Preliminary support for realtime clock (#2772)

* #include an absolute path didn't work - because paths were taken to always be relative.

* Initial support for realtime clock.

* Add realtime-clock render feature where seems appropriate.

* Fixes to make NVAPI compile properly.
Change realtime-clock.slang check to use maths that can't overflow.

7 files changed, 39 insertions, 3 deletions

diff --git a/tools/gfx/cuda/cuda-device.cpp b/tools/gfx/cuda/cuda-device.cpp
index f81bcfe99..4aeecb606 100644
--- a/tools/gfx/cuda/cuda-device.cpp
+++ b/tools/gfx/cuda/cuda-device.cpp
@@ -179,9 +179,12 @@ SLANG_NO_THROW SlangResult SLANG_MCALL DeviceImpl::initialize(const Desc& desc)
     SLANG_CUDA_RETURN_WITH_REPORT_ON_FAIL(
         cuCtxCreate(&m_context->m_context, 0, m_device), reportType);
 
-    // Not clear how to detect half support on CUDA. For now we'll assume we have it
     {
+        // Not clear how to detect half support on CUDA. For now we'll assume we have it
         m_features.add("half");
+        
+        // CUDA has support for realtime clock
+        m_features.add("realtime-clock");
     }
 
     cudaDeviceProp deviceProps;
diff --git a/tools/gfx/d3d11/d3d11-device.cpp b/tools/gfx/d3d11/d3d11-device.cpp
index e32bdf7ed..cc2eda089 100644
--- a/tools/gfx/d3d11/d3d11-device.cpp
+++ b/tools/gfx/d3d11/d3d11-device.cpp
@@ -196,6 +196,11 @@ SlangResult DeviceImpl::initialize(const Desc& desc)
             m_features.add("atomic-float");
         }
 
+        // If we have NVAPI well assume we have realtime clock
+        {
+            m_features.add("realtime-clock");
+        }
+
         m_nvapi = true;
 #endif
     }
diff --git a/tools/gfx/d3d11/d3d11-scopeNVAPI.cpp b/tools/gfx/d3d11/d3d11-scopeNVAPI.cpp
index b230623fe..1a662e999 100644
--- a/tools/gfx/d3d11/d3d11-scopeNVAPI.cpp
+++ b/tools/gfx/d3d11/d3d11-scopeNVAPI.cpp
@@ -20,7 +20,7 @@ SlangResult ScopeNVAPI::init(DeviceImpl* device, Index regIndex)
     }
 
 #ifdef GFX_NVAPI
-    NvAPI_Status nvapiStatus = NvAPI_D3D11_SetNvShaderExtnSlot(renderer->m_device, NvU32(regIndex));
+    NvAPI_Status nvapiStatus = NvAPI_D3D11_SetNvShaderExtnSlot(device->m_device, NvU32(regIndex));
     if (nvapiStatus != NVAPI_OK)
     {
         return SLANG_FAIL;
diff --git a/tools/gfx/d3d12/d3d12-device.cpp b/tools/gfx/d3d12/d3d12-device.cpp
index 8f1a3e366..312c81d75 100644
--- a/tools/gfx/d3d12/d3d12-device.cpp
+++ b/tools/gfx/d3d12/d3d12-device.cpp
@@ -24,6 +24,10 @@
 #    define ENABLE_DEBUG_LAYER 0
 #endif
 
+#ifdef GFX_NVAPI
+#    include "../nvapi/nvapi-include.h"
+#endif
+
 namespace gfx
 {
 namespace d3d12
@@ -597,6 +601,11 @@ Result DeviceImpl::initialize(const Desc& desc)
             m_features.add("atomic-float");
         }
 
+        // If we have NVAPI well assume we have realtime clock
+        {
+            m_features.add("realtime-clock");
+        }
+
         m_nvapi = true;
 #endif
     }
diff --git a/tools/gfx/d3d12/d3d12-pipeline-state.cpp b/tools/gfx/d3d12/d3d12-pipeline-state.cpp
index adfdcd518..35313f676 100644
--- a/tools/gfx/d3d12/d3d12-pipeline-state.cpp
+++ b/tools/gfx/d3d12/d3d12-pipeline-state.cpp
@@ -240,7 +240,7 @@ Result PipelineStateImpl::ensureAPIPipelineStateCreated()
                 SIZE_T(programImpl->m_shaders[0].code.getCount()) };
 
 #ifdef GFX_NVAPI
-            if (m_nvapi)
+            if (m_device->m_nvapi)
             {
                 // Also fill the extension structure.
                 // Use the same UAV slot index and register space that are declared in the shader.
diff --git a/tools/gfx/vulkan/vk-api.h b/tools/gfx/vulkan/vk-api.h
index af2234f55..c34372f45 100644
--- a/tools/gfx/vulkan/vk-api.h
+++ b/tools/gfx/vulkan/vk-api.h
@@ -270,6 +270,11 @@ struct VulkanExtendedFeatureProperties
     // Robustness2 features
     VkPhysicalDeviceRobustness2FeaturesEXT robustness2Features = {
         VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT};
+
+    // Clock features
+    VkPhysicalDeviceShaderClockFeaturesKHR clockFeatures = { 
+        VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_CLOCK_FEATURES_KHR
+    };
 };
 
 struct VulkanApi
diff --git a/tools/gfx/vulkan/vk-device.cpp b/tools/gfx/vulkan/vk-device.cpp
index b5eec0b72..488f39be6 100644
--- a/tools/gfx/vulkan/vk-device.cpp
+++ b/tools/gfx/vulkan/vk-device.cpp
@@ -424,6 +424,10 @@ Result DeviceImpl::initVulkanInstanceAndDevice(
         extendedFeatures.robustness2Features.pNext = deviceFeatures2.pNext;
         deviceFeatures2.pNext = &extendedFeatures.robustness2Features;
 
+        // clock features
+        extendedFeatures.clockFeatures.pNext = deviceFeatures2.pNext;
+        deviceFeatures2.pNext = &extendedFeatures.clockFeatures;
+
         // Atomic Float
         // To detect atomic float we need
         // https://www.khronos.org/registry/vulkan/specs/1.2-extensions/man/html/VkPhysicalDeviceShaderAtomicFloatFeaturesEXT.html
@@ -575,6 +579,16 @@ Result DeviceImpl::initVulkanInstanceAndDevice(
             m_features.add("robustness2");
         }
 
+        if (extendedFeatures.clockFeatures.shaderDeviceClock)
+        {
+            deviceExtensions.add(VK_KHR_SHADER_CLOCK_EXTENSION_NAME);
+
+            extendedFeatures.clockFeatures.pNext = (void*)deviceCreateInfo.pNext;
+            deviceCreateInfo.pNext = &extendedFeatures.clockFeatures;
+
+            m_features.add("realtime-clock");
+        }
+
         VkPhysicalDeviceProperties2 extendedProps = {
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2 };
         VkPhysicalDeviceRayTracingPipelinePropertiesKHR rtProps = {