Commit Graph

17 Commits

Author SHA1 Message Date
Hans-Kristian Arntzen
fc9fe4e480 Fix variable scope when an if or else block dominates a variable.
Just like loops, we need complicated hoisting again to make this work.
2019-07-03 11:18:50 +02:00
Hans-Kristian Arntzen
bf56dc88b9 Rewrite how loop dominators are propagated.
Do this analysis in the CFG stage rather than last minute with the
ad-hoc algorithm we had in place before CFG was introduced.
2019-06-06 12:17:46 +02:00
Hans-Kristian Arntzen
03d93abc1a Deal with case where a variable is dominated by inner part of a loop.
There is a risk that we try to preserve a loop variable through multiple
iterations, even though the dominating block is inside a loop.

Fix this by analyzing if a block starts off by writing to a variable. In
that case, there cannot be any preservation going on. If we don't, pretend the
loop header is reading the variable, which moves the variable to an
appropriate scope.
2019-06-06 11:11:44 +02:00
Hans-Kristian Arntzen
a489ba7fd1 Reduce pressure on global allocation.
- Replace ostringstream with custom implementation.
  ~30% performance uplift on vector-shuffle-oom test.
  Allocations are measurably reduced in Valgrind.

- Replace std::vector with SmallVector.
  Classic malloc optimization, small vectors are backed by inline data.
  ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux.

- Use an object pool for IVariant type.
  We generally allocate a lot of SPIR* objects. We can amortize these
  allocations neatly by pooling them.

- ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-09 15:09:44 +02:00
Hans-Kristian Arntzen
9b92e68d71 Add an option to override the namespace used for spirv_cross.
This is a pragmatic trick to avoid symbol collision where a project
links against SPIRV-Cross statically, while linking to other projects
which also use SPIRV-Cross statically. We can end up with very awkward
symbol collisions which can resolve themselves silently because
SPIRV-Cross is pulled in as necessary. To fix this, we must use
different symbols and embed two copies of SPIRV-Cross in this scenario,
now with different namespaces, which in turn leads to different symbols.
2019-03-29 10:29:44 +01:00
Hans-Kristian Arntzen
57be90321b Use hashmaps in CFG as well.
The arrays are incredibly sparse for the most part as we only need
entires per basic block.

For a small shader with ID bound of 8 million, we now have about 30x
uplift.
2019-01-10 14:35:34 +01:00
Hans-Kristian Arntzen
318c17cbb2 Nonfunctional: Update copyright headers for 2019. 2019-01-04 12:38:35 +01:00
Hans-Kristian Arntzen
382101bd05 Run format_all.sh. 2018-04-04 09:26:53 +02:00
Bill Hollings
1c94715350 Update copyright dates to 2018 in main files. 2018-01-31 17:08:43 -05:00
Hans-Kristian Arntzen
b3f6e3de8e Fix CFG::update_common_dominator.
The algorithm was too conservative causing lots of unnecessary
temporaries to be created.
2018-01-24 20:32:11 +01:00
Hans-Kristian Arntzen
a4fafa0607 Fix edge-case where do/while body is a dominator.
Shows up quite a lot in inlined code, should get more test coverage in
this area ...
2017-08-02 11:58:24 +02:00
Hans-Kristian Arntzen
b2c2e6483b Analyze parameter preservation for functions.
This is kinda tricky, because if we only conditionally write to a
function parameter variable it is implicitly preserved in SPIR-V, so we must force
an in qualifier on the parameter to get the same behavior in GLSL.
2017-03-25 16:25:30 +01:00
Hans-Kristian Arntzen
18c37bcdab Update license headers for 2017. 2017-01-28 09:00:40 +01:00
Hans-Kristian Arntzen
5ff11cc689 Handle more corner cases with the CFG traversal. 2016-11-18 16:45:11 +01:00
Hans-Kristian Arntzen
0c9683cd85 Distinguish between cross edges and back edges.
Handle the weird opcodes CopyMemory/CopyObject.
2016-11-18 09:59:54 +01:00
Hans-Kristian Arntzen
edbe867b74 Improve handling of dead blocks.
All tests pass now.
Still need to handle OpFunction, atomics, OpPhi, etc ...
2016-11-17 22:15:07 +01:00
Hans-Kristian Arntzen
dad4a34072 Start adding CFG analysis.
Not complete yet, but partly working ...
2016-11-17 16:58:05 +01:00