[AArch64] Add support for NEON scalar floating-point convert to fixed-point instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194394 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 56 insertions, 8 deletions

diff --git a/test/CodeGen/AArch64/neon-scalar-cvt.ll b/test/CodeGen/AArch64/neon-scalar-cvt.ll
index 056504a67e..a7f0ac0965 100644
--- a/test/CodeGen/AArch64/neon-scalar-cvt.ll
+++ b/test/CodeGen/AArch64/neon-scalar-cvt.ll
@@ -5,7 +5,7 @@ define float @test_vcvts_f32_s32(i32 %a) {
 ; CHECK: scvtf {{s[0-9]+}}, {{s[0-9]+}}
 entry:
   %vcvtf.i = insertelement <1 x i32> undef, i32 %a, i32 0
-  %vcvtf1.i = tail call <1 x float> @llvm.aarch64.neon.vcvtf32.s32(<1 x i32> %vcvtf.i)
+  %vcvtf1.i = call <1 x float> @llvm.aarch64.neon.vcvtf32.s32(<1 x i32> %vcvtf.i)
   %0 = extractelement <1 x float> %vcvtf1.i, i32 0
   ret float %0
 }
@@ -17,7 +17,7 @@ define double @test_vcvtd_f64_s64(i64 %a) {
 ; CHECK: scvtf {{d[0-9]+}}, {{d[0-9]+}}
 entry:
   %vcvtf.i = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vcvtf1.i = tail call <1 x double> @llvm.aarch64.neon.vcvtf64.s64(<1 x i64> %vcvtf.i)
+  %vcvtf1.i = call <1 x double> @llvm.aarch64.neon.vcvtf64.s64(<1 x i64> %vcvtf.i)
   %0 = extractelement <1 x double> %vcvtf1.i, i32 0
   ret double %0
 }
@@ -29,7 +29,7 @@ define float @test_vcvts_f32_u32(i32 %a) {
 ; CHECK: ucvtf {{s[0-9]+}}, {{s[0-9]+}}
 entry:
   %vcvtf.i = insertelement <1 x i32> undef, i32 %a, i32 0
-  %vcvtf1.i = tail call <1 x float> @llvm.aarch64.neon.vcvtf32.u32(<1 x i32> %vcvtf.i)
+  %vcvtf1.i = call <1 x float> @llvm.aarch64.neon.vcvtf32.u32(<1 x i32> %vcvtf.i)
   %0 = extractelement <1 x float> %vcvtf1.i, i32 0
   ret float %0
 }
@@ -41,7 +41,7 @@ define double @test_vcvtd_f64_u64(i64 %a) {
 ; CHECK: ucvtf {{d[0-9]+}}, {{d[0-9]+}}
 entry:
   %vcvtf.i = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vcvtf1.i = tail call <1 x double> @llvm.aarch64.neon.vcvtf64.u64(<1 x i64> %vcvtf.i)
+  %vcvtf1.i = call <1 x double> @llvm.aarch64.neon.vcvtf64.u64(<1 x i64> %vcvtf.i)
   %0 = extractelement <1 x double> %vcvtf1.i, i32 0
   ret double %0
 }
@@ -53,7 +53,7 @@ define float @test_vcvts_n_f32_s32(i32 %a) {
 ; CHECK: scvtf {{s[0-9]+}}, {{s[0-9]+}}, #1
 entry:
   %vcvtf = insertelement <1 x i32> undef, i32 %a, i32 0
-  %vcvtf1 = tail call <1 x float> @llvm.aarch64.neon.vcvtf32.n.s32(<1 x i32> %vcvtf, i32 1)
+  %vcvtf1 = call <1 x float> @llvm.aarch64.neon.vcvtf32.n.s32(<1 x i32> %vcvtf, i32 1)
   %0 = extractelement <1 x float> %vcvtf1, i32 0
   ret float %0
 }
@@ -65,7 +65,7 @@ define double @test_vcvtd_n_f64_s64(i64 %a) {
 ; CHECK: scvtf {{d[0-9]+}}, {{d[0-9]+}}, #1
 entry:
   %vcvtf = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vcvtf1 = tail call <1 x double> @llvm.aarch64.neon.vcvtf64.n.s64(<1 x i64> %vcvtf, i32 1)
+  %vcvtf1 = call <1 x double> @llvm.aarch64.neon.vcvtf64.n.s64(<1 x i64> %vcvtf, i32 1)
   %0 = extractelement <1 x double> %vcvtf1, i32 0
   ret double %0
 }
@@ -77,7 +77,7 @@ define float @test_vcvts_n_f32_u32(i32 %a) {
 ; CHECK: ucvtf {{s[0-9]+}}, {{s[0-9]+}}, #1
 entry:
   %vcvtf = insertelement <1 x i32> undef, i32 %a, i32 0
-  %vcvtf1 = tail call <1 x float> @llvm.aarch64.neon.vcvtf32.n.u32(<1 x i32> %vcvtf, i32 1)
+  %vcvtf1 = call <1 x float> @llvm.aarch64.neon.vcvtf32.n.u32(<1 x i32> %vcvtf, i32 1)
   %0 = extractelement <1 x float> %vcvtf1, i32 0
   ret float %0
 }
@@ -89,9 +89,57 @@ define double @test_vcvtd_n_f64_u64(i64 %a) {
 ; CHECK: ucvtf {{d[0-9]+}}, {{d[0-9]+}}, #1
 entry:
   %vcvtf = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vcvtf1 = tail call <1 x double> @llvm.aarch64.neon.vcvtf64.n.u64(<1 x i64> %vcvtf, i32 1)
+  %vcvtf1 = call <1 x double> @llvm.aarch64.neon.vcvtf64.n.u64(<1 x i64> %vcvtf, i32 1)
   %0 = extractelement <1 x double> %vcvtf1, i32 0
   ret double %0
 }
 
 declare <1 x double> @llvm.aarch64.neon.vcvtf64.n.u64(<1 x i64>, i32)
+
+define i32 @test_vcvts_n_s32_f32(float %a) {
+; CHECK: test_vcvts_n_s32_f32
+; CHECK: fcvtzs {{s[0-9]+}}, {{s[0-9]+}}, #0
+entry:
+  %fcvtzs = insertelement <1 x float> undef, float %a, i32 0
+  %fcvtzs1 = call <1 x i32> @llvm.aarch64.neon.vcvts.n.s32.f32(<1 x float> %fcvtzs, i32 0)
+  %0 = extractelement <1 x i32> %fcvtzs1, i32 0
+  ret i32 %0
+}
+
+declare <1 x i32> @llvm.aarch64.neon.vcvts.n.s32.f32(<1 x float>, i32)
+
+define i64 @test_vcvtd_n_s64_f64(double %a) {
+; CHECK: test_vcvtd_n_s64_f64
+; CHECK: fcvtzs {{d[0-9]+}}, {{d[0-9]+}}, #0
+entry:
+  %fcvtzs = insertelement <1 x double> undef, double %a, i32 0
+  %fcvtzs1 = call <1 x i64> @llvm.aarch64.neon.vcvtd.n.s64.f64(<1 x double> %fcvtzs, i32 0)
+  %0 = extractelement <1 x i64> %fcvtzs1, i32 0
+  ret i64 %0
+}
+
+declare <1 x i64> @llvm.aarch64.neon.vcvtd.n.s64.f64(<1 x double>, i32)
+
+define i32 @test_vcvts_n_u32_f32(float %a) {
+; CHECK: test_vcvts_n_u32_f32
+; CHECK: fcvtzu {{s[0-9]+}}, {{s[0-9]+}}, #0
+entry:
+  %fcvtzu = insertelement <1 x float> undef, float %a, i32 0
+  %fcvtzu1 = call <1 x i32> @llvm.aarch64.neon.vcvts.n.u32.f32(<1 x float> %fcvtzu, i32 0)
+  %0 = extractelement <1 x i32> %fcvtzu1, i32 0
+  ret i32 %0
+}
+
+declare <1 x i32> @llvm.aarch64.neon.vcvts.n.u32.f32(<1 x float>, i32)
+
+define i64 @test_vcvtd_n_u64_f64(double %a) {
+; CHECK: test_vcvtd_n_u64_f64
+; CHECK: fcvtzu {{d[0-9]+}}, {{d[0-9]+}}, #0
+entry:
+  %fcvtzu = insertelement <1 x double> undef, double %a, i32 0
+  %fcvtzu1 = tail call <1 x i64> @llvm.aarch64.neon.vcvtd.n.u64.f64(<1 x double> %fcvtzu, i32 0)
+  %0 = extractelement <1 x i64> %fcvtzu1, i32 0
+  ret i64 %0
+}
+
+declare <1 x i64> @llvm.aarch64.neon.vcvtd.n.u64.f64(<1 x double>, i32)