Commit graph

2764 commits

Author SHA1 Message Date
Bill Currie
e77d3b9e54 Fix a missed colspan in sblock dumps. 2012-11-15 13:44:08 +09:00
Bill Currie
783b4082a1 Produce tidier dag dumps. 2012-11-15 13:44:08 +09:00
Bill Currie
04a8e19481 Move the flow-var use/define calcs to a better place.
dags.c really isn't the right place. The only reason it wasn't in flow.c in
the first place is find_operands was in the wrong place.
2012-11-15 13:44:08 +09:00
Bill Currie
c4d4174ebc Make some minor improvements to flow graph dumps. 2012-11-15 13:44:08 +09:00
Bill Currie
3298e4d67e Ensure find_operands sets the operands to known values.
This fixes the last of the bogus live variables.
2012-11-15 13:44:08 +09:00
Bill Currie
e3a49df0a8 Set def after setting use.
An instruction that both reads and writes the same variable will read the
variable before writing to it, so the instruction uses the variable rather
than defines it (for live-variable purposes).
2012-11-15 13:44:08 +09:00
Bill Currie
58cb73d6aa Remember to union the node's use with (out - def)
Oops.
2012-11-15 13:44:08 +09:00
Bill Currie
cd224c00d4 Add live variable info to flow graph dumps.
Things are a little bogus, but at least the bogosity is visible :)
2012-11-15 13:44:08 +09:00
Bill Currie
8817c4a1c2 Re-enable the flow graph dumping.
It's still in a bogus place, but it's necessary.
2012-11-15 13:44:08 +09:00
Bill Currie
3c79f07705 Fix an out-by-one error in the set operators.
I have no idea why I added that - 1.
2012-11-15 13:44:08 +09:00
Bill Currie
4b64912793 Begin doing some data flow analysis.
In te beginning, live variables.
2012-11-15 13:44:08 +09:00
Bill Currie
c00e666668 Expose find_operands.
It has proven to be a generally useful function, not just for dags.
2012-11-15 13:44:08 +09:00
Bill Currie
0a2ce6d381 Fix some doxygen comments. 2012-11-15 13:44:08 +09:00
Bill Currie
5514c8db3c Print the statement number in the sblock output. 2012-11-15 13:44:08 +09:00
Bill Currie
0b49bd343c Collect information on statements that set or use vars.
First, it turns out using daglabels wasn't such a workable plan (due to
labels being flushed every sblock). Instead, flowvars are used. Each actual
variable (whether normal or temp) has a pointer to the flowvar attached to
that variable.

For each variable, the statements that use or define the variable are
recorded in the appropriate set attached to each (flow)variable.
2012-11-15 13:44:08 +09:00
Bill Currie
7404aacab7 Resurrect dags building and printing.
However, this time it's done after the statement recording (so the
statement numbers are set).
2012-11-15 13:44:08 +09:00
Bill Currie
acedc65de1 Make an array of statements for the function.
With this, sets can be used to represent statements.
2012-11-15 13:44:07 +09:00
Bill Currie
98245f6d00 Bury a dead field. 2012-11-15 13:44:07 +09:00
Bill Currie
8844ac61a2 Rename dump_flow to dump_sblock and move to dot_sblock.c 2012-11-15 13:44:07 +09:00
Bill Currie
d6d03a63b4 Print the depth-first order of the flow graph nodes. 2012-11-15 13:44:07 +09:00
Bill Currie
acc3beb6ab Make flow_find_dominators a little more robust.
While a non-initial node not having predecessors is really an internal
error, segfaulting because of it is not nice.
2012-11-15 13:44:07 +09:00
Bill Currie
acde6ca0a0 Handle jumpb statements in dot_sblock and flow graphs. 2012-11-15 13:44:07 +09:00
Bill Currie
ffa08a99c2 Store the initializer expression in the def.
Initially, this will be used for collecting the targets of a jumpb
instruction.
2012-11-15 13:44:07 +09:00
Bill Currie
d7177a78e9 Rewrite the flow graph code.
The flow graph nodes are now properly separated from the graph, and edge
information is stored in the graph struct. This actually made for much
cleaner code (partly thanks to the use of sets and set iterators).

Flow graph reduction has been (temporarily) ripped out as the entire
approach was wrong. There was also a bug in that I didn't really understand
the dragon book about selecting nodes and thus messed things up. The
depth-first search tree "fixed" the problem, but was really the wrong
solution (sledge hammer :P).

Also, now that I understand that dot's directed graphs must be acyclic, I
now have much better control over the graphs (back edges need to be
flipped).
2012-11-15 13:44:07 +09:00
Bill Currie
b9599a7119 Better set memory management.
At the cost of one pointer per set, sets can now be allocated and freed
more efficiently (especially since malloc might round things up).
2012-11-15 13:44:07 +09:00
Bill Currie
2385cf65c2 Rename setstate_t to set_iter_t.
After all, it is an iterator :).
2012-11-15 13:44:07 +09:00
Bill Currie
d3d88290b5 Mostly satisfactory output of nested regions.
It turns out dot does not like cyclic graphs (thus some of the weird
layouts), but fixing it by flipping back-edges requires proper recording of
edge info (I guess that's what T is for in the dragon book).
2012-11-15 13:44:07 +09:00
Bill Currie
8c5a2b3d11 Implement depth-first searching for flow graphs.
This fixes the bogusly irreducible graphs.
2012-11-15 13:44:07 +09:00
Bill Currie
53622dbb99 Perform graph reduction on the function's flow graph.
The reduction is performed itteratively until the graph is irreducible, but
such that each reduction wraps the previous graph. Unfortunately, due
depth-first searching not being implemented, graphs that should be reduced
(ie, those with natural loops).
2012-11-15 13:44:07 +09:00
Bill Currie
388b6e6e8f Add set_size ().
Return the number of members in the set.
2012-11-15 13:44:07 +09:00
Bill Currie
8567508963 Rewrite set_first() and add set_next().
set_first() now returns a pointer to a setstate_t struct that holds the
state necessary for scanning a set. set_next() will automatically delete
the state block when the end of the set is reached. set_delstate() is also
provided to allow early termination of the scan.
2012-11-15 13:44:07 +09:00
Bill Currie
2da038ce0f Bring back dot_flow.c.
However, it produces simplified flow graphs (ie, just nodes and edges, no
statements, though later on data flow information will be added).
2012-11-15 13:44:07 +09:00
Bill Currie
8b374305d2 Rename dot_flow.c and print_flow.
They're now dot_sblock.c and print_sblock. The new names both better
reflect their purpose and free up "flow" for outputting the real flow
analysis graphs.
2012-11-15 13:44:07 +09:00
Bill Currie
c483770121 Rework flow.c with the aim of graph reduction.
Much of the data recently added to sblock_t has been moved to flownode_t.
No graph reduction is carried out yet, but the initial (innermost level)
graph has been built.
2012-11-15 13:44:07 +09:00
Bill Currie
da446fd343 Detect natural loops.
Non-nested loops that have the same head-node are merged into one loop.
However, loop nesting is not currently detected (next step).
2012-11-15 13:44:07 +09:00
Bill Currie
8fb958603c Add set_first.
It returns the first element of a set. If the set is empty, -1 (unsigned)
is returned.
2012-11-15 13:44:07 +09:00
Bill Currie
4cc8d4a04b Fix an apparent snafu in set_test.
It looks like a bit of case&paste where I forgot to change != to &.
2012-11-15 13:44:07 +09:00
Bill Currie
1fead50f4e Calculate the dominators of each node in the flow graph.
The dominators are represented by sets using the node numbers.
2012-11-15 13:44:07 +09:00
Bill Currie
964ea7f9fe Add set_as_string for debug output. 2012-11-15 13:44:07 +09:00
Bill Currie
d2fab28d01 Implement set_is_member.
Oops, forgot that one.
2012-11-15 13:44:06 +09:00
Bill Currie
7853bf1859 Calculate a node's successors and predecessors.
The dot graphs are a little odd (arrow heads on the wrong end of the
predecessor edges), but things seem to be correct.
2012-11-15 13:44:06 +09:00
Bill Currie
739ebc522c Begin working on flow analysis.
For now, variable and flow graph node maps are built.
2012-11-15 13:44:06 +09:00
Bill Currie
f30010b367 Add some docs to function.h 2012-11-15 13:44:06 +09:00
Bill Currie
5acf474882 Double quote dag labels.
Dot interprets escape sequences in non-html string, and needs the quotes to
be escaped, so quote the result of operand_string. Unfortunately,
operand_string uses quote_string and quote_string returns a static pointer,
so some hoop-jumping is necessary.
2012-11-15 13:44:06 +09:00
Bill Currie
cdc790d0c8 Add a set type and support code. 2012-11-15 13:44:06 +09:00
Bill Currie
60d03dbdd9 Fix the naive opcode checks.
Confusing . with .=, and < and <= with more <MOVE etc isn't good.
2012-11-15 13:44:06 +09:00
Bill Currie
dc9d2a982e Make the returned dag a linked list of root-nodes.
A dag can have mulitple "root" nodes (ie, nodes with no parents).
2012-11-15 13:44:06 +09:00
Bill Currie
a22260030a Use value for pointer operands.
This allows daglabels on pointer operands without a redundant daglabel
pointer.
2012-11-15 13:44:06 +09:00
Bill Currie
0c58cd9067 Generate only one daglabel for value operands.
As each value is now unique, their labels can also be unique, which will
improve CSE.
2012-11-15 13:44:06 +09:00
Bill Currie
16869cabd1 Fix a slew of uninitialized data problems.
Bah, structs on stacks... good fun :P
2012-11-15 13:44:06 +09:00