diff --git a/qrenderdoc/Widgets/ComputeDebugSelector.cpp b/qrenderdoc/Widgets/ComputeDebugSelector.cpp
index b117cc8f8ca..c78b4f74e14 100644
--- a/qrenderdoc/Widgets/ComputeDebugSelector.cpp
+++ b/qrenderdoc/Widgets/ComputeDebugSelector.cpp
@@ -65,6 +65,21 @@ ComputeDebugSelector::~ComputeDebugSelector()
   delete ui;
 }
 
+void ComputeDebugSelector::SetDefaultDispatch(const rdcfixedarray<uint32_t, 3> &group,
+                                              const rdcfixedarray<uint32_t, 3> &thread)
+{
+  QSignalBlocker blockers[6] = {QSignalBlocker(ui->groupX),  QSignalBlocker(ui->groupY),
+                                QSignalBlocker(ui->groupZ),  QSignalBlocker(ui->threadX),
+                                QSignalBlocker(ui->threadY), QSignalBlocker(ui->threadZ)};
+
+  ui->groupX->setValue(group[0]);
+  ui->groupY->setValue(group[1]);
+  ui->groupZ->setValue(group[2]);
+  ui->threadX->setValue(thread[0]);
+  ui->threadY->setValue(thread[1]);
+  ui->threadZ->setValue(thread[2]);
+}
+
 void ComputeDebugSelector::SetThreadBounds(const rdcfixedarray<uint32_t, 3> &group,
                                            const rdcfixedarray<uint32_t, 3> &thread)
 {
diff --git a/qrenderdoc/Widgets/ComputeDebugSelector.h b/qrenderdoc/Widgets/ComputeDebugSelector.h
index 2657869e2b9..93c0ff63b30 100644
--- a/qrenderdoc/Widgets/ComputeDebugSelector.h
+++ b/qrenderdoc/Widgets/ComputeDebugSelector.h
@@ -40,6 +40,8 @@ class ComputeDebugSelector : public QDialog
   explicit ComputeDebugSelector(QWidget *parent = 0);
   ~ComputeDebugSelector();
 
+  void SetDefaultDispatch(const rdcfixedarray<uint32_t, 3> &group,
+                          const rdcfixedarray<uint32_t, 3> &thread);
   void SetThreadBounds(const rdcfixedarray<uint32_t, 3> &group,
                        const rdcfixedarray<uint32_t, 3> &thread);
 
diff --git a/qrenderdoc/Windows/BufferViewer.cpp b/qrenderdoc/Windows/BufferViewer.cpp
index 874cab7f631..119c0d32f6c 100644
--- a/qrenderdoc/Windows/BufferViewer.cpp
+++ b/qrenderdoc/Windows/BufferViewer.cpp
@@ -39,6 +39,7 @@
 #include "Code/QRDUtils.h"
 #include "Code/Resources.h"
 #include "Widgets/CollapseGroupBox.h"
+#include "Widgets/ComputeDebugSelector.h"
 #include "Widgets/Extended/RDLabel.h"
 #include "Widgets/Extended/RDSplitter.h"
 #include "Windows/Dialogs/AxisMappingDialog.h"
@@ -2361,6 +2362,8 @@ BufferViewer::BufferViewer(ICaptureContext &ctx, bool meshview, QWidget *parent)
   m_ModelOut1 = new BufferItemModel(ui->out1Table, false, meshview, this);
   m_ModelOut2 = new BufferItemModel(ui->out2Table, false, meshview, this);
 
+  m_MeshDebugSelector = new ComputeDebugSelector(this);
+
   // we keep the old UI names for serialised layouts compatibility
   QString containerNames[] = {
       lit("vsinData"),
@@ -2440,6 +2443,9 @@ BufferViewer::BufferViewer(ICaptureContext &ctx, bool meshview, QWidget *parent)
   m_DebugVert = new QAction(tr("&Debug this Vertex"), this);
   m_DebugVert->setIcon(Icons::wrench());
 
+  m_DebugMeshThread = new QAction(tr("&Debug Mesh Thread"), this);
+  m_DebugMeshThread->setIcon(Icons::wrench());
+
   m_FilterMesh = new QAction(tr("&Filter to this Meshlet"), this);
   m_FilterMesh->setIcon(Icons::filter());
 
@@ -2458,6 +2464,8 @@ BufferViewer::BufferViewer(ICaptureContext &ctx, bool meshview, QWidget *parent)
   QObject::connect(m_ExportBytes, &QAction::triggered,
                    [this] { exportData(BufferExport(BufferExport::RawBytes)); });
   QObject::connect(m_DebugVert, &QAction::triggered, this, &BufferViewer::debugVertex);
+  QObject::connect(m_DebugMeshThread, &QAction::triggered, this,
+                   &BufferViewer::debugMeshThread);
   QObject::connect(m_RemoveFilter, &QAction::triggered,
                    [this]() { SetMeshFilter(MeshFilter::None); });
   QObject::connect(m_FilterMesh, &QAction::triggered, [this]() {
@@ -2624,6 +2632,9 @@ BufferViewer::BufferViewer(ICaptureContext &ctx, bool meshview, QWidget *parent)
   ui->fixedVars->setTooltipElidedItems(false);
   ui->fixedVars->installEventFilter(this);
 
+  QObject::connect(m_MeshDebugSelector, &ComputeDebugSelector::beginDebug, this,
+                   &BufferViewer::meshDebugSelector_beginDebug);
+
   Reset();
 
   m_Ctx.AddCaptureViewer(this);
@@ -3093,6 +3104,37 @@ void BufferViewer::stageRowMenu(MeshDataStage stage, QMenu *menu, const QPoint &
       menu->addAction(m_RemoveFilter);
       menu->addAction(m_FilterMesh);
       menu->addAction(m_GotoTask);
+
+      const ShaderReflection *shaderDetails =
+          m_Ctx.CurPipelineState().GetShaderReflection(ShaderStage::Mesh);
+
+      m_DebugMeshThread->setEnabled(false);
+
+      if(!m_Ctx.APIProps().shaderDebugging)
+      {
+        m_DebugMeshThread->setToolTip(tr("This API does not support shader debugging"));
+      }
+      else if(!m_Ctx.CurAction() ||
+              !(m_Ctx.CurAction()->flags & (ActionFlags::Drawcall | ActionFlags::MeshDispatch)))
+      {
+        m_DebugMeshThread->setToolTip(tr("No draw call selected"));
+      }
+      else if(!shaderDetails)
+      {
+        m_DebugMeshThread->setToolTip(tr("No mesh shader bound"));
+      }
+      else if(!shaderDetails->debugInfo.debuggable)
+      {
+        m_DebugMeshThread->setToolTip(
+            tr("This shader doesn't support debugging: %1").arg(shaderDetails->debugInfo.debugStatus));
+      }
+      else
+      {
+        m_DebugMeshThread->setEnabled(true);
+        m_DebugMeshThread->setToolTip(QString());
+      }
+      menu->addAction(m_DebugMeshThread);
+
       menu->addSeparator();
 
       m_GotoTask->setEnabled(m_Ctx.CurPipelineState().GetShaderReflection(ShaderStage::Task));
@@ -6638,6 +6680,133 @@ void BufferViewer::debugVertex()
   m_Ctx.AddDockWindow(s->Widget(), DockReference::AddTo, this);
 }
 
+void BufferViewer::debugMeshThread()
+{
+  if(!m_Ctx.IsCaptureLoaded())
+    return;
+
+  const ActionDescription *action = m_Ctx.CurAction();
+  if(!action)
+    return;
+
+  if(!m_CurView)
+    return;
+
+  QModelIndex idx = m_CurView->selectionModel()->currentIndex();
+
+  if(!idx.isValid())
+  {
+    GUIInvoke::call(this, [this]() {
+      RDDialog::critical(this, tr("Error debugging"),
+                         tr("Error debugging meshlet - make sure a valid meshlet is selected"));
+    });
+    return;
+  }
+
+  uint32_t taskIndex = 0, meshletIndex = 0;
+  GetIndicesForMeshRow((uint32_t)idx.row(), taskIndex, meshletIndex);
+
+  const ShaderReflection *shaderDetails =
+      m_Ctx.CurPipelineState().GetShaderReflection(ShaderStage::Mesh);
+
+  if(!shaderDetails)
+    return;
+
+  rdcfixedarray<uint32_t, 3> threadGroupSize = action->dispatchThreadsDimension[0] == 0 ? shaderDetails->dispatchThreadsDimension : action->dispatchThreadsDimension;
+  m_MeshDebugSelector->SetThreadBounds(action->dispatchDimension, threadGroupSize);
+
+  // Calculate 3d group id from 1d meshlet index and dispatch dimensions
+  // Imagine 8x2x4 with idx 60
+  // 8x2 = 16
+  // 8x2x4 = 64
+  // 60 % x = 4
+  // 60 % (x * y) = 12 / x = 1
+  // 60 / (x * y) = 3
+  // index 60 is id (4,1,3)
+  // 4 + (8 * 1) + (16 * 3) = 60
+  uint32_t xDim = action->dispatchDimension[0];
+  uint32_t yDim = action->dispatchDimension[1];
+  uint32_t zDim = action->dispatchDimension[2];
+  rdcfixedarray<uint32_t, 3> meshletGroup = {
+      meshletIndex % xDim, 
+      (meshletIndex % (xDim * yDim)) / xDim, 
+      meshletIndex / (xDim * yDim),
+  };
+  m_MeshDebugSelector->SetDefaultDispatch(meshletGroup, {0, 0, 0});
+
+  RDDialog::show(m_MeshDebugSelector);
+}
+
+void BufferViewer::meshDebugSelector_beginDebug(const rdcfixedarray<uint32_t, 3> &group,
+                                                   const rdcfixedarray<uint32_t, 3> &thread)
+{
+  const ActionDescription *action = m_Ctx.CurAction();
+
+  if(!action)
+    return;
+
+  const ShaderReflection *shaderDetails =
+      m_Ctx.CurPipelineState().GetShaderReflection(ShaderStage::Mesh);
+
+  if(!shaderDetails)
+    return;
+
+  struct threadSelect
+  {
+    rdcfixedarray<uint32_t, 3> g;
+    rdcfixedarray<uint32_t, 3> t;
+  } debugThread = {
+      // g[]
+      {group[0], group[1], group[2]},
+      // t[]
+      {thread[0], thread[1], thread[2]},
+  };
+
+  bool done = false;
+  ShaderDebugTrace *trace = NULL;
+
+  m_Ctx.Replay().AsyncInvoke([&trace, &done, debugThread](IReplayController *r) {
+    trace = r->DebugMeshThread(debugThread.g, debugThread.t);
+
+    if(trace->debugger == NULL)
+    {
+      r->FreeTrace(trace);
+      trace = NULL;
+    }
+
+    done = true;
+  });
+
+  QString debugContext = lit("Mesh Group [%1,%2,%3] Thread [%4,%5,%6]")
+                             .arg(group[0])
+                             .arg(group[1])
+                             .arg(group[2])
+                             .arg(thread[0])
+                             .arg(thread[1])
+                             .arg(thread[2]);
+
+  // wait a short while before displaying the progress dialog (which won't show if we're already
+  // done by the time we reach it)
+  for(int i = 0; !done && i < 100; i++)
+    QThread::msleep(5);
+
+  ShowProgressDialog(this, tr("Debugging %1").arg(debugContext), [&done]() { return done; });
+
+  if(!trace)
+  {
+    RDDialog::critical(
+        this, tr("Error debugging"),
+        tr("Error debugging thread - make sure a valid group and thread is selected"));
+    return;
+  }
+
+  // viewer takes ownership of the trace
+  IShaderViewer *s = m_Ctx.DebugShader(
+      shaderDetails, m_Ctx.CurPipelineState().GetComputePipelineObject(), trace, debugContext);
+
+  m_Ctx.AddDockWindow(s->Widget(), DockReference::AddTo, this);
+}
+
 void BufferViewer::SyncViews(RDTableView *primary, bool selection, bool scroll)
 {
   if(!ui->syncViews->isChecked())
diff --git a/qrenderdoc/Windows/BufferViewer.h b/qrenderdoc/Windows/BufferViewer.h
index db6dffbfd7c..dc0aef81038 100644
--- a/qrenderdoc/Windows/BufferViewer.h
+++ b/qrenderdoc/Windows/BufferViewer.h
@@ -34,6 +34,7 @@ namespace Ui
 class BufferViewer;
 }
 
+class ComputeDebugSelector;
 class RDSpinBox64;
 class QItemSelection;
 class QMenu;
@@ -164,12 +165,16 @@ private slots:
   void updateExportActionNames();
   void exportData(const BufferExport &params);
   void debugVertex();
+  void debugMeshThread();
+  void meshDebugSelector_beginDebug(const rdcfixedarray<uint32_t, 3> &group,
+                                       const rdcfixedarray<uint32_t, 3> &thread);
   void fixedVars_contextMenu(const QPoint &pos);
 
 private:
   bool eventFilter(QObject *watched, QEvent *event) override;
   Ui::BufferViewer *ui;
   ICaptureContext &m_Ctx;
+  ComputeDebugSelector *m_MeshDebugSelector;
 
   IReplayOutput *m_Output;
 
@@ -323,6 +328,7 @@ private slots:
   QAction *m_ExportCSV = NULL;
   QAction *m_ExportBytes = NULL;
   QAction *m_DebugVert = NULL;
+  QAction *m_DebugMeshThread = NULL;
   QAction *m_FilterMesh = NULL;
   QAction *m_RemoveFilter = NULL;
   QAction *m_GotoTask = NULL;
diff --git a/qrenderdoc/Windows/ShaderMessageViewer.cpp b/qrenderdoc/Windows/ShaderMessageViewer.cpp
index fcee090e770..03d48a62887 100644
--- a/qrenderdoc/Windows/ShaderMessageViewer.cpp
+++ b/qrenderdoc/Windows/ShaderMessageViewer.cpp
@@ -266,6 +266,10 @@ ShaderMessageViewer::ShaderMessageViewer(ICaptureContext &ctx, ShaderStageMask s
           inputs.view = msg.location.pixel.view;
           trace = r->DebugPixel(msg.location.pixel.x, msg.location.pixel.y, inputs);
         }
+        else if(msg.stage == ShaderStage::Mesh)
+        {
+          trace = r->DebugMeshThread(msg.location.mesh.meshGroup, msg.location.mesh.thread);
+        }
 
         if(trace && trace->debugger == NULL)
         {
diff --git a/renderdoc/api/replay/renderdoc_replay.h b/renderdoc/api/replay/renderdoc_replay.h
index 7b0f854504d..88408fbad9f 100644
--- a/renderdoc/api/replay/renderdoc_replay.h
+++ b/renderdoc/api/replay/renderdoc_replay.h
@@ -1021,6 +1021,17 @@ bucket when the pixel values are divided between ``minval`` and ``maxval``.
   virtual ShaderDebugTrace *DebugThread(const rdcfixedarray<uint32_t, 3> &groupid,
                                         const rdcfixedarray<uint32_t, 3> &threadid) = 0;
 
+  DOCUMENT(R"(Retrieve a debugging trace from running a mesh shader.
+
+:param Tuple[int,int,int] groupid: A list containing the 3D workgroup index.
+:param Tuple[int,int,int] threadid: A list containing the 3D thread index within the workgroup.
+:return: The resulting trace resulting from debugging. Destroy with
+  :meth:`FreeTrace`.
+:rtype: ShaderDebugTrace
+)");
+  virtual ShaderDebugTrace *DebugMeshThread(const rdcfixedarray<uint32_t, 3> &groupid,
+                                               const rdcfixedarray<uint32_t, 3> &threadid) = 0;
+
   DOCUMENT(R"(Continue a shader's debugging with a given shader debugger instance. This will run an
 implementation defined number of steps and then return those steps in a list. This may be a fixed
 number of steps or it may run for a fixed length of time and return as many steps as can be
diff --git a/renderdoc/core/image_viewer.cpp b/renderdoc/core/image_viewer.cpp
index 1a8d7422d5f..9c0fdb0324b 100644
--- a/renderdoc/core/image_viewer.cpp
+++ b/renderdoc/core/image_viewer.cpp
@@ -357,6 +357,11 @@ class ImageViewer : public IReplayDriver
   {
     return new ShaderDebugTrace();
   }
+  ShaderDebugTrace *DebugMeshThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
+                                       const rdcfixedarray<uint32_t, 3> &threadid)
+  {
+    return new ShaderDebugTrace();
+  }
   rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger) { return {}; }
   void FreeDebugger(ShaderDebugger *debugger) { delete debugger; }
   void BuildTargetShader(ShaderEncoding sourceEncoding, const bytebuf &source, const rdcstr &entry,
diff --git a/renderdoc/core/replay_proxy.cpp b/renderdoc/core/replay_proxy.cpp
index bd7927bccde..bfd06441cff 100644
--- a/renderdoc/core/replay_proxy.cpp
+++ b/renderdoc/core/replay_proxy.cpp
@@ -1684,6 +1684,44 @@ ShaderDebugTrace *ReplayProxy::DebugThread(uint32_t eventId,
   PROXY_FUNCTION(DebugThread, eventId, groupid, threadid);
 }
 
+template <typename ParamSerialiser, typename ReturnSerialiser>
+ShaderDebugTrace *ReplayProxy::Proxied_DebugMeshThread(ParamSerialiser &paramser,
+                                                          ReturnSerialiser &retser, uint32_t eventId,
+                                                          const rdcfixedarray<uint32_t, 3> &groupid,
+                                                          const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  const ReplayProxyPacket expectedPacket = eReplayProxy_DebugMeshThread;
+  ReplayProxyPacket packet = eReplayProxy_DebugMeshThread;
+  ShaderDebugTrace *ret;
+
+  {
+    BEGIN_PARAMS();
+    SERIALISE_ELEMENT(eventId);
+    SERIALISE_ELEMENT(groupid);
+    SERIALISE_ELEMENT(threadid);
+    END_PARAMS();
+  }
+
+  {
+    REMOTE_EXECUTION();
+    if(paramser.IsReading() && !paramser.IsErrored() && !m_IsErrored)
+      ret = m_Remote->DebugMeshThread(eventId, groupid, threadid);
+    else
+      ret = new ShaderDebugTrace;
+  }
+
+  SERIALISE_RETURN(*ret);
+
+  return ret;
+}
+
+ShaderDebugTrace *ReplayProxy::DebugMeshThread(uint32_t eventId,
+                                                  const rdcfixedarray<uint32_t, 3> &groupid,
+                                                  const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  PROXY_FUNCTION(DebugMeshThread, eventId, groupid, threadid);
+}
+
 template <typename ParamSerialiser, typename ReturnSerialiser>
 rdcarray<ShaderDebugState> ReplayProxy::Proxied_ContinueDebug(ParamSerialiser &paramser,
                                                               ReturnSerialiser &retser,
diff --git a/renderdoc/core/replay_proxy.h b/renderdoc/core/replay_proxy.h
index 014227e8ad0..7f472586804 100644
--- a/renderdoc/core/replay_proxy.h
+++ b/renderdoc/core/replay_proxy.h
@@ -89,6 +89,7 @@ enum ReplayProxyPacket
   eReplayProxy_DebugVertex,
   eReplayProxy_DebugPixel,
   eReplayProxy_DebugThread,
+  eReplayProxy_DebugMeshThread,
 
   eReplayProxy_RenderOverlay,
 
@@ -552,6 +553,9 @@ class ReplayProxy : public IReplayDriver
   IMPLEMENT_FUNCTION_PROXIED(ShaderDebugTrace *, DebugThread, uint32_t eventId,
                              const rdcfixedarray<uint32_t, 3> &groupid,
                              const rdcfixedarray<uint32_t, 3> &threadid);
+  IMPLEMENT_FUNCTION_PROXIED(ShaderDebugTrace *, DebugMeshThread, uint32_t eventId,
+                             const rdcfixedarray<uint32_t, 3> &groupid,
+                             const rdcfixedarray<uint32_t, 3> &threadid);
   IMPLEMENT_FUNCTION_PROXIED(rdcarray<ShaderDebugState>, ContinueDebug, ShaderDebugger *debugger);
   IMPLEMENT_FUNCTION_PROXIED(void, FreeDebugger, ShaderDebugger *debugger);
 
diff --git a/renderdoc/driver/d3d11/d3d11_replay.h b/renderdoc/driver/d3d11/d3d11_replay.h
index dbb1f4058a5..d200c937aed 100644
--- a/renderdoc/driver/d3d11/d3d11_replay.h
+++ b/renderdoc/driver/d3d11/d3d11_replay.h
@@ -288,6 +288,8 @@ class D3D11Replay : public IReplayDriver
                                const DebugPixelInputs &inputs);
   ShaderDebugTrace *DebugThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
                                 const rdcfixedarray<uint32_t, 3> &threadid);
+  ShaderDebugTrace *DebugMeshThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
+                                       const rdcfixedarray<uint32_t, 3> &threadid);
   rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger);
   void FreeDebugger(ShaderDebugger *debugger);
 
diff --git a/renderdoc/driver/d3d11/d3d11_shaderdebug.cpp b/renderdoc/driver/d3d11/d3d11_shaderdebug.cpp
index 65a50c8629d..e3e44213260 100644
--- a/renderdoc/driver/d3d11/d3d11_shaderdebug.cpp
+++ b/renderdoc/driver/d3d11/d3d11_shaderdebug.cpp
@@ -2688,6 +2688,14 @@ ShaderDebugTrace *D3D11Replay::DebugThread(uint32_t eventId,
   return ret;
 }
 
+ShaderDebugTrace *D3D11Replay::DebugMeshThread(uint32_t eventId,
+                                                  const rdcfixedarray<uint32_t, 3> &groupid,
+                                                  const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  // Not supported
+  return new ShaderDebugTrace;
+}
+
 rdcarray<ShaderDebugState> D3D11Replay::ContinueDebug(ShaderDebugger *debugger)
 {
   DXBCDebug::InterpretDebugger *interpreter = (DXBCDebug::InterpretDebugger *)debugger;
diff --git a/renderdoc/driver/d3d12/d3d12_replay.h b/renderdoc/driver/d3d12/d3d12_replay.h
index 51175dfaa20..6c13a9b0878 100644
--- a/renderdoc/driver/d3d12/d3d12_replay.h
+++ b/renderdoc/driver/d3d12/d3d12_replay.h
@@ -251,6 +251,8 @@ class D3D12Replay : public IReplayDriver
                                const DebugPixelInputs &inputs);
   ShaderDebugTrace *DebugThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
                                 const rdcfixedarray<uint32_t, 3> &threadid);
+  ShaderDebugTrace *DebugMeshThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
+                                       const rdcfixedarray<uint32_t, 3> &threadid);
   rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger);
   void FreeDebugger(ShaderDebugger *debugger);
 
diff --git a/renderdoc/driver/d3d12/d3d12_shaderdebug.cpp b/renderdoc/driver/d3d12/d3d12_shaderdebug.cpp
index 2a4a4bb9f1b..9f73cce21a8 100644
--- a/renderdoc/driver/d3d12/d3d12_shaderdebug.cpp
+++ b/renderdoc/driver/d3d12/d3d12_shaderdebug.cpp
@@ -3377,6 +3377,14 @@ ShaderDebugTrace *D3D12Replay::DebugThread(uint32_t eventId,
   return ret;
 }
 
+ShaderDebugTrace *D3D12Replay::DebugMeshThread(uint32_t eventId,
+                                                  const rdcfixedarray<uint32_t, 3> &groupid,
+                                                  const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  // Not implemented yet
+  return new ShaderDebugTrace;
+}
+
 rdcarray<ShaderDebugState> D3D12Replay::ContinueDebug(ShaderDebugger *debugger)
 {
   if(!debugger)
diff --git a/renderdoc/driver/gl/gl_replay.cpp b/renderdoc/driver/gl/gl_replay.cpp
index 2c34e6d6a8c..fe4818fc353 100644
--- a/renderdoc/driver/gl/gl_replay.cpp
+++ b/renderdoc/driver/gl/gl_replay.cpp
@@ -4300,6 +4300,14 @@ ShaderDebugTrace *GLReplay::DebugThread(uint32_t eventId, const rdcfixedarray<ui
   return new ShaderDebugTrace();
 }
 
+ShaderDebugTrace *GLReplay::DebugMeshThread(uint32_t eventId,
+                                               const rdcfixedarray<uint32_t, 3> &groupid,
+                                               const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  GLNOTIMP("DebugMeshThread");
+  return new ShaderDebugTrace();
+}
+
 rdcarray<ShaderDebugState> GLReplay::ContinueDebug(ShaderDebugger *debugger)
 {
   GLNOTIMP("ContinueDebug");
diff --git a/renderdoc/driver/gl/gl_replay.h b/renderdoc/driver/gl/gl_replay.h
index 152f39f4066..56ffdde52c5 100644
--- a/renderdoc/driver/gl/gl_replay.h
+++ b/renderdoc/driver/gl/gl_replay.h
@@ -260,6 +260,8 @@ class GLReplay : public IReplayDriver
                                const DebugPixelInputs &inputs);
   ShaderDebugTrace *DebugThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
                                 const rdcfixedarray<uint32_t, 3> &threadid);
+  ShaderDebugTrace *DebugMeshThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
+                                       const rdcfixedarray<uint32_t, 3> &threadid);
   rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger);
   void FreeDebugger(ShaderDebugger *debugger);
   uint32_t PickVertex(uint32_t eventId, int32_t width, int32_t height, const MeshDisplay &cfg,
diff --git a/renderdoc/driver/shaders/spirv/spirv_debug.cpp b/renderdoc/driver/shaders/spirv/spirv_debug.cpp
index f446e53f082..c34667d4c4a 100644
--- a/renderdoc/driver/shaders/spirv/spirv_debug.cpp
+++ b/renderdoc/driver/shaders/spirv/spirv_debug.cpp
@@ -3521,6 +3521,12 @@ void ThreadState::StepNext(ShaderDebugState *state, const rdcarray<ThreadState>
       break;
     }
 
+    case Op::SetMeshOutputsEXT:
+    {
+      // Ignore mesh outputs, nothing to do, really
+      break;
+    }
+
     // TODO sparse sampling
     case Op::ImageSparseSampleImplicitLod:
     case Op::ImageSparseSampleExplicitLod:
@@ -3750,7 +3756,6 @@ void ThreadState::StepNext(ShaderDebugState *state, const rdcarray<ThreadState>
     case Op::ConvertSampledImageToUNV:
     case Op::SamplerImageAddressingModeNV:
     case Op::EmitMeshTasksEXT:
-    case Op::SetMeshOutputsEXT:
     case Op::HitObjectRecordHitMotionNV:
     case Op::HitObjectRecordHitWithIndexMotionNV:
     case Op::HitObjectRecordMissMotionNV:
diff --git a/renderdoc/driver/shaders/spirv/spirv_debug_setup.cpp b/renderdoc/driver/shaders/spirv/spirv_debug_setup.cpp
index 83b297d277f..4230c373e55 100644
--- a/renderdoc/driver/shaders/spirv/spirv_debug_setup.cpp
+++ b/renderdoc/driver/shaders/spirv/spirv_debug_setup.cpp
@@ -437,6 +437,7 @@ void Reflector::CheckDebuggable(bool &debuggable, rdcstr &debugStatus) const
       "SPV_GOOGLE_user_type",
       "SPV_KHR_physical_storage_buffer",
       "SPV_KHR_relaxed_extended_instruction",
+      "SPV_EXT_mesh_shader",
   };
 
   // whitelist supported extensions
@@ -554,6 +555,7 @@ void Reflector::CheckDebuggable(bool &debuggable, rdcstr &debugStatus) const
       case Capability::UniformDecoration:
       case Capability::SignedZeroInfNanPreserve:
       case Capability::PhysicalStorageBufferAddresses:
+      case Capability::MeshShadingEXT:
       {
         supported = true;
         break;
@@ -646,13 +648,6 @@ void Reflector::CheckDebuggable(bool &debuggable, rdcstr &debugStatus) const
         break;
       }
 
-      // mesh shading
-      case Capability::MeshShadingEXT:
-      {
-        supported = false;
-        break;
-      }
-
       // no plans to support these - mostly Kernel/OpenCL related or vendor extensions
       case Capability::Addresses:
       case Capability::Linkage:
diff --git a/renderdoc/driver/vulkan/vk_replay.h b/renderdoc/driver/vulkan/vk_replay.h
index 1433da516b5..e2c3d9f8d4f 100644
--- a/renderdoc/driver/vulkan/vk_replay.h
+++ b/renderdoc/driver/vulkan/vk_replay.h
@@ -451,6 +451,8 @@ class VulkanReplay : public IReplayDriver
                                const DebugPixelInputs &inputs);
   ShaderDebugTrace *DebugThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
                                 const rdcfixedarray<uint32_t, 3> &threadid);
+  ShaderDebugTrace *DebugMeshThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
+                                       const rdcfixedarray<uint32_t, 3> &threadid);
   rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger);
   void FreeDebugger(ShaderDebugger *debugger);
 
diff --git a/renderdoc/driver/vulkan/vk_shaderdebug.cpp b/renderdoc/driver/vulkan/vk_shaderdebug.cpp
index 01257a26967..73ecb433ac4 100644
--- a/renderdoc/driver/vulkan/vk_shaderdebug.cpp
+++ b/renderdoc/driver/vulkan/vk_shaderdebug.cpp
@@ -1526,7 +1526,15 @@ class VulkanAPIWrapper : public rdcspv::DebugAPIWrapper
     }
   };
 
+  struct TexelBufData
+  {
+    uint32_t texelSize = 0;
+    ResourceFormat fmt;
+    bytebuf bytes;
+  };
+
   std::map<ShaderBindIndex, ImageData> imageCache;
+  std::map<ShaderBindIndex, TexelBufData> texelBufCache;
 
   const Descriptor &GetDescriptor(const rdcstr &access, ShaderBindIndex index, bool &valid)
   {
@@ -1766,6 +1774,43 @@ class VulkanAPIWrapper : public rdcspv::DebugAPIWrapper
     return data;
   }
 
+  TexelBufData &PopulateTexelBuffer(ShaderBindIndex bind)
+  {
+    auto insertIt = texelBufCache.insert(std::make_pair(bind, TexelBufData()));
+    TexelBufData &data = insertIt.first->second;
+    if(insertIt.second)
+    {
+      bool valid = true;
+      const Descriptor &bufData = GetDescriptor("accessing texel buffer value", bind, valid);
+      if(valid)
+      {
+        // if the resources might be dirty from side-effects from the action, replay back to right
+        // before it.
+        if(m_ResourcesDirty)
+        {
+          VkMarkerRegion region("un-dirtying resources");
+          m_pDriver->ReplayLog(0, m_EventID, eReplay_WithoutDraw);
+          m_ResourcesDirty = false;
+        }
+
+        if(bufData.resource != ResourceId())
+        {
+          const VulkanCreationInfo::BufferView &viewProps =
+              m_Creation.m_BufferView[m_pDriver->GetResourceManager()->GetLiveID(bufData.view)];
+
+          data.fmt = MakeResourceFormat(viewProps.format);
+          data.texelSize = (uint32_t)GetByteSize(1, 1, 1, viewProps.format, 0);
+
+          m_pDriver->GetReplay()->GetBufferData(
+              m_pDriver->GetResourceManager()->GetLiveID(bufData.resource), bufData.byteOffset,
+              bufData.byteSize, data.bytes);
+        }
+      }
+    }
+
+    return data;
+  }
+
   VkPipeline MakePipe(const ShaderConstParameters &params, uint32_t floatBitSize, bool depthTex,
                       bool uintTex, bool sintTex)
   {
@@ -5080,6 +5125,87 @@ ShaderDebugTrace *VulkanReplay::DebugThread(uint32_t eventId,
   return ret;
 }
 
+ShaderDebugTrace *VulkanReplay::DebugMeshThread(uint32_t eventId,
+                                                   const rdcfixedarray<uint32_t, 3> &groupid,
+                                                   const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  const VulkanRenderState &state = m_pDriver->GetRenderState();
+  VulkanCreationInfo &c = m_pDriver->m_CreationInfo;
+
+  rdcstr regionName =
+      StringFormat::Fmt("DebugMeshThread @ %u of (%u,%u,%u) (%u,%u,%u)", eventId, groupid[0],
+                        groupid[1], groupid[2], threadid[0], threadid[1], threadid[2]);
+
+  VkMarkerRegion region(regionName);
+
+  if(Vulkan_Debug_ShaderDebugLogging())
+    RDCLOG("%s", regionName.c_str());
+
+  const ActionDescription *action = m_pDriver->GetAction(eventId);
+
+  if(!(action->flags & ActionFlags::MeshDispatch))
+  {
+    RDCLOG("No mesh dispatch selected");
+    return new ShaderDebugTrace();
+  }
+
+  // get ourselves in pristine state before this dispatch (without any side effects it may have had)
+  m_pDriver->ReplayLog(0, eventId, eReplay_WithoutDraw);
+
+  const VulkanCreationInfo::Pipeline &pipe = c.m_Pipeline[state.graphics.pipeline];
+  const VulkanCreationInfo::ShaderEntry &shaderEntry =
+      state.graphics.shaderObject ? c.m_ShaderObject[state.shaderObjects[(size_t)ShaderStage::Mesh]].shad : pipe.shaders[(size_t)ShaderStage::Mesh];
+  VulkanCreationInfo::ShaderModule &shader = c.m_ShaderModule[shaderEntry.module];
+  rdcstr entryPoint = shaderEntry.entryPoint;
+  const rdcarray<SpecConstant> &spec = shaderEntry.specialization;
+
+  VulkanCreationInfo::ShaderModuleReflection &shadRefl =
+      shader.GetReflection(ShaderStage::Mesh, entryPoint, state.graphics.pipeline);
+
+  if(!shadRefl.refl->debugInfo.debuggable)
+  {
+    RDCLOG("Shader is not debuggable: %s", shadRefl.refl->debugInfo.debugStatus.c_str());
+    return new ShaderDebugTrace();
+  }
+
+  shadRefl.PopulateDisassembly(shader.spirv);
+
+  VulkanAPIWrapper *apiWrapper =
+      new VulkanAPIWrapper(m_pDriver, c, ShaderStage::Mesh, eventId, shadRefl.refl->resourceId);
+
+  uint32_t threadDim[3];
+  threadDim[0] = shadRefl.refl->dispatchThreadsDimension[0];
+  threadDim[1] = shadRefl.refl->dispatchThreadsDimension[1];
+  threadDim[2] = shadRefl.refl->dispatchThreadsDimension[2];
+
+  std::map<ShaderBuiltin, ShaderVariable> &builtins = apiWrapper->builtin_inputs;
+  builtins[ShaderBuiltin::DispatchSize] =
+      ShaderVariable(rdcstr(), action->dispatchDimension[0], action->dispatchDimension[1],
+                     action->dispatchDimension[2], 0U);
+  builtins[ShaderBuiltin::DispatchThreadIndex] = ShaderVariable(
+      rdcstr(), groupid[0] * threadDim[0] + threadid[0], groupid[1] * threadDim[1] + threadid[1],
+      groupid[2] * threadDim[2] + threadid[2], 0U);
+  builtins[ShaderBuiltin::GroupIndex] =
+      ShaderVariable(rdcstr(), groupid[0], groupid[1], groupid[2], 0U);
+  builtins[ShaderBuiltin::GroupSize] =
+      ShaderVariable(rdcstr(), threadDim[0], threadDim[1], threadDim[2], 0U);
+  builtins[ShaderBuiltin::GroupThreadIndex] =
+      ShaderVariable(rdcstr(), threadid[0], threadid[1], threadid[2], 0U);
+  builtins[ShaderBuiltin::GroupFlatIndex] = ShaderVariable(
+      rdcstr(), threadid[2] * threadDim[0] * threadDim[1] + threadid[1] * threadDim[0] + threadid[0],
+      0U, 0U, 0U);
+  builtins[ShaderBuiltin::DeviceIndex] = ShaderVariable(rdcstr(), 0U, 0U, 0U, 0U);
+  builtins[ShaderBuiltin::DrawIndex] = ShaderVariable(rdcstr(), action->drawIndex, 0U, 0U, 0U);
+
+  rdcspv::Debugger *debugger = new rdcspv::Debugger;
+  debugger->Parse(shader.spirv.GetSPIRV());
+  ShaderDebugTrace *ret = debugger->BeginDebug(apiWrapper, ShaderStage::Mesh, entryPoint, spec,
+                                               shadRefl.instructionLines, shadRefl.patchData, 0);
+  apiWrapper->ResetReplay();
+
+  return ret;
+}
+
 rdcarray<ShaderDebugState> VulkanReplay::ContinueDebug(ShaderDebugger *debugger)
 {
   rdcspv::Debugger *spvDebugger = (rdcspv::Debugger *)debugger;
diff --git a/renderdoc/replay/dummy_driver.cpp b/renderdoc/replay/dummy_driver.cpp
index d6202ad8937..ca8484e306c 100644
--- a/renderdoc/replay/dummy_driver.cpp
+++ b/renderdoc/replay/dummy_driver.cpp
@@ -296,6 +296,13 @@ ShaderDebugTrace *DummyDriver::DebugVertex(uint32_t eventId, uint32_t vertid, ui
   return new ShaderDebugTrace;
 }
 
+ShaderDebugTrace *DummyDriver::DebugMeshThread(uint32_t eventId,
+                                                  const rdcfixedarray<uint32_t, 3> &groupid,
+                                                  const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  return new ShaderDebugTrace;
+}
+
 ShaderDebugTrace *DummyDriver::DebugPixel(uint32_t eventId, uint32_t x, uint32_t y,
                                           const DebugPixelInputs &inputs)
 {
diff --git a/renderdoc/replay/dummy_driver.h b/renderdoc/replay/dummy_driver.h
index 2ec6c3b1bd0..68f763f9db7 100644
--- a/renderdoc/replay/dummy_driver.h
+++ b/renderdoc/replay/dummy_driver.h
@@ -114,6 +114,8 @@ class DummyDriver : public IReplayDriver
                                const DebugPixelInputs &inputs);
   ShaderDebugTrace *DebugThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
                                 const rdcfixedarray<uint32_t, 3> &threadid);
+  ShaderDebugTrace *DebugMeshThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
+                                       const rdcfixedarray<uint32_t, 3> &threadid);
   rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger);
   void FreeDebugger(ShaderDebugger *debugger);
 
diff --git a/renderdoc/replay/replay_controller.cpp b/renderdoc/replay/replay_controller.cpp
index 4f7b3a3a973..6db8dc9234c 100644
--- a/renderdoc/replay/replay_controller.cpp
+++ b/renderdoc/replay/replay_controller.cpp
@@ -1698,6 +1698,24 @@ ShaderDebugTrace *ReplayController::DebugThread(const rdcfixedarray<uint32_t, 3>
   return ret;
 }
 
+ShaderDebugTrace *ReplayController::DebugMeshThread(const rdcfixedarray<uint32_t, 3> &groupid,
+                                                       const rdcfixedarray<uint32_t, 3> &threadid)
+{
+  CHECK_REPLAY_THREAD();
+
+  RENDERDOC_PROFILEFUNCTION();
+
+  ShaderDebugTrace *ret = m_pDevice->DebugMeshThread(m_EventID, groupid, threadid);
+  FatalErrorCheck();
+
+  SetFrameEvent(m_EventID, true);
+
+  if(ret->debugger)
+    m_Debuggers.push_back(ret->debugger);
+
+  return ret;
+}
+
 rdcarray<ShaderDebugState> ReplayController::ContinueDebug(ShaderDebugger *debugger)
 {
   CHECK_REPLAY_THREAD();
diff --git a/renderdoc/replay/replay_controller.h b/renderdoc/replay/replay_controller.h
index fbd15b4681e..22243e95267 100644
--- a/renderdoc/replay/replay_controller.h
+++ b/renderdoc/replay/replay_controller.h
@@ -214,6 +214,8 @@ struct ReplayController : public IReplayController
   ShaderDebugTrace *DebugPixel(uint32_t x, uint32_t y, const DebugPixelInputs &inputs);
   ShaderDebugTrace *DebugThread(const rdcfixedarray<uint32_t, 3> &groupid,
                                 const rdcfixedarray<uint32_t, 3> &threadid);
+  ShaderDebugTrace *DebugMeshThread(const rdcfixedarray<uint32_t, 3> &groupid,
+                                       const rdcfixedarray<uint32_t, 3> &threadid);
   rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger);
   void FreeTrace(ShaderDebugTrace *trace);
 
diff --git a/renderdoc/replay/replay_driver.h b/renderdoc/replay/replay_driver.h
index 1e7b7f7fab1..c87c1f68fb4 100644
--- a/renderdoc/replay/replay_driver.h
+++ b/renderdoc/replay/replay_driver.h
@@ -231,6 +231,9 @@ class IRemoteDriver
                                        const DebugPixelInputs &inputs) = 0;
   virtual ShaderDebugTrace *DebugThread(uint32_t eventId, const rdcfixedarray<uint32_t, 3> &groupid,
                                         const rdcfixedarray<uint32_t, 3> &threadid) = 0;
+  virtual ShaderDebugTrace *DebugMeshThread(uint32_t eventId,
+                                               const rdcfixedarray<uint32_t, 3> &groupid,
+                                               const rdcfixedarray<uint32_t, 3> &threadid) = 0;
   virtual rdcarray<ShaderDebugState> ContinueDebug(ShaderDebugger *debugger) = 0;
   virtual void FreeDebugger(ShaderDebugger *debugger) = 0;