VulkanPushPool: Sprinkle a couple of asserts.

Common/GPU/Vulkan/VulkanMemory.cpp

@@ -323,6 +323,7 @@ VulkanPushPool::Block VulkanPushPool::CreateBlock(size_t size) {
 	result = vmaMapMemory(vulkan_->Allocator(), block.allocation, (void **)(&block.writePtr));
 	_assert_(result == VK_SUCCESS);
 
+	_assert_msg_(block.writePtr != nullptr, "VulkanPushPool: Failed to map memory on block of size %d", (int)block.size);
 	return block;
 }
 
@@ -384,13 +385,15 @@ void VulkanPushPool::NextBlock(VkDeviceSize allocationSize) {
 			block.used = allocationSize;
 			block.lastUsed = time_now_d();
 			block.frameIndex = curFrameIndex;
+			_assert_(block.writePtr != nullptr);
 			return;
 		}
 		curBlockIndex_++;
 	}
 
 	double start = time_now_d();
 	VkDeviceSize newBlockSize = std::max(originalBlockSize_ * 2, (VkDeviceSize)RoundUpToPowerOf2((uint32_t)allocationSize));
+
 	// We're still here and ran off the end of blocks. Create a new one.
 	blocks_.push_back(CreateBlock(newBlockSize));
 	blocks_.back().frameIndex = curFrameIndex;

Common/GPU/Vulkan/VulkanMemory.h

@@ -107,6 +107,7 @@ class VulkanPushBuffer : public VulkanMemoryManager {
 
 // Simple memory pushbuffer pool that can share blocks between the "frames", to reduce the impact of push memory spikes -
 // a later frame can gobble up redundant buffers from an earlier frame even if they don't share frame index.
+// NOT thread safe! Can only be used from one thread (our main thread).
 class VulkanPushPool : public VulkanMemoryManager {
 public:
 	VulkanPushPool(VulkanContext *vulkan, const char *name, size_t originalBlockSize, VkBufferUsageFlags usage);