video_core: MCI boot fixes and DMA multisized components support

Add workarounds for Marvel Cosmic Invasion boot issues:
- Skip first 2 compute dispatches (xbzk@eden-emu.dev)
- Clamp staging buffers to 2GB to prevent Vulkan failures (xbzk@eden-emu.dev)
- Validate staging buffer sizes before uploads (xbzk@eden-emu.dev)

Also improve DMA engine to support multisized components (1-4 bytes)
instead of hardcoded 4-byte components.

Co-authored-by: xbzk <xbzk@eden-emu.dev>
Signed-off-by: Zephyron <zephyron@citron-emu.org>

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -1,4 +1,5 @@
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 citron Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
@@ -37,6 +38,7 @@
 #include "video_core/texture_cache/texture_cache_base.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
+#include "citron/util/title_ids.h"
 
 namespace Vulkan {
 
@@ -487,6 +489,15 @@ void RasterizerVulkan::Clear(u32 layer_count) {
 }
 
 void RasterizerVulkan::DispatchCompute() {
+    // Skip first 2 dispatches for Marvel Cosmic Invasion to fix boot issues
+    if (program_id == UICommon::TitleID::MarvelCosmicInvasion) {
+        static u32 dispatch_count = 0;
+        if (dispatch_count < 2) {
+            dispatch_count++;
+            return;
+        }
+    }
+
     FlushWork();
     gpu_memory->FlushCaching();
 
@@ -1604,6 +1615,7 @@ void RasterizerVulkan::InitializeChannel(Tegra::Control::ChannelState& channel)
 
 void RasterizerVulkan::BindChannel(Tegra::Control::ChannelState& channel) {
     const s32 channel_id = channel.bind_id;
+    staging_pool.SetProgramId(channel.program_id);
     BindToChannel(channel_id);
     {
         std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};

src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp

@@ -1,4 +1,5 @@
 // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 citron Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #include <algorithm>
@@ -16,6 +17,7 @@
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
+#include "citron/util/title_ids.h"
 
 namespace Vulkan {
 namespace {
@@ -235,7 +237,15 @@ std::optional<StagingBufferRef> StagingBufferPool::TryGetReservedBuffer(size_t s
 
 StagingBufferRef StagingBufferPool::CreateStagingBuffer(size_t size, MemoryUsage usage,
                                                         bool deferred) {
-    const u32 log2 = Common::Log2Ceil64(size);
+    u32 log2 = Common::Log2Ceil64(size);
+
+    // Only apply this workaround for Marvel Cosmic Invasion
+    if (program_id == UICommon::TitleID::MarvelCosmicInvasion) {
+        static constexpr u32 MAX_STAGING_BUFFER_LOG2 = 31U;
+        // Calculate log2 of requested size, but clamp to maximum to prevent overflow
+        // This ensures we still round up to the next power of 2, but cap at 2GB
+        log2 = std::min(log2, MAX_STAGING_BUFFER_LOG2);
+    }
     VkBufferCreateInfo buffer_ci = {
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
         .pNext = nullptr,

src/video_core/renderer_vulkan/vk_staging_buffer_pool.h

@@ -1,4 +1,5 @@
 // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 citron Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #pragma once
@@ -44,6 +45,10 @@ public:
 
     u64 GetMemoryUsage() const;
 
+    void SetProgramId(u64 program_id_) {
+        program_id = program_id_;
+    }
+
 private:
     struct StreamBufferCommit {
         size_t upper_bound;
@@ -121,6 +126,7 @@ private:
     size_t current_delete_level = 0;
     u64 buffer_index = 0;
     u64 unique_ids{};
+    u64 program_id{};
 };
 
 } // namespace Vulkan