[AArch64][SVE2] Asm: add SADALP and UADALP instructions

[c-rhodes]

Summary

This patch adds support for the integer pairwise add and accumulate long
instructions SADALP/UADALP. These instructions are predicated.

The specification can be found here:
https://developer.arm.com/docs/ddi0602/latest

[SjoerdMeijer]

LGTM

llvm/trunk/lib/Target/AArch64/AArch64SVEInstrInfo.td

let Predicates = [HasSVE2] in {
   defm UHADD_ZPmZ  : sve2_int_arith_pred<0b100010, "uhadd">;
   defm SHSUB_ZPmZ  : sve2_int_arith_pred<0b100100, "shsub">;
   defm UHSUB_ZPmZ  : sve2_int_arith_pred<0b100110, "uhsub">;
   defm SRHADD_ZPmZ : sve2_int_arith_pred<0b101000, "srhadd">;
   defm URHADD_ZPmZ : sve2_int_arith_pred<0b101010, "urhadd">;
   defm SHSUBR_ZPmZ : sve2_int_arith_pred<0b101100, "shsubr">;
   defm UHSUBR_ZPmZ : sve2_int_arith_pred<0b101110, "uhsubr">;
 
+  // SVE2 integer pairwise add and accumulate long
+  defm SADALP_ZPmZ : sve2_int_sadd_long_accum_pairwise<0, "sadalp">;
+  defm UADALP_ZPmZ : sve2_int_sadd_long_accum_pairwise<1, "uadalp">;
+
   // SVE2 integer multiply long
   defm SQDMULLB_ZZZ : sve2_wide_int_arith_long<0b11000, "sqdmullb">;
   defm SQDMULLT_ZZZ : sve2_wide_int_arith_long<0b11001, "sqdmullt">;
   defm SMULLB_ZZZ   : sve2_wide_int_arith_long<0b11100, "smullb">;
   defm SMULLT_ZZZ   : sve2_wide_int_arith_long<0b11101, "smullt">;
   defm UMULLB_ZZZ   : sve2_wide_int_arith_long<0b11110, "umullb">;
   defm UMULLT_ZZZ   : sve2_wide_int_arith_long<0b11111, "umullt">;
 }

llvm/trunk/lib/Target/AArch64/SVEInstrFormats.td

 
 multiclass sve2_int_arith_pred<bits<6> opc, string asm> {
   def _B : sve2_int_arith_pred<0b00, opc, asm, ZPR8>;
   def _H : sve2_int_arith_pred<0b01, opc, asm, ZPR16>;
   def _S : sve2_int_arith_pred<0b10, opc, asm, ZPR32>;
   def _D : sve2_int_arith_pred<0b11, opc, asm, ZPR64>;
 }
 
+class sve2_int_sadd_long_accum_pairwise<bits<2> sz, bit U, string asm,
+                                        ZPRRegOp zprty1, ZPRRegOp zprty2>
+: I<(outs zprty1:$Zda), (ins PPR3bAny:$Pg, zprty1:$_Zda, zprty2:$Zn),
+  asm, "\t$Zda, $Pg/m, $Zn", "", []>, Sched<[]> {
+  bits<3> Pg;
+  bits<5> Zn;
+  bits<5> Zda;
+  let Inst{31-24} = 0b01000100;
+  let Inst{23-22} = sz;
+  let Inst{21-17} = 0b00010;
+  let Inst{16}    = U;
+  let Inst{15-13} = 0b101;
+  let Inst{12-10} = Pg;
+  let Inst{9-5}   = Zn;
+  let Inst{4-0}   = Zda;
+
+  let Constraints = "$Zda = $_Zda";
+  let DestructiveInstType = Destructive;
+  let ElementSize = zprty1.ElementSize;
+}
+
+multiclass sve2_int_sadd_long_accum_pairwise<bit U, string asm> {
+  def _H : sve2_int_sadd_long_accum_pairwise<0b01, U, asm, ZPR16, ZPR8>;
+  def _S : sve2_int_sadd_long_accum_pairwise<0b10, U, asm, ZPR32, ZPR16>;
+  def _D : sve2_int_sadd_long_accum_pairwise<0b11, U, asm, ZPR64, ZPR32>;
+}
+
 //===----------------------------------------------------------------------===//
 // SVE2 Widening Integer Arithmetic Group
 //===----------------------------------------------------------------------===//
 
 class sve2_wide_int_arith<bits<2> sz, bits<5> opc, string asm,
                           ZPRRegOp zprty1, ZPRRegOp zprty2, ZPRRegOp zprty3>
 : I<(outs zprty1:$Zd), (ins zprty2:$Zn, zprty3:$Zm),
   asm, "\t$Zd, $Zn, $Zm", "", []>, Sched<[]> {

llvm/trunk/test/MC/AArch64/SVE2/sadalp-diagnostics.s

+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2 2>&1 < %s| FileCheck %s
+
+
+// --------------------------------------------------------------------------//
+// Element sizes must match
+
+sadalp z0.b, p0/m, z1.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sadalp z0.b, p0/m, z1.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sadalp z0.h, p0/m, z1.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sadalp z0.h, p0/m, z1.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sadalp z0.s, p0/m, z1.s
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sadalp z0.s, p0/m, z1.s
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sadalp z0.d, p0/m, z1.d
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: sadalp z0.d, p0/m, z1.d
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+// --------------------------------------------------------------------------//
+// Predicate not in restricted predicate range
+
+sadalp z0.h, p8/m, z1.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: restricted predicate has range [0, 7].
+// CHECK-NEXT: sadalp z0.h, p8/m, z1.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+// --------------------------------------------------------------------------//
+// Negative movprfx tests
+
+movprfx z31.s, p0/z, z6.s // element type of the source operand, rather than destination.
+sadalp z31.d, p0/m, z30.s
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a predicated movprfx with a different element size
+// CHECK-NEXT: sadalp z31.d, p0/m, z30.s
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

llvm/trunk/test/MC/AArch64/SVE2/sadalp.s

+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2 < %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=+sve2 < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve2 - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve2 < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+sadalp z0.h, p0/m, z1.b
+// CHECK-INST: sadalp z0.h, p0/m, z1.b
+// CHECK-ENCODING: [0x20,0xa0,0x44,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 a0 44 44 <unknown>
+
+sadalp z29.s, p0/m, z30.h
+// CHECK-INST: sadalp z29.s, p0/m, z30.h
+// CHECK-ENCODING: [0xdd,0xa3,0x84,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: dd a3 84 44 <unknown>
+
+sadalp z30.d, p7/m, z31.s
+// CHECK-INST: sadalp z30.d, p7/m, z31.s
+// CHECK-ENCODING: [0xfe,0xbf,0xc4,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: fe bf c4 44 <unknown>
+
+// --------------------------------------------------------------------------//
+// Test compatibility with MOVPRFX instruction.
+
+movprfx z31.d, p0/z, z6.d
+// CHECK-INST: movprfx z31.d, p0/z, z6.d
+// CHECK-ENCODING: [0xdf,0x20,0xd0,0x04]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: df 20 d0 04 <unknown>
+
+sadalp z31.d, p0/m, z30.s
+// CHECK-INST: sadalp z31.d, p0/m, z30.s
+// CHECK-ENCODING: [0xdf,0xa3,0xc4,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: df a3 c4 44 <unknown>
+
+movprfx z31, z6
+// CHECK-INST: movprfx z31, z6
+// CHECK-ENCODING: [0xdf,0xbc,0x20,0x04]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: df bc 20 04 <unknown>
+
+sadalp z31.d, p0/m, z30.s
+// CHECK-INST: sadalp z31.d, p0/m, z30.s
+// CHECK-ENCODING: [0xdf,0xa3,0xc4,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: df a3 c4 44 <unknown>

llvm/trunk/test/MC/AArch64/SVE2/uadalp-diagnostics.s

+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2 2>&1 < %s| FileCheck %s
+
+
+// --------------------------------------------------------------------------//
+// Element sizes must match
+
+uadalp z0.b, p0/m, z1.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uadalp z0.b, p0/m, z1.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+uadalp z0.h, p0/m, z1.h
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uadalp z0.h, p0/m, z1.h
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+uadalp z0.s, p0/m, z1.s
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uadalp z0.s, p0/m, z1.s
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+uadalp z0.d, p0/m, z1.d
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: uadalp z0.d, p0/m, z1.d
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+// --------------------------------------------------------------------------//
+// Predicate not in restricted predicate range
+
+uadalp z0.h, p8/m, z1.b
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: restricted predicate has range [0, 7].
+// CHECK-NEXT: uadalp z0.h, p8/m, z1.b
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+// --------------------------------------------------------------------------//
+// Negative movprfx tests
+
+movprfx z31.s, p0/z, z6.s // element type of the source operand, rather than destination.
+uadalp z31.d, p0/m, z30.s
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a predicated movprfx with a different element size
+// CHECK-NEXT: uadalp z31.d, p0/m, z30.s
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

llvm/trunk/test/MC/AArch64/SVE2/uadalp.s

+// RUN: llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2 < %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=+sve2 < %s \
+// RUN:        | llvm-objdump -d -mattr=+sve2 - | FileCheck %s --check-prefix=CHECK-INST
+// RUN: llvm-mc -triple=aarch64 -filetype=obj -mattr=+sve2 < %s \
+// RUN:        | llvm-objdump -d - | FileCheck %s --check-prefix=CHECK-UNKNOWN
+
+uadalp z0.h, p0/m, z1.b
+// CHECK-INST: uadalp z0.h, p0/m, z1.b
+// CHECK-ENCODING: [0x20,0xa0,0x45,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 a0 45 44 <unknown>
+
+uadalp z29.s, p0/m, z30.h
+// CHECK-INST: uadalp z29.s, p0/m, z30.h
+// CHECK-ENCODING: [0xdd,0xa3,0x85,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: dd a3 85 44 <unknown>
+
+uadalp z30.d, p7/m, z31.s
+// CHECK-INST: uadalp z30.d, p7/m, z31.s
+// CHECK-ENCODING: [0xfe,0xbf,0xc5,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: fe bf c5 44 <unknown>
+
+// --------------------------------------------------------------------------//
+// Test compatibility with MOVPRFX instruction.
+
+movprfx z31.d, p0/z, z6.d
+// CHECK-INST: movprfx z31.d, p0/z, z6.d
+// CHECK-ENCODING: [0xdf,0x20,0xd0,0x04]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: df 20 d0 04 <unknown>
+
+uadalp z31.d, p0/m, z30.s
+// CHECK-INST: uadalp z31.d, p0/m, z30.s
+// CHECK-ENCODING: [0xdf,0xa3,0xc5,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: df a3 c5 44 <unknown>
+
+movprfx z31, z6
+// CHECK-INST: movprfx z31, z6
+// CHECK-ENCODING: [0xdf,0xbc,0x20,0x04]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: df bc 20 04 <unknown>
+
+uadalp z31.d, p0/m, z30.s
+// CHECK-INST: uadalp z31.d, p0/m, z30.s
+// CHECK-ENCODING: [0xdf,0xa3,0xc5,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: df a3 c5 44 <unknown>