Add AVX512 patterns for v16i32 broadcast and v2i64 zero extend load.

Patch by Aleksey Bader.

llvm-svn: 196435

3 files changed, 27 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 012934140ff..2604c8e6f97 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -420,6 +420,8 @@ def : Pat <(v8i64 (X86vzext VK8WM:$mask)),
 
 def : Pat<(v16i32 (X86VBroadcast (i32 GR32:$src))),
         (VPBROADCASTDrZrr GR32:$src)>;
+def : Pat<(v16i32 (X86VBroadcastm VK16WM:$mask, (i32 GR32:$src))),
+        (VPBROADCASTDrZkrr VK16WM:$mask, GR32:$src)>;
 def : Pat<(v8i64 (X86VBroadcast (i64 GR64:$src))),
         (VPBROADCASTQrZrr GR64:$src)>;
 def : Pat<(v8i64 (X86VBroadcastm VK8WM:$mask, (i64 GR64:$src))),
@@ -1536,6 +1538,8 @@ let Predicates = [HasAVX512] in {
             (VMOVZPQILo2PQIZrm addr:$src)>;
     def : Pat<(v2f64 (X86vzmovl (v2f64 VR128X:$src))),
             (VMOVZPQILo2PQIZrr VR128X:$src)>;
+    def : Pat<(v2i64 (X86vzload addr:$src)),
+            (VMOVZPQILo2PQIZrm addr:$src)>;
   }
 
   // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
diff --git a/llvm/test/CodeGen/X86/avx512-vbroadcast-crash.ll b/llvm/test/CodeGen/X86/avx512-vbroadcast-crash.ll
new file mode 100644
index 00000000000..0d6d0d7ae5d
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx512-vbroadcast-crash.ll
@@ -0,0 +1,10 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl 
+
+define <16 x i32> @test_vbroadcast() {
+entry:
+  %0 = sext <16 x i1> zeroinitializer to <16 x i32>
+  %1 = fcmp uno <16 x float> undef, zeroinitializer
+  %2 = sext <16 x i1> %1 to <16 x i32>
+  %3 = select <16 x i1> %1, <16 x i32> %0, <16 x i32> %2
+  ret <16 x i32> %3
+}
diff --git a/llvm/test/CodeGen/X86/avx512-zext-load-crash.ll b/llvm/test/CodeGen/X86/avx512-zext-load-crash.ll
new file mode 100644
index 00000000000..5861fbfa0b5
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx512-zext-load-crash.ll
@@ -0,0 +1,13 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl
+
+define <8 x i16> @test_zext_load() {
+entry:
+  %0 = load <2 x i16> ** undef, align 8
+  %1 = getelementptr inbounds <2 x i16>* %0, i64 1
+  %2 = load <2 x i16>* %0, align 1
+  %3 = shufflevector <2 x i16> %2, <2 x i16> undef, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %4 = load <2 x i16>* %1, align 1
+  %5 = shufflevector <2 x i16> %4, <2 x i16> undef, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %6 = shufflevector <8 x i16> %3, <8 x i16> %5, <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x i16> %6
+}