2 files changed, 36 insertions, 21 deletions

diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 073b622ef..e83f25fa1 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -1673,6 +1673,7 @@ private:
             INST("1110111111110---", Id::ST_A, Type::Memory, "ST_A"),
             INST("1110111101011---", Id::ST_S, Type::Memory, "ST_S"),
             INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"),
+            INST("101-------------", Id::ST, Type::Memory, "ST"),
             INST("1110111011011---", Id::STG, Type::Memory, "STG"),
             INST("1110111110100---", Id::AL2P, Type::Memory, "AL2P"),
             INST("110000----111---", Id::TEX, Type::Texture, "TEX"),
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index 8ac2fd4ac..a5ca9a164 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -180,27 +180,6 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
         break;
     }
-    case OpCode::Id::STG: {
-        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(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");
@@ -256,6 +235,41 @@ 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.