1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
|
//
// Copyright (c) 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 NVIDIA D3D11 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};
Feature setDataFasterThanImageUploadOn128bitFormats = {
"set_data_faster_than_image_upload_on_128bit_formats", FeatureCategory::D3DWorkarounds,
"Set data faster than image upload on 128bit formats", &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,
"D3D11 is 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 NVIDIA D3D11 Driver version <=347.88 and >368.81 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,
"NVIDIA 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 NVIDIA 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,
"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",
&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,
"On some AMD drivers, 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,
"On Intel and AMD drivers, 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,
"On some Intel drivers, 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,
"On some Intel drivers, 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,
"On some Intel drivers, using clear() may not take effect", &members,
"https://crbug.com/655534"};
// On Sandybridge, calling ClearView after using dual source blending causes the hardware to hang.
// See: https://bugzilla.mozilla.org/show_bug.cgi?id=1633628
Feature emulateClearViewAfterDualSourceBlending = {"emulate_clear_view_after_dual_source_blending",
FeatureCategory::D3DWorkarounds,
"On Sandybridge, calling ClearView after using dual source blending causes "
"the hardware to hang", &members,
"https://bugzilla.mozilla.org/show_bug.cgi?id=1633628"};
// 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,
"On some Intel drivers, 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 D3D ntel drivers <4815 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,
"On NVIDIA D3D driver v388.59 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 an NVIDIA D3D driver, 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"};
Feature allowES3OnFL10_0 = {
"allowES3OnFL10_0", FeatureCategory::D3DWorkarounds,
"Allow ES3 on 10.0 devices", &members};
};
inline FeaturesD3D::FeaturesD3D() = default;
inline FeaturesD3D::~FeaturesD3D() = default;
} // namespace angle
#endif // ANGLE_PLATFORM_FEATURESD3D_H_
|
