[Fix] (MG_Backend/DirectVulkan): fix Voxy subgroup and indirect draw sync

- Implement Vulkan subgroup capability querying and expose KHR subgroup getter values.

- Fix DirectVulkan memory barriers so GL_COMMAND_BARRIER_BIT makes generated indirect draw commands visible.

- Keep Voxy on the DirectVulkan gpu_shader_int64 quad decode path while filtering unsupported optional int64 usage on backends that do not advertise it.

- Add MG_Test coverage for subgroup getters, Voxy subgroup/int64 shader probes, command barrier mapping, and indirect draw command layout.

- Check for whether driver supports shader subgroup operation, disable on demand, and provide env var `MOBILEGL_DISABLE_SUBGROUP` to explicitly disable subgroup features
This commit is contained in:
2026-06-09 09:46:07 +08:00
parent cf165c0db5
commit dd52f0381a
13 changed files with 595 additions and 44 deletions
@@ -19,6 +19,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
BackendObject_DirectVulkan::~BackendObject_DirectVulkan() = default;
BackendObject_DirectVulkan::BackendObject_DirectVulkan():
m_rendererInfo{
.RendererName = "Magma",
.BackendName = "Direct (Vulkan)",
.ExtraVendor = Nullopt,
.RendererGLInfo =
{
.TargetGLVersion = {3, 3, 0},
.TargetGLSLVersion = {4, 6, 0},
.Extensions = {V_OpenGL30, V_OpenGL31, V_OpenGL32,
V_OpenGL33, E_GL_ARB_draw_buffers_blend, E_GL_ARB_compute_shader,
E_GL_ARB_shader_storage_buffer_object, E_GL_ARB_shader_image_load_store,
E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object,
E_GL_ARB_multi_draw_indirect, E_GL_ARB_indirect_parameters,
E_GL_EXT_framebuffer_object, E_GL_ARB_depth_texture, E_GL_ARB_buffer_storage,
E_GL_ARB_texture_storage, E_GL_ARB_direct_state_access,
E_GL_ARB_shader_draw_parameters, E_GL_ARB_gpu_shader_int64},
.IsCompatibilityProfile = false
},
.StaticBackendCapability = {.AllowVSOnlyPrograms = false}} {}
Bool BackendObject_DirectVulkan::InitWindowSurface() {
if (!m_windowHandle.Handle) {
MGLOG_E("Cannot initialize DirectVulkan window surface: native window handle is null");
@@ -47,8 +68,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return false;
}
MG_Util::BackendLoader::FillInVulkanCapabilities(m_vulkanCaps, pVulkanRenderer->GetPhysicalDevice().properties);
const auto& physicalDevice = pVulkanRenderer->GetPhysicalDevice();
if (!MG_Util::BackendLoader::QueryVulkanCapabilities(m_vulkanCaps, pVulkanRenderer->GetInstance(),
physicalDevice.handle)) {
MGLOG_W("DirectVulkan: failed to query extended Vulkan capabilities, using basic properties");
MG_Util::BackendLoader::FillInVulkanCapabilities(m_vulkanCaps, physicalDevice.properties);
}
UpdateDynamicBackendParameters();
UpdateAdvertisedExtensions();
return true;
}
@@ -113,27 +140,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
const RendererInfo& BackendObject_DirectVulkan::GetRendererInfo() const {
static RendererInfo RendererInfo = {
.RendererName = "Magma", // Renderer Name
.BackendName = "Direct (Vulkan)", // Backend Name
.ExtraVendor = Nullopt, // Extra vendor
.RendererGLInfo =
{
.TargetGLVersion = {3, 3, 0}, // Target OpenGL Version
.TargetGLSLVersion = {4, 6, 0}, // Target Shading Language Version
.Extensions = {V_OpenGL30, V_OpenGL31, V_OpenGL32, // OpenGL Extensions
V_OpenGL33, E_GL_ARB_draw_buffers_blend, E_GL_ARB_compute_shader,
E_GL_ARB_shader_storage_buffer_object, E_GL_ARB_shader_image_load_store,
E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object,
E_GL_ARB_multi_draw_indirect, E_GL_ARB_indirect_parameters,
E_GL_EXT_framebuffer_object, E_GL_ARB_depth_texture, E_GL_ARB_buffer_storage,
E_GL_ARB_texture_storage, E_GL_ARB_direct_state_access,
E_GL_ARB_shader_draw_parameters, E_GL_ARB_gpu_shader_int64},
.IsCompatibilityProfile = false // Is Compatibility Profile
},
.StaticBackendCapability = {.AllowVSOnlyPrograms = false} // Backend Capability
};
return RendererInfo;
return m_rendererInfo;
}
String BackendObject_DirectVulkan::GetBackendAPIVersionString() const {
@@ -214,11 +221,81 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return m_dynamicParameters;
}
void BackendObject_DirectVulkan::ApplyVulkanCapabilitiesForTesting(
const MG_External::VulkanCapabilities& capabilities) {
m_vulkanCaps = capabilities;
UpdateDynamicBackendParameters();
UpdateAdvertisedExtensions();
}
void BackendObject_DirectVulkan::UpdateAdvertisedExtensions() {
auto& extensions = m_rendererInfo.RendererGLInfo.Extensions;
extensions.erase(std::remove(extensions.begin(), extensions.end(), E_GL_KHR_shader_subgroup),
extensions.end());
if (m_vulkanCaps.SupportsShaderSubgroup) {
extensions.push_back(E_GL_KHR_shader_subgroup);
}
}
void BackendObject_DirectVulkan::UpdateDynamicBackendParameters() {
const auto mapShaderStages = [](Uint32 vkStages) {
Uint32 glStages = 0;
if ((vkStages & VK_SHADER_STAGE_VERTEX_BIT) != 0) glStages |= GL_VERTEX_SHADER_BIT;
if ((vkStages & VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT) != 0) glStages |= GL_TESS_CONTROL_SHADER_BIT;
if ((vkStages & VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT) != 0) {
glStages |= GL_TESS_EVALUATION_SHADER_BIT;
}
if ((vkStages & VK_SHADER_STAGE_GEOMETRY_BIT) != 0) glStages |= GL_GEOMETRY_SHADER_BIT;
if ((vkStages & VK_SHADER_STAGE_FRAGMENT_BIT) != 0) glStages |= GL_FRAGMENT_SHADER_BIT;
if ((vkStages & VK_SHADER_STAGE_COMPUTE_BIT) != 0) glStages |= GL_COMPUTE_SHADER_BIT;
return glStages;
};
const auto mapSubgroupFeatures = [](Uint32 vkFeatures) {
Uint32 glFeatures = 0;
if ((vkFeatures & VK_SUBGROUP_FEATURE_BASIC_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_BASIC_BIT_KHR;
}
if ((vkFeatures & VK_SUBGROUP_FEATURE_VOTE_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_VOTE_BIT_KHR;
}
if ((vkFeatures & VK_SUBGROUP_FEATURE_ARITHMETIC_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_ARITHMETIC_BIT_KHR;
}
if ((vkFeatures & VK_SUBGROUP_FEATURE_BALLOT_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_BALLOT_BIT_KHR;
}
if ((vkFeatures & VK_SUBGROUP_FEATURE_SHUFFLE_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_SHUFFLE_BIT_KHR;
}
if ((vkFeatures & VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT_KHR;
}
if ((vkFeatures & VK_SUBGROUP_FEATURE_CLUSTERED_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_CLUSTERED_BIT_KHR;
}
if ((vkFeatures & VK_SUBGROUP_FEATURE_QUAD_BIT) != 0) {
glFeatures |= GL_SUBGROUP_FEATURE_QUAD_BIT_KHR;
}
return glFeatures;
};
static constexpr SizeT kMaxAdvertisedShaderStorageBlockSize = 512ull * 1024ull * 1024ull;
m_dynamicParameters.UniformBufferOffsetAlignment = m_vulkanCaps.UniformBufferOffsetAlignment;
m_dynamicParameters.MaxShaderStorageBlockSize =
std::min(m_vulkanCaps.MaxShaderStorageBlockSize, kMaxAdvertisedShaderStorageBlockSize);
if (m_vulkanCaps.SupportsShaderSubgroup) {
m_dynamicParameters.SubgroupSize = m_vulkanCaps.SubgroupSize;
m_dynamicParameters.SubgroupSupportedStages = mapShaderStages(m_vulkanCaps.SubgroupSupportedStages);
m_dynamicParameters.SubgroupSupportedFeatures = mapSubgroupFeatures(m_vulkanCaps.SubgroupSupportedOperations);
m_dynamicParameters.SubgroupQuadOperationsInAllStages = m_vulkanCaps.SubgroupQuadOperationsInAllStages;
} else {
m_dynamicParameters.SubgroupSize = 0;
m_dynamicParameters.SubgroupSupportedStages = 0;
m_dynamicParameters.SubgroupSupportedFeatures = 0;
m_dynamicParameters.SubgroupQuadOperationsInAllStages = false;
}
if (m_dynamicParameters.MaxShaderStorageBlockSize != m_vulkanCaps.MaxShaderStorageBlockSize) {
MGLOG_I("DirectVulkan: clamped GL_MAX_SHADER_STORAGE_BLOCK_SIZE from %zu to %zu",
m_vulkanCaps.MaxShaderStorageBlockSize,
@@ -14,6 +14,7 @@
namespace MobileGL::MG_Backend::DirectVulkan {
class BackendObject_DirectVulkan : public BackendObject {
public:
BackendObject_DirectVulkan();
~BackendObject_DirectVulkan() override;
void Initialize() override;
@@ -30,12 +31,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const GlobalBackendFunctionsTable& GetBackendFunctions() const override;
const DynamicBackendParameters& GetDynamicParameters() const override;
BackendType GetBackendType() const override;
void ApplyVulkanCapabilitiesForTesting(const MG_External::VulkanCapabilities& capabilities);
private:
void UpdateAdvertisedExtensions();
void UpdateDynamicBackendParameters();
Bool m_initialized = false;
DynamicBackendParameters m_dynamicParameters;
MG_External::VulkanCapabilities m_vulkanCaps;
RendererInfo m_rendererInfo;
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -2920,16 +2920,7 @@ void main() {
vkCmdDispatchIndirect(frame.commandBuffer, slice.buffer, slice.offset + static_cast<VkDeviceSize>(indirect));
}
void VulkanRenderer::MemoryBarrier(GLbitfield barriers) {
auto& frame = m_frameContext.GetCurrent();
if (!frame.isCommandRecording) {
m_frameContext.BeginCommandRecording();
m_uniformManager->BeginFrame(m_frameContext.GetCurrentFrameIndex());
}
if (VkRenderPassManager::GetActiveRenderPass() != nullptr) {
VkRenderPassManager::EndRenderPass(frame.commandBuffer);
}
VkMemoryBarrier VulkanRenderer::BuildMemoryBarrierForGlBarriers(GLbitfield barriers) {
VkMemoryBarrier memoryBarrier{};
memoryBarrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER;
memoryBarrier.srcAccessMask =
@@ -2945,6 +2936,24 @@ void main() {
VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT |
VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
if ((barriers & GL_COMMAND_BARRIER_BIT) != 0) {
memoryBarrier.dstAccessMask |= VK_ACCESS_INDIRECT_COMMAND_READ_BIT;
}
return memoryBarrier;
}
void VulkanRenderer::MemoryBarrier(GLbitfield barriers) {
auto& frame = m_frameContext.GetCurrent();
if (!frame.isCommandRecording) {
m_frameContext.BeginCommandRecording();
m_uniformManager->BeginFrame(m_frameContext.GetCurrentFrameIndex());
}
if (VkRenderPassManager::GetActiveRenderPass() != nullptr) {
VkRenderPassManager::EndRenderPass(frame.commandBuffer);
}
VkMemoryBarrier memoryBarrier = BuildMemoryBarrierForGlBarriers(barriers);
MGLOG_D("DirectVulkan: glMemoryBarrier(0x%x)", static_cast<Uint32>(barriers));
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
@@ -5331,6 +5340,10 @@ void main() {
return m_physicalDevice;
}
VkInstance VulkanRenderer::GetInstance() const {
return m_instance;
}
Bool VulkanRenderer::IsDrawIndirectCountExtensionEnabled() const {
return m_drawIndirectCountExtensionEnabled;
}
@@ -143,6 +143,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void DispatchCompute(GLuint numGroupsX, GLuint numGroupsY, GLuint numGroupsZ);
void DispatchComputeIndirect(GLintptr indirect);
void MemoryBarrier(GLbitfield barriers);
static VkMemoryBarrier BuildMemoryBarrierForGlBarriers(GLbitfield barriers);
void DrawArrays(const DrawCmd& payload);
void DrawElements(const DrawIndexedCmd& payload);
void MultiDrawElements(const MultiDrawIndexedCmd& payloads);
@@ -151,6 +152,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void Present();
const PhysicalDevice& GetPhysicalDevice() const;
VkInstance GetInstance() const;
Bool IsDrawIndirectCountExtensionEnabled() const;
void RecreateSwapchain();