read wrong depth from stage texture? about diligentengine

[TheMostDiligent]

Note that your code for 24-bit is incorrect. You are reading as if 24-bit depth was packed into 32-bit leaving each byte unused, which may not be the case.

How do you copy your depth texture to your staging texture?

[9MW]

Note that your code for 24-bit is incorrect. You are reading as if 24-bit depth was packed into 32-bit leaving each byte unused, which may not be the case.

How do you copy your depth texture to your staging texture?

I stopped use TEX_FORMAT_D24_UNORM_S8_UINT after found this format is not supported to copy back to RAM.
Copied by

			math::v4f cursorZ;
			const float ClearColor[] = { 0.350f, 0.350f, 0.350f, 1.0f };
			auto& m_pSwapChain = this->p_rt->m_pSwapChain;
			auto pRTV = m_pSwapChain->GetCurrentBackBufferRTV();
			auto pDSV = m_pSwapChain->GetDepthBufferDSV();
			auto df = pDSV->GetTexture()->GetDesc();

			auto& pContext = this->p_rt->m_pImmediateContext;
			auto pdepthtex = pDSV->GetTexture();
			pContext->SetRenderTargets(0, nullptr, nullptr, RESOURCE_STATE_TRANSITION_MODE_NONE);
			auto& m_UploadCompleteFence = _editmem.get<1>().m_UploadCompleteFence;
			auto& cam = Camera::Main();
			inputmanager& ipt = *p_ipt;
			cursorZ.head<2>() = ipt.mouseinfo.operator pj2::mousestate::f2 & ();
			editmemCounters& memedit = _editmem.get<1>();
			cursorZ = cursorZ.cwiseMax(math::v4f::Ones()*4);
			cursorZ[0] = std::min((UINT32)cursorZ[0], cam.m_ProjAttribs.Width);
			cursorZ[1] = std::min((UINT32)cursorZ[1], cam.m_ProjAttribs.Height);
			//pContext->UnmapTextureSubresource(pStagingTexture, StartDstMip + mip, StartDstSlice + slice);
			Box SrcBox;
			SrcBox.MaxX = cursorZ[0];
			SrcBox.MinX = std::max(0u, SrcBox.MaxX - 3);
			SrcBox.MaxY = cursorZ[1];
			SrcBox.MinY = std::max(0u, SrcBox.MaxY - 3);
			CopyTextureAttribs CopyAttribs(pdepthtex, RESOURCE_STATE_TRANSITION_MODE_TRANSITION,
				memedit.pStagingTexture, RESOURCE_STATE_TRANSITION_MODE_TRANSITION);
			CopyAttribs.pSrcBox = &SrcBox;
			pContext->CopyTexture(CopyAttribs);
			pContext->EnqueueSignal(m_UploadCompleteFence, ++_editmem.get<1>().m_UploadCompleteFenceValue);

			const StateTransitionDesc Barrier{ pdepthtex, RESOURCE_STATE_COPY_SOURCE, RESOURCE_STATE_DEPTH_WRITE };
			pContext->TransitionResourceStates(1, &Barrier);
			pContext->SetRenderTargets(1, &pRTV, pDSV, RESOURCE_STATE_TRANSITION_MODE_TRANSITION);

[TheMostDiligent]

You may consider using GPUCompletionAwaitQueue

Here is the code that does what you need.

[9MW]

Thanks you reply, I'll try it.

inline void CvtDepth(edtmem& edtm, span<float> num) {
		constexpr int Mip = 0;
		auto& UploadBuffDesc = edtm.get<1>().UploadBuffDesc;
		const Uint32 Width = UploadBuffDesc.Width >> Mip;
		const Uint32 Height = UploadBuffDesc.Height >> Mip;
		auto& MappedData = edtm.get<1>().MappedData;
		constexpr uint xs = 0x00FFFFFF;
		auto pRGBAData = reinterpret_cast<float*>(MappedData.pData);
		{
			for (Uint32 y = 0, cnt = 0; y < Height; ++y)
			{
				auto pRGBA = pRGBAData + MappedData.Stride * y;
				for (Uint32 x = 0; x < Width; ++x)
				{
					// depth bits
					num[cnt++] = static_cast<float>(pRGBA[x]);
				}
			}
		}
	}

		       inputmanager& ipt = *p_ipt;
			auto pdepthtex = rt_.m_pDepthDSV->GetTexture();
			auto& m_UploadCompleteFence = _editmem.get<1>().m_UploadCompleteFence;
			m_UploadCompleteFence->Wait(_editmem.get<1>().m_UploadCompleteFenceValue);
			math::v4f cursorZ, zero4 = {};
			inputmanager::updateParam uppara;

			auto& worldPos = std::get<0>(uppara);
			cursorZ.head<2>() = ipt.mouseinfo.operator pj2::mousestate::f2 & ();
			editmemCounters& memedit = _editmem.get<1>();
			cursorZ = cursorZ.cwiseMax(zero4);
			cursorZ[0] = std::min((UINT32)cursorZ[0], cam.m_ProjAttribs.Width);
			cursorZ[1] = std::min((UINT32)cursorZ[1], cam.m_ProjAttribs.Height);
			pContext->UnmapTextureSubresource(memedit.pStagingTexture, 0,0);
			pContext->MapTextureSubresource(memedit.pStagingTexture, 0, 0,
				MAP_READ,
				MAP_FLAG_DO_NOT_WAIT | MAP_FLAG_DISCARD | MAP_FLAG_NO_OVERWRITE, nullptr, memedit.MappedData);

			auto& UploadBuffDesc = _editmem.get<1>().UploadBuffDesc;
			UploadBuffDesc.Width = 1;
			UploadBuffDesc.Height = 1;
			assert(rt_.DepthBufferFormat != TEX_FORMAT_D24_UNORM_S8_UINT);
			std::array<float, 16> depthNorm;
			CvtDepth(_editmem, span(depthNorm.data(), 1));
			cursorZ[2] = depthNorm[0];
			const math::Matrix4f& inverse_projection_matrix = cam.m_ProjMatrix.inverse();
			const math::Matrix4f& inverse_View_matrix = cam.m_ViewMatrix.matrix().inverse();
			getRays(cursorZ, inverse_projection_matrix, worldPos, cam);
			worldPos = inverse_View_matrix * worldPos;
			freeUIVec = worldPos;

			ipt.update(uppara);

here is update function to read depth before call the CopyTexture Code posted previously

[TheMostDiligent]

Did you run this in Debug? Do you get any messages in the console?

This is incorrect:

			cursorZ[0] = std::min((UINT32)cursorZ[0], cam.m_ProjAttribs.Width);
			cursorZ[1] = std::min((UINT32)cursorZ[1], cam.m_ProjAttribs.Height);

It should be

			cursorZ[0] = std::min((UINT32)cursorZ[0], cam.m_ProjAttribs.Width - 1);
			cursorZ[1] = std::min((UINT32)cursorZ[1], cam.m_ProjAttribs.Height - 1);

[TheMostDiligent]

I suggest you look at the code I sent and do the same way.

[9MW]

I suggest you look at the code I sent and do the same way.

Appreciate you reply, I'll take a look.

[9MW]

solved by remove UnmapTextureSubresource and MapTextureSubresource call in update function.
MapTextureSubresource once at init function would be enough. Maybe I need UnmapTextureSubresource before release, haven't try that

[TheMostDiligent]

You read the data between MapTextureSubresource and UnmapTextureSubresource. In you snippet, you unmap right before mapping again, which does not make a lot of sense.
Once again, take a look at the code that I sent.