work

src/core/libraries/gnmdriver/gnmdriver.cpp

@@ -2155,6 +2155,7 @@ int PS4_SYSV_ABI sceGnmSubmitCommandBuffersForWorkload() {
 
 int PS4_SYSV_ABI sceGnmSubmitDone() {
     LOG_DEBUG(Lib_GnmDriver, "called");
+    WaitGpuIdle();
     if (!liverpool->IsGpuIdle()) {
         submission_lock = true;
     }

src/emulator.cpp

@@ -10,6 +10,7 @@
 #ifdef ENABLE_QT_GUI
 #include "qt_gui/memory_patcher.h"
 #endif
+#include "common/ntapi.h"
 #include "common/path_util.h"
 #include "common/polyfill_thread.h"
 #include "common/scm_rev.h"

src/shader_recompiler/frontend/translate/translate.cpp

@@ -171,7 +171,7 @@ T Translator::GetSrc(const InstOperand& operand) {
         }
     } else {
         if (operand.input_modifier.abs) {
-            LOG_WARNING(Render_Vulkan, "Input abs modifier on integer instruction");
+            value = ir.IAbs(value);
         }
         if (operand.input_modifier.neg) {
             UNREACHABLE();

src/shader_recompiler/info.h

@@ -1,6 +1,5 @@
 // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
-
 #pragma once
 
 #include <span>
@@ -182,6 +181,7 @@ struct Info {
         const u32* base = user_data.data();
         if (ptr_index != IR::NumScalarRegs) {
             std::memcpy(&base, &user_data[ptr_index], sizeof(base));
+            base = reinterpret_cast<const u32*>(VAddr(base) & 0xFFFFFFFFFFFFULL);
         }
         std::memcpy(&data, base + dword_offset, sizeof(T));
         return data;

src/shader_recompiler/ir/passes/lower_shared_mem_to_registers.cpp

@@ -21,8 +21,7 @@ void LowerSharedMemToRegisters(IR::Program& program) {
                 const IR::Inst* prod = inst.Arg(0).InstRecursive();
                 const auto it = std::ranges::find_if(ds_writes, [&](const IR::Inst* write) {
                     const IR::Inst* write_prod = write->Arg(0).InstRecursive();
-                    return write_prod->Arg(1).U32() == prod->Arg(1).U32() &&
+                    return write_prod->Arg(1).U32() == prod->Arg(1).U32();
-                           write_prod->Arg(0) == prod->Arg(0);
                 });
                 ASSERT(it != ds_writes.end());
                 // Replace data read with value written.

src/video_core/amdgpu/liverpool.cpp

@@ -581,9 +581,6 @@ Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, int vqid) {
             const auto* nop = reinterpret_cast<const PM4CmdNop*>(header);
             break;
         }
-        case PM4ItOpcode::DmaData: {
-            break;
-        }
         case PM4ItOpcode::IndirectBuffer: {
             const auto* indirect_buffer = reinterpret_cast<const PM4CmdIndirectBuffer*>(header);
             auto task = ProcessCompute(

src/video_core/amdgpu/pm4_cmds.h

@@ -187,6 +187,11 @@ struct PM4CmdSetData {
         BitField<28, 4, u32> index;      ///< Index for UCONFIG/CONTEXT on CI+
                                          ///< Program to zero for other opcodes and on SI
     };
+    u32 data[0];
+
+    [[nodiscard]] u32 Size() const {
+        return header.count << 2u;
+    }
 
     template <PM4ShaderType type = PM4ShaderType::ShaderGraphics, typename... Args>
     static constexpr u32* SetContextReg(u32* cmdbuf, Args... data) {
@@ -350,6 +355,16 @@ struct PM4CmdEventWriteEop {
     }
 };
 
+struct PM4CmdAcquireMem {
+    PM4Type3Header header;
+    u32 cp_coher_cntl;
+    u32 cp_coher_size_lo;
+    u32 cp_coher_size_hi;
+    u32 cp_coher_base_lo;
+    u32 cp_coher_base_hi;
+    u32 poll_interval;
+};
+
 enum class DmaDataDst : u32 {
     Memory = 0,
     Gds = 1,
@@ -467,6 +482,10 @@ struct PM4CmdWriteData {
     };
     u32 data[0];
 
+    u32 Size() const {
+        return (header.count.Value() - 2) * 4;
+    }
+
     template <typename T>
     void Address(T addr) {
         addr64 = static_cast<u64>(addr);

src/video_core/buffer_cache/buffer_cache.cpp

@@ -586,9 +586,6 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr,
         const u32 depth =
             image.info.props.is_volume ? std::max(image.info.size.depth >> m, 1u) : 1u;
         const auto& [mip_size, mip_pitch, mip_height, mip_ofs] = image.info.mips_layout[m];
-        if (offset + (mip_ofs * num_layers) > buffer.SizeBytes()) {
-            break;
-        }
         copies.push_back({
             .bufferOffset = offset,
             .bufferRowLength = static_cast<u32>(mip_pitch),

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -452,7 +452,7 @@ void GraphicsPipeline::BindResources(const Liverpool::Regs& regs,
         boost::container::static_vector<AmdGpu::Image, 32> tsharps;
         for (const auto& image_desc : stage->images) {
             const auto tsharp = image_desc.GetSharp(*stage);
-            if (tsharp) {
+            if (tsharp.GetDataFmt() != AmdGpu::DataFormat::FormatInvalid) {
                 tsharps.emplace_back(tsharp);
                 VideoCore::ImageInfo image_info{tsharp, image_desc.is_depth};
                 VideoCore::ImageViewInfo view_info{tsharp, image_desc.is_storage};

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -184,7 +184,7 @@ const ComputePipeline* PipelineCache::GetComputePipeline() {
 }
 
 bool ShouldSkipShader(u64 shader_hash, const char* shader_type) {
-    static constexpr std::array<u64, 4> skip_hashes = {0x6f27708a, 0x6af8ef74, 0xdf795c1f, 0xc2c49a3b};
+    static constexpr std::array<u64, 0> skip_hashes = {};
     if (std::ranges::contains(skip_hashes, shader_hash)) {
         LOG_WARNING(Render_Vulkan, "Skipped {} shader hash {:#x}.", shader_type, shader_hash);
         return true;

src/video_core/renderer_vulkan/vk_platform.cpp

@@ -43,8 +43,6 @@ static VKAPI_ATTR VkBool32 VKAPI_CALL DebugUtilsCallback(
     case 0x609a13b: // Vertex attribute at location not consumed by shader
     case 0xc81ad50e:
     case 0xb7c39078:
-    case 0x30b6e267: // TODO remove this
-    case 0xde55a405: // TODO remove this
     case 0x32868fde: // vkCreateBufferView(): pCreateInfo->range does not equal VK_WHOLE_SIZE
     case 0x92d66fc1: // `pMultisampleState is NULL` for depth only passes (confirmed VL error)
         return VK_FALSE;

src/video_core/renderer_vulkan/vk_resource_pool.cpp

@@ -151,8 +151,7 @@ vk::DescriptorSet DescriptorHeap::Commit(vk::DescriptorSetLayout set_layout) {
 
     // The pool has run out. Record current tick and place it in pending list.
     ASSERT_MSG(result == vk::Result::eErrorOutOfPoolMemory,
-               "Unexpected error during descriptor set allocation {}",
+               "Unexpected error during descriptor set allocation {}", vk::to_string(result));
-               vk::to_string(result));
     pending_pools.emplace_back(curr_pool, master_semaphore->CurrentTick());
     if (const auto [pool, tick] = pending_pools.front(); master_semaphore->IsFree(tick)) {
         curr_pool = pool;
@@ -166,8 +165,7 @@ vk::DescriptorSet DescriptorHeap::Commit(vk::DescriptorSetLayout set_layout) {
     alloc_info.descriptorPool = curr_pool;
     result = device.allocateDescriptorSets(&alloc_info, desc_sets.data());
     ASSERT_MSG(result == vk::Result::eSuccess,
-               "Unexpected error during descriptor set allocation {}",
+               "Unexpected error during descriptor set allocation {}", vk::to_string(result));
-               vk::to_string(result));
 
     // We've changed pool so also reset descriptor batch cache.
     descriptor_sets.clear();

src/video_core/renderer_vulkan/vk_resource_pool.h

@@ -66,6 +66,7 @@ private:
 
 class DescriptorHeap final {
     static constexpr u32 DescriptorSetBatch = 32;
+
 public:
     explicit DescriptorHeap(const Instance& instance, MasterSemaphore* master_semaphore,
                             std::span<const vk::DescriptorPoolSize> pool_sizes,

src/video_core/texture_cache/image.cpp

@@ -73,7 +73,6 @@ static vk::ImageUsageFlags ImageUsageFlags(const ImageInfo& info) {
         if (!info.IsBlockCoded() && !info.IsPacked()) {
             usage |= vk::ImageUsageFlagBits::eColorAttachment;
         }
-
         // In cases where an image is created as a render/depth target and cleared with compute,
         // we cannot predict whether it will be used as a storage image. A proper solution would
         // involve re-creating the resource with a new configuration and copying previous content