[ARM] Add ARMv8.3 VCMLA and VCADD instructions

Add support for VCMLA and VCADD advanced SIMD complex number instructions.

The command line option is -march=armv8.3-a+fp16+simd for enabling all
instructions.

In arm-dis.c the formatting syntax was abused a bit to select between
0 vs 90 or 180 vs 270 or 90 vs 270 based on a bit value instead of
duplicating entries in the opcode table.

gas/
	* config/tc-arm.c (do_vcmla, do_vcadd): Define.
	(neon_scalar_for_vcmla): Define.
	(enum operand_parse_code): Add OP_IROT1 and OP_IROT2.
	(NEON_ENC_TAB): Add DDSI and QQSI variants.
	(insns): Add vcmla and vcadd.
	* testsuite/gas/arm/armv8_3-a-simd.d: New.
	* testsuite/gas/arm/armv8_3-a-simd.s: New.
	* testsuite/gas/arm/armv8_3-a-simd-bad.d: New.
	* testsuite/gas/arm/armv8_3-a-simd-bad.l: New.
	* testsuite/gas/arm/armv8_3-a-simd-bad.s: New.

opcodes/
	* arm-dis.c (coprocessor_opcodes): Add vcmla and vcadd.
	(print_insn_coprocessor): Add 'V' format for neon D or Q regs.

1 files changed, 80 insertions, 0 deletions

diff --git a/gas/config/tc-arm.c b/gas/config/tc-arm.c
index 3515fc6..41ab13e 100644
--- a/gas/config/tc-arm.c
+++ b/gas/config/tc-arm.c
@@ -6535,6 +6535,8 @@ enum operand_parse_code
   OP_EXPi,	/* same, with optional immediate prefix */
   OP_EXPr,	/* same, with optional relocation suffix */
   OP_HALF,	/* 0 .. 65535 or low/high reloc.  */
+  OP_IROT1,	/* VCADD rotate immediate: 90, 270.  */
+  OP_IROT2,	/* VCMLA rotate immediate: 0, 90, 180, 270.  */
 
   OP_CPSF,	/* CPS flags */
   OP_ENDI,	/* Endianness specifier */
@@ -13348,6 +13350,8 @@ NEON_ENC_TAB
   X(3, (D, Q, S), MIXED),		\
   X(4, (D, D, D, I), DOUBLE),		\
   X(4, (Q, Q, Q, I), QUAD),		\
+  X(4, (D, D, S, I), DOUBLE),		\
+  X(4, (Q, Q, S, I), QUAD),		\
   X(2, (F, F), SINGLE),			\
   X(3, (F, F, F), SINGLE),		\
   X(2, (F, I), SINGLE),			\
@@ -17261,6 +17265,80 @@ do_vrintm (void)
   do_vrint_1 (neon_cvt_mode_m);
 }
 
+static unsigned
+neon_scalar_for_vcmla (unsigned opnd, unsigned elsize)
+{
+  unsigned regno = NEON_SCALAR_REG (opnd);
+  unsigned elno = NEON_SCALAR_INDEX (opnd);
+
+  if (elsize == 16 && elno < 2 && regno < 16)
+    return regno | (elno << 4);
+  else if (elsize == 32 && elno == 0)
+    return regno;
+
+  first_error (_("scalar out of range"));
+  return 0;
+}
+
+static void
+do_vcmla (void)
+{
+  constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8),
+	      _(BAD_FPU));
+  constraint (inst.reloc.exp.X_op != O_constant, _("expression too complex"));
+  unsigned rot = inst.reloc.exp.X_add_number;
+  constraint (rot != 0 && rot != 90 && rot != 180 && rot != 270,
+	      _("immediate out of range"));
+  rot /= 90;
+  if (inst.operands[2].isscalar)
+    {
+      enum neon_shape rs = neon_select_shape (NS_DDSI, NS_QQSI, NS_NULL);
+      unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
+				       N_KEY | N_F16 | N_F32).size;
+      unsigned m = neon_scalar_for_vcmla (inst.operands[2].reg, size);
+      inst.is_neon = 1;
+      inst.instruction = 0xfe000800;
+      inst.instruction |= LOW4 (inst.operands[0].reg) << 12;
+      inst.instruction |= HI1 (inst.operands[0].reg) << 22;
+      inst.instruction |= LOW4 (inst.operands[1].reg) << 16;
+      inst.instruction |= HI1 (inst.operands[1].reg) << 7;
+      inst.instruction |= LOW4 (m);
+      inst.instruction |= HI1 (m) << 5;
+      inst.instruction |= neon_quad (rs) << 6;
+      inst.instruction |= rot << 20;
+      inst.instruction |= (size == 32) << 23;
+    }
+  else
+    {
+      enum neon_shape rs = neon_select_shape (NS_DDDI, NS_QQQI, NS_NULL);
+      unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
+				       N_KEY | N_F16 | N_F32).size;
+      neon_three_same (neon_quad (rs), 0, -1);
+      inst.instruction &= 0x00ffffff; /* Undo neon_dp_fixup.  */
+      inst.instruction |= 0xfc200800;
+      inst.instruction |= rot << 23;
+      inst.instruction |= (size == 32) << 20;
+    }
+}
+
+static void
+do_vcadd (void)
+{
+  constraint (!ARM_CPU_HAS_FEATURE (cpu_variant, fpu_neon_ext_armv8),
+	      _(BAD_FPU));
+  constraint (inst.reloc.exp.X_op != O_constant, _("expression too complex"));
+  unsigned rot = inst.reloc.exp.X_add_number;
+  constraint (rot != 90 && rot != 270, _("immediate out of range"));
+  enum neon_shape rs = neon_select_shape (NS_DDDI, NS_QQQI, NS_NULL);
+  unsigned size = neon_check_type (3, rs, N_EQK, N_EQK,
+				   N_KEY | N_F16 | N_F32).size;
+  neon_three_same (neon_quad (rs), 0, -1);
+  inst.instruction &= 0x00ffffff; /* Undo neon_dp_fixup.  */
+  inst.instruction |= 0xfc800800;
+  inst.instruction |= (rot == 270) << 24;
+  inst.instruction |= (size == 32) << 20;
+}
+
 /* Crypto v1 instructions.  */
 static void
 do_crypto_2op_1 (unsigned elttype, int op)
@@ -19799,6 +19877,8 @@ static const struct asm_opcode insns[] =
 #undef  THUMB_VARIANT
 #define THUMB_VARIANT & arm_ext_v8_3
  NCE (vjcvt, eb90bc0, 2, (RVS, RVD), vjcvt),
+ NUF (vcmla, 0, 4, (RNDQ, RNDQ, RNDQ_RNSC, EXPi), vcmla),
+ NUF (vcadd, 0, 4, (RNDQ, RNDQ, RNDQ, EXPi), vcadd),
 
 #undef  ARM_VARIANT
 #define ARM_VARIANT  & fpu_fpa_ext_v1  /* Core FPA instruction set (V1).  */