Add VK_ARM_data_graph support

layer_gpu_profile/source/CMakeLists.txt

@@ -49,6 +49,7 @@ add_library(
         layer_device_functions_command_buffer.cpp
         layer_device_functions_command_pool.cpp
         layer_device_functions_debug.cpp
+        layer_device_functions_dispatch_data_graph.cpp
         layer_device_functions_dispatch.cpp
         layer_device_functions_queue.cpp
         layer_device_functions_render_pass.cpp

layer_gpu_profile/source/layer_device_functions.hpp

@@ -194,6 +194,14 @@ VKAPI_ATTR void VKAPI_CALL layer_vkCmdDispatchIndirect<user_tag>(VkCommandBuffer
                                                                  VkBuffer buffer,
                                                                  VkDeviceSize offset);
 
+// Commands for data graph
+
+/* See Vulkan API for documentation. */
+template<>
+VKAPI_ATTR void VKAPI_CALL layer_vkCmdDispatchDataGraphARM<user_tag>(VkCommandBuffer commandBuffer,
+                                                                     VkDataGraphPipelineSessionARM session,
+                                                                     const VkDataGraphPipelineDispatchInfoARM* pInfo);
+
 // Commands for trace rays
 
 /* See Vulkan API for documentation. */

layer_gpu_profile/source/layer_device_functions_dispatch_data_graph.cpp

@@ -0,0 +1,71 @@
+/*
+ * SPDX-License-Identifier: MIT
+ * ----------------------------------------------------------------------------
+ * Copyright (c) 2024-2025 Arm Limited
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ * ----------------------------------------------------------------------------
+ */
+
+#include "device.hpp"
+#include "device_utils.hpp"
+#include "framework/device_dispatch_table.hpp"
+
+#include <mutex>
+
+extern std::mutex g_vulkanLock;
+
+/**
+ * @brief Register a compute dispatch with the tracker.
+ *
+ * @param layer           The layer context for the device.
+ * @param commandBuffer   The command buffer we are recording.
+ */
+static void registerDispatchDataGraph(Device* layer,
+                                      VkCommandBuffer commandBuffer)
+{
+    if (!layer->isFrameOfInterest)
+    {
+        return;
+    }
+
+    auto& tracker = layer->getStateTracker();
+    auto& cb = tracker.getCommandBuffer(commandBuffer);
+    cb.dispatchDataGraph();
+}
+
+/* See Vulkan API for documentation. */
+template<>
+VKAPI_ATTR void VKAPI_CALL layer_vkCmdDispatchDataGraphARM<user_tag>(VkCommandBuffer commandBuffer,
+                                                                     VkDataGraphPipelineSessionARM session,
+                                                                     const VkDataGraphPipelineDispatchInfoARM* pInfo)
+{
+    LAYER_TRACE(__func__);
+
+    // Hold the lock to access layer-wide global store
+    std::unique_lock<std::mutex> lock {g_vulkanLock};
+    auto* layer = Device::retrieve(commandBuffer);
+
+    registerDispatchDataGraph(layer, commandBuffer);
+
+    // Release the lock to call into the driver
+    lock.unlock();
+    layer->driver.vkCmdDispatchDataGraphARM(commandBuffer, session, pInfo);
+    emitCPUTrap(*layer, commandBuffer);
+}

layer_gpu_profile/source/submit_visitor.cpp

@@ -97,7 +97,7 @@ void ProfileSubmitVisitor::operator()(
 ) {
     UNUSED(renderPass);
 
-    handleCPUTrap("renderpass", debugStack);
+    handleCPUTrap("render_pass", debugStack);
 }
 
 /* See header for documentation */
@@ -123,14 +123,24 @@ void ProfileSubmitVisitor::operator()(
     handleCPUTrap("compute", debugStack);
 }
 
+/* See header for documentation */
+void ProfileSubmitVisitor::operator()(
+    const Tracker::LCSDispatchDataGraph& dispatch,
+    const std::vector<std::string>& debugStack
+) {
+    UNUSED(dispatch);
+
+    handleCPUTrap("data_graph", debugStack);
+}
+
 /* See header for documentation */
 void ProfileSubmitVisitor::operator()(
     const Tracker::LCSTraceRays& traceRays,
     const std::vector<std::string>& debugStack
 ) {
     UNUSED(traceRays);
 
-    handleCPUTrap("tracerays", debugStack);
+    handleCPUTrap("trace_rays", debugStack);
 }
 
 /* See header for documentation */

layer_gpu_profile/source/submit_visitor.hpp

@@ -68,6 +68,10 @@ class ProfileSubmitVisitor : public Tracker::SubmitCommandWorkloadVisitor
         const Tracker::LCSDispatch& dispatch,
         const std::vector<std::string>& debugStack) override;
 
+    void operator()(
+        const Tracker::LCSDispatchDataGraph& dispatch,
+        const std::vector<std::string>& debugStack) override;
+
     void operator()(
         const Tracker::LCSTraceRays& traceRays,
         const std::vector<std::string>& debugStack) override;

layer_gpu_timeline/source/CMakeLists.txt

@@ -47,6 +47,7 @@ add_library(
         layer_device_functions_command_buffer.cpp
         layer_device_functions_command_pool.cpp
         layer_device_functions_debug.cpp
+        layer_device_functions_dispatch_data_graph.cpp
         layer_device_functions_dispatch.cpp
         layer_device_functions_draw_call.cpp
         layer_device_functions_queue.cpp

layer_gpu_timeline/source/layer_device_functions.hpp

@@ -279,6 +279,14 @@ VKAPI_ATTR void VKAPI_CALL layer_vkCmdDispatchIndirect<user_tag>(VkCommandBuffer
                                                                  VkBuffer buffer,
                                                                  VkDeviceSize offset);
 
+// Commands for data graph
+
+/* See Vulkan API for documentation. */
+template<>
+VKAPI_ATTR void VKAPI_CALL layer_vkCmdDispatchDataGraphARM<user_tag>(VkCommandBuffer commandBuffer,
+                                                                     VkDataGraphPipelineSessionARM session,
+                                                                     const VkDataGraphPipelineDispatchInfoARM* pInfo);
+
 // Commands for trace rays
 
 /* See Vulkan API for documentation. */

layer_gpu_timeline/source/layer_device_functions_dispatch_data_graph.cpp

@@ -0,0 +1,55 @@
+/*
+ * SPDX-License-Identifier: MIT
+ * ----------------------------------------------------------------------------
+ * Copyright (c) 2025 Arm Limited
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ * ----------------------------------------------------------------------------
+ */
+
+#include "device.hpp"
+#include "device_utils.hpp"
+#include "framework/device_dispatch_table.hpp"
+
+#include <mutex>
+
+extern std::mutex g_vulkanLock;
+
+/* See Vulkan API for documentation. */
+template<>
+VKAPI_ATTR void VKAPI_CALL layer_vkCmdDispatchDataGraphARM<user_tag>(VkCommandBuffer commandBuffer,
+                                                                     VkDataGraphPipelineSessionARM session,
+                                                                     const VkDataGraphPipelineDispatchInfoARM* pInfo)
+{
+    LAYER_TRACE(__func__);
+
+    // Hold the lock to access layer-wide global store
+    std::unique_lock<std::mutex> lock {g_vulkanLock};
+    auto* layer = Device::retrieve(commandBuffer);
+
+    auto& tracker = layer->getStateTracker();
+    auto& cb = tracker.getCommandBuffer(commandBuffer);
+    uint64_t tagID = cb.dispatchDataGraph();
+
+    // Release the lock to call into the driver
+    lock.unlock();
+    emitStartTag(layer, commandBuffer, tagID);
+    layer->driver.vkCmdDispatchDataGraphARM(commandBuffer, session, pInfo);
+    emitEndTag(layer, commandBuffer);
+}

layer_gpu_timeline/source/timeline_protobuf_encoder.cpp

@@ -148,6 +148,13 @@ using Dispatch = pp::message<
     /* Any user defined debug labels associated with the dispatch */
     pp::string_field<"debug_label", 5, pp::repeated>>;
 
+/* A dispatch data graph object submission */
+using DispatchDataGraph = pp::message<
+    /* The unique identifier for this operation */
+    pp::uint64_field<"tag_id", 1>,
+    /* Any user defined debug labels associated with the dispatch */
+    pp::string_field<"debug_label", 2, pp::repeated>>;
+
 /* A trace rays object submission */
 using TraceRays = pp::message<
     /* The unique identifier for this operation */
@@ -251,7 +258,8 @@ using TimelineRecord =
                 pp::message_field<"image_transfer", 9, ImageTransfer>,
                 pp::message_field<"buffer_transfer", 10, BufferTransfer>,
                 pp::message_field<"acceleration_structure_build", 11, AccelerationStructureBuild>,
-                pp::message_field<"acceleration_structure_transfer", 12, AccelerationStructureTransfer>>;
+                pp::message_field<"acceleration_structure_transfer", 12, AccelerationStructureTransfer>,
+                pp::message_field<"dispatch_data_graph", 13, DispatchDataGraph>>;
 
 namespace
 {
@@ -527,6 +535,23 @@ Comms::MessageData serialize(const Tracker::LCSDispatch& dispatch, const std::ve
                       });
 }
 
+/**
+ * @brief Get the metadata for this workload
+ *
+ * @param dispatch     The dispatch data graph to serialize
+ * @param debugLabel   The debug label stack for the VkQueue at submit time.
+ */
+Comms::MessageData serialize(const Tracker::LCSDispatchDataGraph& dispatch, const std::vector<std::string>& debugLabel)
+{
+    using namespace pp;
+
+    return packBuffer("dispatch_data_graph"_f,
+                      DispatchDataGraph {
+                          dispatch.getTagID(),
+                          debugLabel,
+                      });
+}
+
 /**
  * @brief Get the metadata for this workload
  *
@@ -702,6 +727,12 @@ void TimelineProtobufEncoder::operator()(const Tracker::LCSDispatch& dispatch,
     device.txMessage(serialize(dispatch, debugStack));
 }
 
+void TimelineProtobufEncoder::operator()(const Tracker::LCSDispatchDataGraph& dispatch,
+                                         const std::vector<std::string>& debugStack)
+{
+    device.txMessage(serialize(dispatch, debugStack));
+}
+
 void TimelineProtobufEncoder::operator()(const Tracker::LCSTraceRays& traceRays,
                                          const std::vector<std::string>& debugStack)
 {

layer_gpu_timeline/source/timeline_protobuf_encoder.hpp

@@ -112,6 +112,7 @@ class TimelineProtobufEncoder : public Tracker::SubmitCommandWorkloadVisitor
                     const std::vector<std::string>& debugStack,
                     uint64_t renderPassTagID) override;
     void operator()(const Tracker::LCSDispatch& dispatch, const std::vector<std::string>& debugStack) override;
+    void operator()(const Tracker::LCSDispatchDataGraph& dispatch, const std::vector<std::string>& debugStack) override;
     void operator()(const Tracker::LCSTraceRays& traceRays, const std::vector<std::string>& debugStack) override;
     void operator()(const Tracker::LCSImageTransfer& imageTransfer,
                     const std::vector<std::string>& debugStack) override;

layer_gpu_timeline/timeline.proto

@@ -209,6 +209,14 @@ message Dispatch {
     repeated string debug_label = 5;
 }
 
+/* A dispatch data graph object submission */
+message DispatchDataGraph {
+    /* The unique identifier for this operation */
+    uint64 tag_id = 1;
+    /* Any user defined debug labels associated with the dispatch */
+    repeated string debug_label = 2;
+}
+
 /* A trace rays object submission */
 message TraceRays {
     /* The unique identifier for this operation */
@@ -317,4 +325,5 @@ message TimelineRecord {
     BufferTransfer buffer_transfer = 10;
     AccelerationStructureBuild acceleration_structure_build = 11;
     AccelerationStructureTransfer acceleration_structure_transfer = 12;
+    DispatchDataGraph dispatch_data_graph = 13;
 }

source_common/trackers/command_buffer.cpp

@@ -144,6 +144,21 @@ uint64_t CommandBuffer::dispatch(int64_t xGroups, int64_t yGroups, int64_t zGrou
     return tagID;
 }
 
+/* See header for documentation. */
+uint64_t CommandBuffer::dispatchDataGraph()
+{
+    uint64_t tagID = Tracker::LCSWorkload::assignTagID();
+    stats.incDispatchDataGraphCount();
+
+    // Add a workload to the command stream
+    auto workload = std::make_shared<LCSDispatchDataGraph>(tagID);
+
+    // Add a command to the layer-side command stream
+    workloadCommandStream.emplace_back(LCSInstructionWorkload(workload));
+
+    return tagID;
+}
+
 /* See header for documentation. */
 uint64_t CommandBuffer::traceRays(int64_t xItems, int64_t yItems, int64_t zItems)
 {

source_common/trackers/command_buffer.hpp

@@ -116,6 +116,13 @@ class CommandBuffer
      */
     uint64_t dispatch(int64_t xGroups, int64_t yGroups, int64_t zGroups);
 
+    /**
+     * @brief Capture a data graph dispatch.
+     *
+     * @return Returns the tagID assigned to this workload.
+     */
+    uint64_t dispatchDataGraph();
+
     /**
      * @brief Capture a trace rays dispatch.
      *

source_common/trackers/layer_command_stream.cpp

@@ -71,6 +71,12 @@ LCSDispatch::LCSDispatch(uint64_t _tagID, int64_t _xGroups, int64_t _yGroups, in
 {
 }
 
+/* See header for details. */
+LCSDispatchDataGraph::LCSDispatchDataGraph(uint64_t _tagID)
+    : LCSWorkload(_tagID)
+{
+}
+
 /* See header for details. */
 LCSTraceRays::LCSTraceRays(uint64_t _tagID, int64_t _xItems, int64_t _yItems, int64_t _zItems)
     : LCSWorkload(_tagID),

source_common/trackers/layer_command_stream.hpp

@@ -272,6 +272,20 @@ class LCSDispatch : public LCSWorkload
     int64_t zGroups;
 };
 
+/**
+ * @brief Class representing a data graph dispatch workload in the command stream.
+ */
+class LCSDispatchDataGraph : public LCSWorkload
+{
+public:
+    /**
+     * @brief Create a new data graph dispatch workload.
+     *
+     * @param tagID     The assigned tagID.
+     */
+    LCSDispatchDataGraph(uint64_t tagID);
+};
+
 /**
  * @brief Class representing a trace rays workload in the command stream.
  */
@@ -586,8 +600,10 @@ using LCSInstruction = std::variant<
     LCSInstructionWorkload<LCSRenderPass>,
     // The instruction represents a continuation of a render pass workload operation
     LCSInstructionWorkload<LCSRenderPassContinuation>,
-    // The instruction represents a dispatch workload operation
+    // The instruction represents a dispatch compute workload operation
     LCSInstructionWorkload<LCSDispatch>,
+    // The instruction represents a dispatch data graph workload operation
+    LCSInstructionWorkload<LCSDispatchDataGraph>,
     // The instruction represents a trace rays workload operation
     LCSInstructionWorkload<LCSTraceRays>,
     // The instruction represents an image transfer workload operation