[AArch64] Additional SVE instructions

This patch supports some additions to the SVE architecture prior to
its public release.

include/
	* opcode/aarch64.h (AARCH64_OPND_SVE_ADDR_RI_S4x16)
	(AARCH64_OPND_SVE_IMM_ROT1, AARCH64_OPND_SVE_IMM_ROT2)
	(AARCH64_OPND_SVE_Zm3_INDEX, AARCH64_OPND_SVE_Zm3_22_INDEX)
	(AARCH64_OPND_SVE_Zm4_INDEX): New aarch64_opnds.

opcodes/
	* aarch64-tbl.h (OP_SVE_HMH, OP_SVE_VMU_HSD, OP_SVE_VMVU_HSD)
	(OP_SVE_VMVV_HSD, OP_SVE_VMVVU_HSD, OP_SVE_VM_HSD, OP_SVE_VUVV_HSD)
	(OP_SVE_VUV_HSD, OP_SVE_VU_HSD, OP_SVE_VVVU_H, OP_SVE_VVVU_S)
	(OP_SVE_VVVU_HSD, OP_SVE_VVV_D, OP_SVE_VVV_D_H, OP_SVE_VVV_H)
	(OP_SVE_VVV_HSD, OP_SVE_VVV_S, OP_SVE_VVV_S_B, OP_SVE_VVV_SD_BH)
	(OP_SVE_VV_BHSDQ, OP_SVE_VV_HSD, OP_SVE_VZVV_HSD, OP_SVE_VZV_HSD)
	(OP_SVE_V_HSD): New macros.
	(OP_SVE_VMU_SD, OP_SVE_VMVU_SD, OP_SVE_VM_SD, OP_SVE_VUVV_SD)
	(OP_SVE_VU_SD, OP_SVE_VVVU_SD, OP_SVE_VVV_SD, OP_SVE_VZVV_SD)
	(OP_SVE_VZV_SD, OP_SVE_V_SD): Delete.
	(aarch64_opcode_table): Add new SVE instructions.
	(aarch64_opcode_table): Use imm_rotate{1,2} instead of imm_rotate
	for rotation operands.  Add new SVE operands.
	* aarch64-asm.h (ins_sve_addr_ri_s4): New inserter.
	(ins_sve_quad_index): Likewise.
	(ins_imm_rotate): Split into...
	(ins_imm_rotate1, ins_imm_rotate2): ...these two inserters.
	* aarch64-asm.c (aarch64_ins_imm_rotate): Split into...
	(aarch64_ins_imm_rotate1, aarch64_ins_imm_rotate2): ...these two
	functions.
	(aarch64_ins_sve_addr_ri_s4): New function.
	(aarch64_ins_sve_quad_index): Likewise.
	(do_misc_encoding): Handle "MOV Zn.Q, Qm".
	* aarch64-asm-2.c: Regenerate.
	* aarch64-dis.h (ext_sve_addr_ri_s4): New extractor.
	(ext_sve_quad_index): Likewise.
	(ext_imm_rotate): Split into...
	(ext_imm_rotate1, ext_imm_rotate2): ...these two extractors.
	* aarch64-dis.c (aarch64_ext_imm_rotate): Split into...
	(aarch64_ext_imm_rotate1, aarch64_ext_imm_rotate2): ...these two
	functions.
	(aarch64_ext_sve_addr_ri_s4): New function.
	(aarch64_ext_sve_quad_index): Likewise.
	(aarch64_ext_sve_index): Allow quad indices.
	(do_misc_decoding): Likewise.
	* aarch64-dis-2.c: Regenerate.
	* aarch64-opc.h (FLD_SVE_i3h, FLD_SVE_rot1, FLD_SVE_rot2): New
	aarch64_field_kinds.
	(OPD_F_OD_MASK): Widen by one bit.
	(OPD_F_NO_ZR): Bump accordingly.
	(get_operand_field_width): New function.
	* aarch64-opc.c (fields): Add new SVE fields.
	(operand_general_constraint_met_p): Handle new SVE operands.
	(aarch64_print_operand): Likewise.
	* aarch64-opc-2.c: Regenerate.

gas/
	* doc/c-aarch64.texi: Document that sve implies fp16, simd and compnum.
	* config/tc-aarch64.c (parse_vector_type_for_operand): Allow .q
	to be used with SVE registers.
	(parse_operands): Handle new SVE operands.
	(aarch64_features): Make "sve" require F16 rather than FP.  Also
	require COMPNUM.
	* testsuite/gas/aarch64/sve.s: Add tests for new instructions.
	Include compnum tests.
	* testsuite/gas/aarch64/sve.d: Update accordingly.
	* testsuite/gas/aarch64/sve-invalid.s: Add tests for new instructions.
	* testsuite/gas/aarch64/sve-invalid.l: Update accordingly.  Also
	update expected output for new FMOV and MOV alternatives.

1 files changed, 48 insertions, 31 deletions

diff --git a/opcodes/aarch64-dis.c b/opcodes/aarch64-dis.c
index d08e81f..b528af6 100644
--- a/opcodes/aarch64-dis.c
+++ b/opcodes/aarch64-dis.c
@@ -711,36 +711,26 @@ aarch64_ext_fpimm (const aarch64_operand *self, aarch64_opnd_info *info,
   return 1;
 }
 
-/* Decode rotate immediate for FCMLA <Vd>.<T>, <Vn>.<T>, <Vm>.<T>, #rotate.  */
+/* Decode a 1-bit rotate immediate (#90 or #270).  */
 int
-aarch64_ext_imm_rotate (const aarch64_operand *self, aarch64_opnd_info *info,
-			const aarch64_insn code,
-			const aarch64_inst *inst ATTRIBUTE_UNUSED)
+aarch64_ext_imm_rotate1 (const aarch64_operand *self, aarch64_opnd_info *info,
+			 const aarch64_insn code,
+			 const aarch64_inst *inst ATTRIBUTE_UNUSED)
 {
   uint64_t rot = extract_field (self->fields[0], code, 0);
+  assert (rot < 2U);
+  info->imm.value = rot * 180 + 90;
+  return 1;
+}
 
-  switch (info->type)
-    {
-    case AARCH64_OPND_IMM_ROT1:
-    case AARCH64_OPND_IMM_ROT2:
-      /* rot	value
-	 0	0
-	 1	90
-	 2	180
-	 3	270  */
-      assert (rot < 4U);
-      break;
-    case AARCH64_OPND_IMM_ROT3:
-      /* rot	value
-	 0	90
-	 1	270  */
-      assert (rot < 2U);
-      rot = 2 * rot + 1;
-      break;
-    default:
-      assert (0);
-      return 0;
-    }
+/* Decode a 2-bit rotate immediate (#0, #90, #180 or #270).  */
+int
+aarch64_ext_imm_rotate2 (const aarch64_operand *self, aarch64_opnd_info *info,
+			 const aarch64_insn code,
+			 const aarch64_inst *inst ATTRIBUTE_UNUSED)
+{
+  uint64_t rot = extract_field (self->fields[0], code, 0);
+  assert (rot < 4U);
   info->imm.value = rot * 90;
   return 1;
 }
@@ -1364,6 +1354,18 @@ aarch64_ext_sve_addr_reg_imm (const aarch64_operand *self,
   return 1;
 }
 
+/* Decode an SVE address [X<n>, #<SVE_imm4> << <shift>], where <SVE_imm4>
+   is a 4-bit signed number and where <shift> is SELF's operand-dependent
+   value.  fields[0] specifies the base register field.  */
+int
+aarch64_ext_sve_addr_ri_s4 (const aarch64_operand *self,
+			    aarch64_opnd_info *info, aarch64_insn code,
+			    const aarch64_inst *inst ATTRIBUTE_UNUSED)
+{
+  int offset = sign_extend (extract_field (FLD_SVE_imm4, code, 0), 3);
+  return aarch64_ext_sve_addr_reg_imm (self, info, code, offset);
+}
+
 /* Decode an SVE address [X<n>, #<SVE_imm6> << <shift>], where <SVE_imm6>
    is a 6-bit unsigned number and where <shift> is SELF's operand-dependent
    value.  fields[0] specifies the base register field.  */
@@ -1591,7 +1593,7 @@ aarch64_ext_sve_index (const aarch64_operand *self,
 
   info->reglane.regno = extract_field (self->fields[0], code, 0);
   val = extract_fields (code, 0, 2, FLD_SVE_tszh, FLD_imm5);
-  if ((val & 15) == 0)
+  if ((val & 31) == 0)
     return 0;
   while ((val & 1) == 0)
     val /= 2;
@@ -1610,6 +1612,21 @@ aarch64_ext_sve_limm_mov (const aarch64_operand *self,
 	  && aarch64_sve_dupm_mov_immediate_p (info->imm.value, esize));
 }
 
+/* Decode Zn[MM], where Zn occupies the least-significant part of the field
+   and where MM occupies the most-significant part.  The operand-dependent
+   value specifies the number of bits in Zn.  */
+int
+aarch64_ext_sve_quad_index (const aarch64_operand *self,
+			    aarch64_opnd_info *info, aarch64_insn code,
+			    const aarch64_inst *inst ATTRIBUTE_UNUSED)
+{
+  unsigned int reg_bits = get_operand_specific_data (self);
+  unsigned int val = extract_all_fields (self, code);
+  info->reglane.regno = val & ((1 << reg_bits) - 1);
+  info->reglane.index = val >> reg_bits;
+  return 1;
+}
+
 /* Decode {Zn.<T> - Zm.<T>}.  The fields array specifies which field
    to use for Zn.  The opcode-dependent value specifies the number
    of registers in the list.  */
@@ -1907,7 +1924,7 @@ do_misc_decoding (aarch64_inst *inst)
     case OP_MOV_Z_V:
       /* Index must be zero.  */
       value = extract_fields (inst->value, 0, 2, FLD_SVE_tszh, FLD_imm5);
-      return value == 1 || value == 2 || value == 4 || value == 8;
+      return value > 0 && value <= 16 && value == (value & -value);
 
     case OP_MOV_Z_Z:
       return (extract_field (FLD_SVE_Zn, inst->value, 0)
@@ -1916,7 +1933,7 @@ do_misc_decoding (aarch64_inst *inst)
     case OP_MOV_Z_Zi:
       /* Index must be nonzero.  */
       value = extract_fields (inst->value, 0, 2, FLD_SVE_tszh, FLD_imm5);
-      return value != 1 && value != 2 && value != 4 && value != 8;
+      return value > 0 && value != (value & -value);
 
     case OP_MOVM_P_P_P:
       return (extract_field (FLD_SVE_Pd, inst->value, 0)
@@ -2573,8 +2590,8 @@ aarch64_decode_variant_using_iclass (aarch64_inst *inst)
       break;
 
     case sve_index:
-      i = extract_field (FLD_SVE_tsz, inst->value, 0);
-      if (i == 0)
+      i = extract_fields (inst->value, 0, 2, FLD_SVE_tszh, FLD_imm5);
+      if ((i & 31) == 0)
 	return FALSE;
       while ((i & 1) == 0)
 	{