From: Adam Nemet <anemet@apple.com>
Date: Mon, 27 Oct 2014 23:08:40 +0000 (+0000)
Subject: [AVX512] Add vpermil variable version
X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=commitdiff_plain;h=6bc8d95153d1f311a02cf48ca0e5735c80a531a3

[AVX512] Add vpermil variable version

This is implemented via a multiclass that derives from the vperm imm
multiclass.

Fixes <rdar://problem/18426089>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220737 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 07fe310d039..1d1abcfb40f 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -783,14 +783,37 @@ multiclass avx512_perm_imm<bits<8> opc, string OpcodeStr, SDNode OpNode,
 }
 }
 
+multiclass avx512_permil<bits<8> OpcImm, bits<8> OpcVar, X86VectorVTInfo _,
+                         X86VectorVTInfo Ctrl> :
+     avx512_perm_imm<OpcImm, "vpermil" # _.Suffix, X86VPermilpi, _> {
+  let ExeDomain = _.ExeDomain in {
+    def rr : AVX5128I<OpcVar, MRMSrcReg, (outs _.RC:$dst),
+                     (ins _.RC:$src1, _.RC:$src2),
+                     !strconcat("vpermil" # _.Suffix,
+                         " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                     [(set _.RC:$dst,
+                         (_.VT (X86VPermilpv _.RC:$src1,
+                                  (Ctrl.VT Ctrl.RC:$src2))))]>,
+             EVEX_4V;
+    def rm : AVX5128I<OpcVar, MRMSrcMem, (outs _.RC:$dst),
+                     (ins _.RC:$src1, Ctrl.MemOp:$src2),
+                     !strconcat("vpermil" # _.Suffix,
+                         " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                     [(set _.RC:$dst,
+                         (_.VT (X86VPermilpv _.RC:$src1,
+                                  (Ctrl.VT (Ctrl.MemOpFrag addr:$src2)))))]>,
+             EVEX_4V;
+  }
+}
+
 defm VPERMQZ :    avx512_perm_imm<0x00, "vpermq", X86VPermi, v8i64_info>,
                   EVEX_V512, VEX_W;
 defm VPERMPDZ :   avx512_perm_imm<0x01, "vpermpd", X86VPermi, v8f64_info>,
                   EVEX_V512, VEX_W;
 
-defm VPERMILPSZ : avx512_perm_imm<0x04, "vpermilps", X86VPermilpi, v16f32_info>,
+defm VPERMILPSZ : avx512_permil<0x04, 0x0C, v16f32_info, v16i32_info>,
                   EVEX_V512;
-defm VPERMILPDZ : avx512_perm_imm<0x05, "vpermilpd", X86VPermilpi, v8f64_info>,
+defm VPERMILPDZ : avx512_permil<0x05, 0x0D, v8f64_info, v8i64_info>,
                   EVEX_V512, VEX_W;
 
 def : Pat<(v16i32 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
diff --git a/test/MC/X86/avx512-encodings.s b/test/MC/X86/avx512-encodings.s
index 6b2acf6d40d..b60b72fd6bb 100644
--- a/test/MC/X86/avx512-encodings.s
+++ b/test/MC/X86/avx512-encodings.s
@@ -4375,3 +4375,19 @@ vfnmadd231ps %zmm5, %zmm6, %zmm7
 // CHECK: vfnmsub231pd
 // CHECK: encoding: [0x62,0xf2,0xcd,0x48,0xbe,0xfd]
 vfnmsub231pd %zmm5, %zmm6, %zmm7
+
+// CHECK: vpermilps
+// CHECK: encoding: [0x62,0xf2,0x6d,0x48,0x0c,0xd9]
+vpermilps %zmm1, %zmm2, %zmm3
+
+// CHECK: vpermilpd
+// CHECK: encoding: [0x62,0xf2,0xed,0x48,0x0d,0x5b,0x10]
+vpermilpd 0x400(%rbx), %zmm2, %zmm3
+
+// CHECK: vpermilps
+// CHECK: encoding: [0x62,0xf2,0x6d,0x48,0x0c,0x5b,0x10]
+vpermilps 0x400(%rbx), %zmm2, %zmm3
+
+// CHECK: vpermilpd
+// CHECK: encoding: [0x62,0xf3,0xfd,0x48,0x05,0x53,0x10,0x23]
+vpermilpd $0x23, 0x400(%rbx), %zmm2