shader_recompiler: Shared mem has side-effects, fix format component order

src/core/libraries/gnmdriver/gnmdriver.cpp

@@ -1962,7 +1962,7 @@ s32 PS4_SYSV_ABI sceGnmSubmitCommandBuffers(u32 count, const u32* dcb_gpu_addrs[
         if (Config::dumpPM4()) {
             static auto last_frame_num = -1LL;
             static u32 seq_num{};
-            if (last_frame_num == frames_submitted) {
+            if (last_frame_num == frames_submitted && cbpair == 0) {
                 ++seq_num;
             } else {
                 last_frame_num = frames_submitted;

src/shader_recompiler/frontend/control_flow_graph.cpp

@@ -149,9 +149,15 @@ void CFG::LinkBlocks() {
             block.end_class = EndClass::Branch;
         } else if (end_inst.opcode == Opcode::S_ENDPGM) {
             const auto& prev_inst = inst_list[block.end_index - 1];
-            if (prev_inst.opcode == Opcode::EXP && prev_inst.control.exp.en == 0 &&
-                prev_inst.control.exp.target != 9) {
-                block.end_class = EndClass::Kill;
+            if (prev_inst.opcode == Opcode::EXP && prev_inst.control.exp.en == 0) {
+                if (prev_inst.control.exp.target != 9) {
+                    block.end_class = EndClass::Kill;
+                } else if (const auto& exec_mask = inst_list[block.end_index - 2];
+                           exec_mask.src[0].field == OperandField::ConstZero) {
+                    block.end_class = EndClass::Kill;
+                } else {
+                    block.end_class = EndClass::Exit;
+                }
             } else {
                 block.end_class = EndClass::Exit;
             }

src/shader_recompiler/ir/microinstruction.cpp

@@ -40,6 +40,9 @@ Inst::~Inst() {
 
 bool Inst::MayHaveSideEffects() const noexcept {
     switch (op) {
+    case Opcode::Barrier:
+    case Opcode::WorkgroupMemoryBarrier:
+    case Opcode::DeviceMemoryBarrier:
     case Opcode::ConditionRef:
     case Opcode::Reference:
     case Opcode::PhiMove:
@@ -52,6 +55,11 @@ bool Inst::MayHaveSideEffects() const noexcept {
     case Opcode::StoreBufferF32x3:
     case Opcode::StoreBufferF32x4:
     case Opcode::StoreBufferU32:
+    case Opcode::WriteSharedU128:
+    case Opcode::WriteSharedU64:
+    case Opcode::WriteSharedU32:
+    case Opcode::WriteSharedU16:
+    case Opcode::WriteSharedU8:
     case Opcode::ImageWrite:
     case Opcode::ImageAtomicIAdd32:
     case Opcode::ImageAtomicSMin32:

src/video_core/renderer_vulkan/liverpool_to_vk.cpp

@@ -291,11 +291,11 @@ vk::Format SurfaceFormat(AmdGpu::DataFormat data_format, AmdGpu::NumberFormat nu
     }
     if (data_format == AmdGpu::DataFormat::Format8_8_8_8 &&
         num_format == AmdGpu::NumberFormat::Unorm) {
-        return vk::Format::eR8G8B8A8Unorm;
+        return vk::Format::eB8G8R8A8Unorm;
     }
     if (data_format == AmdGpu::DataFormat::Format8_8_8_8 &&
         num_format == AmdGpu::NumberFormat::Srgb) {
-        return vk::Format::eR8G8B8A8Srgb;
+        return vk::Format::eB8G8R8A8Srgb;
     }
     if (data_format == AmdGpu::DataFormat::Format32_32_32 &&
         num_format == AmdGpu::NumberFormat::Float) {

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -23,7 +23,7 @@ Rasterizer::Rasterizer(const Instance& instance_, Scheduler& scheduler_,
     : instance{instance_}, scheduler{scheduler_}, texture_cache{texture_cache_},
       liverpool{liverpool_}, memory{Core::Memory::Instance()},
       pipeline_cache{instance, scheduler, liverpool},
-      vertex_index_buffer{instance, scheduler, VertexIndexFlags, 512_MB, BufferType::Upload} {
+      vertex_index_buffer{instance, scheduler, VertexIndexFlags, 1_GB, BufferType::Upload} {
     if (!Config::nullGpu()) {
         liverpool->BindRasterizer(this);
     }
@@ -176,7 +176,9 @@ u32 Rasterizer::SetupIndexBuffer(bool& is_indexed, u32 index_offset) {
     const auto index_address = regs.index_base_address.Address<const void*>();
     const u32 index_buffer_size = regs.num_indices * index_size;
     const auto [data, offset, _] = vertex_index_buffer.Map(index_buffer_size);
+    static constexpr std::array<u16, 4> test{};
     std::memcpy(data, index_address, index_buffer_size);
+    ASSERT(std::memcmp(data, test.data(), sizeof(test)) != 0);
     vertex_index_buffer.Commit(index_buffer_size);
 
     // Bind index buffer.

src/video_core/renderer_vulkan/vk_stream_buffer.cpp

@@ -226,7 +226,7 @@ void StreamBuffer::WaitPendingOperations(u64 requested_upper_bound) {
     while (requested_upper_bound > wait_bound && wait_cursor < *invalidation_mark) {
         auto& watch = previous_watches[wait_cursor];
         wait_bound = watch.upper_bound;
-        // scheduler.Wait(watch.tick);
+        scheduler.Wait(watch.tick);
         ++wait_cursor;
     }
 }