AuroraMiddleware/SPIRV-Cross
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
b8905bbd95
SPIRV-Cross/spirv_cross_parsed_ir.cpp

911 lines
21 KiB
C++
Raw Normal View History

Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
/*
Update license headers to 2020.
2020-01-16 14:24:37 +00:00
* Copyright 2018-2020 Arm Limited
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "spirv_cross_parsed_ir.hpp"
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
#include <algorithm>
Run format_all.sh.
2019-01-11 11:56:00 +00:00
#include <assert.h>
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
using namespace std;
using namespace spv;
Add an option to override the namespace used for spirv_cross. This is a pragmatic trick to avoid symbol collision where a project links against SPIRV-Cross statically, while linking to other projects which also use SPIRV-Cross statically. We can end up with very awkward symbol collisions which can resolve themselves silently because SPIRV-Cross is pulled in as necessary. To fix this, we must use different symbols and embed two copies of SPIRV-Cross in this scenario, now with different namespaces, which in turn leads to different symbols.
2019-03-29 09:29:44 +00:00
namespace SPIRV_CROSS_NAMESPACE
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Reduce pressure on global allocation. - Replace ostringstream with custom implementation. ~30% performance uplift on vector-shuffle-oom test. Allocations are measurably reduced in Valgrind. - Replace std::vector with SmallVector. Classic malloc optimization, small vectors are backed by inline data. ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux. - Use an object pool for IVariant type. We generally allocate a lot of SPIR* objects. We can amortize these allocations neatly by pooling them. - ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-02 09:19:03 +00:00
ParsedIR::ParsedIR()
{
// If we move ParsedIR, we need to make sure the pointer stays fixed since the child Variant objects consume a pointer to this group,
// so need an extra pointer here.
pool_group.reset(new ObjectPoolGroup);
pool_group->pools[TypeType].reset(new ObjectPool<SPIRType>);
pool_group->pools[TypeVariable].reset(new ObjectPool<SPIRVariable>);
pool_group->pools[TypeConstant].reset(new ObjectPool<SPIRConstant>);
pool_group->pools[TypeFunction].reset(new ObjectPool<SPIRFunction>);
pool_group->pools[TypeFunctionPrototype].reset(new ObjectPool<SPIRFunctionPrototype>);
pool_group->pools[TypeBlock].reset(new ObjectPool<SPIRBlock>);
pool_group->pools[TypeExtension].reset(new ObjectPool<SPIRExtension>);
pool_group->pools[TypeExpression].reset(new ObjectPool<SPIRExpression>);
pool_group->pools[TypeConstantOp].reset(new ObjectPool<SPIRConstantOp>);
pool_group->pools[TypeCombinedImageSampler].reset(new ObjectPool<SPIRCombinedImageSampler>);
pool_group->pools[TypeAccessChain].reset(new ObjectPool<SPIRAccessChain>);
pool_group->pools[TypeUndef].reset(new ObjectPool<SPIRUndef>);
Support emitting OpLine directive. Facilitates easier mapping from source language to cross-compiled output in tooling.
2019-05-28 11:41:46 +00:00
pool_group->pools[TypeString].reset(new ObjectPool<SPIRString>);
Reduce pressure on global allocation. - Replace ostringstream with custom implementation. ~30% performance uplift on vector-shuffle-oom test. Allocations are measurably reduced in Valgrind. - Replace std::vector with SmallVector. Classic malloc optimization, small vectors are backed by inline data. ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux. - Use an object pool for IVariant type. We generally allocate a lot of SPIR* objects. We can amortize these allocations neatly by pooling them. - ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-02 09:19:03 +00:00
}
Explicitly implement move operators for ParsedIR. Workaround for MSVC 2013.
2019-04-09 08:25:41 +00:00
// Should have been default-implemented, but need this on MSVC 2013.
ParsedIR::ParsedIR(ParsedIR &&other) SPIRV_CROSS_NOEXCEPT
{
*this = move(other);
}
ParsedIR &ParsedIR::operator=(ParsedIR &&other) SPIRV_CROSS_NOEXCEPT
{
if (this != &other)
{
Fix various bugs found in testing.
2019-04-09 08:54:28 +00:00
pool_group = move(other.pool_group);
Explicitly implement move operators for ParsedIR. Workaround for MSVC 2013.
2019-04-09 08:25:41 +00:00
spirv = move(other.spirv);
meta = move(other.meta);
for (int i = 0; i < TypeCount; i++)
Make sure ids_for_types are moved correctly in move operator.
2019-04-09 13:16:37 +00:00
ids_for_type[i] = move(other.ids_for_type[i]);
Explicitly implement move operators for ParsedIR. Workaround for MSVC 2013.
2019-04-09 08:25:41 +00:00
ids_for_constant_or_type = move(other.ids_for_constant_or_type);
ids_for_constant_or_variable = move(other.ids_for_constant_or_variable);
declared_capabilities = move(other.declared_capabilities);
declared_extensions = move(other.declared_extensions);
block_meta = move(other.block_meta);
continue_block_to_loop_header = move(other.continue_block_to_loop_header);
entry_points = move(other.entry_points);
ids = move(other.ids);
GLSL: Implement GL_EXT_buffer_reference. Buffer objects can contain arbitrary pointers to blocks. We can also implement ConvertPtrToU and ConvertUToPtr. The latter can cast a uint64_t to any type as it pleases, so we will need to generate fake buffer reference blocks to be able to cast the type.
2019-04-24 12:12:50 +00:00
addressing_model = other.addressing_model;
memory_model = other.memory_model;
Make sure ids_for_types are moved correctly in move operator.
2019-04-09 13:16:37 +00:00
default_entry_point = other.default_entry_point;
source = other.source;
Start considering how to emit physical type ID.
2019-07-18 15:05:28 +00:00
loop_iteration_depth_hard = other.loop_iteration_depth_hard;
loop_iteration_depth_soft = other.loop_iteration_depth_soft;
Explicitly implement move operators for ParsedIR. Workaround for MSVC 2013.
2019-04-09 08:25:41 +00:00
}
return *this;
}
Make sure ids_for_types are moved correctly in move operator.
2019-04-09 13:16:37 +00:00
ParsedIR::ParsedIR(const ParsedIR &other)
: ParsedIR()
{
*this = other;
}
Reduce pressure on global allocation. - Replace ostringstream with custom implementation. ~30% performance uplift on vector-shuffle-oom test. Allocations are measurably reduced in Valgrind. - Replace std::vector with SmallVector. Classic malloc optimization, small vectors are backed by inline data. ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux. - Use an object pool for IVariant type. We generally allocate a lot of SPIR* objects. We can amortize these allocations neatly by pooling them. - ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-02 09:19:03 +00:00
ParsedIR &ParsedIR::operator=(const ParsedIR &other)
{
if (this != &other)
{
spirv = other.spirv;
meta = other.meta;
for (int i = 0; i < TypeCount; i++)
ids_for_type[i] = other.ids_for_type[i];
ids_for_constant_or_type = other.ids_for_constant_or_type;
ids_for_constant_or_variable = other.ids_for_constant_or_variable;
declared_capabilities = other.declared_capabilities;
declared_extensions = other.declared_extensions;
block_meta = other.block_meta;
continue_block_to_loop_header = other.continue_block_to_loop_header;
entry_points = other.entry_points;
default_entry_point = other.default_entry_point;
source = other.source;
Start considering how to emit physical type ID.
2019-07-18 15:05:28 +00:00
loop_iteration_depth_hard = other.loop_iteration_depth_hard;
loop_iteration_depth_soft = other.loop_iteration_depth_soft;
GLSL: Implement GL_EXT_buffer_reference. Buffer objects can contain arbitrary pointers to blocks. We can also implement ConvertPtrToU and ConvertUToPtr. The latter can cast a uint64_t to any type as it pleases, so we will need to generate fake buffer reference blocks to be able to cast the type.
2019-04-24 12:12:50 +00:00
addressing_model = other.addressing_model;
memory_model = other.memory_model;
Reduce pressure on global allocation. - Replace ostringstream with custom implementation. ~30% performance uplift on vector-shuffle-oom test. Allocations are measurably reduced in Valgrind. - Replace std::vector with SmallVector. Classic malloc optimization, small vectors are backed by inline data. ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux. - Use an object pool for IVariant type. We generally allocate a lot of SPIR* objects. We can amortize these allocations neatly by pooling them. - ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-02 09:19:03 +00:00
// Very deliberate copying of IDs. There is no default copy constructor, nor a simple default constructor.
// Construct object first so we have the correct allocator set-up, then we can copy object into our new pool group.
ids.clear();
ids.reserve(other.ids.size());
for (size_t i = 0; i < other.ids.size(); i++)
{
ids.emplace_back(pool_group.get());
ids.back() = other.ids[i];
}
}
return *this;
}
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
void ParsedIR::set_id_bounds(uint32_t bounds)
{
Reduce pressure on global allocation. - Replace ostringstream with custom implementation. ~30% performance uplift on vector-shuffle-oom test. Allocations are measurably reduced in Valgrind. - Replace std::vector with SmallVector. Classic malloc optimization, small vectors are backed by inline data. ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux. - Use an object pool for IVariant type. We generally allocate a lot of SPIR* objects. We can amortize these allocations neatly by pooling them. - ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-02 09:19:03 +00:00
ids.reserve(bounds);
while (ids.size() < bounds)
ids.emplace_back(pool_group.get());
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
block_meta.resize(bounds);
}
Roll custom versions of isalpha/isalnum. Avoids locale issues and avoids unexpected assertions for non-ASCII values.
2020-01-06 12:04:57 +00:00
// Roll our own versions of these functions to avoid potential locale shenanigans.
static bool is_alpha(char c)
{
return (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z');
}
static bool is_alphanumeric(char c)
{
return is_alpha(c) || (c >= '0' && c <= '9');
}
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
static string ensure_valid_identifier(const string &name, bool member)
{
// Functions in glslangValidator are mangled with name(<mangled> stuff.
// Normally, we would never see '(' in any legal identifiers, so just strip them out.
auto str = name.substr(0, name.find('('));
for (uint32_t i = 0; i < str.size(); i++)
{
auto &c = str[i];
if (member)
{
// _m<num> variables are reserved by the internal implementation,
// otherwise, make sure the name is a valid identifier.
if (i == 0)
Roll custom versions of isalpha/isalnum. Avoids locale issues and avoids unexpected assertions for non-ASCII values.
2020-01-06 12:04:57 +00:00
c = is_alpha(c) ? c : '_';
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
else if (i == 2 && str[0] == '_' && str[1] == 'm')
Roll custom versions of isalpha/isalnum. Avoids locale issues and avoids unexpected assertions for non-ASCII values.
2020-01-06 12:04:57 +00:00
c = is_alpha(c) ? c : '_';
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
else
Roll custom versions of isalpha/isalnum. Avoids locale issues and avoids unexpected assertions for non-ASCII values.
2020-01-06 12:04:57 +00:00
c = is_alphanumeric(c) ? c : '_';
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
else
{
// _<num> variables are reserved by the internal implementation,
// otherwise, make sure the name is a valid identifier.
if (i == 0 || (str[0] == '_' && i == 1))
Roll custom versions of isalpha/isalnum. Avoids locale issues and avoids unexpected assertions for non-ASCII values.
2020-01-06 12:04:57 +00:00
c = is_alpha(c) ? c : '_';
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
else
Roll custom versions of isalpha/isalnum. Avoids locale issues and avoids unexpected assertions for non-ASCII values.
2020-01-06 12:04:57 +00:00
c = is_alphanumeric(c) ? c : '_';
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
}
return str;
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
const string &ParsedIR::get_name(ID id) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (m)
return m->decoration.alias;
else
return empty_string;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
const string &ParsedIR::get_member_name(TypeID id, uint32_t index) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (m)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
if (index >= m->members.size())
return empty_string;
return m->members[index].alias;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
else
return empty_string;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::set_name(ID id, const string &name)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
auto &str = meta[id].decoration.alias;
str.clear();
if (name.empty())
return;
// Reserved for temporaries.
if (name[0] == '_' && name.size() >= 2 && isdigit(name[1]))
return;
str = ensure_valid_identifier(name, false);
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::set_member_name(TypeID id, uint32_t index, const string &name)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
meta[id].members.resize(max(meta[id].members.size(), size_t(index) + 1));
auto &str = meta[id].members[index].alias;
str.clear();
if (name.empty())
return;
// Reserved for unnamed members.
if (name[0] == '_' && name.size() >= 3 && name[1] == 'm' && isdigit(name[2]))
return;
str = ensure_valid_identifier(name, true);
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::set_decoration_string(ID id, Decoration decoration, const string &argument)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
auto &dec = meta[id].decoration;
dec.decoration_flags.set(decoration);
switch (decoration)
{
case DecorationHlslSemanticGOOGLE:
dec.hlsl_semantic = argument;
break;
default:
break;
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::set_decoration(ID id, Decoration decoration, uint32_t argument)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
auto &dec = meta[id].decoration;
dec.decoration_flags.set(decoration);
switch (decoration)
{
case DecorationBuiltIn:
dec.builtin = true;
dec.builtin_type = static_cast<BuiltIn>(argument);
break;
case DecorationLocation:
dec.location = argument;
break;
case DecorationComponent:
dec.component = argument;
break;
case DecorationOffset:
dec.offset = argument;
break;
GLSL: Support GL_ARB_enchanced_layouts for XFB.
2020-01-27 11:56:48 +00:00
case DecorationXfbBuffer:
dec.xfb_buffer = argument;
break;
case DecorationXfbStride:
dec.xfb_stride = argument;
break;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
case DecorationArrayStride:
dec.array_stride = argument;
break;
case DecorationMatrixStride:
dec.matrix_stride = argument;
break;
case DecorationBinding:
dec.binding = argument;
break;
case DecorationDescriptorSet:
dec.set = argument;
break;
case DecorationInputAttachmentIndex:
dec.input_attachment = argument;
break;
case DecorationSpecId:
dec.spec_id = argument;
break;
case DecorationIndex:
dec.index = argument;
break;
case DecorationHlslCounterBufferGOOGLE:
meta[id].hlsl_magic_counter_buffer = argument;
Remove old hack for dealing with HLSL counter buffers. No longer needed.
2018-11-22 09:23:58 +00:00
meta[argument].hlsl_is_magic_counter_buffer = true;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
break;
Handle FPRoundingMode decoration. Doesn't do anything, but we can parse it correctly for now.
2019-01-07 09:51:44 +00:00
case DecorationFPRoundingMode:
dec.fp_rounding_mode = static_cast<FPRoundingMode>(argument);
break;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
default:
break;
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::set_member_decoration(TypeID id, uint32_t index, Decoration decoration, uint32_t argument)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
meta[id].members.resize(max(meta[id].members.size(), size_t(index) + 1));
auto &dec = meta[id].members[index];
dec.decoration_flags.set(decoration);
switch (decoration)
{
case DecorationBuiltIn:
dec.builtin = true;
dec.builtin_type = static_cast<BuiltIn>(argument);
break;
case DecorationLocation:
dec.location = argument;
break;
case DecorationComponent:
dec.component = argument;
break;
case DecorationBinding:
dec.binding = argument;
break;
case DecorationOffset:
dec.offset = argument;
break;
GLSL: Support GL_ARB_enchanced_layouts for XFB.
2020-01-27 11:56:48 +00:00
case DecorationXfbBuffer:
dec.xfb_buffer = argument;
break;
case DecorationXfbStride:
dec.xfb_stride = argument;
break;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
case DecorationSpecId:
dec.spec_id = argument;
break;
case DecorationMatrixStride:
dec.matrix_stride = argument;
break;
case DecorationIndex:
dec.index = argument;
break;
default:
break;
}
}
// Recursively marks any constants referenced by the specified constant instruction as being used
// as an array length. The id must be a constant instruction (SPIRConstant or SPIRConstantOp).
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::mark_used_as_array_length(ID id)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
switch (ids[id].get_type())
{
case TypeConstant:
get<SPIRConstant>(id).is_used_as_array_length = true;
break;
case TypeConstantOp:
{
auto &cop = get<SPIRConstantOp>(id);
Fix ParsedIR::mark_used_as_array_length(uint32_t id)
2019-09-04 22:56:05 +00:00
if (cop.opcode == OpCompositeExtract)
mark_used_as_array_length(cop.arguments[0]);
else if (cop.opcode == OpCompositeInsert)
{
mark_used_as_array_length(cop.arguments[0]);
mark_used_as_array_length(cop.arguments[1]);
}
else
for (uint32_t arg_id : cop.arguments)
mark_used_as_array_length(arg_id);
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
break;
}
case TypeUndef:
break;
default:
assert(0);
}
}
Bitset ParsedIR::get_buffer_block_flags(const SPIRVariable &var) const
{
auto &type = get<SPIRType>(var.basetype);
assert(type.basetype == SPIRType::Struct);
// Some flags like non-writable, non-readable are actually found
// as member decorations. If all members have a decoration set, propagate
// the decoration up as a regular variable decoration.
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
Bitset base_flags;
auto *m = find_meta(var.self);
if (m)
base_flags = m->decoration.decoration_flags;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
if (type.member_types.empty())
return base_flags;
Bitset all_members_flags = get_member_decoration_bitset(type.self, 0);
for (uint32_t i = 1; i < uint32_t(type.member_types.size()); i++)
all_members_flags.merge_and(get_member_decoration_bitset(type.self, i));
base_flags.merge_or(all_members_flags);
return base_flags;
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
const Bitset &ParsedIR::get_member_decoration_bitset(TypeID id, uint32_t index) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (m)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
if (index >= m->members.size())
return cleared_bitset;
return m->members[index].decoration_flags;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
else
return cleared_bitset;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
bool ParsedIR::has_decoration(ID id, Decoration decoration) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
return get_decoration_bitset(id).get(decoration);
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
uint32_t ParsedIR::get_decoration(ID id, Decoration decoration) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (!m)
return 0;
auto &dec = m->decoration;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
if (!dec.decoration_flags.get(decoration))
return 0;
switch (decoration)
{
case DecorationBuiltIn:
return dec.builtin_type;
case DecorationLocation:
return dec.location;
case DecorationComponent:
return dec.component;
case DecorationOffset:
return dec.offset;
GLSL: Support GL_ARB_enchanced_layouts for XFB.
2020-01-27 11:56:48 +00:00
case DecorationXfbBuffer:
return dec.xfb_buffer;
case DecorationXfbStride:
return dec.xfb_stride;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
case DecorationBinding:
return dec.binding;
case DecorationDescriptorSet:
return dec.set;
case DecorationInputAttachmentIndex:
return dec.input_attachment;
case DecorationSpecId:
return dec.spec_id;
case DecorationArrayStride:
return dec.array_stride;
case DecorationMatrixStride:
return dec.matrix_stride;
case DecorationIndex:
return dec.index;
Handle FPRoundingMode decoration. Doesn't do anything, but we can parse it correctly for now.
2019-01-07 09:51:44 +00:00
case DecorationFPRoundingMode:
return dec.fp_rounding_mode;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
default:
return 1;
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
const string &ParsedIR::get_decoration_string(ID id, Decoration decoration) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (!m)
return empty_string;
auto &dec = m->decoration;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
if (!dec.decoration_flags.get(decoration))
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
return empty_string;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
switch (decoration)
{
case DecorationHlslSemanticGOOGLE:
return dec.hlsl_semantic;
default:
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
return empty_string;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::unset_decoration(ID id, Decoration decoration)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
auto &dec = meta[id].decoration;
dec.decoration_flags.clear(decoration);
switch (decoration)
{
case DecorationBuiltIn:
dec.builtin = false;
break;
case DecorationLocation:
dec.location = 0;
break;
case DecorationComponent:
dec.component = 0;
break;
case DecorationOffset:
dec.offset = 0;
break;
GLSL: Support GL_ARB_enchanced_layouts for XFB.
2020-01-27 11:56:48 +00:00
case DecorationXfbBuffer:
dec.xfb_buffer = 0;
break;
case DecorationXfbStride:
dec.xfb_stride = 0;
break;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
case DecorationBinding:
dec.binding = 0;
break;
case DecorationDescriptorSet:
dec.set = 0;
break;
case DecorationInputAttachmentIndex:
dec.input_attachment = 0;
break;
case DecorationSpecId:
dec.spec_id = 0;
break;
case DecorationHlslSemanticGOOGLE:
dec.hlsl_semantic.clear();
break;
Handle FPRoundingMode decoration. Doesn't do anything, but we can parse it correctly for now.
2019-01-07 09:51:44 +00:00
case DecorationFPRoundingMode:
dec.fp_rounding_mode = FPRoundingModeMax;
break;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
case DecorationHlslCounterBufferGOOGLE:
{
auto &counter = meta[id].hlsl_magic_counter_buffer;
if (counter)
{
meta[counter].hlsl_is_magic_counter_buffer = false;
counter = 0;
}
break;
}
default:
break;
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
bool ParsedIR::has_member_decoration(TypeID id, uint32_t index, Decoration decoration) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
return get_member_decoration_bitset(id, index).get(decoration);
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
uint32_t ParsedIR::get_member_decoration(TypeID id, uint32_t index, Decoration decoration) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (!m)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
return 0;
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
if (index >= m->members.size())
return 0;
auto &dec = m->members[index];
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
if (!dec.decoration_flags.get(decoration))
return 0;
switch (decoration)
{
case DecorationBuiltIn:
return dec.builtin_type;
case DecorationLocation:
return dec.location;
case DecorationComponent:
return dec.component;
case DecorationBinding:
return dec.binding;
case DecorationOffset:
return dec.offset;
GLSL: Support GL_ARB_enchanced_layouts for XFB.
2020-01-27 11:56:48 +00:00
case DecorationXfbBuffer:
return dec.xfb_buffer;
case DecorationXfbStride:
return dec.xfb_stride;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
case DecorationSpecId:
return dec.spec_id;
case DecorationIndex:
return dec.index;
default:
return 1;
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
const Bitset &ParsedIR::get_decoration_bitset(ID id) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (m)
{
auto &dec = m->decoration;
return dec.decoration_flags;
}
else
return cleared_bitset;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::set_member_decoration_string(TypeID id, uint32_t index, Decoration decoration, const string &argument)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
meta[id].members.resize(max(meta[id].members.size(), size_t(index) + 1));
auto &dec = meta[id].members[index];
dec.decoration_flags.set(decoration);
switch (decoration)
{
case DecorationHlslSemanticGOOGLE:
dec.hlsl_semantic = argument;
break;
default:
break;
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
const string &ParsedIR::get_member_decoration_string(TypeID id, uint32_t index, Decoration decoration) const
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto *m = find_meta(id);
if (m)
{
if (!has_member_decoration(id, index, decoration))
return empty_string;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
auto &dec = m->members[index];
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
switch (decoration)
{
case DecorationHlslSemanticGOOGLE:
return dec.hlsl_semantic;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
default:
return empty_string;
}
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
else
return empty_string;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::unset_member_decoration(TypeID id, uint32_t index, Decoration decoration)
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
{
auto &m = meta[id];
if (index >= m.members.size())
return;
auto &dec = m.members[index];
dec.decoration_flags.clear(decoration);
switch (decoration)
{
case DecorationBuiltIn:
dec.builtin = false;
break;
case DecorationLocation:
dec.location = 0;
break;
case DecorationComponent:
dec.component = 0;
break;
case DecorationOffset:
dec.offset = 0;
break;
GLSL: Support GL_ARB_enchanced_layouts for XFB.
2020-01-27 11:56:48 +00:00
case DecorationXfbBuffer:
dec.xfb_buffer = 0;
break;
case DecorationXfbStride:
dec.xfb_stride = 0;
break;
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
case DecorationSpecId:
dec.spec_id = 0;
break;
case DecorationHlslSemanticGOOGLE:
dec.hlsl_semantic.clear();
break;
default:
break;
}
}
uint32_t ParsedIR::increase_bound_by(uint32_t incr_amount)
{
auto curr_bound = ids.size();
auto new_bound = curr_bound + incr_amount;
Reduce pressure on global allocation. - Replace ostringstream with custom implementation. ~30% performance uplift on vector-shuffle-oom test. Allocations are measurably reduced in Valgrind. - Replace std::vector with SmallVector. Classic malloc optimization, small vectors are backed by inline data. ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux. - Use an object pool for IVariant type. We generally allocate a lot of SPIR* objects. We can amortize these allocations neatly by pooling them. - ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-02 09:19:03 +00:00
ids.reserve(ids.size() + incr_amount);
for (uint32_t i = 0; i < incr_amount; i++)
ids.emplace_back(pool_group.get());
Hoist out parsing module from spirv_cross::Compiler. This is a large refactor which splits out the SPIR-V parser from Compiler and moves it into its more appropriately named Parser module. The Parser is responsible for building a ParsedIR structure which is then consumed by one or more compilers. Compiler can take a ParsedIR by value or move reference. This should allow for optimal case for both multiple compilations and single compilation scenarios.
2018-10-05 09:30:57 +00:00
block_meta.resize(new_bound);
return uint32_t(curr_bound);
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::remove_typed_id(Types type, ID id)
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
{
auto &type_ids = ids_for_type[type];
type_ids.erase(remove(begin(type_ids), end(type_ids), id), end(type_ids));
}
void ParsedIR::reset_all_of_type(Types type)
{
for (auto &id : ids_for_type[type])
if (ids[id].get_type() == type)
ids[id].reset();
ids_for_type[type].clear();
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
void ParsedIR::add_typed_id(Types type, ID id)
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
{
Start considering how to emit physical type ID.
2019-07-18 15:05:28 +00:00
if (loop_iteration_depth_hard != 0)
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
SPIRV_CROSS_THROW("Cannot add typed ID while looping over it.");
Start considering how to emit physical type ID.
2019-07-18 15:05:28 +00:00
if (loop_iteration_depth_soft != 0)
{
if (!ids[id].empty())
SPIRV_CROSS_THROW("Cannot override IDs when loop is soft locked.");
return;
}
GLSL: Implement GL_EXT_buffer_reference. Buffer objects can contain arbitrary pointers to blocks. We can also implement ConvertPtrToU and ConvertUToPtr. The latter can cast a uint64_t to any type as it pleases, so we will need to generate fake buffer reference blocks to be able to cast the type.
2019-04-24 12:12:50 +00:00
if (ids[id].empty() || ids[id].get_type() != type)
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
{
GLSL: Implement GL_EXT_buffer_reference. Buffer objects can contain arbitrary pointers to blocks. We can also implement ConvertPtrToU and ConvertUToPtr. The latter can cast a uint64_t to any type as it pleases, so we will need to generate fake buffer reference blocks to be able to cast the type.
2019-04-24 12:12:50 +00:00
switch (type)
{
case TypeConstant:
ids_for_constant_or_variable.push_back(id);
ids_for_constant_or_type.push_back(id);
break;
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
GLSL: Implement GL_EXT_buffer_reference. Buffer objects can contain arbitrary pointers to blocks. We can also implement ConvertPtrToU and ConvertUToPtr. The latter can cast a uint64_t to any type as it pleases, so we will need to generate fake buffer reference blocks to be able to cast the type.
2019-04-24 12:12:50 +00:00
case TypeVariable:
ids_for_constant_or_variable.push_back(id);
break;
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
GLSL: Implement GL_EXT_buffer_reference. Buffer objects can contain arbitrary pointers to blocks. We can also implement ConvertPtrToU and ConvertUToPtr. The latter can cast a uint64_t to any type as it pleases, so we will need to generate fake buffer reference blocks to be able to cast the type.
2019-04-24 12:12:50 +00:00
case TypeType:
case TypeConstantOp:
ids_for_constant_or_type.push_back(id);
break;
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
GLSL: Implement GL_EXT_buffer_reference. Buffer objects can contain arbitrary pointers to blocks. We can also implement ConvertPtrToU and ConvertUToPtr. The latter can cast a uint64_t to any type as it pleases, so we will need to generate fake buffer reference blocks to be able to cast the type.
2019-04-24 12:12:50 +00:00
default:
break;
}
Rewrite how IDs are iterated over. This is a fairly fundamental change on how IDs are handled. It serves many purposes: - Improve performance. We only need to iterate over IDs which are relevant at any one time. - Makes sure we iterate through IDs in SPIR-V module declaration order rather than ID space. IDs don't have to be monotonically increasing, which was an assumption SPIRV-Cross used to have. It has apparently never been a problem until now. - Support LUTs of structs. We do this by interleaving declaration of constants and struct types in SPIR-V module order. To support this, the ParsedIR interface needed to change slightly. Before setting any ID with variant_set<T> we let ParsedIR know that an ID with a specific type has been added. The surface for change should be minimal. ParsedIR will maintain a per-type list of IDs which the cross-compiler will need to consider for later. Instead of looping over ir.ids[] (which can be extremely large), we loop over types now, using: ir.for_each_typed_id<SPIRVariable>([&](uint32_t id, SPIRVariable &var) { handle_variable(var); }); Now we make sure that we're never looking at irrelevant types.
2019-01-10 08:49:33 +00:00
}
if (ids[id].empty())
{
ids_for_type[type].push_back(id);
}
else if (ids[id].get_type() != type)
{
remove_typed_id(ids[id].get_type(), id);
ids_for_type[type].push_back(id);
}
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
const Meta *ParsedIR::find_meta(ID id) const
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
{
auto itr = meta.find(id);
if (itr != end(meta))
return &itr->second;
else
return nullptr;
}
Refactor into stronger types in public API. Some fallout where internal functions are using stronger types. Overkill to move everything over to strong types right now, but perhaps move over to it slowly over time.
2019-09-05 10:43:40 +00:00
Meta *ParsedIR::find_meta(ID id)
Make meta a hashmap. A flat array was consuming way too much memory and was far too slow to initialize properly with a very large ID bound (8 million IDs, showed up as #1 hotspot in perf). Meta struct does not have to be in-order as we never iterate over it in a meaningful way, so using a hashmap here is reasonable. Very few IDs should need decorations or meta-data, so this should also be a quite decent memory save. For the pathological case, a 6x uplift was observed.
2019-01-10 13:04:01 +00:00
{
auto itr = meta.find(id);
if (itr != end(meta))
return &itr->second;
else
return nullptr;
}
Start considering how to emit physical type ID.
2019-07-18 15:05:28 +00:00
ParsedIR::LoopLock ParsedIR::create_loop_hard_lock() const
{
return ParsedIR::LoopLock(&loop_iteration_depth_hard);
}
ParsedIR::LoopLock ParsedIR::create_loop_soft_lock() const
MSL: Begin rewrite of buffer packing logic.
2019-07-18 11:34:47 +00:00
{
Start considering how to emit physical type ID.
2019-07-18 15:05:28 +00:00
return ParsedIR::LoopLock(&loop_iteration_depth_soft);
MSL: Begin rewrite of buffer packing logic.
2019-07-18 11:34:47 +00:00
}
ParsedIR::LoopLock::~LoopLock()
{
if (lock)
(*lock)--;
}
ParsedIR::LoopLock::LoopLock(uint32_t *lock_)
Run format_all.sh.
2019-07-23 10:23:41 +00:00
: lock(lock_)
MSL: Begin rewrite of buffer packing logic.
2019-07-18 11:34:47 +00:00
{
if (lock)
(*lock)++;
}
ParsedIR::LoopLock::LoopLock(LoopLock &&other) SPIRV_CROSS_NOEXCEPT
{
*this = move(other);
}
ParsedIR::LoopLock &ParsedIR::LoopLock::operator=(LoopLock &&other) SPIRV_CROSS_NOEXCEPT
{
if (lock)
(*lock)--;
lock = other.lock;
other.lock = nullptr;
return *this;
}
Add support for forcefully zero-initialized variables. Useful to better support certain platforms which require all variables to be initialized to something.
2020-03-26 10:21:23 +00:00
void ParsedIR::make_constant_null(uint32_t id, uint32_t type, bool add_to_typed_id_set)
{
auto &constant_type = get<SPIRType>(type);
if (constant_type.pointer)
{
if (add_to_typed_id_set)
add_typed_id(TypeConstant, id);
auto &constant = variant_set<SPIRConstant>(ids[id], type);
constant.self = id;
constant.make_null(constant_type);
}
else if (!constant_type.array.empty())
{
assert(constant_type.parent_type);
uint32_t parent_id = increase_bound_by(1);
make_constant_null(parent_id, constant_type.parent_type, add_to_typed_id_set);
if (!constant_type.array_size_literal.back())
SPIRV_CROSS_THROW("Array size of OpConstantNull must be a literal.");
SmallVector<uint32_t> elements(constant_type.array.back());
for (uint32_t i = 0; i < constant_type.array.back(); i++)
elements[i] = parent_id;
if (add_to_typed_id_set)
add_typed_id(TypeConstant, id);
variant_set<SPIRConstant>(ids[id], type, elements.data(), uint32_t(elements.size()), false).self = id;
}
else if (!constant_type.member_types.empty())
{
uint32_t member_ids = increase_bound_by(uint32_t(constant_type.member_types.size()));
SmallVector<uint32_t> elements(constant_type.member_types.size());
for (uint32_t i = 0; i < constant_type.member_types.size(); i++)
{
make_constant_null(member_ids + i, constant_type.member_types[i], add_to_typed_id_set);
elements[i] = member_ids + i;
}
if (add_to_typed_id_set)
add_typed_id(TypeConstant, id);
variant_set<SPIRConstant>(ids[id], type, elements.data(), uint32_t(elements.size()), false).self = id;
}
else
{
if (add_to_typed_id_set)
add_typed_id(TypeConstant, id);
auto &constant = variant_set<SPIRConstant>(ids[id], type);
constant.self = id;
constant.make_null(constant_type);
}
}
Reduce pressure on global allocation. - Replace ostringstream with custom implementation. ~30% performance uplift on vector-shuffle-oom test. Allocations are measurably reduced in Valgrind. - Replace std::vector with SmallVector. Classic malloc optimization, small vectors are backed by inline data. ~ 7-8% gain on vector-shuffle-oom on GCC 8 on Linux. - Use an object pool for IVariant type. We generally allocate a lot of SPIR* objects. We can amortize these allocations neatly by pooling them. - ~15% overall uplift on ./test_shaders.py --iterations 10000 shaders/.
2019-04-02 09:19:03 +00:00
} // namespace SPIRV_CROSS_NAMESPACE
Reference in New Issue Copy Permalink