purgatory/yuzu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

shader_recompiler: Use vector arithmetic rather than component-wise in ImageGatherSubpixelOffset

Browse Source 
Should be more efficient and better readable
This commit is contained in:
Wollnashorn 2023-04-05 18:00:35 +02:00
parent fe91066f46
commit 82b78cde73
1 changed files with 9 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
View File

@ -266,30 +266,21 @@ Id ImageGatherSubpixelOffset(EmitContext& ctx, const IR::TextureInstInfo& info,
Id coords) { Id coords) {
// Apply a subpixel offset of 1/512 the texel size of the texture to ensure same rounding on // Apply a subpixel offset of 1/512 the texel size of the texture to ensure same rounding on
// AMD hardware as on Maxwell or other Nvidia architectures. // AMD hardware as on Maxwell or other Nvidia architectures.
const auto calculate_offset{[&](size_t dim) -> std::array<Id, 2> { const auto calculate_coords{[&](size_t dim) {
const Id nudge{ctx.Const(0x1p-9f)}; const Id nudge{ctx.Const(0x1p-9f)};
const Id image_size{ctx.OpImageQuerySizeLod(ctx.U32[dim], texture, ctx.u32_zero_value)}; const Id image_size{ctx.OpImageQuerySizeLod(ctx.U32[dim], texture, ctx.u32_zero_value)};
const Id offset_x{ctx.OpFDiv( Id offset{dim == 2 ? ctx.ConstantComposite(ctx.F32[dim], nudge, nudge)
ctx.F32[1], nudge, : ctx.ConstantComposite(ctx.F32[dim], nudge, nudge, ctx.f32_zero_value)};
ctx.OpConvertUToF(ctx.F32[1], ctx.OpCompositeExtract(ctx.U32[1], image_size, 0)))}; offset = ctx.OpFDiv(ctx.F32[dim], offset, ctx.OpConvertUToF(ctx.F32[dim], image_size));
const Id offset_y{ctx.OpFDiv( return ctx.OpFAdd(ctx.F32[dim], coords, offset);
ctx.F32[1], nudge,
ctx.OpConvertUToF(ctx.F32[1], ctx.OpCompositeExtract(ctx.U32[1], image_size, 1)))};
return {ctx.OpFAdd(ctx.F32[1], ctx.OpCompositeExtract(ctx.F32[1], coords, 0), offset_x),
ctx.OpFAdd(ctx.F32[1], ctx.OpCompositeExtract(ctx.F32[1], coords, 1), offset_y)};
}}; }};
switch (info.type) { switch (info.type) {
case TextureType::Color2D: case TextureType::Color2D:
case TextureType::Color2DRect: { case TextureType::Color2DRect:
const auto offset{calculate_offset(2)}; return calculate_coords(2);
return ctx.OpCompositeConstruct(ctx.F32[2], offset[0], offset[1]);
}
case TextureType::ColorArray2D: case TextureType::ColorArray2D:
case TextureType::ColorCube: { case TextureType::ColorCube:
const auto offset{calculate_offset(3)}; return calculate_coords(3);
return ctx.OpCompositeConstruct(ctx.F32[3], offset[0], offset[1],
ctx.OpCompositeExtract(ctx.F32[1], coords, 2));
}
default: default:
return coords; return coords;
} }