AVX-512: Added X86vzmovl patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@191733 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index c9901bca0b119c28d0de387f9ac6eece64a6c67f..e73c73e3bfdde46589c42c2bfc5db3436a3fe507 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -1467,6 +1467,10 @@ let Predicates = [HasAVX512] in {
   let AddedComplexity = 20 in {
     def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
               (VMOVDI2PDIZrm addr:$src)>;
+    def : Pat<(v2i64 (X86vzmovl (v2i64 (scalar_to_vector GR64:$src)))),
+              (VMOV64toPQIZrr GR64:$src)>;
+    def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
+              (VMOVDI2PDIZrr GR32:$src)>;
               
     def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
               (VMOVDI2PDIZrm addr:$src)>;
@@ -1477,6 +1481,7 @@ let Predicates = [HasAVX512] in {
     def : Pat<(v2f64 (X86vzmovl (v2f64 VR128X:$src))),
             (VMOVZPQILo2PQIZrr VR128X:$src)>;
   }
+
   // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
   def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
                                (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),

diff --git a/test/CodeGen/X86/avx512-mov.ll b/test/CodeGen/X86/avx512-mov.ll
index 6ac487251edf84bd6eab829aed3652f2c3000cff..c44107be2c9dbe9dbe0325c8b2ed2b1f298e1f7a 100644 (file)
--- a/test/CodeGen/X86/avx512-mov.ll
+++ b/test/CodeGen/X86/avx512-mov.ll
@@ -100,3 +100,19 @@ define <2 x double> @test12(double* %x) {
    %res = insertelement <2 x double>zeroinitializer, double %y, i32 0
    ret <2 x double>%res
 }
+
+; CHECK-LABEL: @test13
+; CHECK: vmovqz  %rdi
+; CHECK: ret
+define <2 x i64> @test13(i64 %x) {
+   %res = insertelement <2 x i64>zeroinitializer, i64 %x, i32 0
+   ret <2 x i64>%res
+}
+
+; CHECK-LABEL: @test14
+; CHECK: vmovdz  %edi
+; CHECK: ret
+define <4 x i32> @test14(i32 %x) {
+   %res = insertelement <4 x i32>zeroinitializer, i32 %x, i32 0
+   ret <4 x i32>%res
+}