Hardware testing revealed that SSY and PBK push to a different stack,
allowing code like this:
SSY label1;
PBK label2;
SYNC;
label1: PBK;
label2: EXIT;
Analysis passes do not have a good reason to depend on shader_ir.h to
work on top of nodes. This splits node-related declarations to their own
file and leaves the IR in shader_ir.h
Instead of having a vector of unique_ptr stored in a vector and
returning star pointers to this, use shared_ptr. While changing
initialization code, move it to a separate file when possible.
This is a first step to allow code analysis and node generation beyond
the ShaderIR class.
"position" was being written but not read anywhere besides geometry
shaders, where it had the same value as gl_Position.
This commit replaces "position" with gl_Position, reducing the
complexity of our code and the emitted GLSL code.
Fix missing OpSelectionMerge instruction. This caused devices loses on
most hardware, Intel didn't care.
Fix [-1;1] -> [0;1] depth conversions.
Conditionally use VK_EXT_scalar_block_layout. This allows us to use
non-std140 layouts on UBOs.
Update external Vulkan headers.
Keeps track of native ASTC support, VK_EXT_scalar_block_layout
availability and SSBO range.
Check for independentBlend and vertexPipelineStorageAndAtomics as a
required feature. Always enable it.
Use vk::to_string format to log Vulkan enums.
Style changes.
Uses a std::string_view instead of a std::string, given the pointed to
string isn't modified and is only used in a formatting operation.
This is nice because a few usages directly supply a string literal to
the function, allowing these usages to otherwise not heap allocate,
unlike the std::string overloads.
While we're at it, we can combine the address formatting into a single
formatting call.
The following code is broken on AMD's proprietary GLSL compiler:
```glsl
uint idx = ...;
vec4 values = ...;
float some_value = values[idx & 3];
```
It index the wrong components, to fix this the following pessimized code
is emitted when that bug is present:
```glsl
uint idx = ...;
vec4 values = ...;
float some_value;
if ((idx & 3) == 0) some_value = values.x;
if ((idx & 3) == 1) some_value = values.y;
if ((idx & 3) == 2) some_value = values.z;
if ((idx & 3) == 3) some_value = values.w;
```
Component indexing on AMD's proprietary driver is broken. This commit adds
a test to detect when we are on a driver that can't successfully manage
component indexing.
It dispatches a dummy draw with just one vertex shader that writes to an
indexed SSBO from the GPU with data sent through uniforms, it then reads
that data from the CPU and compares the expected output.
This allows for forming comment nodes without making unnecessary copies
of the std::string instance.
e.g. previously:
Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]",
cbuf->GetIndex(), cbuf_offset));
Would result in a copy of the string being created, as CommentNode()
takes a std::string by value (a const ref passed to a value parameter
results in a copy).
Now, only one instance of the string is ever moved around. (fmt::format
returns a std::string, and since it's returned from a function by value,
this is a prvalue (which can be treated like an rvalue), so it's moved
into Comment's string parameter), we then move it into the CommentNode
constructor, which then moves the string into its member variable).
Amends cases where we were using things that were indirectly being
satisfied through other headers. This way, if those headers change and
eliminate dependencies on other headers in the future, we don't have
cascading compilation errors.
Gets rid of the need to special-case brace handling depending on the
overload used, and makes it consistent across the board with how fmt
handles them.
Strings with compile-time deducible strings are directly forwarded to
std::string's constructor, so we don't need to worry about the
performance difference here, as it'll be identical.
In a lot of places throughout the decompiler, string concatenation via
operator+ is used quite heavily. This is usually fine, when not heavily
used, but when used extensively, can be a problem. operator+ creates an
entirely new heap allocated temporary string and given we perform
expressions like:
std::string thing = a + b + c + d;
this ends up with a lot of unnecessary temporary strings being created
and discarded, which kind of thrashes the heap more than we need to.
Given we utilize fmt in some AddLine calls, we can make this a part of
the ShaderWriter's API. We can make an overload that simply acts as a
passthrough to fmt.
This way, whenever things need to be appended to a string, the operation
can be done via a single string formatting operation instead of
discarding numerous temporary strings. This also has the benefit of
making the strings themselves look nicer and makes it easier to spot
errors in them.
Many of these constructors don't even need to be templated. The only
ones that need to be templated are the ones that actually make use of
the parameter pack.
Even then, since std::vector accepts an initializer list, we can supply
the parameter pack directly to it instead of creating our own copy of
the list, then copying it again into the std::vector.
Given the class contains quite a lot of non-trivial types, place the
constructor and destructor within the cpp file to avoid inlining
construction and destruction code everywhere the class is used.
Avoids performing copies into the pair being returned. Instead, we can
just move the resources into the pair, avoiding the need to make copies
of both the std::string and ShaderEntries struct.
Given the offset is assigned a fixed value in the constructor, we can
just assign it directly and get rid of the need to write the name of the
variable again in the constructor initializer list.
Given the disk shader cache contains non-trivial types, we should
default it in the cpp file in order to prevent inlining of the
complex destruction logic.
The standard library expects hash specializations that don't throw
exceptions. Make this explicit in the type to allow selection of better
code paths if possible in implementations.
We don't need to load the code into a vector and then construct a string
over the data. We can just create a string with the necessary size ahead
of time, and read the data directly into it, getting rid of an
unnecessary heap allocation.
std::move does nothing when applied to a const variable. Resources can't
be moved if the object is immutable. With this change, we don't end up
making several unnecessary heap allocations and copies.
Booleans don't have a guaranteed size, but we still want to have them
integrate into the disk cache system without needing to actually use a
different type. We can do this by supplying non-template overloads for
the bool type.
Non-template overloads always have precedence during function
resolution, so this is safe to provide.
This gets rid of the need to smatter ternary conditionals, as well as
the need to use u8 types to store the value in.
This was reduced due to happening on most games and at such constant
rate that it affected performance heavily for the end user. In general,
we are well aware of the assert and an implementation is already
planned.
Avoids inlining destruction logic where applicable, and also makes
forward declarations not cause unexpected compilation errors depending
on where the State class is used.
Lessens the amount of code that needs to be read, and gets rid of the
need to introduce an indexing variable. Instead, we just operate on the
objects directly.
std::memset is used to clear the entire register structure, which
requires that the Regs struct be trivially copyable (otherwise undefined
behavior is invoked). This prevents the case where a non-trivial type is
potentially added to the struct.
std::move within a copy constructor (on a data member that isn't
mutable) will always result in a copy. Because of that, the behavior of
this copy constructor is identical to the one that would be generated
automatically by the compiler, so we can remove it.
