From 31d5fa9eb3bcd088da3a34185f3f9049d9925de5 Mon Sep 17 00:00:00 2001
From: baldurk <baldurk@baldurk.org>
Date: Tue, 14 Mar 2023 15:07:28 +0000
Subject: [PATCH] Fix non-coherent atom alignment issues

---
 renderdoc/driver/vulkan/vk_initstate.cpp | 8 +++++---
 renderdoc/driver/vulkan/vk_memory.cpp    | 4 ++++
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/renderdoc/driver/vulkan/vk_initstate.cpp b/renderdoc/driver/vulkan/vk_initstate.cpp
index cbed89402..1312015f8 100644
--- a/renderdoc/driver/vulkan/vk_initstate.cpp
+++ b/renderdoc/driver/vulkan/vk_initstate.cpp
@@ -1306,9 +1306,11 @@ bool WrappedVulkan::Serialise_InitialState(SerialiserType &ser, ResourceId id, V
     {
       if(initial && initial->mem.mem != VK_NULL_HANDLE)
       {
+        VkDeviceSize size = AlignUp(initial->mem.size, nonCoherentAtomSize);
+
         mappedMem = initial->mem;
-        vkr = ObjDisp(d)->MapMemory(Unwrap(d), Unwrap(mappedMem.mem), initial->mem.offs,
-                                    initial->mem.size, 0, (void **)&Contents);
+        vkr = ObjDisp(d)->MapMemory(Unwrap(d), Unwrap(mappedMem.mem), initial->mem.offs, size, 0,
+                                    (void **)&Contents);
         CheckVkResult(vkr);
 
         // invalidate the cpu cache for this memory range to avoid reading stale data
@@ -1317,7 +1319,7 @@ bool WrappedVulkan::Serialise_InitialState(SerialiserType &ser, ResourceId id, V
             NULL,
             Unwrap(mappedMem.mem),
             mappedMem.offs,
-            mappedMem.size,
+            size,
         };
 
         vkr = ObjDisp(d)->InvalidateMappedMemoryRanges(Unwrap(d), 1, &range);
diff --git a/renderdoc/driver/vulkan/vk_memory.cpp b/renderdoc/driver/vulkan/vk_memory.cpp
index 860748dd9..9f6db8095 100644
--- a/renderdoc/driver/vulkan/vk_memory.cpp
+++ b/renderdoc/driver/vulkan/vk_memory.cpp
@@ -207,6 +207,10 @@ MemoryAllocation WrappedVulkan::AllocateMemoryForResource(bool buffer, VkMemoryR
     // offs is where we can put our next sub-allocation
     VkDeviceSize offs = block.offs;
 
+    // for ease, ensure all allocations are allocated to the non-coherent atom size, so we can
+    // invalidate/flush safely. This is at most 256 bytes which is likely already satisfied.
+    offs = AlignUp(offs, nonCoherentAtomSize);
+
     // if we are on a buffer/image, account for any alignment we might have to do
     if(ret.buffer != block.buffer)
       offs = AlignUp(offs, m_PhysicalDeviceData.props.limits.bufferImageGranularity);