[AArch64][SVE2] Asm: implement CDOT instruction

[c-rhodes]

Summary

The complex DOT instructions perform a dot-product on quadtuplets from
two source vectors and the resuling wide real or wide imaginary is
accumulated into the destination register. The instructions come in two
forms:

Vector form, e.g.

cdot z0.s, z1.b, z2.b, #90    - complex dot product on four 8-bit quad-tuplets,
                                accumulating results in 32-bit elements. The
                                complex numbers in the second source vector are
                                rotated by 90 degrees.

cdot z0.d, z1.h, z2.h, #180   - complex dot product on four 16-bit quad-tuplets,
                                accumulating results in 64-bit elements.
                                The complex numbers in the second source
                                vector are rotated by 180 degrees.

Indexed form, e.g.

cdot z0.s, z1.b, z2.b[3], #0  - complex dot product on four 8-bit quad-tuplets,
                                with specified quadtuplet from second source vector,
                                accumulating results in 32-bit elements.
cdot z0.d, z1.h, z2.h[1], #0  - complex dot product on four 16-bit quad-tuplets,
                                with specified quadtuplet from second source vector,
                                accumulating results in 64-bit elements.

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

[rovka]

LGTM

[SjoerdMeijer]

LGTM

[SjoerdMeijer]

Ah, sorry, didn't noticed the LGTM until I clicked the submit button.

[rovka]

In D61903#1502532, @SjoerdMeijer wrote:

Ah, sorry, didn't noticed the LGTM until I clicked the submit button.

No problem, two LGTMs are better than one ;)

lib/Target/AArch64/AArch64SVEInstrInfo.td

let Predicates = [HasSVE2] in {
   defm SQDMULH_ZZZ  : sve2_int_mul<0b100, "sqdmulh">;
   defm SQRDMULH_ZZZ : sve2_int_mul<0b101, "sqrdmulh">;
 
   // SVE2 integer multiply vectors (unpredicated)
   defm MUL_ZZZ    : sve2_int_mul<0b000, "mul">;
   defm SMULH_ZZZ  : sve2_int_mul<0b010, "smulh">;
   defm UMULH_ZZZ  : sve2_int_mul<0b011, "umulh">;
   def  PMUL_ZZZ_B : sve2_int_mul<0b00, 0b001, "pmul", ZPR8>;
+
+  // SVE2 complex integer dot product (indexed)
+  defm CDOT_ZZZI : sve2_cintx_dot_by_indexed_elem<"cdot">;
+
+  // SVE2 complex integer dot product
+  defm CDOT_ZZZ : sve2_cintx_dot<"cdot">;
 }

lib/Target/AArch64/SVEInstrFormats.td

def _D : sve_intx_dot_by_indexed_elem<0b1, opc, asm, ZPR64, ZPR16, ZPR4b16, VectorIndexD> {
     bits<1> iop;
     bits<4> Zm;
     let Inst{20} = iop;
     let Inst{19-16} = Zm;
   }
 }
 
 //===----------------------------------------------------------------------===//
+// SVE2 Complex Integer Dot Product Group
+//===----------------------------------------------------------------------===//
+
+class sve2_complex_int_arith<bits<2> sz, bits<4> opc, string asm,
+                             ZPRRegOp zprty1, ZPRRegOp zprty2>
+: I<(outs zprty1:$Zda), (ins zprty1:$_Zda, zprty2:$Zn, zprty2:$Zm,
+                         complexrotateop:$rot),
+  asm, "\t$Zda, $Zn, $Zm, $rot", "", []>, Sched<[]> {
+  bits<5> Zda;
+  bits<5> Zn;
+  bits<5> Zm;
+  bits<2> rot;
+  let Inst{31-24} = 0b01000100;
+  let Inst{23-22} = sz;
+  let Inst{21}    = 0b0;
+  let Inst{20-16} = Zm;
+  let Inst{15-12} = opc;
+  let Inst{11-10} = rot;
+  let Inst{9-5}   = Zn;
+  let Inst{4-0}   = Zda;
+
+  let Constraints = "$Zda = $_Zda";
+  let DestructiveInstType = Destructive;
+  let ElementSize = ElementSizeNone;
+}
+
+multiclass sve2_cintx_dot<string asm> {
+  def _S : sve2_complex_int_arith<0b10, 0b0001, asm, ZPR32, ZPR8>;
+  def _D : sve2_complex_int_arith<0b11, 0b0001, asm, ZPR64, ZPR16>;
+}
+
+//===----------------------------------------------------------------------===//
+// SVE2 Complex Integer Dot Product - Indexed Group
+//===----------------------------------------------------------------------===//
+
+class sve2_complex_int_arith_indexed<bits<2> sz, bits<4> opc, string asm,
+                                     ZPRRegOp zprty1, ZPRRegOp zprty2,
+                                     ZPRRegOp zprty3, Operand itype>
+: I<(outs zprty1:$Zda), (ins zprty1:$_Zda, zprty2:$Zn, zprty3:$Zm, itype:$iop,
+                         complexrotateop:$rot),
+  asm, "\t$Zda, $Zn, $Zm$iop, $rot", "", []>, Sched<[]> {
+  bits<5> Zda;
+  bits<5> Zn;
+  bits<2> rot;
+  let Inst{31-24} = 0b01000100;
+  let Inst{23-22} = sz;
+  let Inst{21}    = 0b1;
+  let Inst{15-12} = opc;
+  let Inst{11-10} = rot;
+  let Inst{9-5}   = Zn;
+  let Inst{4-0}   = Zda;
+
+  let Constraints = "$Zda = $_Zda";
+  let DestructiveInstType = Destructive;
+  let ElementSize = ElementSizeNone;
+}
+
+multiclass sve2_cintx_dot_by_indexed_elem<string asm> {
+  def _S : sve2_complex_int_arith_indexed<0b10, 0b0100, asm, ZPR32, ZPR8, ZPR3b8, VectorIndexS> {
+    bits<2> iop;
+    bits<3> Zm;
+    let Inst{20-19} = iop;
+    let Inst{18-16} = Zm;
+  }
+  def _D : sve2_complex_int_arith_indexed<0b11, 0b0100, asm, ZPR64, ZPR16, ZPR4b16, VectorIndexD> {
+    bit iop;
+    bits<4> Zm;
+    let Inst{20} = iop;
+    let Inst{19-16} = Zm;
+  }
+}
+
+//===----------------------------------------------------------------------===//
 // SVE2 Integer Multiply - Unpredicated Group
 //===----------------------------------------------------------------------===//
 
 class sve2_int_mul<bits<2> sz, bits<3> opc, string asm, ZPRRegOp zprty>
 : I<(outs zprty:$Zd), (ins zprty:$Zn, zprty:$Zm),
   asm, "\t$Zd, $Zn, $Zm", "", []>, Sched<[]> {
   bits<5> Zd;
   bits<5> Zm;

test/MC/AArch64/SVE2/cdot-diagnostics.s

+// RUN: not llvm-mc -triple=aarch64 -show-encoding -mattr=+sve2  2>&1 < %s| FileCheck %s
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element size
+
+cdot  z0.s, z1.h, z31.h, #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: cdot  z0.s, z1.h, z31.h, #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.s, z1.s, z31.s, #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: cdot  z0.s, z1.s, z31.s, #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.s, z1.d, z31.d, #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: cdot  z0.s, z1.d, z31.d, #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.d, z1.b, z31.b, #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: cdot  z0.d, z1.b, z31.b, #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.d, z1.s, z31.s, #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: cdot  z0.d, z1.s, z31.s, #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.d, z1.d, z31.d, #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid element width
+// CHECK-NEXT: cdot  z0.d, z1.d, z31.d, #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// ------------------------------------------------------------------------- //
+// Invalid restricted register for indexed vector.
+
+cdot  z0.s, z1.b, z8.b[3], #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// CHECK-NEXT: cdot  z0.s, z1.b, z8.b[3], #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.d, z1.h, z16.h[1], #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// CHECK-NEXT: cdot  z0.d, z1.h, z16.h[1], #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// ------------------------------------------------------------------------- //
+// Invalid element index
+
+cdot  z0.s, z1.b, z7.b[-1], #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 3].
+// CHECK-NEXT: cdot  z0.s, z1.b, z7.b[-1], #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.s, z1.b, z7.b[4], #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 3].
+// CHECK-NEXT: cdot  z0.s, z1.b, z7.b[4], #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.d, z1.h, z15.h[-1], #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 1].
+// CHECK-NEXT: cdot  z0.d, z1.h, z15.h[-1], #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot  z0.d, z1.h, z15.h[2], #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: vector lane must be an integer in range [0, 1].
+// CHECK-NEXT: cdot  z0.d, z1.h, z15.h[2], #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// --------------------------------------------------------------------------//
+// Invalid rotation
+
+cdot z0.s, z1.b, z2.b[0], #360
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: complex rotation must be 0, 90, 180 or 270.
+// CHECK-NEXT: cdot z0.s, z1.b, z2.b[0], #360
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+cdot z0.d, z1.h, z2.h[0], #450
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: complex rotation must be 0, 90, 180 or 270.
+// CHECK-NEXT: cdot z0.d, z1.h, z2.h[0], #450
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
+// --------------------------------------------------------------------------//
+// Negative tests for instructions that are incompatible with movprfx
+
+movprfx z0.d, p0/z, z7.d
+cdot  z0.d, z1.h, z31.h, #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a predicated movprfx, suggest using unpredicated movprfx
+// CHECK-NEXT: cdot  z0.d, z1.h, z31.h, #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+movprfx z0.d, p0/z, z7.d
+cdot  z0.d, z1.h, z15.h[1], #0
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: instruction is unpredictable when following a predicated movprfx, suggest using unpredicated movprfx
+// CHECK-NEXT: cdot  z0.d, z1.h, z15.h[1], #0
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:

test/MC/AArch64/SVE2/cdot.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
+
+cdot  z0.s, z1.b, z31.b, #0
+// CHECK-INST: cdot z0.s, z1.b, z31.b, #0
+// CHECK-ENCODING: [0x20,0x10,0x9f,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 10 9f 44 <unknown>
+
+cdot  z0.d, z1.h, z31.h, #0
+// CHECK-INST: cdot z0.d, z1.h, z31.h, #0
+// CHECK-ENCODING: [0x20,0x10,0xdf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 10 df 44 <unknown>
+
+cdot  z0.d, z1.h, z31.h, #90
+// CHECK-INST: cdot z0.d, z1.h, z31.h, #90
+// CHECK-ENCODING: [0x20,0x14,0xdf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 14 df 44 <unknown>
+
+cdot  z0.d, z1.h, z31.h, #180
+// CHECK-INST: cdot z0.d, z1.h, z31.h, #180
+// CHECK-ENCODING: [0x20,0x18,0xdf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 18 df 44 <unknown>
+
+cdot  z0.d, z1.h, z31.h, #270
+// CHECK-INST: cdot z0.d, z1.h, z31.h, #270
+// CHECK-ENCODING: [0x20,0x1c,0xdf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 1c df 44 <unknown>
+
+cdot  z0.s, z1.b, z7.b[3], #0
+// CHECK-INST: cdot z0.s, z1.b, z7.b[3], #0
+// CHECK-ENCODING: [0x20,0x40,0xbf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 40 bf 44 <unknown>
+
+cdot  z0.d, z1.h, z15.h[1], #0
+// CHECK-INST: cdot z0.d, z1.h, z15.h[1], #0
+// CHECK-ENCODING: [0x20,0x40,0xff,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 40 ff 44 <unknown>
+
+cdot  z5.d, z6.h, z3.h[0], #90
+// CHECK-INST: cdot z5.d, z6.h, z3.h[0], #90
+// CHECK-ENCODING: [0xc5,0x44,0xe3,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: c5 44 e3 44 <unknown>
+
+cdot  z29.d, z30.h, z0.h[0], #180
+// CHECK-INST: cdot z29.d, z30.h, z0.h[0], #180
+// CHECK-ENCODING: [0xdd,0x4b,0xe0,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: dd 4b e0 44 <unknown>
+
+cdot  z31.d, z30.h, z7.h[1], #270
+// CHECK-INST: cdot z31.d, z30.h, z7.h[1], #270
+// CHECK-ENCODING: [0xdf,0x4f,0xf7,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: df 4f 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>
+
+cdot  z0.d, z1.h, z31.h, #0
+// CHECK-INST: cdot z0.d, z1.h, z31.h, #0
+// CHECK-ENCODING: [0x20,0x10,0xdf,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 10 df 44 <unknown>
+
+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>
+
+cdot  z0.d, z1.h, z15.h[1], #0
+// CHECK-INST: cdot z0.d, z1.h, z15.h[1], #0
+// CHECK-ENCODING: [0x20,0x40,0xff,0x44]
+// CHECK-ERROR: instruction requires: sve2
+// CHECK-UNKNOWN: 20 40 ff 44 <unknown>