Implemented the per-image-view usages to no longer get errors about extended usages of the source image, @achalpandeyy I'm ready for your compute blits!

P.S. Also fixed the setting of IGPUImageView's format to always match IGPUImages, now only if format got promoted.

Author: devshgraphicsprogramming

include/nbl/video/utilities/IGPUObjectFromAssetConverter.h

@@ -1059,6 +1059,7 @@ auto IGPUObjectFromAssetConverter::create(const asset::ICPUImage** const _begin,
         {
             newFormat = physDev->promoteImageFormat(promotionRequest, video::IGPUImage::ET_LINEAR);
             newFormatIsStorable = physDev->getImageFormatUsagesLinearTiling()[newFormat].storageImage;
+            params.tiling = video::IGPUImage::ET_LINEAR;
         }
 
         assert(newFormat != asset::EF_UNKNOWN); // No feasible supported format found for creating this image
@@ -1548,20 +1549,55 @@ inline created_gpu_object_array<asset::ICPUImageView> IGPUObjectFromAssetConvert
     core::vector<size_t> redirs = eliminateDuplicatesAndGenRedirs(cpuDeps);
 
     auto gpuDeps = getGPUObjectsFromAssets<asset::ICPUImage>(cpuDeps.data(), cpuDeps.data() + cpuDeps.size(), _params);
-
+    const auto physDev = _params.device->getPhysicalDevice();
+    const auto& optimalUsages = physDev->getImageFormatUsagesOptimalTiling();
+    const auto& linearUsages = physDev->getImageFormatUsagesLinearTiling();
     for (ptrdiff_t i = 0; i < assetCount; ++i)
     {
         if (gpuDeps->begin()[redirs[i]])
         {
-            const asset::ICPUImageView::SCreationParams& cpuparams = _begin[i]->getCreationParameters();
+            const auto& cpuParams = _begin[i]->getCreationParameters();
+
             IGPUImageView::SCreationParams params = {};
-            params.flags = static_cast<IGPUImageView::E_CREATE_FLAGS>(cpuparams.flags);
+            params.flags = static_cast<IGPUImageView::E_CREATE_FLAGS>(cpuParams.flags);
+            params.viewType = static_cast<IGPUImageView::E_TYPE>(cpuParams.viewType);
             params.image = (*gpuDeps)[redirs[i]];
-            params.viewType = static_cast<IGPUImageView::E_TYPE>(cpuparams.viewType);
-            params.format = params.image->getCreationParameters().format;
-            memcpy(&params.components, &cpuparams.components, sizeof(params.components));
-            params.subresourceRange = cpuparams.subresourceRange;
-            params.subresourceRange.levelCount = (*gpuDeps)[redirs[i]]->getCreationParameters().mipLevels - params.subresourceRange.baseMipLevel;
+            const auto& gpuImgParams = params.image->getCreationParameters();
+            // override the view's format if the source image got promoted, a bit crude, but don't want to scratch my head about how to promote the views and guess semantics
+            const bool formatGotPromoted = asset::getFormatClass(cpuParams.format)!=asset::getFormatClass(gpuImgParams.format);
+            params.format = formatGotPromoted ? gpuImgParams.format:cpuParams.format;
+            params.subUsages = cpuParams.subUsages;
+            // TODO: In Asset Converter 2.0 we'd pass through all descriptor sets etc and propagate the adding usages backwards to views, but here we need to trim the image's usages instead
+            {
+                IPhysicalDevice::SFormatImageUsages::SUsage validUsages(gpuImgParams.usage);
+                if (params.image->getTiling()!=IGPUImage::ET_LINEAR)
+                    validUsages = validUsages & optimalUsages[params.format];
+                else
+                    validUsages = validUsages & linearUsages[params.format];
+                // add them after trimming
+                if (validUsages.sampledImage)
+                    params.subUsages |= IGPUImage::EUF_SAMPLED_BIT;
+                if (validUsages.storageImage)
+                    params.subUsages |= IGPUImage::EUF_STORAGE_BIT;
+                if (validUsages.attachment)
+                {
+                    if (asset::isDepthOrStencilFormat(params.format))
+                        params.subUsages |= IGPUImage::EUF_DEPTH_STENCIL_ATTACHMENT_BIT;
+                    else
+                        params.subUsages |= IGPUImage::EUF_COLOR_ATTACHMENT_BIT;
+                }
+                if (validUsages.transferSrc)
+                    params.subUsages |= IGPUImage::EUF_TRANSFER_SRC_BIT;
+                if (validUsages.transferDst)
+                    params.subUsages |= IGPUImage::EUF_TRANSFER_DST_BIT;
+                // stuff thats not dependent on device caps
+                const auto uncappedUsages = IGPUImage::EUF_TRANSIENT_ATTACHMENT_BIT|IGPUImage::EUF_INPUT_ATTACHMENT_BIT|IGPUImage::EUF_SHADING_RATE_IMAGE_BIT_NV|IGPUImage::EUF_FRAGMENT_DENSITY_MAP_BIT_EXT;
+                params.subUsages |= gpuImgParams.usage&uncappedUsages;
+            }
+            memcpy(&params.components, &cpuParams.components, sizeof(params.components));
+            params.subresourceRange = cpuParams.subresourceRange;
+            // TODO: Undo this, make all loaders set the level and layer counts on image views to `ICPUImageView::remaining_...`
+            params.subresourceRange.levelCount = gpuImgParams.mipLevels-params.subresourceRange.baseMipLevel;
             (*res)[i] = _params.device->createImageView(std::move(params));
         }
     }