[llvm] r235148 - [AArch64] Don't assert on f16 in DUP PerfectShuffle generator.

[Ahmed Bougacha]

Author: ab
Date: Thu Apr 16 18:57:07 2015
New Revision: 235148

URL: http://llvm.org/viewvc/llvm-project?rev=235148&view=rev
Log:
[AArch64] Don't assert on f16 in DUP PerfectShuffle generator.

Found by code inspection, but breaking i16 at least breaks other tests.
They aren't checking this in particular though, so also add some
explicit tests for the already working types.

Modified:
    llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp
    llvm/trunk/test/CodeGen/AArch64/arm64-dup.ll

Modified: llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp?rev=235148&r1=235147&r2=235148&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp Thu Apr 16 18:57:07 2015
@@ -5046,7 +5046,7 @@ static SDValue GeneratePerfectShuffle(un
     unsigned Opcode;
     if (EltTy == MVT::i8)
       Opcode = AArch64ISD::DUPLANE8;
-    else if (EltTy == MVT::i16)
+    else if (EltTy == MVT::i16 || EltTy == MVT::f16)
       Opcode = AArch64ISD::DUPLANE16;
     else if (EltTy == MVT::i32 || EltTy == MVT::f32)
       Opcode = AArch64ISD::DUPLANE32;

Modified: llvm/trunk/test/CodeGen/AArch64/arm64-dup.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/arm64-dup.ll?rev=235148&r1=235147&r2=235148&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AArch64/arm64-dup.ll (original)
+++ llvm/trunk/test/CodeGen/AArch64/arm64-dup.ll Thu Apr 16 18:57:07 2015
@@ -321,3 +321,40 @@ entry:
   %sub = sub <4 x i16> %a, %mul
   ret <4 x i16> %sub
 }
+
+; Also test the DUP path in the PerfectShuffle generator.
+
+; CHECK-LABEL: test_perfectshuffle_dupext_v4i16:
+; CHECK-NEXT: dup.4h v0, v0[0]
+; CHECK-NEXT: ext.8b v0, v0, v1, #4
+define <4 x i16> @test_perfectshuffle_dupext_v4i16(<4 x i16> %a, <4 x i16> %b) nounwind {
+  %r = shufflevector <4 x i16> %a, <4 x i16> %b, <4 x i32> <i32 0, i32 0, i32 4, i32 5>
+  ret <4 x i16> %r
+}
+
+; CHECK-LABEL: test_perfectshuffle_dupext_v4f16:
+; CHECK-NEXT: dup.4h v0, v0[0]
+; CHECK-NEXT: ext.8b v0, v0, v1, #4
+; CHECK-NEXT: ret
+define <4 x half> @test_perfectshuffle_dupext_v4f16(<4 x half> %a, <4 x half> %b) nounwind {
+  %r = shufflevector <4 x half> %a, <4 x half> %b, <4 x i32> <i32 0, i32 0, i32 4, i32 5>
+  ret <4 x half> %r
+}
+
+; CHECK-LABEL: test_perfectshuffle_dupext_v4i32:
+; CHECK-NEXT: dup.4s v0, v0[0]
+; CHECK-NEXT: ext.16b v0, v0, v1, #8
+; CHECK-NEXT: ret
+define <4 x i32> @test_perfectshuffle_dupext_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
+  %r = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 0, i32 4, i32 5>
+  ret <4 x i32> %r
+}
+
+; CHECK-LABEL: test_perfectshuffle_dupext_v4f32:
+; CHECK-NEXT: dup.4s v0, v0[0]
+; CHECK-NEXT: ext.16b v0, v0, v1, #8
+; CHECK-NEXT: ret
+define <4 x float> @test_perfectshuffle_dupext_v4f32(<4 x float> %a, <4 x float> %b) nounwind {
+  %r = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 4, i32 5>
+  ret <4 x float> %r
+}