| // |
| // Copyright 2015 The ANGLE Project Authors. All rights reserved. |
| // Use of this source code is governed by a BSD-style license that can be |
| // found in the LICENSE file. |
| // |
| |
| // FeaturesD3D.h: Features and workarounds for D3D driver bugs and other issues. |
| |
| #ifndef ANGLE_PLATFORM_FEATURESD3D_H_ |
| #define ANGLE_PLATFORM_FEATURESD3D_H_ |
| |
| #include "platform/Feature.h" |
| |
| namespace angle |
| { |
| |
| // Workarounds attached to each shader. Do not need to expose information about these workarounds so |
| // a simple bool struct suffices. |
| struct CompilerWorkaroundsD3D |
| { |
| bool skipOptimization = false; |
| |
| bool useMaxOptimization = false; |
| |
| // IEEE strictness needs to be enabled for NANs to work. |
| bool enableIEEEStrictness = false; |
| }; |
| |
| struct FeaturesD3D : FeatureSetBase |
| { |
| FeaturesD3D(); |
| ~FeaturesD3D(); |
| |
| // On some systems, having extra rendertargets than necessary slows down the shader. |
| // We can fix this by optimizing those out of the shader. At the same time, we can |
| // work around a bug on some nVidia drivers that they ignore "null" render targets |
| // in D3D11, by compacting the active color attachments list to omit null entries. |
| Feature mrtPerfWorkaround = {"mrt_perf_workaround", FeatureCategory::D3DWorkarounds, |
| "Some drivers have a bug where they ignore null render targets", |
| &members}; |
| |
| Feature setDataFasterThanImageUpload = {"set_data_faster_than_image_upload", |
| FeatureCategory::D3DWorkarounds, |
| "Set data faster than image upload", &members}; |
| |
| // Some renderers can't disable mipmaps on a mipmapped texture (i.e. solely sample from level |
| // zero, and ignore the other levels). D3D11 Feature Level 10+ does this by setting MaxLOD to |
| // 0.0f in the Sampler state. D3D9 sets D3DSAMP_MIPFILTER to D3DTEXF_NONE. There is no |
| // equivalent to this in D3D11 Feature Level 9_3. This causes problems when (for example) an |
| // application creates a mipmapped texture2D, but sets GL_TEXTURE_MIN_FILTER to GL_NEAREST |
| // (i.e disables mipmaps). To work around this, D3D11 FL9_3 has to create two copies of the |
| // texture. The textures' level zeros are identical, but only one texture has mips. |
| Feature zeroMaxLodWorkaround = {"zero_max_lod", FeatureCategory::D3DWorkarounds, |
| "Missing an option to disable mipmaps on a mipmapped texture", |
| &members}; |
| |
| // Some renderers do not support Geometry Shaders so the Geometry Shader-based PointSprite |
| // emulation will not work. To work around this, D3D11 FL9_3 has to use a different pointsprite |
| // emulation that is implemented using instanced quads. |
| Feature useInstancedPointSpriteEmulation = { |
| "use_instanced_point_sprite_emulation", FeatureCategory::D3DWorkarounds, |
| "Some D3D11 renderers do not support geometry shaders for pointsprite emulation", &members}; |
| |
| // A bug fixed in NVIDIA driver version 347.88 < x <= 368.81 triggers a TDR when using |
| // CopySubresourceRegion from a staging texture to a depth/stencil in D3D11. The workaround |
| // is to use UpdateSubresource to trigger an extra copy. We disable this workaround on newer |
| // NVIDIA driver versions because of a second driver bug present with the workaround enabled. |
| // (See: http://anglebug.com/1452) |
| Feature depthStencilBlitExtraCopy = { |
| "depth_stencil_blit_extra_copy", FeatureCategory::D3DWorkarounds, |
| "Bug in some drivers triggers a TDR when using CopySubresourceRegion from a staging " |
| "texture to a depth/stencil", |
| &members, "http://anglebug.com/1452"}; |
| |
| // The HLSL optimizer has a bug with optimizing "pow" in certain integer-valued expressions. |
| // We can work around this by expanding the pow into a series of multiplies if we're running |
| // under the affected compiler. |
| Feature expandIntegerPowExpressions = { |
| "expand_integer_pow_expressions", FeatureCategory::D3DWorkarounds, |
| "The HLSL optimizer has a bug with optimizing 'pow' in certain integer-valued expressions", |
| &members}; |
| |
| // NVIDIA drivers sometimes write out-of-order results to StreamOut buffers when transform |
| // feedback is used to repeatedly write to the same buffer positions. |
| Feature flushAfterEndingTransformFeedback = { |
| "flush_after_ending_transform_feedback", FeatureCategory::D3DWorkarounds, |
| "Some drivers sometimes write out-of-order results to StreamOut buffers when transform " |
| "feedback is used to repeatedly write to the same buffer positions", |
| &members}; |
| |
| // Some drivers (NVIDIA) do not take into account the base level of the texture in the results |
| // of the HLSL GetDimensions builtin. |
| Feature getDimensionsIgnoresBaseLevel = { |
| "get_dimensions_ignores_base_level", FeatureCategory::D3DWorkarounds, |
| "Some drivers do not take into account the base level of the " |
| "texture in the results of the HLSL GetDimensions builtin", |
| &members}; |
| |
| // On some Intel drivers, HLSL's function texture.Load returns 0 when the parameter Location |
| // is negative, even if the sum of Offset and Location is in range. This may cause errors when |
| // translating GLSL's function texelFetchOffset into texture.Load, as it is valid for |
| // texelFetchOffset to use negative texture coordinates as its parameter P when the sum of P |
| // and Offset is in range. To work around this, we translate texelFetchOffset into texelFetch |
| // by adding Offset directly to Location before reading the texture. |
| Feature preAddTexelFetchOffsets = { |
| "pre_add_texel_fetch_offsets", FeatureCategory::D3DWorkarounds, |
| "HLSL's function texture.Load returns 0 when the parameter Location is negative, even if " |
| "the sum of Offset and Location is in range", |
| &members}; |
| |
| // On some AMD drivers, 1x1 and 2x2 mips of depth/stencil textures aren't sampled correctly. |
| // We can work around this bug by doing an internal blit to a temporary single-channel texture |
| // before we sample. |
| Feature emulateTinyStencilTextures = { |
| "emulate_tiny_stencil_textures", FeatureCategory::D3DWorkarounds, |
| "1x1 and 2x2 mips of depth/stencil textures aren't sampled correctly", &members}; |
| |
| // In Intel driver, the data with format DXGI_FORMAT_B5G6R5_UNORM will be parsed incorrectly. |
| // This workaroud will disable B5G6R5 support when it's Intel driver. By default, it will use |
| // R8G8B8A8 format. This bug is fixed in version 4539 on Intel drivers. |
| // On older AMD drivers, the data in DXGI_FORMAT_B5G6R5_UNORM becomes corrupted for unknown |
| // reasons. |
| Feature disableB5G6R5Support = {"disable_b5g6r5_support", FeatureCategory::D3DWorkarounds, |
| "Textures with the format " |
| "DXGI_FORMAT_B5G6R5_UNORM have incorrect data", |
| &members}; |
| |
| // On some Intel drivers, evaluating unary minus operator on integer may get wrong answer in |
| // vertex shaders. To work around this bug, we translate -(int) into ~(int)+1. |
| // This driver bug is fixed in 20.19.15.4624. |
| Feature rewriteUnaryMinusOperator = { |
| "rewrite_unary_minus_operator", FeatureCategory::D3DWorkarounds, |
| "Evaluating unary minus operator on integer may get wrong answer in vertex shaders", |
| &members}; |
| |
| // On some Intel drivers, using isnan() on highp float will get wrong answer. To work around |
| // this bug, we use an expression to emulate function isnan(). |
| // Tracking bug: https://crbug.com/650547 |
| // This driver bug is fixed in 21.20.16.4542. |
| Feature emulateIsnanFloat = {"emulate_isnan_float", FeatureCategory::D3DWorkarounds, |
| "Using isnan() on highp float will get wrong answer", &members, |
| "https://crbug.com/650547"}; |
| |
| // On some Intel drivers, using clear() may not take effect. To work around this bug, we call |
| // clear() twice on these platforms. |
| // Tracking bug: https://crbug.com/655534 |
| Feature callClearTwice = {"call_clear_twice", FeatureCategory::D3DWorkarounds, |
| "Using clear() may not take effect", &members, |
| "https://crbug.com/655534"}; |
| |
| // On some Intel drivers, copying from staging storage to constant buffer storage does not |
| // seem to work. Work around this by keeping system memory storage as a canonical reference |
| // for buffer data. |
| // D3D11-only workaround. See http://crbug.com/593024. |
| Feature useSystemMemoryForConstantBuffers = {"use_system_memory_for_constant_buffers", |
| FeatureCategory::D3DWorkarounds, |
| "Copying from staging storage to constant buffer " |
| "storage does not work", |
| &members, "https://crbug.com/593024"}; |
| |
| // This workaround is for the ANGLE_multiview extension. If enabled the viewport or render |
| // target slice will be selected in the geometry shader stage. The workaround flag is added to |
| // make it possible to select the code path in end2end and performance tests. |
| Feature selectViewInGeometryShader = { |
| "select_view_in_geometry_shader", FeatureCategory::D3DWorkarounds, |
| "The viewport or render target slice will be selected in the geometry shader stage for " |
| "the ANGLE_multiview extension", |
| &members}; |
| |
| // When rendering with no render target on D3D, two bugs lead to incorrect behavior on Intel |
| // drivers < 4815. The rendering samples always pass neglecting discard statements in pixel |
| // shader. |
| // 1. If rendertarget is not set, the pixel shader will be recompiled to drop 'SV_TARGET'. |
| // When using a pixel shader with no 'SV_TARGET' in a draw, the pixels are always generated even |
| // if they should be discard by 'discard' statements. |
| // 2. If ID3D11BlendState.RenderTarget[].RenderTargetWriteMask is 0 and rendertarget is not set, |
| // then rendering samples also pass neglecting discard statements in pixel shader. |
| // So we add a dummy texture as render target in such case. See http://anglebug.com/2152 |
| Feature addDummyTextureNoRenderTarget = { |
| "add_dummy_texture_no_render_target", FeatureCategory::D3DWorkarounds, |
| "On some drivers when rendering with no render target, two bugs lead to incorrect behavior", |
| &members, "http://anglebug.com/2152"}; |
| |
| // Don't use D3D constant register zero when allocating space for uniforms in the vertex shader. |
| // This is targeted to work around a bug in NVIDIA D3D driver version 388.59 where in very |
| // specific cases the driver would not handle constant register zero correctly. |
| Feature skipVSConstantRegisterZero = { |
| "skip_vs_constant_register_zero", FeatureCategory::D3DWorkarounds, |
| "In specific cases the driver doesn't handle constant register zero correctly", &members}; |
| |
| // Forces the value returned from an atomic operations to be always be resolved. This is |
| // targeted to workaround a bug in NVIDIA D3D driver where the return value from |
| // RWByteAddressBuffer.InterlockedAdd does not get resolved when used in the .yzw components of |
| // a RWByteAddressBuffer.Store operation. Only has an effect on HLSL translation. |
| // http://anglebug.com/3246 |
| Feature forceAtomicValueResolution = { |
| "force_atomic_value_resolution", FeatureCategory::D3DWorkarounds, |
| "On some drivers the return value from RWByteAddressBuffer.InterlockedAdd does not resolve " |
| "when used in the .yzw components of a RWByteAddressBuffer.Store operation", |
| &members, "http://anglebug.com/3246"}; |
| |
| // Match chromium's robust resource init behaviour by always prefering to upload texture data |
| // instead of clearing. Clear calls have been observed to cause texture corruption for some |
| // formats. |
| Feature allowClearForRobustResourceInit = { |
| "allow_clear_for_robust_resource_init", FeatureCategory::D3DWorkarounds, |
| "Some drivers corrupt texture data when clearing for robust resource initialization.", |
| &members, "http://crbug.com/941620"}; |
| }; |
| |
| inline FeaturesD3D::FeaturesD3D() = default; |
| inline FeaturesD3D::~FeaturesD3D() = default; |
| |
| } // namespace angle |
| |
| #endif // ANGLE_PLATFORM_FEATURESD3D_H_ |