[AArch64] Add the constraint to NEON scalar mla/mls instructions.

llvm-svn: 193117

1 files changed, 24 insertions, 20 deletions

diff --git a/llvm/test/CodeGen/AArch64/neon-scalar-mul.ll b/llvm/test/CodeGen/AArch64/neon-scalar-mul.ll
index 3129df20f0e..a58294b209c 100644
--- a/llvm/test/CodeGen/AArch64/neon-scalar-mul.ll
+++ b/llvm/test/CodeGen/AArch64/neon-scalar-mul.ll
@@ -69,55 +69,59 @@ define double @test_vmulxd_f64(double %a, double %b) {
 declare <1 x float> @llvm.aarch64.neon.vmulx.v1f32(<1 x float>, <1 x float>)
 declare <1 x double> @llvm.aarch64.neon.vmulx.v1f64(<1 x double>, <1 x double>)
 
-define i32 @test_vqdmlalh_s16(i16 %a, i16 %b) {
+define i32 @test_vqdmlalh_s16(i32 %a, i16 %b, i16 %c) {
 ; CHECK: test_vqdmlalh_s16
 ; CHECK: sqdmlal {{s[0-9]+}}, {{h[0-9]+}}, {{h[0-9]+}}
 entry:
-  %vqdmlal.i = insertelement <1 x i16> undef, i16 %a, i32 0
+  %vqdmlal.i = insertelement <1 x i32> undef, i32 %a, i32 0
   %vqdmlal1.i = insertelement <1 x i16> undef, i16 %b, i32 0
-  %vqdmlal2.i = call <1 x i32> @llvm.aarch64.neon.vqdmlal.v1i32(<1 x i16> %vqdmlal.i, <1 x i16> %vqdmlal1.i)
-  %0 = extractelement <1 x i32> %vqdmlal2.i, i32 0
+  %vqdmlal2.i = insertelement <1 x i16> undef, i16 %c, i32 0
+  %vqdmlal3.i = call <1 x i32> @llvm.aarch64.neon.vqdmlal.v1i32(<1 x i32> %vqdmlal.i, <1 x i16> %vqdmlal1.i, <1 x i16> %vqdmlal2.i)
+  %0 = extractelement <1 x i32> %vqdmlal3.i, i32 0
   ret i32 %0
 }
 
-define i64 @test_vqdmlals_s32(i32 %a, i32 %b) {
+define i64 @test_vqdmlals_s32(i64 %a, i32 %b, i32 %c) {
 ; CHECK: test_vqdmlals_s32
 ; CHECK: sqdmlal {{d[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}
 entry:
-  %vqdmlal.i = insertelement <1 x i32> undef, i32 %a, i32 0
+  %vqdmlal.i = insertelement <1 x i64> undef, i64 %a, i32 0
   %vqdmlal1.i = insertelement <1 x i32> undef, i32 %b, i32 0
-  %vqdmlal2.i = call <1 x i64> @llvm.aarch64.neon.vqdmlal.v1i64(<1 x i32> %vqdmlal.i, <1 x i32> %vqdmlal1.i)
-  %0 = extractelement <1 x i64> %vqdmlal2.i, i32 0
+  %vqdmlal2.i = insertelement <1 x i32> undef, i32 %c, i32 0
+  %vqdmlal3.i = call <1 x i64> @llvm.aarch64.neon.vqdmlal.v1i64(<1 x i64> %vqdmlal.i, <1 x i32> %vqdmlal1.i, <1 x i32> %vqdmlal2.i)
+  %0 = extractelement <1 x i64> %vqdmlal3.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i32> @llvm.aarch64.neon.vqdmlal.v1i32(<1 x i16>, <1 x i16>)
-declare <1 x i64> @llvm.aarch64.neon.vqdmlal.v1i64(<1 x i32>, <1 x i32>)
+declare <1 x i32> @llvm.aarch64.neon.vqdmlal.v1i32(<1 x i32>, <1 x i16>, <1 x i16>)
+declare <1 x i64> @llvm.aarch64.neon.vqdmlal.v1i64(<1 x i64>, <1 x i32>, <1 x i32>)
 
-define i32 @test_vqdmlslh_s16(i16 %a, i16 %b) {
+define i32 @test_vqdmlslh_s16(i32 %a, i16 %b, i16 %c) {
 ; CHECK: test_vqdmlslh_s16
 ; CHECK: sqdmlsl {{s[0-9]+}}, {{h[0-9]+}}, {{h[0-9]+}}
 entry:
-  %vqdmlsl.i = insertelement <1 x i16> undef, i16 %a, i32 0
+  %vqdmlsl.i = insertelement <1 x i32> undef, i32 %a, i32 0
   %vqdmlsl1.i = insertelement <1 x i16> undef, i16 %b, i32 0
-  %vqdmlsl2.i = call <1 x i32> @llvm.aarch64.neon.vqdmlsl.v1i32(<1 x i16> %vqdmlsl.i, <1 x i16> %vqdmlsl1.i)
-  %0 = extractelement <1 x i32> %vqdmlsl2.i, i32 0
+  %vqdmlsl2.i = insertelement <1 x i16> undef, i16 %c, i32 0
+  %vqdmlsl3.i = call <1 x i32> @llvm.aarch64.neon.vqdmlsl.v1i32(<1 x i32> %vqdmlsl.i, <1 x i16> %vqdmlsl1.i, <1 x i16> %vqdmlsl2.i)
+  %0 = extractelement <1 x i32> %vqdmlsl3.i, i32 0
   ret i32 %0
 }
 
-define i64 @test_vqdmlsls_s32(i32 %a, i32 %b) {
+define i64 @test_vqdmlsls_s32(i64 %a, i32 %b, i32 %c) {
 ; CHECK: test_vqdmlsls_s32
 ; CHECK: sqdmlsl {{d[0-9]+}}, {{s[0-9]+}}, {{s[0-9]+}}
 entry:
-  %vqdmlsl.i = insertelement <1 x i32> undef, i32 %a, i32 0
+  %vqdmlsl.i = insertelement <1 x i64> undef, i64 %a, i32 0
   %vqdmlsl1.i = insertelement <1 x i32> undef, i32 %b, i32 0
-  %vqdmlsl2.i = call <1 x i64> @llvm.aarch64.neon.vqdmlsl.v1i64(<1 x i32> %vqdmlsl.i, <1 x i32> %vqdmlsl1.i)
-  %0 = extractelement <1 x i64> %vqdmlsl2.i, i32 0
+  %vqdmlsl2.i = insertelement <1 x i32> undef, i32 %c, i32 0
+  %vqdmlsl3.i = call <1 x i64> @llvm.aarch64.neon.vqdmlsl.v1i64(<1 x i64> %vqdmlsl.i, <1 x i32> %vqdmlsl1.i, <1 x i32> %vqdmlsl2.i)
+  %0 = extractelement <1 x i64> %vqdmlsl3.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i32> @llvm.aarch64.neon.vqdmlsl.v1i32(<1 x i16>, <1 x i16>)
-declare <1 x i64> @llvm.aarch64.neon.vqdmlsl.v1i64(<1 x i32>, <1 x i32>)
+declare <1 x i32> @llvm.aarch64.neon.vqdmlsl.v1i32(<1 x i32>, <1 x i16>, <1 x i16>)
+declare <1 x i64> @llvm.aarch64.neon.vqdmlsl.v1i64(<1 x i64>, <1 x i32>, <1 x i32>)
 
 define i32 @test_vqdmullh_s16(i16 %a, i16 %b) {
 ; CHECK: test_vqdmullh_s16