AVX-512: Fixed a bug in extracting subvector from v64i1

By Igor Breger (igor.breger@intel.com)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@238322 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 8b26603e250244e2a542a36c7696ce4c063dff2d..9d11d3c7050f8dd72a455f9068e71283f6d3fa83 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -1893,6 +1893,8 @@ let Predicates = [HasAVX512] in {
 
   def : Pat<(i32 (zext VK1:$src)),
             (AND32ri (KMOVWrk (COPY_TO_REGCLASS VK1:$src, VK16)), (i32 1))>;
+  def : Pat<(i32 (anyext VK1:$src)),
+            (KMOVWrk (COPY_TO_REGCLASS VK1:$src, VK16))>;
   def : Pat<(i8 (zext VK1:$src)),
             (EXTRACT_SUBREG
              (AND32ri (KMOVWrk
@@ -2223,6 +2225,12 @@ def : Pat<(v16i1 (insert_subvector undef, (v8i1 VK8:$src), (iPTR 0))),
 def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))),
           (v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>;
 
+def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 0))),
+          (v32i1 (COPY_TO_REGCLASS VK64:$src, VK32))>;
+
+def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 32))),
+          (v32i1 (COPY_TO_REGCLASS (KSHIFTRQri VK64:$src, (i8 32)), VK32))>;
+
 let Predicates = [HasVLX] in {
   def : Pat<(v8i1 (insert_subvector undef, (v4i1 VK4:$src), (iPTR 0))),
             (v8i1 (COPY_TO_REGCLASS VK4:$src, VK8))>;

diff --git a/test/CodeGen/X86/avx512-trunc-ext.ll b/test/CodeGen/X86/avx512-trunc-ext.ll
index f5e910c4a8e8e19196fcd0177a2e8cadbfde2dda..f25458972e42a957d7ea6bc351c887d32ccea34d 100644 (file)
--- a/test/CodeGen/X86/avx512-trunc-ext.ll
+++ b/test/CodeGen/X86/avx512-trunc-ext.ll
@@ -950,3 +950,12 @@ define void @extload_v8i64(<8 x i8>* %a, <8 x i64>* %res) {
   ret void
 }
 
+;SKX-LABEL: test21:
+;SKX:       vmovdqu16 %zmm0, %zmm3 {%k1}
+;SKX-NEXT:  kshiftrq  $32, %k1, %k1
+;SKX-NEXT:  vmovdqu16 %zmm1, %zmm2 {%k1}
+define <64 x i16> @test21(<64 x i16> %x , <64 x i1> %mask) nounwind readnone {
+  %ret = select <64 x i1> %mask, <64 x i16> %x, <64 x i16> zeroinitializer
+  ret <64 x i16> %ret
+}  
+