VK/WGPU: drain a view's blits and uniform cache before its dispatches. VK: block on the swapchain image of window framebuffers on a headless device (#3989)

src/renderer_vk.cpp

@@ -3476,8 +3476,12 @@ VK_IMPORT_DEVICE
 						);
 				}
 
+				const bool block = !isValid(_fbh)
+					|| NULL == m_backBuffer.m_swapChain.m_nwh
+					;
+
 				int64_t start = bx::getHPCounter();
-				newFrameBuffer.acquire(m_commandBuffer, !isValid(_fbh) );
+				newFrameBuffer.acquire(m_commandBuffer, block);
 				m_presentElapsed += bx::getHPCounter() - start;
 			}
 
@@ -10778,6 +10782,12 @@ VK_DESTROY
 						}
 					}
 
+					if (viewChanged)
+					{
+						submitUniformCache(ucs, view);
+						submitBlit(bs, view);
+					}
+
 					// renderpass external subpass dependencies handle graphics -> compute and compute -> graphics
 					// but not compute -> compute (possibly also across views if they contain no draw calls)
 					setMemoryBarrier(

src/renderer_webgpu.cpp

@@ -7220,6 +7220,13 @@ WGPU_IMPORT
 
 				if (isCompute)
 				{
+					if (viewChanged
+					&&  NULL != computePassEncoder)
+					{
+						WGPU_CHECK(wgpuComputePassEncoderEnd(computePassEncoder) );
+						wgpuRelease(computePassEncoder);
+					}
+
 					if (NULL == computePassEncoder)
 					{
 						BGFX_WGPU_PROFILER_END();
@@ -7232,6 +7239,12 @@ WGPU_IMPORT
 							wgpuRelease(renderPassEncoder);
 						}
 
+						if (viewChanged)
+						{
+							submitUniformCache(ucs, view);
+							submitBlit(bs, view);
+						}
+
 						WGPUCommandEncoder cmdEncoder = m_cmd.alloc();
 						computePassEncoder = WGPU_CHECK(wgpuCommandEncoderBeginComputePass(cmdEncoder, NULL) );
 					}