Add v16i16 and v32i8 store patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@137166 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 56b05e17ac2ba5e5116782b1272eb58c8280fd5c..d8239bc993043d2cc4859677adc78365349eaabf 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -4063,10 +4063,18 @@ let Predicates = [HasAVX] in {
             (VMOVAPSYmr addr:$dst, VR256:$src)>;
   def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
             (VMOVAPSYmr addr:$dst, VR256:$src)>;
+  def : Pat<(alignedstore (v16i16 VR256:$src), addr:$dst),
+            (VMOVAPSYmr addr:$dst, VR256:$src)>;
+  def : Pat<(alignedstore (v32i8 VR256:$src), addr:$dst),
+            (VMOVAPSYmr addr:$dst, VR256:$src)>;
   def : Pat<(store (v4i64 VR256:$src), addr:$dst),
             (VMOVUPSYmr addr:$dst, VR256:$src)>;
   def : Pat<(store (v8i32 VR256:$src), addr:$dst),
             (VMOVUPSYmr addr:$dst, VR256:$src)>;
+  def : Pat<(store (v16i16 VR256:$src), addr:$dst),
+            (VMOVUPSYmr addr:$dst, VR256:$src)>;
+  def : Pat<(store (v32i8 VR256:$src), addr:$dst),
+            (VMOVUPSYmr addr:$dst, VR256:$src)>;
 }
 
 //===----------------------------------------------------------------------===//

diff --git a/test/CodeGen/X86/avx-load-store.ll b/test/CodeGen/X86/avx-load-store.ll
index 0d591ec27f71d65cc425726bd765d00ce27cfddf..d191af88638bbfacf54ad88d582f610fa7794518 100644 (file)
--- a/test/CodeGen/X86/avx-load-store.ll
+++ b/test/CodeGen/X86/avx-load-store.ll
@@ -40,3 +40,27 @@ define <4 x double> @mov01(<4 x double> %v, double * %ptr) nounwind {
   ret <4 x double> %i0
 }
 
+; CHECK: vmovaps  %ymm
+define void @storev16i16(<16 x i16> %a) nounwind {
+  store <16 x i16> %a, <16 x i16>* undef, align 32
+  unreachable
+}
+
+; CHECK: vmovups  %ymm
+define void @storev16i16_01(<16 x i16> %a) nounwind {
+  store <16 x i16> %a, <16 x i16>* undef, align 4
+  unreachable
+}
+
+; CHECK: vmovaps  %ymm
+define void @storev32i8(<32 x i8> %a) nounwind {
+  store <32 x i8> %a, <32 x i8>* undef, align 32
+  unreachable
+}
+
+; CHECK: vmovups  %ymm
+define void @storev32i8_01(<32 x i8> %a) nounwind {
+  store <32 x i8> %a, <32 x i8>* undef, align 4
+  unreachable
+}
+