Add a few patterns to match allzeros without having to use the fp unit.

Take advantage that the 128-bit vpxor zeros the higher part and use it.
This also fixes PR10491

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@136321 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index e15d7bd76b89d384b61efa8ac30d56a0f2907dee..cb4dbcc96919bfe5cb65de209a7f1ac4ad29676b 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -2098,6 +2098,16 @@ def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
           (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
 
+// AVX has no support for 256-bit integer instructions, but since the 128-bit
+// VPXOR instruction writes zero to its upper part, it's safe build zeros.
+def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
+def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
+          (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
+
+def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
+def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
+          (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
+
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Load/Store XCSR register
 //===----------------------------------------------------------------------===//

diff --git a/test/CodeGen/X86/avx-cast.ll b/test/CodeGen/X86/avx-cast.ll
index e24cfeb50852dbdf0b2d6ebeb192fcbaa1310c5c..7e368b3d9ee827467bc7401bad6240ea3ff56a24 100644 (file)
--- a/test/CodeGen/X86/avx-cast.ll
+++ b/test/CodeGen/X86/avx-cast.ll
@@ -16,7 +16,7 @@ entry:
   ret <4 x double> %shuffle.i
 }
 
-; CHECK: vxorps
+; CHECK: vpxor
 ; CHECK-NEXT: vinsertf128 $0
 define <4 x i64> @castC(<2 x i64> %m) nounwind uwtable readnone ssp {
 entry: