From 5a50f8b39f4e02cc05e78f1faf3c82c9c7449026 Mon Sep 17 00:00:00 2001
From: Amara Emerson <amara@apple.com>
Date: Fri, 18 Sep 2020 23:31:30 -0700
Subject: [PATCH] [AArch64][GlobalISel] Add legalization and selection support
 for <4 x s16> G_SHL.

---
 .../AArch64/GISel/AArch64InstructionSelector.cpp    |  2 ++
 .../Target/AArch64/GISel/AArch64LegalizerInfo.cpp   | 18 +++++++++++-------
 .../CodeGen/AArch64/GlobalISel/legalize-shift.mir   | 14 ++++++++++++++
 .../AArch64/GlobalISel/select-vector-shift.mir      | 21 +++++++++++++++++++++
 4 files changed, 48 insertions(+), 7 deletions(-)

diff --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
index 2a85ede..5469456 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -1493,6 +1493,8 @@ bool AArch64InstructionSelector::selectVectorSHL(
     Opc = ImmVal ? AArch64::SHLv4i32_shift : AArch64::USHLv4i32;
   } else if (Ty == LLT::vector(2, 32)) {
     Opc = ImmVal ? AArch64::SHLv2i32_shift : AArch64::USHLv2i32;
+  } else if (Ty == LLT::vector(4, 16)) {
+    Opc = ImmVal ? AArch64::SHLv4i16_shift : AArch64::USHLv4i16;
   } else {
     LLVM_DEBUG(dbgs() << "Unhandled G_SHL type");
     return false;
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index b69a70b..bbdd238 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -98,13 +98,17 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
       .moreElementsToNextPow2(0);
 
   getActionDefinitionsBuilder(G_SHL)
-      .legalFor({{s32, s32},
-                 {s64, s64},
-                 {v2s32, v2s32},
-                 {v4s32, v4s32},
-                 {v2s64, v2s64},
-                 {v16s8, v16s8},
-                 {v8s16, v8s16}})
+      .legalFor({
+          {s32, s32},
+          {s64, s64},
+          {v16s8, v16s8},
+          {v4s16, v4s16},
+          {v8s16, v8s16},
+          {v2s32, v2s32},
+          {v4s32, v4s32},
+          {v2s64, v2s64},
+
+      })
       .clampScalar(1, s32, s64)
       .clampScalar(0, s32, s64)
       .widenScalarToNextPow2(0)
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-shift.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-shift.mir
index 05cb4cb..ce7c8e7 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-shift.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-shift.mir
@@ -368,3 +368,17 @@ body:             |
     %2:_(<8 x s16>) = G_LSHR %0, %1
     $q0 = COPY %2
 ...
+---
+name: test_shl_v4i16
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name: test_shl_v4i16
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $d0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $d1
+    ; CHECK: [[SHL:%[0-9]+]]:_(<4 x s16>) = G_SHL [[COPY]], [[COPY1]](<4 x s16>)
+    ; CHECK: $d0 = COPY [[SHL]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $d0
+    %1:_(<4 x s16>) = COPY $d1
+    %2:_(<4 x s16>) = G_SHL %0, %1
+    $d0 = COPY %2
+...
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/select-vector-shift.mir b/llvm/test/CodeGen/AArch64/GlobalISel/select-vector-shift.mir
index 29e6f442..d0717ee 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/select-vector-shift.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/select-vector-shift.mir
@@ -351,3 +351,24 @@ body:             |
     RET_ReallyLR implicit $q0
 
 ...
+---
+name:            shl_v4i16
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.1:
+    liveins: $d0, $d1
+    ; CHECK-LABEL: name: shl_v4i16
+    ; CHECK: liveins: $d0, $d1
+    ; CHECK: [[COPY:%[0-9]+]]:fpr64 = COPY $d0
+    ; CHECK: [[COPY1:%[0-9]+]]:fpr64 = COPY $d1
+    ; CHECK: [[USHLv4i16_:%[0-9]+]]:fpr64 = USHLv4i16 [[COPY]], [[COPY1]]
+    ; CHECK: $d0 = COPY [[USHLv4i16_]]
+    ; CHECK: RET_ReallyLR implicit $d0
+    %0:fpr(<4 x s16>) = COPY $d0
+    %1:fpr(<4 x s16>) = COPY $d1
+    %2:fpr(<4 x s16>) = G_SHL %0, %1(<4 x s16>)
+    $d0 = COPY %2(<4 x s16>)
+    RET_ReallyLR implicit $d0
+...
-- 
2.7.4