first when the cvar is modified. The
combination of proper listeners and direct access to the controlled
variable greatly simplifies the more complex cvar interactions as much
less null checking is required, and there's no need for one cvar's
callback to call another's.
nq-x11 is known to work at least well enough for the demos. More testing
will come.
This means that a tex_t object is passed in instead of just raw bytes
and width and height, but it means the texture can specify whether it's
flipped or uses BGR instead of RGB. This fixes the upside down
screenshots for vulkan.
QFS_NextFilename was renamed to QFS_NextFile to reflect the fact it now
returns a QFile pointer for the newly created file (as well as the
name). This necessitated updating WritePNG to take a file pointer
instead of a file name, with the advantage that WritePNGqfs is no longer
necessary and callers have much more control over the creation of the
file.
This makes QFS_NextFile much more secure against file system race
conditions and attacks (at least in theory). If nothing else, it will
make it more robust in a multi-threaded environment.
The "not" because I'm pretty sure they're false positives due to when
the function is called, but clang doesn't know that (wonder why gcc was
ok with it).
clang doesn't like anything but a bare 0 as null (and in some of the
cases, it was quite right: '\0' should not be treated as a null
pointer). And the crashers were just for paranoia and probably aren't
needed any more (kept for now, though).
It seems clang defaults to unsigned for enums. Interestingly, gcc was ok
with the checks being either way. I guess gcc treats enums that *can* be
unsigned as DWIM.
In working with vectors and matrices while testing the scene wrappers, I
found that there was a fair bit of confusion about how large something
could be. Return values can be up to 32 words (but qfcc wasn't aware of
that), parameters were limited to 4 words still (and possibly should be
for varargs), and temp defs were limited to 8 words (1 lvec4). Temps are
used for handling return values (at least when not optimizing) and thus
must be capable of holding a return value, and passing large arguments
through *formal* parameters should be allowed. It seems reasonable to
limit parameter sizes to return value sizes.
A temp and a move are still used for large return values (4x4 matrix),
but that's an optimization issue: the code itself is at least correct.
This is the bulk of the work for recording the resource pointer with
with builtin data. I don't know how much of a difference it makes for
most things, but it's probably pretty big for qwaq-curses due to the
very high number of calls to the curses builtins.
Closes#26
When the def can be found. This fixes direct assignments to arrays (and
probably structs) getting lost when the array is later read using a
variable index.
Float is not int, and Ruamoko has only int ifz/ifnz, which will fail for
-0.0 (0x80000000 when viewed as an int). And then there's nan, but I
haven't seen too many of those in quake.
I suspect this is an ancient bug that wasn't noticed due to not looking
at progs.src compiled code enough, but it makes the first statements of
the function point to the correct line instead of a forward declaration.
Currently only via pragma (not command line options), but I needed to
test the concept. Converting legacy code is just too error prone.
Telling the compiler how to treat the operator makes more sense. When *
acts as @dot with Ruamoko progs, the result is automatically aliased as
a float as this is the legacy meaning (ie, float result for dot
product).
This is a very tiny optimization, but there's no point in adjust the
stack if there's no actual adjustment. I didn't bother with it initially
because I thought it wouldn't happen (and I was more interested in
getting things working first), but it turns out that simple getters that
result in a zero adjustment are quite common (70/535 in qwaq-app.dat).
It now takes the function name to print in error message (passed on to
PR_Sprintf) and the argument number of the format string. The variable
arguments (in ...) are assumed to be immediately after the format
argument.
This is achieved by marking a void function with the void_return
attribute and then calling that function in an @return expression.
@return can be used only inside a void function and only with void
functions marked with the void_return attribute. As this is intended for
Objective-QC message forwarding, it is deliberately "difficult" to use
as returning a larger than expected value is unlikely to end well for
the calling function.
However, as a convenience, "@return nil" is allowed (in a void
function). It always returns an integer (which, of course,can be
interpreted as a pointer). This is safe because if the return value is
ignored, it will go into the progs return buffer, and if it is not
ignored, it is the smallest value that can be returned.
Having to remember to copy yet another specifier bit was getting
tedious, so use a union of a struct with the bitfields and an unsigned
int to access them in parallel. Makes for a tidier spec_merge, and one
less headache.
The command line option works the same way as
--advanced/traditional/extended, as does the pragma. As well, raumoko
(alternative spelling) can be used because both are legitimate and some
people may prefer one spelling over the other.
As always, use of the pragma is at one's own risk: its intended use is
forcing the target in the unit tests.
dvec4, lvec4 and ulvec4 need to be aligned to 8 words (32 bytes) in
order to avoid hardware exceptions. Rather than dealing with possibly
mixed alignment when a function has 8-word aligned locals but only
4-word aligned parameters, simply keep the stack frame 8-word aligned at
all times.
As for sizes, the temp def recycler was written before the Ruamoko ISA
was even a pipe dream and thus never expected temp def sizes over 4. At
least now any future adjustments can be done in one place.
My quick and dirty test program works :)
dvec4 xy = {1d, 2d, 0d, 0.5};
void printf(string fmt, ...) = #0;
int main()
{
dvec4 u = {3, 4, 3.14};
dvec4 v = {3, 4, 0, 1};
dvec4 w = v * xy + u;
printf ("[%g, %g, %g, %g]\n", w[0], w[1], w[2], w[3]);
return 0;
}
They're now properly part of the type system and can be used for
declaring variables, initialized (using {} block initializers), operated
on (=, *, + tested) though much work needs to be done on binary
expressions, and indexed. So far, only ivec2 has been tested.
When possible, of course. However, this tightens up struct and constant
index array accesses, and avoids issues with flow analysis losing track
of the def (such trucking is something I want to do, but haven't decided
out to get the information out to the right statements).
Since address expressions always product a pointer type, aliasing one to
another pointer type is redundant. Instead, simply return an address
expression with the desired type.
The FIXME was there because I couldn't remember why the test was
type_compatible but the internal error complains about the types being
the same size. The compatibility check is to see if the op can be used
directly or whether a temp is required. The offset check is because
types that are the same size (which they must be if they are
compatible) is because it is not possible to create an offset alias def
that escapes the bounds of the real def, which any non-zero offset will
do if the types are the same size.
This is the intended purpose of the offset field in address expressions,
and will make struct and array accesses more efficient when I sort out
the code generation side.
Ruamoko passes va_list (@args) through the ... parameter (as such), but
IMP uses ... to defeat parameter type and count checking and doesn't
want va_list. While possibly not the best solution, adding a no_va_list
flag to function types and skipping ex_args entirely does take care of
the problem without hard-coding anything specific to IMP.
The system currently just sets some bits in the type specifier (the
attribute list should probably be carried around with the specifier),
but it gets the job done for now, and at least gets things started.
This makes it much easier to check (and more robust to name changes),
allowing for effectively killing the node to which the variable being
addressed is attached. This fixes the incorrect address being used for
va_list, which is what caused double-alias to fail.
In order to not waste instructions, the Ruamoko ISA does not provide 1
and 2 component 64-bit load/store instructions since they can be
implemented using 2 and 4 component 32-bit instructions (load and store
are independent of the interpretation of the data). This fixes the
double test, and technically the double-alias test, but it fails due to
a problem with the optimizer causing lea to use the wrong reference for
the address. It also breaks the quaternion test due to what seems to be
a type error that may have been lurking for a while, further
investigation is needed there.
Since the call instruction in the Ruamoko ISA specifies the destination
of the return value of the called function, it is much like any
expression type instruction in that the def referenced by its c operand
is both defined and killed by the instruction. However, unlike other
instructions, it really has many pseudo-operands: the arguments placed
on the stack. The problem is that when one of the arguments is also the
destination of the return value, the dags code wants to use the stack
argument as it was the last use of the real argument. Thus, instead of
using the value of the child node for the result, use the value label
attached to the call node (there should be only one such label).
This fixes iterfunc, typedef, zerolinker and vkgen when optimizing. Now
all but the double tests and return postop tests pass (and the retun
postop test is not related to the Ruamoko ISA, so fails either way).
That is, updating a variable using a function that takes the same
variable, probably very common in iterators, thus the name. It happens
to be the first qfcc test specific to Ruamoko. It's really just the
typedef, zerolinker, and vkgen type encoding loop stripped down for ease
of debugging.
Of course, it fails :)
I really need to come up with a better way to get the result type into
the flow analyser. However, this fixes the aliasing ICE when optimizing
Ruamoko code that uses struct assignment.
Many math instructions don't care about the difference between signed
and unsigned operands and are thus specified using int, but need to be
usable with uint. div is NOT mapped because there is a difference:
0x8000 / 2 (16-bit) is 0x4000 unsigned but 0xc000 signed, and 0x8000 /
0xfffe is 0 unsigned and 0x4000 signed. This means I'll need to add some
more instructions. Not sure what to do about % and %% though as that's a
lot of instructions (12).
Thanks to the size of the type encoding being explicit in the encoding,
anything that tries to read the encodings without expecting the width
will simply skip over the width, as it is placed after the ev type in
the encoding.
Any code that needs to read both the old encodings and the new can check
the size of the basic encodings to see if the width field is present.
It's full of evil hacks, but has always been an evil hack relying on
undefined behavior. The weird shenanigans with local variables are
because Ruamoko doesn't copy the parameters like v6p does and thus v and
z are NOT adjacent as parameters. Worse, the padding is uninitialized
and thus should not be relied upon to be any particular value. Still
does a nice job of testing dot products, though.
With explicit operators, even. While they're a tad verbose, they're at
least unambiguous and most importantly have the right precedence (or at
least adjustable precedence if I got it wrong, but vector ops having
high precedence than scalar or component seems reasonable to me).
I don't remember why I did this originally, but it causes the dags code
to lose the offset temp alias when accessing fields on structural temps
(known to be the case for vectors (temp-component.r), and I seem to
remember having problems with structs).
While it specifically checks vectors, I'm pretty sure it applies to
structs, too. Also, it's a little redundant with vecaddr.r, but is much
more specific and far less evil in what it does (no horrible pointer
shenanigans): just something that is fairly common practice.
Since Ruamoko progs must use lea to get the address of a local variable,
add use/def/kill references to the move instruction in order to inform
flow analysis of the variable since it is otherwise lost via the
resulting pointer (not an issue when direct var reference move can be
used).
The test and digging for the def can probably do with being more
aggressive, but this did nicely as a proof of concept.
This code now reaches into one level of the expression tree and
rearranges the nodes to allow the constant folder to do its things, but
only for ints, and only when the folding is trivially correct (* and *,
+/- and +/-). There may be more opportunities, but these cover what I
needed for now and anything more will need code generation or smarter
tree manipulation as things are getting out of hand.
It now addressing_mode cleaning up store instructions to use ptr+offset
instead of lea;store ptr...
Entity.field addressing has been impelmented as well.
Move instructions still generate sub-optimal code in that they use an
add instruction instead of lea.
This allows the code handling simple pointer dereferences to recurse
along an alias chain that resulted from casting between different
pointer types (such chains could probably be eliminated by replacing the
type in the original pointer expression, but it wasn't worth it at this
stage).
Aliasing an alias expression to the same type as the original aliased
expression is a no-op, so drop the alias entirely in order to simplify
code generation.
Simply dereferencing a pointer does not need to go through array_expr
and thus collect a 0 offset that will only be constant-folded out again.
Really just a minor optimization in qfcc, but at one stage in today's
modification, it resulted in some unwanted aliasing chains.
While this does make the generated code a little worse, load is behaving
nicely), the two are at least consistent with each other and when I fix
one, I'll fix both. I missed this change the other day when I did the
address_expr cleanup. Yay near-duplicate code :P
This is what using new_ret_expr would result in, but new_ret_expr is no
longer used for referencing .return (except in pascal, but I haven't
gotten around to sorting that out) due to the recent changes for Ruamoko
progs. Fixes an ICE when compiling (with optimization) something like
the following (dir is a vector):
dir /= sqrt (dir * dir);
return dir * speed;
It turns out the sorting wasn't working properly and I've decided that
anything that actually needs the defs to be sorted by address (such as a
debugger searching for defs by address) can do the sorting itself. Fixes
a weird swapping of def names.
This is necessary to get statement disassembly working, and likely
debugging in general. locals is the total size of the stack frame and
thus reaches above the function-entry stack pointer, and params_start is
the local space relative start of the parameters. Thus, knowing the
function-entry stack pointer, the bottom of the locals space can be
found by subtracting params_start, and the top of the locals space by
adding (locals - params_start).
This gets all the sections of the progs file nicely aligned and the code
easier to read with the offset and size calculations not being spread
through the function. ivar-struct-return now works when compiled for
Ruamoko.
This cleans up dprograms_t, making it easier to read and see what chunks
are in it (I was surprised to see only 6, the explicit pairs made it
seem to have more).
While I think the reason the dags code moved an instruction before
adjstk and with was they shared a constant with that instruction (which
is a different bug), this ensures other instructions cannot get
reordered in front of adjstk and with, as doing so would cause any such
instructions to access incorrect data.
The goal was to get lea being used for locals in ruamoko progs because
lea takes the base registers into account while the constant pointer
defs used by v6p cannot. Pointer defs are still used for gobals as they
may be out of reach of 16-bit addressing.
address_expr() has been simplified in that it no longer takes an offset:
the vast majority of the callers never passed one, and the few that did
have been reworked to use other mechanisms. In particular,
offset_pointer_expr does the manipulations needed to add an offset
(unscaled by type size) to a pointer. High-level pointer offsets still
apply a scale, though.
Alias expressions now do a better job of hanling aliasing of aliases by
simply replacing the target type when possible.
It's possible I lost the child printing when creating the return
expressions, but dot diagrams are much more useful when they don't have
nodes with just pointer values.
The parameter defs are allocated from the parameter space using a
minimum alignment of 4, and varargs functions get a va_list struct in
place of the ...
An "args" expression is unconditionally injected into the call arguments
list at the place where ... is in the list, with arguments passed
through ... coming after the ...
Arguments get through to functions now, but there's problems with taking
the address of local variables: currently done using constant pointer
defs, which can't work for the base register addressing used in Ruamoko
progs.
With the update to test-bi's printf (and a hack to qfcc for lea),
triangle.r actually works, printing the expected results (but -1 instead
of 1 for equality, though that too is actually expected). qfcc will take
a bit longer because it seems there are some design issues in address
expressions (ambiguity, and a few other things) that have pretty much
always been there.
The aux use ops need to be counted and given nodes explicitly as they
may refer to defs that are not accessed by other statements other than
by aliases, and those aliases need to be marked live as well as the used
def.
This is part of the work for #26 (Record resource pointer with builtin
function data). Currently, the data pointer gets as far as the
per-instance VM function table (I don't feel like tackling the job of
converting all the builtin functions tonight). All the builtin modules
that register a resources data block pass that block on to
PR_RegisterBuiltins.
This will make it possible for the engine to set up their parameter
pointers when running Ruamoko progs. At this stage, it doesn't matter
*too* much, except for varargs functions, because no builtin yet takes
anything larger than a float quaternion, but it will be critical when
double or long vec3 and vec4 values are passed.
Storing a variable into a dereference pointer (*p = x) is not marking
the variable as used (due to a mistake while converting to Ruamoko
statement format) resulting in assignments to that variable being
dropped due to it being a dead assignment as the assignment to the
variable and the storing need to be in separate basic blocks (thus the
call in the test, though an if would have worked, I think) for the bug
to trigger.
The problem was a missed change when switching the internal statement
format to Ruamoko: I "used" the statement's operands directly rather
than the rotated ones when emitting v6p progs. Fixes a compile segfault
when NOT optimizing.
There was an out-by-one where attempting to run a program with only one
argument would result in the argument not being passed to the program
(two worked). This is actually the source of the error fixed in
9347e4f901 because test-harness.c was the
basis for qwaq's main.c
While all base registers can be used for any purpose at any time (this
is why the with instruction has hard-absolute modes: you can never get
permanently lost), qfcc currently uses the convention of register 0 for
globals and register 1 for stack locals (params, locals, function args).
The register used to access a def is stored in the def and that is used
to set the register bits in the instruction opcode.
The def code actually doesn't know anything about any conventions: it
assumes all defs are global for non-temp defs (the function code updates
the defs before emitting code) and the current function provides the
register to use for any temp defs allocated while emitting code.
Seems to work well, but debug is utterly messed up (not surprised, that
will be tricky).
Still need to get the base register index into the instructions, but I
think this is it for basic code generation. I should be able to start
testing Ruamoko properly fairly soon :)
Thanks to the use/def/kill lists attached to statements for pseudo-ops,
it turned out to be a lot easier to implement flow analysis (and thus
dags processing) than I expected. I suspect I should go back and make
the old call code use them too, and probably several other places, as
that will greatly simplify the edge setting.
The means that the actual call expression is not in the statement lint
of the enclosing block expression, but just its result, whether the call
is void or not. This actually simplifies several things, but most
importantly will make Ruamoko calls easier to implement.
The test is because I had some trouble with double-calls, and is how I
found the return-postop issue :P
Commit 76b3bedb72 broke more than just the
swap test, but at least I know I need to get an edge in the dag.
Currently, the following code is generated: return and add are reversed.
../tools/qfcc/test/return-postop.r:8: return counter++;
0001 store.i counter, .tmp0
0002 return .tmp0
0003 add.i .tmp0, (1), counter
However, I don't want to deal with it right now, so it's marked XFAIL.
Since Ruamoko now uses the stack for parameters and locals, parameters
need to come after locals in the address space (instead of before, as in
v6 progs). Thus use separate spaces for parameters and locals regardless
of the target, then stitch them together appropriately for the target.
The third space is used for allocating stack space for arguments to
called functions. It us not used for v6 progs, and comes before locals
in Ruamoko progs.
Other than the return value, and optimization (ice, not implemented)
calls in Ruamoko look like they'll work.
Thanks to me having done something right 20 years ago, that was pretty
easy :). The two boolean types aren't supported yet because I haven't
decided on just how to represent their types in qfcc.
This seems to be the most reasonable approach to allocating space for
function call parameters without using push and pop (or adding to the
stack pointer), though it's probably good even when using push and pop
to help keep things aligned.
My little test program now builds with the Ruamoko ISA :)
void cp (int *dst, int *src, int count)
{
while (count--) {
*dst++ = *src++;
}
}
Calls are broken (unimplemented), and non-void returns are not likely to
work either (only partially implemented).
Operand width is encoded in the instruction opcode, so the width needs
to be accounted for in order to select the correct instruction. With
this, my little test generates correct code for the ruamoko ISA (except
for return, still fails).
For the most part, it wasn't too bad as it's just a rotation of the
operands for some instructions (store, assign, branch), but dealing with
all the direct accesses to specific operands was a small pain. I am very
glad I made all those automated tests :)
This makes the v6p instruction table consistent with the ruamoko
instruction table, and clears up some of the ugliness with the load,
store, and assign instructions (. .= and = are now spelled out). I think
I'd still prefer an enum code (faster) but at least this is more
readable.
Missed this case in duplicate_type. Allows "short foo" and
"sizeof(short)" (even though qfcc and the engine have two ideas of the
size: I expect trouble later).
long is ignored for double, and v6p progs are stuck with 32 bits for
longs (don't feel like extending v6p any further), but the basics are
there for Ruamoko.
short is ignored for ints because the minimum size is 32, and signed is
just noise for ints anyway (and no chars, so...).
unsigned, however, is finally implemented properly (or at least seems to
be working correctly: tests pass after getting things compiling again,
and lt.u is used where it should be :)
Attempting to add ev_ushort caused ptraliasenc to break, but that was
because it was already broken: I had implemented the scan of the xdef
table incorrectly, thus adding only 1 ev type resulted in the walked
pointer being out of phase with its data due to it first passing over
the type encodings (which is why adding long and ulong didn't cause any
obvious trouble).
And other related fields so integer is now int (and uinteger is uint). I
really don't know why I went with integer in the first place, but this
will make using macros easier for dealing with types.
They are both gone, and pr_pointer_t is now pr_ptr_t (pointer may be a
little clearer than ptr, but ptr is consistent with things like intptr,
and keeps the type name short).
I don't know why they were ever signed (oversight at id and just
propagated?). Anyway, this resulted in "unsigned" spreading a bit, but
all to reasonable places.
This includes calls and unconditional jumps, relative and through a
table. The parameters are all lumped into the one object, with some
being unused by the different types (eg, args and ret_type used only by
call expressions). Just having nice names for the parameters (instead of
e1 and e2) makes it nice, even with all the sub-types lumped together.
No mysterious type aliasing bugs this time ;)
The move operator names are definitely obsolete (due to dropping the
expressions a year or two ago) and the precedence checks seem to be
handled elsewhere. Memset and state expressions went away a while back
too.
While this was a pain to get working, that pain only went to prove the
value of using proper "types" (even if only an enum) for different
expression types: just finding all the places to edit was a chore, and
easy to make mistakes (forgetting bits here and there).
Strangely enough, this exposed a pile of *type* aliasing bugs (next
commit).
v6 vs v6p are more or less as before, with ruamoko added in. qfcc will
now try (and fail, due to the opcode table opnames being wrong) to
create ruamoko progs when given the ruamoko target option.
At this stage, I doubt emit.c will need to know the details of the
target (v6, v6p, ruamoko) since the instruction formats are identical,
just different meanings for the opcode itself.
This allows v6, v6p (older QF VM) or ruamoko (new QF VM) to be targeted.
Currently defaults to v6p to allow QF to continue building without too
much hassle.
While qfcc dealing sensibly with mixed target VMs in the object files
has always been an outstanding issue, with the new instruction set it
has become a priority. Most importantly, this should allow QF to
continue building while I work on qfcc targeting the new IS.
It does little good for documentation to refer to fields that don't
exist (because a certain someone forgot to change the docs when changing
the field names, I wonder who :P).
And partial implementations in qfcc (most places will generate an
internal error (not implemented) or segfault, but some low-hanging fruit
has already been implemented).
This allows the VM to select the right execution loop and qfcc currently
still produces only the old IS (it doesn't know how to deal with the new
IS yet)
build_struct was unconditionally setting the type's alignment. This was
not a problem before because no types were requesting alignments larger
than those requested by their members (for structs). However, with the
upcoming new instruction set, quaternions need to be 4-word aligned.
For int, long, float and double. I've been meaning to add them for a
while, and they're part of the new Ruamoko instructions set (which is
progressing nicely).
The opcode table is a nightmare to maintain, but this does clean it up
and speed up opcode lookups since they can now be indexed. Of course, it
turns out I had missed adding several instructions, so had to fix that,
and qfcc needed a bit of a re-jigger to get the opcode out of the table.
The assignment to the node's variable must come after any uses of that
node, which the node's parent set indicates. In the swap test, this was
not a problem as the node had no parents, and in the link order test, it
just happened(?) to work.
While using just the label node's reachable set was sufficient for a
simple swap (t = a; a = b; b = t;), it is not sufficient for
read-before-write dependencies such as found in linked-list building:
{ o = array[ind]; o.next = obj; obj = o; }
The assignment to o.next uses obj, but that use is hidden because obj's
reachable nodes does not include o thus assigning o to obj causes the
array dereference to be assigned directly to obj and thus o.next winds
up pointing to o instead of whatever obj was. The parent nodes of obj's
node are its users, so any new assigned to obj must come after those
parents as well as any node reachable by obj's node.
Fixes a runaway loop error when adding a frikbot to the server.
qfo_to_progs was modifying the space data pointers in the input qfo,
making it impossible to reuse the qfo. However, qfo_relocate_refs needs
the updated pointers, thus do a shallow copy of the qfo and its spaces
(but not any of the data)
build_builtin_function does the right thing, and it was only legacy
syntax functions that were affected anyway. Certainly, external
variables should not be initialized, but klik uses @extern { } wrapped
around several builtin functions and I had added the feature to allow
just this as it is rather convenient.
I decided that the check for whether control reaches the end of the
function without performing some necessary action (eg, invoking
[super dealoc] in a derived -dealoc) is conceptually the return
statement using a pseudo operand and the necessary action defining that
pseudo operand and thus is the same as checking for uninitialised
variables. Thus, add a pseudo operand type and use one to represent the
invocation of [super alloc], with a special function to call when the
"used" pseudo operand is "uninitialised".
While I currently don't know what else pseudo operands could be used
for, the system should be flexible enough to add any check.
Fixes#24
I want to use the function's pseudo address that was used for managing
aliased temporary variables for other pseudo operands as well. The new
name seems to better reflect the variable's purpose even without the
other pseudo operands as temporary variables are, effectively, pseudo
operands until they are properly allocated.
Forgetting to invoke [super dealloc] in a derived class's -dealloc
method has caused me to waste far too much time chasing down the
resulting memory leaks and crashes. This is actually the main focus of
issue #24, but I want to take care of multiple paths before I consider
the issue to be done.
However, as a bonus, four cases were found :)
While get_selector does the job of getting a selector from a selector
reference expression, I have long considered lumping various expression
types under ex_expr to be a mistake. Not only is this a step towards
sorting that out, it will make working on #24 easier.
I have gotten tired of chasing memory leaks caused by me forgetting to
add [super dealloc] to my dealloc methods, so getting qfcc to chew me
out when I do seems to be a good idea (having such a warning would have
saved me many hours, just as missing return warnings have).
Well... it could be done better, but this works for now assuming it's in
/usr/include (and it's correct for mxe builts). Does need proper
autoconfiscation, though.
The portal flow stack nodes contain a simd vector, which requires
16-byte alignment. However, on 32-bit Windows, malloc returns 8-byte
aligned memory, leading to eventual segfaults. Since pstack_t is 48
bytes on 32-bit systems, it fits nicely into a 64-byte aligned cache
line (or two on 64-bit systems due to being 80 bytes).
For most (if not all) maps. The heapsort is needed only if the clustered
leafs are not contiguous, but most bsp compilers output contiguous leaf
clusters, so is just a bit of protection. The difference isn't really
noticeable on a fast machine, but no point in doing more work than
necessary.
Now that only 3852 clusters need to be checked for each cluster, fat-pvs
construction for ad_tears completes in about 0.7s, most of which seems
to be loading, conversion, compression and writing. O(N^3) cuts both
ways (hurts like crazy when N increases, does wonders when N decreases,
especially by a factor of 25). And then throw in improved cache
performance...
I suspect having an off-line compiler is still useful, but even if
qfvis's implementation never actually gets used, if cluster
reconstruction is put in the engine, large maps will be feasible even
for quakeworld. Just the reduced memory requirements alone will be a
huge benefit (~3GB down to 1.8MB).
This is only the first half (vertical) in that the vis bits are still
for the leafs rather than the clusters, but ad_tears goes from 500s to
7s for calculating the fat pvs (3852 clusters).
While this doesn't give as much of a boost as does basic sphere culling
(since it's just culling sphere tests), it took ad_tears' base vis from
1000s to 720s on my machine.
This removes the last of the arbitrary limits from qfvis. The goal is
not so much supporting crazy maps, but more about better data usage
(cluster_t is now 24 (or 16) bytes instead of 1048 (or 528). And
passages isn't used (yet?)...
It turns out cmem is not so good for many large allocations (probably a
bug in handling the blocks), but was really meant for lots of little
churning allocations anyway. After an analysis of winding lifetimes, it
became clear that the hunk allocator would work very well. The base
windings are allocated from a global hunk (currently 1GB, plenty for
even ad_tears), and ephemeral windings are allocated from a per-thread
hunk of 1MB (seems to be way more than enough: gmsp3v2 uses a maximum of
only 56064 bytes, and ad_tears got through 30% before I gave up on it).
Any speed difference (for gmsp3v2) seems to be lost in the noise: still
completing in 38.4s on my machine.
The output fat-pvs data is the *difference* between the base pvs and fat
pvs. This currently makes for about 64kB savings for marcher.bsp, and
about 233MB savings for ad_tears.bsp (or about 50% (470.7MB->237.1MB)).
I expect using utf-8 encoding for the run lengths to make for even
bigger savings (the second output fat-pvs leaf of marcher.bsp is all 0s,
or 6 bytes in the file, which would reduce to 3 bytes using utf-8).
After seeing set_size and thinking it redundant (thought it returned the
capacity of the set until I checked), I realized set_count would be a
much better name (set_count (node->successors) in qfcc does make much
more sense).
Extremely large maps take a very long time to process their PVS sets for
PHS or shadows, so having an off-line compiler seems like a good idea.
The data isn't written out yet, and the fat pvs code may not be optimal
for cache access, but it gets through ad_tears in about 500s (12
threads, compared to 2100s single-threaded in the qw server).
This reduces the overhead needed to manage the memory blocks as the
blocks are guaranteed to be page-aligned. Also, the superblock is now
alllocated from within one of the memory blocks it manages. While this
does slightly reduce the available cachelines within the first block (by
one or two depending on 32 vs 64 bit pointers), it removes the need for
an extra memory allocation (probably via malloc) for the superblock.
When moving an identifier label from one node to another, the first node
must be evaluated before the second node, which the edge guarantees.
However, code for swapping two variables
t = a; a = b; b = t;
creates a dependency cycle. The solution is to create a new leaf node
for the source operand of the assignment. This fixes the swap.r test
without pessimizing postop code.
This takes care of the core problem in #3, but there is still room for
improvement in that the load/store can be combined into a move.
This reverts commit 2fcda44ab0.
Killing the node is not the correcgt answer as it blocks many
optimization opportunities. The correct answer is adding edges to
describe the temporal dependencies. Of course, this breaks the swap.r
test.
In order to correctly handle swap-style code
{ t = a; a = b; b = t; }
edges need to be created for each of the assignments moving an
identifier lable, but the dag must remain acyclic (the above example
wants to create a cycle). Having the reachable nodes recorded makes
checking for potential loops a quick operation.
Identifiers can be constants. I don't remember quite what it fixed other
than some bogus kill relations in the dags (which might have caused
issues later).
If the src type is not a class, there is no inheritance chain to walk.
Fixes a segfault when returning self after a syntax error in the
following:
+(EditStatus *)withRect:(Rect)rect
{
return [[[self alloc] initWithRect:rect]:
}
-setCursorMode:(CursorMode)mode
{
cursorMode = mode;
return self;
}
GCC does a fairly nice job of producing code for vector types when the
hardware doesn't support SIMD, but it seems to break certain math
optimization rules due to excess precision (?). Still, it works well
enough for the core engine, but may not be well suited to the tools.
However, so far, only qfvis uses vector types (and it's not tested yet),
and tools should probably be used on suitable machines anyway (not
forces, of course).
This fixes the mightsee updates never occurring, but it doesn't make a
huge difference (though I suppose it might have back in the 90s, or with
a different map).
The stats were being updated before UpdateMightsee was getting called,
and it was incrementing the wrong value (so it would not have been
thread-safe).
While whether it's any faster is debatable (it's slightly slower, but
many more portals are being tested due to different rounding in the base
vis stage), it's certainly easier to read.
While the main bulk of the improvement (36s down from 42s for
gmsp3v2.bsp on my i7-6850K) comes from using a high-tide allocator for
the windings (which necessitated using a fixed size), it is ever so
slightly faster than using malloc as the back-end.
This is for the conversion /to/ paletted textures. The conversion is
necessary for csqc support. In the process, the conversion has been sped up
by implementing a color cache for the conversion process. I haven't
measured the difference yet, but Mr Fixit does seem to load much faster for
the sw renderer than it did before the change (many months old memory).
The server edict arrays are now stored outside of progs memory, only the
entity data itself (ie data accessible to progs via ent.fld) is stored in
progs memory. Many of the changes were due to code accessing edicts and
entity fields directly rather than through the provided macros.
Double benefit, actually: faster when building a fat PVS (don't need to
copy as much) and can be used in multiple threads. Also, default visiblity
can be set, and the buffer size has its own macro.
Sort of at the request of leileilol (a utility to create quakepal.py was
asked for, but this seems to be better approach). However, the feature is
not used yet (needs hooks in the import and export modules).
It now takes a context pointer (opaque data) that holds the buffers it
uses for the temporary strings. If the context pointer is null, a static
context is used (making those uses of va NOT thread-safe). Most calls to
va use the static context, but all such calls have been formatted
consistently so they are easy to find when it comes time to do a full
audit.
Block expressions hide ex_error, but get_type() always returns null when
it finds one (which it does by recursing into block expression), so just
check the type itself.
