[AArch64][SVE] Asm: Vector Unpack Low/High instructions.

This patch adds support for the following unpack instructions:
  
- PUNPKLO, PUNPKHI   Unpack elements from low/high half and
                     place into elements of twice their size.

  e.g. punpklo p0.h, p0.b

- UUNPKLO, UUNPKHI   Unpack elements from low/high half and 
  SUNPKLO, SUNPKHI   place into elements of twice their size
                     after zero- or sign-extending the values.

  e.g. uunpklo z0.h, z0.b

llvm-svn: 336982

lib/Target/AArch64/AArch64SVEInstrInfo.td

@@ -116,6 +116,14 @@ let Predicates = [HasSVE] in {
   defm INSR_ZR : sve_int_perm_insrs<"insr">;
   defm INSR_ZV : sve_int_perm_insrv<"insr">;
 
+  defm SUNPKLO_ZZ : sve_int_perm_unpk<0b00, "sunpklo">;
+  defm SUNPKHI_ZZ : sve_int_perm_unpk<0b01, "sunpkhi">;
+  defm UUNPKLO_ZZ : sve_int_perm_unpk<0b10, "uunpklo">;
+  defm UUNPKHI_ZZ : sve_int_perm_unpk<0b11, "uunpkhi">;
+
+  def  PUNPKLO_PP : sve_int_perm_punpk<0b0, "punpklo">;
+  def  PUNPKHI_PP : sve_int_perm_punpk<0b1, "punpkhi">;
+
   def AND_PPzPP   : sve_int_pred_log<0b0000, "and">;
   def BIC_PPzPP   : sve_int_pred_log<0b0001, "bic">;
   def EOR_PPzPP   : sve_int_pred_log<0b0010, "eor">;

lib/Target/AArch64/SVEInstrFormats.td

@@ -658,6 +658,28 @@ multiclass sve_int_perm_tbl<string asm> {
                  (!cast<Instruction>(NAME # _D) ZPR64:$Zd, ZPR64:$Zn, ZPR64:$Zm), 0>;
 }
 
+class sve_int_perm_unpk<bits<2> sz16_64, bits<2> opc, string asm,
+                        ZPRRegOp zprty1, ZPRRegOp zprty2>
+: I<(outs zprty1:$Zd), (ins zprty2:$Zn),
+  asm, "\t$Zd, $Zn",
+  "", []>, Sched<[]> {
+  bits<5> Zd;
+  bits<5> Zn;
+  let Inst{31-24} = 0b00000101;
+  let Inst{23-22} = sz16_64;
+  let Inst{21-18} = 0b1100;
+  let Inst{17-16} = opc;
+  let Inst{15-10} = 0b001110;
+  let Inst{9-5}   = Zn;
+  let Inst{4-0}   = Zd;
+}
+
+multiclass sve_int_perm_unpk<bits<2> opc, string asm> {
+  def _H : sve_int_perm_unpk<0b01, opc, asm, ZPR16, ZPR8>;
+  def _S : sve_int_perm_unpk<0b10, opc, asm, ZPR32, ZPR16>;
+  def _D : sve_int_perm_unpk<0b11, opc, asm, ZPR64, ZPR32>;
+}
+
 class sve_int_perm_insrs<bits<2> sz8_64, string asm, ZPRRegOp zprty,
                          RegisterClass srcRegType>
 : I<(outs zprty:$Zdn), (ins zprty:$_Zdn, srcRegType:$Rm),
@@ -2358,6 +2380,21 @@ multiclass sve_int_perm_bin_perm_pp<bits<3> opc, string asm> {
   def _D : sve_int_perm_bin_perm_pp<opc, 0b11, asm, PPR64>;
 }
 
+class sve_int_perm_punpk<bit opc, string asm>
+: I<(outs PPR16:$Pd), (ins PPR8:$Pn),
+  asm, "\t$Pd, $Pn",
+  "",
+  []>, Sched<[]> {
+  bits<4> Pd;
+  bits<4> Pn;
+  let Inst{31-17} = 0b000001010011000;
+  let Inst{16}    = opc;
+  let Inst{15-9}  = 0b0100000;
+  let Inst{8-5}   = Pn;
+  let Inst{4}     = 0b0;
+  let Inst{3-0}   = Pd;
+}
+
 class sve_int_rdffr_pred<bit s, string asm>
 : I<(outs PPR8:$Pd), (ins PPRAny:$Pg),
   asm, "\t$Pd, $Pg/z",

test/MC/AArch64/SVE/punpkhi-diagnostics.s

@@ -0,0 +1,15 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element widths.
+
+punpkhi p0.b, p0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid predicate register
+// CHECK-NEXT: punpkhi p0.b, p0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+punpkhi p0.s, p0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid predicate register
+// CHECK-NEXT: punpkhi p0.s, p0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

test/MC/AArch64/SVE/punpkhi.s

@@ -0,0 +1,20 @@
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve < %s \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+// RUN: not llvm-mc -triple=aarch64 -show-encoding < %s 2>&1 \
+// RUN:        | FileCheck %s --check-prefix=CHECK-ERROR
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+punpkhi p0.h, p0.b
+// CHECK-INST: punpkhi	p0.h, p0.b
+// CHECK-ENCODING: [0x00,0x40,0x31,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: 00 40 31 05 <unknown>
+
+punpkhi p15.h, p15.b
+// CHECK-INST: punpkhi	p15.h, p15.b
+// CHECK-ENCODING: [0xef,0x41,0x31,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ef 41 31 05 <unknown>

test/MC/AArch64/SVE/punpklo-diagnostics.s

@@ -0,0 +1,15 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element widths.
+
+punpklo p0.b, p0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid predicate register
+// CHECK-NEXT: punpklo p0.b, p0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+punpklo p0.s, p0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid predicate register
+// CHECK-NEXT: punpklo p0.s, p0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

test/MC/AArch64/SVE/punpklo.s

@@ -0,0 +1,20 @@
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve < %s \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+// RUN: not llvm-mc -triple=aarch64 -show-encoding < %s 2>&1 \
+// RUN:        | FileCheck %s --check-prefix=CHECK-ERROR
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+punpklo p0.h, p0.b
+// CHECK-INST: punpklo	p0.h, p0.b
+// CHECK-ENCODING: [0x00,0x40,0x30,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: 00 40 30 05 <unknown>
+
+punpklo p15.h, p15.b
+// CHECK-INST: punpklo	p15.h, p15.b
+// CHECK-ENCODING: [0xef,0x41,0x30,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ef 41 30 05 <unknown>

test/MC/AArch64/SVE/sunpkhi-diagnostics.s

@@ -0,0 +1,20 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element widths.
+
+sunpkhi z0.b, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sunpkhi z0.b, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sunpkhi z0.s, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sunpkhi z0.s, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sunpkhi z0.d, z0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sunpkhi z0.d, z0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

test/MC/AArch64/SVE/sunpkhi.s

@@ -0,0 +1,26 @@
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve < %s \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+// RUN: not llvm-mc -triple=aarch64 -show-encoding < %s 2>&1 \
+// RUN:        | FileCheck %s --check-prefix=CHECK-ERROR
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+sunpkhi z31.h, z31.b
+// CHECK-INST: sunpkhi	z31.h, z31.b
+// CHECK-ENCODING: [0xff,0x3b,0x71,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b 71 05 <unknown>
+
+sunpkhi z31.s, z31.h
+// CHECK-INST: sunpkhi	z31.s, z31.h
+// CHECK-ENCODING: [0xff,0x3b,0xb1,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b b1 05 <unknown>
+
+sunpkhi z31.d, z31.s
+// CHECK-INST: sunpkhi	z31.d, z31.s
+// CHECK-ENCODING: [0xff,0x3b,0xf1,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b f1 05 <unknown>

test/MC/AArch64/SVE/sunpklo-diagnostics.s

@@ -0,0 +1,20 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element widths.
+
+sunpklo z0.b, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sunpklo z0.b, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sunpklo z0.s, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sunpklo z0.s, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sunpklo z0.d, z0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sunpklo z0.d, z0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

test/MC/AArch64/SVE/sunpklo.s

@@ -0,0 +1,26 @@
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve < %s \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+// RUN: not llvm-mc -triple=aarch64 -show-encoding < %s 2>&1 \
+// RUN:        | FileCheck %s --check-prefix=CHECK-ERROR
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+sunpklo z31.h, z31.b
+// CHECK-INST: sunpklo	z31.h, z31.b
+// CHECK-ENCODING: [0xff,0x3b,0x70,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b 70 05 <unknown>
+
+sunpklo z31.s, z31.h
+// CHECK-INST: sunpklo	z31.s, z31.h
+// CHECK-ENCODING: [0xff,0x3b,0xb0,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b b0 05 <unknown>
+
+sunpklo z31.d, z31.s
+// CHECK-INST: sunpklo	z31.d, z31.s
+// CHECK-ENCODING: [0xff,0x3b,0xf0,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b f0 05 <unknown>

test/MC/AArch64/SVE/uunpkhi-diagnostics.s

@@ -0,0 +1,20 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element widths.
+
+uunpkhi z0.b, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uunpkhi z0.b, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+uunpkhi z0.s, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uunpkhi z0.s, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+uunpkhi z0.d, z0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uunpkhi z0.d, z0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

test/MC/AArch64/SVE/uunpkhi.s

@@ -0,0 +1,26 @@
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve < %s \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+// RUN: not llvm-mc -triple=aarch64 -show-encoding < %s 2>&1 \
+// RUN:        | FileCheck %s --check-prefix=CHECK-ERROR
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+uunpkhi z31.h, z31.b
+// CHECK-INST: uunpkhi	z31.h, z31.b
+// CHECK-ENCODING: [0xff,0x3b,0x73,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b 73 05 <unknown>
+
+uunpkhi z31.s, z31.h
+// CHECK-INST: uunpkhi	z31.s, z31.h
+// CHECK-ENCODING: [0xff,0x3b,0xb3,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b b3 05 <unknown>
+
+uunpkhi z31.d, z31.s
+// CHECK-INST: uunpkhi	z31.d, z31.s
+// CHECK-ENCODING: [0xff,0x3b,0xf3,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b f3 05 <unknown>

test/MC/AArch64/SVE/uunpklo-diagnostics.s

@@ -0,0 +1,20 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element widths.
+
+uunpklo z0.b, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uunpklo z0.b, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+uunpklo z0.s, z0.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uunpklo z0.s, z0.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+uunpklo z0.d, z0.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uunpklo z0.d, z0.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

test/MC/AArch64/SVE/uunpklo.s

@@ -0,0 +1,26 @@
+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve < %s \
+// RUN:        | FileCheck %s --check-prefixes=CHECK-ENCODING,CHECK-INST
+// RUN: not llvm-mc -triple=aarch64 -show-encoding < %s 2>&1 \
+// RUN:        | FileCheck %s --check-prefix=CHECK-ERROR
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+uunpklo z31.h, z31.b
+// CHECK-INST: uunpklo	z31.h, z31.b
+// CHECK-ENCODING: [0xff,0x3b,0x72,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b 72 05 <unknown>
+
+uunpklo z31.s, z31.h
+// CHECK-INST: uunpklo	z31.s, z31.h
+// CHECK-ENCODING: [0xff,0x3b,0xb2,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b b2 05 <unknown>
+
+uunpklo z31.d, z31.s
+// CHECK-INST: uunpklo	z31.d, z31.s
+// CHECK-ENCODING: [0xff,0x3b,0xf2,0x05]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: ff 3b f2 05 <unknown>