From e7fba004ce02b6adbcc2e04d6adde18d63d0f228 Mon Sep 17 00:00:00 2001
From: Juergen Ributzka <juergen@apple.com>
Date: Thu, 18 Sep 2014 07:04:54 +0000
Subject: [PATCH] [FastISel][AArch64] Simplify XALU multiplies.

Simplify {s|u}mul.with.overflow to {s|u}add.with.overflow when possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218033 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64FastISel.cpp | 23 ++++++++++++++++++++++-
 test/CodeGen/AArch64/arm64-xaluo.ll    | 24 ++++++++++++++++++++++++
 2 files changed, 46 insertions(+), 1 deletion(-)

diff --git a/lib/Target/AArch64/AArch64FastISel.cpp b/lib/Target/AArch64/AArch64FastISel.cpp
index cb0fa958a5a..e64f1bdacd8 100644
--- a/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/lib/Target/AArch64/AArch64FastISel.cpp
@@ -3050,9 +3050,30 @@ bool AArch64FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
         isCommutativeIntrinsic(II))
       std::swap(LHS, RHS);
 
+    // Simplify multiplies.
+    unsigned IID = II->getIntrinsicID();
+    switch (IID) {
+    default:
+      break;
+    case Intrinsic::smul_with_overflow:
+      if (const auto *C = dyn_cast<ConstantInt>(RHS))
+        if (C->getValue() == 2) {
+          IID = Intrinsic::sadd_with_overflow;
+          RHS = LHS;
+        }
+      break;
+    case Intrinsic::umul_with_overflow:
+      if (const auto *C = dyn_cast<ConstantInt>(RHS))
+        if (C->getValue() == 2) {
+          IID = Intrinsic::uadd_with_overflow;
+          RHS = LHS;
+        }
+      break;
+    }
+
     unsigned ResultReg1 = 0, ResultReg2 = 0, MulReg = 0;
     AArch64CC::CondCode CC = AArch64CC::Invalid;
-    switch (II->getIntrinsicID()) {
+    switch (IID) {
     default: llvm_unreachable("Unexpected intrinsic!");
     case Intrinsic::sadd_with_overflow:
       ResultReg1 = emitAdd(VT, LHS, RHS, /*SetFlags=*/true);
diff --git a/test/CodeGen/AArch64/arm64-xaluo.ll b/test/CodeGen/AArch64/arm64-xaluo.ll
index d23eb877b99..687a2f735b2 100644
--- a/test/CodeGen/AArch64/arm64-xaluo.ll
+++ b/test/CodeGen/AArch64/arm64-xaluo.ll
@@ -217,6 +217,18 @@ entry:
   ret i1 %obit
 }
 
+define zeroext i1 @smulo2.i64(i64 %v1, i64* %res) {
+entry:
+; CHECK-LABEL:  smulo2.i64
+; CHECK:        adds [[MREG:x[0-9]+]], x0, x0
+; CHECK-NEXT:   cset {{w[0-9]+}}, vs
+  %t = call {i64, i1} @llvm.smul.with.overflow.i64(i64 %v1, i64 2)
+  %val = extractvalue {i64, i1} %t, 0
+  %obit = extractvalue {i64, i1} %t, 1
+  store i64 %val, i64* %res
+  ret i1 %obit
+}
+
 define zeroext i1 @umulo.i32(i32 %v1, i32 %v2, i32* %res) {
 entry:
 ; CHECK-LABEL:  umulo.i32
@@ -243,6 +255,18 @@ entry:
   ret i1 %obit
 }
 
+define zeroext i1 @umulo2.i64(i64 %v1, i64* %res) {
+entry:
+; CHECK-LABEL:  umulo2.i64
+; CHECK:        adds [[MREG:x[0-9]+]], x0, x0
+; CHECK-NEXT:   cset {{w[0-9]+}}, hs
+  %t = call {i64, i1} @llvm.umul.with.overflow.i64(i64 %v1, i64 2)
+  %val = extractvalue {i64, i1} %t, 0
+  %obit = extractvalue {i64, i1} %t, 1
+  store i64 %val, i64* %res
+  ret i1 %obit
+}
+
 
 ;
 ; Check the use of the overflow bit in combination with a select instruction.
-- 
2.34.1