From ac7908164a2c5f7b127f3dc17105ea279d505a03 Mon Sep 17 00:00:00 2001
From: Merry <git@mary.rs>
Date: Wed, 27 Jul 2022 20:08:00 +0100
Subject: [PATCH] emit_arm64_packed: Implement packed halving operations

---
 .../backend/arm64/emit_arm64_packed.cpp       | 52 +++++++------------
 1 file changed, 20 insertions(+), 32 deletions(-)

diff --git a/src/dynarmic/backend/arm64/emit_arm64_packed.cpp b/src/dynarmic/backend/arm64/emit_arm64_packed.cpp
index c5d16584..5876910b 100644
--- a/src/dynarmic/backend/arm64/emit_arm64_packed.cpp
+++ b/src/dynarmic/backend/arm64/emit_arm64_packed.cpp
@@ -18,6 +18,18 @@ namespace Dynarmic::Backend::Arm64 {
 
 using namespace oaknut::util;
 
+template<typename EmitFn>
+static void EmitPackedOp(oaknut::CodeGenerator&, EmitContext& ctx, IR::Inst* inst, EmitFn emit) {
+    auto args = ctx.reg_alloc.GetArgumentInfo(inst);
+
+    auto Vresult = ctx.reg_alloc.WriteD(inst);
+    auto Va = ctx.reg_alloc.ReadD(args[0]);
+    auto Vb = ctx.reg_alloc.ReadD(args[1]);
+    RegAlloc::Realize(Vresult, Va, Vb);
+
+    emit(Vresult, Va, Vb);
+}
+
 template<typename EmitFn>
 static void EmitSaturatedPackedOp(oaknut::CodeGenerator&, EmitContext& ctx, IR::Inst* inst, EmitFn emit) {
     auto args = ctx.reg_alloc.GetArgumentInfo(inst);
@@ -266,66 +278,42 @@ void EmitIR<IR::Opcode::PackedSubAddS16>(oaknut::CodeGenerator& code, EmitContex
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingAddU8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.UHADD(Vresult->B8(), Va->B8(), Vb->B8()); });
 }
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingAddS8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.SHADD(Vresult->B8(), Va->B8(), Vb->B8()); });
 }
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingSubU8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.UHSUB(Vresult->B8(), Va->B8(), Vb->B8()); });
 }
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingSubS8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.SHSUB(Vresult->B8(), Va->B8(), Vb->B8()); });
 }
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingAddU16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.UHADD(Vresult->H4(), Va->H4(), Vb->H4()); });
 }
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingAddS16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.SHADD(Vresult->H4(), Va->H4(), Vb->H4()); });
 }
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingSubU16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.UHSUB(Vresult->H4(), Va->H4(), Vb->H4()); });
 }
 
 template<>
 void EmitIR<IR::Opcode::PackedHalvingSubS16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
-    (void)code;
-    (void)ctx;
-    (void)inst;
-    ASSERT_FALSE("Unimplemented");
+    EmitPackedOp(code, ctx, inst, [&](auto& Vresult, auto& Va, auto& Vb) { code.SHSUB(Vresult->H4(), Va->H4(), Vb->H4()); });
 }
 
 template<>