[AArch64] Refactor the Neon vector/scalar floating-point convert intrinsics so

that they use float/double rather than the vector equivalents when appropriate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@196930 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 46 insertions, 68 deletions

diff --git a/test/CodeGen/AArch64/neon-misc.ll b/test/CodeGen/AArch64/neon-misc.ll
index 8510076482..ed0cc0fc41 100644
--- a/test/CodeGen/AArch64/neon-misc.ll
+++ b/test/CodeGen/AArch64/neon-misc.ll
@@ -894,13 +894,13 @@ define <4 x float> @test_vcvt_high_f32_f64(<2 x float> %a, <2 x double> %b) #0 {
 
 define <2 x float> @test_vcvtx_f32_f64(<2 x double> %a) #0 {
 ; CHECK: fcvtxn v{{[0-9]+}}.2s, v{{[0-9]+}}.2d
-  %vcvtx_f32_f641.i = tail call <2 x float> @llvm.aarch64.neon.fcvtxn.v2f32.v2f64(<2 x double> %a) #4
+  %vcvtx_f32_f641.i = call <2 x float> @llvm.aarch64.neon.vcvtxn.v2f32.v2f64(<2 x double> %a) #4
   ret <2 x float> %vcvtx_f32_f641.i
 }
 
 define <4 x float> @test_vcvtx_high_f32_f64(<2 x float> %a, <2 x double> %b) #0 {
 ; CHECK: fcvtxn2 v{{[0-9]+}}.4s, v{{[0-9]+}}.2d
-  %vcvtx_f32_f641.i.i = tail call <2 x float> @llvm.aarch64.neon.fcvtxn.v2f32.v2f64(<2 x double> %b) #4
+  %vcvtx_f32_f641.i.i = tail call <2 x float> @llvm.aarch64.neon.vcvtxn.v2f32.v2f64(<2 x double> %b) #4
   %shuffle.i = shufflevector <2 x float> %a, <2 x float> %vcvtx_f32_f641.i.i, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   ret <4 x float> %shuffle.i
 }
@@ -1462,7 +1462,7 @@ declare <4 x float> @llvm.aarch64.neon.frintn.v4f32(<4 x float>) #2
 
 declare <2 x float> @llvm.aarch64.neon.frintn.v2f32(<2 x float>) #2
 
-declare <2 x float> @llvm.aarch64.neon.fcvtxn.v2f32.v2f64(<2 x double>) #2
+declare <2 x float> @llvm.aarch64.neon.vcvtxn.v2f32.v2f64(<2 x double>) #2
 
 declare <2 x float> @llvm.aarch64.neon.fcvtn.v2f32.v2f64(<2 x double>) #2
 
diff --git a/test/CodeGen/AArch64/neon-scalar-fcvt.ll b/test/CodeGen/AArch64/neon-scalar-fcvt.ll
index 460be2c2c3..6cf30a7df3 100644
--- a/test/CodeGen/AArch64/neon-scalar-fcvt.ll
+++ b/test/CodeGen/AArch64/neon-scalar-fcvt.ll
@@ -6,250 +6,228 @@ define float @test_vcvtxn(double %a) {
 ; CHECK: test_vcvtxn
 ; CHECK: fcvtxn {{s[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtf.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtf1.i = tail call <1 x float> @llvm.aarch64.neon.fcvtxn.v1f32.v1f64(<1 x double> %vcvtf.i)
-  %0 = extractelement <1 x float> %vcvtf1.i, i32 0
-  ret float %0
+  %vcvtf = call float @llvm.aarch64.neon.fcvtxn(double %a)
+  ret float %vcvtf
 }
 
-declare <1 x float> @llvm.aarch64.neon.fcvtxn.v1f32.v1f64(<1 x double>)
+declare float @llvm.aarch64.neon.fcvtxn(double)
 
 define i32 @test_vcvtass(float %a) {
 ; CHECK: test_vcvtass
 ; CHECK: fcvtas {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtas.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtas1.i = call <1 x i32> @llvm.arm.neon.vcvtas.v1i32.v1f32(<1 x float> %vcvtas.i)
+  %vcvtas1.i = call <1 x i32> @llvm.aarch64.neon.fcvtas.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtas1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtas.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtas.v1i32.f32(float)
 
 define i64 @test_test_vcvtasd(double %a) {
 ; CHECK: test_test_vcvtasd
 ; CHECK: fcvtas {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtas.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtas1.i = call <1 x i64> @llvm.arm.neon.vcvtas.v1i64.v1f64(<1 x double> %vcvtas.i)
+  %vcvtas1.i = call <1 x i64> @llvm.aarch64.neon.fcvtas.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtas1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtas.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtas.v1i64.f64(double)
 
 define i32 @test_vcvtaus(float %a) {
 ; CHECK: test_vcvtaus
 ; CHECK: fcvtau {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtau.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtau1.i = call <1 x i32> @llvm.arm.neon.vcvtau.v1i32.v1f32(<1 x float> %vcvtau.i)
+  %vcvtau1.i = call <1 x i32> @llvm.aarch64.neon.fcvtau.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtau1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtau.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtau.v1i32.f32(float)
 
 define i64 @test_vcvtaud(double %a) {
 ; CHECK: test_vcvtaud
 ; CHECK: fcvtau {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtau.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtau1.i = call <1 x i64> @llvm.arm.neon.vcvtau.v1i64.v1f64(<1 x double> %vcvtau.i)
+  %vcvtau1.i = call <1 x i64> @llvm.aarch64.neon.fcvtau.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtau1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtau.v1i64.v1f64(<1 x double>) 
+declare <1 x i64> @llvm.aarch64.neon.fcvtau.v1i64.f64(double) 
 
 define i32 @test_vcvtmss(float %a) {
 ; CHECK: test_vcvtmss
 ; CHECK: fcvtms {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtms.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtms1.i = call <1 x i32> @llvm.arm.neon.vcvtms.v1i32.v1f32(<1 x float> %vcvtms.i)
+  %vcvtms1.i = call <1 x i32> @llvm.aarch64.neon.fcvtms.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtms1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtms.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtms.v1i32.f32(float)
 
 define i64 @test_vcvtmd_s64_f64(double %a) {
 ; CHECK: test_vcvtmd_s64_f64
 ; CHECK: fcvtms {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtms.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtms1.i = call <1 x i64> @llvm.arm.neon.vcvtms.v1i64.v1f64(<1 x double> %vcvtms.i)
+  %vcvtms1.i = call <1 x i64> @llvm.aarch64.neon.fcvtms.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtms1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtms.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtms.v1i64.f64(double)
 
 define i32 @test_vcvtmus(float %a) {
 ; CHECK: test_vcvtmus
 ; CHECK: fcvtmu {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtmu.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtmu1.i = call <1 x i32> @llvm.arm.neon.vcvtmu.v1i32.v1f32(<1 x float> %vcvtmu.i)
+  %vcvtmu1.i = call <1 x i32> @llvm.aarch64.neon.fcvtmu.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtmu1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtmu.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtmu.v1i32.f32(float)
 
 define i64 @test_vcvtmud(double %a) {
 ; CHECK: test_vcvtmud
 ; CHECK: fcvtmu {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtmu.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtmu1.i = call <1 x i64> @llvm.arm.neon.vcvtmu.v1i64.v1f64(<1 x double> %vcvtmu.i)
+  %vcvtmu1.i = call <1 x i64> @llvm.aarch64.neon.fcvtmu.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtmu1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtmu.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtmu.v1i64.f64(double)
 
 define i32 @test_vcvtnss(float %a) {
 ; CHECK: test_vcvtnss
 ; CHECK: fcvtns {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtns.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtns1.i = call <1 x i32> @llvm.arm.neon.vcvtns.v1i32.v1f32(<1 x float> %vcvtns.i)
+  %vcvtns1.i = call <1 x i32> @llvm.aarch64.neon.fcvtns.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtns1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtns.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtns.v1i32.f32(float)
 
 define i64 @test_vcvtnd_s64_f64(double %a) {
 ; CHECK: test_vcvtnd_s64_f64
 ; CHECK: fcvtns {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtns.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtns1.i = call <1 x i64> @llvm.arm.neon.vcvtns.v1i64.v1f64(<1 x double> %vcvtns.i)
+  %vcvtns1.i = call <1 x i64> @llvm.aarch64.neon.fcvtns.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtns1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtns.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtns.v1i64.f64(double)
 
 define i32 @test_vcvtnus(float %a) {
 ; CHECK: test_vcvtnus
 ; CHECK: fcvtnu {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtnu.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtnu1.i = call <1 x i32> @llvm.arm.neon.vcvtnu.v1i32.v1f32(<1 x float> %vcvtnu.i)
+  %vcvtnu1.i = call <1 x i32> @llvm.aarch64.neon.fcvtnu.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtnu1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtnu.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtnu.v1i32.f32(float)
 
 define i64 @test_vcvtnud(double %a) {
 ; CHECK: test_vcvtnud
 ; CHECK: fcvtnu {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtnu.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtnu1.i = call <1 x i64> @llvm.arm.neon.vcvtnu.v1i64.v1f64(<1 x double> %vcvtnu.i)
+  %vcvtnu1.i = call <1 x i64> @llvm.aarch64.neon.fcvtnu.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtnu1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtnu.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtnu.v1i64.f64(double)
 
 define i32 @test_vcvtpss(float %a) {
 ; CHECK: test_vcvtpss
 ; CHECK: fcvtps {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtps.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtps1.i = call <1 x i32> @llvm.arm.neon.vcvtps.v1i32.v1f32(<1 x float> %vcvtps.i)
+  %vcvtps1.i = call <1 x i32> @llvm.aarch64.neon.fcvtps.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtps1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtps.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtps.v1i32.f32(float)
 
 define i64 @test_vcvtpd_s64_f64(double %a) {
 ; CHECK: test_vcvtpd_s64_f64
 ; CHECK: fcvtps {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtps.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtps1.i = call <1 x i64> @llvm.arm.neon.vcvtps.v1i64.v1f64(<1 x double> %vcvtps.i)
+  %vcvtps1.i = call <1 x i64> @llvm.aarch64.neon.fcvtps.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtps1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtps.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtps.v1i64.f64(double)
 
 define i32 @test_vcvtpus(float %a) {
 ; CHECK: test_vcvtpus
 ; CHECK: fcvtpu {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtpu.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtpu1.i = call <1 x i32> @llvm.arm.neon.vcvtpu.v1i32.v1f32(<1 x float> %vcvtpu.i)
+  %vcvtpu1.i = call <1 x i32> @llvm.aarch64.neon.fcvtpu.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtpu1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.arm.neon.vcvtpu.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtpu.v1i32.f32(float)
 
 define i64 @test_vcvtpud(double %a) {
 ; CHECK: test_vcvtpud
 ; CHECK: fcvtpu {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtpu.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtpu1.i = tail call <1 x i64> @llvm.arm.neon.vcvtpu.v1i64.v1f64(<1 x double> %vcvtpu.i)
+  %vcvtpu1.i = call <1 x i64> @llvm.aarch64.neon.fcvtpu.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtpu1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.arm.neon.vcvtpu.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtpu.v1i64.f64(double)
 
 define i32 @test_vcvtss(float %a) {
 ; CHECK: test_vcvtss
 ; CHECK: fcvtzs {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtzs.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtzs1.i = tail call <1 x i32> @llvm.aarch64.neon.fcvtzs.v1i32.v1f32(<1 x float> %vcvtzs.i)
+  %vcvtzs1.i = call <1 x i32> @llvm.aarch64.neon.fcvtzs.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtzs1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.aarch64.neon.fcvtzs.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtzs.v1i32.f32(float)
 
 define i64 @test_vcvtd_s64_f64(double %a) {
 ; CHECK: test_vcvtd_s64_f64
 ; CHECK: fcvtzs {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvzs.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvzs1.i = tail call <1 x i64> @llvm.aarch64.neon.fcvtzs.v1i64.v1f64(<1 x double> %vcvzs.i)
+  %vcvzs1.i = call <1 x i64> @llvm.aarch64.neon.fcvtzs.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvzs1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.aarch64.neon.fcvtzs.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtzs.v1i64.f64(double)
 
 define i32 @test_vcvtus(float %a) {
 ; CHECK: test_vcvtus
 ; CHECK: fcvtzu {{s[0-9]}}, {{s[0-9]}}
 entry:
-  %vcvtzu.i = insertelement <1 x float> undef, float %a, i32 0
-  %vcvtzu1.i = tail call <1 x i32> @llvm.aarch64.neon.fcvtzu.v1i32.v1f32(<1 x float> %vcvtzu.i)
+  %vcvtzu1.i = call <1 x i32> @llvm.aarch64.neon.fcvtzu.v1i32.f32(float %a)
   %0 = extractelement <1 x i32> %vcvtzu1.i, i32 0
   ret i32 %0
 }
 
-declare <1 x i32> @llvm.aarch64.neon.fcvtzu.v1i32.v1f32(<1 x float>)
+declare <1 x i32> @llvm.aarch64.neon.fcvtzu.v1i32.f32(float)
 
 define i64 @test_vcvtud(double %a) {
 ; CHECK: test_vcvtud
 ; CHECK: fcvtzu {{d[0-9]}}, {{d[0-9]}}
 entry:
-  %vcvtzu.i = insertelement <1 x double> undef, double %a, i32 0
-  %vcvtzu1.i = tail call <1 x i64> @llvm.aarch64.neon.fcvtzu.v1i64.v1f64(<1 x double> %vcvtzu.i)
+  %vcvtzu1.i = call <1 x i64> @llvm.aarch64.neon.fcvtzu.v1i64.f64(double %a)
   %0 = extractelement <1 x i64> %vcvtzu1.i, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.aarch64.neon.fcvtzu.v1i64.v1f64(<1 x double>)
+declare <1 x i64> @llvm.aarch64.neon.fcvtzu.v1i64.f64(double)