From 5e4b45361f5da80f42a502e120bd3352a0f3d68a Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Sat, 27 Jan 2018 23:49:14 +0000
Subject: [PATCH] [X86] Add patterns for using masked vptestnmd for 256-bit
 vectors without VLX.

We can widen the mask and extract it back down.

llvm-svn: 323610
---
 llvm/lib/Target/X86/X86InstrAVX512.td     | 37 +++++++++++++++--------
 llvm/test/CodeGen/X86/avx512vl-vec-cmp.ll |  5 ++-
 2 files changed, 26 insertions(+), 16 deletions(-)

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index b4e68f20c119..e095011d8757 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -2958,19 +2958,19 @@ defm KSHIFTR : avx512_mask_shiftop_w<0x30, 0x31, "kshiftr", X86kshiftr, SSE_PSHU
 multiclass axv512_icmp_packed_no_vlx_lowering<PatFrag Frag, string InstStr,
                                               X86VectorVTInfo Narrow,
                                               X86VectorVTInfo Wide> {
-def : Pat<(Narrow.KVT (Frag (Narrow.VT Narrow.RC:$src1),
+  def : Pat<(Narrow.KVT (Frag (Narrow.VT Narrow.RC:$src1),
                               (Narrow.VT Narrow.RC:$src2))),
           (COPY_TO_REGCLASS
-           (!cast<Instruction>(InstStr##Zrr)
+           (!cast<Instruction>(InstStr#"Zrr")
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src2, Narrow.SubRegIdx))),
            Narrow.KRC)>;
 
-def : Pat<(Narrow.KVT (and Narrow.KRC:$mask,
-                           (Frag (Narrow.VT Narrow.RC:$src1),
+  def : Pat<(Narrow.KVT (and Narrow.KRC:$mask,
+                             (Frag (Narrow.VT Narrow.RC:$src1),
                                    (Narrow.VT Narrow.RC:$src2)))),
           (COPY_TO_REGCLASS
-           (!cast<Instruction>(InstStr##Zrrk)
+           (!cast<Instruction>(InstStr#"Zrrk")
             (COPY_TO_REGCLASS Narrow.KRC:$mask, Wide.KRC),
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src1, Narrow.SubRegIdx)),
             (Wide.VT (INSERT_SUBREG (IMPLICIT_DEF), Narrow.RC:$src2, Narrow.SubRegIdx))),
@@ -5232,14 +5232,25 @@ multiclass avx512_vptest_mb<bits<8> opc, string OpcodeStr, SDNode OpNode,
 // Use 512bit version to implement 128/256 bit in case NoVLX.
 multiclass avx512_vptest_lowering<SDNode OpNode, X86VectorVTInfo ExtendInfo,
                                   X86VectorVTInfo _, string Suffix> {
-    def : Pat<(_.KVT (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2))),
-              (_.KVT (COPY_TO_REGCLASS
-                       (!cast<Instruction>(NAME # Suffix # "Zrr")
-                         (INSERT_SUBREG (ExtendInfo.VT (IMPLICIT_DEF)),
-                                        _.RC:$src1, _.SubRegIdx),
-                         (INSERT_SUBREG (ExtendInfo.VT (IMPLICIT_DEF)),
-                                        _.RC:$src2, _.SubRegIdx)),
-                     _.KRC))>;
+  def : Pat<(_.KVT (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2))),
+            (_.KVT (COPY_TO_REGCLASS
+                     (!cast<Instruction>(NAME # Suffix # "Zrr")
+                       (INSERT_SUBREG (ExtendInfo.VT (IMPLICIT_DEF)),
+                                      _.RC:$src1, _.SubRegIdx),
+                       (INSERT_SUBREG (ExtendInfo.VT (IMPLICIT_DEF)),
+                                      _.RC:$src2, _.SubRegIdx)),
+                   _.KRC))>;
+
+  def : Pat<(_.KVT (and _.KRC:$mask,
+                        (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2)))),
+            (COPY_TO_REGCLASS
+             (!cast<Instruction>(NAME # Suffix # "Zrrk")
+              (COPY_TO_REGCLASS _.KRC:$mask, ExtendInfo.KRC),
+              (INSERT_SUBREG (ExtendInfo.VT (IMPLICIT_DEF)),
+                             _.RC:$src1, _.SubRegIdx),
+              (INSERT_SUBREG (ExtendInfo.VT (IMPLICIT_DEF)),
+                             _.RC:$src2, _.SubRegIdx)),
+             _.KRC)>;
 }
 
 multiclass avx512_vptest_dq_sizes<bits<8> opc, string OpcodeStr, SDNode OpNode,
diff --git a/llvm/test/CodeGen/X86/avx512vl-vec-cmp.ll b/llvm/test/CodeGen/X86/avx512vl-vec-cmp.ll
index 82c1054a9802..e75e6ea0ad53 100644
--- a/llvm/test/CodeGen/X86/avx512vl-vec-cmp.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-vec-cmp.ll
@@ -1134,9 +1134,8 @@ define <8 x i32> @testnm_and(<8 x i32> %a, <8 x i32> %b, <8 x i32> %x, <8 x i32>
 ; NoVLX-NEXT:    # kill: def %ymm2 killed %ymm2 def %zmm2
 ; NoVLX-NEXT:    # kill: def %ymm1 killed %ymm1 def %zmm1
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 def %zmm0
-; NoVLX-NEXT:    vptestnmd %zmm0, %zmm0, %k0
-; NoVLX-NEXT:    vptestnmd %zmm1, %zmm1, %k1
-; NoVLX-NEXT:    kandw %k1, %k0, %k1
+; NoVLX-NEXT:    vptestnmd %zmm0, %zmm0, %k1
+; NoVLX-NEXT:    vptestnmd %zmm1, %zmm1, %k1 {%k1}
 ; NoVLX-NEXT:    vpblendmd %zmm2, %zmm3, %zmm0 {%k1}
 ; NoVLX-NEXT:    # kill: def %ymm0 killed %ymm0 killed %zmm0
 ; NoVLX-NEXT:    retq