platform/upstream/SPIRV-Tools.git
7 years agoCreate the dead function elimination pass
Steven Perron [Tue, 19 Sep 2017 14:12:13 +0000 (10:12 -0400)]
Create the dead function elimination pass

Creates a pass called eliminate dead functions that looks for functions
that could never be called, and deletes them from the module.

To support this change a new function was added to the Pass class to
traverse the call trees from diffent starting points.

Includes a test to ensure that annotations are removed when deleting a
dead function.  They were not, so fixed that up as well.

Did some cleanup of the assembly for the test in pass_test.cpp.  Trying
to make them smaller and easier to read.

7 years agoDetach MARK-V from the validator
Andrey Tuganov [Mon, 25 Sep 2017 18:55:44 +0000 (14:55 -0400)]
Detach MARK-V from the validator

MARK-V codec was previously dependent on the validation state.
Now it doesn't need the validator to function, but can still optionally
create it and validate every instruction once it's decoded.

7 years agoAvoid using global static variables
Lei Zhang [Thu, 21 Sep 2017 21:24:57 +0000 (17:24 -0400)]
Avoid using global static variables

Previously we have several grammar tables defined as global static
variables and these grammar table entries contains non-POD struct
fields (CapabilitySet/ExtensionSet). The initialization of these
non-POD struct fields may require calling operator new. If used
as a library and the caller defines its own operator new, things
can screw up.

This pull request changes all global static variables into
function static variables, which is lazy evaluated in a thread
safe way as guaranteed by C++11.

7 years agoAdd SPIRV_SPIRV_COMPRESSION option to cmake
Andrey Tuganov [Mon, 25 Sep 2017 16:45:48 +0000 (12:45 -0400)]
Add SPIRV_SPIRV_COMPRESSION option to cmake

The option is off by default.
cmake -DSPIRV_BUILD_COMPRESSION=ON ..
enables the compression lib, executable, and test build.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/834

7 years agoMerge pull request #829 from atgoo/fix_val_logicals
Ehsan [Thu, 21 Sep 2017 22:50:09 +0000 (18:50 -0400)]
Merge pull request #829 from atgoo/fix_val_logicals

Validator: fix logicals pass for OpSelect pointers

7 years agoValidator: fix logicals pass for OpSelect pointers
Andrey Tuganov [Thu, 21 Sep 2017 20:12:14 +0000 (16:12 -0400)]
Validator: fix logicals pass for OpSelect pointers

OpSelect works with pointers also when capability
VariablePointersStorageBuffer is declared (before worked only with
capability VariablePointers).

7 years agoelim-multi-store: only patch loop header phis that we created
David Neto [Wed, 20 Sep 2017 15:07:55 +0000 (11:07 -0400)]
elim-multi-store: only patch loop header phis that we created

There can already be OpPhi instructions in a loop header that
are unrelated to the optimization.  We should not be patching those.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/826

7 years agoAdd validate logicals pass to the validator
Andrey Tuganov [Thu, 7 Sep 2017 21:27:57 +0000 (17:27 -0400)]
Add validate logicals pass to the validator

New pass checks operands of all instructions listed under
3.32.15. Relational and Logical Instructions

7 years agoRefactored validate_aritmetics.cpp
Andrey Tuganov [Mon, 18 Sep 2017 21:07:00 +0000 (17:07 -0400)]
Refactored validate_aritmetics.cpp

Improved error messages and readability.

7 years agoUpdated markv_autogen
Andrey Tuganov [Wed, 6 Sep 2017 20:44:15 +0000 (16:44 -0400)]
Updated markv_autogen

- now includes a table of all descriptors with coding scheme
(improves performance by 5% by allowing to avoid creation of
move-to-front sequences which will never be used)
- increased the size of markv_autogen.inc, clang doesn't seem
to have the long compilation time problem now
(probably was inadvertently fixed by using Huffman codec
serialization)

7 years agoElimLocalMultiStore: Reset structured successors for each function
Greg Fischer [Tue, 19 Sep 2017 19:47:28 +0000 (13:47 -0600)]
ElimLocalMultiStore: Reset structured successors for each function

7 years agoAdd strength reduction; for now replace multiply by power of 2
Steven Perron [Fri, 8 Sep 2017 16:08:03 +0000 (12:08 -0400)]
Add strength reduction; for now replace multiply by power of 2

Create a new optimization pass, strength reduction, which will replace
integer multiplication by a constant power of 2 with an equivalent bit
shift.  More changes could be added later.

- Does not duplicate constants

- Adds vector |Concat| utility function to a common test header.

7 years agoExtractInsert: Handle rudimentary CompositeConstruct and ConstantComposite
GregF [Wed, 13 Sep 2017 00:15:09 +0000 (18:15 -0600)]
ExtractInsert: Handle rudimentary CompositeConstruct and ConstantComposite

This optimizes a single index extract whose composite value terminates with a
CompositeConstruct (or ConstantComposite) by evaluating to the correct
component. This was needed for opaque legalization.

This highlights the need/opportunity to improve this optimization to deal
with more complex composite expressions including currently handled ops
plus Null ops and special vector composition. A TODO has been added.

7 years agoRecognize SPV_AMD_shader_fragment_mask
David Neto [Thu, 14 Sep 2017 14:35:22 +0000 (10:35 -0400)]
Recognize SPV_AMD_shader_fragment_mask

7 years agoAdd new checks to validate arithmetics pass
Andrey Tuganov [Wed, 6 Sep 2017 18:30:27 +0000 (14:30 -0400)]
Add new checks to validate arithmetics pass

New operations:
- OpDot
- OpVectorTimesScalar
- OpMatrixTimesScalar
- OpVectorTimesMatrix
- OpMatrixTimesVector
- OpMatrixTimesMatrix
- OpOuterProduct

7 years agoUpdate CHANGES for OpModuleProcessed validation rule
David Neto [Fri, 8 Sep 2017 13:16:15 +0000 (09:16 -0400)]
Update CHANGES for OpModuleProcessed validation rule

7 years agovalidator: OpModuleProcessed allowed in layout section 7c
David Neto [Thu, 7 Sep 2017 21:38:31 +0000 (17:38 -0400)]
validator: OpModuleProcessed allowed in layout section 7c

Recent spec fix from SPIR Working group:
  Allow OpModuleProcessed after debug names, but before any
  annotation instructions.

7 years agoUpdate MARK-V to version 1.01
Andrey Tuganov [Wed, 9 Aug 2017 18:01:12 +0000 (14:01 -0400)]
Update MARK-V to version 1.01

Includes:
- Multi-sequence move-to-front
- Coding by id descriptor
- Statistical coding of non-id words
- Joint coding of opcode and num_operands

Removed explicit form Huffman codec constructor
- The standard use case for it is to be constructed from initializer list.

Using serialization for Huffman codecs

7 years agospirv-as: Fail for unrecognized long option
David Neto [Fri, 1 Sep 2017 22:12:15 +0000 (18:12 -0400)]
spirv-as: Fail for unrecognized long option

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/798

7 years agoInliner: Fix LoopMerge when inline into loop header of multi block loop
David Neto [Sat, 2 Sep 2017 23:01:03 +0000 (19:01 -0400)]
Inliner: Fix LoopMerge when inline into loop header of multi block loop

This adapts the fix for the single-block loop.  Split the loop like
before.  But when we move the OpLoopMerge back to the loop header,
redirect the continue target only when the original loop was a single
block loop.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/800

7 years agoAdd validation pass for arithmetic operations
Andrey Tuganov [Wed, 30 Aug 2017 14:13:10 +0000 (10:13 -0400)]
Add validation pass for arithmetic operations

The pass checks if arithmetic operations (such as OpFMul) receive
correct operands.

7 years agoMake enums for all currently published extensions
David Neto [Thu, 17 Aug 2017 17:53:14 +0000 (13:53 -0400)]
Make enums for all currently published extensions

Use the list from the SPIR-V registry page.  Also, capture it as
a string so it's much easier to update via copy-paste.

The validator will accept modules that declare these known
extensions.  However, we might not know about new tokens or
instructions declared in them.  For that we need grammar updates
applied to SPIRV-Headers.

7 years agoStarge v2017.1-dev
David Neto [Fri, 1 Sep 2017 20:28:22 +0000 (16:28 -0400)]
Starge v2017.1-dev

7 years agoCreate v2017.0
David Neto [Fri, 1 Sep 2017 20:24:27 +0000 (16:24 -0400)]
Create v2017.0

Update README to describe that we understand SPIR-V syntax based on the
grammar files included from the SPIRV-Headers repo.

(Also, it's high time we issue a v2017 release!

7 years agoFinalize v2016.7
David Neto [Fri, 1 Sep 2017 20:14:25 +0000 (16:14 -0400)]
Finalize v2016.7

7 years agoUpdate CHANGES for recent inliner fixes
David Neto [Fri, 1 Sep 2017 20:13:12 +0000 (16:13 -0400)]
Update CHANGES for recent inliner fixes

7 years agoFix mingw build (source/print.cpp)
Andrey Tuganov [Mon, 28 Aug 2017 22:36:52 +0000 (18:36 -0400)]
Fix mingw build (source/print.cpp)

source/print.cpp doesn't compile due to integer conversion.

Tested by @dneto0 on a Windows machine.

7 years agoInliner: Remap callee entry block id to single-trip loop header
David Neto [Thu, 31 Aug 2017 21:33:44 +0000 (17:33 -0400)]
Inliner: Remap callee entry block id to single-trip loop header

Otherwise cloned phis can be invalid.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/790

7 years agoInline: Fix single-block loop caller cases
David Neto [Thu, 31 Aug 2017 19:47:31 +0000 (15:47 -0400)]
Inline: Fix single-block loop caller cases

If the caller block is a single-block loop and inlining will
replace the caller block by several blocks, then:
- The original OpLoopMerge instruction will end up in the *last*
  such block.  That's the wrong place to put it.
- Move it back to the end of the first block.
- Update its Continue Target ID to point to the last block

We also have to take care of cases where the inlined code
begins with a structured header block.  In this case
we need to ensure the restored OpLoopMerge does not appear
in the same block as the merge instruction from the callee's
first block.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/787

7 years agoBasicBlock: add ctail, GetMergeInst, GetLoopMergeInst
David Neto [Thu, 31 Aug 2017 15:37:17 +0000 (11:37 -0400)]
BasicBlock: add ctail, GetMergeInst, GetLoopMergeInst

7 years agoUpdate tests for new preferred name as ShaderViewportIndexLayerEXT
David Neto [Fri, 1 Sep 2017 14:26:41 +0000 (10:26 -0400)]
Update tests for new preferred name as ShaderViewportIndexLayerEXT

This reacts to a recent update to SPIRV-Headers

7 years agoUpdate CHANGES for fix to 781
David Neto [Fri, 1 Sep 2017 14:15:42 +0000 (10:15 -0400)]
Update CHANGES for fix to 781

7 years agoExtension allows multiple same OpTypePointer types
Andrey Tuganov [Tue, 29 Aug 2017 20:25:23 +0000 (16:25 -0400)]
Extension allows multiple same OpTypePointer types

SPV_KHR_variable_pointers allows OpTypePointer to declare multiple
pointer identical types.

https://github.com/KhronosGroup/SPIRV-Tools/issues/781

7 years agoUpdate CHANGES for fix to 776
David Neto [Wed, 30 Aug 2017 18:25:23 +0000 (14:25 -0400)]
Update CHANGES for fix to 776

7 years agoDeadBranchElim: Fix dead block detection to ignore backedges
GregF [Wed, 23 Aug 2017 23:05:38 +0000 (17:05 -0600)]
DeadBranchElim: Fix dead block detection to ignore backedges

- DeadBranchElim: Make sure to mark orphan'd merge blocks and continue
targets as live.
- Add test with loop in dead branch
- Add test that orphan'd merge block is handled.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/776

7 years agoInlineOpaque: Remove from usage until complete Opaque policy is designed.
GregF [Tue, 29 Aug 2017 00:02:14 +0000 (18:02 -0600)]
InlineOpaque: Remove from usage until complete Opaque policy is designed.

7 years agoUpdate changes for fix to issue 782
David Neto [Wed, 30 Aug 2017 03:22:07 +0000 (23:22 -0400)]
Update changes for fix to issue 782

7 years agoInline: Fix remapping of non-label forward references in callee phi
GregF [Wed, 30 Aug 2017 00:35:05 +0000 (18:35 -0600)]
Inline: Fix remapping of non-label forward references in callee phi

7 years agoDisassembler: Print colour codes only when writing to a terminal
David Neto [Mon, 28 Aug 2017 14:30:58 +0000 (10:30 -0400)]
Disassembler: Print colour codes only when writing to a terminal

7 years agoFix encode zero bits on word boundary bug
Andrey Tuganov [Mon, 28 Aug 2017 17:18:45 +0000 (13:18 -0400)]
Fix encode zero bits on word boundary bug

Bit stream writer was manifesting incorrect behaviour when the following
two conditions were met:
- writer was on 64-bit word boundary
- WriteBits was invoked with num_bits=0 (can happen when a Huffman codec has only one
value)

The bug was causing very rare sporadic corruption which was detected by
tests after a random experimental change in MARK-V model.

7 years agoShow result id for CompositeInsert validation failure
David Neto [Thu, 24 Aug 2017 22:11:23 +0000 (18:11 -0400)]
Show result id for CompositeInsert validation failure

7 years agoREADME: Add reference to SPIR-V size reduction white paper
GregF [Thu, 24 Aug 2017 22:36:53 +0000 (16:36 -0600)]
README: Add reference to SPIR-V size reduction white paper

7 years agoWindows: Increase intensity of blue text
David Neto [Thu, 24 Aug 2017 14:34:00 +0000 (10:34 -0400)]
Windows: Increase intensity of blue text

7 years agoFix: background color was erroneously reset on Win32 platform.
Lukas Hermanns [Sun, 20 Aug 2017 12:11:50 +0000 (14:11 +0200)]
Fix: background color was erroneously reset on Win32 platform.

Fix: background color was erroneously reset on Win32 platform.

7 years agoUpdated capabilites for SampleMask
David Neto [Wed, 23 Aug 2017 17:28:48 +0000 (13:28 -0400)]
Updated capabilites for SampleMask

SPIRV-Headers recently fixed the capability dependency
for SampleMask.  It depends on Shader, not SampleRateShading

7 years agoOpt: Create InlineOpaquePass
GregF [Tue, 15 Aug 2017 23:58:28 +0000 (17:58 -0600)]
Opt: Create InlineOpaquePass

Only inline calls to functions with opaque params or return

TODO: Handle parameter type or return type where the opqaue
type is buried within an array.

7 years agoOpt: Have "size" passes process full entry point call tree.
GregF [Thu, 10 Aug 2017 22:42:16 +0000 (16:42 -0600)]
Opt: Have "size" passes process full entry point call tree.

Includes code to deal correctly with OpFunctionParameter. This
is needed by opaque propagation which may not exhaustively inline
entry point functions.

Adds ProcessEntryPointCallTree: a method to do work on the
functions in the entry point call trees in a deterministic order.

7 years agoSupport supplying extra definitions via CMAKE variable
Lei Zhang [Wed, 16 Aug 2017 20:42:50 +0000 (16:42 -0400)]
Support supplying extra definitions via CMAKE variable

7 years agoHuffman codec can serialize to text
Andrey Tuganov [Fri, 11 Aug 2017 20:51:24 +0000 (16:51 -0400)]
Huffman codec can serialize to text

Refactored the Huffman codec implementation and added ability to
serialize to C++-like text format. This would reduce the time-complexity
if loading hard-coded codecs.

7 years agoOpt: Add opaque tests
GregF [Tue, 15 Aug 2017 21:54:41 +0000 (15:54 -0600)]
Opt: Add opaque tests

7 years agoAdd id descriptor feature to SPIR-V
Andrey Tuganov [Wed, 2 Aug 2017 20:47:25 +0000 (16:47 -0400)]
Add id descriptor feature to SPIR-V

Id descriptors are computed as a recursive hash of all instructions used
to define an id. Descriptors are invarint of actual id values and
the similar code in different files would produce the same descriptors.

Multiple ids can have the same descriptor. For example
%1 = OpConstant %u32 1
%2 = OpConstant %u32 1
would produce two ids with the same descriptor. But
%3 = OpConstant %s32 1
%4 = OpConstant %u32 2
would have descriptors different from %1 and %2.

Descriptors will be used as handles of move-to-front sequences in SPIR-V
compression.

7 years agoADCE: Add support for function calls
GregF [Fri, 4 Aug 2017 21:04:37 +0000 (15:04 -0600)]
ADCE: Add support for function calls

ADCE will now generate correct code in the presence of function calls.
This is needed for opaque type optimization needed by glslang. Currently
all function calls are marked as live. TODO: mark calls live only if they
write a non-local.

7 years agoUpdate CHANGE to list fix for issue 755
David Neto [Thu, 10 Aug 2017 17:13:33 +0000 (13:13 -0400)]
Update CHANGE to list fix for issue 755

7 years agoInliner: callee can have early return that isn't multi-return
David Neto [Wed, 9 Aug 2017 18:59:04 +0000 (14:59 -0400)]
Inliner: callee can have early return that isn't multi-return

Avoid generating an invalid OpLabel.
Create the continue target for the single-trip loop only if
you actually created the header for the single-trip loop.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/755

7 years agoAccessChainConvert: Add HasOnlySupportedRefs()
GregF [Fri, 4 Aug 2017 18:00:30 +0000 (12:00 -0600)]
AccessChainConvert: Add HasOnlySupportedRefs()

This avoids conversion on variables which will not ultimately be optimized.
Also removed an obsolete restriction from FindTargetVars(). Also added
decorates to supported refs (eg. RelaxedPrecision). Also fixed name to
IsNonTypeDecorate().

7 years agoInline: Split out InlineExhaustivePass from InlinePass
GregF [Tue, 1 Aug 2017 23:20:13 +0000 (17:20 -0600)]
Inline: Split out InlineExhaustivePass from InlinePass

7 years agoMem2Reg: Allow Image and Sampler types as base target types.
GregF [Thu, 3 Aug 2017 23:27:02 +0000 (17:27 -0600)]
Mem2Reg: Allow Image and Sampler types as base target types.

7 years agoUpdate CHANGES with common uniform loads pass
David Neto [Fri, 4 Aug 2017 21:48:22 +0000 (17:48 -0400)]
Update CHANGES with common uniform loads pass

7 years agoUpdate README.md with new code-reduction passes
GregF [Thu, 3 Aug 2017 16:37:26 +0000 (10:37 -0600)]
Update README.md with new code-reduction passes

7 years agoAdd CommonUniformElim pass
GregF [Mon, 3 Jul 2017 23:23:04 +0000 (17:23 -0600)]
Add CommonUniformElim pass

- UniformElim: Only process reachable blocks

- UniformElim: Don't reuse loads of samplers and images across blocks.
  Added a second phase which only reuses loads within a block for samplers
  and images.

- UniformElim: Upgrade CopyObject skipping in GetPtr

- UniformElim: Add extensions whitelist
  Currently disallowing SPV_KHR_variable_pointers because it doesn't
  handle extended pointer forms.

- UniformElim: Do not process shaders with GroupDecorate

- UniformElim: Bail on shaders with non-32-bit ints.

- UniformElim: Document support for only single index and add TODO.

7 years agoAdd MemPass, move all shared functions to it.
GregF [Thu, 27 Jul 2017 19:30:12 +0000 (13:30 -0600)]
Add MemPass, move all shared functions to it.

7 years agoAdd multi-sequence move-to-front implementation
Andrey Tuganov [Mon, 24 Jul 2017 18:29:35 +0000 (14:29 -0400)]
Add multi-sequence move-to-front implementation

Add MultiMoveToFront class which supports multiple move-to-front
sequences and allows to promote value in all sequences at once.

Added caching for last accessed sequence handle and last accessed value
in each sequence.

7 years agoUpdate CHANGES with recently added optimizations
David Neto [Tue, 1 Aug 2017 19:33:15 +0000 (15:33 -0400)]
Update CHANGES with recently added optimizations

7 years agoOpt: Add new size-reduction passes to usage message.
GregF [Tue, 1 Aug 2017 18:55:29 +0000 (12:55 -0600)]
Opt: Add new size-reduction passes to usage message.

7 years agoAdded C++ code generation to spirv-stats
Andrey Tuganov [Mon, 31 Jul 2017 17:08:38 +0000 (13:08 -0400)]
Added C++ code generation to spirv-stats

The tool can now generate C++ code returning some of the historgrams and
Huffman codecs generated from those historgrams.

7 years agoAdd Appveyor config that uses VS 2017
David Neto [Tue, 1 Aug 2017 15:42:28 +0000 (11:42 -0400)]
Add Appveyor config that uses VS 2017

7 years agoOpt: Delete names and decorations of dead instructions
GregF [Wed, 26 Jul 2017 22:34:41 +0000 (18:34 -0400)]
Opt: Delete names and decorations of dead instructions

7 years agoOpt: HasOnlySupportedRefs should consider OpCopyObject
Lei Zhang [Mon, 24 Jul 2017 15:43:47 +0000 (11:43 -0400)]
Opt: HasOnlySupportedRefs should consider OpCopyObject

This fixes test failure after merging the previous pull request.

7 years agoRevert "Revert "Opt: LocalBlockElim: Add HasOnlySupportedRefs""
Lei Zhang [Mon, 24 Jul 2017 14:37:38 +0000 (10:37 -0400)]
Revert "Revert "Opt: LocalBlockElim: Add HasOnlySupportedRefs""

This reverts commit df96e243c633b939690a42d89904e04f50126e86.

7 years agoAdd extension whitelists to size-reduction passes.
GregF [Wed, 19 Jul 2017 00:57:26 +0000 (18:57 -0600)]
Add extension whitelists to size-reduction passes.

Currently only SPV_KHR_variable_pointers is disallowed in passes which
do pointer analysis. Positive and negative tests of the general extensions
mechanism were added to aggressive_dce but cover all passes.

7 years agoRevert "Opt: LocalBlockElim: Add HasOnlySupportedRefs"
Lei Zhang [Sat, 22 Jul 2017 14:48:28 +0000 (10:48 -0400)]
Revert "Opt: LocalBlockElim: Add HasOnlySupportedRefs"

This reverts commit 2d0f7fbc1151004c4cf111b947f1070a03dfe092.

7 years agoOpt: LocalBlockElim: Add HasOnlySupportedRefs
greg-lunarg [Sat, 22 Jul 2017 14:32:19 +0000 (08:32 -0600)]
Opt: LocalBlockElim: Add HasOnlySupportedRefs

Verifies that targeted variables have only access chain and direct
loads and stores as references.

7 years agoFix handling of CopyObject in GetPtr and its call sites
GregF [Tue, 18 Jul 2017 20:42:51 +0000 (14:42 -0600)]
Fix handling of CopyObject in GetPtr and its call sites

7 years agoFix Visual Studio size_t cast compiler warning
Lenny Komow [Thu, 13 Jul 2017 17:13:14 +0000 (11:13 -0600)]
Fix Visual Studio size_t cast compiler warning

Visual Studio was complaining about possible loss of data on 64-bit
builds, due to an implicit cast from size_t to int. This changes the
data to use an int with no cast.

7 years agoLocalMultiStore: Always put varId for backedge on loop phi function.
Greg Fischer [Tue, 11 Jul 2017 22:52:19 +0000 (16:52 -0600)]
LocalMultiStore: Always put varId for backedge on loop phi function.

And always patch the backedge operand when patching phi functions. This
approach is more correct and cleaner. The previous code was generating
incorrect phis when the backedge block had no predecessors.

7 years agoDeadBranchElim: Improve algorithm to only remove blocks with no predecessors
GregF [Mon, 10 Jul 2017 23:20:35 +0000 (17:20 -0600)]
DeadBranchElim: Improve algorithm to only remove blocks with no predecessors

Must be careful not to remove blocks pointed at by unreachable blocks

7 years agoMinor code review feedback on AggressiveDCE
David Neto [Mon, 10 Jul 2017 15:45:59 +0000 (11:45 -0400)]
Minor code review feedback on AggressiveDCE

7 years agoAdd AggressiveDCEPass
GregF [Thu, 8 Jun 2017 16:37:21 +0000 (10:37 -0600)]
Add AggressiveDCEPass

Create aggressive dead code elimination pass
This pass eliminates unused code from functions. In addition,
it detects and eliminates code which may have spurious uses but which do
not contribute to the output of the function. The most common cause of
such code sequences is summations in loops whose result is no longer used
due to dead code elimination. This optimization has additional compile
time cost over standard dead code elimination.

This pass only processes entry point functions. It also only processes
shaders with logical addressing. It currently will not process functions
with function calls. It currently only supports the GLSL.std.450 extended
instruction set. It currently does not support any extensions.

This pass will be made more effective by first running passes that remove
dead control flow and inlines function calls.

This pass can be especially useful after running Local Access Chain
Conversion, which tends to cause cycles of dead code to be left after
Store/Load elimination passes are completed. These cycles cannot be
eliminated with standard dead code elimination.

Additionally: This transform uses a whitelist of instructions that it
knows do have side effects, (a.k.a. combinators).  It assumes other
instructions have side effects: it will not remove them, and assumes
they have side effects via their ID operands.

7 years agoAdd LocalMultiStoreElim pass
GregF [Fri, 16 Jun 2017 21:37:31 +0000 (15:37 -0600)]
Add LocalMultiStoreElim pass

A SSA local variable load/store elimination pass.
For every entry point function, eliminate all loads and stores of function
scope variables only referenced with non-access-chain loads and stores.
Eliminate the variables as well.

The presence of access chain references and function calls can inhibit
the above optimization.

Only shader modules with logical addressing are currently processed.
Currently modules with any extensions enabled are not processed. This
is left for future work.

This pass is most effective if preceeded by Inlining and
LocalAccessChainConvert. LocalSingleStoreElim and LocalSingleBlockElim
will reduce the work that this pass has to do.

7 years agoDeadBranchElim: Add DeadBranchElimPass
GregF [Fri, 2 Jun 2017 19:23:20 +0000 (13:23 -0600)]
DeadBranchElim: Add DeadBranchElimPass

7 years agoInclude memory and semantics IDs when iterating over inbound IDs
David Neto [Tue, 4 Jul 2017 20:24:46 +0000 (16:24 -0400)]
Include memory and semantics IDs when iterating over inbound IDs

Fixes Instruction::ForEachInId so it covers
SPV_OPERAND_TYPE_MEMORY_SEMANTICS_ID and SPV_OPERAND_TYPE_SCOPE_ID.
Future proof a bit by using the common spvIsIdType routine.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/697

7 years agoMARK-V decoder supports extended instructions
Andrey Tuganov [Fri, 30 Jun 2017 18:59:24 +0000 (14:59 -0400)]
MARK-V decoder supports extended instructions

7 years agoUpdate CHANGES to say we use GNUIntallDirs
David Neto [Tue, 4 Jul 2017 16:27:31 +0000 (12:27 -0400)]
Update CHANGES to say we use GNUIntallDirs

7 years agoDefine variable to skip installation
d3x0r [Tue, 4 Jul 2017 13:11:40 +0000 (06:11 -0700)]
Define variable to skip installation

If this is used as a static library in another project, this does not
need to be installed, and otherwise will just clutter the application's install.
To use, define SKIP_SPIRV_TOOLS_INSTALL which internally defines
ENABLE_SPIRV_TOOLS_INSTALL to control installation.

Also include GNUInstallDirs to get standard output 'lib' directory which is sometimes 'lib64' and not 'lib'

7 years agoConvert pattern stack from deque to vector, and share it
Chris Forbes [Tue, 27 Jun 2017 23:28:22 +0000 (16:28 -0700)]
Convert pattern stack from deque to vector, and share it

Also move various vector::reserve calls to State ctor
Negligible perf benefit, but more tidy.

7 years agoAdded fixed width encoding to bit_stream
Andrey Tuganov [Thu, 29 Jun 2017 19:05:34 +0000 (15:05 -0400)]
Added fixed width encoding to bit_stream

Fixed width encoding is intended to be used for small unsigned integers
when the upper bound is known both to the encoder and the decoder
(for example move-to-front rank).

7 years agoAdded compression tool tools/spirv-markv. Work in progress.
Andrey Tuganov [Wed, 31 May 2017 17:07:51 +0000 (13:07 -0400)]
Added compression tool tools/spirv-markv. Work in progress.

Command line application is located at tools/spirv-markv
API at include/spirv-tools/markv.h

At the moment only very basic compression is implemented, mostly varint.
Scope of supported SPIR-V opcodes is also limited.

Using a simple move-to-front implementation instead of encoding mapped
ids.

Work in progress:
- Does not cover all of SPIR-V
- Does not promise compatibility of compression/decompression across
  different versions of the code.

7 years agoAdded log(n) move-to-front implementation
Andrey Tuganov [Mon, 12 Jun 2017 16:01:39 +0000 (12:01 -0400)]
Added log(n) move-to-front implementation

The implementation is based on AVL and order statistic tree.

It accepts all kinds of values and the implementation
doesn't expect the behaviour to be consistent with id coding.

Intended by SPIR-V compression algorithms.

7 years agoAdded Huffman codec to utils
Andrey Tuganov [Thu, 25 May 2017 15:21:12 +0000 (11:21 -0400)]
Added Huffman codec to utils

Attached ids to Huffman nodes for deterministic internal node
comparison.

7 years agoTravis CI: stop requiring sudo and use make instead of ninja
Lei Zhang [Tue, 27 Jun 2017 15:20:34 +0000 (11:20 -0400)]
Travis CI: stop requiring sudo and use make instead of ninja

This gives us shorter testing time since we can use the container
based testing environment, which starts faster. Also using make
means we don't need to install ninja separately.

7 years agoDon't do hash lookup twice in FindDef
Chris Forbes [Tue, 27 Jun 2017 21:21:33 +0000 (14:21 -0700)]
Don't do hash lookup twice in FindDef

7 years agoMove spv_instruction_t's into vector
Chris Forbes [Tue, 27 Jun 2017 21:21:00 +0000 (14:21 -0700)]
Move spv_instruction_t's into vector

No need to incur another copy here. These guys have embedded vectors
we'd rather not copy.

7 years agoReserve expected length of instructions vector
Chris Forbes [Tue, 27 Jun 2017 18:55:14 +0000 (11:55 -0700)]
Reserve expected length of instructions vector

7 years agoMove some temp vectors into parser state
Chris Forbes [Tue, 27 Jun 2017 18:00:06 +0000 (11:00 -0700)]
Move some temp vectors into parser state

We don't need to churn the allocations for these every instruction.

7 years agoBlockMerge: Add BlockMergePass
GregF [Wed, 7 Jun 2017 21:28:53 +0000 (15:28 -0600)]
BlockMerge: Add BlockMergePass

Also, add BasicBlock::tail()

7 years agoUpdate CHANGES
David Neto [Thu, 22 Jun 2017 14:22:55 +0000 (10:22 -0400)]
Update CHANGES

- Describe support for AMD extensions
- Describe recently added optimizer passes

7 years agoAdd support for SPV AMD extensions
Rex Xu [Wed, 21 Jun 2017 07:58:00 +0000 (15:58 +0800)]
Add support for SPV AMD extensions

7 years agomem2reg: Add InsertExtractElimPass
GregF [Fri, 26 May 2017 16:33:11 +0000 (10:33 -0600)]
mem2reg: Add InsertExtractElimPass

7 years agoTransform multiple entry points
David Neto [Tue, 20 Jun 2017 19:57:47 +0000 (15:57 -0400)]
Transform multiple entry points

Don't stop just after one because of short-circuiting logical-or.

7 years agomem2reg: Add LocalSingleStoreElimPass
GregF [Fri, 19 May 2017 23:31:28 +0000 (17:31 -0600)]
mem2reg: Add LocalSingleStoreElimPass

Eliminate function scope variables with one store, if possible.