shader_recompiler: Use shared memory buffer for non-compute shared memory.

CMakeLists.txt

@@ -754,7 +754,6 @@ set(SHADER_RECOMPILER src/shader_recompiler/exception.h
                       src/shader_recompiler/ir/passes/identity_removal_pass.cpp
                       src/shader_recompiler/ir/passes/ir_passes.h
                       src/shader_recompiler/ir/passes/lower_buffer_format_to_raw.cpp
-                      src/shader_recompiler/ir/passes/lower_shared_mem_to_registers.cpp
                       src/shader_recompiler/ir/passes/resource_tracking_pass.cpp
                       src/shader_recompiler/ir/passes/ring_access_elimination.cpp
                       src/shader_recompiler/ir/passes/shader_info_collection_pass.cpp

src/shader_recompiler/backend/spirv/spirv_emit_context.cpp

@@ -813,22 +813,22 @@ void EmitContext::DefineSharedMemory() {
     if (!info.uses_shared) {
         return;
     }
+
     const u32 max_shared_memory_size = profile.max_shared_memory_size;
     u32 shared_memory_size = runtime_info.cs_info.shared_memory_size;
     if (shared_memory_size == 0) {
         shared_memory_size = DefaultSharedMemSize;
     }
-
+    if (info.stage == Stage::Compute && shared_memory_size <= max_shared_memory_size) {
-    const u32 num_elements{Common::DivCeil(shared_memory_size, 4U)};
+        const u32 num_elements{Common::DivCeil(shared_memory_size, 4U)};
-    const Id type{TypeArray(U32[1], ConstU32(num_elements))};
+        const Id type{TypeArray(U32[1], ConstU32(num_elements))};
-
-    if (shared_memory_size <= max_shared_memory_size) {
         shared_memory_u32_type = TypePointer(spv::StorageClass::Workgroup, type);
         shared_u32 = TypePointer(spv::StorageClass::Workgroup, U32[1]);
         shared_memory_u32 = AddGlobalVariable(shared_memory_u32_type, spv::StorageClass::Workgroup);
         Name(shared_memory_u32, "shared_mem");
         interfaces.push_back(shared_memory_u32);
     } else {
+        const Id type{TypeRuntimeArray(U32[1])};
         shared_memory_u32_type = TypePointer(spv::StorageClass::StorageBuffer, type);
         shared_u32 = TypePointer(spv::StorageClass::StorageBuffer, U32[1]);
 
@@ -849,7 +849,6 @@ void EmitContext::DefineSharedMemory() {
         shared_memory_u32 = ssbo_id;
 
         info.has_emulated_shared_memory = true;
-        info.shared_memory_size = shared_memory_size;
         interfaces.push_back(ssbo_id);
     }
 }

src/shader_recompiler/info.h

@@ -196,7 +196,6 @@ struct Info {
     bool translation_failed{}; // indicates that shader has unsupported instructions
     bool has_emulated_shared_memory{};
     bool has_readconst{};
-    u32 shared_memory_size{};
     u8 mrt_mask{0u};
     bool has_fetch_shader{false};
     u32 fetch_shader_sgpr_base{0u};
@@ -233,7 +232,8 @@ struct Info {
     }
 
     void AddBindings(Backend::Bindings& bnd) const {
-        const auto total_buffers = buffers.size() + (has_readconst ? 1 : 0);
+        const auto total_buffers =
+            buffers.size() + (has_readconst ? 1 : 0) + (has_emulated_shared_memory ? 1 : 0);
         bnd.buffer += total_buffers;
         bnd.unified += total_buffers + images.size() + samplers.size();
         bnd.user_data += ud_mask.NumRegs();

src/shader_recompiler/ir/passes/ir_passes.h

@@ -20,7 +20,6 @@ void FlattenExtendedUserdataPass(IR::Program& program);
 void ResourceTrackingPass(IR::Program& program);
 void CollectShaderInfoPass(IR::Program& program);
 void LowerBufferFormatToRaw(IR::Program& program);
-void LowerSharedMemToRegisters(IR::Program& program);
 void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtime_info,
                            Stage stage);
 void TessellationPreprocess(IR::Program& program, RuntimeInfo& runtime_info);

src/shader_recompiler/ir/passes/lower_shared_mem_to_registers.cpp

@@ -1,38 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#include <boost/container/small_vector.hpp>
-#include "shader_recompiler/ir/program.h"
-
-namespace Shader::Optimization {
-
-void LowerSharedMemToRegisters(IR::Program& program) {
-    boost::container::small_vector<IR::Inst*, 8> ds_writes;
-    Info& info{program.info};
-    for (IR::Block* const block : program.blocks) {
-        for (IR::Inst& inst : block->Instructions()) {
-            const auto opcode = inst.GetOpcode();
-            if (opcode == IR::Opcode::WriteSharedU32 || opcode == IR::Opcode::WriteSharedU64) {
-                ds_writes.emplace_back(&inst);
-                continue;
-            }
-            if (opcode == IR::Opcode::LoadSharedU32 || opcode == IR::Opcode::LoadSharedU64) {
-                // Search for write instruction with same offset
-                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();
-                });
-                ASSERT(it != ds_writes.end());
-                // Replace data read with value written.
-                inst.ReplaceUsesWithAndRemove((*it)->Arg(1));
-            }
-        }
-    }
-    // We should have eliminated everything. Invalidate data write instructions.
-    for (const auto inst : ds_writes) {
-        inst->Invalidate();
-    }
-}
-
-} // namespace Shader::Optimization

src/shader_recompiler/recompiler.cpp

@@ -82,9 +82,6 @@ IR::Program TranslateProgram(std::span<const u32> code, Pools& pools, Info& info
     }
     Shader::Optimization::ConstantPropagationPass(program.post_order_blocks);
     Shader::Optimization::RingAccessElimination(program, runtime_info, stage);
-    if (stage != Stage::Compute) {
-        Shader::Optimization::LowerSharedMemToRegisters(program);
-    }
     Shader::Optimization::ConstantPropagationPass(program.post_order_blocks);
     Shader::Optimization::FlattenExtendedUserdataPass(program);
     Shader::Optimization::ResourceTrackingPass(program);

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -357,6 +357,14 @@ void GraphicsPipeline::BuildDescSetLayout() {
         if (!stage) {
             continue;
         }
+        if (stage->has_emulated_shared_memory) {
+            bindings.push_back({
+                .binding = binding++,
+                .descriptorType = vk::DescriptorType::eStorageBuffer,
+                .descriptorCount = 1,
+                .stageFlags = gp_stage_flags,
+            });
+        }
         if (stage->has_readconst) {
             bindings.push_back({
                 .binding = binding++,

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -535,6 +535,7 @@ void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Binding
             .descriptorType = vk::DescriptorType::eStorageBuffer,
             .pBufferInfo = &buffer_infos.back(),
         });
+        ++binding.buffer;
     }
 
     // Bind the flattened user data buffer as a UBO so it's accessible to the shader