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
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
|
/*
* Copyright 2015 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "GrVkCaps.h"
#include "GrRenderTargetProxy.h"
#include "GrShaderCaps.h"
#include "GrVkUtil.h"
#include "vk/GrVkBackendContext.h"
#include "vk/GrVkInterface.h"
GrVkCaps::GrVkCaps(const GrContextOptions& contextOptions, const GrVkInterface* vkInterface,
VkPhysicalDevice physDev, uint32_t featureFlags, uint32_t extensionFlags)
: INHERITED(contextOptions) {
fCanUseGLSLForShaderModule = false;
fMustDoCopiesFromOrigin = false;
fSupportsCopiesAsDraws = false;
fMustSubmitCommandsBeforeCopyOp = false;
fMustSleepOnTearDown = false;
fNewSecondaryCBOnPipelineChange = false;
/**************************************************************************
* GrDrawTargetCaps fields
**************************************************************************/
fMipMapSupport = true; // always available in Vulkan
fSRGBSupport = true; // always available in Vulkan
fNPOTTextureTileSupport = true; // always available in Vulkan
fTwoSidedStencilSupport = true; // always available in Vulkan
fStencilWrapOpsSupport = true; // always available in Vulkan
fDiscardRenderTargetSupport = true;
fReuseScratchTextures = true; //TODO: figure this out
fGpuTracingSupport = false; //TODO: figure this out
fCompressedTexSubImageSupport = false; //TODO: figure this out
fOversizedStencilSupport = false; //TODO: figure this out
fUseDrawInsteadOfClear = false;
fFenceSyncSupport = true; // always available in Vulkan
fCrossContextTextureSupport = false; // TODO: Add thread-safe memory pools so we can enable this
fMapBufferFlags = kNone_MapFlags; //TODO: figure this out
fBufferMapThreshold = SK_MaxS32; //TODO: figure this out
fMaxRenderTargetSize = 4096; // minimum required by spec
fMaxTextureSize = 4096; // minimum required by spec
fMaxColorSampleCount = 4; // minimum required by spec
fMaxStencilSampleCount = 4; // minimum required by spec
fShaderCaps.reset(new GrShaderCaps(contextOptions));
this->init(contextOptions, vkInterface, physDev, featureFlags, extensionFlags);
}
bool GrVkCaps::initDescForDstCopy(const GrRenderTargetProxy* src, GrSurfaceDesc* desc,
bool* rectsMustMatch, bool* disallowSubrect) const {
// Vk doesn't use rectsMustMatch or disallowSubrect. Always return false.
*rectsMustMatch = false;
*disallowSubrect = false;
// We can always succeed here with either a CopyImage (none msaa src) or ResolveImage (msaa).
// For CopyImage we can make a simple texture, for ResolveImage we require the dst to be a
// render target as well.
desc->fOrigin = src->origin();
desc->fConfig = src->config();
if (src->numColorSamples() > 1 || (src->asTextureProxy() && this->supportsCopiesAsDraws())) {
desc->fFlags = kRenderTarget_GrSurfaceFlag;
} else {
// Just going to use CopyImage here
desc->fFlags = kNone_GrSurfaceFlags;
}
return true;
}
void GrVkCaps::init(const GrContextOptions& contextOptions, const GrVkInterface* vkInterface,
VkPhysicalDevice physDev, uint32_t featureFlags, uint32_t extensionFlags) {
VkPhysicalDeviceProperties properties;
GR_VK_CALL(vkInterface, GetPhysicalDeviceProperties(physDev, &properties));
VkPhysicalDeviceMemoryProperties memoryProperties;
GR_VK_CALL(vkInterface, GetPhysicalDeviceMemoryProperties(physDev, &memoryProperties));
this->initGrCaps(properties, memoryProperties, featureFlags);
this->initShaderCaps(properties, featureFlags);
this->initConfigTable(vkInterface, physDev);
this->initStencilFormat(vkInterface, physDev);
if (SkToBool(extensionFlags & kNV_glsl_shader_GrVkExtensionFlag)) {
// Currently disabling this feature since it does not play well with validation layers which
// expect a SPIR-V shader
// fCanUseGLSLForShaderModule = true;
}
if (kQualcomm_VkVendor == properties.vendorID) {
fMustDoCopiesFromOrigin = true;
}
if (kNvidia_VkVendor == properties.vendorID) {
fSupportsCopiesAsDraws = true;
fMustSubmitCommandsBeforeCopyOp = true;
}
#if defined(SK_BUILD_FOR_WIN)
if (kNvidia_VkVendor == properties.vendorID) {
fMustSleepOnTearDown = true;
}
#elif defined(SK_BUILD_FOR_ANDROID)
if (kImagination_VkVendor == properties.vendorID) {
fMustSleepOnTearDown = true;
}
#endif
this->applyOptionsOverrides(contextOptions);
fShaderCaps->applyOptionsOverrides(contextOptions);
}
int get_max_sample_count(VkSampleCountFlags flags) {
SkASSERT(flags & VK_SAMPLE_COUNT_1_BIT);
if (!(flags & VK_SAMPLE_COUNT_2_BIT)) {
return 0;
}
if (!(flags & VK_SAMPLE_COUNT_4_BIT)) {
return 2;
}
if (!(flags & VK_SAMPLE_COUNT_8_BIT)) {
return 4;
}
if (!(flags & VK_SAMPLE_COUNT_16_BIT)) {
return 8;
}
if (!(flags & VK_SAMPLE_COUNT_32_BIT)) {
return 16;
}
if (!(flags & VK_SAMPLE_COUNT_64_BIT)) {
return 32;
}
return 64;
}
void GrVkCaps::initSampleCount(const VkPhysicalDeviceProperties& properties) {
VkSampleCountFlags colorSamples = properties.limits.framebufferColorSampleCounts;
VkSampleCountFlags stencilSamples = properties.limits.framebufferStencilSampleCounts;
fMaxColorSampleCount = get_max_sample_count(colorSamples);
if (kImagination_VkVendor == properties.vendorID) {
fMaxColorSampleCount = 0;
}
fMaxStencilSampleCount = get_max_sample_count(stencilSamples);
}
void GrVkCaps::initGrCaps(const VkPhysicalDeviceProperties& properties,
const VkPhysicalDeviceMemoryProperties& memoryProperties,
uint32_t featureFlags) {
// So GPUs, like AMD, are reporting MAX_INT support vertex attributes. In general, there is no
// need for us ever to support that amount, and it makes tests which tests all the vertex
// attribs timeout looping over that many. For now, we'll cap this at 64 max and can raise it if
// we ever find that need.
static const uint32_t kMaxVertexAttributes = 64;
fMaxVertexAttributes = SkTMin(properties.limits.maxVertexInputAttributes, kMaxVertexAttributes);
// AMD advertises support for MAX_UINT vertex input attributes, but in reality only supports 32.
if (kAMD_VkVendor == properties.vendorID) {
fMaxVertexAttributes = SkTMin(fMaxVertexAttributes, 32);
}
// We could actually query and get a max size for each config, however maxImageDimension2D will
// give the minimum max size across all configs. So for simplicity we will use that for now.
fMaxRenderTargetSize = SkTMin(properties.limits.maxImageDimension2D, (uint32_t)INT_MAX);
fMaxTextureSize = SkTMin(properties.limits.maxImageDimension2D, (uint32_t)INT_MAX);
this->initSampleCount(properties);
// Assuming since we will always map in the end to upload the data we might as well just map
// from the get go. There is no hard data to suggest this is faster or slower.
fBufferMapThreshold = 0;
fMapBufferFlags = kCanMap_MapFlag | kSubset_MapFlag;
fStencilWrapOpsSupport = true;
fOversizedStencilSupport = true;
fSampleShadingSupport = SkToBool(featureFlags & kSampleRateShading_GrVkFeatureFlag);
// AMD seems to have issues binding new VkPipelines inside a secondary command buffer.
// Current workaround is to use a different secondary command buffer for each new VkPipeline.
if (kAMD_VkVendor == properties.vendorID) {
fNewSecondaryCBOnPipelineChange = true;
}
}
void GrVkCaps::initShaderCaps(const VkPhysicalDeviceProperties& properties, uint32_t featureFlags) {
GrShaderCaps* shaderCaps = fShaderCaps.get();
shaderCaps->fVersionDeclString = "#version 330\n";
// fConfigOutputSwizzle will default to RGBA so we only need to set it for alpha only config.
for (int i = 0; i < kGrPixelConfigCnt; ++i) {
GrPixelConfig config = static_cast<GrPixelConfig>(i);
if (GrPixelConfigIsAlphaOnly(config)) {
shaderCaps->fConfigTextureSwizzle[i] = GrSwizzle::RRRR();
shaderCaps->fConfigOutputSwizzle[i] = GrSwizzle::AAAA();
} else {
if (kGray_8_GrPixelConfig == config) {
shaderCaps->fConfigTextureSwizzle[i] = GrSwizzle::RRRA();
} else if (kRGBA_4444_GrPixelConfig == config) {
// The vulkan spec does not require R4G4B4A4 to be supported for texturing so we
// store the data in a B4G4R4A4 texture and then swizzle it when doing texture reads
// or writing to outputs. Since we're not actually changing the data at all, the
// only extra work is the swizzle in the shader for all operations.
shaderCaps->fConfigTextureSwizzle[i] = GrSwizzle::BGRA();
shaderCaps->fConfigOutputSwizzle[i] = GrSwizzle::BGRA();
} else {
shaderCaps->fConfigTextureSwizzle[i] = GrSwizzle::RGBA();
}
}
}
if (kImagination_VkVendor == properties.vendorID) {
shaderCaps->fAtan2ImplementedAsAtanYOverX = true;
}
// Vulkan is based off ES 3.0 so the following should all be supported
shaderCaps->fUsesPrecisionModifiers = true;
shaderCaps->fFlatInterpolationSupport = true;
// GrShaderCaps
shaderCaps->fShaderDerivativeSupport = true;
shaderCaps->fGeometryShaderSupport = SkToBool(featureFlags & kGeometryShader_GrVkFeatureFlag);
shaderCaps->fDualSourceBlendingSupport = SkToBool(featureFlags & kDualSrcBlend_GrVkFeatureFlag);
if (kAMD_VkVendor == properties.vendorID) {
// Currently DualSourceBlending is not working on AMD. vkCreateGraphicsPipeline fails when
// using a draw with dual source. Looking into whether it is driver bug or issue with our
// SPIR-V. Bug skia:6405
shaderCaps->fDualSourceBlendingSupport = false;
}
shaderCaps->fIntegerSupport = true;
// Assume the minimum precisions mandated by the SPIR-V spec.
shaderCaps->fShaderPrecisionVaries = true;
for (int s = 0; s < kGrShaderTypeCount; ++s) {
auto& highp = shaderCaps->fFloatPrecisions[s][kHigh_GrSLPrecision];
highp.fLogRangeLow = highp.fLogRangeHigh = 127;
highp.fBits = 23;
auto& mediump = shaderCaps->fFloatPrecisions[s][kMedium_GrSLPrecision];
mediump.fLogRangeLow = mediump.fLogRangeHigh = 14;
mediump.fBits = 10;
shaderCaps->fFloatPrecisions[s][kLow_GrSLPrecision] = mediump;
}
shaderCaps->initSamplerPrecisionTable();
shaderCaps->fMaxVertexSamplers =
shaderCaps->fMaxGeometrySamplers =
shaderCaps->fMaxFragmentSamplers = SkTMin(
SkTMin(properties.limits.maxPerStageDescriptorSampledImages,
properties.limits.maxPerStageDescriptorSamplers),
(uint32_t)INT_MAX);
shaderCaps->fMaxCombinedSamplers = SkTMin(
SkTMin(properties.limits.maxDescriptorSetSampledImages,
properties.limits.maxDescriptorSetSamplers),
(uint32_t)INT_MAX);
}
bool stencil_format_supported(const GrVkInterface* interface,
VkPhysicalDevice physDev,
VkFormat format) {
VkFormatProperties props;
memset(&props, 0, sizeof(VkFormatProperties));
GR_VK_CALL(interface, GetPhysicalDeviceFormatProperties(physDev, format, &props));
return SkToBool(VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT & props.optimalTilingFeatures);
}
void GrVkCaps::initStencilFormat(const GrVkInterface* interface, VkPhysicalDevice physDev) {
// List of legal stencil formats (though perhaps not supported on
// the particular gpu/driver) from most preferred to least. We are guaranteed to have either
// VK_FORMAT_D24_UNORM_S8_UINT or VK_FORMAT_D32_SFLOAT_S8_UINT. VK_FORMAT_D32_SFLOAT_S8_UINT
// can optionally have 24 unused bits at the end so we assume the total bits is 64.
static const StencilFormat
// internal Format stencil bits total bits packed?
gS8 = { VK_FORMAT_S8_UINT, 8, 8, false },
gD24S8 = { VK_FORMAT_D24_UNORM_S8_UINT, 8, 32, true },
gD32S8 = { VK_FORMAT_D32_SFLOAT_S8_UINT, 8, 64, true };
if (stencil_format_supported(interface, physDev, VK_FORMAT_S8_UINT)) {
fPreferedStencilFormat = gS8;
} else if (stencil_format_supported(interface, physDev, VK_FORMAT_D24_UNORM_S8_UINT)) {
fPreferedStencilFormat = gD24S8;
} else {
SkASSERT(stencil_format_supported(interface, physDev, VK_FORMAT_D32_SFLOAT_S8_UINT));
fPreferedStencilFormat = gD32S8;
}
}
void GrVkCaps::initConfigTable(const GrVkInterface* interface, VkPhysicalDevice physDev) {
for (int i = 0; i < kGrPixelConfigCnt; ++i) {
VkFormat format;
if (GrPixelConfigToVkFormat(static_cast<GrPixelConfig>(i), &format)) {
fConfigTable[i].init(interface, physDev, format);
}
}
// We currently do not support compressed textures in Vulkan
const uint16_t kFlagsToRemove = ConfigInfo::kTextureable_Flag|ConfigInfo::kRenderable_Flag;
fConfigTable[kETC1_GrPixelConfig].fOptimalFlags &= ~kFlagsToRemove;
fConfigTable[kETC1_GrPixelConfig].fLinearFlags &= ~kFlagsToRemove;
}
void GrVkCaps::ConfigInfo::InitConfigFlags(VkFormatFeatureFlags vkFlags, uint16_t* flags) {
if (SkToBool(VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT & vkFlags) &&
SkToBool(VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT & vkFlags)) {
*flags = *flags | kTextureable_Flag;
}
if (SkToBool(VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT & vkFlags)) {
*flags = *flags | kRenderable_Flag;
}
if (SkToBool(VK_FORMAT_FEATURE_BLIT_SRC_BIT & vkFlags)) {
*flags = *flags | kBlitSrc_Flag;
}
if (SkToBool(VK_FORMAT_FEATURE_BLIT_DST_BIT & vkFlags)) {
*flags = *flags | kBlitDst_Flag;
}
}
void GrVkCaps::ConfigInfo::init(const GrVkInterface* interface,
VkPhysicalDevice physDev,
VkFormat format) {
VkFormatProperties props;
memset(&props, 0, sizeof(VkFormatProperties));
GR_VK_CALL(interface, GetPhysicalDeviceFormatProperties(physDev, format, &props));
InitConfigFlags(props.linearTilingFeatures, &fLinearFlags);
InitConfigFlags(props.optimalTilingFeatures, &fOptimalFlags);
}
|