Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

AtlasEngine: Improve glyph generation performance #13477

Merged
1 commit merged into from
Jul 11, 2022
Merged

AtlasEngine: Improve glyph generation performance #13477

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 1 addition & 5 deletions src/renderer/atlas/AtlasEngine.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -929,8 +929,7 @@ void AtlasEngine::_recreateFontDependentResources()
{
// We're likely resizing the atlas anyways and can
// thus also release any of these buffers prematurely.
_r.d2dRenderTarget.reset(); // depends on _r.atlasScratchpad
_r.atlasScratchpad.reset();
_r.d2dRenderTarget.reset(); // depends on _r.atlasBuffer
_r.atlasView.reset();
_r.atlasBuffer.reset();
}
Expand Down Expand Up @@ -970,8 +969,6 @@ void AtlasEngine::_recreateFontDependentResources()
_r.strikethroughPos = _api.fontMetrics.strikethroughPos;
_r.lineThickness = _api.fontMetrics.lineThickness;
_r.dpi = _api.dpi;
_r.maxEncounteredCellCount = 0;
_r.scratchpadCellWidth = 0;
}
{
// See AtlasEngine::UpdateFont.
Expand Down Expand Up @@ -1446,7 +1443,6 @@ void AtlasEngine::_emplaceGlyph(IDWriteFontFace* fontFace, size_t bufferPos1, si
const auto it = _r.glyphs.insert(std::move(key), std::move(value));
valueRef = &it->second;
_r.glyphQueue.emplace_back(&it->first, &it->second);
_r.maxEncounteredCellCount = std::max(_r.maxEncounteredCellCount, cellCount);
}

// For some reason MSVC doesn't understand that valueRef is overwritten in the branch above, resulting in:
Expand Down
5 changes: 0 additions & 5 deletions src/renderer/atlas/AtlasEngine.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -857,11 +857,9 @@ namespace Microsoft::Console::Render
void _setShaderResources() const;
void _updateConstantBuffer() const noexcept;
void _adjustAtlasSize();
void _reserveScratchpadSize(u16 minWidth);
void _processGlyphQueue();
void _drawGlyph(const AtlasQueueItem& item) const;
void _drawCursor();
void _copyScratchpadTile(uint32_t scratchpadIndex, u16x2 target, uint32_t copyFlags = 0) const noexcept;

static constexpr bool debugGlyphGenerationPerformance = false;
static constexpr bool debugGeneralPerformance = false || debugGlyphGenerationPerformance;
Expand Down Expand Up @@ -906,7 +904,6 @@ namespace Microsoft::Console::Render
// D2D resources
wil::com_ptr<ID3D11Texture2D> atlasBuffer;
wil::com_ptr<ID3D11ShaderResourceView> atlasView;
wil::com_ptr<ID3D11Texture2D> atlasScratchpad;
wil::com_ptr<ID2D1RenderTarget> d2dRenderTarget;
wil::com_ptr<ID2D1Brush> brush;
wil::com_ptr<IDWriteTextFormat> textFormats[2][2];
Expand All @@ -921,8 +918,6 @@ namespace Microsoft::Console::Render
u16 strikethroughPos = 0;
u16 lineThickness = 0;
u16 dpi = USER_DEFAULT_SCREEN_DPI; // invalidated by ApiInvalidations::Font, caches _api.dpi
u16 maxEncounteredCellCount = 0;
u16 scratchpadCellWidth = 0;
u16x2 atlasSizeInPixel; // invalidated by ApiInvalidations::Font
TileHashMap glyphs;
TileAllocator tileAllocator;
Expand Down
91 changes: 23 additions & 68 deletions src/renderer/atlas/AtlasEngine.r.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,6 @@ using namespace Microsoft::Console::Render;
try
{
_adjustAtlasSize();
_reserveScratchpadSize(_r.maxEncounteredCellCount);
_processGlyphQueue();

if (WI_IsFlagSet(_r.invalidations, RenderInvalidations::Cursor))
Expand Down Expand Up @@ -86,6 +85,7 @@ try
}
catch (const wil::ResultException& exception)
{
// TODO: this writes to _api.
return _handleException(exception);
}
CATCH_RETURN()
Expand Down Expand Up @@ -159,7 +159,7 @@ void AtlasEngine::_adjustAtlasSize()
desc.ArraySize = 1;
desc.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
desc.SampleDesc = { 1, 0 };
desc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
desc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_RENDER_TARGET;
THROW_IF_FAILED(_r.device->CreateTexture2D(&desc, nullptr, atlasBuffer.addressof()));
THROW_IF_FAILED(_r.device->CreateShaderResourceView(atlasBuffer.get(), nullptr, atlasView.addressof()));
}
Expand All @@ -184,38 +184,8 @@ void AtlasEngine::_adjustAtlasSize()
_r.atlasView = std::move(atlasView);
_setShaderResources();

WI_SetFlagIf(_r.invalidations, RenderInvalidations::Cursor, !copyFromExisting);
}

void AtlasEngine::_reserveScratchpadSize(u16 minWidth)
{
if (minWidth <= _r.scratchpadCellWidth)
{
return;
}

// The new size is the greater of ... cells wide:
// * 2
// * minWidth
// * current size * 1.5
const auto newWidth = std::max<UINT>(std::max<UINT>(2, minWidth), _r.scratchpadCellWidth + (_r.scratchpadCellWidth >> 1));

_r.d2dRenderTarget.reset();
_r.atlasScratchpad.reset();

{
D3D11_TEXTURE2D_DESC desc{};
desc.Width = _r.cellSize.x * newWidth;
desc.Height = _r.cellSize.y;
desc.MipLevels = 1;
desc.ArraySize = 1;
desc.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
desc.SampleDesc = { 1, 0 };
desc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_RENDER_TARGET;
THROW_IF_FAILED(_r.device->CreateTexture2D(&desc, nullptr, _r.atlasScratchpad.put()));
}
{
const auto surface = _r.atlasScratchpad.query<IDXGISurface>();
const auto surface = _r.atlasBuffer.query<IDXGISurface>();

wil::com_ptr<IDWriteRenderingParams1> renderingParams;
DWrite_GetRenderParams(_sr.dwriteFactory.get(), &_r.gamma, &_r.cleartypeEnhancedContrast, &_r.grayscaleEnhancedContrast, renderingParams.addressof());
Expand Down Expand Up @@ -243,8 +213,8 @@ void AtlasEngine::_reserveScratchpadSize(u16 minWidth)
_r.brush = brush.query<ID2D1Brush>();
}

_r.scratchpadCellWidth = _r.maxEncounteredCellCount;
WI_SetAllFlags(_r.invalidations, RenderInvalidations::ConstBuffer);
WI_SetFlagIf(_r.invalidations, RenderInvalidations::Cursor, !copyFromExisting);
}

void AtlasEngine::_processGlyphQueue()
Expand All @@ -254,10 +224,12 @@ void AtlasEngine::_processGlyphQueue()
return;
}

_r.d2dRenderTarget->BeginDraw();
for (const auto& pair : _r.glyphQueue)
{
_drawGlyph(pair);
}
THROW_IF_FAILED(_r.d2dRenderTarget->EndDraw());

_r.glyphQueue.clear();
}
Expand All @@ -280,7 +252,7 @@ void AtlasEngine::_drawGlyph(const AtlasQueueItem& item) const
textLayout->SetTypography(_r.typography.get(), { 0, charsLength });
}

auto options = D2D1_DRAW_TEXT_OPTIONS_NONE;
auto options = D2D1_DRAW_TEXT_OPTIONS_CLIP;
// D2D1_DRAW_TEXT_OPTIONS_ENABLE_COLOR_FONT enables a bunch of internal machinery
// which doesn't have to run if we know we can't use it anyways in the shader.
WI_SetFlagIf(options, D2D1_DRAW_TEXT_OPTIONS_ENABLE_COLOR_FONT, coloredGlyph);
Expand All @@ -294,31 +266,29 @@ void AtlasEngine::_drawGlyph(const AtlasQueueItem& item) const
_r.d2dRenderTarget->SetTextAntialiasMode(coloredGlyph ? D2D1_TEXT_ANTIALIAS_MODE_GRAYSCALE : D2D1_TEXT_ANTIALIAS_MODE_CLEARTYPE);
}

_r.d2dRenderTarget->BeginDraw();
// We could call
// _r.d2dRenderTarget->PushAxisAlignedClip(&rect, D2D1_ANTIALIAS_MODE_ALIASED);
// now to reduce the surface that needs to be cleared, but this decreases
// performance by 10% (tested using debugGlyphGenerationPerformance).
_r.d2dRenderTarget->Clear();
_r.d2dRenderTarget->DrawTextLayout({}, textLayout.get(), _r.brush.get(), options);
THROW_IF_FAILED(_r.d2dRenderTarget->EndDraw());

for (u32 i = 0; i < cells; ++i)
{
// Specifying NO_OVERWRITE means that the system can assume that existing references to the surface that
// may be in flight on the GPU will not be affected by the update, so the copy can proceed immediately
// (avoiding either a batch flush or the system maintaining multiple copies of the resource behind the scenes).
//
// Since our shader only draws whatever is in the atlas, and since we don't replace glyph tiles that are in use,
// we can safely (?) tell the GPU that we don't overwrite parts of our atlas that are in use.
_copyScratchpadTile(i, coords[i], D3D11_COPY_NO_OVERWRITE);
const auto coord = coords[i];

D2D1_RECT_F rect;
rect.left = static_cast<float>(coord.x) * static_cast<float>(USER_DEFAULT_SCREEN_DPI) / static_cast<float>(_r.dpi);
rect.top = static_cast<float>(coord.y) * static_cast<float>(USER_DEFAULT_SCREEN_DPI) / static_cast<float>(_r.dpi);
Comment on lines +274 to +275
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This bit is a bit unrefined, but in the grand scheme of things isn't the biggest issue about this class. 😅
I'd much prefer if D2D would be DPI-unaware so that I can set rect and origin coordinates straight to pixel coordinates, but IIRC that's problematic for AA when using the non-text primitives (for instance when drawing the cursor). Might be worthwhile to investigate at another time.

rect.right = rect.left + _r.cellSizeDIP.x;
rect.bottom = rect.top + _r.cellSizeDIP.y;

D2D1_POINT_2F origin;
origin.x = rect.left - i * _r.cellSizeDIP.x;
origin.y = rect.top;

_r.d2dRenderTarget->PushAxisAlignedClip(&rect, D2D1_ANTIALIAS_MODE_ALIASED);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh, I see. Axis aligned clips -- i thought they were somewhat expensive. They're not?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

They are expensive, but copying the texture tiles seperately incurs a higher latency, whereas this approach can potentially do batched operations.

_r.d2dRenderTarget->Clear();
_r.d2dRenderTarget->DrawTextLayout(origin, textLayout.get(), _r.brush.get(), options);
_r.d2dRenderTarget->PopAxisAlignedClip();
}
}

void AtlasEngine::_drawCursor()
{
_reserveScratchpadSize(1);

// lineWidth is in D2D's DIPs. For instance if we have a 150-200% zoom scale we want to draw a 2px wide line.
// At 150% scale lineWidth thus needs to be 1.33333... because at a zoom scale of 1.5 this results in a 2px wide line.
const auto lineWidth = std::max(1.0f, static_cast<float>((_r.dpi + USER_DEFAULT_SCREEN_DPI / 2) / USER_DEFAULT_SCREEN_DPI * USER_DEFAULT_SCREEN_DPI) / static_cast<float>(_r.dpi));
Expand Down Expand Up @@ -377,19 +347,4 @@ void AtlasEngine::_drawCursor()
}

THROW_IF_FAILED(_r.d2dRenderTarget->EndDraw());

_copyScratchpadTile(0, {});
}

void AtlasEngine::_copyScratchpadTile(uint32_t scratchpadIndex, u16x2 target, uint32_t copyFlags) const noexcept
{
D3D11_BOX box;
box.left = scratchpadIndex * _r.cellSize.x;
box.top = 0;
box.front = 0;
box.right = box.left + _r.cellSize.x;
box.bottom = _r.cellSize.y;
box.back = 1;
#pragma warning(suppress : 26447) // The function is declared 'noexcept' but calls function '...' which may throw exceptions (f.6).
_r.deviceContext->CopySubresourceRegion1(_r.atlasBuffer.get(), 0, target.x, target.y, 0, _r.atlasScratchpad.get(), 0, &box, copyFlags);
}