Adding Fma instruction can speed up the code. This was requested by
swiftshader, so they do not have to do this analysis themselves. It can
also help reduce the code size, and the work the ICD compilers have to
do.
spread-volatile-semantics pass spreads Volatile semantics for builtin
variables used by certain execution models based on
VUID-StandaloneSpirv-VulkanMemoryModel-04678 and
VUID-StandaloneSpirv-VulkanMemoryModel-04679 (See "Standalone SPIR-V
Validation" section of Vulkan spec "Appendix A: Vulkan Environment for
SPIR-V"). Therefore, shaders without execution model (e.g., used only
for linkage) are not the target of the pass. This commit lets the pass
just return SuccessWithoutChange in that case.
Clang added -Wunqualified-std-cast-call in
https://reviews.llvm.org/D119670, which warns on unqualified std::move
and std::forward calls. This change qualifies these calls to allow the
project to build on HEAD Clang -Werror.
If the body of the module does not have any ids change, compact ids will
not change the id bound. This can cause problems because the id bound
could be much higher than the largest id in that is used. It should be
reset any time it is not the larger id used + 1.
Fixes#4604
When folding a vector shuffle feeding a vector shuffle, we do not
propagate an 0xFFFFFFFF, which has a special meaning, correctly. We
adjust the value making it lose it meaning as an undefined value.
Fixes#4581
CCP does not want to fold an instruction unless it folds to a constant.
There is an asser to check for this. The question if a spec constant
counts as a constant. The constant folder considers a spec constant a
constand, but CCP does not. I've fixed the assert in CCP to match what
the folder does. It should not require any new changes to CCP.
Swift shader needs a way to inline all functions, even those marked as
DontInline. See https://github.com/KhronosGroup/SPIRV-Tools/pull/4471.
This implements the suggestion I made in the PR. We add a pass that
will remove the DontInline function control, so that the inlining passes
will inline them.
SwiftShader will still have to modify their code to add this pass before
the other passes are run.
The function `BuildInvalideAnalyses` will be rebuilt for every analysis that
has been requested, but it is not necessary. It also can cause problems
because if the CFG needs to be rebuilt, so do the dominator trees.
This change will make the functionality match the description of the
function.
* Optimize DefUseManager allocations
Saves around 30-35% of compilation time.
For inst->use_ids, use a pool linked list instead of allocating vectors for every instruction. For inst->uses, use a "PooledLinkedList"' -- a linked list that has shared storage for all nodes. Neither re-use nodes, instead we do a bulk compaction operation when too much memory is being wasted (tuneable).
Includes separate PooledLinkedList templated datastructure, a very special case construct, but split out to make the code a little easier to understand.
Incrementally compute the hash instead of collecting words
Avoids allocating temporary space in a std::vector and std::u32string, and making three passes over all the hashed data.
Switch to using std::vector to prevent processing duplicates instead of std::unordered_set: avoids an allocation/deletion every call to ComputeHashValue, and ends up faster due to much better cache behaviour and smaller constant-factor when searching the (generally very small) list.
In my test case, made Type::HashValue go from 7.5% of compilation time to .5%
Scalar replacement generates a null when there value for a member will
not be used. The null is used to make sure things are
deterministic in case there is an error.
However, some type cannot be null, so we will change that to use undef.
To keep the code simpler we will always use the undef.
Fixes#3996
spirv-diff is a new tool that produces diff-style output comparing two
SPIR-V modules. The instructions between the src and dst modules are
matched as best as the tool can, and output is produced (in src
id-space) that shows which instructions are removed in src, added in dst
or modified between them. The order of instructions are not retained.
Matching instructions between two SPIR-V modules is not trivial, and
thus a number of heuristics are applied in this tool. In particular,
without debug information, it's hard to match functions as they can be
reordered. As such, this tool is primarily useful to produce the diff
of two SPIR-V modules derived from the same source.
This tool can be useful in a number of scenarios:
- Compare the SPIR-V before and after modifying a shader
- Compare the SPIR-V produced from a shader before and after compiler
codegen changes.
- Compare the SPIR-V produced from a shader before and after some
transformation or optimization.
- Compare the SPIR-V produced from a shader with different compilers.
The handling of the RayQueryKHR type is not complete in the type
manager. The tests were not picking this up. I've added a test to make
sure that the `GenerateAllTypes` function actually does generate all of
the types. Once it is added there other tests should pick up on the
other parts that were missing.
Add a pass to spread Volatile semantics to variables with SMIDNV,
WarpIDNV, SubgroupSize, SubgroupLocalInvocationId, SubgroupEqMask,
SubgroupGeMask, SubgroupGtMask, SubgroupLeMask, or SubgroupLtMask BuiltIn
decorations or OpLoad for them when the shader model is the ray
generation, closest hit, miss, intersection, or callable shaders. This
pass can be used for VUID-StandaloneSpirv-VulkanMemoryModel-04678 and
VUID-StandaloneSpirv-VulkanMemoryModel-04679 (See "Standalone SPIR-V
Validation" section of Vulkan spec "Appendix A: Vulkan Environment for
SPIR-V").
Handle variables used by multiple entry points:
1. Update error check to make it working regardless of the order of
entry points.
2. For a variable, if it is used by two entry points E1 and E2 and
it needs the Volatile semantics for E1 while it does not for E2
- If VulkanMemoryModel capability is enabled, which means we have to
set memory operation of load instructions for the variable, we
update load instructions in E1, but do not update the ones in E2.
- If VulkanMemoryModel capability is disabled, which means we have
to add Volatile decoration for the variable, we report an error
because E1 needs to add Volatile decoration for the variable while
E2 does not.
For the simplicity of the implementation, we assume that all functions
other than entry point functions are inlined.
* PassManager: Print errors occurring during disassembly
Otherwise one could be greeted by the following text when running
spirv-opt withe the `--print-all` flag:
; IR before pass wrap-opkill
; IR before pass eliminate-dead-branches
; IR before pass merge-return
With this commit, one will instead get:
error: line 143: Invalid opcode: 400
warning: line 0: Disassembly failed before pass wrap-opkill
error: line 143: Invalid opcode: 400
warning: line 0: Disassembly failed before pass eliminate-dead-branches
error: line 143: Invalid opcode: 400
warning: line 0: Disassembly failed before pass merge-return
* PassManager: Use the right target environment when disassembling
Disassembly would fail if features from a newer version of SPIR-V than
1.2 were used.
The comment in `Array::GetExtraHashWords` is misleading because getting
the hash words is split up between the generic `Type::GetHashWords` and
the type specific `Type::GetExtraHashWords`. While `IsSameImpl` is
self-contained. Removing the comment since it is misleading and no
comment is really needed.
Fixes#3248
The pass to remove the nonsemantic information and instructions
is used for drivers or tools that may not support them. Debug
information was only partially handle, which is causing a
problem. We need to either fully remove debug information or
not remove it all. Since I can see it being useful to keep the
debug information even when the nonsemantic instructions are
removed, I propose we do not remove debug info.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/4269
In https://github.com/KhronosGroup/SPIRV-Tools/pull/3110, the strip reflect
pass was changed to also remove all explicitly nonsemantic instructions. This
makes it so that the name of the pass no longer reflects what the pass actually
does. This change renames the pass so that it reflects what the pass actaully does.
The change in
commit 4ac8e5e541
Author: Greg Fischer <greg@lunarg.com>
Date: Wed Sep 15 12:38:34 2021 -0600
Add preserve_interface mode to aggressive_dead_code_elim (#4520)
Broke the C++ ABI for spirv-tools shared libraries on Linux, for not a great reason.
Restore the previous ABI.
* Fix endianness of string literals
To get correct and consistent encoding and decoding of string literals
on big-endian platforms, use spvtools::utils::MakeString and MakeVector
(or wrapper functions) consistently for handling string literals.
- add variant of MakeVector that encodes a string literal into an
existing vector of words
- add variants of MakeString
- add a wrapper spvDecodeLiteralStringOperand in source/
- fix wrapper Operand::AsString to use MakeString (source/opt)
- remove Operand::AsCString as broken and unused
- add a variant of GetOperandAs for string literals (source/val)
... and apply those wrappers throughout the code.
Fixes #149
* Extend round trip test for StringLiterals to flip word order
In the encoding/decoding roundtrip tests for string literals, include
a case that flips byte order in words after encoding and then checks for
successful decoding. That is, on a little-endian host flip to big-endian
byte order and then decode, and vice versa.
* BinaryParseTest.InstructionWithStringOperand: also flip byte order
Test binary parsing of string operands both with the host's and with the
reversed byte order.
Currently if an ID overflow occurs, spirv-opt (and other users of
IRContext) emits a warning and starts returning 0 when fresh ids are
requested. This tends to lead to crashes - such as null pointer
exceptions. When these arise during fuzzing they lead to auto-reported
bugs.
This change uses an ifdef guard to instead gracefully exit as soon as an
ID overflow occurs when the build is a fuzzing build.
Related issue: #4539.
This prevents CCP from making constant -> constant transitions when
evaluating instruction values. In this case, FClamp is evaluated twice.
On the first evaluation, if computes FClamp(0.5, 0.5, -1) which returns
-1. On the second evaluation, it computes FClamp(0.5, 0.5, VARYING)
which returns 0.5.
Both fold() computations are correct given the semantics of FClamp() but
this causes a lateral transition in the constant lattice which was not
being considered VARYING by CCP.
Along with OpDecorate, also clone the OpDecorateString instructions for
variables created in the descriptor scalar replacement pass.
Fixesmicrosoft/DirectXShaderCompiler#3705
* https://github.com/KhronosGroup/Vulkan-Docs/issues/666 clearly
specified that interfaces do not require an input if there is an
associated output
* ADCE can now remove unused input variables (though they are kept if
the preserve interfaces option is used)
If the ids overflow when creating an integer constant in the ir_builder, there will be a nullptr dereference. This is happening from inside merge return.
We need to propagate the error up, and make sure it is handled appropriately.
Consider the new test case. The conditional branch in the continue
block is never marked as live. However, `IsDead` will say it is not
dead, so it does not get deleted. Because it was never marked as live,
`%false` was not mark as live either, but it gets deleted. This results
in invalid code.
To fix this properly, we had to reconsider how branches are handle. We
make the following changes:
1) Terminator instructions that are not branch or OpUnreachable must be
kept, so they are marked as live when initializing the worklist.
2) Branches and OpUnreachable instructions are marked as live if
a) the block does not have a merge instruction and another instruction
in the block is marked as live, or
b) the merge instruction in the same block is marked as live.
3) Any instruction that is not marked as live is removed.
4) If a terminator is to be removed, an OpUnreachable is added. This
happens when the entire block is dead, and the block will be removed.
The OpUnreachable is generated to make sure the block still has a
terminator, and is valid.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/4509.
Spirv-opt has not had to handle module with function declarations. This
lead many passes to assume that every function has a body. This is not
always true. This commit will modify a number of passes to handle
function declarations.
Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/4443
