When folding an OpVectorShuffle where the first operand is defined by
an OpVectorShuffle, is unused, and is equal to the second, we end up
with an infinite loop. This is because we think we change the
instruction, but it does not actually change. So we keep trying to
folding the same instruction.
This commit fixes up that specific issue. When the operand is unused,
we replace it with Null.
When folding a vector shuffle that feeds another vector shuffle causes
the size of the first operand to change, when other indices have to be
adjusted reletive to the new size.
The function class provides a {Set|Get}Parent call in order to provide
the context to the LoopDescriptor methods. This CL removes the module
from Function and provides the needed context directly to LoopDescriptor
on creation.
This CL removes the context() method from opt::Function. In the places
where the context() was used we can retrieve, or provide, the context in
another fashion.
Currently the IRContext is passed into the Pass::Process method. It is
then up to the individual pass to store the context into the context_
variable. This CL changes the Run method to store the context before
calling Process which no-longer receives the context as a parameter.
This CL moves the files in opt/ to consistenly be under the opt::
namespace. This frees up the ir:: namespace so it can be used to make a
shared ir represenation.
Currently the utils/ folder uses both spvutils:: and spvtools::utils.
This CL changes the namespace to consistenly be spvtools::utils to match
the rest of the codebase.
The folding routines are currently global functions. They also rely on
data in an std::map that holds the folding rules for each opcode. This
causes that map to not have a clear owner, and therefore never gets
deleted.
There has been a request to delete this map. To implement this, we will
create a InstructionFolder class that owns the maps. The IRContext will
own the InstructionFolder instance. Then the global functions will
become public memeber functions of the InstructionFolder.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1659.
There are a few locations where we need to handle duplicate types. We
cannot merge them because they may be needed for reflection. When this
happens we need do some extra lookups in the type manager.
The specific fixes are:
1) When generating a constant through `GetDefiningInstruction` accept
and use an id for the desired type of the constant. This will make sure
you get the type that is needed.
2) In Private-to-local, make sure we to update the def-use chains when a
new pointer type is created.
3) In the type manager, make sure that `FindPointerToType` returns a
pointer that points to the given type and not a duplicate type.
4) In scalar replacment, make sure the null constants that are created
are the correct type.
Many optimization will run on function scope symbols only. When symbols
are moved from private scope to function scople, then these optimizations
can do more.
I believe it is a good idea to run this pass with both -O and -Os. To
get the most out of it it should be run ASAP after inlining and something
that remove all of the dead functions.
Revert "Don't merge types of resources"
This reverts commit f393b0e480, but leaves
the tests that were added. Added new test. These test are the so that,
if someone tries the same change I made, they will see the test that
they need to handle.
Don't run remove duplicates in -O and -Os
Romve duplicates was run to help reduce compile time when looking for
types in the type manager. I've run compile time test on three sets
of shaders, and the compile time does not seem to change.
It should be safe to remove it.
During the compact IDs optimization pass, the result IDs of some
basic blocks can change. In spite of this, GetPreservedAnalyses
indicated that the CFG was preserved. But the CFG relies on
the basic blocks having the same IDs. Simply removing this flag
resolves the issue by preventing the CFG check.
Also Removes combinators and namemap preserved analyses from
compact IDs pass.
When doing reflection users care about the names of the variable, the
name of the type, and the name of the members. Remove duplicates breaks
this because it removes the names one of the types when merging.
To fix this we have to keep the different types around for each
resource. This commit adds code to remove duplicates to look for the
types uses to describe resources, and make sure they do not get merged.
However, allow merging of a type used in a resource with something not
used in a resource. Was done when the non resource type came second.
This could have a negative effect on compile time, but it was not
expected to be much.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1372.
We replace the std::vector in the Operand class by a new class that does
a small size optimization. This helps improve compile time on Windows.
Tested on three sets of shaders. Trying various values for the small
vector. The optimal value for the operand class was 2. However, for
the Instruction class, using an std::vector was optimal. Size of "0"
means that an std::vector was used.
Instruction size
0 4 8
Operand Size
0 489 544 684
1 593 487
2 469 570
4 473
8 505
This is a single thread run of ~120 shaders. For the multithreaded run
the results were the similar. The basline time was ~62sec. The
optimal configuration was an 2 for the OperandData and an
std::vector for the OperandList with a compile time of ~38sec. Similar
expiriments were done with other sets of shaders. The compile time still
improved, but not as much.
Contributes to https://github.com/KhronosGroup/SPIRV-Tools/issues/1609.
The following passes are updated to preserve the inst-to-block and
def-use analysies:
private-to-local
aggressive dead-code elimination
dead branch elimination
local-single-block elimination
local-single-store elimination
reduce load size
compact ids (inst-to-block only)
merge block
dead-insert elimination
ccp
The one execption is that compact ids still kills the def-use manager.
This is because it changes so many ids it is faster to kill and rebuild.
Does everything in
https://github.com/KhronosGroup/SPIRV-Tools/issues/1593 except for the
changes to merge return.
By using forward pointers, we are able to define a struct that has a
pointer to itself. This could be directly or indirectly. The current
implementation of the type manager did not handle this case. There are
three changes that are made in this commit inorder to handle this case:
1) Change the handling of OpTypeForwardPointer
The current handling of OpTypeForwardsPointer is broken if there is a
reference to the pointer before the real definition. When build the
type that contain the forward delared pointer, the type manager will ask
for the type for that ID, and will get a nullptr because it does not
exists. This nullptr is not handleded very well.
The change is to keep track of the incomplete types the first time
through all of the types. An incomplete type is a ForwardPointer or any
type that references an incomplete type.
Then we implement a second pass through the incomplete types that will
complete them.
2) Hashing types.
When hashing a type, we want to uses all of the subtypes as part of the
hash. However, with types that reference them selves, this creates an
infinite recursion. To get around this, we keep track of which types
have been seen on the path from the root type. If we have see the
current type already then we can stop the recursion.
3) Comparing types.
In order to check if two types are the same, we must check that all of
their subtypes are the same as well. This also causes an infinit
recursion. The solution is to stop comparing the subtypes if we are
trying to compare two pointer types that we are already in the middle of
comparing. The ideas is that if the two pointer are different, then in
progress compare will return false itself.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1578.
We add a new rule to the folding rules to fold an FMix feeding an
extract when the alpha value for the element being extracted is either
0 or 1. In those case, we can simple extract from one of the operands
to the FMix.
With that change the simplification pass completely subsumes the
insert-extract elimination pass. So we remove the insert-extract
elimination passes and replce them with calls to the simplification
pass.
In a follow up PR, we should delete the insert-extract elimination pass.
Contributes to https://github.com/KhronosGroup/SPIRV-Tools/issues/1570.
Currently it's impossible for external code to register a pass because
the only source file that can create pass tokens is optimizer.cpp. This
makes it hard to add passes that can't be upstreamed since you can't run
them from the usual pass sequence without reimplementing Optimizer.
This change adds a PassToken constructor that takes unique_ptr to
opt::Pass; if out-of-tree code implements opt::Pass it can register a
custom pass without having to add it to SPIRV-Tools source code.
Removes the limit on scalar replacement for the lagalization passes.
This is done by adding an option to the pass (and command line option)
to set the limit on maximum size of the composite that scalar
replacement is willing to divide.
Fixes#1494.
ADCE does not treat OpCopyMemory as an instruction that references
memory. Because of that stores are removed that should not be.
This change teaches ADCE that OpCopyMemory and OpCopyMemorySize both
loads from and stores to memory. This will keep other stores live when
needed, and will allows ADCE to remove OpCopyMemory instructions as
well.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1556.
Currently in scalar replacement, we create a new variable for every
memeber of the composite being divided. It is often overkill, because
not all of those members will be used. This change will check which
elements are used and only create variable for the members that are
used.
This reduces the compile time for one set of shader from 248s to 165s.
Part of https://github.com/KhronosGroup/SPIRV-Tools/issues/1494.
The code patterns generated by DXC around function calls can cause many
store to be storing the same value that was just loaded from the same
location:
```
%10 = OpLoad %type %var
OpStore %var %10
```
We want to clean these up very early on because they can cause other
transformations to do a lot of work. For the cases I see, they can be
removed during local-single-block-elim.
For one set of shaders the compile time goes from 248s to 182s. A 26%
improvement.
Part of https://github.com/KhronosGroup/SPIRV-Tools/issues/1494.
We have already disabled common uniform elimination because it created
sequences of loads an entire uniform object, then we extract just a
single element. This caused problems in some drivers, and is just
generally slow because it loads more memory than needed.
However, there are other way to get into this situation, so I've added
a pass that looks specifically for this pattern and removes it when only
a portion of the load is used.
Fixes#1547.
An FClamp instruction forces a values to be within a certain interval.
When the upper or lower bound of the FClamp is a constant and the value
being compared with is a constant, then in some case we can fold the
compared because the entire range is say less than the value.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1549.
If there is a shader with a variable in the workgroup storage class that
is stored to, but not loadeds, then we know nothing will read those
loads. It should be safe to remove them.
This is implemented in ADCE by treating workgroup variables the same
way that private variables are treated.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1550.
At this time, DCE will only remove an instruction if it is a combinator.
However, there are certain non-combinator instructions that can be
safely removed if their results are not used. The derivative
instructions are on example.
We are also missing some instructions from the list of combinators
those are added as the same time.
When doing if-conversion, we do not currently move code out of the side
nodes. The reason for this is that it can increase the number of
instructions that get executed because both side nods will have to be
executed now.
In this commit, we add code to move an instruction, and all of the
instructions it depends on, out of a side node and into the header of
the selection construct. However to keep the cost down, we only do it
when the two values in the OpPhi node compute the same value. This way
we have to move only one of the instructions and the other becomes
unused most of the time. So no real extra cost.
Makes the value number table an alalysis in the ir context.
Added more opcodes to list of code motion safe opcodes.
Fixes#1526.
Previously, the loop class used the terms latch and continue block
interchangeably. This patch splits the two and corrects and tests some
uses of the old uses of GetLatchBlock.
This pass will look for adjacent loops that are compatible and legal to
be fused.
Loops are compatible if:
- they both have one induction variable
- they have the same upper and lower bounds
- same initial value
- same condition
- they have the same update step
- they are adjacent
- there are no break/continue in either of them
Fusion is legal if:
- fused loops do not have any dependencies with dependence distance
greater than 0 that did not exist in the original loops.
- there are no function calls in the loops (could have side-effects)
- there are no barriers in the loops
It will fuse all such loops as long as the number of registers used for
the fused loop stays under the threshold defined by
max_registers_per_loop.
Adds support for spliting loops whose register pressure exceeds a user
provided level. This pass will split a loop into two or more loops given
that the loop is a top level loop and that spliting the loop is legal.
Control flow is left intact for dead code elimination to remove.
This pass is enabled with the --loop-fission flag to spirv-opt.
Track live scalars in VDCE as if they were single element vectors.
Handle the extended instructions for GLSL in VDCE.
Handle composite construct instructions in VDCE.
If one of the operands to an OpVectorTimesScalar instruction is zero,
then the result will be the 0 vector. Currently we do not fold the
insturction unless both operands are constants. This change fixes that.
We also allow folding of OpPhi instructions where the incoming values
are either an OpUndef or the OpPhi instruction itself. As with other
cases, this can be simplified to the OpUndef.
Track live scalars in VDCE as if they were single element vectors.
Handle the extended instructions for GLSL in VDCE.
Handle composite construct instructions in VDCE.
Fixes#1511.
Eliminate unused store to variable if followed by store to same
variable in same block.
Most significantly, this cleans up stores made unused by this pass.
These useless stores can inhibit subsequent optimizations, specifically
LocalSingleStoreElim. Eliminating them makes subsequent optimization more
effective.
The main effect of this pass is to simplify the work done by the SSA
rewriter. It catches many local loads/stores that help speeding up the
work done by the main rewriter.
Introduce a pass that does a DCE type analysis for vector elements
instead of the whole vector as a single element.
It will then rewrite instructions that are not used with something else.
For example, an instruction whose value are not used, even though it is
referenced, is replaced with an OpUndef.
For each function, the analysis determine which SSA registers are live
at the beginning of each basic block and which one are killed at
the end of the basic block.
It also includes utilities to simulate the register pressure for loop
fusion and fission.
The implementation is based on the paper "A non-iterative data-flow
algorithm for computing liveness sets in strict ssa programs" from
Boissinot et al.
The local-single-store-elim algorithm is not fundamentally bad.
However, when there are a large number of variables, some of the
maps that are used can become very large. These large data structures
then take a very long time to be destroyed. I've seen cases around 40%
if the time.
I've rewritten that algorithm to not use as much memory. This give a
significant improvement when running a large number of shader through
DXC.
I've also made a small change to local-single-block-elim to delete the
loads that is has replaced. That way local-single-store-elim will not
have to look at those. local-single-store-elim now does the same thing.
The time for one set goes from 309s down to 126s. For another set, the
time goes from 102s down to 88s.
GCD MIV test as described in Chapter 3 of "Optimizing Compilers for
Modern Architectures: A Dependence-Based Approach" by Randy Allen, and
Ken Kennedy.
Delta test as described in Figure 3 of "Practical Dependence Testing" by
Gina Goff, Ken Kennedy, and Chau-Wen Tseng from PLDI '91.
The unordered_set in ADCE that holds all of the live instructions takes
a very long time to be destroyed. In some shaders, it takes over 40% of
the time.
If we look at the unique ids of the live instructions, I believe they
are dense enough make a simple bit vector a good choice for to hold that
data. When I check the density of the bit vector for larger shaders, we
are usually using less than 4 bytes per element in the vector, and
almost always less than 16.
So, in this commit, I introduce a simple bit vector class, and
use it in ADCE.
This help improve the compile time for some shaders on windows by the
40% mentioned above.
Contributes to https://github.com/KhronosGroup/SPIRV-Tools/issues/1328.
For each loop in a function, the pass walks the loops from inner to outer most loop
and tries to peel loop for which a certain amount of iteration can be done before or after the loop.
To limit code growth, peeling will not happen if the growth in code size goes above a configurable threshold.
Provides functionality to perform ZIV and SIV dependency analysis tests
between a load and store within the same loop.
Dependency tests rely on scalar analysis to prove and disprove dependencies
with regard to the loop being analysed.
Based on the 1990 paper Practical Dependence Testing by Goff, Kennedy, Tseng
Adds support for marking loops in the loop nest as IRRELEVANT.
Loops are marked IRRELEVANT if the analysed instructions contain
no induction variables for the loops, i.e. the loops induction
variable is not relevent to the dependence of the store and load.
Adding three rules to fold OpDot (implemented as two).
- When an OpDot has two constants, then fold to the resulting const.
- When one of the inputs is the 0 vector, then fold to zero.
- When one of the inputs is a single 1 with 0s, then rewrite to an
OpCompositeExtract of the appropriate element. This will help find
even more folding opportunities.
Contributes to #709.
From the test case, the slice of the CFG that is interesting for the bug
is
25
|
v
30
|
v
31<-+
| |
v |
34--+
1. In block 25, we have a Phi candidate for %f with arguments
%47 = Phi[%float_0, %0]. This merges %float_0 and a yet unknown
argument from the external loop backedge.
2. We are now processing block 34:
i. The load %35 = OpLoad %f triggers a Phi candidate to be placed in
block 31.
ii. The Phi candidate %50 = Phi needs two arguments. The one coming
from block 30 is %47. But the one coming from block 34 (which we
are now processing and have marked sealed), finds %50 itself as
the reaching def for %f.
3. This wrongfully marks %50 as a copy-of Phi, which ultimately makes
both %47 and %50 copy-of Phis that get eliminated.
Currently OpImageTexelPointer operations are treat like a use of the
pointer, but it does
not look for the memory being referenced to make sure stores are not
removed.
This change teaches it so identify the memory being accessed, and
treats it as if that memory is loaded.
Fixes to #1445.
OpImageTexelPointer acts like a special kind of load. It is not an
array load, but it also cannot be removed the same way a regular
load can. The type of propagation that needs to be done is similar
to what we do for arrays, so I want to merge that code into that
optmization.
Contributers to #1445.
OpImageTexelPointer acts like a special kind of load. It is still
safe to change the storage class of a variable used in a
OpImageTexalPointer instruction.
Contributes to #1445.
When the original code copies an entire array or struct one element at a
time, this turns into a series of OpCompositeInsert instruction followed
by a store of the whole array. We currently miss opportunities in copy
propagate arrays because we do not recognize this as a copy.
This commit adds code to copy propagate arrays to identify this code
pattern.
Also updates the performance passed to run array copy propagation.
The first implementation of MemroyObject, which is used in copy
propagate arrays, forced the access chain to be like the access chains
in OpCompositeExtract. This excluded the possibility of the memory
object from representing an array element that was extracted with a
variable index. Looking at the code, that restriction is not
neccessary. I also see some opportunities for doing this in some real
shaders.
Contributes to #1430.
This patch adds support for the analysis of scalars in loops. It works
by traversing the defuse chain to build a DAG of scalar operations and
then simplifies the DAG by folding constants and grouping like terms.
It represents induction variables as recurrent expressions with respect
to a given loop and can simplify DAGs containing recurrent expression by
rewritting the entire DAG to be a recurrent expression with respect to
the same loop.
When we change the type of an object that gets stored, we do not want to
change the type of the memory location being stored to. In order to
still be able to do the rewrite, we will decompose and rebuild the
object so it is the type that can be stored.
Fixes#1416.
The sprir-v generated from HLSL code contain many copyies of very large
arrays. Not only are these time consumming, but they also cause
problems for drivers because they require too much space.
To work around this, we will implement an array copy propagation. Note
that we will not implement a complete array data flow analysis in order
to implement this. We will be looking for very simple cases:
1) The source must never be stored to.
2) The target must be stored to exactly once.
3) The store to the target must be a store to the entire array, and be a
copy of the entire source.
4) All loads of the target must be dominated by the store.
The hard part is keeping all of the types correct. We do not want to
have to do too large a search to update everything, which may not be
possible, do we give up if we see any instruction that might be hard to
update.
Also in types.h, the element decorations are not stored in an std::map.
This change was done so the hashing algorithm for a Struct is
consistent. With the std::unordered_map, the traversal order was
non-deterministic leading to the same type getting hashed to different
values. See |Struct::GetExtraHashWords|.
Contributes to #1416.
This patch adds a new option --time-report to spirv-opt. For each pass
executed by spirv-opt, the flag prints resource utilization for the pass
(CPU time, wall time, RSS and page faults)
This fixes issue #1378
This pass replaces the load/store elimination passes. It implements the
SSA re-writing algorithm proposed in
Simple and Efficient Construction of Static Single Assignment Form.
Braun M., Buchwald S., Hack S., Leißa R., Mallon C., Zwinkau A. (2013)
In: Jhala R., De Bosschere K. (eds)
Compiler Construction. CC 2013.
Lecture Notes in Computer Science, vol 7791.
Springer, Berlin, Heidelberg
https://link.springer.com/chapter/10.1007/978-3-642-37051-9_6
In contrast to common eager algorithms based on dominance and dominance
frontier information, this algorithm works backwards from load operations.
When a target variable is loaded, it queries the variable's reaching
definition. If the reaching definition is unknown at the current location,
it searches backwards in the CFG, inserting Phi instructions at join points
in the CFG along the way until it finds the desired store instruction.
The algorithm avoids repeated lookups using memoization.
For reducible CFGs, which are a superset of the structured CFGs in SPIRV,
this algorithm is proven to produce minimal SSA. That is, it inserts the
minimal number of Phi instructions required to ensure the SSA property, but
some Phi instructions may be dead
(https://en.wikipedia.org/wiki/Static_single_assignment_form).
The loop peeler util takes a loop as input and create a new one before.
The iterator of the duplicated loop then set to accommodate the number
of iteration required for the peeling.
The loop peeling pass that decided to do the peeling and profitability
analysis is left for a follow-up PR.
We are seeing shaders that have multiple returns in a functions. These
functions must get inlined for legalization purposes; however, the
inliner does not know how to inline functions that have multiple
returns.
The solution we will go with it to improve the merge return pass to
handle structured control flow.
Note that the merge return pass will assume the cfg has been cleanedup
by dead branch elimination.
Fixes#857.
Strips reflection info. This is limited to decorations and
decoration instructions related to the SPV_GOOGLE_hlsl_functionality1
extension.
It will remove the OpExtension for SPV_GOOGLE_hlsl_functionality1.
It will also remove the OpExtension for SPV_GOOGLE_decorate_string
if there are no further remaining uses of OpDecorateStringGOOGLE.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1398
Optimizations should work in the presence of recent
SPV_GOOGLE_decorate_string and SPV_GOOGLE_hlsl_functionality1
SPV_GOOGLE_decorate_string:
- Adds operation OpDecorateStringGOOGLE to decorate an object with decorations
having string operands.
SPV_GOOGLE_hlsl_functionality1:
- Adds HlslSemanticGOOGLE, used to decorate an interface variable with
an HLSL semantic string. Optimizations already preserve those variables
as required because they are interface variables (with uses), independent
of whether they have HLSL decorations.
- Adds HlslCounterBufferGOOGLE, used to associate a buffer with a
counter variable.
Fixes#1391
This reimplementation fixes several issues when removing decorations associated
to an ID (partially addresses #1174 and gives tools for fixing #898), as well
as making it easier to remove groups; a few additional tests have been added.
DecorationManager::RemoveDecoration() will still not delete dead decorations it
created, but I do not think it is its job either; given the following input
```
OpCapability Shader
OpCapability Linkage
OpMemoryModel Logical GLSL450
OpDecorate %2 Restrict
%2 = OpDecorationGroup
OpGroupDecorate %2 %1 %3
OpDecorate %4 Invariant
%4 = OpDecorationGroup
OpGroupDecorate %4 %2
%uint = OpTypeInt 32 0
%1 = OpVariable %uint Uniform
%3 = OpVariable %uint Uniform
```
which of the following two outputs would you expect RemoveDecoration(2) to produce:
```
OpCapability Shader
OpCapability Linkage
OpMemoryModel Logical GLSL450
%uint = OpTypeInt 32 0
%1 = OpVariable %uint Uniform
%3 = OpVariable %uint Uniform
```
or
```
OpCapability Shader
OpCapability Linkage
OpMemoryModel Logical GLSL450
OpDecorate %4 Invariant
%4 = OpDecorationGroup
%uint = OpTypeInt 32 0
%1 = OpVariable %uint Uniform
%3 = OpVariable %uint Uniform
```
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/924
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1174
Remove extension whitelists from transforms that are essentially
combinatorial (and avoiding pointers) or which affect only control flow.
It's very very unlikely an extension will add a new control flow construct.
Remove from:
- dead branch elimination
- dead insertion elimination
- insert extract elimination
- block merge
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/1392
The merging types we do not remove other information related to the
types. We simply leave it duplicated, and hope it is removed later.
This is what happens with decorations. They are removed in the next
phase of remove duplicates. However, for OpNames that is not the case.
We end up with two different names for the same id, which does not make
sense.
The solution is to remove the names and decorations for the type being
removed instead of rewriting them to refer to the other type.
Note that it is possible that if the first type does not have a name,
then the types will end up with no name. That is fine because the names
should not have any semantic significance anyway.
The was identified in issue #1372, but this does not fix that issue.
* Also mark function parameters as varying
* Conservatively mark assignment instructions as varying if any input is
varying after attempting to fold
* Added a test to catch this case
* getFloatConstantKind() now handles OpConstantNull
* PerformOperation() now handles OpConstantNull for vectors
* Fixed some instances where we would attempt to merge a division by 0
* added tests
The algorithm used in DCEInst to remove dead code is very slow. It is
fine if you only want to remove a small number of instructions, but, if
you need to remove a large number of instructions, then the algorithm in
ADCE is much faster.
This PR removes the calls to DCEInst in the load-store removal passes
and adds a pass of ADCE afterwards.
A number of different iterations of the order of optimization, and I
believe this is the best I could find.
The results I have on 3 sets of shaders are:
Legalization:
Set 1: 5.39 -> 5.01
Set 2: 13.98 -> 8.38
Set 3: 98.00 -> 96.26
Performance passes:
Set 1: 6.90 -> 5.23
Set 2: 10.11 -> 6.62
Set 3: 253.69 -> 253.74
Size reduction passes:
Set 1: 7.16 -> 7.25
Set 2: 17.17 -> 16.81
Set 3: 112.06 -> 107.71
Note that the third set's compile time is large because of the large
number of basic blocks, not so much because of the number of
instructions. That is why we don't see much gain there.
Use indirection through latest_version_spirv.h
Also, when generating enum tables, use the unified1 JSON grammar since
it now has FragmentFullyCoveredEXT but the other JSON grammars don't.
They are starting to fall behind.
Adding basis of arithmetic merging
* Refactored constant collection in ConstantManager
* New rules:
* consecutive negates
* negate of arithmetic op with a constant
* consecutive muls
* reciprocal of div
* Removed IRContext::CanFoldFloatingPoint
* replaced by Instruction::IsFloatingPointFoldingAllowed
* Fixed some bad tests
* added some header comments
Added PerformIntegerOperation
* minor fixes to constants and tests
* fixed IntMultiplyBy1 to work with 64 bit ints
* added tests for integer mul merging
Adding test for vector integer multiply merging
Adding support for merging integer add and sub through negate
* Added tests
Adding rules to merge mult with preceding divide
* Has a couple tests, but needs more
* Added more comments
Fixed bug in integer division folding
* Will no longer merge through integer division if there would be a
remainder in the division
* Added a bunch more tests
Adding rules to merge divide and multiply through divide
* Improved comments
* Added tests
Adding rules to handle mul or div of a negation
* Added tests
Changes for review
* Early exit if no constants are involved in more functions
* fixed some comments
* removed unused declaration
* clarified some logic
Adding new rules for add and subtract
* Fold adds of adds, subtracts or negates
* Fold subtracts of adds, subtracts or negates
* Added tests
It moves all conditional branching and switch whose conditions are loop
invariant and uniform. Before performing the loop unswitch we check that
the loop does not contain any instruction that would prevent it
(barriers, group instructions etc.).
In some shaders there are a lot of very large and deeply nested
structures. This creates a lot of work for scalar replacement. Also,
since commit ca4457b we have been very aggressive as rewriting
variables. This has causes a large increase in compile time in creating
and then deleting the instructions.
To help low the costs, I want to run a cleanup of some of the easy loads
and stores to remove. This reduces the number of symbols sroa has to
work on. It also reduces the amount of code the simplifier has to
simplify because it was not generated by sroa.
To confirm the improvement, I ran numbers on three different sets of
shaders:
Time to run --legalize-hlsl:
Set #1: 55.89s -> 12.0s
Set #2: 1m44s -> 1m40.5s
Set #3: 6.8s -> 5.7s
Time to run -O
Set #1: 18.8s -> 10.9s
Set #2: 5m44s -> 4m17s
Set #3: 7.8s -> 7.8s
Contributes to #1328.
Fixes a bug at the same time. In `UpdateDefUse`, if the definition
already exists, we are not suppose to analyse it again. When you do
the entries for the definition are deleted, and we don't want that.
The check for this was wrong.
This function now checks for side-effects before adding operand
instructions to the dead instruction work list.
Because this fix puts more pressure on IsCombinatorInstruction() to
be correct, this commit adds all OpConstant* and OpType* instructions
to combinator_ops_ set.
Fixes#1341.
When inlining a function call the instructions in the same basic block
as the call get cloned. The clone is added to the set of new blocks
containing the inlined code, and the original instructions are deleted.
This PR will change this so that we simply move the instructions to the
new blocks. This saves on the creation and deletion of the
instructions.
Contributes to #1328.
This change implements instruction folding for arithmetic operations
that are redundant, specifically:
x + 0 = 0 + x = x
x - 0 = x
0 - x = -x
x * 0 = 0 * x = 0
x * 1 = 1 * x = x
0 / x = 0
x / 1 = x
mix(a, b, 0) = a
mix(a, b, 1) = b
Cache ExtInst import id in feature manager
This allows us to avoid string lookups during optimization; for now we
just cache GLSL std450 import id but I can imagine caching more sets as
they become utilized by the optimizer.
Add tests for add/sub/mul/div/mix folding
The tests cover scalar float/double cases, and some vector cases.
Since most of the code for floating point folding is shared, the tests
for vector folding are not as exhaustive as scalar.
To test sub->negate folding I had to implement a custom fixture.
Building the def-use chains is very expensive, so we do not want to
invalidate them it if is not necessary. At the moment, it seems like
most optimizatoins are good at not invalidating the def-use chains, but
simplification does.
This PR get the simlification pass to keep the analysies valid.
Contributes to #1328.
On some shader code we have in our testsuite, Phi insertion is showing
massive compile time slowdowns, particularly during destruction. The
specific shader I was looking at has about 600 variables to keep track
of and around 3200 basic blocks. The algorithm is currently O(var x
blocks), which means maps with around 2M entries. This was taking about
8 minutes of compile time.
This patch changes the tracking of stored variables to be more sparse.
Instead of having every basic block contain all the tracked variables in
the map, they now have only the variables actually stored in that block.
This speeds up deallocation, which brings down compile time to about
1m20s.
Note that this is not the definite fix for this. I will re-write Phi
insertion to use a standard SSA rewriting algorithm
(https://github.com/KhronosGroup/SPIRV-Tools/issues/893).
This contributes to
https://github.com/KhronosGroup/SPIRV-Tools/issues/1328.
I mixed up two cases when folding an OpCompositeExtract that is feed by
and OpCompositeInsert. The specific cases are demonstracted in the new
test. I mixed up the conditions for the cases, and treated one like the
other.
Fixes#1323.
* Now track propagation status and assert on bad statuses
* Added helper methods to access instruction propagation status
* Modified the phi meet operator to properly reflect the paper it is
based on
* Modified SSA edge addition so that all edge are added, but only on
state changes
* Fixed a bug in instruction simulation where interesting conditional
branches would not mark the interesting edge as executed
* Added a test to catch this bug
* Added an ostream operator for SSAPropagator::PropStatus
The simplification pass works better after all of the dead branches are
removed. So swapping them around in the legalization passes. Also
adding the simplification pass to performance passes right after dead
branch elimination.
Added CCP to the legalization passes so we can propagate the constants
into the branchs, and remove as many branches a possible. CCP is
designed to still get opportunities even if the branches are dead, so it
is a good place for it.
Fixes#1118
This change handles all 6 regular comparison types in two variations,
ordered (true if values are ordered *and* comparison is true) and
unordered (true if values are unordered *or* comparison is true).
Ordered comparison matches the default floating-point behavior on host
but we use std::isnan to check ordering explicitly anyway.
This change also slightly reworks the floating-point folding support
code to make it possible to define a folding operation that returns
boolean instead of floating point.
These tests exhaustively test ordered/unordered comparisons for
float/double.
Since for NaN inputs the comparison result doesn't depend on the
comparison function, we just test == and !=; NaN inputs result in true
unordered comparisons and false ordered comparisons.
Registering a constant in constant manager establishes a relation
between instruction that defined it and constant object. On complex
shaders this could result in the constant definition getting removed as
part of one of the DCE pass, and a subsequent simplification pass trying
to use the defining instruction for the constant.
To fix this, we now remove associated constant entries from constant
manager when killing constant instructions; the constant object is still
registered and can be remapped to a new instruction later.
GetDefiningInstruction shouldn't ever return nullptr after this change
so add an assertion to check for that.
In dead branch elimination, we already recognize unreachable continue
blocks, and update OpPhi instruction accordingly. This change adds an
extra check: if the head block has exactly 1 other incoming edge, then
replace the OpPhi with the value from that edge.
Fixes#1314.
unordered_map is not POD. Using it as static may cause problems
when operator new() and operator delete() is customized.
Also changed some function signatures to use const char* instead
of std::string, which will give caller the flexibility to avoid
creating a std::string.
We can fold OpSelect into one of the operands in two cases:
- condition is constant
- both results are the same
Even if the original shader doesn't have either of these, if-conversion
pass sometimes ends up generating instructions like
%7127 = OpSelect %int %3220 %7058 %7058
And this optimization cleans them up.
Adding a map from an id to it set of OpName and OpMemberName
instructions. This will be used in KillNameAndDecorates to kill the
names for the ids that are being removed.
In my test, the compile time for 50 shaders went from 1m57s to 55s.
This was on linux using the release build.
Fixes#1290.
This patch adds initial support for loop unrolling in the form of a
series of utility classes which perform the unrolling. The pass can
be run with the command spirv-opt --loop-unroll. This will unroll
loops within the module which have the unroll hint set. The unroller
imposes a number of requirements on the loops it can unroll. These are
documented in the comments for the LoopUtils::CanPerformUnroll method in
loop_utils.h. Some of the restrictions will be lifted in future patches.
There seems to only be a single location where the def-use manager is
used. It is to get information about a type. We can do that with the
type manager instead.
Fixes#1285
Implementation of the simplification pass.
- Create pass that calls the instruction folder on each instruction and
propagate instructions that fold to a copy. This will do copy
propagation as well.
- Did not use the propagator engine because I want to modify the instruction
as we go along.
- Change folding to not allocate new instructions, but make changes in
place. This change had a big impact on compile time.
- Add simplification pass to the legalization passes in place of
insert-extract elimination.
- Added test cases for new folding rules.
- Added tests for the simplification pass
- Added a method to the CFG to apply a function to the basic blocks in
reverse post order.
Contributes to #1164.
