tree a2199dc18a700b046e69f1296e349016539285f4
parent b41615cfce1bc895b3ead88baa69c7ca345def51
author Adam Sawicki <adam.sawicki@amd.com> 1645105446 +0100
committer Adam Sawicki <adam.sawicki@amd.com> 1645105446 +0100

Added new API for selecting preferred memory type: VMA_MEMORY_USAGE_AUTO, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, VMA_MEMORY_USAGE_AUTO_PREFER_HOST, VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT, VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT, VMA_ALLOCATION_CREATE_HOST_ACCESS_ALLOW_TRANSFER_INSTEAD_BIT

Old values like VMA_MEMORY_USAGE_GPU_ONLY still work as before, for backward compatibility.
Updated documentation accordingly.

Introduced a concept of mappable and non-mappable allocations - with VMA_MEMORY_USAGE_AUTO* an intent to map has to be specified using VMA_ALLOCATION_CREATE_HOST_ACCESS*.

Added mapping hysteresis logic (internal class VmaMappingHysteresis, config macro VMA_MAPPING_HYSTERESIS_ENABLED) that prevents too many calls vkMapMemory/vkUnmapMemory back and forth.

Internal improvement in VmaBlockVector::AllocatePage to try to keep mappable and non-mappable allocations separate.