Ryujinx/moltenvk
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

MVKPhysicalDevice: Correct subgroup properties.

Browse Source 
On systems not supporting this, the subgroup size is set to 1.

Make sure the subgroup size is fixed in the shader, at least until we
implement `VK_EXT_subgroup_size_control`.

According to the Metal feature set tables, SIMD-group reduction is only
supported on Mac family 2 GPUs and Apple family 7 GPUs. Previously, we
were exposing these on all Mac GPUs.

Quadgroup permutation is supported on all Apple GPUs starting from
family 4. We use them for regular group non-uniform ops as well, so
these are considered to have a subgroup size of 4. On Mac, it's a bit
more complicated. The 2.1 tables say that all Mac GPUs support this, but
the 3.0 and 4.0 tables say that only family 2 supports quadgroup ops.
I've allowed quad ops on family 1 for now.

Unfortunately, my testing shows that SIMD-group functions don't work in
fragment shaders on Mojave, so no fragment shader support until Metal 3.

Update SPIRV-Cross to pull in changes needed for all this.
This commit is contained in:
Chip Davis 2020-11-17 14:22:06 -06:00
parent ce85a96d80
commit 8e11c41c40
4 changed files with 62 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ExternalRevisions/SPIRV-Cross_repo_revision
View File

@ -1 +1 @@
b3c59263a0e8f921df15757c7b28407306a16830
3d16060c3243e9f7bfd027de6e2e27c348d1791e

3
MoltenVK/MoltenVK/API/vk_mvk_moltenvk.h
View File

@ -631,6 +631,9 @@ typedef struct {
VkBool32 renderLinearTextures; /**< If true, linear textures are renderable. */
VkBool32 pullModelInterpolation; /**< If true, explicit interpolation functions are supported. */
VkBool32 samplerMirrorClampToEdge; /**< If true, the mirrored clamp to edge address mode is supported in samplers. */
VkBool32 quadPermute; /**< If true, quadgroup permutation functions (vote, ballot, shuffle) are supported in shaders. */
VkBool32 simdPermute; /**< If true, SIMD-group permutation functions (vote, ballot, shuffle) are supported in shaders. */
VkBool32 simdReduction; /**< If true, SIMD-group reduction functions (arithmetic) are supported in shaders. */
} MVKPhysicalDeviceMetalFeatures;
/** MoltenVK performance of a particular type of activity. */

59
MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm
View File

@ -115,7 +115,7 @@ void MVKPhysicalDevice::getFeatures(VkPhysicalDeviceFeatures2* features) {
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_SUBGROUP_EXTENDED_TYPES_FEATURES: {
auto* shaderSGTypesFeatures = (VkPhysicalDeviceShaderSubgroupExtendedTypesFeatures*)next;
shaderSGTypesFeatures->shaderSubgroupExtendedTypes = _metalFeatures.subgroupSize != 0;
shaderSGTypesFeatures->shaderSubgroupExtendedTypes = _metalFeatures.simdPermute || _metalFeatures.quadPermute;
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES: {
@ -317,28 +317,32 @@ void MVKPhysicalDevice::getProperties(VkPhysicalDeviceProperties2* properties) {
pushDescProps->maxPushDescriptors = _properties.limits.maxPerStageResources;
break;
}
#if MVK_MACOS
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES:
if (mvkOSVersionIsAtLeast(10.14)) {
auto* subgroupProps = (VkPhysicalDeviceSubgroupProperties*)next;
subgroupProps->subgroupSize = _metalFeatures.subgroupSize;
subgroupProps->supportedStages =
VK_SHADER_STAGE_COMPUTE_BIT |
VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT |
VK_SHADER_STAGE_FRAGMENT_BIT;
subgroupProps->supportedOperations =
VK_SUBGROUP_FEATURE_BASIC_BIT |
VK_SUBGROUP_FEATURE_VOTE_BIT |
VK_SUBGROUP_FEATURE_ARITHMETIC_BIT |
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES: {
auto* subgroupProps = (VkPhysicalDeviceSubgroupProperties*)next;
subgroupProps->subgroupSize = _metalFeatures.subgroupSize;
subgroupProps->supportedStages = VK_SHADER_STAGE_COMPUTE_BIT;
if (_features.tessellationShader) {
subgroupProps->supportedStages |= VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT;
}
if (mvkOSVersionIsAtLeast(10.15, 13.0)) {
subgroupProps->supportedStages |= VK_SHADER_STAGE_FRAGMENT_BIT;
}
subgroupProps->supportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT;
if (_metalFeatures.simdPermute || _metalFeatures.quadPermute) {
subgroupProps->supportedOperations |= VK_SUBGROUP_FEATURE_VOTE_BIT |
VK_SUBGROUP_FEATURE_BALLOT_BIT |
VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT |
// VK_SUBGROUP_FEATURE_CLUSTERED_BIT |
VK_SUBGROUP_FEATURE_QUAD_BIT;
subgroupProps->quadOperationsInAllStages = true;
VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT;
}
if (_metalFeatures.simdReduction) {
subgroupProps->supportedOperations |= VK_SUBGROUP_FEATURE_ARITHMETIC_BIT;
}
if (_metalFeatures.quadPermute) {
subgroupProps->supportedOperations |= VK_SUBGROUP_FEATURE_QUAD_BIT;
}
subgroupProps->quadOperationsInAllStages = _metalFeatures.quadPermute;
break;
#endif
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_PROPERTIES: {
auto* timelineSem4Props = (VkPhysicalDeviceTimelineSemaphoreProperties*)next;
timelineSem4Props->maxTimelineSemaphoreValueDifference = std::numeric_limits<uint64_t>::max();
@ -1202,6 +1206,7 @@ void MVKPhysicalDevice::initMetalFeatures() {
}
if (supportsMTLGPUFamily(Apple4)) {
_metalFeatures.nativeTextureSwizzle = true;
_metalFeatures.quadPermute = true;
}
}
@ -1291,9 +1296,11 @@ void MVKPhysicalDevice::initMetalFeatures() {
}
if (supportsMTLGPUFamily(Apple4)) {
_metalFeatures.nativeTextureSwizzle = true;
_metalFeatures.quadPermute = true;
}
if (supportsMTLGPUFamily(Apple6) ) {
_metalFeatures.astcHDRTextures = true;
_metalFeatures.simdPermute = true;
}
}
@ -1312,6 +1319,7 @@ void MVKPhysicalDevice::initMetalFeatures() {
_metalFeatures.multisampleLayeredRendering = _metalFeatures.layeredRendering;
_metalFeatures.samplerClampToBorder = true;
_metalFeatures.samplerMirrorClampToEdge = true;
_metalFeatures.simdReduction = true;
}
}
#endif
@ -1359,6 +1367,8 @@ void MVKPhysicalDevice::initMetalFeatures() {
_metalFeatures.events = true;
_metalFeatures.memoryBarriers = true;
_metalFeatures.textureBuffers = true;
_metalFeatures.quadPermute = true;
_metalFeatures.simdPermute = true;
}
if (supportsMTLFeatureSet(macOS_GPUFamily2_v1)) {
@ -1366,6 +1376,7 @@ void MVKPhysicalDevice::initMetalFeatures() {
_metalFeatures.stencilFeedback = true;
_metalFeatures.depthResolve = true;
_metalFeatures.stencilResolve = true;
_metalFeatures.simdReduction = true;
}
if ( mvkOSVersionIsAtLeast(10.15) ) {
@ -1424,12 +1435,20 @@ void MVKPhysicalDevice::initMetalFeatures() {
}
}
_metalFeatures.subgroupSize = 1;
#if MVK_MACOS
if (mvkOSVersionIsAtLeast(10.14)) {
if (_metalFeatures.simdPermute) {
static const uint32_t kAMDVendorId = 0x1002;
_metalFeatures.subgroupSize = (_properties.vendorID == kAMDVendorId) ? 64 : 32;
}
#endif
#if MVK_IOS
if (_metalFeatures.simdPermute) {
_metalFeatures.subgroupSize = 32;
} else if (_metalFeatures.quadPermute) {
_metalFeatures.subgroupSize = 4;
}
#endif
#define setMSLVersion(maj, min) \
_metalFeatures.mslVersion = SPIRV_CROSS_NAMESPACE::CompilerMSL::Options::make_msl_version(maj, min);

19
MoltenVK/MoltenVK/GPUObjects/MVKPipeline.mm
View File

@ -1481,6 +1481,14 @@ void MVKGraphicsPipeline::initMVKShaderConverterContext(SPIRVToMSLConversionConf
shaderContext.options.mslOptions.multiview = mvkRendPass->isMultiview();
shaderContext.options.mslOptions.multiview_layered_rendering = getDevice()->getPhysicalDevice()->canUseInstancingForMultiview();
shaderContext.options.mslOptions.view_index_from_device_index = mvkAreAllFlagsEnabled(pCreateInfo->flags, VK_PIPELINE_CREATE_VIEW_INDEX_FROM_DEVICE_INDEX_BIT);
shaderContext.options.mslOptions.fixed_subgroup_size = _device->_pMetalFeatures->subgroupSize;
#if MVK_MACOS
shaderContext.options.mslOptions.emulate_subgroups = !_device->_pMetalFeatures->simdPermute;
#endif
#if MVK_IOS_OR_TVOS
shaderContext.options.mslOptions.emulate_subgroups = !_device->_pMetalFeatures->quadPermute;
shaderContext.options.mslOptions.ios_use_simdgroup_functions = !!_device->_pMetalFeatures->simdPermute;
#endif
shaderContext.options.tessPatchKind = reflectData.patchKind;
shaderContext.options.numTessControlPoints = reflectData.numControlPoints;
@ -1679,6 +1687,14 @@ MVKMTLFunction MVKComputePipeline::getMTLFunction(const VkComputePipelineCreateI
shaderContext.options.mslOptions.texture_buffer_native = _device->_pMetalFeatures->textureBuffers;
shaderContext.options.mslOptions.dispatch_base = _allowsDispatchBase;
shaderContext.options.mslOptions.texture_1D_as_2D = mvkTreatTexture1DAs2D();
shaderContext.options.mslOptions.fixed_subgroup_size = _device->_pMetalFeatures->subgroupSize;
#if MVK_MACOS
shaderContext.options.mslOptions.emulate_subgroups = !_device->_pMetalFeatures->simdPermute;
#endif
#if MVK_IOS_OR_TVOS
shaderContext.options.mslOptions.emulate_subgroups = !_device->_pMetalFeatures->quadPermute;
shaderContext.options.mslOptions.ios_use_simdgroup_functions = !!_device->_pMetalFeatures->simdPermute;
#endif
MVKPipelineLayout* layout = (MVKPipelineLayout*)pCreateInfo->layout;
layout->populateShaderConverterContext(shaderContext);
@ -1954,6 +1970,7 @@ namespace SPIRV_CROSS_NAMESPACE {
opt.device_index,
opt.enable_frag_output_mask,
opt.additional_fixed_sample_mask,
opt.fixed_subgroup_size,
opt.enable_point_size_builtin,
opt.enable_frag_depth_builtin,
opt.enable_frag_stencil_ref_builtin,
@ -1981,6 +1998,8 @@ namespace SPIRV_CROSS_NAMESPACE {
opt.multi_patch_workgroup,
opt.vertex_for_tessellation,
opt.arrayed_subpass_input,
opt.ios_use_simdgroup_functions,
opt.emulate_subgroups,
opt.vertex_index_type);
}