[host] dxgi: d3d12 now writes direclty into ivshmem

This is still incomplete as d3d12 doesn't have any proper sync with the
captured frame and as such is still not suggested for general usage. This
change though is monumental for this project as it removes a full memory
copy reducing bandwidth consumption enormously.

host/include/interface/capture.h

@@ -116,17 +116,18 @@ typedef struct CaptureInterface
   void         (*initOptions    )(void);
 
   bool(*create)(
+    void                   * ivshmemBase,
     CaptureGetPointerBuffer  getPointerBufferFn,
     CapturePostPointerBuffer postPointerBufferFn
   );
 
-  bool          (*init         )(void);
+  bool          (*init         )(unsigned * alignSize);
   bool          (*start        )(void);
   void          (*stop         )(void);
   bool          (*deinit       )(void);
   void          (*free         )(void);
 
-  CaptureResult (*capture   )(void);
+  CaptureResult (*capture   )(unsigned frameBufferIndex, FrameBuffer * frame);
   CaptureResult (*waitFrame )(CaptureFrame * frame, const size_t maxFrameSize);
   CaptureResult (*getFrame  )(FrameBuffer  * frame, int frameIndex);
 }

host/platform/Windows/capture/DXGI/CMakeLists.txt

@@ -14,7 +14,7 @@ add_library(capture_DXGI STATIC
   src/pp/rgb24.c
 )
 
-add_definitions("-DCOBJMACROS -DINITGUID")
+add_definitions("-DCOBJMACROS -DINITGUID -DWIDL_C_INLINE_WRAPPERS")
 
 target_link_libraries(capture_DXGI
   lg_common

host/platform/Windows/capture/DXGI/src/backend.h

@@ -36,7 +36,11 @@ struct DXGICopyBackend
   const char * code;
 
   // create the copy backend
-  bool (*create)(unsigned textures);
+  bool (*create)(
+    void * ivshmemBase,
+    unsigned * alignSize,
+    unsigned frameBuffers,
+    unsigned textures);
 
   // configure the copy backend with the specified format
   bool (*configure)(
@@ -50,7 +54,10 @@ struct DXGICopyBackend
   void (*free)(void);
 
   // called just before the copy starts
-  bool (*preCopy)(ID3D11Texture2D * src, unsigned textureIndex);
+  bool (*preCopy)(ID3D11Texture2D * src,
+    unsigned textureIndex,
+    unsigned frameBufferIndex,
+    FrameBuffer * frameBuffer);
 
   // called to copy the full frame
   bool (*copyFull)(ID3D11Texture2D * src, unsigned textureIndex);
@@ -65,6 +72,9 @@ struct DXGICopyBackend
   // maps the copied frame into memory
   CaptureResult (*mapTexture)(unsigned textureIndex, void ** map);
 
+  // [optional] backend specific write into the FrameBuffer
+  CaptureResult (*writeFrame)(int textureIndex, FrameBuffer * frame);
+
   // unmaps the copied frame from memory
   void (*unmapTexture)(unsigned textureIndex);
 

host/platform/Windows/capture/DXGI/src/d3d11.c

@@ -43,7 +43,11 @@ struct D3D11Backend
 
 static struct D3D11Backend  * this = NULL;
 
-static bool d3d11_create(unsigned textures)
+static bool d3d11_create(
+  void     * ivshmemBase,
+  unsigned * alignSize,
+  unsigned   frameBuffers,
+  unsigned   textures)
 {
   DEBUG_ASSERT(!this);
   this = calloc(1,
@@ -128,7 +132,11 @@ static void d3d11_free(void)
   this = NULL;
 }
 
-static bool d3d11_preCopy(ID3D11Texture2D * src, unsigned textureIndex)
+static bool d3d11_preCopy(
+  ID3D11Texture2D * src,
+  unsigned          textureIndex,
+  unsigned          frameBufferIndex,
+  FrameBuffer     * frameBuffer)
 {
   dxgi_contextLock();
   this->texture[textureIndex].copyTime = microtime();

host/platform/Windows/capture/DXGI/src/d3d12.c

@@ -35,7 +35,6 @@
 
 struct D3D12Texture
 {
-  ID3D12Resource            ** tex;
   ID3D12CommandAllocator    ** commandAllocator;
   ID3D12CommandList         ** commandList;
   ID3D12GraphicsCommandList ** graphicsCommandList;
@@ -58,6 +57,11 @@ struct D3D12Backend
 
   float                  copySleep;
   ID3D12Device        ** device;
+  ID3D12Device3       ** device3;
+  void                 * ivshmemBase;
+  ID3D12Heap          ** heap;
+  ID3D12Resource     *** frameBuffers;
+  ID3D12Resource      ** dstResource;
   ID3D12CommandQueue  ** commandQueue;
   UINT64                 fenceValue;
   ID3D12Fence         ** fence;
@@ -88,7 +92,11 @@ typedef HRESULT (*D3D12GetDebugInterface_t)(
 
 static void d3d12_free(void);
 
-static bool d3d12_create(unsigned textures)
+static bool d3d12_create(
+  void     * ivshmemBase,
+  unsigned * alignSize,
+  unsigned   frameBuffers,
+  unsigned   textures)
 {
   DEBUG_ASSERT(!this);
   comRef_scopePush();
@@ -109,6 +117,13 @@ static bool d3d12_create(unsigned textures)
   if (!this->d3d12)
     goto exit;
 
+  this->frameBuffers = calloc(frameBuffers, sizeof(*this->frameBuffers));
+  if (!this->frameBuffers)
+  {
+    DEBUG_ERROR("Failed to allocate the ID3D12Resource frame buffer array");
+    goto exit;
+  }
+
   if (dxgi_debug())
   {
     D3D12GetDebugInterface_t D3D12GetDebugInterface = (D3D12GetDebugInterface_t)
@@ -145,6 +160,33 @@ static bool d3d12_create(unsigned textures)
     goto exit;
   }
 
+  comRef_defineLocal(ID3D12Device3, device3);
+  status = ID3D12Device_QueryInterface(
+    *device, &IID_ID3D12Device3, (void **)device3);
+  if (FAILED(status))
+  {
+    DEBUG_WINERROR("Failed to obtain the ID3D12Device3 interface", status);
+    goto exit;
+  }
+
+  comRef_defineLocal(ID3D12Heap, heap);
+  status = ID3D12Device3_OpenExistingHeapFromAddress(
+    *device3,
+    ivshmemBase,
+    &IID_ID3D12Heap,
+    (void **)heap);
+
+  if (FAILED(status))
+  {
+    DEBUG_WINERROR("Failed to open the framebuffer as a heap", status);
+    return false;
+  }
+
+  D3D12_HEAP_DESC heapDesc = ID3D12Heap_GetDesc(*heap);
+  DEBUG_INFO("ID3D12Heap        : Size:%I64u Alignment:%I64u",
+    heapDesc.SizeInBytes, heapDesc.Alignment);
+  *alignSize = heapDesc.Alignment;
+
   D3D12_COMMAND_QUEUE_DESC queueDesc =
   {
     .Type     = D3D12_COMMAND_LIST_TYPE_COPY,
@@ -186,7 +228,11 @@ static bool d3d12_create(unsigned textures)
     goto exit;
   }
 
+  this->ivshmemBase = ivshmemBase;
+
   comRef_toGlobal(this->device      , device      );
+  comRef_toGlobal(this->device3     , device3     );
+  comRef_toGlobal(this->heap        , heap        );
   comRef_toGlobal(this->commandQueue, commandQueue);
   comRef_toGlobal(this->fence       , fence       );
 
@@ -213,48 +259,12 @@ static bool d3d12_configure(
   this->format = format;
   this->pitch  = ALIGN_TO(width * bpp, D3D12_TEXTURE_DATA_PITCH_ALIGNMENT);
 
-  D3D12_HEAP_PROPERTIES readbackHeapProperties =
-  {
-    .Type = D3D12_HEAP_TYPE_READBACK,
-  };
-
-  D3D12_RESOURCE_DESC texDesc =
-  {
-    .Dimension          = D3D12_RESOURCE_DIMENSION_BUFFER,
-    .Alignment          = D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT,
-    .Width              = this->pitch * height,
-    .Height             = 1,
-    .DepthOrArraySize   = 1,
-    .MipLevels          = 1,
-    .Format             = DXGI_FORMAT_UNKNOWN,
-    .SampleDesc.Count   = 1,
-    .SampleDesc.Quality = 0,
-    .Layout             = D3D12_TEXTURE_LAYOUT_ROW_MAJOR,
-    .Flags              = D3D12_RESOURCE_FLAG_NONE
-  };
-
-  comRef_defineLocal(ID3D12Resource           , texture            );
   comRef_defineLocal(ID3D12Fence              , fence              );
   comRef_defineLocal(ID3D12CommandAllocator   , commandAllocator   );
   comRef_defineLocal(ID3D12GraphicsCommandList, graphicsCommandList);
   comRef_defineLocal(ID3D12CommandList        , commandList        );
   for (int i = 0; i < this->textures; ++i)
   {
-    status = ID3D12Device_CreateCommittedResource(*this->device,
-        &readbackHeapProperties,
-        D3D12_HEAP_FLAG_NONE,
-        &texDesc,
-        D3D12_RESOURCE_STATE_COPY_DEST,
-        NULL,
-        &IID_ID3D12Resource,
-        (void **)texture);
-
-    if (FAILED(status))
-    {
-      DEBUG_WINERROR("Failed to create texture", status);
-      goto exit;
-    }
-
     this->texture[i].event = CreateEvent(NULL, FALSE, FALSE, NULL);
     if (!this->texture[i].event)
     {
@@ -307,7 +317,6 @@ static bool d3d12_configure(
       goto exit;
     }
 
-    comRef_toGlobal(this->texture[i].tex                , texture            );
     comRef_toGlobal(this->texture[i].fence              , fence              );
     comRef_toGlobal(this->texture[i].commandAllocator   , commandAllocator   );
     comRef_toGlobal(this->texture[i].graphicsCommandList, graphicsCommandList);
@@ -320,6 +329,7 @@ static bool d3d12_configure(
 
 exit:
   comRef_scopePop();
+
   return result;
 }
 
@@ -335,6 +345,8 @@ static void d3d12_free(void)
   if (this->event)
     CloseHandle(this->event);
 
+  free(this->frameBuffers);
+
   if (this->d3d12)
     FreeLibrary(this->d3d12);
 
@@ -342,7 +354,11 @@ static void d3d12_free(void)
   this = NULL;
 }
 
-static bool d3d12_preCopy(ID3D11Texture2D * src, unsigned textureIndex)
+static bool d3d12_preCopy(
+  ID3D11Texture2D * src,
+  unsigned          textureIndex,
+  unsigned          frameBufferIndex,
+  FrameBuffer     * frameBuffer)
 {
   comRef_scopePush();
   bool result = false;
@@ -352,9 +368,6 @@ static bool d3d12_preCopy(ID3D11Texture2D * src, unsigned textureIndex)
   ID3D11DeviceContext_Flush(dxgi_getContext());
   dxgi_contextUnlock();
 
-  if (this->copySleep > 0)
-    nsleep((uint64_t)(this->copySleep * 1000000));
-
   this->d12src = NULL;
   if (this->sharedCacheCount > -1)
   {
@@ -363,8 +376,7 @@ static bool d3d12_preCopy(ID3D11Texture2D * src, unsigned textureIndex)
       if (this->sharedCache[i].tex == src)
       {
         this->d12src = *this->sharedCache[i].d12src;
-        result = true;
+        goto framebuffer;
-        goto exit;
       }
   }
 
@@ -419,8 +431,52 @@ static bool d3d12_preCopy(ID3D11Texture2D * src, unsigned textureIndex)
     this->sharedCacheCount = -1;
   }
 
+framebuffer:
+  if (this->frameBuffers[frameBufferIndex])
+    goto done;
+
+  // prepare a IVSHMEM backed texture for the provided framebuffer
+  D3D12_RESOURCE_DESC desc =
+  {
+    .Dimension          = D3D12_RESOURCE_DIMENSION_BUFFER,
+    .Alignment          = D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT,
+    .Width              = this->height * this->pitch,
+    .Height             = 1,
+    .DepthOrArraySize   = 1,
+    .MipLevels          = 1,
+    .Format             = DXGI_FORMAT_UNKNOWN,
+    .SampleDesc.Count   = 1,
+    .SampleDesc.Quality = 0,
+    .Layout             = D3D12_TEXTURE_LAYOUT_ROW_MAJOR,
+    .Flags              = D3D12_RESOURCE_FLAG_ALLOW_CROSS_ADAPTER
+  };
+
+  comRef_defineLocal(ID3D12Resource, resource);
+  status = ID3D12Device3_CreatePlacedResource(
+    *this->device3,
+    *this->heap,
+    (uintptr_t)framebuffer_get_data(frameBuffer) - (uintptr_t)this->ivshmemBase,
+    &desc,
+    D3D12_RESOURCE_STATE_COPY_DEST,
+    NULL,
+    &IID_ID3D12Resource,
+    (void**)resource);
+
+  if (FAILED(status))
+  {
+    DEBUG_WINERROR("Failed to create D3D12 resource from heap", status);
+    goto exit;
+  }
+
+  comRef_toGlobal(this->frameBuffers[frameBufferIndex], resource);
+
+done:
+  this->dstResource = this->frameBuffers[frameBufferIndex];
   result = true;
 
+  if (this->copySleep > 0)
+    nsleep((uint64_t)(this->copySleep * 1000000));
+
 exit:
   if (!result)
     comRef_scopePop();
@@ -441,7 +497,7 @@ static bool d3d12_copyFull(ID3D11Texture2D * src, unsigned textureIndex)
 
   D3D12_TEXTURE_COPY_LOCATION destLoc =
   {
-    .pResource       = *tex->tex,
+    .pResource       = *this->dstResource,
     .Type            = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT,
     .PlacedFootprint =
     {
@@ -463,7 +519,6 @@ static bool d3d12_copyFull(ID3D11Texture2D * src, unsigned textureIndex)
   return true;
 }
 
-
 static bool d3d12_copyRect(ID3D11Texture2D * src, unsigned textureIndex,
   FrameDamageRect * rect)
 {
@@ -478,7 +533,7 @@ static bool d3d12_copyRect(ID3D11Texture2D * src, unsigned textureIndex,
 
   D3D12_TEXTURE_COPY_LOCATION destLoc =
   {
-    .pResource       = *tex->tex,
+    .pResource       = *this->dstResource,
     .Type            = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT,
     .PlacedFootprint =
     {
@@ -561,7 +616,8 @@ static bool d3d12_postCopy(ID3D11Texture2D * src, unsigned textureIndex)
   result = true;
 
 exit:
-  comRef_scopePop(); //push is in pre-copy
+  //push is in preCopy
+  comRef_scopePop();
   return result;
 }
 
@@ -587,32 +643,17 @@ static CaptureResult d3d12_mapTexture(unsigned textureIndex, void ** map)
     return CAPTURE_RESULT_ERROR;
   }
 
-  D3D12_RANGE range =
+  return CAPTURE_RESULT_OK;
-  {
-    .Begin = 0,
-    .End   = this->pitch * this->height
-  };
-  status = ID3D12Resource_Map(*tex->tex, 0, &range, map);
-
-  if (FAILED(status))
-  {
-    DEBUG_WINERROR("Failed to map the texture", status);
-    return CAPTURE_RESULT_ERROR;
 }
 
+static CaptureResult d3d12_writeFrame(int textureIndex, FrameBuffer * frame)
+{
+  framebuffer_set_write_ptr(frame, this->height * this->pitch);
   return CAPTURE_RESULT_OK;
 }
 
 static void d3d12_unmapTexture(unsigned textureIndex)
 {
-  struct D3D12Texture * tex = &this->texture[textureIndex];
-
-  D3D12_RANGE range =
-  {
-    .Begin = 0,
-    .End   = 0
-  };
-  ID3D12Resource_Unmap(*tex->tex, 0, &range);
 }
 
 static void d3d12_preRelease(void)
@@ -632,6 +673,7 @@ struct DXGICopyBackend copyBackendD3D12 =
   .copyRect     = d3d12_copyRect,
   .postCopy     = d3d12_postCopy,
   .mapTexture   = d3d12_mapTexture,
+  .writeFrame   = d3d12_writeFrame,
   .unmapTexture = d3d12_unmapTexture,
   .preRelease   = d3d12_preRelease,
 };

host/platform/Windows/capture/DXGI/src/dxgi.c

@@ -125,6 +125,7 @@ struct DXGIInterface
   D3D_FEATURE_LEVEL          featureLevel;
   IDXGIOutputDuplication  ** dup;
   int                        maxTextures;
+  void                     * ivshmemBase;
   Texture                  * texture;
   int                        texRIndex;
   int                        texWIndex;
@@ -275,7 +276,10 @@ static void dxgi_initOptions(void)
   ppEarlyInit();
 }
 
-static bool dxgi_create(CaptureGetPointerBuffer getPointerBufferFn, CapturePostPointerBuffer postPointerBufferFn)
+static bool dxgi_create(
+  void                   * ivshmemBase,
+  CaptureGetPointerBuffer  getPointerBufferFn,
+  CapturePostPointerBuffer postPointerBufferFn)
 {
   DEBUG_ASSERT(!this);
   this = calloc(1, sizeof(*this));
@@ -302,6 +306,7 @@ static bool dxgi_create(CaptureGetPointerBuffer getPointerBufferFn, CapturePostP
   this->allowRGB24          = option_get_bool("dxgi", "allowRGB24");
   this->dwmFlush            = option_get_bool("dxgi", "dwmFlush");
   this->disableDamage       = option_get_bool("dxgi", "disableDamage");
+  this->ivshmemBase         = ivshmemBase;
   this->texture             = calloc(this->maxTextures, sizeof(*this->texture));
   this->getPointerBufferFn  = getPointerBufferFn;
   this->postPointerBufferFn = postPointerBufferFn;
@@ -360,7 +365,7 @@ static bool initVertexShader(void)
   return true;
 }
 
-static bool dxgi_init(void)
+static bool dxgi_init(unsigned * alignSize)
 {
   DEBUG_ASSERT(this);
 
@@ -800,7 +805,11 @@ static bool dxgi_init(void)
   {
     if (!strcasecmp(copyBackend, backends[i]->code))
     {
-      if (!backends[i]->create(this->maxTextures))
+      if (!backends[i]->create(
+        this->ivshmemBase,
+        alignSize,
+        LGMP_Q_FRAME_LEN,
+        this->maxTextures))
       {
         DEBUG_ERROR("Failed to initialize selected capture backend: %s", backends[i]->name);
         backends[i]->free();
@@ -1052,7 +1061,8 @@ static void computeTexDamage(Texture * tex)
   }
 }
 
-static CaptureResult dxgi_capture(void)
+static CaptureResult dxgi_capture(unsigned frameBufferIndex,
+  FrameBuffer * frameBuffer)
 {
   DEBUG_ASSERT(this);
   DEBUG_ASSERT(this->initialized);
@@ -1101,6 +1111,7 @@ static CaptureResult dxgi_capture(void)
   {
     if (result == CAPTURE_RESULT_ERROR)
       DEBUG_WINERROR("AcquireNextFrame failed", status);
+
     goto exit;
   }
 
@@ -1230,7 +1241,8 @@ static CaptureResult dxgi_capture(void)
         computeFrameDamage(tex);
       computeTexDamage(tex);
 
-      if (!this->backend->preCopy(dst, this->texWIndex))
+      if (!this->backend->preCopy(dst, this->texWIndex,
+        frameBufferIndex, frameBuffer))
       {
         result = CAPTURE_RESULT_ERROR;
         goto exit;
@@ -1440,6 +1452,14 @@ static CaptureResult dxgi_getFrame(FrameBuffer * frame, int frameIndex)
   Texture     * tex    = &this->texture[this->texRIndex];
   FrameDamage * damage = &this->frameDamage[frameIndex];
 
+  if (this->backend->writeFrame)
+  {
+    CaptureResult result = this->backend->writeFrame(frameIndex, frame);
+    if (result != CAPTURE_RESULT_OK)
+      return result;
+  }
+  else
+  {
     if (tex->damageRectsCount                 == 0 ||
         damage->count                          < 0 ||
         damage->count + tex->damageRectsCount  > KVMFR_MAX_DAMAGE_RECTS)
@@ -1475,6 +1495,7 @@ static CaptureResult dxgi_getFrame(FrameBuffer * frame, int frameIndex)
           this->pitch, this->dataHeight, tex->map, this->pitch);
       }
     }
+  }
 
   for (int i = 0; i < LGMP_Q_FRAME_LEN; ++i)
   {

host/platform/Windows/capture/NVFBC/src/nvfbc.c

@@ -185,6 +185,7 @@ static void nvfbc_initOptions(void)
 }
 
 static bool nvfbc_create(
+  void                   * ivshmemBase,
   CaptureGetPointerBuffer  getPointerBufferFn,
   CapturePostPointerBuffer postPointerBufferFn)
 {
@@ -220,7 +221,7 @@ static void updateScale(void)
   this->targetHeight = this->height;
 }
 
-static bool nvfbc_init(void)
+static bool nvfbc_init(unsigned * alignSize)
 {
   int adapterIndex = option_get_int("nvfbc", "adapterIndex");
 
@@ -447,7 +448,8 @@ static void nvfbc_free(void)
   NvFBCFree();
 }
 
-static CaptureResult nvfbc_capture(void)
+static CaptureResult nvfbc_capture(unsigned frameBufferIndex,
+  FrameBuffer * frameBuffer)
 {
   // this is a bit of a hack as it causes this thread to block until the next
   // present keeping us locked with the refresh rate of the monitor being

host/src/app.c

@@ -89,10 +89,13 @@ struct app
   unsigned int   pointerIndex;
   unsigned int   pointerShapeIndex;
 
-  long           pageSize;
+  unsigned       alignSize;
   size_t         maxFrameSize;
   PLGMPHostQueue frameQueue;
   PLGMPMemory    frameMemory[LGMP_Q_FRAME_LEN];
+  KVMFRFrame   * frame      [LGMP_Q_FRAME_LEN];
+  FrameBuffer  * frameBuffer[LGMP_Q_FRAME_LEN];
+
   unsigned int   frameIndex;
   bool           frameValid;
   uint32_t       frameSerial;
@@ -243,7 +246,7 @@ static bool sendFrame(void)
   if (++app.frameIndex == LGMP_Q_FRAME_LEN)
     app.frameIndex = 0;
 
-  KVMFRFrame * fi = lgmpHostMemPtr(app.frameMemory[app.frameIndex]);
+  KVMFRFrame * fi = app.frame[app.frameIndex];
   KVMFRFrameFlags flags =
     (frame.hdr   ? FRAME_FLAG_HDR    : 0) |
     (frame.hdrPQ ? FRAME_FLAG_HDR_PQ : 0);
@@ -311,7 +314,7 @@ static bool sendFrame(void)
   fi->frameHeight       = frame.frameHeight;
   fi->stride            = frame.stride;
   fi->pitch             = frame.pitch;
-  fi->offset            = app.pageSize - sizeof(FrameBuffer);
+  // fi->offset is initialized at startup
   fi->flags             = flags;
   fi->damageRectsCount  = frame.damageRectsCount;
   memcpy(fi->damageRects, frame.damageRects,
@@ -319,10 +322,7 @@ static bool sendFrame(void)
 
   app.frameValid = true;
 
-  // put the framebuffer on the border of the next page
+  framebuffer_prepare(app.frameBuffer[app.frameIndex]);
-  // this is to allow for aligned DMA transfers by the receiver
-  FrameBuffer * fb = (FrameBuffer *)(((uint8_t*)fi) + fi->offset);
-  framebuffer_prepare(fb);
 
   /* we post and then get the frame, this is intentional! */
   if ((status = lgmpHostQueuePost(app.frameQueue, 0,
@@ -332,7 +332,7 @@ static bool sendFrame(void)
     return true;
   }
 
-  app.iface->getFrame(fb, app.frameIndex);
+  app.iface->getFrame(app.frameBuffer[app.frameIndex], app.frameIndex);
   return true;
 }
 
@@ -392,13 +392,13 @@ static bool captureStart(void)
 {
   if (app.state == APP_STATE_IDLE)
   {
-    if (!app.iface->init())
+    if (!app.iface->init(&app.alignSize))
     {
       DEBUG_ERROR("Failed to initialize the capture device");
       return false;
     }
 
-    if (app.iface->start && !app.iface->start())
+    if (app.iface->start && app.iface->start())
     {
       DEBUG_ERROR("Failed to start the capture device");
       return false;
@@ -720,18 +720,26 @@ static bool lgmpSetup(struct IVSHMEM * shmDev)
   }
 
   app.maxFrameSize = lgmpHostMemAvail(app.lgmp);
-  app.maxFrameSize = (app.maxFrameSize - (app.pageSize - 1)) & ~(app.pageSize - 1);
+  app.maxFrameSize = (app.maxFrameSize - (app.alignSize - 1)) & ~(app.alignSize - 1);
   app.maxFrameSize /= LGMP_Q_FRAME_LEN;
   DEBUG_INFO("Max Frame Size   : %u MiB", (unsigned int)(app.maxFrameSize / 1048576LL));
 
   for(int i = 0; i < LGMP_Q_FRAME_LEN; ++i)
   {
     if ((status = lgmpHostMemAllocAligned(app.lgmp, app.maxFrameSize,
-            app.pageSize, &app.frameMemory[i])) != LGMP_OK)
+            app.alignSize, &app.frameMemory[i])) != LGMP_OK)
     {
       DEBUG_ERROR("lgmpHostMemAlloc Failed (Frame): %s", lgmpStatusString(status));
       goto fail_lgmp;
     }
+
+    app.frame[i] = lgmpHostMemPtr(app.frameMemory[i]);
+
+    /* put the framebuffer on the border of the next page, this is to allow for
+       aligned DMA transfers by the receiver */
+    const unsigned alignOffset = app.alignSize - sizeof(FrameBuffer);
+    app.frame[i]->offset = alignOffset;
+    app.frameBuffer[i] = (FrameBuffer *)(((uint8_t*)app.frame[i]) + alignOffset);
   }
 
   if (!lgCreateTimer(10, lgmpTimer, NULL, &app.lgmpTimer))
@@ -828,7 +836,7 @@ int app_main(int argc, char * argv[])
   DEBUG_INFO("Max Pointer Size : %u KiB", (unsigned int)MAX_POINTER_SIZE / 1024);
   DEBUG_INFO("KVMFR Version    : %u", KVMFR_VERSION);
 
-  app.pageSize          = sysinfo_getPageSize();
+  app.alignSize         = sysinfo_getPageSize();
   app.frameValid        = false;
   app.pointerShapeValid = false;
 
@@ -847,13 +855,16 @@ int app_main(int argc, char * argv[])
 
       DEBUG_INFO("Trying           : %s", iface->getName());
 
-      if (!iface->create(captureGetPointerBuffer, capturePostPointerBuffer))
+      if (!iface->create(
+        shmDev.mem,
+        captureGetPointerBuffer,
+        capturePostPointerBuffer))
       {
         iface = NULL;
         continue;
       }
 
-      if (iface->init())
+      if (iface->init(&app.alignSize))
         break;
 
       iface->free();
@@ -953,7 +964,14 @@ int app_main(int argc, char * argv[])
       }
 
       const uint64_t captureStart = microtime();
-      const CaptureResult result = app.iface->capture();
+
+      unsigned nextIndex = app.frameIndex + 1;
+      if (nextIndex == LGMP_Q_FRAME_LEN)
+        nextIndex = 0;
+
+      const CaptureResult result = app.iface->capture(
+        nextIndex, app.frameBuffer[nextIndex]);
+
       if (likely(result == CAPTURE_RESULT_OK))
         previousFrameTime = captureStart;
       else if (likely(result == CAPTURE_RESULT_TIMEOUT))