shader/memory: Implement RED.E.ADD

Implements a reduction operation. It's an atomic operation that doesn't
return a value.

This commit introduces another primitive because some shading languages
might have a primitive for reduction operations.

2 files changed, 29 insertions, 1 deletions

diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index 1a93540fe..8112ead3e 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -378,13 +378,27 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
 
             if (IsUnaligned(type)) {
                 const u32 mask = GetUnalignedMask(type);
-                value = InsertUnaligned(gmem, std::move(value), real_address, mask, size);
+                value = InsertUnaligned(gmem, move(value), real_address, mask, size);
             }
 
             bb.push_back(Operation(OperationCode::Assign, gmem, value));
         }
         break;
     }
+    case OpCode::Id::RED: {
+        UNIMPLEMENTED_IF_MSG(instr.red.type != GlobalAtomicType::U32);
+        UNIMPLEMENTED_IF_MSG(instr.red.operation != AtomicOp::Add);
+        const auto [real_address, base_address, descriptor] =
+            TrackGlobalMemory(bb, instr, true, true);
+        if (!real_address || !base_address) {
+            // Tracking failed, skip atomic.
+            break;
+        }
+        Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
+        Node value = GetRegister(instr.gpr0);
+        bb.push_back(Operation(OperationCode::ReduceIAdd, move(gmem), move(value)));
+        break;
+    }
     case OpCode::Id::ATOM: {
         UNIMPLEMENTED_IF_MSG(instr.atom.operation == AtomicOp::Inc ||
                                  instr.atom.operation == AtomicOp::Dec ||
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index 5fcc9da60..3eee961f5 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -178,6 +178,20 @@ enum class OperationCode {
     AtomicIOr,       /// (memory, int) -> int
     AtomicIXor,      /// (memory, int) -> int
 
+    ReduceUAdd, /// (memory, uint) -> void
+    ReduceUMin, /// (memory, uint) -> void
+    ReduceUMax, /// (memory, uint) -> void
+    ReduceUAnd, /// (memory, uint) -> void
+    ReduceUOr,  /// (memory, uint) -> void
+    ReduceUXor, /// (memory, uint) -> void
+
+    ReduceIAdd, /// (memory, int) -> void
+    ReduceIMin, /// (memory, int) -> void
+    ReduceIMax, /// (memory, int) -> void
+    ReduceIAnd, /// (memory, int) -> void
+    ReduceIOr,  /// (memory, int) -> void
+    ReduceIXor, /// (memory, int) -> void
+
     Branch,         /// (uint branch_target) -> void
     BranchIndirect, /// (uint branch_target) -> void
     PushFlowStack,  /// (uint branch_target) -> void