From e868005a274ca7d7bc12b75771fb85625c86749f Mon Sep 17 00:00:00 2001
From: Adam Nemet <anemet@apple.com>
Date: Wed, 8 Oct 2014 23:25:33 +0000
Subject: [PATCH] [AVX512] Add asm-only support for vextract*x4 masking
 variants

These derive from the new asm-only masking definitions.

Unfortunately I wasn't able to find a ISel pattern that we could legally
generate for the masking variants.  The problem is that since the destination
is v4* we would need VK4 register classes and v4i1 value types to express the
masking.  These are however not legal types/classes in AVX512f but only in VL,
so things get complicated pretty quickly.  We can revisit this question later
if we have a more pressing need to express something like this.

So the ISel patterns are empty for the masking instructions and the next patch
will add Pat<>s instead to match the intrinsics calls with instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219361 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86InstrAVX512.td | 26 +++++++++++++++++++-------
 test/MC/X86/avx512-encodings.s   |  8 ++++++++
 2 files changed, 27 insertions(+), 7 deletions(-)

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 4930db9563d..987ff6cab2c 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -209,6 +209,18 @@ multiclass AVX512_masking_3src<bits<8> O, Format F, X86VectorVTInfo _,
                          OpcodeStr, AttSrcAsm, IntelSrcAsm, RHS,
                          (vselect _.KRCWM:$mask, RHS, _.RC:$src1)>;
 
+
+multiclass AVX512_masking_in_asm<bits<8> O, Format F, X86VectorVTInfo _,
+                                 dag Outs, dag Ins,
+                                 string OpcodeStr,
+                                 string AttSrcAsm, string IntelSrcAsm,
+                                 list<dag> Pattern> :
+   AVX512_masking_custom<O, F, Outs, Ins,
+                         !con((ins _.RC:$src0, _.KRCWM:$mask), Ins),
+                         !con((ins _.KRCWM:$mask), Ins),
+                         OpcodeStr, AttSrcAsm, IntelSrcAsm, Pattern, [], [],
+                         "$src0 = $dst">;
+
 // Bitcasts between 512-bit vector types. Return the original type since
 // no instruction is needed for the conversion
 let Predicates = [HasAVX512] in {
@@ -405,13 +417,13 @@ multiclass vextract_for_size<int Opcode,
                              PatFrag vextract_extract,
                              SDNodeXForm EXTRACT_get_vextract_imm> {
   let hasSideEffects = 0, ExeDomain = To.ExeDomain in {
-    def rr : AVX512AIi8<Opcode, MRMDestReg, (outs To.RC:$dst),
-            (ins VR512:$src1, i8imm:$idx),
-            "vextract" # To.EltTypeName # "x4\t{$idx, $src1, $dst|"
-                                               "$dst, $src1, $idx}",
-            [(set To.RC:$dst, (vextract_extract:$idx (From.VT VR512:$src1),
-                                                     (iPTR imm)))]>,
-            EVEX, EVEX_V512;
+    defm rr : AVX512_masking_in_asm<Opcode, MRMDestReg, To, (outs To.RC:$dst),
+                (ins VR512:$src1, i8imm:$idx),
+                "vextract" # To.EltTypeName # "x4",
+                "$idx, $src1", "$src1, $idx",
+                [(set To.RC:$dst, (vextract_extract:$idx (From.VT VR512:$src1),
+                                                         (iPTR imm)))]>,
+              AVX512AIi8Base, EVEX, EVEX_V512;
     let mayStore = 1 in
     def rm : AVX512AIi8<Opcode, MRMDestMem, (outs),
             (ins To.MemOp:$dst, VR512:$src1, i8imm:$src2),
diff --git a/test/MC/X86/avx512-encodings.s b/test/MC/X86/avx512-encodings.s
index 7e968b77fec..8326a7c9289 100644
--- a/test/MC/X86/avx512-encodings.s
+++ b/test/MC/X86/avx512-encodings.s
@@ -4335,3 +4335,11 @@ valignq  $2, 0x100(%rsp), %zmm0, %zmm1
 // CHECK: valignq $3
 // CHECK: encoding: [0x62,0xf3,0xfd,0x49,0x03,0xcb,0x03]
 valignq  $3, %zmm3, %zmm0, %zmm1 {%k1}
+
+// CHECK: vextractf32x4 $3
+// CHECK: encoding: [0x62,0xf3,0x7d,0x49,0x19,0xd9,0x03]
+vextractf32x4  $3, %zmm3, %xmm1 {%k1}
+
+// CHECK: vextracti64x4 $1
+// CHECK: encoding: [0x62,0x53,0xfd,0xcb,0x3b,0xf4,0x01]
+vextracti64x4  $1, %zmm14, %ymm12 {%k3} {z}
-- 
2.40.0