[AArch64] Add ARMv8.3 FCMLA and FCADD instructions

Message ID 58207461.9080006@arm.com
State New
Headers show

Commit Message

Szabolcs Nagy Nov. 7, 2016, 12:32 p.m.
Add support for FCMLA and FCADD complex arithmetic SIMD instructions.
FCMLA has an indexed element variant where the index range has to be
treated specially because a complex number takes two elements and the
indexed vector size depends on the other operands.

These complex number SIMD instructions are part of ARMv8.3
https://community.arm.com/groups/processors/blog/2016/10/27/armv8-a-architecture-2016-additions

(generated files are not in the diff.)

gas/
2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

	* config/tc-aarch64.c (parse_operands): Handle AARCH64_OPND_IMM_ROT*.

include/
2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

	* opcode/aarch64.h (enum aarch64_opnd): Add AARCH64_OPND_IMM_ROT1,
	AARCH64_OPND_IMM_ROT2, AARCH64_OPND_IMM_ROT3.
	(enum aarch64_op): Add OP_FCMLA_ELEM.

opcodes/
2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

	* aarch64-tbl.h (QL_V3SAMEHSD_ROT, QL_ELEMENT_ROT): Define.
	(aarch64_feature_simd_v8_3, SIMD_V8_3): Define.
	(aarch64_opcode_table): Add fcmla and fcadd.
	(AARCH64_OPERANDS): Add IMM_ROT{1,2,3}.
	* aarch64-asm.h (aarch64_ins_imm_rotate): Declare.
	* aarch64-asm.c (aarch64_ins_imm_rotate): Define.
	* aarch64-dis.h (aarch64_ext_imm_rotate): Declare.
	* aarch64-dis.c (aarch64_ext_imm_rotate): Define.
	* aarch64-opc.h (enum aarch64_field_kind): Add FLD_rotate{1,2,3}.
	* aarch64-opc.c (fields): Add FLD_rotate{1,2,3}.
	(operand_general_constraint_met_p): Rotate and index range check.
	(aarch64_print_operand): Handle rotate operand.
	* aarch64-asm-2.c: Regenerate.
	* aarch64-dis-2.c: Likewise.
	* aarch64-opc-2.c: Likewise.

gas/testsuite/
2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

	* gas/aarch64/advsimd-armv8_3.d: New.
	* gas/aarch64/advsimd-armv8_3.s: New.
	* gas/aarch64/illegal-fcmla.s: New.
	* gas/aarch64/illegal-fcmla.l: New.
	* gas/aarch64/illegal-fcmla.d: New.

Comments

Nick Clifton Nov. 10, 2016, 2:46 p.m. | #1
Hi  Szabolcs,

> gas/

> 2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

> 

> 	* config/tc-aarch64.c (parse_operands): Handle AARCH64_OPND_IMM_ROT*.

> 

> include/

> 2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

> 

> 	* opcode/aarch64.h (enum aarch64_opnd): Add AARCH64_OPND_IMM_ROT1,

> 	AARCH64_OPND_IMM_ROT2, AARCH64_OPND_IMM_ROT3.

> 	(enum aarch64_op): Add OP_FCMLA_ELEM.

> 

> opcodes/

> 2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

> 

> 	* aarch64-tbl.h (QL_V3SAMEHSD_ROT, QL_ELEMENT_ROT): Define.

> 	(aarch64_feature_simd_v8_3, SIMD_V8_3): Define.

> 	(aarch64_opcode_table): Add fcmla and fcadd.

> 	(AARCH64_OPERANDS): Add IMM_ROT{1,2,3}.

> 	* aarch64-asm.h (aarch64_ins_imm_rotate): Declare.

> 	* aarch64-asm.c (aarch64_ins_imm_rotate): Define.

> 	* aarch64-dis.h (aarch64_ext_imm_rotate): Declare.

> 	* aarch64-dis.c (aarch64_ext_imm_rotate): Define.

> 	* aarch64-opc.h (enum aarch64_field_kind): Add FLD_rotate{1,2,3}.

> 	* aarch64-opc.c (fields): Add FLD_rotate{1,2,3}.

> 	(operand_general_constraint_met_p): Rotate and index range check.

> 	(aarch64_print_operand): Handle rotate operand.

> 	* aarch64-asm-2.c: Regenerate.

> 	* aarch64-dis-2.c: Likewise.

> 	* aarch64-opc-2.c: Likewise.

> 

> gas/testsuite/

> 2016-11-07  Szabolcs Nagy  <szabolcs.nagy@arm.com>

> 

> 	* gas/aarch64/advsimd-armv8_3.d: New.

> 	* gas/aarch64/advsimd-armv8_3.s: New.

> 	* gas/aarch64/illegal-fcmla.s: New.

> 	* gas/aarch64/illegal-fcmla.l: New.

> 	* gas/aarch64/illegal-fcmla.d: New.


Approved - please apply.

Cheers
  Nick

Patch hide | download patch | download mbox

diff --git a/gas/config/tc-aarch64.c b/gas/config/tc-aarch64.c
index a9cee79..8b839c6 100644
--- a/gas/config/tc-aarch64.c
+++ b/gas/config/tc-aarch64.c
@@ -5561,6 +5561,9 @@  parse_operands (char *str, const aarch64_opcode *opcode)
 	case AARCH64_OPND_SVE_UIMM7:
 	case AARCH64_OPND_SVE_UIMM8:
 	case AARCH64_OPND_SVE_UIMM8_53:
+	case AARCH64_OPND_IMM_ROT1:
+	case AARCH64_OPND_IMM_ROT2:
+	case AARCH64_OPND_IMM_ROT3:
 	  po_imm_nc_or_fail ();
 	  info->imm.value = val;
 	  break;
diff --git a/gas/testsuite/gas/aarch64/advsimd-armv8_3.d b/gas/testsuite/gas/aarch64/advsimd-armv8_3.d
new file mode 100644
index 0000000..0e477e4
--- /dev/null
+++ b/gas/testsuite/gas/aarch64/advsimd-armv8_3.d
@@ -0,0 +1,33 @@ 
+#as: -march=armv8.3-a
+#objdump: -dr
+
+.*:     file format .*
+
+Disassembly of section \.text:
+
+0000000000000000 <.*>:
+[^:]+:	6ec3c441 	fcmla	v1.2d, v2.2d, v3.2d, #0
+[^:]+:	6ec3cc41 	fcmla	v1.2d, v2.2d, v3.2d, #90
+[^:]+:	6ec3d441 	fcmla	v1.2d, v2.2d, v3.2d, #180
+[^:]+:	6ec3dc41 	fcmla	v1.2d, v2.2d, v3.2d, #270
+[^:]+:	2e83cc41 	fcmla	v1.2s, v2.2s, v3.2s, #90
+[^:]+:	6e83cc41 	fcmla	v1.4s, v2.4s, v3.4s, #90
+[^:]+:	2e43cc41 	fcmla	v1.4h, v2.4h, v3.4h, #90
+[^:]+:	6e43cc41 	fcmla	v1.8h, v2.8h, v3.8h, #90
+[^:]+:	6f831041 	fcmla	v1.4s, v2.4s, v3.s\[0\], #0
+[^:]+:	6f833041 	fcmla	v1.4s, v2.4s, v3.s\[0\], #90
+[^:]+:	6f835041 	fcmla	v1.4s, v2.4s, v3.s\[0\], #180
+[^:]+:	6f837041 	fcmla	v1.4s, v2.4s, v3.s\[0\], #270
+[^:]+:	6f833841 	fcmla	v1.4s, v2.4s, v3.s\[1\], #90
+[^:]+:	2f433041 	fcmla	v1.4h, v2.4h, v3.h\[0\], #90
+[^:]+:	2f633041 	fcmla	v1.4h, v2.4h, v3.h\[1\], #90
+[^:]+:	6f433041 	fcmla	v1.8h, v2.8h, v3.h\[0\], #90
+[^:]+:	6f633041 	fcmla	v1.8h, v2.8h, v3.h\[1\], #90
+[^:]+:	6f433841 	fcmla	v1.8h, v2.8h, v3.h\[2\], #90
+[^:]+:	6f633841 	fcmla	v1.8h, v2.8h, v3.h\[3\], #90
+[^:]+:	6ec3e441 	fcadd	v1.2d, v2.2d, v3.2d, #90
+[^:]+:	6ec3f441 	fcadd	v1.2d, v2.2d, v3.2d, #270
+[^:]+:	2e83e441 	fcadd	v1.2s, v2.2s, v3.2s, #90
+[^:]+:	6e83e441 	fcadd	v1.4s, v2.4s, v3.4s, #90
+[^:]+:	2e43e441 	fcadd	v1.4h, v2.4h, v3.4h, #90
+[^:]+:	6e43e441 	fcadd	v1.8h, v2.8h, v3.8h, #90
diff --git a/gas/testsuite/gas/aarch64/advsimd-armv8_3.s b/gas/testsuite/gas/aarch64/advsimd-armv8_3.s
new file mode 100644
index 0000000..bfb94e5
--- /dev/null
+++ b/gas/testsuite/gas/aarch64/advsimd-armv8_3.s
@@ -0,0 +1,36 @@ 
+/* Test file for ARMv8.3 complex arithmetics instructions.  */
+	.text
+
+	/* Three-same operands FCMLA.  */
+	fcmla	v1.2d, v2.2d, v3.2d, #0
+	fcmla	v1.2d, v2.2d, v3.2d, #90
+	fcmla	v1.2d, v2.2d, v3.2d, #180
+	fcmla	v1.2d, v2.2d, v3.2d, #270
+
+	fcmla	v1.2s, v2.2s, v3.2s, #90
+	fcmla	v1.4s, v2.4s, v3.4s, #90
+	fcmla	v1.4h, v2.4h, v3.4h, #90
+	fcmla	v1.8h, v2.8h, v3.8h, #90
+
+	/* Indexed element FCMLA.  */
+	fcmla	v1.4s, v2.4s, v3.s[0], #0
+	fcmla	v1.4s, v2.4s, v3.s[0], #90
+	fcmla	v1.4s, v2.4s, v3.s[0], #180
+	fcmla	v1.4s, v2.4s, v3.s[0], #270
+	fcmla	v1.4s, v2.4s, v3.s[1], #90
+
+	fcmla	v1.4h, v2.4h, v3.h[0], #90
+	fcmla	v1.4h, v2.4h, v3.h[1], #90
+	fcmla	v1.8h, v2.8h, v3.h[0], #90
+	fcmla	v1.8h, v2.8h, v3.h[1], #90
+	fcmla	v1.8h, v2.8h, v3.h[2], #90
+	fcmla	v1.8h, v2.8h, v3.h[3], #90
+
+	/* Three-same operands FADD.  */
+	fcadd	v1.2d, v2.2d, v3.2d, #90
+	fcadd	v1.2d, v2.2d, v3.2d, #270
+
+	fcadd	v1.2s, v2.2s, v3.2s, #90
+	fcadd	v1.4s, v2.4s, v3.4s, #90
+	fcadd	v1.4h, v2.4h, v3.4h, #90
+	fcadd	v1.8h, v2.8h, v3.8h, #90
diff --git a/gas/testsuite/gas/aarch64/illegal-fcmla.d b/gas/testsuite/gas/aarch64/illegal-fcmla.d
new file mode 100644
index 0000000..af94c51
--- /dev/null
+++ b/gas/testsuite/gas/aarch64/illegal-fcmla.d
@@ -0,0 +1,2 @@ 
+#as: -march=armv8.3-a -mno-verbose-error
+#error-output: illegal-fcmla.l
diff --git a/gas/testsuite/gas/aarch64/illegal-fcmla.l b/gas/testsuite/gas/aarch64/illegal-fcmla.l
new file mode 100644
index 0000000..1c0f1e4
--- /dev/null
+++ b/gas/testsuite/gas/aarch64/illegal-fcmla.l
@@ -0,0 +1,17 @@ 
+[^:]*: Assembler messages:
+[^:]+:10: Error: rotate expected to be 0, 90, 180 or 270 at operand 4 -- `fcmla v0\.4s,v1\.4s,v2\.s\[0\],#-90'
+[^:]+:11: Error: rotate expected to be 0, 90, 180 or 270 at operand 4 -- `fcmla v0\.4s,v1\.4s,v2\.s\[0\],#30'
+[^:]+:12: Error: rotate expected to be 0, 90, 180 or 270 at operand 4 -- `fcmla v0\.4s,v1\.4s,v2\.s\[0\],#360'
+[^:]+:13: Error: register element index out of range 0 to 1 at operand 3 -- `fcmla v0\.4h,v1\.4h,v2\.h\[2\],#90'
+[^:]+:14: Error: register element index out of range 0 to 3 at operand 3 -- `fcmla v0\.8h,v1\.8h,v2\.h\[4\],#90'
+[^:]+:15: Error: register element index out of range 0 to 1 at operand 3 -- `fcmla v0\.4s,v1\.4s,v2\.s\[2\],#90'
+[^:]+:16: Error: operand mismatch -- `fcmla v0\.2s,v1\.2s,v2\.s\[0\],#90'
+[^:]+:17: Error: operand mismatch -- `fcmla v0\.4s,v1\.4s,v2\.d\[0\],#90'
+[^:]+:18: Error: operand mismatch -- `fcmla v0\.2d,v1\.2d,v2\.d\[0\],#0'
+[^:]+:19: Error: rotate expected to be 0, 90, 180 or 270 at operand 4 -- `fcmla v0\.4s,v1\.4s,v2\.4s,#-90'
+[^:]+:20: Error: rotate expected to be 0, 90, 180 or 270 at operand 4 -- `fcmla v0\.4s,v1\.4s,v2\.4s,#30'
+[^:]+:21: Error: rotate expected to be 0, 90, 180 or 270 at operand 4 -- `fcmla v0\.4s,v1\.4s,v2\.4s,#360'
+[^:]+:22: Error: invalid element size 8 and vector size combination s at operand 1 -- `fcmla v0\.8s,v1\.8s,v2\.8s,#0'
+[^:]+:23: Error: operand mismatch -- `fcmla v0\.1d,v1\.1d,v2\.1d,#0'
+[^:]+:24: Error: rotate expected to be 90 or 270 at operand 4 -- `fcadd v0\.4h,v1\.4h,v2\.4h,#0'
+[^:]+:25: Error: rotate expected to be 90 or 270 at operand 4 -- `fcadd v0\.4h,v1\.4h,v2\.4h,#180'
diff --git a/gas/testsuite/gas/aarch64/illegal-fcmla.s b/gas/testsuite/gas/aarch64/illegal-fcmla.s
new file mode 100644
index 0000000..882c7ba
--- /dev/null
+++ b/gas/testsuite/gas/aarch64/illegal-fcmla.s
@@ -0,0 +1,25 @@ 
+// Test illegal ARMv8.3 FCMLA and FCADD instructions with -march=armv8.3-a.
+.text
+
+	// Good.
+	fcmla v0.4s, v1.4s, v2.s[0], #90
+	fcmla v0.4s, v1.4s, v2.4s, #90
+	fcadd v0.4h, v1.4h, v2.4h, #90
+
+	// Bad.
+	fcmla v0.4s, v1.4s, v2.s[0], #-90
+	fcmla v0.4s, v1.4s, v2.s[0], #30
+	fcmla v0.4s, v1.4s, v2.s[0], #360
+	fcmla v0.4h, v1.4h, v2.h[2], #90
+	fcmla v0.8h, v1.8h, v2.h[4], #90
+	fcmla v0.4s, v1.4s, v2.s[2], #90
+	fcmla v0.2s, v1.2s, v2.s[0], #90
+	fcmla v0.4s, v1.4s, v2.d[0], #90
+	fcmla v0.2d, v1.2d, v2.d[0], #0
+	fcmla v0.4s, v1.4s, v2.4s, #-90
+	fcmla v0.4s, v1.4s, v2.4s, #30
+	fcmla v0.4s, v1.4s, v2.4s, #360
+	fcmla v0.8s, v1.8s, v2.8s, #0
+	fcmla v0.1d, v1.1d, v2.1d, #0
+	fcadd v0.4h, v1.4h, v2.4h, #0
+	fcadd v0.4h, v1.4h, v2.4h, #180
diff --git a/include/opcode/aarch64.h b/include/opcode/aarch64.h
index a35a813..69645d0 100644
--- a/include/opcode/aarch64.h
+++ b/include/opcode/aarch64.h
@@ -204,6 +204,9 @@  enum aarch64_opnd
   AARCH64_OPND_HALF,	/* #<imm16>{, LSL #<shift>} operand in move wide.  */
   AARCH64_OPND_FBITS,	/* FP #<fbits> operand in e.g. SCVTF */
   AARCH64_OPND_IMM_MOV,	/* Immediate operand for the MOV alias.  */
+  AARCH64_OPND_IMM_ROT1,	/* Immediate rotate operand for FCMLA.  */
+  AARCH64_OPND_IMM_ROT2,	/* Immediate rotate operand for indexed FCMLA.  */
+  AARCH64_OPND_IMM_ROT3,	/* Immediate rotate operand for FCADD.  */
 
   AARCH64_OPND_COND,	/* Standard condition as the last operand.  */
   AARCH64_OPND_COND1,	/* Same as the above, but excluding AL and NV.  */
@@ -596,6 +599,8 @@  enum aarch64_op
   OP_NOTS_P_P_P_Z,
   OP_NOT_P_P_P_Z,
 
+  OP_FCMLA_ELEM,	/* ARMv8.3, indexed element version.  */
+
   OP_TOTAL_NUM,		/* Pseudo.  */
 };
 
diff --git a/opcodes/aarch64-asm.c b/opcodes/aarch64-asm.c
index 64051a9..cfe6630 100644
--- a/opcodes/aarch64-asm.c
+++ b/opcodes/aarch64-asm.c
@@ -125,19 +125,28 @@  aarch64_ins_reglane (const aarch64_operand *self, const aarch64_opnd_info *info,
     {
       /* index for e.g. SQDMLAL <Va><d>, <Vb><n>, <Vm>.<Ts>[<index>]
          or SQDMLAL <Va><d>, <Vb><n>, <Vm>.<Ts>[<index>].  */
+      unsigned index = info->reglane.index;
+
+      if (inst->opcode->op == OP_FCMLA_ELEM)
+	/* Complex operand takes two elements.  */
+	index *= 2;
+
       switch (info->qualifier)
 	{
 	case AARCH64_OPND_QLF_S_H:
 	  /* H:L:M */
-	  insert_fields (code, info->reglane.index, 0, 3, FLD_M, FLD_L, FLD_H);
+	  assert (index < 8);
+	  insert_fields (code, index, 0, 3, FLD_M, FLD_L, FLD_H);
 	  break;
 	case AARCH64_OPND_QLF_S_S:
 	  /* H:L */
-	  insert_fields (code, info->reglane.index, 0, 2, FLD_L, FLD_H);
+	  assert (index < 4);
+	  insert_fields (code, index, 0, 2, FLD_L, FLD_H);
 	  break;
 	case AARCH64_OPND_QLF_S_D:
 	  /* H */
-	  insert_field (FLD_H, code, info->reglane.index, 0);
+	  assert (index < 2);
+	  insert_field (FLD_H, code, index, 0);
 	  break;
 	default:
 	  assert (0);
@@ -427,6 +436,41 @@  aarch64_ins_fpimm (const aarch64_operand *self, const aarch64_opnd_info *info,
   return NULL;
 }
 
+/* Insert field rot for the rotate immediate in
+   FCMLA <Vd>.<T>, <Vn>.<T>, <Vm>.<T>, #<rotate>.  */
+const char *
+aarch64_ins_imm_rotate (const aarch64_operand *self,
+			const aarch64_opnd_info *info,
+			aarch64_insn *code, const aarch64_inst *inst)
+{
+  uint64_t rot = info->imm.value / 90;
+
+  switch (info->type)
+    {
+    case AARCH64_OPND_IMM_ROT1:
+    case AARCH64_OPND_IMM_ROT2:
+      /* value	rot
+	 0	0
+	 90	1
+	 180	2
+	 270	3  */
+      assert (rot < 4U);
+      break;
+    case AARCH64_OPND_IMM_ROT3:
+      /* value	rot
+	 90	0
+	 270	1  */
+      rot = (rot - 1) / 2;
+      assert (rot < 2U);
+      break;
+    default:
+      assert (0);
+    }
+  insert_field (self->fields[0], code, rot, inst->opcode->mask);
+
+  return NULL;
+}
+
 /* Insert #<fbits> for the immediate operand in fp fix-point instructions,
    e.g.  SCVTF <Dd>, <Wn>, #<fbits>.  */
 const char *
diff --git a/opcodes/aarch64-asm.h b/opcodes/aarch64-asm.h
index b49d3d5..7fce657 100644
--- a/opcodes/aarch64-asm.h
+++ b/opcodes/aarch64-asm.h
@@ -92,6 +92,7 @@  AARCH64_DECL_OPD_INSERTER (ins_sve_reglist);
 AARCH64_DECL_OPD_INSERTER (ins_sve_scale);
 AARCH64_DECL_OPD_INSERTER (ins_sve_shlimm);
 AARCH64_DECL_OPD_INSERTER (ins_sve_shrimm);
+AARCH64_DECL_OPD_INSERTER (ins_imm_rotate);
 
 #undef AARCH64_DECL_OPD_INSERTER
 
diff --git a/opcodes/aarch64-dis.c b/opcodes/aarch64-dis.c
index 245749c..ed58058 100644
--- a/opcodes/aarch64-dis.c
+++ b/opcodes/aarch64-dis.c
@@ -351,6 +351,14 @@  aarch64_ext_reglane (const aarch64_operand *self, aarch64_opnd_info *info,
 	default:
 	  return 0;
 	}
+
+      if (inst->opcode->op == OP_FCMLA_ELEM)
+	{
+	  /* Complex operand takes two elements.  */
+	  if (info->reglane.index & 1)
+	    return 0;
+	  info->reglane.index /= 2;
+	}
     }
 
   return 1;
@@ -703,6 +711,40 @@  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.  */
+int
+aarch64_ext_imm_rotate (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);
+
+  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;
+    }
+  info->imm.value = rot * 90;
+  return 1;
+}
+
 /* Decode scale for e.g. SCVTF <Dd>, <Wn>, #<fbits>.  */
 int
 aarch64_ext_fbits (const aarch64_operand *self ATTRIBUTE_UNUSED,
diff --git a/opcodes/aarch64-dis.h b/opcodes/aarch64-dis.h
index 48be5e4..b11f261 100644
--- a/opcodes/aarch64-dis.h
+++ b/opcodes/aarch64-dis.h
@@ -114,6 +114,7 @@  AARCH64_DECL_OPD_EXTRACTOR (ext_sve_reglist);
 AARCH64_DECL_OPD_EXTRACTOR (ext_sve_scale);
 AARCH64_DECL_OPD_EXTRACTOR (ext_sve_shlimm);
 AARCH64_DECL_OPD_EXTRACTOR (ext_sve_shrimm);
+AARCH64_DECL_OPD_EXTRACTOR (ext_imm_rotate);
 
 #undef AARCH64_DECL_OPD_EXTRACTOR
 
diff --git a/opcodes/aarch64-opc.c b/opcodes/aarch64-opc.c
index c85abc6..5b9eb27 100644
--- a/opcodes/aarch64-opc.c
+++ b/opcodes/aarch64-opc.c
@@ -309,7 +309,10 @@  const aarch64_field fields[] =
     {  8,  2 }, /* SVE_tszl_8: triangular size select low, bits [9,8].  */
     { 19,  2 }, /* SVE_tszl_19: triangular size select low, bits [20,19].  */
     { 14,  1 }, /* SVE_xs_14: UXTW/SXTW select (bit 14).  */
-    { 22,  1 }  /* SVE_xs_22: UXTW/SXTW select (bit 22).  */
+    { 22,  1 }, /* SVE_xs_22: UXTW/SXTW select (bit 22).  */
+    { 11,  2 }, /* rotate1: FCMLA immediate rotate.  */
+    { 13,  2 }, /* rotate2: Indexed element FCMLA immediate rotate.  */
+    { 12,  1 }, /* rotate3: FCADD immediate rotate.  */
 };
 
 enum aarch64_operand_class
@@ -2097,6 +2100,28 @@  operand_general_constraint_met_p (const aarch64_opnd_info *opnds, int idx,
 	    }
 	  break;
 
+	case AARCH64_OPND_IMM_ROT1:
+	case AARCH64_OPND_IMM_ROT2:
+	  if (opnd->imm.value != 0
+	      && opnd->imm.value != 90
+	      && opnd->imm.value != 180
+	      && opnd->imm.value != 270)
+	    {
+	      set_other_error (mismatch_detail, idx,
+			       _("rotate expected to be 0, 90, 180 or 270"));
+	      return 0;
+	    }
+	  break;
+
+	case AARCH64_OPND_IMM_ROT3:
+	  if (opnd->imm.value != 90 && opnd->imm.value != 270)
+	    {
+	      set_other_error (mismatch_detail, idx,
+			       _("rotate expected to be 90 or 270"));
+	      return 0;
+	    }
+	  break;
+
 	case AARCH64_OPND_SHLL_IMM:
 	  assert (idx == 2);
 	  size = 8 * aarch64_get_qualifier_esize (opnds[idx - 1].qualifier);
@@ -2436,7 +2461,15 @@  operand_general_constraint_met_p (const aarch64_opnd_info *opnds, int idx,
 
     case AARCH64_OPND_CLASS_SIMD_ELEMENT:
       /* Get the upper bound for the element index.  */
-      num = 16 / aarch64_get_qualifier_esize (qualifier) - 1;
+      if (opcode->op == OP_FCMLA_ELEM)
+	/* FCMLA index range depends on the vector size of other operands
+	   and is halfed because complex numbers take two elements.  */
+	num = aarch64_get_qualifier_nelem (opnds[0].qualifier)
+	      * aarch64_get_qualifier_esize (opnds[0].qualifier) / 2;
+      else
+	num = 16;
+      num = num / aarch64_get_qualifier_esize (qualifier) - 1;
+
       /* Index out-of-range.  */
       if (!value_in_range_p (opnd->reglane.index, 0, num))
 	{
@@ -3185,6 +3218,9 @@  aarch64_print_operand (char *buf, size_t size, bfd_vma pc,
     case AARCH64_OPND_SVE_UIMM7:
     case AARCH64_OPND_SVE_UIMM8:
     case AARCH64_OPND_SVE_UIMM8_53:
+    case AARCH64_OPND_IMM_ROT1:
+    case AARCH64_OPND_IMM_ROT2:
+    case AARCH64_OPND_IMM_ROT3:
       snprintf (buf, size, "#%" PRIi64, opnd->imm.value);
       break;
 
diff --git a/opcodes/aarch64-opc.h b/opcodes/aarch64-opc.h
index 7bb895a..fe14d87 100644
--- a/opcodes/aarch64-opc.h
+++ b/opcodes/aarch64-opc.h
@@ -137,6 +137,9 @@  enum aarch64_field_kind
   FLD_SVE_tszl_19,
   FLD_SVE_xs_14,
   FLD_SVE_xs_22,
+  FLD_rotate1,
+  FLD_rotate2,
+  FLD_rotate3,
 };
 
 /* Field description.  */
diff --git a/opcodes/aarch64-tbl.h b/opcodes/aarch64-tbl.h
index 6960a56..0efd98e 100644
--- a/opcodes/aarch64-tbl.h
+++ b/opcodes/aarch64-tbl.h
@@ -969,6 +969,16 @@ 
   QLF3(V_2D , V_2D , V_2D )	\
 }
 
+/* e.g. FCMLA <Vd>.<T>, <Vn>.<T>, <Vm>.<T>, #<rotate>.  */
+#define QL_V3SAMEHSD_ROT	\
+{				\
+  QLF4 (V_4H, V_4H, V_4H, NIL),	\
+  QLF4 (V_8H, V_8H, V_8H, NIL),	\
+  QLF4 (V_2S, V_2S, V_2S, NIL),	\
+  QLF4 (V_4S, V_4S, V_4S, NIL),	\
+  QLF4 (V_2D, V_2D, V_2D, NIL),	\
+}
+
 /* e.g. FMAXNM <Vd>.<T>, <Vn>.<T>, <Vm>.<T>.  */
 #define QL_V3SAMEH		\
 {				\
@@ -1308,6 +1318,14 @@ 
   QLF3 (V_8H, V_8H, S_H),	\
 }
 
+/* e.g. FCMLA <Vd>.<T>, <Vn>.<T>, <Vm>.<Ts>[<index>], #<rotate>.  */
+#define QL_ELEMENT_ROT		\
+{				\
+  QLF4 (V_4H, V_4H, S_H, NIL),	\
+  QLF4 (V_8H, V_8H, S_H, NIL),	\
+  QLF4 (V_4S, V_4S, S_S, NIL),	\
+}
+
 /* e.g. MOVI <Vd>.4S, #<imm8> {, LSL #<amount>}.  */
 #define QL_SIMD_IMM_S0W		\
 {				\
@@ -1920,6 +1938,8 @@  static const aarch64_feature_set aarch64_feature_v8_3 =
   AARCH64_FEATURE (AARCH64_FEATURE_V8_3, 0);
 static const aarch64_feature_set aarch64_feature_fp_v8_3 =
   AARCH64_FEATURE (AARCH64_FEATURE_V8_3 | AARCH64_FEATURE_FP, 0);
+static const aarch64_feature_set aarch64_feature_simd_v8_3 =
+  AARCH64_FEATURE (AARCH64_FEATURE_V8_3 | AARCH64_FEATURE_SIMD, 0);
 
 #define CORE		&aarch64_feature_v8
 #define FP		&aarch64_feature_fp
@@ -1937,6 +1957,7 @@  static const aarch64_feature_set aarch64_feature_fp_v8_3 =
 #define SVE		&aarch64_feature_sve
 #define ARMV8_3		&aarch64_feature_v8_3
 #define FP_V8_3		&aarch64_feature_fp_v8_3
+#define SIMD_V8_3	&aarch64_feature_simd_v8_3
 
 #define CORE_INSN(NAME,OPCODE,MASK,CLASS,OP,OPS,QUALS,FLAGS) \
   { NAME, OPCODE, MASK, CLASS, OP, CORE, OPS, QUALS, FLAGS, 0, NULL }
@@ -2104,6 +2125,7 @@  struct aarch64_opcode aarch64_opcode_table[] =
   SF16_INSN ("fmulx",   0x2f009000, 0xbfc0f400, asimdelem, OP3 (Vd, Vn, Em), QL_ELEMENT_FP_H, F_SIZEQ),
   RDMA_INSN ("sqrdmlah",0x2f00d000, 0xbf00f400, asimdelem, OP3 (Vd, Vn, Em), QL_ELEMENT,      F_SIZEQ),
   RDMA_INSN ("sqrdmlsh",0x2f00f000, 0xbf00f400, asimdelem, OP3 (Vd, Vn, Em), QL_ELEMENT,      F_SIZEQ),
+  {"fcmla", 0x2f001000, 0xbf009400, asimdelem, OP_FCMLA_ELEM, SIMD_V8_3, OP4 (Vd, Vn, Em, IMM_ROT2), QL_ELEMENT_ROT, F_SIZEQ, 0, NULL},
   /* AdvSIMD EXT.  */
   SIMD_INSN ("ext",  0x2e000000, 0xbfe08400, asimdext, 0, OP4 (Vd, Vn, Vm, IDX), QL_VEXT, F_SIZEQ),
   /* AdvSIMD modified immediate.  */
@@ -2347,6 +2369,8 @@  struct aarch64_opcode aarch64_opcode_table[] =
   /* AdvSIMD three same extension.  */
   RDMA_INSN ("sqrdmlah",0x2e008400, 0xbf20fe00, asimdsame, OP3 (Vd, Vn, Vm), QL_V3SAMEHS, F_SIZEQ),
   RDMA_INSN ("sqrdmlsh",0x2e008c00, 0xbf20fe00, asimdsame, OP3 (Vd, Vn, Vm), QL_V3SAMEHS, F_SIZEQ),
+  {"fcmla", 0x2e00c400, 0xbf20e400, asimdsame, 0, SIMD_V8_3, OP4 (Vd, Vn, Vm, IMM_ROT1), QL_V3SAMEHSD_ROT, F_SIZEQ, 0, NULL},
+  {"fcadd", 0x2e00e400, 0xbf20ec00, asimdsame, 0, SIMD_V8_3, OP4 (Vd, Vn, Vm, IMM_ROT3), QL_V3SAMEHSD_ROT, F_SIZEQ, 0, NULL},
   /* AdvSIMD shift by immediate.  */
   SIMD_INSN ("sshr", 0xf000400, 0xbf80fc00, asimdshf, 0, OP3 (Vd, Vn, IMM_VLSR), QL_VSHIFT, 0),
   SIMD_INSN ("ssra", 0xf001400, 0xbf80fc00, asimdshf, 0, OP3 (Vd, Vn, IMM_VLSR), QL_VSHIFT, 0),
@@ -4116,6 +4140,12 @@  struct aarch64_opcode aarch64_opcode_table[] =
     Y(IMMEDIATE, fbits, "FBITS", 0, F(FLD_scale),			\
       "the number of bits after the binary point in the fixed-point value")\
     X(IMMEDIATE, 0, 0, "IMM_MOV", 0, F(), "an immediate")		\
+    Y(IMMEDIATE, imm_rotate, "IMM_ROT1", 0, F(FLD_rotate1),		\
+      "a 2-bit rotation specifier for complex arithmetic operations")	\
+    Y(IMMEDIATE, imm_rotate, "IMM_ROT2", 0, F(FLD_rotate2),		\
+      "a 2-bit rotation specifier for complex arithmetic operations")	\
+    Y(IMMEDIATE, imm_rotate, "IMM_ROT3", 0, F(FLD_rotate3),		\
+      "a 1-bit rotation specifier for complex arithmetic operations")	\
     Y(COND, cond, "COND", 0, F(), "a condition")			\
     Y(COND, cond, "COND1", 0, F(),					\
       "one of the standard conditions, excluding AL and NV.")		\