1 files changed, 55 insertions, 29 deletions

diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index 6a992c543..e6a010a7d 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -12,8 +12,6 @@
 #include "video_core/engines/shader_bytecode.h"
 #include "video_core/shader/shader_ir.h"
 
-#pragma optimize("", off)
-
 namespace VideoCommon::Shader {
 
 using Tegra::Shader::Attribute;
@@ -148,12 +146,25 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
         break;
     }
+    case OpCode::Id::LD:
     case OpCode::Id::LDG: {
+        const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType {
+            switch (opcode->get().GetId()) {
+            case OpCode::Id::LD:
+                UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended LD is not implemented");
+                return instr.generic.type;
+            case OpCode::Id::LDG:
+                return instr.ldg.type;
+            default:
+                UNREACHABLE();
+                return {};
+            }
+        }();
+
         const auto [real_address_base, base_address, descriptor] =
-            TrackAndGetGlobalMemory(bb, GetRegister(instr.gpr8),
-                                    static_cast<u32>(instr.ldg.immediate_offset.Value()), false);
+            TrackAndGetGlobalMemory(bb, instr, false);
 
-        const u32 count = GetUniformTypeElementsCount(instr.ldg.type);
+        const u32 count = GetUniformTypeElementsCount(type);
         for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
             const Node real_address =
@@ -167,28 +178,6 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
         break;
     }
-    case OpCode::Id::STG: {
-        const auto [real_address_base, base_address, descriptor] =
-            TrackAndGetGlobalMemory(bb, GetRegister(instr.gpr8),
-                                    static_cast<u32>(instr.stg.immediate_offset.Value()), true);
-
-        // Encode in temporary registers like this: real_base_address, {registers_to_be_written...}
-        SetTemporal(bb, 0, real_address_base);
-
-        const u32 count = GetUniformTypeElementsCount(instr.stg.type);
-        for (u32 i = 0; i < count; ++i) {
-            SetTemporal(bb, i + 1, GetRegister(instr.gpr0.Value() + i));
-        }
-        for (u32 i = 0; i < count; ++i) {
-            const Node it_offset = Immediate(i * 4);
-            const Node real_address =
-                Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
-            const Node gmem = StoreNode(GmemNode(real_address, base_address, descriptor));
-
-            bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporal(i + 1)));
-        }
-        break;
-    }
     case OpCode::Id::ST_A: {
         UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex,
                              "Indirect attribute loads are not supported");
@@ -244,6 +233,41 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
         break;
     }
+    case OpCode::Id::ST:
+    case OpCode::Id::STG: {
+        const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType {
+            switch (opcode->get().GetId()) {
+            case OpCode::Id::ST:
+                UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended ST is not implemented");
+                return instr.generic.type;
+            case OpCode::Id::STG:
+                return instr.stg.type;
+            default:
+                UNREACHABLE();
+                return {};
+            }
+        }();
+
+        const auto [real_address_base, base_address, descriptor] =
+            TrackAndGetGlobalMemory(bb, instr, true);
+
+        // Encode in temporary registers like this: real_base_address, {registers_to_be_written...}
+        SetTemporal(bb, 0, real_address_base);
+
+        const u32 count = GetUniformTypeElementsCount(type);
+        for (u32 i = 0; i < count; ++i) {
+            SetTemporal(bb, i + 1, GetRegister(instr.gpr0.Value() + i));
+        }
+        for (u32 i = 0; i < count; ++i) {
+            const Node it_offset = Immediate(i * 4);
+            const Node real_address =
+                Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
+            const Node gmem = StoreNode(GmemNode(real_address, base_address, descriptor));
+
+            bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporal(i + 1)));
+        }
+        break;
+    }
     case OpCode::Id::AL2P: {
         // Ignore al2p.direction since we don't care about it.
 
@@ -267,9 +291,11 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
 }
 
 std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackAndGetGlobalMemory(NodeBlock& bb,
-                                                                           Node addr_register,
-                                                                           u32 immediate_offset,
+                                                                           Instruction instr,
                                                                            bool is_write) {
+    const auto addr_register{GetRegister(instr.gmem.gpr)};
+    const auto immediate_offset{static_cast<u32>(instr.gmem.offset)};
+
     const Node base_address{
         TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()))};
     const auto cbuf = std::get_if<CbufNode>(base_address);