AVX512: Add extract_subvector patterns v8i1->v4i1 , v4i1->v2i1.

Differential Revision: http://reviews.llvm.org/D17953

llvm-svn: 262929

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 3412ae3..306d0e3 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -2512,6 +2512,14 @@ defm : operation_subvector_mask_lowering<VK16, v16i1, VK64, v64i1>;
 
 defm : operation_subvector_mask_lowering<VK32, v32i1, VK64, v64i1>;
 
+def : Pat<(v2i1 (extract_subvector (v4i1 VK4:$src), (iPTR 2))),
+          (v2i1 (COPY_TO_REGCLASS 
+                  (KSHIFTRWri (COPY_TO_REGCLASS VK4:$src, VK16), (i8 2)),
+                   VK2))>;
+def : Pat<(v4i1 (extract_subvector (v8i1 VK8:$src), (iPTR 4))),
+          (v4i1 (COPY_TO_REGCLASS 
+                  (KSHIFTRWri (COPY_TO_REGCLASS VK8:$src, VK16), (i8 4)),
+                   VK4))>;
 def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))),
           (v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>;
 def : Pat<(v16i1 (extract_subvector (v32i1 VK32:$src), (iPTR 16))),

diff --git a/llvm/test/CodeGen/X86/avx512-skx-insert-subvec.ll b/llvm/test/CodeGen/X86/avx512-skx-insert-subvec.ll
index 8016bd4..c1d0a9a 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512-skx-insert-subvec.ll
+++ b/llvm/test/CodeGen/X86/avx512-skx-insert-subvec.ll
@@ -145,3 +145,26 @@ define <64 x i1> @test8(<8 x i1> %a, <8 x i1>%b) {
   ret <64 x i1> %res
 }
 
+define <4 x i1> @test9(<8 x i1> %a, <8 x i1> %b) {
+; CHECK-LABEL: test9:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vpsllw $15, %xmm0, %xmm0
+; CHECK-NEXT:    vpmovw2m %xmm0, %k0
+; CHECK-NEXT:    kshiftrw $4, %k0, %k0
+; CHECK-NEXT:    vpmovm2d %k0, %xmm0
+; CHECK-NEXT:    retq
+  %res = shufflevector <8 x i1> %a, <8 x i1> %b, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  ret <4 x i1> %res
+}
+
+define <2 x i1> @test10(<4 x i1> %a, <4 x i1> %b) {
+; CHECK-LABEL: test10:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vpslld $31, %xmm0, %xmm0
+; CHECK-NEXT:    vptestmd %xmm0, %xmm0, %k0
+; CHECK-NEXT:    kshiftrw $2, %k0, %k0
+; CHECK-NEXT:    vpmovm2q %k0, %xmm0
+; CHECK-NEXT:    retq
+  %res = shufflevector <4 x i1> %a, <4 x i1> %b, <2 x i32> <i32 2, i32 3>
+  ret <2 x i1> %res
+}