diff --git a/src/Vulkan/BUILD.gn b/src/Vulkan/BUILD.gn
index 8ba0b9b..1612e1a 100644
--- a/src/Vulkan/BUILD.gn
+++ b/src/Vulkan/BUILD.gn
@@ -94,6 +94,7 @@
     "VkShaderModule.hpp",
     "VkSpecializationInfo.hpp",
     "VkStringify.hpp",
+    "VkTimelineSemaphore.hpp",
     "VulkanPlatform.hpp",
   ]
   if (is_linux || is_chromeos || is_android) {
@@ -143,6 +144,7 @@
     "VkShaderModule.cpp",
     "VkSpecializationInfo.cpp",
     "VkStringify.cpp",
+    "VkTimelineSemaphore.cpp",
     "libVulkan.cpp",
     "main.cpp",
     "resource.h",
diff --git a/src/Vulkan/CMakeLists.txt b/src/Vulkan/CMakeLists.txt
index 1eb2fd4..2c1ce24 100644
--- a/src/Vulkan/CMakeLists.txt
+++ b/src/Vulkan/CMakeLists.txt
@@ -91,6 +91,8 @@
     VkShaderModule.hpp
     VkStringify.cpp
     VkStringify.hpp
+    VkTimelineSemaphore.cpp
+    VkTimelineSemaphore.hpp
     VulkanPlatform.hpp
 )
 
diff --git a/src/Vulkan/VkDevice.cpp b/src/Vulkan/VkDevice.cpp
index 7b65016..e420921 100644
--- a/src/Vulkan/VkDevice.cpp
+++ b/src/Vulkan/VkDevice.cpp
@@ -18,6 +18,8 @@
 #include "VkDescriptorSetLayout.hpp"
 #include "VkFence.hpp"
 #include "VkQueue.hpp"
+#include "VkSemaphore.hpp"
+#include "VkTimelineSemaphore.hpp"
 #include "Debug/Context.hpp"
 #include "Debug/Server.hpp"
 #include "Device/Blitter.hpp"
@@ -29,11 +31,21 @@
 
 namespace {
 
-std::chrono::time_point<std::chrono::system_clock, std::chrono::nanoseconds> now()
+using time_point = std::chrono::time_point<std::chrono::system_clock, std::chrono::nanoseconds>;
+
+time_point now()
 {
 	return std::chrono::time_point_cast<std::chrono::nanoseconds>(std::chrono::system_clock::now());
 }
 
+const time_point getEndTimePoint(uint64_t timeout, bool &infiniteTimeout)
+{
+	const time_point start = now();
+	const uint64_t max_timeout = (LLONG_MAX - start.time_since_epoch().count());
+	infiniteTimeout = (timeout > max_timeout);
+	return start + std::chrono::nanoseconds(std::min(max_timeout, timeout));
+}
+
 }  // anonymous namespace
 
 namespace vk {
@@ -206,11 +218,8 @@
 
 VkResult Device::waitForFences(uint32_t fenceCount, const VkFence *pFences, VkBool32 waitAll, uint64_t timeout)
 {
-	using time_point = std::chrono::time_point<std::chrono::system_clock, std::chrono::nanoseconds>;
-	const time_point start = now();
-	const uint64_t max_timeout = (LLONG_MAX - start.time_since_epoch().count());
-	bool infiniteTimeout = (timeout > max_timeout);
-	const time_point end_ns = start + std::chrono::nanoseconds(std::min(max_timeout, timeout));
+	bool infiniteTimeout = false;
+	const time_point end_ns = getEndTimePoint(timeout, infiniteTimeout);
 
 	if(waitAll != VK_FALSE)  // All fences must be signaled
 	{
@@ -267,6 +276,63 @@
 	}
 }
 
+VkResult Device::waitForSemaphores(const VkSemaphoreWaitInfo *pWaitInfo, uint64_t timeout)
+{
+	bool infiniteTimeout = false;
+	const time_point end_ns = getEndTimePoint(timeout, infiniteTimeout);
+
+	if(pWaitInfo->flags & VK_SEMAPHORE_WAIT_ANY_BIT)
+	{
+		TimelineSemaphore any = TimelineSemaphore();
+
+		for(uint32_t i = 0; i < pWaitInfo->semaphoreCount; i++)
+		{
+			TimelineSemaphore *semaphore = DynamicCast<TimelineSemaphore>(pWaitInfo->pSemaphores[i]);
+			uint64_t waitValue = pWaitInfo->pValues[i];
+
+			if(semaphore->getCounterValue() == waitValue)
+			{
+				return VK_SUCCESS;
+			}
+
+			semaphore->addDependent(any, waitValue);
+		}
+
+		if(infiniteTimeout)
+		{
+			any.wait(1ull);
+			return VK_SUCCESS;
+		}
+		else
+		{
+			if(any.wait(1, end_ns) == VK_SUCCESS)
+			{
+				return VK_SUCCESS;
+			}
+		}
+
+		return VK_TIMEOUT;
+	}
+	else
+	{
+		ASSERT(pWaitInfo->flags == 0);
+		for(uint32_t i = 0; i < pWaitInfo->semaphoreCount; i++)
+		{
+			TimelineSemaphore *semaphore = DynamicCast<TimelineSemaphore>(pWaitInfo->pSemaphores[i]);
+			uint64_t value = pWaitInfo->pValues[i];
+			if(infiniteTimeout)
+			{
+				semaphore->wait(value);
+			}
+			else if(semaphore->wait(pWaitInfo->pValues[i], end_ns) != VK_SUCCESS)
+			{
+				return VK_TIMEOUT;
+			}
+		}
+		return VK_SUCCESS;
+	}
+}
+
 VkResult Device::waitIdle()
 {
 	for(uint32_t i = 0; i < queueCount; i++)
diff --git a/src/Vulkan/VkDevice.hpp b/src/Vulkan/VkDevice.hpp
index 0129657..fdcce63 100644
--- a/src/Vulkan/VkDevice.hpp
+++ b/src/Vulkan/VkDevice.hpp
@@ -58,6 +58,7 @@
 	bool hasExtension(const char *extensionName) const;
 	VkQueue getQueue(uint32_t queueFamilyIndex, uint32_t queueIndex) const;
 	VkResult waitForFences(uint32_t fenceCount, const VkFence *pFences, VkBool32 waitAll, uint64_t timeout);
+	VkResult waitForSemaphores(const VkSemaphoreWaitInfo *pWaitInfo, uint64_t timeout);
 	VkResult waitIdle();
 	void getDescriptorSetLayoutSupport(const VkDescriptorSetLayoutCreateInfo *pCreateInfo,
 	                                   VkDescriptorSetLayoutSupport *pSupport) const;
diff --git a/src/Vulkan/VkGetProcAddress.cpp b/src/Vulkan/VkGetProcAddress.cpp
index 0edcfcf..1ed42cb 100644
--- a/src/Vulkan/VkGetProcAddress.cpp
+++ b/src/Vulkan/VkGetProcAddress.cpp
@@ -315,6 +315,10 @@
 	MAKE_VULKAN_DEVICE_ENTRY(vkCmdEndRenderPass2),
 	MAKE_VULKAN_DEVICE_ENTRY(vkCmdNextSubpass2),
 	MAKE_VULKAN_DEVICE_ENTRY(vkResetQueryPool),
+	// VK_KHR_timeline_semaphore
+	MAKE_VULKAN_DEVICE_ENTRY(vkGetSemaphoreCounterValue),
+	MAKE_VULKAN_DEVICE_ENTRY(vkSignalSemaphore),
+	MAKE_VULKAN_DEVICE_ENTRY(vkWaitSemaphores),
 };
 
 static const std::vector<std::pair<const char *, std::unordered_map<std::string, PFN_vkVoidFunction>>> deviceExtensionFunctionPointers = {
@@ -377,6 +381,14 @@
 	        MAKE_VULKAN_DEVICE_ENTRY(vkCmdNextSubpass2KHR),
 	        MAKE_VULKAN_DEVICE_ENTRY(vkCmdEndRenderPass2KHR),
 	    } },
+	// VK_KHR_timeline_semaphore
+	{
+	    VK_KHR_TIMELINE_SEMAPHORE_EXTENSION_NAME,
+	    {
+	        MAKE_VULKAN_DEVICE_ENTRY(vkGetSemaphoreCounterValueKHR),
+	        MAKE_VULKAN_DEVICE_ENTRY(vkSignalSemaphoreKHR),
+	        MAKE_VULKAN_DEVICE_ENTRY(vkWaitSemaphoresKHR),
+	    } },
 	// VK_EXT_line_rasterization
 	{
 	    VK_EXT_LINE_RASTERIZATION_EXTENSION_NAME,
diff --git a/src/Vulkan/VkPhysicalDevice.cpp b/src/Vulkan/VkPhysicalDevice.cpp
index 581157a..b9e3e54 100644
--- a/src/Vulkan/VkPhysicalDevice.cpp
+++ b/src/Vulkan/VkPhysicalDevice.cpp
@@ -276,6 +276,12 @@
 }
 
 template<typename T>
+static void getPhysicalDeviceTimelineSemaphoreFeatures(T *features)
+{
+	features->timelineSemaphore = VK_TRUE;
+}
+
+template<typename T>
 static void getPhysicalDeviceVulkan12Features(T *features)
 {
 	features->samplerMirrorClampToEdge = VK_FALSE;
@@ -294,7 +300,7 @@
 	getPhysicalDeviceShaderSubgroupExtendedTypesFeatures(features);
 	getPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR(features);
 	getPhysicalDeviceHostQueryResetFeatures(features);
-	features->timelineSemaphore = VK_FALSE;
+	getPhysicalDeviceTimelineSemaphoreFeatures(features);
 	features->bufferDeviceAddress = VK_FALSE;
 	features->bufferDeviceAddressCaptureReplay = VK_FALSE;
 	features->bufferDeviceAddressMultiDevice = VK_FALSE;
@@ -375,6 +381,9 @@
 			case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_MEMORY_MODEL_FEATURES:
 				getPhysicalDeviceVulkanMemoryModelFeatures(reinterpret_cast<VkPhysicalDeviceVulkanMemoryModelFeatures *>(curExtension));
 				break;
+			case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES:
+				getPhysicalDeviceTimelineSemaphoreFeatures(reinterpret_cast<VkPhysicalDeviceTimelineSemaphoreFeatures *>(curExtension));
+				break;
 			default:
 				LOG_TRAP("curExtension->pNext->sType = %s", vk::Stringify(curExtension->sType).c_str());
 				break;
@@ -787,6 +796,30 @@
 
 void PhysicalDevice::getProperties(const VkPhysicalDeviceExternalSemaphoreInfo *pExternalSemaphoreInfo, VkExternalSemaphoreProperties *pExternalSemaphoreProperties) const
 {
+	for(const auto *nextInfo = reinterpret_cast<const VkBaseInStructure *>(pExternalSemaphoreInfo->pNext);
+	    nextInfo != nullptr; nextInfo = nextInfo->pNext)
+	{
+		switch(nextInfo->sType)
+		{
+			case VK_STRUCTURE_TYPE_SEMAPHORE_TYPE_CREATE_INFO:
+			{
+				const auto *tlsInfo = reinterpret_cast<const VkSemaphoreTypeCreateInfo *>(nextInfo);
+				// Timeline Semaphore does not support external semaphore
+				if(tlsInfo->semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE)
+				{
+					pExternalSemaphoreProperties->compatibleHandleTypes = 0;
+					pExternalSemaphoreProperties->exportFromImportedHandleTypes = 0;
+					pExternalSemaphoreProperties->externalSemaphoreFeatures = 0;
+					return;
+				}
+			}
+			break;
+			default:
+				WARN("nextInfo->sType = %s", vk::Stringify(nextInfo->sType).c_str());
+				break;
+		}
+	}
+
 #if SWIFTSHADER_EXTERNAL_SEMAPHORE_OPAQUE_FD
 	if(pExternalSemaphoreInfo->handleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT)
 	{
@@ -939,7 +972,8 @@
 template<typename T>
 static void getTimelineSemaphoreProperties(T *properties)
 {
-	properties->maxTimelineSemaphoreValueDifference = 0x7FFFFFFFull;
+	// Our implementation of Timeline Semaphores allows the timeline to advance to any value from any value.
+	properties->maxTimelineSemaphoreValueDifference = (uint64_t)-1;
 }
 
 void PhysicalDevice::getProperties(VkPhysicalDeviceTimelineSemaphoreProperties *properties) const
diff --git a/src/Vulkan/VkPromotedExtensions.cpp b/src/Vulkan/VkPromotedExtensions.cpp
index d406a80..f150221 100644
--- a/src/Vulkan/VkPromotedExtensions.cpp
+++ b/src/Vulkan/VkPromotedExtensions.cpp
@@ -215,4 +215,20 @@
 {
 	vkResetQueryPool(device, queryPool, firstQuery, queryCount);
 }
+
+// VK_KHR_timeline_semaphore
+VKAPI_ATTR VkResult VKAPI_CALL vkGetSemaphoreCounterValueKHR(VkDevice device, VkSemaphore semaphore, uint64_t *pValue)
+{
+	return vkGetSemaphoreCounterValue(device, semaphore, pValue);
+}
+
+VKAPI_ATTR VkResult VKAPI_CALL vkSignalSemaphoreKHR(VkDevice device, const VkSemaphoreSignalInfo *pSignalInfo)
+{
+	return vkSignalSemaphore(device, pSignalInfo);
+}
+
+VKAPI_ATTR VkResult VKAPI_CALL vkWaitSemaphoresKHR(VkDevice device, const VkSemaphoreWaitInfo *pWaitInfo, uint64_t timeout)
+{
+	return vkWaitSemaphores(device, pWaitInfo, timeout);
+}
 }
diff --git a/src/Vulkan/VkQueue.cpp b/src/Vulkan/VkQueue.cpp
index fd8bc6d..7995eff 100644
--- a/src/Vulkan/VkQueue.cpp
+++ b/src/Vulkan/VkQueue.cpp
@@ -16,6 +16,8 @@
 #include "VkCommandBuffer.hpp"
 #include "VkFence.hpp"
 #include "VkSemaphore.hpp"
+#include "VkStringify.hpp"
+#include "VkTimelineSemaphore.hpp"
 #include "Device/Renderer.hpp"
 #include "WSI/VkSwapchainKHR.hpp"
 
@@ -38,6 +40,25 @@
 		totalSize += pSubmits[i].waitSemaphoreCount * sizeof(VkPipelineStageFlags);
 		totalSize += pSubmits[i].signalSemaphoreCount * sizeof(VkSemaphore);
 		totalSize += pSubmits[i].commandBufferCount * sizeof(VkCommandBuffer);
+
+		for(const auto *extension = reinterpret_cast<const VkBaseInStructure *>(pSubmits[i].pNext);
+		    extension != nullptr; extension = reinterpret_cast<const VkBaseInStructure *>(extension->pNext))
+		{
+			switch(extension->sType)
+			{
+				case VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO:
+				{
+					const auto *tlsSubmitInfo = reinterpret_cast<const VkTimelineSemaphoreSubmitInfo *>(extension);
+					totalSize += sizeof(VkTimelineSemaphoreSubmitInfo);
+					totalSize += tlsSubmitInfo->waitSemaphoreValueCount * sizeof(uint64_t);
+					totalSize += tlsSubmitInfo->signalSemaphoreValueCount * sizeof(uint64_t);
+				}
+				break;
+				default:
+					WARN("submitInfo[%d]->pNext sType: %s", i, vk::Stringify(extension->sType).c_str());
+					break;
+			}
+		}
 	}
 
 	uint8_t *mem = static_cast<uint8_t *>(
@@ -68,6 +89,41 @@
 		submits[i].pCommandBuffers = reinterpret_cast<const VkCommandBuffer *>(mem);
 		memcpy(mem, pSubmits[i].pCommandBuffers, size);
 		mem += size;
+
+		for(const auto *extension = reinterpret_cast<const VkBaseInStructure *>(pSubmits[i].pNext);
+		    extension != nullptr; extension = reinterpret_cast<const VkBaseInStructure *>(extension->pNext))
+		{
+			switch(extension->sType)
+			{
+				case VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO:
+				{
+					const VkTimelineSemaphoreSubmitInfo *tlsSubmitInfo = reinterpret_cast<const VkTimelineSemaphoreSubmitInfo *>(extension);
+
+					size = sizeof(VkTimelineSemaphoreSubmitInfo);
+					VkTimelineSemaphoreSubmitInfo *tlsSubmitInfoCopy = reinterpret_cast<VkTimelineSemaphoreSubmitInfo *>(mem);
+					memcpy(mem, extension, size);
+					// Don't copy the pNext pointer at all.
+					tlsSubmitInfoCopy->pNext = nullptr;
+					mem += size;
+
+					size = tlsSubmitInfo->waitSemaphoreValueCount * sizeof(uint64_t);
+					tlsSubmitInfoCopy->pWaitSemaphoreValues = reinterpret_cast<uint64_t *>(mem);
+					memcpy(mem, tlsSubmitInfo->pWaitSemaphoreValues, size);
+					mem += size;
+
+					size = tlsSubmitInfo->signalSemaphoreValueCount * sizeof(uint64_t);
+					tlsSubmitInfoCopy->pSignalSemaphoreValues = reinterpret_cast<uint64_t *>(mem);
+					memcpy(mem, tlsSubmitInfo->pSignalSemaphoreValues, size);
+					mem += size;
+
+					submits[i].pNext = tlsSubmitInfoCopy;
+				}
+				break;
+				default:
+					WARN("submitInfo[%d]->pNext sType: %s", i, vk::Stringify(extension->sType).c_str());
+					break;
+			}
+		}
 	}
 
 	return submits;
@@ -122,10 +178,38 @@
 
 	for(uint32_t i = 0; i < task.submitCount; i++)
 	{
-		auto &submitInfo = task.pSubmits[i];
+		VkSubmitInfo &submitInfo = task.pSubmits[i];
+		const VkTimelineSemaphoreSubmitInfo *timelineInfo = nullptr;
+		for(const auto *nextInfo = reinterpret_cast<const VkBaseInStructure *>(submitInfo.pNext);
+		    nextInfo != nullptr; nextInfo = nextInfo->pNext)
+		{
+			switch(nextInfo->sType)
+			{
+				case VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO:
+					timelineInfo = reinterpret_cast<const VkTimelineSemaphoreSubmitInfo *>(submitInfo.pNext);
+					break;
+				default:
+					WARN("submitInfo.pNext->sType = %s", vk::Stringify(nextInfo->sType).c_str());
+					break;
+			}
+		}
+
 		for(uint32_t j = 0; j < submitInfo.waitSemaphoreCount; j++)
 		{
-			vk::Cast(submitInfo.pWaitSemaphores[j])->wait(submitInfo.pWaitDstStageMask[j]);
+			if(auto *sem = DynamicCast<TimelineSemaphore>(submitInfo.pWaitSemaphores[j]))
+			{
+				ASSERT_MSG(timelineInfo != nullptr,
+				           "the pNext chain must include a VkTimelineSemaphoreSubmitInfo if timeline semaphores are used");
+				sem->wait(timelineInfo->pWaitSemaphoreValues[j]);
+			}
+			else if(auto *sem = DynamicCast<BinarySemaphore>(submitInfo.pWaitSemaphores[j]))
+			{
+				sem->wait(submitInfo.pWaitDstStageMask[j]);
+			}
+			else
+			{
+				UNSUPPORTED("Unknown semaphore type");
+			}
 		}
 
 		{
@@ -134,13 +218,26 @@
 			executionState.events = task.events.get();
 			for(uint32_t j = 0; j < submitInfo.commandBufferCount; j++)
 			{
-				vk::Cast(submitInfo.pCommandBuffers[j])->submit(executionState);
+				Cast(submitInfo.pCommandBuffers[j])->submit(executionState);
 			}
 		}
 
 		for(uint32_t j = 0; j < submitInfo.signalSemaphoreCount; j++)
 		{
-			vk::Cast(submitInfo.pSignalSemaphores[j])->signal();
+			if(auto *sem = DynamicCast<TimelineSemaphore>(submitInfo.pSignalSemaphores[j]))
+			{
+				ASSERT_MSG(timelineInfo != nullptr,
+				           "the pNext chain must include a VkTimelineSemaphoreSubmitInfo if timeline semaphores are used");
+				sem->signal(timelineInfo->pSignalSemaphoreValues[j]);
+			}
+			else if(auto *sem = DynamicCast<BinarySemaphore>(submitInfo.pSignalSemaphores[j]))
+			{
+				sem->signal();
+			}
+			else
+			{
+				UNSUPPORTED("Unknown semaphore type");
+			}
 		}
 	}
 
@@ -220,7 +317,7 @@
 
 	for(uint32_t i = 0; i < presentInfo->waitSemaphoreCount; i++)
 	{
-		vk::Cast(presentInfo->pWaitSemaphores[i])->wait();
+		vk::DynamicCast<BinarySemaphore>(presentInfo->pWaitSemaphores[i])->wait();
 	}
 
 	VkResult commandResult = VK_SUCCESS;
diff --git a/src/Vulkan/VkSemaphore.cpp b/src/Vulkan/VkSemaphore.cpp
index aa9a7fc..c82543a 100644
--- a/src/Vulkan/VkSemaphore.cpp
+++ b/src/Vulkan/VkSemaphore.cpp
@@ -16,10 +16,13 @@
 
 #include "VkConfig.hpp"
 #include "VkStringify.hpp"
+#include "VkTimelineSemaphore.hpp"
 
 #include "marl/blockingcall.h"
 #include "marl/conditionvariable.h"
 
+#include <chrono>
+#include <climits>
 #include <functional>
 #include <memory>
 #include <utility>
@@ -28,7 +31,7 @@
 
 // This is a base abstract class for all external semaphore implementations
 // used in this source file.
-class Semaphore::External
+class BinarySemaphore::External
 {
 public:
 	virtual ~External() = default;
@@ -90,45 +93,53 @@
 #endif
     0;
 
-namespace {
-
-struct SemaphoreCreateInfo
+// Create a new instance. The external instance will be allocated only
+// the pCreateInfo->pNext chain indicates it needs to be exported.
+SemaphoreCreateInfo::SemaphoreCreateInfo(const VkSemaphoreCreateInfo *pCreateInfo)
 {
-	bool exportSemaphore = false;
-	VkExternalSemaphoreHandleTypeFlags exportHandleTypes = 0;
-
-	// Create a new instance. The external instance will be allocated only
-	// the pCreateInfo->pNext chain indicates it needs to be exported.
-	SemaphoreCreateInfo(const VkSemaphoreCreateInfo *pCreateInfo)
+	for(const auto *nextInfo = reinterpret_cast<const VkBaseInStructure *>(pCreateInfo->pNext);
+	    nextInfo != nullptr; nextInfo = nextInfo->pNext)
 	{
-		for(const auto *nextInfo = reinterpret_cast<const VkBaseInStructure *>(pCreateInfo->pNext);
-		    nextInfo != nullptr; nextInfo = nextInfo->pNext)
+		switch(nextInfo->sType)
 		{
-			switch(nextInfo->sType)
+			case VK_STRUCTURE_TYPE_EXPORT_SEMAPHORE_CREATE_INFO:
 			{
-				case VK_STRUCTURE_TYPE_EXPORT_SEMAPHORE_CREATE_INFO:
+				const auto *exportInfo = reinterpret_cast<const VkExportSemaphoreCreateInfo *>(nextInfo);
+				exportSemaphore = true;
+				exportHandleTypes = exportInfo->handleTypes;
+				if((exportHandleTypes & ~kSupportedTypes) != 0)
 				{
-					const auto *exportInfo = reinterpret_cast<const VkExportSemaphoreCreateInfo *>(nextInfo);
-					exportSemaphore = true;
-					exportHandleTypes = exportInfo->handleTypes;
-					if((exportHandleTypes & ~kSupportedTypes) != 0)
-					{
-						UNSUPPORTED("exportInfo->handleTypes 0x%X (supports 0x%X)",
-						            int(exportHandleTypes),
-						            int(kSupportedTypes));
-					}
+					UNSUPPORTED("exportInfo->handleTypes 0x%X (supports 0x%X)",
+					            int(exportHandleTypes),
+					            int(kSupportedTypes));
 				}
-				break;
-				default:
-					WARN("nextInfo->sType = %s", vk::Stringify(nextInfo->sType).c_str());
 			}
+			break;
+			case VK_STRUCTURE_TYPE_SEMAPHORE_TYPE_CREATE_INFO:
+			{
+				const auto *tlsInfo = reinterpret_cast<const VkSemaphoreTypeCreateInfo *>(nextInfo);
+				semaphoreType = tlsInfo->semaphoreType;
+				initialPayload = tlsInfo->initialValue;
+			}
+			break;
+			default:
+				WARN("nextInfo->sType = %s", vk::Stringify(nextInfo->sType).c_str());
+				break;
 		}
 	}
-};
+}
 
-}  // namespace
+Semaphore::Semaphore(VkSemaphoreType type)
+    : type(type)
+{
+}
 
-void Semaphore::wait()
+VkSemaphoreType Semaphore::getSemaphoreType() const
+{
+	return type;
+}
+
+void BinarySemaphore::wait()
 {
 	marl::lock lock(mutex);
 	External *ext = tempExternal ? tempExternal : external;
@@ -149,7 +160,7 @@
 		}
 
 		// If the import was temporary, reset the semaphore to its previous state.
-		// See "6.4.5. Importing Semaphore Payloads" in Vulkan 1.1 spec.
+		// See "6.4.5. Importing BinarySemaphore Payloads" in Vulkan 1.1 spec.
 		if(ext == tempExternal)
 		{
 			tempExternal = ext->previous;
@@ -162,8 +173,9 @@
 	}
 }
 
-void Semaphore::signal()
+void BinarySemaphore::signal()
 {
+	ASSERT(type == VK_SEMAPHORE_TYPE_BINARY);
 	marl::lock lock(mutex);
 	External *ext = tempExternal ? tempExternal : external;
 	if(ext)
@@ -178,14 +190,17 @@
 	}
 }
 
-Semaphore::Semaphore(const VkSemaphoreCreateInfo *pCreateInfo, void *mem, const VkAllocationCallbacks *pAllocator)
-    : allocator(pAllocator)
+BinarySemaphore::BinarySemaphore(const VkSemaphoreCreateInfo *pCreateInfo, void *mem, const VkAllocationCallbacks *pAllocator)
+    : Semaphore(VK_SEMAPHORE_TYPE_BINARY)
+    , allocator(pAllocator)
 {
 	SemaphoreCreateInfo info(pCreateInfo);
 	exportableHandleTypes = info.exportHandleTypes;
+	ASSERT(info.semaphoreType == VK_SEMAPHORE_TYPE_BINARY);
+	type = info.semaphoreType;
 }
 
-void Semaphore::destroy(const VkAllocationCallbacks *pAllocator)
+void BinarySemaphore::destroy(const VkAllocationCallbacks *pAllocator)
 {
 	marl::lock lock(mutex);
 	while(tempExternal)
@@ -201,31 +216,31 @@
 	}
 }
 
-size_t Semaphore::ComputeRequiredAllocationSize(const VkSemaphoreCreateInfo *pCreateInfo)
+size_t BinarySemaphore::ComputeRequiredAllocationSize(const VkSemaphoreCreateInfo *pCreateInfo)
 {
 	// Semaphore::External instance is created and destroyed on demand so return 0 here.
 	return 0;
 }
 
 template<class EXTERNAL>
-Semaphore::External *Semaphore::allocateExternal()
+BinarySemaphore::External *BinarySemaphore::allocateExternal()
 {
-	auto *ext = reinterpret_cast<Semaphore::External *>(
+	auto *ext = reinterpret_cast<BinarySemaphore::External *>(
 	    vk::allocate(sizeof(EXTERNAL), alignof(EXTERNAL), allocator));
 	new(ext) EXTERNAL();
 	return ext;
 }
 
-void Semaphore::deallocateExternal(Semaphore::External *ext)
+void BinarySemaphore::deallocateExternal(BinarySemaphore::External *ext)
 {
 	ext->~External();
 	vk::deallocate(ext, allocator);
 }
 
 template<typename ALLOC_FUNC, typename IMPORT_FUNC>
-VkResult Semaphore::importPayload(bool temporaryImport,
-                                  ALLOC_FUNC alloc_func,
-                                  IMPORT_FUNC import_func)
+VkResult BinarySemaphore::importPayload(bool temporaryImport,
+                                        ALLOC_FUNC alloc_func,
+                                        IMPORT_FUNC import_func)
 {
 	marl::lock lock(mutex);
 
@@ -258,7 +273,7 @@
 }
 
 template<typename ALLOC_FUNC, typename EXPORT_FUNC>
-VkResult Semaphore::exportPayload(ALLOC_FUNC alloc_func, EXPORT_FUNC export_func)
+VkResult BinarySemaphore::exportPayload(ALLOC_FUNC alloc_func, EXPORT_FUNC export_func)
 {
 	marl::lock lock(mutex);
 	// Sanity check, do not try to export a semaphore that has a temporary import.
@@ -283,7 +298,7 @@
 }
 
 #if SWIFTSHADER_EXTERNAL_SEMAPHORE_OPAQUE_FD
-VkResult Semaphore::importFd(int fd, bool temporaryImport)
+VkResult BinarySemaphore::importFd(int fd, bool temporaryImport)
 {
 	return importPayload(
 	    temporaryImport,
@@ -295,7 +310,7 @@
 	    });
 }
 
-VkResult Semaphore::exportFd(int *pFd)
+VkResult BinarySemaphore::exportFd(int *pFd)
 {
 	if((exportableHandleTypes & VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT) == 0)
 	{
@@ -314,7 +329,7 @@
 #endif  // SWIFTSHADER_EXTERNAL_SEMAPHORE_OPAQUE_FD
 
 #if VK_USE_PLATFORM_FUCHSIA
-VkResult Semaphore::importHandle(zx_handle_t handle, bool temporaryImport)
+VkResult BinarySemaphore::importHandle(zx_handle_t handle, bool temporaryImport)
 {
 	return importPayload(
 	    temporaryImport,
@@ -326,7 +341,7 @@
 	    });
 }
 
-VkResult Semaphore::exportHandle(zx_handle_t *pHandle)
+VkResult BinarySemaphore::exportHandle(zx_handle_t *pHandle)
 {
 	if((exportableHandleTypes & VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_TEMP_ZIRCON_EVENT_BIT_FUCHSIA) == 0)
 	{
diff --git a/src/Vulkan/VkSemaphore.hpp b/src/Vulkan/VkSemaphore.hpp
index 3f32111..c188a93 100644
--- a/src/Vulkan/VkSemaphore.hpp
+++ b/src/Vulkan/VkSemaphore.hpp
@@ -22,16 +22,45 @@
 #include "marl/mutex.h"
 #include "marl/tsa.h"
 
+#include "System/Synchronization.hpp"
+
 #if VK_USE_PLATFORM_FUCHSIA
 #	include <zircon/types.h>
 #endif
 
 namespace vk {
 
-class Semaphore : public Object<Semaphore, VkSemaphore>
+class BinarySemaphore;
+class TimelineSemaphore;
+
+class Semaphore
 {
 public:
-	Semaphore(const VkSemaphoreCreateInfo *pCreateInfo, void *mem, const VkAllocationCallbacks *pAllocator);
+	Semaphore(VkSemaphoreType type);
+
+	virtual ~Semaphore() = default;
+
+	static inline Semaphore *Cast(VkSemaphore semaphore)
+	{
+		return static_cast<Semaphore *>(static_cast<void *>(semaphore));
+	}
+
+	virtual void destroy(const VkAllocationCallbacks *pAllocator)
+	{
+	}
+
+	VkSemaphoreType getSemaphoreType() const;
+	//static size_t ComputeRequiredAllocationSize(const VkSemaphoreCreateInfo *pCreateInfo);
+
+protected:
+	VkSemaphoreType type;
+	marl::mutex mutex;
+};
+
+class BinarySemaphore : public Semaphore, public Object<BinarySemaphore, VkSemaphore>
+{
+public:
+	BinarySemaphore(const VkSemaphoreCreateInfo *pCreateInfo, void *mem, const VkAllocationCallbacks *pAllocator);
 	void destroy(const VkAllocationCallbacks *pAllocator);
 
 	static size_t ComputeRequiredAllocationSize(const VkSemaphoreCreateInfo *pCreateInfo);
@@ -149,7 +178,6 @@
 	const VkAllocationCallbacks *allocator = nullptr;
 	VkExternalSemaphoreHandleTypeFlags exportableHandleTypes = (VkExternalSemaphoreHandleTypeFlags)0;
 	marl::Event internal;
-	marl::mutex mutex;
 	External *external GUARDED_BY(mutex) = nullptr;
 	External *tempExternal GUARDED_BY(mutex) = nullptr;
 };
@@ -159,6 +187,48 @@
 	return Semaphore::Cast(object);
 }
 
+template<typename T>
+static inline T *DynamicCast(VkSemaphore object)
+{
+	Semaphore *semaphore = vk::Cast(object);
+	if(semaphore == nullptr)
+	{
+		return nullptr;
+	}
+
+	static_assert(std::is_same_v<T, BinarySemaphore> || std::is_same_v<T, TimelineSemaphore>);
+	if constexpr(std::is_same_v<T, BinarySemaphore>)
+	{
+		if(semaphore->getSemaphoreType() != VK_SEMAPHORE_TYPE_BINARY)
+		{
+			return nullptr;
+		}
+	}
+	else
+	{
+		if(semaphore->getSemaphoreType() != VK_SEMAPHORE_TYPE_TIMELINE)
+		{
+			return nullptr;
+		}
+	}
+	return static_cast<T *>(semaphore);
+}
+
+// This struct helps parse VkSemaphoreCreateInfo. It also looks at the pNext
+// structures and stores their data flatly in a single struct. The default
+// values of each data member are what the absence of a pNext struct implies
+// for those values.
+struct SemaphoreCreateInfo
+{
+	bool exportSemaphore = false;
+	VkExternalSemaphoreHandleTypeFlags exportHandleTypes = 0;
+
+	VkSemaphoreType semaphoreType = VK_SEMAPHORE_TYPE_BINARY;
+	uint64_t initialPayload = 0;
+
+	SemaphoreCreateInfo(const VkSemaphoreCreateInfo *pCreateInfo);
+};
+
 }  // namespace vk
 
 #endif  // VK_SEMAPHORE_HPP_
diff --git a/src/Vulkan/VkSemaphoreExternalFuchsia.hpp b/src/Vulkan/VkSemaphoreExternalFuchsia.hpp
index 1e37e68..d204476 100644
--- a/src/Vulkan/VkSemaphoreExternalFuchsia.hpp
+++ b/src/Vulkan/VkSemaphoreExternalFuchsia.hpp
@@ -26,7 +26,7 @@
 
 namespace vk {
 
-class ZirconEventExternalSemaphore : public Semaphore::External
+class ZirconEventExternalSemaphore : public BinarySemaphore::External
 {
 public:
 	~ZirconEventExternalSemaphore()
diff --git a/src/Vulkan/VkSemaphoreExternalLinux.hpp b/src/Vulkan/VkSemaphoreExternalLinux.hpp
index d4e4cd2..3689ad9 100644
--- a/src/Vulkan/VkSemaphoreExternalLinux.hpp
+++ b/src/Vulkan/VkSemaphoreExternalLinux.hpp
@@ -130,7 +130,7 @@
 
 namespace vk {
 
-class OpaqueFdExternalSemaphore : public Semaphore::External
+class OpaqueFdExternalSemaphore : public BinarySemaphore::External
 {
 public:
 	~OpaqueFdExternalSemaphore() { unmapRegion(); }
diff --git a/src/Vulkan/VkTimelineSemaphore.cpp b/src/Vulkan/VkTimelineSemaphore.cpp
new file mode 100644
index 0000000..7aa85db
--- /dev/null
+++ b/src/Vulkan/VkTimelineSemaphore.cpp
@@ -0,0 +1,145 @@
+// Copyright 2021 The SwiftShader Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "VkTimelineSemaphore.hpp"
+#include "VkSemaphore.hpp"
+
+#include "marl/blockingcall.h"
+#include "marl/conditionvariable.h"
+
+#include <vector>
+
+namespace vk {
+
+TimelineSemaphore::TimelineSemaphore(const VkSemaphoreCreateInfo *pCreateInfo, void *mem, const VkAllocationCallbacks *pAllocator)
+    : Semaphore(VK_SEMAPHORE_TYPE_TIMELINE)
+{
+	SemaphoreCreateInfo info(pCreateInfo);
+	ASSERT(info.semaphoreType == VK_SEMAPHORE_TYPE_TIMELINE);
+	type = info.semaphoreType;
+	shared = marl::Allocator::Default->make_shared<TimelineSemaphore::Shared>(marl::Allocator::Default, info.initialPayload);
+}
+
+TimelineSemaphore::TimelineSemaphore()
+    : Semaphore(VK_SEMAPHORE_TYPE_TIMELINE)
+{
+	type = VK_SEMAPHORE_TYPE_TIMELINE;
+	shared = marl::Allocator::Default->make_shared<TimelineSemaphore::Shared>(marl::Allocator::Default, 0);
+}
+
+size_t TimelineSemaphore::ComputeRequiredAllocationSize(const VkSemaphoreCreateInfo *pCreateInfo)
+{
+	return 0;
+}
+
+void TimelineSemaphore::destroy(const VkAllocationCallbacks *pAllocator)
+{
+}
+
+void TimelineSemaphore::signal(uint64_t value)
+{
+	return shared->signal(value);
+}
+
+void TimelineSemaphore::Shared::signal(uint64_t value)
+{
+	marl::lock lock(mutex);
+	if(counter < value)
+	{
+		counter = value;
+		cv.notify_all();
+		for(auto dep : deps)
+		{
+			dep->signal(id, counter);
+		}
+	}
+}
+
+void TimelineSemaphore::wait(uint64_t value)
+{
+	shared->wait(value);
+}
+
+void TimelineSemaphore::Shared::wait(uint64_t value)
+{
+	marl::lock lock(mutex);
+	cv.wait(lock, [&]() { return counter == value; });
+}
+
+uint64_t TimelineSemaphore::getCounterValue()
+{
+	return shared->getCounterValue();
+}
+
+uint64_t TimelineSemaphore::Shared::getCounterValue()
+{
+	marl::lock lock(mutex);
+	return counter;
+}
+
+std::atomic<int> TimelineSemaphore::Shared::nextId;
+
+TimelineSemaphore::Shared::Shared(marl::Allocator *allocator, uint64_t initialState)
+    : cv(allocator)
+    , counter(initialState)
+    , id(nextId++)
+{
+}
+
+void TimelineSemaphore::Shared::signal(int parentId, uint64_t value)
+{
+	marl::lock lock(mutex);
+	auto it = waitMap.find(parentId);
+	// Either we aren't waiting for a signal, or parentId is not something we're waiting for
+	// Reject any signals that we aren't waiting on
+	if(counter == 0 && it != waitMap.end() && value == it->second)
+	{
+		// Stop waiting on all parents once we find a signal
+		waitMap.clear();
+		counter = 1;
+		cv.notify_all();
+		for(auto dep : deps)
+		{
+			dep->signal(id, counter);
+		}
+	}
+}
+
+void TimelineSemaphore::addDependent(TimelineSemaphore &other, uint64_t waitValue)
+{
+	shared->addDependent(other);
+	other.addDependency(shared->id, waitValue);
+}
+
+void TimelineSemaphore::Shared::addDependent(TimelineSemaphore &other)
+{
+	marl::lock lock(mutex);
+	deps.push_back(other.shared);
+}
+
+void TimelineSemaphore::addDependency(int id, uint64_t waitValue)
+{
+	shared->addDependency(id, waitValue);
+}
+
+void TimelineSemaphore::Shared::addDependency(int id, uint64_t waitValue)
+{
+	marl::lock lock(mutex);
+	auto mapPos = waitMap.find(id);
+	ASSERT(mapPos == waitMap.end());
+
+	waitMap.insert(mapPos, std::make_pair(id, waitValue));
+}
+
+}  // namespace vk
diff --git a/src/Vulkan/VkTimelineSemaphore.hpp b/src/Vulkan/VkTimelineSemaphore.hpp
new file mode 100644
index 0000000..55c3f45
--- /dev/null
+++ b/src/Vulkan/VkTimelineSemaphore.hpp
@@ -0,0 +1,166 @@
+// Copyright 2021 The SwiftShader Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef VK_TIMELINE_SEMAPHORE_HPP_
+#define VK_TIMELINE_SEMAPHORE_HPP_
+
+#include "VkConfig.hpp"
+#include "VkObject.hpp"
+#include "VkSemaphore.hpp"
+
+#include "marl/conditionvariable.h"
+#include "marl/mutex.h"
+
+#include "System/Synchronization.hpp"
+
+#include <chrono>
+
+namespace vk {
+
+struct Shared;
+
+// Timeline Semaphores track a 64-bit payload instead of a binary payload.
+//
+// A timeline does not have a "signaled" and "unsignalled" state. Threads instead wait
+// for the payload to become a certain value. When a thread signals the timeline, it provides
+// a new payload that is greater than the current payload.
+//
+// There is no way to reset a timeline or to decrease the payload's value. A user must instead
+// create a new timeline with a new initial payload if they desire this behavior.
+class TimelineSemaphore : public Semaphore, public Object<TimelineSemaphore, VkSemaphore>
+{
+public:
+	TimelineSemaphore(const VkSemaphoreCreateInfo *pCreateInfo, void *mem, const VkAllocationCallbacks *pAllocator);
+	TimelineSemaphore();
+
+	static size_t ComputeRequiredAllocationSize(const VkSemaphoreCreateInfo *pCreateInfo);
+
+	// Block until this semaphore is signaled with the specified value;
+	void wait(uint64_t value);
+
+	// Wait until a certain amount of time has passed or until the specified value is signaled.
+	template<class CLOCK, class DURATION>
+	VkResult wait(uint64_t value, const std::chrono::time_point<CLOCK, DURATION> end_ns);
+
+	// Set the payload to the specified value and signal all waiting threads.
+	void signal(uint64_t value);
+
+	// Retrieve the current payload. This should not be used to make thread execution decisions
+	// as there's no guarantee that the value returned here matches the actual payload's value.
+	uint64_t getCounterValue();
+
+	// Dependent timeline semaphores allow an 'any' semaphore to be created that can wait on the
+	// state of multiple other timeline semaphores and be signaled like a binary semaphore
+	// if any of its parent semaphores are signaled with a certain value.
+	//
+	// Since a timeline semaphore can be signalled with nearly any value, but threads waiting
+	// on a timeline semaphore only unblock when a specific value is signaled, dependents can't
+	// naively become signaled whenever their parent semaphores are signaled with a new value.
+	// Instead, the dependent semaphore needs to wait for its parent semaphore to be signaled
+	// with a specific value as well. This specific value may differ for each parent semaphore.
+	//
+	// So this function adds other as a dependent semaphore, and tells it to only become unsignaled
+	// by this semaphore when this semaphore is signaled with waitValue.
+	void addDependent(TimelineSemaphore &other, uint64_t waitValue);
+	void addDependency(int id, uint64_t waitValue);
+
+	// Tells this semaphore to become signaled as part of a dependency chain when the parent semaphore
+	// with the specified id is signaled with the specified waitValue.
+	void addToWaitMap(int parentId, uint64_t waitValue);
+
+	// Clean up any allocated resources
+	void destroy(const VkAllocationCallbacks *pAllocator);
+
+private:
+	// Track the 64-bit payload. Timeline Semaphores have a shared_ptr<Shared>
+	// that they can pass to other Timeline Semaphores to create dependency chains.
+	struct Shared
+	{
+	private:
+		// Guards access to all the resources that may be accessed by other threads.
+		// No clang Thread Safety Analysis is used on variables guarded by mutex
+		// as there is an issue with TSA. Despite instrumenting everything properly,
+		// compilation will fail when a lambda function uses a guarded resource.
+		marl::mutex mutex;
+
+		static std::atomic<int> nextId;
+
+	public:
+		Shared(marl::Allocator *allocator, uint64_t initialState);
+
+		// Block until this semaphore is signaled with the specified value;
+		void wait(uint64_t value);
+		// Wait until a certain amount of time has passed or until the specified value is signaled.
+		template<class CLOCK, class DURATION>
+		VkResult wait(uint64_t value, const std::chrono::time_point<CLOCK, DURATION> end_ns);
+
+		// Pass a signal down to a dependent.
+		void signal(int parentId, uint64_t value);
+		// Set the payload to the specified value and signal all waiting threads.
+		void signal(uint64_t value);
+
+		// Retrieve the current payload. This should not be used to make thread execution decisions
+		// as there's no guarantee that the value returned here matches the actual payload's value.
+		uint64_t getCounterValue();
+
+		// Add the other semaphore's Shared to deps.
+		void addDependent(TimelineSemaphore &other);
+
+		// Add {id, waitValue} as a key-value pair to waitMap.
+		void addDependency(int id, uint64_t waitValue);
+
+		// Entry point to the marl threading library that handles blocking and unblocking.
+		marl::ConditionVariable cv;
+
+		// TODO(b/181683382) -- Add Thread Safety Analysis instrumentation when it can properly
+		// analyze lambdas.
+		// The 64-bit payload.
+		uint64_t counter;
+
+		// A list of this semaphore's dependents.
+		marl::containers::vector<std::shared_ptr<Shared>, 1> deps;
+
+		// A map of {parentId: waitValue} pairs that tracks when this semaphore should unblock if it's
+		// signaled as a dependent by another semaphore.
+		std::map<int, uint64_t> waitMap;
+
+		// An ID that's unique for each instance of Shared
+		const int id;
+	};
+
+	std::shared_ptr<Shared> shared;
+};
+
+template<typename Clock, typename Duration>
+VkResult TimelineSemaphore::wait(uint64_t value,
+                                 const std::chrono::time_point<Clock, Duration> timeout)
+{
+	return shared->wait(value, timeout);
+}
+
+template<typename Clock, typename Duration>
+VkResult TimelineSemaphore::Shared::wait(uint64_t value,
+                                         const std::chrono::time_point<Clock, Duration> timeout)
+{
+	marl::lock lock(mutex);
+	if(!cv.wait_until(lock, timeout, [&]() { return counter == value; }))
+	{
+		return VK_TIMEOUT;
+	}
+	return VK_SUCCESS;
+}
+
+}  // namespace vk
+
+#endif  // VK_TIMELINE_SEMAPHORE_HPP_
diff --git a/src/Vulkan/libVulkan.cpp b/src/Vulkan/libVulkan.cpp
index 482e178..a32c3ff 100644
--- a/src/Vulkan/libVulkan.cpp
+++ b/src/Vulkan/libVulkan.cpp
@@ -42,6 +42,7 @@
 #include "VkSemaphore.hpp"
 #include "VkShaderModule.hpp"
 #include "VkStringify.hpp"
+#include "VkTimelineSemaphore.hpp"
 
 #include "System/Debug.hpp"
 
@@ -421,6 +422,7 @@
 	{ { VK_KHR_SHADER_SUBGROUP_EXTENDED_TYPES_EXTENSION_NAME, VK_KHR_SHADER_SUBGROUP_EXTENDED_TYPES_SPEC_VERSION } },
 	{ { VK_KHR_SPIRV_1_4_EXTENSION_NAME, VK_KHR_SPIRV_1_4_SPEC_VERSION } },
 	{ { VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_SPEC_VERSION } },
+	{ { VK_KHR_TIMELINE_SEMAPHORE_EXTENSION_NAME, VK_KHR_TIMELINE_SEMAPHORE_SPEC_VERSION } },
 };
 
 static uint32_t numSupportedExtensions(const ExtensionProperties *extensionProperties, uint32_t extensionPropertiesCount)
@@ -1428,7 +1430,35 @@
 		UNSUPPORTED("pCreateInfo->flags %d", int(pCreateInfo->flags));
 	}
 
-	return vk::Semaphore::Create(pAllocator, pCreateInfo, pSemaphore, pAllocator);
+	VkSemaphoreType type = VK_SEMAPHORE_TYPE_BINARY;
+	for(const auto *nextInfo = reinterpret_cast<const VkBaseInStructure *>(pCreateInfo->pNext);
+	    nextInfo != nullptr; nextInfo = nextInfo->pNext)
+	{
+		switch(nextInfo->sType)
+		{
+			case VK_STRUCTURE_TYPE_EXPORT_SEMAPHORE_CREATE_INFO:
+				// Let the semaphore constructor handle this
+				break;
+			case VK_STRUCTURE_TYPE_SEMAPHORE_TYPE_CREATE_INFO:
+			{
+				const VkSemaphoreTypeCreateInfo *info = reinterpret_cast<const VkSemaphoreTypeCreateInfo *>(nextInfo);
+				type = info->semaphoreType;
+			}
+			break;
+			default:
+				WARN("nextInfo->sType = %s", vk::Stringify(nextInfo->sType).c_str());
+				break;
+		}
+	}
+
+	if(type == VK_SEMAPHORE_TYPE_BINARY)
+	{
+		return vk::BinarySemaphore::Create(pAllocator, pCreateInfo, pSemaphore, pAllocator);
+	}
+	else
+	{
+		return vk::TimelineSemaphore::Create(pAllocator, pCreateInfo, pSemaphore, pAllocator);
+	}
 }
 
 VKAPI_ATTR void VKAPI_CALL vkDestroySemaphore(VkDevice device, VkSemaphore semaphore, const VkAllocationCallbacks *pAllocator)
@@ -1450,7 +1480,9 @@
 		UNSUPPORTED("pGetFdInfo->handleType %d", int(pGetFdInfo->handleType));
 	}
 
-	return vk::Cast(pGetFdInfo->semaphore)->exportFd(pFd);
+	auto *sem = vk::DynamicCast<vk::BinarySemaphore>(pGetFdInfo->semaphore);
+	ASSERT(sem != nullptr);
+	return sem->exportFd(pFd);
 }
 
 VKAPI_ATTR VkResult VKAPI_CALL vkImportSemaphoreFdKHR(VkDevice device, const VkImportSemaphoreFdInfoKHR *pImportSemaphoreInfo)
@@ -1464,7 +1496,9 @@
 	}
 	bool temporaryImport = (pImportSemaphoreInfo->flags & VK_SEMAPHORE_IMPORT_TEMPORARY_BIT) != 0;
 
-	return vk::Cast(pImportSemaphoreInfo->semaphore)->importFd(pImportSemaphoreInfo->fd, temporaryImport);
+	auto *sem = vk::DynamicCast<vk::BinarySemaphore>(pImportSemaphoreInfo->semaphore);
+	ASSERT(sem != nullptr);
+	return sem->importFd(pImportSemaphoreInfo->fd, temporaryImport);
 }
 #endif  // SWIFTSHADER_EXTERNAL_SEMAPHORE_OPAQUE_FD
 
@@ -1481,8 +1515,9 @@
 		UNSUPPORTED("pImportSemaphoreZirconHandleInfo->handleType %d", int(pImportSemaphoreZirconHandleInfo->handleType));
 	}
 	bool temporaryImport = (pImportSemaphoreZirconHandleInfo->flags & VK_SEMAPHORE_IMPORT_TEMPORARY_BIT) != 0;
-
-	return vk::Cast(pImportSemaphoreZirconHandleInfo->semaphore)->importHandle(pImportSemaphoreZirconHandleInfo->handle, temporaryImport);
+	auto *sem = vk::DynamicCast<vk::BinarySemaphore>(pImportSemaphoreZirconHandleInfo->semaphore);
+	ASSERT(sem != nullptr);
+	return sem->importHandle(pImportSemaphoreZirconHandleInfo->handle, temporaryImport);
 }
 
 VKAPI_ATTR VkResult VKAPI_CALL vkGetSemaphoreZirconHandleFUCHSIA(
@@ -1498,10 +1533,35 @@
 		UNSUPPORTED("pGetZirconHandleInfo->handleType %d", int(pGetZirconHandleInfo->handleType));
 	}
 
-	return vk::Cast(pGetZirconHandleInfo->semaphore)->exportHandle(pZirconHandle);
+	auto *sem = vk::DynamicCast<vk::BinarySemaphore>(pGetZirconHandleInfo->semaphore);
+	ASSERT(sem != nullptr);
+	return sem->exportHandle(pZirconHandle);
 }
 #endif  // VK_USE_PLATFORM_FUCHSIA
 
+VKAPI_ATTR VkResult VKAPI_CALL vkGetSemaphoreCounterValue(VkDevice device, VkSemaphore semaphore, uint64_t *pValue)
+{
+	TRACE("(VkDevice device = %p, VkSemaphore semaphore = %p, uint64_t* pValue = %p)",
+	      device, static_cast<void *>(semaphore), pValue);
+	*pValue = vk::DynamicCast<vk::TimelineSemaphore>(semaphore)->getCounterValue();
+	return VK_SUCCESS;
+}
+
+VKAPI_ATTR VkResult VKAPI_CALL vkSignalSemaphore(VkDevice device, const VkSemaphoreSignalInfo *pSignalInfo)
+{
+	TRACE("(VkDevice device = %p, const VkSemaphoreSignalInfo *pSignalInfo = %p)",
+	      device, pSignalInfo);
+	vk::DynamicCast<vk::TimelineSemaphore>(pSignalInfo->semaphore)->signal(pSignalInfo->value);
+	return VK_SUCCESS;
+}
+
+VKAPI_ATTR VkResult VKAPI_CALL vkWaitSemaphores(VkDevice device, const VkSemaphoreWaitInfo *pWaitInfo, uint64_t timeout)
+{
+	TRACE("(VkDevice device = %p, const VkSemaphoreWaitInfo *pWaitInfo = %p, uint64_t timeout = %" PRIu64 ")",
+	      device, pWaitInfo, timeout);
+	return vk::Cast(device)->waitForSemaphores(pWaitInfo, timeout);
+}
+
 VKAPI_ATTR VkResult VKAPI_CALL vkCreateEvent(VkDevice device, const VkEventCreateInfo *pCreateInfo, const VkAllocationCallbacks *pAllocator, VkEvent *pEvent)
 {
 	TRACE("(VkDevice device = %p, const VkEventCreateInfo* pCreateInfo = %p, const VkAllocationCallbacks* pAllocator = %p, VkEvent* pEvent = %p)",
@@ -3949,7 +4009,7 @@
 	TRACE("(VkDevice device = %p, VkSwapchainKHR swapchain = %p, uint64_t timeout = %" PRIu64 ", VkSemaphore semaphore = %p, VkFence fence = %p, uint32_t* pImageIndex = %p)",
 	      device, static_cast<void *>(swapchain), timeout, static_cast<void *>(semaphore), static_cast<void *>(fence), pImageIndex);
 
-	return vk::Cast(swapchain)->getNextImage(timeout, vk::Cast(semaphore), vk::Cast(fence), pImageIndex);
+	return vk::Cast(swapchain)->getNextImage(timeout, vk::DynamicCast<vk::BinarySemaphore>(semaphore), vk::Cast(fence), pImageIndex);
 }
 
 VKAPI_ATTR VkResult VKAPI_CALL vkQueuePresentKHR(VkQueue queue, const VkPresentInfoKHR *pPresentInfo)
@@ -3965,7 +4025,7 @@
 	TRACE("(VkDevice device = %p, const VkAcquireNextImageInfoKHR *pAcquireInfo = %p, uint32_t *pImageIndex = %p",
 	      device, pAcquireInfo, pImageIndex);
 
-	return vk::Cast(pAcquireInfo->swapchain)->getNextImage(pAcquireInfo->timeout, vk::Cast(pAcquireInfo->semaphore), vk::Cast(pAcquireInfo->fence), pImageIndex);
+	return vk::Cast(pAcquireInfo->swapchain)->getNextImage(pAcquireInfo->timeout, vk::DynamicCast<vk::BinarySemaphore>(pAcquireInfo->semaphore), vk::Cast(pAcquireInfo->fence), pImageIndex);
 }
 
 VKAPI_ATTR VkResult VKAPI_CALL vkGetDeviceGroupPresentCapabilitiesKHR(VkDevice device, VkDeviceGroupPresentCapabilitiesKHR *pDeviceGroupPresentCapabilities)
diff --git a/src/Vulkan/vk_swiftshader.def b/src/Vulkan/vk_swiftshader.def
index 77fb8c6..f0c2c9c 100644
--- a/src/Vulkan/vk_swiftshader.def
+++ b/src/Vulkan/vk_swiftshader.def
@@ -226,6 +226,13 @@
 	vkGetPhysicalDeviceSurfaceCapabilitiesKHR
 	vkGetPhysicalDeviceSurfaceFormatsKHR
 	vkGetPhysicalDeviceSurfacePresentModesKHR
+	; VK_KHR_timeline_semaphore
+	vkGetSemaphoreCounterValue
+	vkSignalSemaphore
+	vkWaitSemaphores
+	vkGetSemaphoreCounterValueKHR
+	vkSignalSemaphoreKHR
+	vkWaitSemaphoresKHR
 	; VK_KHR_win32_surface
 	vkCreateWin32SurfaceKHR
 	vkGetPhysicalDeviceWin32PresentationSupportKHR
diff --git a/src/Vulkan/vk_swiftshader.lds b/src/Vulkan/vk_swiftshader.lds
index 768114e..0268f98 100644
--- a/src/Vulkan/vk_swiftshader.lds
+++ b/src/Vulkan/vk_swiftshader.lds
@@ -230,7 +230,13 @@
 	vkResetQueryPool;
 	# VK_EXT_headless_surface
 	vkCreateHeadlessSurfaceEXT;
-
+	# VK_KHR_timeline_semaphore
+	vkGetSemaphoreCounterValue;
+	vkSignalSemaphore;
+	vkWaitSemaphores;
+	vkGetSemaphoreCounterValueKHR;
+	vkSignalSemaphoreKHR;
+	vkWaitSemaphoresKHR;
 	# Android HAL module info object
 	HMI;
 
diff --git a/src/WSI/VkSwapchainKHR.cpp b/src/WSI/VkSwapchainKHR.cpp
index aed8c03..57d9cdd 100644
--- a/src/WSI/VkSwapchainKHR.cpp
+++ b/src/WSI/VkSwapchainKHR.cpp
@@ -170,7 +170,7 @@
 	return VK_SUCCESS;
 }
 
-VkResult SwapchainKHR::getNextImage(uint64_t timeout, Semaphore *semaphore, Fence *fence, uint32_t *pImageIndex)
+VkResult SwapchainKHR::getNextImage(uint64_t timeout, BinarySemaphore *semaphore, Fence *fence, uint32_t *pImageIndex)
 {
 	for(uint32_t i = 0; i < imageCount; i++)
 	{
diff --git a/src/WSI/VkSwapchainKHR.hpp b/src/WSI/VkSwapchainKHR.hpp
index e5aca6e..b4dcbad 100644
--- a/src/WSI/VkSwapchainKHR.hpp
+++ b/src/WSI/VkSwapchainKHR.hpp
@@ -24,7 +24,7 @@
 namespace vk {
 
 class Fence;
-class Semaphore;
+class BinarySemaphore;
 
 class SwapchainKHR : public Object<SwapchainKHR, VkSwapchainKHR>
 {
@@ -42,7 +42,7 @@
 	uint32_t getImageCount() const;
 	VkResult getImages(uint32_t *pSwapchainImageCount, VkImage *pSwapchainImages) const;
 
-	VkResult getNextImage(uint64_t timeout, Semaphore *semaphore, Fence *fence, uint32_t *pImageIndex);
+	VkResult getNextImage(uint64_t timeout, BinarySemaphore *semaphore, Fence *fence, uint32_t *pImageIndex);
 
 	VkResult present(uint32_t index);
 	PresentImage const &getImage(uint32_t imageIndex) { return images[imageIndex]; }