[AArch64][SVE2] Asm: add integer multiply-add/subtract (indexed) inst…

…ructions

Summary:
This patch adds support for the following instructions:

  MLA mul-add, writing addend (Zda = Zda +  Zn * Zm[idx])
  MLS mul-sub, writing addend (Zda = Zda + -Zn * Zm[idx])

Predicated forms of these instructions were added in SVE.

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

Reviewed By: rovka

Differential Revision: https://reviews.llvm.org/D61514

llvm-svn: 360682

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

@@ -1021,3 +1021,9 @@ let Predicates = [HasSVE] in {
   def : InstAlias<"fcmlt $Zd, $Pg/z, $Zm, $Zn",
                   (FCMGT_PPzZZ_D PPR64:$Zd, PPR3bAny:$Pg, ZPR64:$Zn, ZPR64:$Zm), 0>;
 }
+
+let Predicates = [HasSVE2] in {
+  // SVE2 integer multiply-add (indexed)
+  defm MLA_ZZZI : sve2_int_mla_by_indexed_elem<0b01, 0b0, "mla">;
+  defm MLS_ZZZI : sve2_int_mla_by_indexed_elem<0b01, 0b1, "mls">;
+}

llvm/lib/Target/AArch64/SVEInstrFormats.td

@@ -1690,6 +1690,51 @@ multiclass sve_int_mlas_vvv_pred<bits<1> opc, string asm> {
   def _D : sve_int_mlas_vvv_pred<0b11, opc, asm, ZPR64>;
 }
 
+//===----------------------------------------------------------------------===//
+// SVE2 Integer Multiply-Add - Indexed Group
+//===----------------------------------------------------------------------===//
+
+class sve2_int_mla_by_indexed_elem<bits<2> sz, bits<6> opc, string asm,
+                                   ZPRRegOp zprty1, ZPRRegOp zprty2,
+                                   ZPRRegOp zprty3, Operand itype>
+: I<(outs zprty1:$Zda), (ins zprty1:$_Zda, zprty2:$Zn, zprty3:$Zm, itype:$iop),
+  asm, "\t$Zda, $Zn, $Zm$iop", "", []>, Sched<[]> {
+  bits<5> Zda;
+  bits<5> Zn;
+  let Inst{31-24} = 0b01000100;
+  let Inst{23-22} = sz;
+  let Inst{21}    = 0b1;
+  let Inst{15-10} = opc;
+  let Inst{9-5}   = Zn;
+  let Inst{4-0}   = Zda;
+
+  let Constraints = "$Zda = $_Zda";
+  let DestructiveInstType = Destructive;
+  let ElementSize = ElementSizeNone;
+}
+
+multiclass sve2_int_mla_by_indexed_elem<bits<2> opc, bit S, string asm> {
+  def _H : sve2_int_mla_by_indexed_elem<{0, ?}, { 0b000, opc, S }, asm, ZPR16, ZPR16, ZPR3b16, VectorIndexH> {
+    bits<3> Zm;
+    bits<3> iop;
+    let Inst{22} = iop{2};
+    let Inst{20-19} = iop{1-0};
+    let Inst{18-16} = Zm;
+  }
+  def _S : sve2_int_mla_by_indexed_elem<0b10, { 0b000, opc, S }, asm, ZPR32, ZPR32, ZPR3b32, VectorIndexS> {
+    bits<3> Zm;
+    bits<2> iop;
+    let Inst{20-19} = iop;
+    let Inst{18-16} = Zm;
+  }
+  def _D : sve2_int_mla_by_indexed_elem<0b11, { 0b000, opc, S }, asm, ZPR64, ZPR64, ZPR4b64, VectorIndexD> {
+    bits<4> Zm;
+    bit iop;
+    let Inst{20} = iop;
+    let Inst{19-16} = Zm;
+  }
+}
+
 //===----------------------------------------------------------------------===//
 // SVE Integer Dot Product Group
 //===----------------------------------------------------------------------===//

llvm/test/MC/AArch64/SVE2/mla-diagnostics.s

@@ -0,0 +1,64 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// z register out of range for index
+
+mla z0.h, z1.h, z8.h[0]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: Invalid restricted vector register, expected z0.h..z7.h
+// CHECK-NEXT: mla z0.h, z1.h, z8.h[0]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mla z0.s, z1.s, z8.s[0]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: Invalid restricted vector register, expected z0.s..z7.s
+// CHECK-NEXT: mla z0.s, z1.s, z8.s[0]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mla z0.d, z1.d, z16.d[0]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: Invalid restricted vector register, expected z0.d..z15.d
+// CHECK-NEXT: mla z0.d, z1.d, z16.d[0]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element index
+
+mla z0.h, z1.h, z2.h[-1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 7].
+// CHECK-NEXT: mla z0.h, z1.h, z2.h[-1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mla z0.h, z1.h, z2.h[8]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 7].
+// CHECK-NEXT: mla z0.h, z1.h, z2.h[8]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mla z0.s, z1.s, z2.s[-1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 3].
+// CHECK-NEXT: mla z0.s, z1.s, z2.s[-1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mla z0.s, z1.s, z2.s[4]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 3].
+// CHECK-NEXT: mla z0.s, z1.s, z2.s[4]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mla z0.d, z1.d, z2.d[-1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 1].
+// CHECK-NEXT: mla z0.d, z1.d, z2.d[-1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mla z0.d, z1.d, z2.d[2]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 1].
+// CHECK-NEXT: mla z0.d, z1.d, z2.d[2]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// --------------------------------------------------------------------------//
+// Negative tests for instructions that are incompatible with movprfx
+
+movprfx z0.d, p0/z, z7.d
+mla z0.d, z1.d, z7.d[1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a predicated movprfx, suggest using unpredicated movprfx
+// CHECK-NEXT: mla z0.d, z1.d, z7.d[1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

llvm/test/MC/AArch64/SVE2/mla.s

@@ -0,0 +1,42 @@
+// 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
+
+mla z0.h, z1.h, z7.h[7]
+// CHECK-INST: mla	z0.h, z1.h, z7.h[7]
+// CHECK-ENCODING: [0x20,0x08,0x7f,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 08 7f 44 <unknown>
+
+mla z0.s, z1.s, z7.s[3]
+// CHECK-INST: mla	z0.s, z1.s, z7.s[3]
+// CHECK-ENCODING: [0x20,0x08,0xbf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 08 bf 44 <unknown>
+
+mla z0.d, z1.d, z7.d[1]
+// CHECK-INST: mla	z0.d, z1.d, z7.d[1]
+// CHECK-ENCODING: [0x20,0x08,0xf7,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 08 f7 44 <unknown>
+
+
+// --------------------------------------------------------------------------//
+// Test compatibility with MOVPRFX instruction.
+
+movprfx z0, z7
+// CHECK-INST: movprfx	z0, z7
+// CHECK-ENCODING: [0xe0,0xbc,0x20,0x04]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: e0 bc 20 04 <unknown>
+
+mla z0.d, z1.d, z7.d[1]
+// CHECK-INST: mla	z0.d, z1.d, z7.d[1]
+// CHECK-ENCODING: [0x20,0x08,0xf7,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 08 f7 44 <unknown>

llvm/test/MC/AArch64/SVE2/mls-diagnostics.s

@@ -0,0 +1,65 @@
+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2  2>&1 < %s| FileCheck %s
+
+
+
+// ------------------------------------------------------------------------- //
+// z register out of range for index
+
+mls z0.h, z1.h, z8.h[0]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: Invalid restricted vector register, expected z0.h..z7.h
+// CHECK-NEXT: mls z0.h, z1.h, z8.h[0]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mls z0.s, z1.s, z8.s[0]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: Invalid restricted vector register, expected z0.s..z7.s
+// CHECK-NEXT: mls z0.s, z1.s, z8.s[0]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mls z0.d, z1.d, z16.d[0]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: Invalid restricted vector register, expected z0.d..z15.d
+// CHECK-NEXT: mls z0.d, z1.d, z16.d[0]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element index
+
+mls z0.h, z1.h, z2.h[-1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 7].
+// CHECK-NEXT: mls z0.h, z1.h, z2.h[-1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mls z0.h, z1.h, z2.h[8]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 7].
+// CHECK-NEXT: mls z0.h, z1.h, z2.h[8]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mls z0.s, z1.s, z2.s[-1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 3].
+// CHECK-NEXT: mls z0.s, z1.s, z2.s[-1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mls z0.s, z1.s, z2.s[4]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 3].
+// CHECK-NEXT: mls z0.s, z1.s, z2.s[4]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mls z0.d, z1.d, z2.d[-1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 1].
+// CHECK-NEXT: mls z0.d, z1.d, z2.d[-1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+mls z0.d, z1.d, z2.d[2]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 1].
+// CHECK-NEXT: mls z0.d, z1.d, z2.d[2]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// --------------------------------------------------------------------------//
+// Negative tests for instructions that are incompatible with movprfx
+
+movprfx z0.d, p0/z, z7.d
+mls z0.d, z1.d, z7.d[1]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a predicated movprfx, suggest using unpredicated movprfx
+// CHECK-NEXT: mls z0.d, z1.d, z7.d[1]
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

llvm/test/MC/AArch64/SVE2/mls.s

@@ -0,0 +1,42 @@
+// 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
+
+mls z0.h, z1.h, z7.h[7]
+// CHECK-INST: mls	z0.h, z1.h, z7.h[7]
+// CHECK-ENCODING: [0x20,0x0c,0x7f,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 0c 7f 44 <unknown>
+
+mls z0.s, z1.s, z7.s[3]
+// CHECK-INST: mls	z0.s, z1.s, z7.s[3]
+// CHECK-ENCODING: [0x20,0x0c,0xbf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 0c bf 44 <unknown>
+
+mls z0.d, z1.d, z7.d[1]
+// CHECK-INST: mls	z0.d, z1.d, z7.d[1]
+// CHECK-ENCODING: [0x20,0x0c,0xf7,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 0c f7 44 <unknown>
+
+
+// --------------------------------------------------------------------------//
+// Test compatibility with MOVPRFX instruction.
+
+movprfx z0, z7
+// CHECK-INST: movprfx	z0, z7
+// CHECK-ENCODING: [0xe0,0xbc,0x20,0x04]
+// CHECK-ERROR: instruction requires: sve
+// CHECK-UNKNOWN: e0 bc 20 04 <unknown>
+
+mls z0.d, z1.d, z7.d[1]
+// CHECK-INST: mls	z0.d, z1.d, z7.d[1]
+// CHECK-ENCODING: [0x20,0x0c,0xf7,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 0c f7 44 <unknown>