Implemented AreEquivalentConstants method to check equivalency of
constants, changing IsApplicable method of
TransformationRecordSynonymousConstants to allow recording equivalence
of composite constants; added some tests to check this.
Tests with arrays and matrices still need to be added.
Fixes#3533.
Add TransformationAddRelaxedDecoration, which adds the RelaxedPrecision decoration to ids of numeric instructions (those yielding 32-bit ints or floats) in dead blocks.
Fixes#3502
Now that WebGPU ingests WGSL instead of SPIR-V,
there is no need to be so strict about the memory model.
Allow any memory model that is already allowed by Vulkan 1.0,
either directly or via an existing.
Fixes#3529
* Make BasicBlock::reachable() only consider static reachability
* Fix reachability calculation to be independent of block order
* add tests
This pass basically follows the same process as ssa-rewrite: it adds a DebugValue after each Store and removes the DebugDeclare or DebugValue Deref. It only does this if all instructions that are dependent on the Store are Loads and are replaced.
This commit lets the vector DCE pass preserve the OpenCL.DebugInfo.100
information properly. When the vector DCE pass determines the liveness
of instructions, the debug instructions must not affect the decision. In
addition, when it kills some instructions, it has to kill DebugValue
instructions that use the killed instructions. When it updates some
composite values to meaningful values (not undef), it has to remove
DebugValue because the value information becomes incorrect.
The decision to reduce the load must be not affected by debug
instructions. For example, even when a DebugValue references a
result id of a loaded composite value, this change lets the
reduce-load-size pass reduce the load if the full composite value is not
used anywhere other than the DebugValue.
When the pass replaces the local variable `OpVariable` ids to their
corresponding pointers, we have to update operands of DebugValue or
DebugDeclare instructions.
When there are multiple entries and the shader has a variable with
WorkGroup storage class, those multiple entry functions store values to
the variable. Since ADCE pass uses def-use chains to propagate the work
list, some of instructions in the work list are not actually a part of
the currently processed function. As a result, it adds instructions in
other functions and put them in |live_insts_|. However, it does not
have the control flow information for those instructions in other
functions i.e., |block2headerBranch_| and |header2nextHeaderBranch_|.
When it processes those instructions (they are added when it processes a
different function), it skips handling them because they are already in
|live_insts_| and does not check |block2headerBranch_| and
|header2nextHeaderBranch_|, which results in skipping some branches.
Even though those branches are live branches, it considers they are dead
branches.
For many spirv-opt passes such as simplify-instructions pass, we have to
correctly clear the OpenCL.DebugInfo.100 debug information for
KillInst() and ReplaceAllUses(). If we keep some debug information that
disappeared because of KillInst() and ReplaceAllUses(), adding new
DebugValue instructions based on the existing DebugDeclare information
will generate incorrect information. This CL update DebugInfoManager
and IRContext to correctly clear debug information.
* Add validation that input/output locations are assigned correctly
* Account for component assignment
* Account for 4 components per location and track the combined
coordinate
* Account for multiple output indexes
* handle specifically arrayed variables
* Arrayed variables that specify a component get locations determined
per index of the array for the sub type
* Added tests that check locations and components can be assigned
to interleave an array's locations and components
* Fix up which interfaces are allowed to be arrayed for various shader
stages based on glslang
* Support load and extract pattern in desc_sroa.
* Fix typo in comments.
* Load replacement var before use; and added test.
* fix formatting
* Address code review comments.
Add OpenCL.DebugInfo.100 `DebugValue` instructions for store
and phi instructions of local variables to provide the debugger with
the updated values of local variables correctly.
* Debug info preservation in dead branch elimination
The DeadBranchElimPass class already handles the OpenCL.DebugInfo.100
instructions correctly. This commit just adds a unit test to make sure
it preserves the information properly.
* Add unit test of ReplaceAllWith for debug instruction
* Debug info preservation in ccp pass
For constant propagation, the ccp pass already replaces the result id of
a value with a result id of the corresponding constant value. As a part
of the replacement, it correctly updates the operands of
DebugValue/DebugDeclare as well. Since we do not have to any addition
work other than the ccp pass itself, this commit just adds unit tests to
check the debug information preservation.
Need to actually instantiate them with TEST_P in the same source file.
Remove the instantiation of RoundTripTest from unit.cpp because
it's viewed as having no instantiations.
ImageGatherBiasLodAMD makes it possible to use the Bias and Lod image
operands with OpImageGather and OpImageSparseGather. This commit makes
sure the validator checks for that capability before reporting errors
and adds a few positive tests.
Handles the OpenCL100Debug extension in inlining. It preserves the information that is available while also adding the debug inlined at for all of the inlining that it does.
Reject folding comparisons with unfoldable types.
Fixes#3343
When CCP is evaluating an instruction, it was trying to fold a
comparison with 64 bit integers. This was causing a fold failure later
since the folder still cannot deal with 64 bit integers.
ssa-rewrite fails in `MemPass::GetPtr()` when the SPIR-V code contains
`OpLoad` for the result id of `OpConstantNull` because of the out of
index access for an operand to get the base address. This commit fixes
it.
Fixes#3344
In this PR, the classes that represent the adjust branch weights
transformation and fuzzer pass were implemented. This transformation
adjusts the branch weights of a OpBranchConditional instruction.
- No longer inline functions with early exits. Merge return can modify them so they can be inlined.
- Otherwise no functional change, should be just refactoring.
Extends the pass for removing unused instructions so that it can
remove global declarations (such as types and variables) that are only
used by decorations with which they are intimately connected, such as
descriptor set and binding decorations.
* Do merge return if the return is not at the end of the function.
We will remove the code in inlining to handle a return in the middle of
a function. To inline those functions, we need to run merge return to
move the return to the end of the function.
Generalizes the IsReadOnlyVariable() method, and related methods, so
that they can be used to ask whether pointer result ids are read-only.
Fixes#3324.
Many high-level languages like HLSL and GLSL generate termination
instructions such as return and branch from the actual part of the
high-level language code like return and if statements. This commit lets
IrLoader set `DebugScope` for termination instructions.
The outliner would outline regions ending with a loop header, making
the block containing the call to the outlined function serve as the
loop header. This, however, is incorrect in general, since the whole
outlined function -- rather than just the exit block for the region --
would end up getting called every time the loop would iterate.
This change restricts the outliner so that the last block in a region
cannot be a loop header.
We need an analysis for OpenCL.DebugInfo.100 extension instructions such
as a map between function id and its DebugFunction. This commit add an
analysis for it.
It has been resolved that statically out-of-bounds accesses are not
invalid in SPIR-V (they lead to undefind behaviour at runtime but
should not cause a module to be rejected during validation). This
change tolerates such accesses in donated code, clamping them in-bound
as part of making a function live-safe.
The Sample argument of OpImageTexelPointer is sometimes required to be
a zero constant. It thus cannot be replaced with a synonym in
general. This change avoids replacing this argument with a synonym.
The fact manager maintains an equivalence relation on data descriptors
that tracks when one data descriptor could be used in place of
another. An algorithm to compute the closure of such facts allows
deducing new synonym facts from existing facts. E.g., for two 2D
vectors u and v it is known that u.x is synonymous with v.x and u.y is
synonymous with v.y, it can be deduced that u and v are synonymous.
The closure computation algorithm is very expensive if we get large
equivalence relations.
This change addresses this in three ways:
- The size of equivalence relations is reduced by limiting the extent
to which the components of a composite are recursively noted as
being equivalent, so that when we have large synonymous arrays we do
not record all array elements as being pairwise equivalent.
- When computing the closure of facts, equivalence classes above a
certain size are simply skipped (which can lead to missed facts)
- The closure computation is performed less frequently - it is invoked
explicitly before fuzzer passes that will benefit from data synonym
facts. A new transformation is used to control its invocation, so
that fuzzing and replaying do not get out of sync.
The change also tidies up the order in which some getters are declared
in FuzzerContext.
Adds an extra condition on when a region can be outlined to avoid the
case where a region ends with a loop head but such that the loop's
continue target is in the region. (Outlining such a region would mean
that the loop merge is in the original function and the continue target
in the outlined function.)
The function outliner uses a struct to return ids that a region
generates and that are used outside that region. If these ids have
pointer type this would result in a struct with pointer members, which
leads to illegal loading from non-logical pointers if logical
addressing is used. This change bans that outlining possibility.
Provides support for runtime arrays in the code that traverses
composite types when checking applicability of transformations that
replace ids with synonyms.
Demotes the image storage class to Private during donation. Also
fixes an issue where instructions that depended on non-donated global
values would not be handled properly.
The SPIR-V data rules say that all uses of an OpSampledImage
instruction must be in the same block as the instruction, and highly
restrict those instructions that can consume the result id of an
OpSampledImage.
This adapts the transformations that split blocks and create synonyms
to avoid separating an OpSampledImage use from its definition, and to
avoid synonym-creation instructions such as OpCopyObject consuming an
OpSampledImage result id.
* Preserve debug info in eliminate-dead-functions
The elimination of dead functions makes OpFunction operand of
DebugFunction invalid. This commit replaces the operand with
DebugInfoNone.
* Handle more cases in dead member elim
- Rewrite composite insert and extract operations on SpecConstnatOp.
- Leaves assert for Access chain instructions, which are only allowed
for kernels.
- Other operations do not require any extra code will no longer cause an
assert.
Fixes#3284.
Fixes#3282.
The management of equation facts suffered from two problems:
(1) The processing of an equation fact required the data descriptors
used in the equation to be in canonical form. However, during
fact processing it can be deduced that certain data descriptors
are equivalent, causing their equivalence classes to be merged,
and that could cause previously canonical data descriptors to no
longer be canonical.
(2) Related to this, if id equations were known about a canonical data
descriptor dd1, and other id equations known about a different
canonical data descriptor dd2, the equation facts about these data
descriptors were not being merged in the event that dd1 and dd2
were deduced to be equivalent.
This changes solves (1) by not requiring equation facts to be in
canonical form while processing them, but instead always checking
whether (not necessary canonical) data descriptors are equivalent when
looking for corollaries of equation facts, rather than comparing them
using ==.
Problem (2) is solved by adding logic to merge sets of equations when
data descriptors are made equivalent.
In addition, the change also requires elements to be registered in an
equivalence relation before they can be made equivalent, rather than
being added (if not already present) at the point of being made
equivalent.
This change increases the extent to which arbitrary SPIR-V can be used
by the fuzzer pass that donates modules. It handles the case where
various ingredients (such as types, variables and particular
instructions) cannot be donated by omitting them, and then either
omitting their dependencies or replacing their dependencies with
alternative instructions.
The change pays particular attention to allowing code that manipulates
image types to be handled (by skipping anything image-specific).
(1) Runtime arrays are turned into fixed-size arrays, by turning
OpTypeRuntimeArray into OpTypeArray and uses of OpArrayLength into
uses of the constant used for the length of the fixed-size array.
(2) Atomic instructions are not donated, and uses of their results are
replaced with uses of constants of the result type.
The fuzzer pass that constructs composites had an issue where it would
regard isomorphic but distinct structs (similarly arrays) as being
interchangeable when constructing composites. This change fixes the
problem by relying less on the type manager.
Some transformations (e.g. TransformationAddFunction) rely on running
the validator to decide whether the transformation is applicable. A
recent change allowed spirv-fuzz to take validator options, to cater
for the case where a module should be considered valid under
particular conditions. However, validation during the checking of
transformations had no access to these validator options.
This change introduced TransformationContext, which currently consists
of a fact manager and a set of validator options, but could in the
future have other fields corresponding to other objects that it is
useful to have access to when applying transformations. Now, instead
of checking and applying transformations in the context of a
FactManager, a TransformationContext is used. This gives access to
the fact manager as before, and also access to the validator options
when they are needed.
When DebugScope is given in SPIR-V, each instruction following the
DebugScope is from the lexical scope pointed by the DebugScope in
the high level language. We add DebugScope struction to keep the
scope information in Instruction class. When ir_loader loads
DebugScope/DebugNoScope, it keeps the scope information in
|last_dbg_scope_| and lets following instructions have that scope
information.
In terms of DebugDeclare/DebugValue, if it is in a function body
but outside of a basic block, we keep it in |debug_insts_in_header_|
of Function class. If it is in a basic block, we keep it as a normal
instruction i.e., in a instruction list of BasicBlock.
Create a pass to instrument OpDebugPrintf instructions. This pass replaces all OpDebugPrintf instructions with instructions to write a record containing the string id and the all specified values into a special printf output buffer (if space allows). This pass is designed to support the printf validation in the Vulkan validation layers.
Fixes#3210
In this PR, the classes that represent the toggle access chain
instruction transformation and fuzzer pass were implemented. This
transformation toggles the instructions OpAccessChain and
OpInBoundsAccessChain between them.
Fixes#3193.
This introduces a new fuzzer pass to add instructions to the module
that define equations, and support in the fact manager for recording
equation facts and deducing synonym facts from equation facts.
Initially the only equations that are supported involve OpIAdd,
OpISub, OpSNegate and OpLogicalNot, but there is scope for adding
support for equations over various other operators.
According to SPV_AMD_shader_image_load_store_lod spec, Lod operand is
valid with OpImageRead, OpImageWrite, or OpImageSparseRead if the
extension is enabled.
This change adds a fuzzer pass that sprinkles access chain
instructions into a module at random. This allows other passes to
have a richer set of pointers available to them, in particular the
passes that add loads and stores.
Adds a fuzzer pass that inserts function calls into the module at
random. Calls from dead blocks can be arbitrary (so long as they do
not introduce recursion), while calls from other blocks can only be to
livesafe functions.
The change fixes some oversights in transformations to replace
constants with uniforms and to obfuscate constants which testing of
this fuzzer pass identified.
This change ensures that global and local variables donated from other
modules are always initialized at their declaration in the module
being transformed. This is to help limit issues related to undefined
behaviour that might arise due to accessing uninitialized memory.
The change also introduces some helper functions in fuzzer_util to
make it easier to find the pointee types of pointer types.
This change adds fuzzer passes that sprinkle loads and stores into a
module at random, with stores restricted to occur in either dead
blocks, or to use pointers for which it is known that the pointee
value does not influence the module's overall behaviour.
The change also generalises the VariableValueIsArbitrary fact to
PointeeValueIsIrrelevant, to allow stores through access chains or
object copies of variables whose values are known to be irrelevant.
The change includes some other minor refactorings.
Adds two new fuzzer passes to add variables to a module: one that adds
Private storage class global variables, another that adds Function
storage class local variables.
If the fuzzer object-copies a pointer we would like to be able to
perform loads from the copy (and stores to it, if its value is known
not to matter). Undefined and null pointers present a problem here,
so this change disallows copying them.
This adds support for replacing TimeAMD with OpReadClockKHR. The scope
for OpReadClockKHR is fixed to be a subgroup because TimeAMD operates
only on subgroup.
* Implement constant folding for many transcendentals
This change adds support for folding of sin/cos/tan/asin/acos/atan,
exp/log/exp2/log2, sqrt, atan2 and pow.
The mechanism allows to use any C function to implement folding in the
future; for now I limited the actual additions to the most commonly used
intrinsics in the shaders.
Unary folder had to be tweaked to work with extended instructions - for
extended instructions, constants.size() == 2 and constants[0] ==
nullptr. This adjustment is similar to the one binary folder already
performs.
Fixes#1390.
* Fix Android build
On old versions of Android NDK, we don't get std::exp2/std::log2
because of partial C++11 support.
We do get ::exp2, but not ::log2 so we need to emulate that.
This change adds a new kind of fact to the fact manager, which records
when a variable (or pointer parameter) refers to an arbitrary value,
so that anything can be stored to it, without affecting the observable
behaviour of the module, and nothing can be guaranteed about values
loaded from it. Donated modules are the current source of such
variables, and other transformations, such as outlining, have been
adapted to propagate these facts appropriately.
This change allows the generator to (optionally and at random) make
the functions of a module "livesafe" during donation. This involves
introducing a loop limiter variable to each function and gating the
number of total loop iterations for the function using that variable.
It also involves eliminating OpKill and OpUnreachable instructions
(changing them to OpReturn/OpReturnValue), and clamping access chain
indices so that they are always in-bounds.
We must treat a branch to the merge node of a switch that is in the
header of a construct as a nested construced. The original merge
instruction is still needed in that case.
Fixes#3139
* If the header of the construct is also a merge block, jump to the
associated header instead of the immediate dominator
* prevents spurious failures from unrelated constructs
* new tests
This new API lets clients request a minimal spv_target_env value
that supports a given Vulkan and SPIR-V version, by a generic
numbering scheme (as already defined by Vulkan and SPIR-V specs).
This breaks a formal source dependency from Glslang to SPIRV-Tools.
When a new API is rolled out, such as Vulkan 1.2, Glslang currently
needs to reference a specific SPIRV-Tools enum by name.
* Allow OpExtInst for DebugInfo between secion 9 and 10
Fixes#3086
* Handle spirv-opt errors on DebugInfo Ext
* Add IR Loader test
* Fix ir loader bug
* Handle DebugFunction/DebugTypeMember forward reference
* Add test cases (forward reference to function)
* Support old DebugInfo extension
* Validate local debug info out of function
This change refactors some code for walking access chain indexes to
make it mirror the structure of other code (to improve readability in
the first instance and potentially enable a future refactoring to
extract common code), and fixes a problem related to module donation
and function types.
As explained in #3118, spirv-opt merge-blocks pass causes a
spirv-val error when an OpBranch has an OpLine in front of it.
OpLoopMerge
OpBranch ; Will be killed by merge-blocks pass
OpLabel ; Will be killed by merge-blocks pass
OpLine ; will be placed between OpLoopMerge and OpBranch - error!
OpBranch
To fix this issue, this commit moves line info of OpBranch to
OpLoopMerge.
Fixes#3118
This adds a new kind of fact to the fact manager that knows whether a
block is dead - i.e. guaranteed to be statically unreachable - and a
new transformation for adding a selection construct to a CFG that
conditionally branches to a fresh, dead block, such that the branch
will never be dynamically taken. Transformations that may create new
blocks ('split block' and 'outline function') are updated to propagate
dead block facts to newly-created blocks where appropriate. A fuzzer
pass randomly adds dead blocks to the module.
Future transformations will be able to exploit the fact that such
blocks are known to be dead.
This change adds a fuzzer pass that allows code from other SPIR-V
modules to be donated into the module under transformation. It also
changes the command-line options of the tools so that, in fuzzing
mode, a file must be specified that contains the names of available
donor modules.