[AArch64] Add support for NEON scalar integer compare instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192596 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 128 insertions, 0 deletions

diff --git a/test/CodeGen/AArch64/neon-scalar-compare.ll b/test/CodeGen/AArch64/neon-scalar-compare.ll
new file mode 100644
index 0000000..831c10b
--- /dev/null
+++ b/test/CodeGen/AArch64/neon-scalar-compare.ll
@@ -0,0 +1,128 @@
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=+neon < %s | FileCheck %s
+
+;; Scalar Integer Compare
+
+define i64 @test_vceqd(i64 %a, i64 %b) {
+; CHECK: test_vceqd
+; CHECK: cmeq {{d[0-9]+}}, {{d[0-9]}}, {{d[0-9]}}
+entry:
+  %vceq.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vceq1.i = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vceq2.i = call <1 x i64> @llvm.aarch64.neon.vceq(<1 x i64> %vceq.i, <1 x i64> %vceq1.i)
+  %0 = extractelement <1 x i64> %vceq2.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vceqzd(i64 %a) {
+; CHECK: test_vceqzd
+; CHECK: cmeq {{d[0-9]}}, {{d[0-9]}}, #0x0
+entry:
+  %vceqz.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vceqz1.i = call <1 x i64> @llvm.aarch64.neon.vceq(<1 x i64> %vceqz.i, <1 x i64> zeroinitializer)
+  %0 = extractelement <1 x i64> %vceqz1.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vcged(i64 %a, i64 %b) {
+; CHECK: test_vcged
+; CHECK: cmge {{d[0-9]}}, {{d[0-9]}}, {{d[0-9]}}
+entry:
+  %vcge.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vcge1.i = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vcge2.i = call <1 x i64> @llvm.aarch64.neon.vcge(<1 x i64> %vcge.i, <1 x i64> %vcge1.i)
+  %0 = extractelement <1 x i64> %vcge2.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vcgezd(i64 %a) {
+; CHECK: test_vcgezd
+; CHECK: cmge {{d[0-9]}}, {{d[0-9]}}, #0x0
+entry:
+  %vcgez.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vcgez1.i = call <1 x i64> @llvm.aarch64.neon.vcge(<1 x i64> %vcgez.i, <1 x i64> zeroinitializer)
+  %0 = extractelement <1 x i64> %vcgez1.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vcgtd(i64 %a, i64 %b) {
+; CHECK: test_vcgtd
+; CHECK: cmgt {{d[0-9]}}, {{d[0-9]}}, {{d[0-9]}}
+entry:
+  %vcgt.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vcgt1.i = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vcgt2.i = call <1 x i64> @llvm.aarch64.neon.vcgt(<1 x i64> %vcgt.i, <1 x i64> %vcgt1.i)
+  %0 = extractelement <1 x i64> %vcgt2.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vcgtzd(i64 %a) {
+; CHECK: test_vcgtzd
+; CHECK: cmgt {{d[0-9]}}, {{d[0-9]}}, #0x0
+entry:
+  %vcgtz.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vcgtz1.i = call <1 x i64> @llvm.aarch64.neon.vcgt(<1 x i64> %vcgtz.i, <1 x i64> zeroinitializer)
+  %0 = extractelement <1 x i64> %vcgtz1.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vcled(i64 %a, i64 %b) {
+; CHECK: test_vcled
+; CHECK: cmgt {{d[0-9]}}, {{d[0-9]}}, {{d[0-9]}}
+entry:
+  %vcgt.i = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vcgt1.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vcgt2.i = call <1 x i64> @llvm.aarch64.neon.vcgt(<1 x i64> %vcgt.i, <1 x i64> %vcgt1.i)
+  %0 = extractelement <1 x i64> %vcgt2.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vclezd(i64 %a) {
+; CHECK: test_vclezd
+; CHECK: cmle {{d[0-9]}}, {{d[0-9]}}, #0x0
+entry:
+  %vclez.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vclez1.i = call <1 x i64> @llvm.aarch64.neon.vclez(<1 x i64> %vclez.i, <1 x i64> zeroinitializer)
+  %0 = extractelement <1 x i64> %vclez1.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vcltd(i64 %a, i64 %b) {
+; CHECK: test_vcltd
+; CHECK: cmge {{d[0-9]}}, {{d[0-9]}}, {{d[0-9]}}
+entry:
+  %vcge.i = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vcge1.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vcge2.i = call <1 x i64> @llvm.aarch64.neon.vcge(<1 x i64> %vcge.i, <1 x i64> %vcge1.i)
+  %0 = extractelement <1 x i64> %vcge2.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vcltzd(i64 %a) {
+; CHECK: test_vcltzd
+; CHECK: cmlt {{d[0-9]}}, {{d[0-9]}}, #0x0
+entry:
+  %vcltz.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vcltz1.i = call <1 x i64> @llvm.aarch64.neon.vcltz(<1 x i64> %vcltz.i, <1 x i64> zeroinitializer)
+  %0 = extractelement <1 x i64> %vcltz1.i, i32 0
+  ret i64 %0
+}
+
+define i64 @test_vtstd(i64 %a, i64 %b) {
+; CHECK: test_vtstd
+; CHECK: cmtst {{d[0-9]}}, {{d[0-9]}}, {{d[0-9]}}
+entry:
+  %vtst.i = insertelement <1 x i64> undef, i64 %a, i32 0
+  %vtst1.i = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vtst2.i = call <1 x i64> @llvm.aarch64.neon.vtstd(<1 x i64> %vtst.i, <1 x i64> %vtst1.i)
+  %0 = extractelement <1 x i64> %vtst2.i, i32 0
+  ret i64 %0
+}
+
+declare <1 x i64> @llvm.aarch64.neon.vtstd(<1 x i64>, <1 x i64>)
+declare <1 x i64> @llvm.aarch64.neon.vcltz(<1 x i64>, <1 x i64>)
+declare <1 x i64> @llvm.aarch64.neon.vchs(<1 x i64>, <1 x i64>)
+declare <1 x i64> @llvm.aarch64.neon.vcge(<1 x i64>, <1 x i64>)
+declare <1 x i64> @llvm.aarch64.neon.vclez(<1 x i64>, <1 x i64>)
+declare <1 x i64> @llvm.aarch64.neon.vchi(<1 x i64>, <1 x i64>)
+declare <1 x i64> @llvm.aarch64.neon.vcgt(<1 x i64>, <1 x i64>)
+declare <1 x i64> @llvm.aarch64.neon.vceq(<1 x i64>, <1 x i64>)