moonpower/azahar-UWP
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

shader_jit: Fix/optimize conditional evaluation (#234)

Browse Source 
* shader_jit: Add conditional unit-tests

Tests all permutations of X, Y, AND, OR with each possible input value.

* video_core: Fix shader-interpreter conditional-code initialization

Rather than reserving the incoming state of the conditional codes, the
shader-interpreter was setting them both to false. In pretty much all
cases, the initial state of a shaderunit can be zero-initialized
statically. Just running the interpreter shouldn't necessarily reset the
conditional codes though.  The JIT loads incoming conditional codes
while the shader-interpreter resets them to false. This makes the
interpreter match the behavior of the shader-jit.

* shader_jit_a64: Fix/optimize conditional evaluation

Fix some of the regressions introduced by the previous optimization.
EOR does not support a constant of `0` in its immediate. In these cases
the COND{0,1} registers can be utilized immediately.

* shader_jit_x64: Fix conditional evaluation extended-bit hazard

The unit test seems to have identified a bug in the x64 jit too. The x64
jit was doing 32-bit comparisons despite the condition flags being 8-bit
values and is sensitive to garbage being in the upper 24 bits of the
register. This is fixed by using the proper 8-bit register types rather
than the 32-bit ones(`eax,`ebx` -> `al`, `bl`).

* shader_jit_x64: Zero-extend conditional-code bytes

`mov` was doing a partial update of bits within the register, allowing
garbage to be introduced in the upper bits of the register.
This commit is contained in:
Wunk
2024-08-20 01:19:04 -07:00
committed by PabloMK7
parent cfc74d2d08
commit 0ea9f23f0c
6 changed files with 144 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/video_core/pica/shader_unit.cpp
View File

@@ -9,10 +9,7 @@
namespace Pica {
ShaderUnit::ShaderUnit(GeometryEmitter* emitter) : emitter_ptr{emitter} {
const Common::Vec4<f24> temp_vec{f24::Zero(), f24::Zero(), f24::Zero(), f24::One()};
temporary.fill(temp_vec);
}
ShaderUnit::ShaderUnit(GeometryEmitter* emitter) : emitter_ptr{emitter} {}
ShaderUnit::~ShaderUnit() = default;

10
src/video_core/pica/shader_unit.h
View File

@@ -46,11 +46,11 @@ struct ShaderUnit {
}
public:
s32 address_registers[3];
bool conditional_code[2];
alignas(16) std::array<Common::Vec4<f24>, 16> input;
alignas(16) std::array<Common::Vec4<f24>, 16> temporary;
alignas(16) std::array<Common::Vec4<f24>, 16> output;
s32 address_registers[3] = {};
bool conditional_code[2] = {};
alignas(16) std::array<Common::Vec4<f24>, 16> input = {};
alignas(16) std::array<Common::Vec4<f24>, 16> temporary = {};
alignas(16) std::array<Common::Vec4<f24>, 16> output = {};
GeometryEmitter* emitter_ptr;
private: