[vk] Bring Vulkan closer to Spec (#180)

The changes noted below bring Vulkan closer to 1.3 spec and get rid of validation errors and enable us to properly use one or two more functions.

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/180
Reviewed-by: crueter <crueter@eden-emu.dev>
Reviewed-by: Maufeat <sahyno1996@gmail.com>
Co-authored-by: Shinmegumi <shinmegumi@eden-emu.dev>
Co-committed-by: Shinmegumi <shinmegumi@eden-emu.dev>

10 files changed, 176 insertions, 122 deletions

diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp
index f2bf5a3228..37f0c74d26 100644
--- a/src/video_core/renderer_vulkan/blit_image.cpp
+++ b/src/video_core/renderer_vulkan/blit_image.cpp
@@ -518,7 +518,7 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([this, dst_framebuffer, src_image_view, src_image, src_sampler, dst_region,
                       src_region, src_size, pipeline, layout](vk::CommandBuffer cmdbuf) {
-        TransitionImageLayout(cmdbuf, src_image, VK_IMAGE_LAYOUT_READ_ONLY_OPTIMAL);
+        TransitionImageLayout(cmdbuf, src_image,  VK_IMAGE_LAYOUT_GENERAL);
         BeginRenderPass(cmdbuf, dst_framebuffer);
         const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
         UpdateOneTextureDescriptorSet(device, descriptor_set, src_sampler, src_image_view);
diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
index 6c40ff1bab..e73e885e66 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
@@ -502,7 +502,6 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
 void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
                                      const RenderAreaPushConstant& render_area) {
     scheduler.RequestRenderpass(texture_cache.GetFramebuffer());
-
     if (!is_built.load(std::memory_order::relaxed)) {
         // Wait for the pipeline to be built
         scheduler.Record([this](vk::CommandBuffer) {
diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp
index 1f71bc68c6..d6ecc2b65c 100644
--- a/src/video_core/renderer_vulkan/vk_query_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp
@@ -13,7 +13,7 @@
 #include <unordered_map>
 #include <utility>
 #include <vector>
-
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "common/bit_util.h"
 #include "common/common_types.h"
 #include "video_core/engines/draw_manager.h"
@@ -116,11 +116,11 @@ struct HostSyncValues {
 class SamplesStreamer : public BaseStreamer {
 public:
     explicit SamplesStreamer(size_t id_, QueryCacheRuntime& runtime_,
-                             VideoCore::RasterizerInterface* rasterizer_, const Device& device_,
+                             VideoCore::RasterizerInterface* rasterizer_, TextureCache& texture_cache_, const Device& device_,
                              Scheduler& scheduler_, const MemoryAllocator& memory_allocator_,
                              ComputePassDescriptorQueue& compute_pass_descriptor_queue,
                              DescriptorPool& descriptor_pool)
-        : BaseStreamer(id_), runtime{runtime_}, rasterizer{rasterizer_}, device{device_},
+        : BaseStreamer(id_), texture_cache{texture_cache_}, runtime{runtime_}, rasterizer{rasterizer_}, device{device_},
           scheduler{scheduler_}, memory_allocator{memory_allocator_} {
         current_bank = nullptr;
         current_query = nullptr;
@@ -153,16 +153,33 @@ public:
         if (has_started) {
             return;
         }
+
         ReserveHostQuery();
+
+        // Ensure outside render pass
+        scheduler.RequestOutsideRenderPassOperationContext();
+
+        // Reset query pool outside render pass
         scheduler.Record([query_pool = current_query_pool,
-                          query_index = current_bank_slot](vk::CommandBuffer cmdbuf) {
+                                 query_index = current_bank_slot](vk::CommandBuffer cmdbuf) {
+            cmdbuf.ResetQueryPool(query_pool, static_cast<u32>(query_index), 1);
+        });
+
+        // Manually restart the render pass (required for vkCmdClearAttachments, etc.)
+        scheduler.RequestRenderpass(texture_cache.GetFramebuffer());
+
+        // Begin query inside the newly started render pass
+        scheduler.Record([query_pool = current_query_pool,
+                                 query_index = current_bank_slot](vk::CommandBuffer cmdbuf) {
             const bool use_precise = Settings::IsGPULevelHigh();
             cmdbuf.BeginQuery(query_pool, static_cast<u32>(query_index),
                               use_precise ? VK_QUERY_CONTROL_PRECISE_BIT : 0);
         });
+
         has_started = true;
     }
 
+
     void PauseCounter() override {
         if (!has_started) {
             return;
@@ -404,7 +421,7 @@ private:
             size_slots -= amount;
         }
     }
-
+    TextureCache& texture_cache;
     template <bool is_ordered, typename Func>
     void ApplyBanksWideOp(std::vector<size_t>& queries, Func&& func) {
         std::conditional_t<is_ordered, std::map<size_t, std::pair<size_t, size_t>>,
@@ -1163,13 +1180,13 @@ struct QueryCacheRuntimeImpl {
                           const MemoryAllocator& memory_allocator_, Scheduler& scheduler_,
                           StagingBufferPool& staging_pool_,
                           ComputePassDescriptorQueue& compute_pass_descriptor_queue,
-                          DescriptorPool& descriptor_pool)
+                          DescriptorPool& descriptor_pool, TextureCache& texture_cache_)
         : rasterizer{rasterizer_}, device_memory{device_memory_}, buffer_cache{buffer_cache_},
           device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
           staging_pool{staging_pool_}, guest_streamer(0, runtime),
           sample_streamer(static_cast<size_t>(QueryType::ZPassPixelCount64), runtime, rasterizer,
-                          device, scheduler, memory_allocator, compute_pass_descriptor_queue,
-                          descriptor_pool),
+                          texture_cache_, device, scheduler, memory_allocator,
+                          compute_pass_descriptor_queue, descriptor_pool),
           tfb_streamer(static_cast<size_t>(QueryType::StreamingByteCount), runtime, device,
                        scheduler, memory_allocator, staging_pool),
           primitives_succeeded_streamer(
@@ -1240,10 +1257,10 @@ QueryCacheRuntime::QueryCacheRuntime(VideoCore::RasterizerInterface* rasterizer,
                                      const MemoryAllocator& memory_allocator_,
                                      Scheduler& scheduler_, StagingBufferPool& staging_pool_,
                                      ComputePassDescriptorQueue& compute_pass_descriptor_queue,
-                                     DescriptorPool& descriptor_pool) {
+                                     DescriptorPool& descriptor_pool, TextureCache& texture_cache_) {
     impl = std::make_unique<QueryCacheRuntimeImpl>(
         *this, rasterizer, device_memory_, buffer_cache_, device_, memory_allocator_, scheduler_,
-        staging_pool_, compute_pass_descriptor_queue, descriptor_pool);
+        staging_pool_, compute_pass_descriptor_queue, descriptor_pool, texture_cache_);
 }
 
 void QueryCacheRuntime::Bind3DEngine(Maxwell3D* maxwell3d) {
diff --git a/src/video_core/renderer_vulkan/vk_query_cache.h b/src/video_core/renderer_vulkan/vk_query_cache.h
index f6151123ec..b8dae9bc2d 100644
--- a/src/video_core/renderer_vulkan/vk_query_cache.h
+++ b/src/video_core/renderer_vulkan/vk_query_cache.h
@@ -7,7 +7,7 @@
 
 #include "video_core/query_cache/query_cache_base.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
-
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
 namespace VideoCore {
 class RasterizerInterface;
 }
@@ -17,7 +17,6 @@ class StreamerInterface;
 }
 
 namespace Vulkan {
-
 class Device;
 class Scheduler;
 class StagingBufferPool;
@@ -32,7 +31,7 @@ public:
                                const MemoryAllocator& memory_allocator_, Scheduler& scheduler_,
                                StagingBufferPool& staging_pool_,
                                ComputePassDescriptorQueue& compute_pass_descriptor_queue,
-                               DescriptorPool& descriptor_pool);
+                               DescriptorPool& descriptor_pool, TextureCache& texture_cache_);
     ~QueryCacheRuntime();
 
     template <typename SyncValuesType>
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index dbe2ce66c9..c511a51720 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -189,7 +189,7 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
                            guest_descriptor_queue, compute_pass_descriptor_queue, descriptor_pool),
       buffer_cache(device_memory, buffer_cache_runtime),
       query_cache_runtime(this, device_memory, buffer_cache, device, memory_allocator, scheduler,
-                          staging_pool, compute_pass_descriptor_queue, descriptor_pool),
+                          staging_pool, compute_pass_descriptor_queue, descriptor_pool, texture_cache),
       query_cache(gpu, *this, device_memory, query_cache_runtime),
       pipeline_cache(device_memory, device, scheduler, descriptor_pool, guest_descriptor_queue,
                      render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()),
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index ea032635c2..30780b9cbd 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -136,7 +136,6 @@ public:
     void BindChannel(Tegra::Control::ChannelState& channel) override;
 
     void ReleaseChannel(s32 channel_id) override;
-
     std::optional<FramebufferTextureInfo> AccelerateDisplay(const Tegra::FramebufferConfig& config,
                                                             VAddr framebuffer_addr,
                                                             u32 pixel_stride);
@@ -147,7 +146,6 @@ private:
         0x0100E95004038000ULL, // XC2
         0x0100A6301214E000ULL, // FE:Engage
     };
-
     static constexpr size_t MAX_TEXTURES = 192;
     static constexpr size_t MAX_IMAGES = 48;
     static constexpr size_t MAX_IMAGE_VIEWS = MAX_TEXTURES + MAX_IMAGES;
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp
index aa56a6affd..5ba6ab98b5 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.cpp
+++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp
@@ -270,46 +270,73 @@ void Scheduler::EndPendingOperations() {
     EndRenderPass();
 }
 
-void Scheduler::EndRenderPass() {
-    if (!state.renderpass) {
-        return;
-    }
-
-    query_cache->CounterEnable(VideoCommon::QueryType::ZPassPixelCount64, false);
-    query_cache->NotifySegment(false);
-
-    Record([num_images = num_renderpass_images, images = renderpass_images,
-            ranges = renderpass_image_ranges](vk::CommandBuffer cmdbuf) {
-        std::array<VkImageMemoryBarrier, 9> barriers;
-        for (size_t i = 0; i < num_images; ++i) {
-            barriers[i] = VkImageMemoryBarrier{
-                .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-                .pNext = nullptr,
-                .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
-                .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT |
-                                 VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
-                                 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
-                                 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
-                .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
-                .newLayout = VK_IMAGE_LAYOUT_GENERAL,
-                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .image = images[i],
-                .subresourceRange = ranges[i],
-            };
+    void Scheduler::EndRenderPass() {
+        if (!state.renderpass) {
+            return;
         }
-        cmdbuf.EndRenderPass();
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                                   VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                                   VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
-                               VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, nullptr, nullptr,
-                               vk::Span(barriers.data(), num_images));
-    });
-    state.renderpass = nullptr;
-    num_renderpass_images = 0;
-}
+
+        query_cache->CounterEnable(VideoCommon::QueryType::ZPassPixelCount64, false);
+        query_cache->NotifySegment(false);
+
+        Record([num_images = num_renderpass_images, images = renderpass_images,
+                       ranges = renderpass_image_ranges](vk::CommandBuffer cmdbuf) {
+            std::array<VkImageMemoryBarrier, 9> barriers;
+            VkPipelineStageFlags src_stages = 0;
+
+            for (size_t i = 0; i < num_images; ++i) {
+                const VkImageSubresourceRange& range = ranges[i];
+                const bool is_color = range.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT;
+                const bool is_depth_stencil = range.aspectMask &
+                                              (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT);
+
+                VkAccessFlags src_access = 0;
+                VkPipelineStageFlags this_stage = 0;
+
+                if (is_color) {
+                    src_access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+                    this_stage |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
+                }
+
+                if (is_depth_stencil) {
+                    src_access |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+                    this_stage |= VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                  VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
+                }
+
+                src_stages |= this_stage;
+
+                barriers[i] = VkImageMemoryBarrier{
+                        .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+                        .pNext = nullptr,
+                        .srcAccessMask = src_access,
+                        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT |
+                                         VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
+                                         VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                                         VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
+                                         VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+                        .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+                        .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                        .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                        .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                        .image = images[i],
+                        .subresourceRange = range,
+                };
+            }
+
+            cmdbuf.EndRenderPass();
+
+            for (size_t i = 0; i < num_images; ++i) {
+                cmdbuf.PipelineBarrier(
+                        src_stages, // OR compute per-image if needed
+                        VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+                        0,
+                        barriers[i]);
+            }
+        });
+
+        state.renderpass = nullptr;
+        num_renderpass_images = 0;
+    }
 
 void Scheduler::AcquireNewChunk() {
     std::scoped_lock rl{reserve_mutex};
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index 98eb608aa5..ba58060d20 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -508,58 +508,84 @@ TransformBufferCopies(std::span<const VideoCommon::BufferCopy> copies, size_t bu
         return value;
     }
 }
+struct RangedBarrierRange {
+    u32 min_mip = std::numeric_limits<u32>::max();
+    u32 max_mip = std::numeric_limits<u32>::min();
+    u32 min_layer = std::numeric_limits<u32>::max();
+    u32 max_layer = std::numeric_limits<u32>::min();
 
+    void AddLayers(const VkImageSubresourceLayers& layers) {
+        min_mip = std::min(min_mip, layers.mipLevel);
+        max_mip = std::max(max_mip, layers.mipLevel + 1);
+        min_layer = std::min(min_layer, layers.baseArrayLayer);
+        max_layer = std::max(max_layer, layers.baseArrayLayer + layers.layerCount);
+    }
+
+    VkImageSubresourceRange SubresourceRange(VkImageAspectFlags aspect_mask) const noexcept {
+        return VkImageSubresourceRange{
+            .aspectMask = aspect_mask,
+            .baseMipLevel = min_mip,
+            .levelCount = max_mip - min_mip,
+            .baseArrayLayer = min_layer,
+            .layerCount = max_layer - min_layer,
+        };
+    }
+};
 void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage image,
                        VkImageAspectFlags aspect_mask, bool is_initialized,
                        std::span<const VkBufferImageCopy> copies) {
     static constexpr VkAccessFlags WRITE_ACCESS_FLAGS =
-        VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-        VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+                                           VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                                           VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
     static constexpr VkAccessFlags READ_ACCESS_FLAGS = VK_ACCESS_SHADER_READ_BIT |
                                                        VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
                                                        VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
+
+    //  Compute exact mip/layer range being written to
+    RangedBarrierRange range;
+    for (const auto& region : copies) {
+        range.AddLayers(region.imageSubresource);
+    }
+    const VkImageSubresourceRange subresource_range = range.SubresourceRange(aspect_mask);
+
     const VkImageMemoryBarrier read_barrier{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-        .pNext = nullptr,
-        .srcAccessMask = WRITE_ACCESS_FLAGS,
-        .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
-        .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-        .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .image = image,
-        .subresourceRange{
-            .aspectMask = aspect_mask,
-            .baseMipLevel = 0,
-            .levelCount = VK_REMAINING_MIP_LEVELS,
-            .baseArrayLayer = 0,
-            .layerCount = VK_REMAINING_ARRAY_LAYERS,
-        },
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = WRITE_ACCESS_FLAGS,
+            .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange = subresource_range,
     };
+
     const VkImageMemoryBarrier write_barrier{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-        .pNext = nullptr,
-        .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS,
-        .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-        .newLayout = VK_IMAGE_LAYOUT_GENERAL,
-        .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .image = image,
-        .subresourceRange{
-            .aspectMask = aspect_mask,
-            .baseMipLevel = 0,
-            .levelCount = VK_REMAINING_MIP_LEVELS,
-            .baseArrayLayer = 0,
-            .layerCount = VK_REMAINING_ARRAY_LAYERS,
-        },
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange = subresource_range,
     };
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
+
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+                           VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
+                           VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
                            read_barrier);
     cmdbuf.CopyBufferToImage(src_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copies);
     // TODO: Move this to another API
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
-                           write_barrier);
+    cmdbuf.PipelineBarrier(
+            VK_PIPELINE_STAGE_TRANSFER_BIT,
+            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
+            0, nullptr, nullptr, write_barrier);
 }
 
 [[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region,
@@ -652,29 +678,7 @@ void TryTransformSwizzleIfNeeded(PixelFormat format, std::array<SwizzleSource, 4
     }
 }
 
-struct RangedBarrierRange {
-    u32 min_mip = std::numeric_limits<u32>::max();
-    u32 max_mip = std::numeric_limits<u32>::min();
-    u32 min_layer = std::numeric_limits<u32>::max();
-    u32 max_layer = std::numeric_limits<u32>::min();
 
-    void AddLayers(const VkImageSubresourceLayers& layers) {
-        min_mip = std::min(min_mip, layers.mipLevel);
-        max_mip = std::max(max_mip, layers.mipLevel + 1);
-        min_layer = std::min(min_layer, layers.baseArrayLayer);
-        max_layer = std::max(max_layer, layers.baseArrayLayer + layers.layerCount);
-    }
-
-    VkImageSubresourceRange SubresourceRange(VkImageAspectFlags aspect_mask) const noexcept {
-        return VkImageSubresourceRange{
-            .aspectMask = aspect_mask,
-            .baseMipLevel = min_mip,
-            .levelCount = max_mip - min_mip,
-            .baseArrayLayer = min_layer,
-            .layerCount = max_layer - min_layer,
-        };
-    }
-};
 
 [[nodiscard]] VkFormat Format(Shader::ImageFormat format) {
     switch (format) {
@@ -1458,12 +1462,18 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
                 .subresourceRange = dst_range.SubresourceRange(aspect_mask),
             },
         };
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               0, {}, {}, pre_barriers);
+        cmdbuf.PipelineBarrier(
+                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT,
+                VK_PIPELINE_STAGE_TRANSFER_BIT,
+                0, nullptr, nullptr, pre_barriers);
         cmdbuf.CopyImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image,
                          VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, vk_copies);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
-                               0, {}, {}, post_barriers);
+        cmdbuf.PipelineBarrier(
+                VK_PIPELINE_STAGE_TRANSFER_BIT,
+                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT,
+                0, nullptr, nullptr, post_barriers);
     });
 }
 
@@ -2377,7 +2387,7 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) {
         };
         scheduler.RequestOutsideRenderPassOperationContext();
         scheduler.Record([barrier](vk::CommandBuffer cmdbuf) {
-            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
                                    VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, barrier);
         });
     }
diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp
index 5d80531b47..106630182f 100644
--- a/src/video_core/vulkan_common/vulkan_wrapper.cpp
+++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp
@@ -120,6 +120,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkCmdEndConditionalRenderingEXT);
     X(vkCmdEndQuery);
     X(vkCmdEndRenderPass);
+    X(vkCmdResetQueryPool);
     X(vkCmdEndTransformFeedbackEXT);
     X(vkCmdEndDebugUtilsLabelEXT);
     X(vkCmdFillBuffer);
diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h
index 5e99444e61..8fd0bff6af 100644
--- a/src/video_core/vulkan_common/vulkan_wrapper.h
+++ b/src/video_core/vulkan_common/vulkan_wrapper.h
@@ -219,6 +219,7 @@ struct DeviceDispatch : InstanceDispatch {
     PFN_vkCmdEndConditionalRenderingEXT vkCmdEndConditionalRenderingEXT{};
     PFN_vkCmdEndDebugUtilsLabelEXT vkCmdEndDebugUtilsLabelEXT{};
     PFN_vkCmdEndQuery vkCmdEndQuery{};
+    PFN_vkCmdResetQueryPool vkCmdResetQueryPool{};
     PFN_vkCmdEndRenderPass vkCmdEndRenderPass{};
     PFN_vkCmdEndTransformFeedbackEXT vkCmdEndTransformFeedbackEXT{};
     PFN_vkCmdFillBuffer vkCmdFillBuffer{};
@@ -1137,7 +1138,9 @@ public:
     VkCommandBuffer operator*() const noexcept {
         return handle;
     }
-
+    void ResetQueryPool(VkQueryPool query_pool, uint32_t first, uint32_t count) const noexcept {
+        dld->vkCmdResetQueryPool(handle, query_pool, first, count);
+    }
     void Begin(const VkCommandBufferBeginInfo& begin_info) const {
         Check(dld->vkBeginCommandBuffer(handle, &begin_info));
     }