;; ARM NEON coprocessor Machine Description ;; Copyright (C) 2006-2015 Free Software Foundation, Inc. ;; Written by CodeSourcery. ;; ;; This file is part of GCC. ;; ;; GCC is free software; you can redistribute it and/or modify it ;; under the terms of the GNU General Public License as published by ;; the Free Software Foundation; either version 3, or (at your option) ;; any later version. ;; ;; GCC is distributed in the hope that it will be useful, but ;; WITHOUT ANY WARRANTY; without even the implied warranty of ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU ;; General Public License for more details. ;; ;; You should have received a copy of the GNU General Public License ;; along with GCC; see the file COPYING3. If not see ;; http://www.gnu.org/licenses/.
;; Attribute used to permit string comparisons against <VQH_mnem> in ;; type attribute definitions. (define_attr “vqh_mnem” “vadd,vmin,vmax” (const_string “vadd”))
(define_insn “*neon_mov” [(set (match_operand:VDX 0 “nonimmediate_operand” “=w,Un,w, w, ?r,?w,?r,?r, ?Us”) (match_operand:VDX 1 “general_operand” " w,w, Dn,Uni, w, r, r, Usi,r"))] “TARGET_NEON && (register_operand (operands[0], mode) || register_operand (operands[1], mode))” { if (which_alternative == 2) { int width, is_valid; static char templ[40];
is_valid = neon_immediate_valid_for_move (operands[1], <MODE>mode, &operands[1], &width); gcc_assert (is_valid != 0); if (width == 0) return "vmov.f32\t%P0, %1 @ <mode>"; else sprintf (templ, "vmov.i%d\t%%P0, %%x1 @ <mode>", width); return templ; }
switch (which_alternative) { case 0: return “vmov\t%P0, %P1 @ ”; case 1: case 3: return output_move_neon (operands); case 2: gcc_unreachable (); case 4: return “vmov\t%Q0, %R0, %P1 @ ”; case 5: return “vmov\t%P0, %Q1, %R1 @ ”; default: return output_move_double (operands, true, NULL); } } [(set_attr “type” “neon_move,neon_store1_1reg,neon_move,
neon_load1_1reg, neon_to_gp,neon_from_gp,mov_reg,
neon_load1_2reg, neon_store1_2reg”) (set_attr “length” “4,4,4,4,4,4,8,8,8”) (set_attr “arm_pool_range” “,,,1020,,,,1020,*”) (set_attr “thumb2_pool_range” “,,,1018,,,,1018,*”) (set_attr “neg_pool_range” “,,,1004,,,,1004,*”)])
(define_insn “*neon_mov” [(set (match_operand:VQXMOV 0 “nonimmediate_operand” “=w,Un,w, w, ?r,?w,?r,?r, ?Us”) (match_operand:VQXMOV 1 “general_operand” " w,w, Dn,Uni, w, r, r, Usi, r"))] “TARGET_NEON && (register_operand (operands[0], mode) || register_operand (operands[1], mode))” { if (which_alternative == 2) { int width, is_valid; static char templ[40];
is_valid = neon_immediate_valid_for_move (operands[1], <MODE>mode, &operands[1], &width); gcc_assert (is_valid != 0); if (width == 0) return "vmov.f32\t%q0, %1 @ <mode>"; else sprintf (templ, "vmov.i%d\t%%q0, %%1 @ <mode>", width); return templ; }
switch (which_alternative) { case 0: return “vmov\t%q0, %q1 @ ”; case 1: case 3: return output_move_neon (operands); case 2: gcc_unreachable (); case 4: return “vmov\t%Q0, %R0, %e1 @ ;vmov\t%J0, %K0, %f1”; case 5: return “vmov\t%e0, %Q1, %R1 @ ;vmov\t%f0, %J1, %K1”; default: return output_move_quad (operands); } } [(set_attr “type” “neon_move_q,neon_store2_2reg_q,neon_move_q,
neon_load2_2reg_q,neon_to_gp_q,neon_from_gp_q,
mov_reg,neon_load1_4reg,neon_store1_4reg”) (set_attr “length” “4,8,4,8,8,8,16,8,16”) (set_attr “arm_pool_range” “,,,1020,,,,1020,*”) (set_attr “thumb2_pool_range” “,,,1018,,,,1018,*”) (set_attr “neg_pool_range” “,,,996,,,,996,*”)])
(define_expand “movti” [(set (match_operand:TI 0 “nonimmediate_operand” "") (match_operand:TI 1 “general_operand” ""))] “TARGET_NEON” { if (can_create_pseudo_p ()) { if (!REG_P (operands[0])) operands[1] = force_reg (TImode, operands[1]); } })
(define_expand “mov” [(set (match_operand:VSTRUCT 0 “nonimmediate_operand” "") (match_operand:VSTRUCT 1 “general_operand” ""))] “TARGET_NEON” { if (can_create_pseudo_p ()) { if (!REG_P (operands[0])) operands[1] = force_reg (mode, operands[1]); } })
(define_insn “*neon_mov” [(set (match_operand:VSTRUCT 0 “nonimmediate_operand” “=w,Ut,w”) (match_operand:VSTRUCT 1 “general_operand” " w,w, Ut"))] “TARGET_NEON && (register_operand (operands[0], mode) || register_operand (operands[1], mode))” { switch (which_alternative) { case 0: return “#”; case 1: case 2: return output_move_neon (operands); default: gcc_unreachable (); } } [(set_attr “type” “neon_move_q,neon_store2_2reg_q,neon_load2_2reg_q”) (set (attr “length”) (symbol_ref “arm_attr_length_move_neon (insn)”))])
(define_split [(set (match_operand:EI 0 “s_register_operand” "") (match_operand:EI 1 “s_register_operand” ""))] “TARGET_NEON && reload_completed” [(set (match_dup 0) (match_dup 1)) (set (match_dup 2) (match_dup 3))] { int rdest = REGNO (operands[0]); int rsrc = REGNO (operands[1]); rtx dest[2], src[2];
dest[0] = gen_rtx_REG (TImode, rdest); src[0] = gen_rtx_REG (TImode, rsrc); dest[1] = gen_rtx_REG (DImode, rdest + 4); src[1] = gen_rtx_REG (DImode, rsrc + 4);
neon_disambiguate_copy (operands, dest, src, 2); })
(define_split [(set (match_operand:OI 0 “s_register_operand” "") (match_operand:OI 1 “s_register_operand” ""))] “TARGET_NEON && reload_completed” [(set (match_dup 0) (match_dup 1)) (set (match_dup 2) (match_dup 3))] { int rdest = REGNO (operands[0]); int rsrc = REGNO (operands[1]); rtx dest[2], src[2];
dest[0] = gen_rtx_REG (TImode, rdest); src[0] = gen_rtx_REG (TImode, rsrc); dest[1] = gen_rtx_REG (TImode, rdest + 4); src[1] = gen_rtx_REG (TImode, rsrc + 4);
neon_disambiguate_copy (operands, dest, src, 2); })
(define_split [(set (match_operand:CI 0 “s_register_operand” "") (match_operand:CI 1 “s_register_operand” ""))] “TARGET_NEON && reload_completed” [(set (match_dup 0) (match_dup 1)) (set (match_dup 2) (match_dup 3)) (set (match_dup 4) (match_dup 5))] { int rdest = REGNO (operands[0]); int rsrc = REGNO (operands[1]); rtx dest[3], src[3];
dest[0] = gen_rtx_REG (TImode, rdest); src[0] = gen_rtx_REG (TImode, rsrc); dest[1] = gen_rtx_REG (TImode, rdest + 4); src[1] = gen_rtx_REG (TImode, rsrc + 4); dest[2] = gen_rtx_REG (TImode, rdest + 8); src[2] = gen_rtx_REG (TImode, rsrc + 8);
neon_disambiguate_copy (operands, dest, src, 3); })
(define_split [(set (match_operand:XI 0 “s_register_operand” "") (match_operand:XI 1 “s_register_operand” ""))] “TARGET_NEON && reload_completed” [(set (match_dup 0) (match_dup 1)) (set (match_dup 2) (match_dup 3)) (set (match_dup 4) (match_dup 5)) (set (match_dup 6) (match_dup 7))] { int rdest = REGNO (operands[0]); int rsrc = REGNO (operands[1]); rtx dest[4], src[4];
dest[0] = gen_rtx_REG (TImode, rdest); src[0] = gen_rtx_REG (TImode, rsrc); dest[1] = gen_rtx_REG (TImode, rdest + 4); src[1] = gen_rtx_REG (TImode, rsrc + 4); dest[2] = gen_rtx_REG (TImode, rdest + 8); src[2] = gen_rtx_REG (TImode, rsrc + 8); dest[3] = gen_rtx_REG (TImode, rdest + 12); src[3] = gen_rtx_REG (TImode, rsrc + 12);
neon_disambiguate_copy (operands, dest, src, 4); })
(define_expand “movmisalign” [(set (match_operand:VDQX 0 “neon_perm_struct_or_reg_operand”) (unspec:VDQX [(match_operand:VDQX 1 “neon_perm_struct_or_reg_operand”)] UNSPEC_MISALIGNED_ACCESS))] “TARGET_NEON && !BYTES_BIG_ENDIAN && unaligned_access” { rtx adjust_mem; /* This pattern is not permitted to fail during expansion: if both arguments are non-registers (e.g. memory := constant, which can be created by the auto-vectorizer), force operand 1 into a register. */ if (!s_register_operand (operands[0], mode) && !s_register_operand (operands[1], mode)) operands[1] = force_reg (mode, operands[1]);
if (s_register_operand (operands[0], mode)) adjust_mem = operands[1]; else adjust_mem = operands[0];
/* Legitimize address. */ if (!neon_vector_mem_operand (adjust_mem, 2, true)) XEXP (adjust_mem, 0) = force_reg (Pmode, XEXP (adjust_mem, 0));
})
(define_insn “*movmisalign_neon_store” [(set (match_operand:VDX 0 “neon_permissive_struct_operand” “=Um”) (unspec:VDX [(match_operand:VDX 1 “s_register_operand” " w")] UNSPEC_MISALIGNED_ACCESS))] “TARGET_NEON && !BYTES_BIG_ENDIAN && unaligned_access” “vst1.<V_sz_elem>\t{%P1}, %A0” [(set_attr “type” “neon_store1_1reg”)])
(define_insn “*movmisalign_neon_load” [(set (match_operand:VDX 0 “s_register_operand” “=w”) (unspec:VDX [(match_operand:VDX 1 “neon_permissive_struct_operand” " Um")] UNSPEC_MISALIGNED_ACCESS))] “TARGET_NEON && !BYTES_BIG_ENDIAN && unaligned_access” “vld1.<V_sz_elem>\t{%P0}, %A1” [(set_attr “type” “neon_load1_1reg”)])
(define_insn “*movmisalign_neon_store” [(set (match_operand:VQX 0 “neon_permissive_struct_operand” “=Um”) (unspec:VQX [(match_operand:VQX 1 “s_register_operand” " w")] UNSPEC_MISALIGNED_ACCESS))] “TARGET_NEON && !BYTES_BIG_ENDIAN && unaligned_access” “vst1.<V_sz_elem>\t{%q1}, %A0” [(set_attr “type” “neon_store1_1reg”)])
(define_insn “*movmisalign_neon_load” [(set (match_operand:VQX 0 “s_register_operand” “=w”) (unspec:VQX [(match_operand:VQX 1 “neon_permissive_struct_operand” " Um")] UNSPEC_MISALIGNED_ACCESS))] “TARGET_NEON && !BYTES_BIG_ENDIAN && unaligned_access” “vld1.<V_sz_elem>\t{%q0}, %A1” [(set_attr “type” “neon_load1_1reg”)])
(define_insn “vec_set_internal” [(set (match_operand:VD 0 “s_register_operand” “=w,w”) (vec_merge:VD (vec_duplicate:VD (match_operand:<V_elem> 1 “nonimmediate_operand” “Um,r”)) (match_operand:VD 3 “s_register_operand” “0,0”) (match_operand:SI 2 “immediate_operand” “i,i”)))] “TARGET_NEON” { int elt = ffs ((int) INTVAL (operands[2])) - 1; if (BYTES_BIG_ENDIAN) elt = GET_MODE_NUNITS (mode) - 1 - elt; operands[2] = GEN_INT (elt);
if (which_alternative == 0) return “vld1.<V_sz_elem>\t{%P0[%c2]}, %A1”; else return “vmov.<V_sz_elem>\t%P0[%c2], %1”; } [(set_attr “type” “neon_load1_all_lanes,neon_from_gp”)])
(define_insn “vec_set_internal” [(set (match_operand:VQ 0 “s_register_operand” “=w,w”) (vec_merge:VQ (vec_duplicate:VQ (match_operand:<V_elem> 1 “nonimmediate_operand” “Um,r”)) (match_operand:VQ 3 “s_register_operand” “0,0”) (match_operand:SI 2 “immediate_operand” “i,i”)))] “TARGET_NEON” { HOST_WIDE_INT elem = ffs ((int) INTVAL (operands[2])) - 1; int half_elts = GET_MODE_NUNITS (mode) / 2; int elt = elem % half_elts; int hi = (elem / half_elts) * 2; int regno = REGNO (operands[0]);
if (BYTES_BIG_ENDIAN) elt = half_elts - 1 - elt;
operands[0] = gen_rtx_REG (<V_HALF>mode, regno + hi); operands[2] = GEN_INT (elt);
if (which_alternative == 0) return “vld1.<V_sz_elem>\t{%P0[%c2]}, %A1”; else return “vmov.<V_sz_elem>\t%P0[%c2], %1”; } [(set_attr “type” “neon_load1_all_lanes,neon_from_gp”)] )
(define_insn “vec_setv2di_internal” [(set (match_operand:V2DI 0 “s_register_operand” “=w,w”) (vec_merge:V2DI (vec_duplicate:V2DI (match_operand:DI 1 “nonimmediate_operand” “Um,r”)) (match_operand:V2DI 3 “s_register_operand” “0,0”) (match_operand:SI 2 “immediate_operand” “i,i”)))] “TARGET_NEON” { HOST_WIDE_INT elem = ffs ((int) INTVAL (operands[2])) - 1; int regno = REGNO (operands[0]) + 2 * elem;
operands[0] = gen_rtx_REG (DImode, regno);
if (which_alternative == 0) return “vld1.64\t%P0, %A1”; else return “vmov\t%P0, %Q1, %R1”; } [(set_attr “type” “neon_load1_all_lanes_q,neon_from_gp_q”)] )
(define_expand “vec_set” [(match_operand:VDQ 0 “s_register_operand” "") (match_operand:<V_elem> 1 “s_register_operand” "") (match_operand:SI 2 “immediate_operand” "")] “TARGET_NEON” { HOST_WIDE_INT elem = (HOST_WIDE_INT) 1 << INTVAL (operands[2]); emit_insn (gen_vec_set_internal (operands[0], operands[1], GEN_INT (elem), operands[0])); DONE; })
(define_insn “vec_extract” [(set (match_operand:<V_elem> 0 “nonimmediate_operand” “=Um,r”) (vec_select:<V_elem> (match_operand:VD 1 “s_register_operand” “w,w”) (parallel [(match_operand:SI 2 “immediate_operand” “i,i”)])))] “TARGET_NEON” { if (BYTES_BIG_ENDIAN) { int elt = INTVAL (operands[2]); elt = GET_MODE_NUNITS (mode) - 1 - elt; operands[2] = GEN_INT (elt); }
if (which_alternative == 0) return “vst1.<V_sz_elem>\t{%P1[%c2]}, %A0”; else return “vmov.<V_uf_sclr>\t%0, %P1[%c2]”; } [(set_attr “type” “neon_store1_one_lane,neon_to_gp”)] )
(define_insn “vec_extract” [(set (match_operand:<V_elem> 0 “nonimmediate_operand” “=Um,r”) (vec_select:<V_elem> (match_operand:VQ 1 “s_register_operand” “w,w”) (parallel [(match_operand:SI 2 “immediate_operand” “i,i”)])))] “TARGET_NEON” { int half_elts = GET_MODE_NUNITS (mode) / 2; int elt = INTVAL (operands[2]) % half_elts; int hi = (INTVAL (operands[2]) / half_elts) * 2; int regno = REGNO (operands[1]);
if (BYTES_BIG_ENDIAN) elt = half_elts - 1 - elt;
operands[1] = gen_rtx_REG (<V_HALF>mode, regno + hi); operands[2] = GEN_INT (elt);
if (which_alternative == 0) return “vst1.<V_sz_elem>\t{%P1[%c2]}, %A0”; else return “vmov.<V_uf_sclr>\t%0, %P1[%c2]”; } [(set_attr “type” “neon_store1_one_lane,neon_to_gp”)] )
(define_insn “vec_extractv2di” [(set (match_operand:DI 0 “nonimmediate_operand” “=Um,r”) (vec_select:DI (match_operand:V2DI 1 “s_register_operand” “w,w”) (parallel [(match_operand:SI 2 “immediate_operand” “i,i”)])))] “TARGET_NEON” { int regno = REGNO (operands[1]) + 2 * INTVAL (operands[2]);
operands[1] = gen_rtx_REG (DImode, regno);
if (which_alternative == 0) return “vst1.64\t{%P1}, %A0 @ v2di”; else return “vmov\t%Q0, %R0, %P1 @ v2di”; } [(set_attr “type” “neon_store1_one_lane_q,neon_to_gp_q”)] )
(define_expand “vec_init” [(match_operand:VDQ 0 “s_register_operand” "") (match_operand 1 "" "")] “TARGET_NEON” { neon_expand_vector_init (operands[0], operands[1]); DONE; })
;; Doubleword and quadword arithmetic.
;; NOTE: some other instructions also support 64-bit integer ;; element size, which we could potentially use for “long long” operations.
(define_insn “*add3_neon” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (plus:VDQ (match_operand:VDQ 1 “s_register_operand” “w”) (match_operand:VDQ 2 “s_register_operand” “w”)))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” “vadd.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_addsub_s”) (const_string “neon_add”)))] )
(define_insn “adddi3_neon” [(set (match_operand:DI 0 “s_register_operand” “=w,?&r,?&r,?w,?&r,?&r,?&r”) (plus:DI (match_operand:DI 1 “s_register_operand” “%w,0,0,w,r,0,r”) (match_operand:DI 2 “arm_adddi_operand” “w,r,0,w,r,Dd,Dd”))) (clobber (reg:CC CC_REGNUM))] “TARGET_NEON” { switch (which_alternative) { case 0: /* fall through / case 3: return “vadd.i64\t%P0, %P1, %P2”; case 1: return “#”; case 2: return “#”; case 4: return “#”; case 5: return “#”; case 6: return “#”; default: gcc_unreachable (); } } [(set_attr “type” “neon_add,multiple,multiple,neon_add,
multiple,multiple,multiple”) (set_attr “conds” ",clob,clob,,clob,clob,clob") (set_attr “length” ",8,8,,8,8,8") (set_attr “arch” "neon_for_64bits,,,avoid_neon_for_64bits,,,")] )
(define_insn “*sub3_neon” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (minus:VDQ (match_operand:VDQ 1 “s_register_operand” “w”) (match_operand:VDQ 2 “s_register_operand” “w”)))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” “vsub.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_addsub_s”) (const_string “neon_sub”)))] )
(define_insn “subdi3_neon” [(set (match_operand:DI 0 “s_register_operand” “=w,?&r,?&r,?&r,?w”) (minus:DI (match_operand:DI 1 “s_register_operand” “w,0,r,0,w”) (match_operand:DI 2 “s_register_operand” “w,r,0,0,w”))) (clobber (reg:CC CC_REGNUM))] “TARGET_NEON” { switch (which_alternative) { case 0: /* fall through / case 4: return “vsub.i64\t%P0, %P1, %P2”; case 1: / fall through / case 2: / fall through / case 3: return “subs\t%Q0, %Q1, %Q2;sbc\t%R0, %R1, %R2”; default: gcc_unreachable (); } } [(set_attr “type” “neon_sub,multiple,multiple,multiple,neon_sub”) (set_attr “conds” ",clob,clob,clob,") (set_attr “length” ",8,8,8,") (set_attr “arch” "neon_for_64bits,,,,avoid_neon_for_64bits")] )
(define_insn “*mul3_neon” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (mult:VDQW (match_operand:VDQW 1 “s_register_operand” “w”) (match_operand:VDQW 2 “s_register_operand” “w”)))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” “vmul.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mul_s”) (const_string “neon_mul_<V_elem_ch>”)))] )
(define_insn “mul3add_neon” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (plus:VDQW (mult:VDQW (match_operand:VDQW 2 “s_register_operand” “w”) (match_operand:VDQW 3 “s_register_operand” “w”)) (match_operand:VDQW 1 “s_register_operand” “0”)))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” “vmla.<V_if_elem>\t%<V_reg>0, %<V_reg>2, %<V_reg>3” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s”) (const_string “neon_mla_<V_elem_ch>”)))] )
(define_insn “mul3negadd_neon” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (minus:VDQW (match_operand:VDQW 1 “s_register_operand” “0”) (mult:VDQW (match_operand:VDQW 2 “s_register_operand” “w”) (match_operand:VDQW 3 “s_register_operand” “w”))))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” “vmls.<V_if_elem>\t%<V_reg>0, %<V_reg>2, %<V_reg>3” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s”) (const_string “neon_mla_<V_elem_ch>”)))] )
;; Fused multiply-accumulate ;; We define each insn twice here: ;; 1: with flag_unsafe_math_optimizations for the widening multiply phase ;; to be able to use when converting to FMA. ;; 2: without flag_unsafe_math_optimizations for the intrinsics to use. (define_insn “fmaVCVTF:mode4” [(set (match_operand:VCVTF 0 “register_operand” “=w”) (fma:VCVTF (match_operand:VCVTF 1 “register_operand” “w”) (match_operand:VCVTF 2 “register_operand” “w”) (match_operand:VCVTF 3 “register_operand” “0”)))] “TARGET_NEON && TARGET_FMA && flag_unsafe_math_optimizations” “vfma%?.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_mla_s”)] )
(define_insn “fmaVCVTF:mode4_intrinsic” [(set (match_operand:VCVTF 0 “register_operand” “=w”) (fma:VCVTF (match_operand:VCVTF 1 “register_operand” “w”) (match_operand:VCVTF 2 “register_operand” “w”) (match_operand:VCVTF 3 “register_operand” “0”)))] “TARGET_NEON && TARGET_FMA” “vfma%?.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_mla_s”)] )
(define_insn “*fmsubVCVTF:mode4” [(set (match_operand:VCVTF 0 “register_operand” “=w”) (fma:VCVTF (neg:VCVTF (match_operand:VCVTF 1 “register_operand” “w”)) (match_operand:VCVTF 2 “register_operand” “w”) (match_operand:VCVTF 3 “register_operand” “0”)))] “TARGET_NEON && TARGET_FMA && flag_unsafe_math_optimizations” “vfms%?.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_mla_s”)] )
(define_insn “fmsubVCVTF:mode4_intrinsic” [(set (match_operand:VCVTF 0 “register_operand” “=w”) (fma:VCVTF (neg:VCVTF (match_operand:VCVTF 1 “register_operand” “w”)) (match_operand:VCVTF 2 “register_operand” “w”) (match_operand:VCVTF 3 “register_operand” “0”)))] “TARGET_NEON && TARGET_FMA” “vfms%?.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_mla_s”)] )
(define_insn “neon_vrint<NEON_VRINT:nvrint_variant>VCVTF:mode” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”)] NEON_VRINT))] “TARGET_NEON && TARGET_FPU_ARMV8” “vrint<nvrint_variant>%?.f32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_fp_round_<V_elem_ch>”)] )
(define_insn “neon_vcvt<NEON_VCVT:nvrint_variant><su_optab>VCVTF:mode<v_cmp_result>” [(set (match_operand:<V_cmp_result> 0 “register_operand” “=w”) (FIXUORS:<V_cmp_result> (unspec:VCVTF [(match_operand:VCVTF 1 “register_operand” “w”)] NEON_VCVT)))] “TARGET_NEON && TARGET_FPU_ARMV8” “vcvt<nvrint_variant>.32.f32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_fp_to_int_<V_elem_ch>”) (set_attr “predicable” “no”)] )
(define_insn “ior3” [(set (match_operand:VDQ 0 “s_register_operand” “=w,w”) (ior:VDQ (match_operand:VDQ 1 “s_register_operand” “w,0”) (match_operand:VDQ 2 “neon_logic_op2” “w,Dl”)))] “TARGET_NEON” { switch (which_alternative) { case 0: return “vorr\t%<V_reg>0, %<V_reg>1, %<V_reg>2”; case 1: return neon_output_logic_immediate (“vorr”, &operands[2], mode, 0, VALID_NEON_QREG_MODE (mode)); default: gcc_unreachable (); } } [(set_attr “type” “neon_logic”)] )
;; The concrete forms of the Neon immediate-logic instructions are vbic and ;; vorr. We support the pseudo-instruction vand instead, because that ;; corresponds to the canonical form the middle-end expects to use for ;; immediate bitwise-ANDs.
(define_insn “and3” [(set (match_operand:VDQ 0 “s_register_operand” “=w,w”) (and:VDQ (match_operand:VDQ 1 “s_register_operand” “w,0”) (match_operand:VDQ 2 “neon_inv_logic_op2” “w,DL”)))] “TARGET_NEON” { switch (which_alternative) { case 0: return “vand\t%<V_reg>0, %<V_reg>1, %<V_reg>2”; case 1: return neon_output_logic_immediate (“vand”, &operands[2], mode, 1, VALID_NEON_QREG_MODE (mode)); default: gcc_unreachable (); } } [(set_attr “type” “neon_logic”)] )
(define_insn “orn3_neon” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (ior:VDQ (not:VDQ (match_operand:VDQ 2 “s_register_operand” “w”)) (match_operand:VDQ 1 “s_register_operand” “w”)))] “TARGET_NEON” “vorn\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_logic”)] )
;; TODO: investigate whether we should disable ;; this and bicdi3_neon for the A8 in line with the other ;; changes above. (define_insn_and_split “orndi3_neon” [(set (match_operand:DI 0 “s_register_operand” “=w,?&r,?&r,?&r”) (ior:DI (not:DI (match_operand:DI 2 “s_register_operand” “w,0,0,r”)) (match_operand:DI 1 “s_register_operand” “w,r,r,0”)))] “TARGET_NEON” "@ vorn\t%P0, %P1, %P2
#" “reload_completed && (TARGET_NEON && !(IS_VFP_REGNUM (REGNO (operands[0]))))” [(set (match_dup 0) (ior:SI (not:SI (match_dup 2)) (match_dup 1))) (set (match_dup 3) (ior:SI (not:SI (match_dup 4)) (match_dup 5)))] " { if (TARGET_THUMB2) { operands[3] = gen_highpart (SImode, operands[0]); operands[0] = gen_lowpart (SImode, operands[0]); operands[4] = gen_highpart (SImode, operands[2]); operands[2] = gen_lowpart (SImode, operands[2]); operands[5] = gen_highpart (SImode, operands[1]); operands[1] = gen_lowpart (SImode, operands[1]); } else { emit_insn (gen_one_cmpldi2 (operands[0], operands[2])); emit_insn (gen_iordi3 (operands[0], operands[1], operands[0])); DONE; } }" [(set_attr “type” “neon_logic,multiple,multiple,multiple”) (set_attr “length” “*,16,8,8”) (set_attr “arch” “any,a,t2,t2”)] )
(define_insn “bic3_neon” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (and:VDQ (not:VDQ (match_operand:VDQ 2 “s_register_operand” “w”)) (match_operand:VDQ 1 “s_register_operand” “w”)))] “TARGET_NEON” “vbic\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_logic”)] )
;; Compare to *anddi_notdi_di. (define_insn “bicdi3_neon” [(set (match_operand:DI 0 “s_register_operand” “=w,?&r,?&r”) (and:DI (not:DI (match_operand:DI 2 “s_register_operand” “w,r,0”)) (match_operand:DI 1 “s_register_operand” “w,0,r”)))] “TARGET_NEON” "@ vbic\t%P0, %P1, %P2
#" [(set_attr “type” “neon_logic,multiple,multiple”) (set_attr “length” “*,8,8”)] )
(define_insn “xor3” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (xor:VDQ (match_operand:VDQ 1 “s_register_operand” “w”) (match_operand:VDQ 2 “s_register_operand” “w”)))] “TARGET_NEON” “veor\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_logic”)] )
(define_insn “one_cmpl2” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (not:VDQ (match_operand:VDQ 1 “s_register_operand” “w”)))] “TARGET_NEON” “vmvn\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_move”)] )
(define_insn “abs2” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (abs:VDQW (match_operand:VDQW 1 “s_register_operand” “w”)))] “TARGET_NEON” “vabs.<V_s_elem>\t%<V_reg>0, %<V_reg>1” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_abs_s”) (const_string “neon_abs”)))] )
(define_insn “neg2” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (neg:VDQW (match_operand:VDQW 1 “s_register_operand” “w”)))] “TARGET_NEON” “vneg.<V_s_elem>\t%<V_reg>0, %<V_reg>1” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_neg_s”) (const_string “neon_neg”)))] )
(define_insn “negdi2_neon” [(set (match_operand:DI 0 “s_register_operand” “=&w, w,r,&r”) (neg:DI (match_operand:DI 1 “s_register_operand” " w, w,0, r"))) (clobber (match_scratch:DI 2 “= X,&w,X, X”)) (clobber (reg:CC CC_REGNUM))] “TARGET_NEON” “#” [(set_attr “length” “8”) (set_attr “type” “multiple”)] )
; Split negdi2_neon for vfp registers (define_split [(set (match_operand:DI 0 “s_register_operand” "") (neg:DI (match_operand:DI 1 “s_register_operand” ""))) (clobber (match_scratch:DI 2 "")) (clobber (reg:CC CC_REGNUM))] “TARGET_NEON && reload_completed && IS_VFP_REGNUM (REGNO (operands[0]))” [(set (match_dup 2) (const_int 0)) (parallel [(set (match_dup 0) (minus:DI (match_dup 2) (match_dup 1))) (clobber (reg:CC CC_REGNUM))])] { if (!REG_P (operands[2])) operands[2] = operands[0]; } )
; Split negdi2_neon for core registers (define_split [(set (match_operand:DI 0 “s_register_operand” "") (neg:DI (match_operand:DI 1 “s_register_operand” ""))) (clobber (match_scratch:DI 2 "")) (clobber (reg:CC CC_REGNUM))] “TARGET_32BIT && reload_completed && arm_general_register_operand (operands[0], DImode)” [(parallel [(set (match_dup 0) (neg:DI (match_dup 1))) (clobber (reg:CC CC_REGNUM))])] "" )
(define_insn “*umin3_neon” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (umin:VDQIW (match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)))] “TARGET_NEON” “vmin.<V_u_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_minmax”)] )
(define_insn “*umax3_neon” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (umax:VDQIW (match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)))] “TARGET_NEON” “vmax.<V_u_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_minmax”)] )
(define_insn “*smin3_neon” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (smin:VDQW (match_operand:VDQW 1 “s_register_operand” “w”) (match_operand:VDQW 2 “s_register_operand” “w”)))] “TARGET_NEON” “vmin.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_minmax_s”) (const_string “neon_minmax”)))] )
(define_insn “*smax3_neon” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (smax:VDQW (match_operand:VDQW 1 “s_register_operand” “w”) (match_operand:VDQW 2 “s_register_operand” “w”)))] “TARGET_NEON” “vmax.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_minmax_s”) (const_string “neon_minmax”)))] )
; TODO: V2DI shifts are current disabled because there are bugs in the ; generic vectorizer code. It ends up creating a V2DI constructor with ; SImode elements.
(define_insn “vashl3” [(set (match_operand:VDQIW 0 “s_register_operand” “=w,w”) (ashift:VDQIW (match_operand:VDQIW 1 “s_register_operand” “w,w”) (match_operand:VDQIW 2 “imm_lshift_or_reg_neon” “w,Dn”)))] “TARGET_NEON” { switch (which_alternative) { case 0: return “vshl.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2”; case 1: return neon_output_shift_immediate (“vshl”, ‘i’, &operands[2], mode, VALID_NEON_QREG_MODE (mode), true); default: gcc_unreachable (); } } [(set_attr “type” “neon_shift_reg, neon_shift_imm”)] )
(define_insn “vashr3_imm” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (ashiftrt:VDQIW (match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “imm_for_neon_rshift_operand” “Dn”)))] “TARGET_NEON” { return neon_output_shift_immediate (“vshr”, ‘s’, &operands[2], mode, VALID_NEON_QREG_MODE (mode), false); } [(set_attr “type” “neon_shift_imm”)] )
(define_insn “vlshr3_imm” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (lshiftrt:VDQIW (match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “imm_for_neon_rshift_operand” “Dn”)))] “TARGET_NEON” { return neon_output_shift_immediate (“vshr”, ‘u’, &operands[2], mode, VALID_NEON_QREG_MODE (mode), false); }
[(set_attr “type” “neon_shift_imm”)]
)
; Used for implementing logical shift-right, which is a left-shift by a negative ; amount, with signed operands. This is essentially the same as ashl3 ; above, but using an unspec in case GCC tries anything tricky with negative ; shift amounts.
(define_insn “ashl3_signed” [(set (match_operand:VDQI 0 “s_register_operand” “=w”) (unspec:VDQI [(match_operand:VDQI 1 “s_register_operand” “w”) (match_operand:VDQI 2 “s_register_operand” “w”)] UNSPEC_ASHIFT_SIGNED))] “TARGET_NEON” “vshl.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_shift_reg”)] )
; Used for implementing logical shift-right, which is a left-shift by a negative ; amount, with unsigned operands.
(define_insn “ashl3_unsigned” [(set (match_operand:VDQI 0 “s_register_operand” “=w”) (unspec:VDQI [(match_operand:VDQI 1 “s_register_operand” “w”) (match_operand:VDQI 2 “s_register_operand” “w”)] UNSPEC_ASHIFT_UNSIGNED))] “TARGET_NEON” “vshl.<V_u_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_shift_reg”)] )
(define_expand “vashr3” [(set (match_operand:VDQIW 0 “s_register_operand” "") (ashiftrt:VDQIW (match_operand:VDQIW 1 “s_register_operand” "") (match_operand:VDQIW 2 “imm_rshift_or_reg_neon” "")))] “TARGET_NEON” { if (s_register_operand (operands[2], mode)) { rtx neg = gen_reg_rtx (mode); emit_insn (gen_neg2 (neg, operands[2])); emit_insn (gen_ashl3_signed (operands[0], operands[1], neg)); } else emit_insn (gen_vashr3_imm (operands[0], operands[1], operands[2])); DONE; })
(define_expand “vlshr3” [(set (match_operand:VDQIW 0 “s_register_operand” "") (lshiftrt:VDQIW (match_operand:VDQIW 1 “s_register_operand” "") (match_operand:VDQIW 2 “imm_rshift_or_reg_neon” "")))] “TARGET_NEON” { if (s_register_operand (operands[2], mode)) { rtx neg = gen_reg_rtx (mode); emit_insn (gen_neg2 (neg, operands[2])); emit_insn (gen_ashl3_unsigned (operands[0], operands[1], neg)); } else emit_insn (gen_vlshr3_imm (operands[0], operands[1], operands[2])); DONE; })
;; 64-bit shifts
;; This pattern loads a 32-bit shift count into a 64-bit NEON register, ;; leaving the upper half uninitalized. This is OK since the shift ;; instruction only looks at the low 8 bits anyway. To avoid confusing ;; data flow analysis however, we pretend the full register is set ;; using an unspec. (define_insn “neon_load_count” [(set (match_operand:DI 0 “s_register_operand” “=w,w”) (unspec:DI [(match_operand:SI 1 “nonimmediate_operand” “Um,r”)] UNSPEC_LOAD_COUNT))] “TARGET_NEON” “@ vld1.32\t{%P0[0]}, %A1 vmov.32\t%P0[0], %1” [(set_attr “type” “neon_load1_1reg,neon_from_gp”)] )
(define_insn “ashldi3_neon_noclobber” [(set (match_operand:DI 0 “s_register_operand” “=w,w”) (ashift:DI (match_operand:DI 1 “s_register_operand” " w,w") (match_operand:DI 2 “reg_or_int_operand” " i,w")))] “TARGET_NEON && reload_completed && (!CONST_INT_P (operands[2]) || (INTVAL (operands[2]) >= 0 && INTVAL (operands[2]) < 64))” “@ vshl.u64\t%P0, %P1, %2 vshl.u64\t%P0, %P1, %P2” [(set_attr “type” “neon_shift_imm, neon_shift_reg”)] )
(define_insn_and_split “ashldi3_neon” [(set (match_operand:DI 0 “s_register_operand” “= w, w,?&r,?r,?&r, ?w,w”) (ashift:DI (match_operand:DI 1 “s_register_operand” " 0w, w, 0r, 0, r, 0w,w") (match_operand:SI 2 “general_operand” “rUm, i, r, i, i,rUm,i”))) (clobber (match_scratch:SI 3 “= X, X,?&r, X, X, X,X”)) (clobber (match_scratch:SI 4 “= X, X,?&r, X, X, X,X”)) (clobber (match_scratch:DI 5 “=&w, X, X, X, X, &w,X”)) (clobber (reg:CC_C CC_REGNUM))] “TARGET_NEON” “#” “TARGET_NEON && reload_completed” [(const_int 0)] " { if (IS_VFP_REGNUM (REGNO (operands[0]))) { if (CONST_INT_P (operands[2])) { if (INTVAL (operands[2]) < 1) { emit_insn (gen_movdi (operands[0], operands[1])); DONE; } else if (INTVAL (operands[2]) > 63) operands[2] = gen_rtx_CONST_INT (VOIDmode, 63); } else { emit_insn (gen_neon_load_count (operands[5], operands[2])); operands[2] = operands[5]; }
/* Ditch the unnecessary clobbers. */ emit_insn (gen_ashldi3_neon_noclobber (operands[0], operands[1], operands[2])); } else { /* The shift expanders support either full overlap or no overlap. */ gcc_assert (!reg_overlap_mentioned_p (operands[0], operands[1]) || REGNO (operands[0]) == REGNO (operands[1])); if (operands[2] == CONST1_RTX (SImode)) /* This clobbers CC. */ emit_insn (gen_arm_ashldi3_1bit (operands[0], operands[1])); else arm_emit_coreregs_64bit_shift (ASHIFT, operands[0], operands[1], operands[2], operands[3], operands[4]); } DONE;
}" [(set_attr “arch” “neon_for_64bits,neon_for_64bits,,,*,avoid_neon_for_64bits,avoid_neon_for_64bits”) (set_attr “opt” “,,speed,speed,speed,,”) (set_attr “type” “multiple”)] )
; The shift amount needs to be negated for right-shifts (define_insn “signed_shift_di3_neon” [(set (match_operand:DI 0 “s_register_operand” “=w”) (unspec:DI [(match_operand:DI 1 “s_register_operand” " w") (match_operand:DI 2 “s_register_operand” " w")] UNSPEC_ASHIFT_SIGNED))] “TARGET_NEON && reload_completed” “vshl.s64\t%P0, %P1, %P2” [(set_attr “type” “neon_shift_reg”)] )
; The shift amount needs to be negated for right-shifts (define_insn “unsigned_shift_di3_neon” [(set (match_operand:DI 0 “s_register_operand” “=w”) (unspec:DI [(match_operand:DI 1 “s_register_operand” " w") (match_operand:DI 2 “s_register_operand” " w")] UNSPEC_ASHIFT_UNSIGNED))] “TARGET_NEON && reload_completed” “vshl.u64\t%P0, %P1, %P2” [(set_attr “type” “neon_shift_reg”)] )
(define_insn “ashrdi3_neon_imm_noclobber” [(set (match_operand:DI 0 “s_register_operand” “=w”) (ashiftrt:DI (match_operand:DI 1 “s_register_operand” " w") (match_operand:DI 2 “const_int_operand” " i")))] “TARGET_NEON && reload_completed && INTVAL (operands[2]) > 0 && INTVAL (operands[2]) <= 64” “vshr.s64\t%P0, %P1, %2” [(set_attr “type” “neon_shift_imm”)] )
(define_insn “lshrdi3_neon_imm_noclobber” [(set (match_operand:DI 0 “s_register_operand” “=w”) (lshiftrt:DI (match_operand:DI 1 “s_register_operand” " w") (match_operand:DI 2 “const_int_operand” " i")))] “TARGET_NEON && reload_completed && INTVAL (operands[2]) > 0 && INTVAL (operands[2]) <= 64” “vshr.u64\t%P0, %P1, %2” [(set_attr “type” “neon_shift_imm”)] )
;; ashrdi3_neon ;; lshrdi3_neon (define_insn_and_split “di3_neon” [(set (match_operand:DI 0 “s_register_operand” “= w, w,?&r,?r,?&r,?w,?w”) (rshifts:DI (match_operand:DI 1 “s_register_operand” " 0w, w, 0r, 0, r,0w, w") (match_operand:SI 2 “reg_or_int_operand” " r, i, r, i, i, r, i"))) (clobber (match_scratch:SI 3 “=2r, X, &r, X, X,2r, X”)) (clobber (match_scratch:SI 4 “= X, X, &r, X, X, X, X”)) (clobber (match_scratch:DI 5 “=&w, X, X, X, X,&w, X”)) (clobber (reg:CC CC_REGNUM))] “TARGET_NEON” “#” “TARGET_NEON && reload_completed” [(const_int 0)] " { if (IS_VFP_REGNUM (REGNO (operands[0]))) { if (CONST_INT_P (operands[2])) { if (INTVAL (operands[2]) < 1) { emit_insn (gen_movdi (operands[0], operands[1])); DONE; } else if (INTVAL (operands[2]) > 64) operands[2] = gen_rtx_CONST_INT (VOIDmode, 64);
/* Ditch the unnecessary clobbers. */ emit_insn (gen_<shift>di3_neon_imm_noclobber (operands[0], operands[1], operands[2])); } else { /* We must use a negative left-shift. */ emit_insn (gen_negsi2 (operands[3], operands[2])); emit_insn (gen_neon_load_count (operands[5], operands[3])); emit_insn (gen_<shifttype>_shift_di3_neon (operands[0], operands[1], operands[5])); } } else { /* The shift expanders support either full overlap or no overlap. */ gcc_assert (!reg_overlap_mentioned_p (operands[0], operands[1]) || REGNO (operands[0]) == REGNO (operands[1])); if (operands[2] == CONST1_RTX (SImode)) /* This clobbers CC. */ emit_insn (gen_arm_<shift>di3_1bit (operands[0], operands[1])); else /* This clobbers CC (ASHIFTRT by register only). */ arm_emit_coreregs_64bit_shift (<CODE>, operands[0], operands[1], operands[2], operands[3], operands[4]); } DONE;
}" [(set_attr “arch” “neon_for_64bits,neon_for_64bits,,,*,avoid_neon_for_64bits,avoid_neon_for_64bits”) (set_attr “opt” “,,speed,speed,speed,,”) (set_attr “type” “multiple”)] )
;; Widening operations
(define_insn “widen_ssum3” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (plus:<V_widen> (sign_extend:<V_widen> (match_operand:VW 1 “s_register_operand” “%w”)) (match_operand:<V_widen> 2 “s_register_operand” “w”)))] “TARGET_NEON” “vaddw.<V_s_elem>\t%q0, %q2, %P1” [(set_attr “type” “neon_add_widen”)] )
(define_insn “widen_usum3” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (plus:<V_widen> (zero_extend:<V_widen> (match_operand:VW 1 “s_register_operand” “%w”)) (match_operand:<V_widen> 2 “s_register_operand” “w”)))] “TARGET_NEON” “vaddw.<V_u_elem>\t%q0, %q2, %P1” [(set_attr “type” “neon_add_widen”)] )
;; VEXT can be used to synthesize coarse whole-vector shifts with 8-bit ;; shift-count granularity. That‘s good enough for the middle-end’s current ;; needs.
;; Note that it's not safe to perform such an operation in big-endian mode, ;; due to element-ordering issues.
(define_expand “vec_shr_” [(match_operand:VDQ 0 “s_register_operand” "") (match_operand:VDQ 1 “s_register_operand” "") (match_operand:SI 2 “const_multiple_of_8_operand” "")] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtx zero_reg; HOST_WIDE_INT num_bits = INTVAL (operands[2]); const int width = GET_MODE_BITSIZE (mode); const machine_mode bvecmode = (width == 128) ? V16QImode : V8QImode; rtx (*gen_ext) (rtx, rtx, rtx, rtx) = (width == 128) ? gen_neon_vextv16qi : gen_neon_vextv8qi;
if (num_bits == width) { emit_move_insn (operands[0], operands[1]); DONE; }
zero_reg = force_reg (bvecmode, CONST0_RTX (bvecmode)); operands[0] = gen_lowpart (bvecmode, operands[0]); operands[1] = gen_lowpart (bvecmode, operands[1]);
emit_insn (gen_ext (operands[0], operands[1], zero_reg, GEN_INT (num_bits / BITS_PER_UNIT))); DONE; })
(define_expand “vec_shl_” [(match_operand:VDQ 0 “s_register_operand” "") (match_operand:VDQ 1 “s_register_operand” "") (match_operand:SI 2 “const_multiple_of_8_operand” "")] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtx zero_reg; HOST_WIDE_INT num_bits = INTVAL (operands[2]); const int width = GET_MODE_BITSIZE (mode); const machine_mode bvecmode = (width == 128) ? V16QImode : V8QImode; rtx (*gen_ext) (rtx, rtx, rtx, rtx) = (width == 128) ? gen_neon_vextv16qi : gen_neon_vextv8qi;
if (num_bits == 0) { emit_move_insn (operands[0], CONST0_RTX (mode)); DONE; }
num_bits = width - num_bits;
zero_reg = force_reg (bvecmode, CONST0_RTX (bvecmode)); operands[0] = gen_lowpart (bvecmode, operands[0]); operands[1] = gen_lowpart (bvecmode, operands[1]);
emit_insn (gen_ext (operands[0], zero_reg, operands[1], GEN_INT (num_bits / BITS_PER_UNIT))); DONE; })
;; Helpers for quad-word reduction operations
; Add (or smin, smax...) the low N/2 elements of the N-element vector ; operand[1] to the high N/2 elements of same. Put the result in operand[0], an ; N/2-element vector.
(define_insn “quad_halves_v4si” [(set (match_operand:V2SI 0 “s_register_operand” “=w”) (vqh_ops:V2SI (vec_select:V2SI (match_operand:V4SI 1 “s_register_operand” “w”) (parallel [(const_int 0) (const_int 1)])) (vec_select:V2SI (match_dup 1) (parallel [(const_int 2) (const_int 3)]))))] “TARGET_NEON” “<VQH_mnem>.<VQH_sign>32\t%P0, %e1, %f1” [(set_attr “vqh_mnem” “<VQH_mnem>”) (set_attr “type” “neon_reduc_<VQH_type>_q”)] )
(define_insn “quad_halves_v4sf” [(set (match_operand:V2SF 0 “s_register_operand” “=w”) (vqhs_ops:V2SF (vec_select:V2SF (match_operand:V4SF 1 “s_register_operand” “w”) (parallel [(const_int 0) (const_int 1)])) (vec_select:V2SF (match_dup 1) (parallel [(const_int 2) (const_int 3)]))))] “TARGET_NEON && flag_unsafe_math_optimizations” “<VQH_mnem>.f32\t%P0, %e1, %f1” [(set_attr “vqh_mnem” “<VQH_mnem>”) (set_attr “type” “neon_fp_reduc_<VQH_type>_s_q”)] )
(define_insn “quad_halves_v8hi” [(set (match_operand:V4HI 0 “s_register_operand” “+w”) (vqh_ops:V4HI (vec_select:V4HI (match_operand:V8HI 1 “s_register_operand” “w”) (parallel [(const_int 0) (const_int 1) (const_int 2) (const_int 3)])) (vec_select:V4HI (match_dup 1) (parallel [(const_int 4) (const_int 5) (const_int 6) (const_int 7)]))))] “TARGET_NEON” “<VQH_mnem>.<VQH_sign>16\t%P0, %e1, %f1” [(set_attr “vqh_mnem” “<VQH_mnem>”) (set_attr “type” “neon_reduc_<VQH_type>_q”)] )
(define_insn “quad_halves_v16qi” [(set (match_operand:V8QI 0 “s_register_operand” “+w”) (vqh_ops:V8QI (vec_select:V8QI (match_operand:V16QI 1 “s_register_operand” “w”) (parallel [(const_int 0) (const_int 1) (const_int 2) (const_int 3) (const_int 4) (const_int 5) (const_int 6) (const_int 7)])) (vec_select:V8QI (match_dup 1) (parallel [(const_int 8) (const_int 9) (const_int 10) (const_int 11) (const_int 12) (const_int 13) (const_int 14) (const_int 15)]))))] “TARGET_NEON” “<VQH_mnem>.<VQH_sign>8\t%P0, %e1, %f1” [(set_attr “vqh_mnem” “<VQH_mnem>”) (set_attr “type” “neon_reduc_<VQH_type>_q”)] )
(define_expand “move_hi_quad_” [(match_operand:ANY128 0 “s_register_operand” "") (match_operand:<V_HALF> 1 “s_register_operand” "")] “TARGET_NEON” { emit_move_insn (simplify_gen_subreg (<V_HALF>mode, operands[0], mode, GET_MODE_SIZE (<V_HALF>mode)), operands[1]); DONE; })
(define_expand “move_lo_quad_” [(match_operand:ANY128 0 “s_register_operand” "") (match_operand:<V_HALF> 1 “s_register_operand” "")] “TARGET_NEON” { emit_move_insn (simplify_gen_subreg (<V_HALF>mode, operands[0], mode, 0), operands[1]); DONE; })
;; Reduction operations
(define_expand “reduc_plus_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VD 1 “s_register_operand” "")] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” { rtx vec = gen_reg_rtx (mode); neon_pairwise_reduce (vec, operands[1], mode, &gen_neon_vpadd_internal); /* The same result is actually computed into every element. */ emit_insn (gen_vec_extract (operands[0], vec, const0_rtx)); DONE; })
(define_expand “reduc_plus_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VQ 1 “s_register_operand” "")] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations) && !BYTES_BIG_ENDIAN” { rtx step1 = gen_reg_rtx (<V_HALF>mode);
emit_insn (gen_quad_halves_plus (step1, operands[1])); emit_insn (gen_reduc_plus_scal_<V_half> (operands[0], step1));
DONE; })
(define_expand “reduc_plus_scal_v2di” [(match_operand:DI 0 “nonimmediate_operand” “=w”) (match_operand:V2DI 1 “s_register_operand” "")] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtx vec = gen_reg_rtx (V2DImode);
emit_insn (gen_arm_reduc_plus_internal_v2di (vec, operands[1])); emit_insn (gen_vec_extractv2di (operands[0], vec, const0_rtx));
DONE; })
(define_insn “arm_reduc_plus_internal_v2di” [(set (match_operand:V2DI 0 “s_register_operand” “=w”) (unspec:V2DI [(match_operand:V2DI 1 “s_register_operand” “w”)] UNSPEC_VPADD))] “TARGET_NEON && !BYTES_BIG_ENDIAN” “vadd.i64\t%e0, %e1, %f1” [(set_attr “type” “neon_add_q”)] )
(define_expand “reduc_smin_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VD 1 “s_register_operand” "")] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” { rtx vec = gen_reg_rtx (mode);
neon_pairwise_reduce (vec, operands[1], mode, &gen_neon_vpsmin); /* The result is computed into every element of the vector. */ emit_insn (gen_vec_extract (operands[0], vec, const0_rtx)); DONE; })
(define_expand “reduc_smin_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VQ 1 “s_register_operand” "")] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations) && !BYTES_BIG_ENDIAN” { rtx step1 = gen_reg_rtx (<V_HALF>mode);
emit_insn (gen_quad_halves_smin (step1, operands[1])); emit_insn (gen_reduc_smin_scal_<V_half> (operands[0], step1));
DONE; })
(define_expand “reduc_smax_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VD 1 “s_register_operand” "")] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” { rtx vec = gen_reg_rtx (mode); neon_pairwise_reduce (vec, operands[1], mode, &gen_neon_vpsmax); /* The result is computed into every element of the vector. */ emit_insn (gen_vec_extract (operands[0], vec, const0_rtx)); DONE; })
(define_expand “reduc_smax_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VQ 1 “s_register_operand” "")] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations) && !BYTES_BIG_ENDIAN” { rtx step1 = gen_reg_rtx (<V_HALF>mode);
emit_insn (gen_quad_halves_smax (step1, operands[1])); emit_insn (gen_reduc_smax_scal_<V_half> (operands[0], step1));
DONE; })
(define_expand “reduc_umin_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VDI 1 “s_register_operand” "")] “TARGET_NEON” { rtx vec = gen_reg_rtx (mode); neon_pairwise_reduce (vec, operands[1], mode, &gen_neon_vpumin); /* The result is computed into every element of the vector. */ emit_insn (gen_vec_extract (operands[0], vec, const0_rtx)); DONE; })
(define_expand “reduc_umin_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VQI 1 “s_register_operand” "")] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtx step1 = gen_reg_rtx (<V_HALF>mode);
emit_insn (gen_quad_halves_umin (step1, operands[1])); emit_insn (gen_reduc_umin_scal_<V_half> (operands[0], step1));
DONE; })
(define_expand “reduc_umax_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VDI 1 “s_register_operand” "")] “TARGET_NEON” { rtx vec = gen_reg_rtx (mode); neon_pairwise_reduce (vec, operands[1], mode, &gen_neon_vpumax); /* The result is computed into every element of the vector. */ emit_insn (gen_vec_extract (operands[0], vec, const0_rtx)); DONE; })
(define_expand “reduc_umax_scal_” [(match_operand:<V_elem> 0 “nonimmediate_operand” "") (match_operand:VQI 1 “s_register_operand” "")] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtx step1 = gen_reg_rtx (<V_HALF>mode);
emit_insn (gen_quad_halves_umax (step1, operands[1])); emit_insn (gen_reduc_umax_scal_<V_half> (operands[0], step1));
DONE; })
(define_insn “neon_vpadd_internal” [(set (match_operand:VD 0 “s_register_operand” “=w”) (unspec:VD [(match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)] UNSPEC_VPADD))] “TARGET_NEON” “vpadd.<V_if_elem>\t%P0, %P1, %P2” ;; Assume this schedules like vadd. [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_reduc_add_s”) (const_string “neon_reduc_add”)))] )
(define_insn “neon_vpsmin” [(set (match_operand:VD 0 “s_register_operand” “=w”) (unspec:VD [(match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)] UNSPEC_VPSMIN))] “TARGET_NEON” “vpmin.<V_s_elem>\t%P0, %P1, %P2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_reduc_minmax_s”) (const_string “neon_reduc_minmax”)))] )
(define_insn “neon_vpsmax” [(set (match_operand:VD 0 “s_register_operand” “=w”) (unspec:VD [(match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)] UNSPEC_VPSMAX))] “TARGET_NEON” “vpmax.<V_s_elem>\t%P0, %P1, %P2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_reduc_minmax_s”) (const_string “neon_reduc_minmax”)))] )
(define_insn “neon_vpumin” [(set (match_operand:VDI 0 “s_register_operand” “=w”) (unspec:VDI [(match_operand:VDI 1 “s_register_operand” “w”) (match_operand:VDI 2 “s_register_operand” “w”)] UNSPEC_VPUMIN))] “TARGET_NEON” “vpmin.<V_u_elem>\t%P0, %P1, %P2” [(set_attr “type” “neon_reduc_minmax”)] )
(define_insn “neon_vpumax” [(set (match_operand:VDI 0 “s_register_operand” “=w”) (unspec:VDI [(match_operand:VDI 1 “s_register_operand” “w”) (match_operand:VDI 2 “s_register_operand” “w”)] UNSPEC_VPUMAX))] “TARGET_NEON” “vpmax.<V_u_elem>\t%P0, %P1, %P2” [(set_attr “type” “neon_reduc_minmax”)] )
;; Saturating arithmetic
; NOTE: Neon supports many more saturating variants of instructions than the ; following, but these are all GCC currently understands. ; FIXME: Actually, GCC doesn‘t know how to create saturating add/sub by itself ; yet either, although these patterns may be used by intrinsics when they’re ; added.
(define_insn “*ss_add_neon” [(set (match_operand:VD 0 “s_register_operand” “=w”) (ss_plus:VD (match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)))] “TARGET_NEON” “vqadd.<V_s_elem>\t%P0, %P1, %P2” [(set_attr “type” “neon_qadd”)] )
(define_insn “*us_add_neon” [(set (match_operand:VD 0 “s_register_operand” “=w”) (us_plus:VD (match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)))] “TARGET_NEON” “vqadd.<V_u_elem>\t%P0, %P1, %P2” [(set_attr “type” “neon_qadd”)] )
(define_insn “*ss_sub_neon” [(set (match_operand:VD 0 “s_register_operand” “=w”) (ss_minus:VD (match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)))] “TARGET_NEON” “vqsub.<V_s_elem>\t%P0, %P1, %P2” [(set_attr “type” “neon_qsub”)] )
(define_insn “*us_sub_neon” [(set (match_operand:VD 0 “s_register_operand” “=w”) (us_minus:VD (match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)))] “TARGET_NEON” “vqsub.<V_u_elem>\t%P0, %P1, %P2” [(set_attr “type” “neon_qsub”)] )
;; Conditional instructions. These are comparisons with conditional moves for ;; vectors. They perform the assignment: ;;
;; Vop0 = (Vop4 Vop5) ? Vop1 : Vop2; ;; ;; where op3 is <, <=, ==, !=, >= or >. Operations are performed ;; element-wise.
(define_expand “vcond” [(set (match_operand:VDQW 0 “s_register_operand” "") (if_then_else:VDQW (match_operator 3 “comparison_operator” [(match_operand:VDQW 4 “s_register_operand” "") (match_operand:VDQW 5 “nonmemory_operand” "")]) (match_operand:VDQW 1 “s_register_operand” "") (match_operand:VDQW 2 “s_register_operand” "")))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” { int inverse = 0; int use_zero_form = 0; int swap_bsl_operands = 0; rtx mask = gen_reg_rtx (<V_cmp_result>mode); rtx tmp = gen_reg_rtx (<V_cmp_result>mode);
rtx (*base_comparison) (rtx, rtx, rtx); rtx (*complimentary_comparison) (rtx, rtx, rtx);
switch (GET_CODE (operands[3])) { case GE: case GT: case LE: case LT: case EQ: if (operands[5] == CONST0_RTX (mode)) { use_zero_form = 1; break; } /* Fall through. */ default: if (!REG_P (operands[5])) operands[5] = force_reg (mode, operands[5]); }
switch (GET_CODE (operands[3])) { case LT: case UNLT: inverse = 1; /* Fall through. / case GE: case UNGE: case ORDERED: case UNORDERED: base_comparison = gen_neon_vcge; complimentary_comparison = gen_neon_vcgt; break; case LE: case UNLE: inverse = 1; / Fall through. */ case GT: case UNGT: base_comparison = gen_neon_vcgt; complimentary_comparison = gen_neon_vcge; break; case EQ: case NE: case UNEQ: base_comparison = gen_neon_vceq; complimentary_comparison = gen_neon_vceq; break; default: gcc_unreachable (); }
switch (GET_CODE (operands[3])) { case LT: case LE: case GT: case GE: case EQ: /* The easy case. Here we emit one of vcge, vcgt or vceq. As a LT b <=> b GE a && a LE b <=> b GT a. Our transformations are: a GE b -> a GE b a GT b -> a GT b a LE b -> b GE a a LT b -> b GT a a EQ b -> a EQ b Note that there also exist direct comparison against 0 forms, so catch those as a special case. / if (use_zero_form) { inverse = 0; switch (GET_CODE (operands[3])) { case LT: base_comparison = gen_neon_vclt; break; case LE: base_comparison = gen_neon_vcle; break; default: / Do nothing, other zero form cases already have the correct base_comparison. */ break; } }
if (!inverse) emit_insn (base_comparison (mask, operands[4], operands[5])); else emit_insn (complimentary_comparison (mask, operands[5], operands[4])); break; case UNLT: case UNLE: case UNGT: case UNGE: case NE: /* Vector compare returns false for lanes which are unordered, so if we use the inverse of the comparison we actually want to emit, then swap the operands to BSL, we will end up with the correct result. Note that a NE NaN and NaN NE b are true for all a, b. Our transformations are: a GE b -> !(b GT a) a GT b -> !(b GE a) a LE b -> !(a GT b) a LT b -> !(a GE b) a NE b -> !(a EQ b) */ if (inverse) emit_insn (base_comparison (mask, operands[4], operands[5])); else emit_insn (complimentary_comparison (mask, operands[5], operands[4])); swap_bsl_operands = 1; break; case UNEQ: /* We check (a > b || b > a). combining these comparisons give us true iff !(a != b && a ORDERED b), swapping the operands to BSL will then give us (a == b || a UNORDERED b) as intended. */ emit_insn (gen_neon_vcgt<mode> (mask, operands[4], operands[5])); emit_insn (gen_neon_vcgt<mode> (tmp, operands[5], operands[4])); emit_insn (gen_ior<v_cmp_result>3 (mask, mask, tmp)); swap_bsl_operands = 1; break; case UNORDERED: /* Operands are ORDERED iff (a > b || b >= a). Swapping the operands to BSL will give the UNORDERED case. */ swap_bsl_operands = 1; /* Fall through. */ case ORDERED: emit_insn (gen_neon_vcgt<mode> (tmp, operands[4], operands[5])); emit_insn (gen_neon_vcge<mode> (mask, operands[5], operands[4])); emit_insn (gen_ior<v_cmp_result>3 (mask, mask, tmp)); break; default: gcc_unreachable (); }
if (swap_bsl_operands) emit_insn (gen_neon_vbsl (operands[0], mask, operands[2], operands[1])); else emit_insn (gen_neon_vbsl (operands[0], mask, operands[1], operands[2])); DONE; })
(define_expand “vcondu” [(set (match_operand:VDQIW 0 “s_register_operand” "") (if_then_else:VDQIW (match_operator 3 “arm_comparison_operator” [(match_operand:VDQIW 4 “s_register_operand” "") (match_operand:VDQIW 5 “s_register_operand” "")]) (match_operand:VDQIW 1 “s_register_operand” "") (match_operand:VDQIW 2 “s_register_operand” "")))] “TARGET_NEON” { rtx mask; int inverse = 0, immediate_zero = 0;
mask = gen_reg_rtx (<V_cmp_result>mode);
if (operands[5] == CONST0_RTX (mode)) immediate_zero = 1; else if (!REG_P (operands[5])) operands[5] = force_reg (mode, operands[5]);
switch (GET_CODE (operands[3])) { case GEU: emit_insn (gen_neon_vcgeu (mask, operands[4], operands[5])); break;
case GTU: emit_insn (gen_neon_vcgtu<mode> (mask, operands[4], operands[5])); break; case EQ: emit_insn (gen_neon_vceq<mode> (mask, operands[4], operands[5])); break; case LEU: if (immediate_zero) emit_insn (gen_neon_vcle<mode> (mask, operands[4], operands[5])); else emit_insn (gen_neon_vcgeu<mode> (mask, operands[5], operands[4])); break; case LTU: if (immediate_zero) emit_insn (gen_neon_vclt<mode> (mask, operands[4], operands[5])); else emit_insn (gen_neon_vcgtu<mode> (mask, operands[5], operands[4])); break; case NE: emit_insn (gen_neon_vceq<mode> (mask, operands[4], operands[5])); inverse = 1; break; default: gcc_unreachable (); }
if (inverse) emit_insn (gen_neon_vbsl (operands[0], mask, operands[2], operands[1])); else emit_insn (gen_neon_vbsl (operands[0], mask, operands[1], operands[2]));
DONE; })
;; Patterns for builtins.
; good for plain vadd, vaddq.
(define_expand “neon_vadd” [(match_operand:VCVTF 0 “s_register_operand” “=w”) (match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] “TARGET_NEON” { if (!<Is_float_mode> || flag_unsafe_math_optimizations) emit_insn (gen_add3 (operands[0], operands[1], operands[2])); else emit_insn (gen_neon_vadd_unspec (operands[0], operands[1], operands[2])); DONE; })
; Note that NEON operations don't support the full IEEE 754 standard: in ; particular, denormal values are flushed to zero. This means that GCC cannot ; use those instructions for autovectorization, etc. unless ; -funsafe-math-optimizations is in effect (in which case flush-to-zero ; behaviour is permissible). Intrinsic operations (provided by the arm_neon.h ; header) must work in either case: if -funsafe-math-optimizations is given, ; intrinsics expand to “canonical” RTL where possible, otherwise intrinsics ; expand to unspecs (which may potentially limit the extent to which they might ; be optimized by generic code).
; Used for intrinsics when flag_unsafe_math_optimizations is false.
(define_insn “neon_vadd_unspec” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] UNSPEC_VADD))] “TARGET_NEON” “vadd.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_addsub_s”) (const_string “neon_add”)))] )
(define_insn “neon_vaddl” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VDI 1 “s_register_operand” “w”) (match_operand:VDI 2 “s_register_operand” “w”)] VADDL))] “TARGET_NEON” “vaddl.%#<V_sz_elem>\t%q0, %P1, %P2” [(set_attr “type” “neon_add_long”)] )
(define_insn “neon_vaddw” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “w”) (match_operand:VDI 2 “s_register_operand” “w”)] VADDW))] “TARGET_NEON” “vaddw.%#<V_sz_elem>\t%q0, %q1, %P2” [(set_attr “type” “neon_add_widen”)] )
; vhadd and vrhadd.
(define_insn “neon_vhadd” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)] VHADD))] “TARGET_NEON” “vhadd.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_add_halve_q”)] )
(define_insn “neon_vqadd” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:VDQIX 2 “s_register_operand” “w”)] VQADD))] “TARGET_NEON” “vqadd.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_qadd”)] )
(define_insn “neon_vaddhn” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”) (match_operand:VN 2 “s_register_operand” “w”)] VADDHN))] “TARGET_NEON” “vaddhn.<V_if_elem>\t%P0, %q1, %q2” [(set_attr “type” “neon_add_halve_narrow_q”)] )
;; Polynomial and Float multiplication. (define_insn “neon_vmul” [(set (match_operand:VPF 0 “s_register_operand” “=w”) (unspec:VPF [(match_operand:VPF 1 “s_register_operand” “w”) (match_operand:VPF 2 “s_register_operand” “w”)] UNSPEC_VMUL))] “TARGET_NEON” “vmul.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mul_s”) (const_string “neon_mul_<V_elem_ch>”)))] )
(define_expand “neon_vmla” [(match_operand:VDQW 0 “s_register_operand” “=w”) (match_operand:VDQW 1 “s_register_operand” “0”) (match_operand:VDQW 2 “s_register_operand” “w”) (match_operand:VDQW 3 “s_register_operand” “w”)] “TARGET_NEON” { if (!<Is_float_mode> || flag_unsafe_math_optimizations) emit_insn (gen_mul3add_neon (operands[0], operands[1], operands[2], operands[3])); else emit_insn (gen_neon_vmla_unspec (operands[0], operands[1], operands[2], operands[3])); DONE; })
(define_expand “neon_vfmaVCVTF:mode” [(match_operand:VCVTF 0 “s_register_operand”) (match_operand:VCVTF 1 “s_register_operand”) (match_operand:VCVTF 2 “s_register_operand”) (match_operand:VCVTF 3 “s_register_operand”)] “TARGET_NEON && TARGET_FMA” { emit_insn (gen_fma4_intrinsic (operands[0], operands[2], operands[3], operands[1])); DONE; })
(define_expand “neon_vfmsVCVTF:mode” [(match_operand:VCVTF 0 “s_register_operand”) (match_operand:VCVTF 1 “s_register_operand”) (match_operand:VCVTF 2 “s_register_operand”) (match_operand:VCVTF 3 “s_register_operand”)] “TARGET_NEON && TARGET_FMA” { emit_insn (gen_fmsub4_intrinsic (operands[0], operands[2], operands[3], operands[1])); DONE; })
; Used for intrinsics when flag_unsafe_math_optimizations is false.
(define_insn “neon_vmla_unspec” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” “0”) (match_operand:VDQW 2 “s_register_operand” “w”) (match_operand:VDQW 3 “s_register_operand” “w”)] UNSPEC_VMLA))] “TARGET_NEON” “vmla.<V_if_elem>\t%<V_reg>0, %<V_reg>2, %<V_reg>3” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s”) (const_string “neon_mla_<V_elem_ch>”)))] )
(define_insn “neon_vmlal” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VW 2 “s_register_operand” “w”) (match_operand:VW 3 “s_register_operand” “w”)] VMLAL))] “TARGET_NEON” “vmlal.%#<V_sz_elem>\t%q0, %P2, %P3” [(set_attr “type” “neon_mla_<V_elem_ch>_long”)] )
(define_expand “neon_vmls” [(match_operand:VDQW 0 “s_register_operand” “=w”) (match_operand:VDQW 1 “s_register_operand” “0”) (match_operand:VDQW 2 “s_register_operand” “w”) (match_operand:VDQW 3 “s_register_operand” “w”)] “TARGET_NEON” { if (!<Is_float_mode> || flag_unsafe_math_optimizations) emit_insn (gen_mul3negadd_neon (operands[0], operands[1], operands[2], operands[3])); else emit_insn (gen_neon_vmls_unspec (operands[0], operands[1], operands[2], operands[3])); DONE; })
; Used for intrinsics when flag_unsafe_math_optimizations is false.
(define_insn “neon_vmls_unspec” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” “0”) (match_operand:VDQW 2 “s_register_operand” “w”) (match_operand:VDQW 3 “s_register_operand” “w”)] UNSPEC_VMLS))] “TARGET_NEON” “vmls.<V_if_elem>\t%<V_reg>0, %<V_reg>2, %<V_reg>3” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s”) (const_string “neon_mla_<V_elem_ch>”)))] )
(define_insn “neon_vmlsl” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VW 2 “s_register_operand” “w”) (match_operand:VW 3 “s_register_operand” “w”)] VMLSL))] “TARGET_NEON” “vmlsl.%#<V_sz_elem>\t%q0, %P2, %P3” [(set_attr “type” “neon_mla_<V_elem_ch>_long”)] )
;; vqdmulh, vqrdmulh (define_insn “neon_vqdmulh” [(set (match_operand:VMDQI 0 “s_register_operand” “=w”) (unspec:VMDQI [(match_operand:VMDQI 1 “s_register_operand” “w”) (match_operand:VMDQI 2 “s_register_operand” “w”)] VQDMULH))] “TARGET_NEON” “vqdmulh.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_sat_mul_<V_elem_ch>”)] )
(define_insn “neon_vqdmlal” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VMDI 2 “s_register_operand” “w”) (match_operand:VMDI 3 “s_register_operand” “w”)] UNSPEC_VQDMLAL))] “TARGET_NEON” “vqdmlal.<V_s_elem>\t%q0, %P2, %P3” [(set_attr “type” “neon_sat_mla_<V_elem_ch>_long”)] )
(define_insn “neon_vqdmlsl” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VMDI 2 “s_register_operand” “w”) (match_operand:VMDI 3 “s_register_operand” “w”)] UNSPEC_VQDMLSL))] “TARGET_NEON” “vqdmlsl.<V_s_elem>\t%q0, %P2, %P3” [(set_attr “type” “neon_sat_mla_<V_elem_ch>_long”)] )
(define_insn “neon_vmull” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VW 1 “s_register_operand” “w”) (match_operand:VW 2 “s_register_operand” “w”)] VMULL))] “TARGET_NEON” “vmull.%#<V_sz_elem>\t%q0, %P1, %P2” [(set_attr “type” “neon_mul_<V_elem_ch>_long”)] )
(define_insn “neon_vqdmull” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VMDI 1 “s_register_operand” “w”) (match_operand:VMDI 2 “s_register_operand” “w”)] UNSPEC_VQDMULL))] “TARGET_NEON” “vqdmull.<V_s_elem>\t%q0, %P1, %P2” [(set_attr “type” “neon_sat_mul_<V_elem_ch>_long”)] )
(define_expand “neon_vsub” [(match_operand:VCVTF 0 “s_register_operand” “=w”) (match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] “TARGET_NEON” { if (!<Is_float_mode> || flag_unsafe_math_optimizations) emit_insn (gen_sub3 (operands[0], operands[1], operands[2])); else emit_insn (gen_neon_vsub_unspec (operands[0], operands[1], operands[2])); DONE; })
; Used for intrinsics when flag_unsafe_math_optimizations is false.
(define_insn “neon_vsub_unspec” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] UNSPEC_VSUB))] “TARGET_NEON” “vsub.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_addsub_s”) (const_string “neon_sub”)))] )
(define_insn “neon_vsubl” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VDI 1 “s_register_operand” “w”) (match_operand:VDI 2 “s_register_operand” “w”)] VSUBL))] “TARGET_NEON” “vsubl.%#<V_sz_elem>\t%q0, %P1, %P2” [(set_attr “type” “neon_sub_long”)] )
(define_insn “neon_vsubw” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “w”) (match_operand:VDI 2 “s_register_operand” “w”)] VSUBW))] “TARGET_NEON” “vsubw.%#<V_sz_elem>\t%q0, %q1, %P2” [(set_attr “type” “neon_sub_widen”)] )
(define_insn “neon_vqsub” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:VDQIX 2 “s_register_operand” “w”)] VQSUB))] “TARGET_NEON” “vqsub.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_qsub”)] )
(define_insn “neon_vhsub” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)] VHSUB))] “TARGET_NEON” “vhsub.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_sub_halve”)] )
(define_insn “neon_vsubhn” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”) (match_operand:VN 2 “s_register_operand” “w”)] VSUBHN))] “TARGET_NEON” “vsubhn.<V_if_elem>\t%P0, %q1, %q2” [(set_attr “type” “neon_sub_halve_narrow_q”)] )
(define_insn “neon_vceq” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w,w”) (unspec:<V_cmp_result> [(match_operand:VDQW 1 “s_register_operand” “w,w”) (match_operand:VDQW 2 “reg_or_zero_operand” “w,Dz”)] UNSPEC_VCEQ))] “TARGET_NEON” “@ vceq.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2 vceq.<V_if_elem>\t%<V_reg>0, %<V_reg>1, #0” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_compare_s”) (if_then_else (match_operand 2 “zero_operand”) (const_string “neon_compare_zero”) (const_string “neon_compare”))))] )
(define_insn “neon_vcge” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w,w”) (unspec:<V_cmp_result> [(match_operand:VDQW 1 “s_register_operand” “w,w”) (match_operand:VDQW 2 “reg_or_zero_operand” “w,Dz”)] UNSPEC_VCGE))] “TARGET_NEON” “@ vcge.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2 vcge.<V_s_elem>\t%<V_reg>0, %<V_reg>1, #0” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_compare_s”) (if_then_else (match_operand 2 “zero_operand”) (const_string “neon_compare_zero”) (const_string “neon_compare”))))] )
(define_insn “neon_vcgeu” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w”) (unspec:<V_cmp_result> [(match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)] UNSPEC_VCGEU))] “TARGET_NEON” “vcge.u%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_compare”)] )
(define_insn “neon_vcgt” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w,w”) (unspec:<V_cmp_result> [(match_operand:VDQW 1 “s_register_operand” “w,w”) (match_operand:VDQW 2 “reg_or_zero_operand” “w,Dz”)] UNSPEC_VCGT))] “TARGET_NEON” “@ vcgt.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2 vcgt.<V_s_elem>\t%<V_reg>0, %<V_reg>1, #0” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_compare_s”) (if_then_else (match_operand 2 “zero_operand”) (const_string “neon_compare_zero”) (const_string “neon_compare”))))] )
(define_insn “neon_vcgtu” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w”) (unspec:<V_cmp_result> [(match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)] UNSPEC_VCGTU))] “TARGET_NEON” “vcgt.u%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_compare”)] )
;; VCLE and VCLT only support comparisons with immediate zero (register ;; variants are VCGE and VCGT with operands reversed).
(define_insn “neon_vcle” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w”) (unspec:<V_cmp_result> [(match_operand:VDQW 1 “s_register_operand” “w”) (match_operand:VDQW 2 “zero_operand” “Dz”)] UNSPEC_VCLE))] “TARGET_NEON” “vcle.<V_s_elem>\t%<V_reg>0, %<V_reg>1, #0” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_compare_s”) (if_then_else (match_operand 2 “zero_operand”) (const_string “neon_compare_zero”) (const_string “neon_compare”))))] )
(define_insn “neon_vclt” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w”) (unspec:<V_cmp_result> [(match_operand:VDQW 1 “s_register_operand” “w”) (match_operand:VDQW 2 “zero_operand” “Dz”)] UNSPEC_VCLT))] “TARGET_NEON” “vclt.<V_s_elem>\t%<V_reg>0, %<V_reg>1, #0” [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_compare_s”) (if_then_else (match_operand 2 “zero_operand”) (const_string “neon_compare_zero”) (const_string “neon_compare”))))] )
(define_insn “neon_vcage” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w”) (unspec:<V_cmp_result> [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] UNSPEC_VCAGE))] “TARGET_NEON” “vacge.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_compare_s”)] )
(define_insn “neon_vcagt” [(set (match_operand:<V_cmp_result> 0 “s_register_operand” “=w”) (unspec:<V_cmp_result> [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] UNSPEC_VCAGT))] “TARGET_NEON” “vacgt.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_compare_s”)] )
(define_insn “neon_vtst” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)] UNSPEC_VTST))] “TARGET_NEON” “vtst.<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_tst”)] )
(define_insn “neon_vabd” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)] VABD))] “TARGET_NEON” “vabd.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_abd”)] )
(define_insn “neon_vabdf” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] UNSPEC_VABD_F))] “TARGET_NEON” “vabd.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_abd_s”)] )
(define_insn “neon_vabdl” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VW 1 “s_register_operand” “w”) (match_operand:VW 2 “s_register_operand” “w”)] VABDL))] “TARGET_NEON” “vabdl.%#<V_sz_elem>\t%q0, %P1, %P2” [(set_attr “type” “neon_abd_long”)] )
(define_insn “neon_vaba” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (plus:VDQIW (unspec:VDQIW [(match_operand:VDQIW 2 “s_register_operand” “w”) (match_operand:VDQIW 3 “s_register_operand” “w”)] VABD) (match_operand:VDQIW 1 “s_register_operand” “0”)))] “TARGET_NEON” “vaba.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>2, %<V_reg>3” [(set_attr “type” “neon_arith_acc”)] )
(define_insn “neon_vabal” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (plus:<V_widen> (unspec:<V_widen> [(match_operand:VW 2 “s_register_operand” “w”) (match_operand:VW 3 “s_register_operand” “w”)] VABDL) (match_operand:<V_widen> 1 “s_register_operand” “0”)))] “TARGET_NEON” “vabal.%#<V_sz_elem>\t%q0, %P2, %P3” [(set_attr “type” “neon_arith_acc”)] )
(define_insn “neon_v” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”) (match_operand:VDQIW 2 “s_register_operand” “w”)] VMAXMIN))] “TARGET_NEON” “v.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_minmax”)] )
(define_insn “neon_vf” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] VMAXMINF))] “TARGET_NEON” “v.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_minmax_s”)] )
(define_expand “neon_vpadd” [(match_operand:VD 0 “s_register_operand” “=w”) (match_operand:VD 1 “s_register_operand” “w”) (match_operand:VD 2 “s_register_operand” “w”)] “TARGET_NEON” { emit_insn (gen_neon_vpadd_internal (operands[0], operands[1], operands[2])); DONE; })
(define_insn “neon_vpaddl” [(set (match_operand:<V_double_width> 0 “s_register_operand” “=w”) (unspec:<V_double_width> [(match_operand:VDQIW 1 “s_register_operand” “w”)] VPADDL))] “TARGET_NEON” “vpaddl.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_reduc_add_long”)] )
(define_insn “neon_vpadal” [(set (match_operand:<V_double_width> 0 “s_register_operand” “=w”) (unspec:<V_double_width> [(match_operand:<V_double_width> 1 “s_register_operand” “0”) (match_operand:VDQIW 2 “s_register_operand” “w”)] VPADAL))] “TARGET_NEON” “vpadal.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>2” [(set_attr “type” “neon_reduc_add_acc”)] )
(define_insn “neon_vp” [(set (match_operand:VDI 0 “s_register_operand” “=w”) (unspec:VDI [(match_operand:VDI 1 “s_register_operand” “w”) (match_operand:VDI 2 “s_register_operand” “w”)] VPMAXMIN))] “TARGET_NEON” “vp.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_reduc_minmax”)] )
(define_insn “neon_vpf” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] VPMAXMINF))] “TARGET_NEON” “vp.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_reduc_minmax_s”)] )
(define_insn “neon_vrecps” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] UNSPEC_VRECPS))] “TARGET_NEON” “vrecps.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_recps_s”)] )
(define_insn “neon_vrsqrts” [(set (match_operand:VCVTF 0 “s_register_operand” “=w”) (unspec:VCVTF [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:VCVTF 2 “s_register_operand” “w”)] UNSPEC_VRSQRTS))] “TARGET_NEON” “vrsqrts.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_fp_rsqrts_s”)] )
(define_expand “neon_vabs” [(match_operand:VDQW 0 “s_register_operand” "") (match_operand:VDQW 1 “s_register_operand” "")] “TARGET_NEON” { emit_insn (gen_abs2 (operands[0], operands[1])); DONE; })
(define_insn “neon_vqabs” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”)] UNSPEC_VQABS))] “TARGET_NEON” “vqabs.<V_s_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_qabs”)] )
(define_insn “neon_bswap” [(set (match_operand:VDQHSD 0 “register_operand” “=w”) (bswap:VDQHSD (match_operand:VDQHSD 1 “register_operand” “w”)))] “TARGET_NEON” “vrev<V_sz_elem>.8\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_rev”)] )
(define_expand “neon_vneg” [(match_operand:VDQW 0 “s_register_operand” "") (match_operand:VDQW 1 “s_register_operand” "")] “TARGET_NEON” { emit_insn (gen_neg2 (operands[0], operands[1])); DONE; })
(define_expand “neon_copysignf” [(match_operand:VCVTF 0 “register_operand”) (match_operand:VCVTF 1 “register_operand”) (match_operand:VCVTF 2 “register_operand”)] “TARGET_NEON” "{ rtx v_bitmask_cast; rtx v_bitmask = gen_reg_rtx (VCVTF:V_cmp_resultmode); int i, n_elt = GET_MODE_NUNITS (mode); rtvec v = rtvec_alloc (n_elt);
/* Create bitmask for vector select. */ for (i = 0; i < n_elt; ++i) RTVEC_ELT (v, i) = GEN_INT (0x80000000); emit_move_insn (v_bitmask, gen_rtx_CONST_VECTOR (<VCVTF:V_cmp_result>mode, v)); emit_move_insn (operands[0], operands[2]); v_bitmask_cast = simplify_gen_subreg (<MODE>mode, v_bitmask, <VCVTF:V_cmp_result>mode, 0); emit_insn (gen_neon_vbsl<mode> (operands[0], v_bitmask_cast, operands[0], operands[1])); DONE;
}" )
(define_insn “neon_vqneg” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”)] UNSPEC_VQNEG))] “TARGET_NEON” “vqneg.<V_s_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_qneg”)] )
(define_insn “neon_vcls” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (unspec:VDQIW [(match_operand:VDQIW 1 “s_register_operand” “w”)] UNSPEC_VCLS))] “TARGET_NEON” “vcls.<V_s_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_cls”)] )
(define_insn “clz2” [(set (match_operand:VDQIW 0 “s_register_operand” “=w”) (clz:VDQIW (match_operand:VDQIW 1 “s_register_operand” “w”)))] “TARGET_NEON” “vclz.<V_if_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_cnt”)] )
(define_expand “neon_vclz” [(match_operand:VDQIW 0 “s_register_operand” "") (match_operand:VDQIW 1 “s_register_operand” "")] “TARGET_NEON” { emit_insn (gen_clz2 (operands[0], operands[1])); DONE; })
(define_insn “popcount2” [(set (match_operand:VE 0 “s_register_operand” “=w”) (popcount:VE (match_operand:VE 1 “s_register_operand” “w”)))] “TARGET_NEON” “vcnt.<V_sz_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_cnt”)] )
(define_expand “neon_vcnt” [(match_operand:VE 0 “s_register_operand” “=w”) (match_operand:VE 1 “s_register_operand” “w”)] “TARGET_NEON” { emit_insn (gen_popcount2 (operands[0], operands[1])); DONE; })
(define_insn “neon_vrecpe” [(set (match_operand:V32 0 “s_register_operand” “=w”) (unspec:V32 [(match_operand:V32 1 “s_register_operand” “w”)] UNSPEC_VRECPE))] “TARGET_NEON” “vrecpe.<V_u_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_fp_recpe_s”)] )
(define_insn “neon_vrsqrte” [(set (match_operand:V32 0 “s_register_operand” “=w”) (unspec:V32 [(match_operand:V32 1 “s_register_operand” “w”)] UNSPEC_VRSQRTE))] “TARGET_NEON” “vrsqrte.<V_u_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_fp_rsqrte_s”)] )
(define_expand “neon_vmvn” [(match_operand:VDQIW 0 “s_register_operand” "") (match_operand:VDQIW 1 “s_register_operand” "")] “TARGET_NEON” { emit_insn (gen_one_cmpl2 (operands[0], operands[1])); DONE; })
(define_insn “neon_vget_lane_sext_internal” [(set (match_operand:SI 0 “s_register_operand” “=r”) (sign_extend:SI (vec_select:<V_elem> (match_operand:VD 1 “s_register_operand” “w”) (parallel [(match_operand:SI 2 “immediate_operand” “i”)]))))] “TARGET_NEON” { if (BYTES_BIG_ENDIAN) { int elt = INTVAL (operands[2]); elt = GET_MODE_NUNITS (mode) - 1 - elt; operands[2] = GEN_INT (elt); } return “vmov.s<V_sz_elem>\t%0, %P1[%c2]”; } [(set_attr “type” “neon_to_gp”)] )
(define_insn “neon_vget_lane_zext_internal” [(set (match_operand:SI 0 “s_register_operand” “=r”) (zero_extend:SI (vec_select:<V_elem> (match_operand:VD 1 “s_register_operand” “w”) (parallel [(match_operand:SI 2 “immediate_operand” “i”)]))))] “TARGET_NEON” { if (BYTES_BIG_ENDIAN) { int elt = INTVAL (operands[2]); elt = GET_MODE_NUNITS (mode) - 1 - elt; operands[2] = GEN_INT (elt); } return “vmov.u<V_sz_elem>\t%0, %P1[%c2]”; } [(set_attr “type” “neon_to_gp”)] )
(define_insn “neon_vget_lane_sext_internal” [(set (match_operand:SI 0 “s_register_operand” “=r”) (sign_extend:SI (vec_select:<V_elem> (match_operand:VQ 1 “s_register_operand” “w”) (parallel [(match_operand:SI 2 “immediate_operand” “i”)]))))] “TARGET_NEON” { rtx ops[3]; int regno = REGNO (operands[1]); unsigned int halfelts = GET_MODE_NUNITS (mode) / 2; unsigned int elt = INTVAL (operands[2]); unsigned int elt_adj = elt % halfelts;
if (BYTES_BIG_ENDIAN) elt_adj = halfelts - 1 - elt_adj;
ops[0] = operands[0]; ops[1] = gen_rtx_REG (<V_HALF>mode, regno + 2 * (elt / halfelts)); ops[2] = GEN_INT (elt_adj); output_asm_insn (“vmov.s<V_sz_elem>\t%0, %P1[%c2]”, ops);
return ""; } [(set_attr “type” “neon_to_gp_q”)] )
(define_insn “neon_vget_lane_zext_internal” [(set (match_operand:SI 0 “s_register_operand” “=r”) (zero_extend:SI (vec_select:<V_elem> (match_operand:VQ 1 “s_register_operand” “w”) (parallel [(match_operand:SI 2 “immediate_operand” “i”)]))))] “TARGET_NEON” { rtx ops[3]; int regno = REGNO (operands[1]); unsigned int halfelts = GET_MODE_NUNITS (mode) / 2; unsigned int elt = INTVAL (operands[2]); unsigned int elt_adj = elt % halfelts;
if (BYTES_BIG_ENDIAN) elt_adj = halfelts - 1 - elt_adj;
ops[0] = operands[0]; ops[1] = gen_rtx_REG (<V_HALF>mode, regno + 2 * (elt / halfelts)); ops[2] = GEN_INT (elt_adj); output_asm_insn (“vmov.u<V_sz_elem>\t%0, %P1[%c2]”, ops);
return ""; } [(set_attr “type” “neon_to_gp_q”)] )
(define_expand “neon_vget_lane” [(match_operand:<V_ext> 0 “s_register_operand” "") (match_operand:VDQW 1 “s_register_operand” "") (match_operand:SI 2 “immediate_operand” "")] “TARGET_NEON” { neon_lane_bounds (operands[2], 0, GET_MODE_NUNITS (mode));
if (BYTES_BIG_ENDIAN) { /* The intrinsics are defined in terms of a model where the element ordering in memory is vldm order, whereas the generic RTL is defined in terms of a model where the element ordering in memory is array order. Convert the lane number to conform to this model. */ unsigned int elt = INTVAL (operands[2]); unsigned int reg_nelts = 64 / GET_MODE_BITSIZE (GET_MODE_INNER (mode)); elt ^= reg_nelts - 1; operands[2] = GEN_INT (elt); }
if (GET_MODE_BITSIZE (GET_MODE_INNER (mode)) == 32) emit_insn (gen_vec_extract (operands[0], operands[1], operands[2])); else emit_insn (gen_neon_vget_lane_sext_internal (operands[0], operands[1], operands[2])); DONE; })
(define_expand “neon_vget_laneu” [(match_operand:<V_ext> 0 “s_register_operand” "") (match_operand:VDQIW 1 “s_register_operand” "") (match_operand:SI 2 “immediate_operand” "")] “TARGET_NEON” { neon_lane_bounds (operands[2], 0, GET_MODE_NUNITS (mode));
if (BYTES_BIG_ENDIAN) { /* The intrinsics are defined in terms of a model where the element ordering in memory is vldm order, whereas the generic RTL is defined in terms of a model where the element ordering in memory is array order. Convert the lane number to conform to this model. */ unsigned int elt = INTVAL (operands[2]); unsigned int reg_nelts = 64 / GET_MODE_BITSIZE (GET_MODE_INNER (mode)); elt ^= reg_nelts - 1; operands[2] = GEN_INT (elt); }
if (GET_MODE_BITSIZE (GET_MODE_INNER (mode)) == 32) emit_insn (gen_vec_extract (operands[0], operands[1], operands[2])); else emit_insn (gen_neon_vget_lane_zext_internal (operands[0], operands[1], operands[2])); DONE; })
(define_expand “neon_vget_lanedi” [(match_operand:DI 0 “s_register_operand” “=r”) (match_operand:DI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” "")] “TARGET_NEON” { neon_lane_bounds (operands[2], 0, 1); emit_move_insn (operands[0], operands[1]); DONE; })
(define_expand “neon_vget_lanev2di” [(match_operand:DI 0 “s_register_operand” "") (match_operand:V2DI 1 “s_register_operand” "") (match_operand:SI 2 “immediate_operand” "")] “TARGET_NEON” { switch (INTVAL (operands[2])) { case 0: emit_move_insn (operands[0], gen_lowpart (DImode, operands[1])); break; case 1: emit_move_insn (operands[0], gen_highpart (DImode, operands[1])); break; default: neon_lane_bounds (operands[2], 0, 1); FAIL; } DONE; })
(define_expand “neon_vset_lane” [(match_operand:VDQ 0 “s_register_operand” “=w”) (match_operand:<V_elem> 1 “s_register_operand” “r”) (match_operand:VDQ 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”)] “TARGET_NEON” { unsigned int elt = INTVAL (operands[3]); neon_lane_bounds (operands[3], 0, GET_MODE_NUNITS (mode));
if (BYTES_BIG_ENDIAN) { unsigned int reg_nelts = 64 / GET_MODE_BITSIZE (GET_MODE_INNER (mode)); elt ^= reg_nelts - 1; }
emit_insn (gen_vec_set_internal (operands[0], operands[1], GEN_INT (1 << elt), operands[2])); DONE; })
; See neon_vget_lanedi comment for reasons operands 2 & 3 are ignored.
(define_expand “neon_vset_lanedi” [(match_operand:DI 0 “s_register_operand” “=w”) (match_operand:DI 1 “s_register_operand” “r”) (match_operand:DI 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”)] “TARGET_NEON” { neon_lane_bounds (operands[3], 0, 1); emit_move_insn (operands[0], operands[1]); DONE; })
(define_expand “neon_vcreate” [(match_operand:VDX 0 “s_register_operand” "") (match_operand:DI 1 “general_operand” "")] “TARGET_NEON” { rtx src = gen_lowpart (mode, operands[1]); emit_move_insn (operands[0], src); DONE; })
(define_insn “neon_vdup_n” [(set (match_operand:VX 0 “s_register_operand” “=w”) (vec_duplicate:VX (match_operand:<V_elem> 1 “s_register_operand” “r”)))] “TARGET_NEON” “vdup.<V_sz_elem>\t%<V_reg>0, %1” [(set_attr “type” “neon_from_gp”)] )
(define_insn “neon_vdup_n” [(set (match_operand:V32 0 “s_register_operand” “=w,w”) (vec_duplicate:V32 (match_operand:<V_elem> 1 “s_register_operand” “r,t”)))] “TARGET_NEON” “@ vdup.<V_sz_elem>\t%<V_reg>0, %1 vdup.<V_sz_elem>\t%<V_reg>0, %y1” [(set_attr “type” “neon_from_gp,neon_dup”)] )
(define_expand “neon_vdup_ndi” [(match_operand:DI 0 “s_register_operand” “=w”) (match_operand:DI 1 “s_register_operand” “r”)] “TARGET_NEON” { emit_move_insn (operands[0], operands[1]); DONE; } )
(define_insn “neon_vdup_nv2di” [(set (match_operand:V2DI 0 “s_register_operand” “=w,w”) (vec_duplicate:V2DI (match_operand:DI 1 “s_register_operand” “r,w”)))] “TARGET_NEON” “@ vmov\t%e0, %Q1, %R1;vmov\t%f0, %Q1, %R1 vmov\t%e0, %P1;vmov\t%f0, %P1” [(set_attr “length” “8”) (set_attr “type” “multiple”)] )
(define_insn “neon_vdup_lane_internal” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (vec_duplicate:VDQW (vec_select:<V_elem> (match_operand:<V_double_vector_mode> 1 “s_register_operand” “w”) (parallel [(match_operand:SI 2 “immediate_operand” “i”)]))))] “TARGET_NEON” { if (BYTES_BIG_ENDIAN) { int elt = INTVAL (operands[2]); elt = GET_MODE_NUNITS (<V_double_vector_mode>mode) - 1 - elt; operands[2] = GEN_INT (elt); } if (<Is_d_reg>) return “vdup.<V_sz_elem>\t%P0, %P1[%c2]”; else return “vdup.<V_sz_elem>\t%q0, %P1[%c2]”; } [(set_attr “type” “neon_dup”)] )
(define_expand “neon_vdup_lane” [(match_operand:VDQW 0 “s_register_operand” “=w”) (match_operand:<V_double_vector_mode> 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] “TARGET_NEON” { neon_lane_bounds (operands[2], 0, GET_MODE_NUNITS (<V_double_vector_mode>mode)); if (BYTES_BIG_ENDIAN) { unsigned int elt = INTVAL (operands[2]); unsigned int reg_nelts = 64 / GET_MODE_BITSIZE (GET_MODE_INNER (<V_double_vector_mode>mode)); elt ^= reg_nelts - 1; operands[2] = GEN_INT (elt); } emit_insn (gen_neon_vdup_lane_internal (operands[0], operands[1], operands[2])); DONE; })
; Scalar index is ignored, since only zero is valid here. (define_expand “neon_vdup_lanedi” [(match_operand:DI 0 “s_register_operand” “=w”) (match_operand:DI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] “TARGET_NEON” { neon_lane_bounds (operands[2], 0, 1); emit_move_insn (operands[0], operands[1]); DONE; })
; Likewise for v2di, as the DImode second operand has only a single element. (define_expand “neon_vdup_lanev2di” [(match_operand:V2DI 0 “s_register_operand” “=w”) (match_operand:DI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] “TARGET_NEON” { neon_lane_bounds (operands[2], 0, 1); emit_insn (gen_neon_vdup_nv2di (operands[0], operands[1])); DONE; })
; Disabled before reload because we don't want combine doing something silly, ; but used by the post-reload expansion of neon_vcombine. (define_insn “*neon_vswp” [(set (match_operand:VDQX 0 “s_register_operand” “+w”) (match_operand:VDQX 1 “s_register_operand” “+w”)) (set (match_dup 1) (match_dup 0))] “TARGET_NEON && reload_completed” “vswp\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_permute”)] )
;; In this insn, operand 1 should be low, and operand 2 the high part of the ;; dest vector. ;; FIXME: A different implementation of this builtin could make it much ;; more likely that we wouldn't actually need to output anything (we could make ;; it so that the reg allocator puts things in the right places magically ;; instead). Lack of subregs for vectors makes that tricky though, I think.
(define_insn_and_split “neon_vcombine” [(set (match_operand:<V_DOUBLE> 0 “s_register_operand” “=w”) (vec_concat:<V_DOUBLE> (match_operand:VDX 1 “s_register_operand” “w”) (match_operand:VDX 2 “s_register_operand” “w”)))] “TARGET_NEON” “#” “&& reload_completed” [(const_int 0)] { neon_split_vcombine (operands); DONE; } [(set_attr “type” “multiple”)] )
(define_expand “neon_vget_high” [(match_operand:<V_HALF> 0 “s_register_operand”) (match_operand:VQX 1 “s_register_operand”)] “TARGET_NEON” { emit_move_insn (operands[0], simplify_gen_subreg (<V_HALF>mode, operands[1], mode, GET_MODE_SIZE (<V_HALF>mode))); DONE; })
(define_expand “neon_vget_low” [(match_operand:<V_HALF> 0 “s_register_operand”) (match_operand:VQX 1 “s_register_operand”)] “TARGET_NEON” { emit_move_insn (operands[0], simplify_gen_subreg (<V_HALF>mode, operands[1], mode, 0)); DONE; })
(define_insn “float<V_cvtto>2” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (float:<V_CVTTO> (match_operand:VCVTI 1 “s_register_operand” “w”)))] “TARGET_NEON && !flag_rounding_math” “vcvt.f32.s32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_int_to_fp_<V_elem_ch>”)] )
(define_insn “floatuns<V_cvtto>2” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (unsigned_float:<V_CVTTO> (match_operand:VCVTI 1 “s_register_operand” “w”)))] “TARGET_NEON && !flag_rounding_math” “vcvt.f32.u32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_int_to_fp_<V_elem_ch>”)] )
(define_insn “fix_trunc<V_cvtto>2” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (fix:<V_CVTTO> (match_operand:VCVTF 1 “s_register_operand” “w”)))] “TARGET_NEON” “vcvt.s32.f32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_fp_to_int_<V_elem_ch>”)] )
(define_insn “fixuns_trunc<V_cvtto>2” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (unsigned_fix:<V_CVTTO> (match_operand:VCVTF 1 “s_register_operand” “w”)))] “TARGET_NEON” “vcvt.u32.f32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_fp_to_int_<V_elem_ch>”)] )
(define_insn “neon_vcvt” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (unspec:<V_CVTTO> [(match_operand:VCVTF 1 “s_register_operand” “w”)] VCVT_US))] “TARGET_NEON” “vcvt.%#32.f32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_fp_to_int_<V_elem_ch>”)] )
(define_insn “neon_vcvt” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (unspec:<V_CVTTO> [(match_operand:VCVTI 1 “s_register_operand” “w”)] VCVT_US))] “TARGET_NEON” “vcvt.f32.%#32\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_int_to_fp_<V_elem_ch>”)] )
(define_insn “neon_vcvtv4sfv4hf” [(set (match_operand:V4SF 0 “s_register_operand” “=w”) (unspec:V4SF [(match_operand:V4HF 1 “s_register_operand” “w”)] UNSPEC_VCVT))] “TARGET_NEON && TARGET_FP16” “vcvt.f32.f16\t%q0, %P1” [(set_attr “type” “neon_fp_cvt_widen_h”)] )
(define_insn “neon_vcvtv4hfv4sf” [(set (match_operand:V4HF 0 “s_register_operand” “=w”) (unspec:V4HF [(match_operand:V4SF 1 “s_register_operand” “w”)] UNSPEC_VCVT))] “TARGET_NEON && TARGET_FP16” “vcvt.f16.f32\t%P0, %q1” [(set_attr “type” “neon_fp_cvt_narrow_s_q”)] )
(define_insn “neon_vcvt_n” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (unspec:<V_CVTTO> [(match_operand:VCVTF 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VCVT_US_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 1, 33); return “vcvt.%#32.f32\t%<V_reg>0, %<V_reg>1, %2”; } [(set_attr “type” “neon_fp_to_int_<V_elem_ch>”)] )
(define_insn “neon_vcvt_n” [(set (match_operand:<V_CVTTO> 0 “s_register_operand” “=w”) (unspec:<V_CVTTO> [(match_operand:VCVTI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VCVT_US_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 1, 33); return “vcvt.f32.%#32\t%<V_reg>0, %<V_reg>1, %2”; } [(set_attr “type” “neon_int_to_fp_<V_elem_ch>”)] )
(define_insn “neon_vmovn” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”)] UNSPEC_VMOVN))] “TARGET_NEON” “vmovn.<V_if_elem>\t%P0, %q1” [(set_attr “type” “neon_shift_imm_narrow_q”)] )
(define_insn “neon_vqmovn” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”)] VQMOVN))] “TARGET_NEON” “vqmovn.%#<V_sz_elem>\t%P0, %q1” [(set_attr “type” “neon_sat_shift_imm_narrow_q”)] )
(define_insn “neon_vqmovun” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”)] UNSPEC_VQMOVUN))] “TARGET_NEON” “vqmovun.<V_s_elem>\t%P0, %q1” [(set_attr “type” “neon_sat_shift_imm_narrow_q”)] )
(define_insn “neon_vmovl” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VW 1 “s_register_operand” “w”)] VMOVL))] “TARGET_NEON” “vmovl.%#<V_sz_elem>\t%q0, %P1” [(set_attr “type” “neon_shift_imm_long”)] )
(define_insn “neon_vmul_lane” [(set (match_operand:VMD 0 “s_register_operand” “=w”) (unspec:VMD [(match_operand:VMD 1 “s_register_operand” “w”) (match_operand:VMD 2 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VMUL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[3], 0, GET_MODE_NUNITS (mode)); return “vmul.<V_if_elem>\t%P0, %P1, %P2[%c3]”; } [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mul_s_scalar”) (const_string “neon_mul_<V_elem_ch>_scalar”)))] )
(define_insn “neon_vmul_lane” [(set (match_operand:VMQ 0 “s_register_operand” “=w”) (unspec:VMQ [(match_operand:VMQ 1 “s_register_operand” “w”) (match_operand:<V_HALF> 2 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VMUL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[3], 0, GET_MODE_NUNITS (<V_HALF>mode)); return “vmul.<V_if_elem>\t%q0, %q1, %P2[%c3]”; } [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mul_s_scalar”) (const_string “neon_mul_<V_elem_ch>_scalar”)))] )
(define_insn “neon_vmull_lane” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VMDI 1 “s_register_operand” “w”) (match_operand:VMDI 2 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 3 “immediate_operand” “i”)] VMULL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[3], 0, GET_MODE_NUNITS (mode)); return “vmull.%#<V_sz_elem>\t%q0, %P1, %P2[%c3]”; } [(set_attr “type” “neon_mul_<V_elem_ch>_scalar_long”)] )
(define_insn “neon_vqdmull_lane” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VMDI 1 “s_register_operand” “w”) (match_operand:VMDI 2 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VQDMULL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[3], 0, GET_MODE_NUNITS (mode)); return “vqdmull.<V_s_elem>\t%q0, %P1, %P2[%c3]”; } [(set_attr “type” “neon_sat_mul_<V_elem_ch>_scalar_long”)] )
(define_insn “neon_vqdmulh_lane” [(set (match_operand:VMQI 0 “s_register_operand” “=w”) (unspec:VMQI [(match_operand:VMQI 1 “s_register_operand” “w”) (match_operand:<V_HALF> 2 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 3 “immediate_operand” “i”)] VQDMULH_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[3], 0, GET_MODE_NUNITS (mode)); return “vqdmulh.<V_s_elem>\t%q0, %q1, %P2[%c3]”; } [(set_attr “type” “neon_sat_mul_<V_elem_ch>_scalar_q”)] )
(define_insn “neon_vqdmulh_lane” [(set (match_operand:VMDI 0 “s_register_operand” “=w”) (unspec:VMDI [(match_operand:VMDI 1 “s_register_operand” “w”) (match_operand:VMDI 2 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 3 “immediate_operand” “i”)] VQDMULH_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[3], 0, GET_MODE_NUNITS (mode)); return “vqdmulh.<V_s_elem>\t%P0, %P1, %P2[%c3]”; } [(set_attr “type” “neon_sat_mul_<V_elem_ch>_scalar_q”)] )
(define_insn “neon_vmla_lane” [(set (match_operand:VMD 0 “s_register_operand” “=w”) (unspec:VMD [(match_operand:VMD 1 “s_register_operand” “0”) (match_operand:VMD 2 “s_register_operand” “w”) (match_operand:VMD 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] UNSPEC_VMLA_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vmla.<V_if_elem>\t%P0, %P2, %P3[%c4]”; } [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s_scalar”) (const_string “neon_mla_<V_elem_ch>_scalar”)))] )
(define_insn “neon_vmla_lane” [(set (match_operand:VMQ 0 “s_register_operand” “=w”) (unspec:VMQ [(match_operand:VMQ 1 “s_register_operand” “0”) (match_operand:VMQ 2 “s_register_operand” “w”) (match_operand:<V_HALF> 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] UNSPEC_VMLA_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vmla.<V_if_elem>\t%q0, %q2, %P3[%c4]”; } [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s_scalar”) (const_string “neon_mla_<V_elem_ch>_scalar”)))] )
(define_insn “neon_vmlal_lane” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VMDI 2 “s_register_operand” “w”) (match_operand:VMDI 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] VMLAL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vmlal.%#<V_sz_elem>\t%q0, %P2, %P3[%c4]”; } [(set_attr “type” “neon_mla_<V_elem_ch>_scalar_long”)] )
(define_insn “neon_vqdmlal_lane” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VMDI 2 “s_register_operand” “w”) (match_operand:VMDI 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] UNSPEC_VQDMLAL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vqdmlal.<V_s_elem>\t%q0, %P2, %P3[%c4]”; } [(set_attr “type” “neon_sat_mla_<V_elem_ch>_scalar_long”)] )
(define_insn “neon_vmls_lane” [(set (match_operand:VMD 0 “s_register_operand” “=w”) (unspec:VMD [(match_operand:VMD 1 “s_register_operand” “0”) (match_operand:VMD 2 “s_register_operand” “w”) (match_operand:VMD 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] UNSPEC_VMLS_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vmls.<V_if_elem>\t%P0, %P2, %P3[%c4]”; } [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s_scalar”) (const_string “neon_mla_<V_elem_ch>_scalar”)))] )
(define_insn “neon_vmls_lane” [(set (match_operand:VMQ 0 “s_register_operand” “=w”) (unspec:VMQ [(match_operand:VMQ 1 “s_register_operand” “0”) (match_operand:VMQ 2 “s_register_operand” “w”) (match_operand:<V_HALF> 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] UNSPEC_VMLS_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vmls.<V_if_elem>\t%q0, %q2, %P3[%c4]”; } [(set (attr “type”) (if_then_else (match_test “<Is_float_mode>”) (const_string “neon_fp_mla_s_scalar”) (const_string “neon_mla_<V_elem_ch>_scalar”)))] )
(define_insn “neon_vmlsl_lane” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VMDI 2 “s_register_operand” “w”) (match_operand:VMDI 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] VMLSL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vmlsl.%#<V_sz_elem>\t%q0, %P2, %P3[%c4]”; } [(set_attr “type” “neon_mla_<V_elem_ch>_scalar_long”)] )
(define_insn “neon_vqdmlsl_lane” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:<V_widen> 1 “s_register_operand” “0”) (match_operand:VMDI 2 “s_register_operand” “w”) (match_operand:VMDI 3 “s_register_operand” “<scalar_mul_constraint>”) (match_operand:SI 4 “immediate_operand” “i”)] UNSPEC_VQDMLSL_LANE))] “TARGET_NEON” { neon_lane_bounds (operands[4], 0, GET_MODE_NUNITS (mode)); return “vqdmlsl.<V_s_elem>\t%q0, %P2, %P3[%c4]”; } [(set_attr “type” “neon_sat_mla_<V_elem_ch>_scalar_long”)] )
; FIXME: For the “_n” multiply/multiply-accumulate insns, we copy a value in a ; core register into a temp register, then use a scalar taken from that. This ; isn‘t an optimal solution if e.g. the scalar has just been read from memory ; or extracted from another vector. The latter case it’s currently better to ; use the “_lane” variant, and the former case can probably be implemented ; using vld1_lane, but that hasn't been done yet.
(define_expand “neon_vmul_n” [(match_operand:VMD 0 “s_register_operand” "") (match_operand:VMD 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vmul_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmul_n” [(match_operand:VMQ 0 “s_register_operand” "") (match_operand:VMQ 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (<V_HALF>mode); emit_insn (gen_neon_vset_lane<V_half> (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vmul_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmulls_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:VMDI 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vmulls_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmullu_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:VMDI 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vmullu_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vqdmull_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:VMDI 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vqdmull_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vqdmulh_n” [(match_operand:VMDI 0 “s_register_operand” "") (match_operand:VMDI 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vqdmulh_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vqrdmulh_n” [(match_operand:VMDI 0 “s_register_operand” "") (match_operand:VMDI 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vqrdmulh_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vqdmulh_n” [(match_operand:VMQI 0 “s_register_operand” "") (match_operand:VMQI 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (<V_HALF>mode); emit_insn (gen_neon_vset_lane<V_half> (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vqdmulh_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vqrdmulh_n” [(match_operand:VMQI 0 “s_register_operand” "") (match_operand:VMQI 1 “s_register_operand” "") (match_operand:<V_elem> 2 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (<V_HALF>mode); emit_insn (gen_neon_vset_lane<V_half> (tmp, operands[2], tmp, const0_rtx)); emit_insn (gen_neon_vqrdmulh_lane (operands[0], operands[1], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmla_n” [(match_operand:VMD 0 “s_register_operand” "") (match_operand:VMD 1 “s_register_operand” "") (match_operand:VMD 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmla_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmla_n” [(match_operand:VMQ 0 “s_register_operand” "") (match_operand:VMQ 1 “s_register_operand” "") (match_operand:VMQ 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (<V_HALF>mode); emit_insn (gen_neon_vset_lane<V_half> (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmla_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmlals_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:<V_widen> 1 “s_register_operand” "") (match_operand:VMDI 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmlals_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmlalu_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:<V_widen> 1 “s_register_operand” "") (match_operand:VMDI 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmlalu_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vqdmlal_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:<V_widen> 1 “s_register_operand” "") (match_operand:VMDI 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vqdmlal_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmls_n” [(match_operand:VMD 0 “s_register_operand” "") (match_operand:VMD 1 “s_register_operand” "") (match_operand:VMD 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmls_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmls_n” [(match_operand:VMQ 0 “s_register_operand” "") (match_operand:VMQ 1 “s_register_operand” "") (match_operand:VMQ 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (<V_HALF>mode); emit_insn (gen_neon_vset_lane<V_half> (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmls_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmlsls_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:<V_widen> 1 “s_register_operand” "") (match_operand:VMDI 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmlsls_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vmlslu_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:<V_widen> 1 “s_register_operand” "") (match_operand:VMDI 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vmlslu_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_expand “neon_vqdmlsl_n” [(match_operand:<V_widen> 0 “s_register_operand” "") (match_operand:<V_widen> 1 “s_register_operand” "") (match_operand:VMDI 2 “s_register_operand” "") (match_operand:<V_elem> 3 “s_register_operand” "")] “TARGET_NEON” { rtx tmp = gen_reg_rtx (mode); emit_insn (gen_neon_vset_lane (tmp, operands[3], tmp, const0_rtx)); emit_insn (gen_neon_vqdmlsl_lane (operands[0], operands[1], operands[2], tmp, const0_rtx)); DONE; })
(define_insn “neon_vext” [(set (match_operand:VDQX 0 “s_register_operand” “=w”) (unspec:VDQX [(match_operand:VDQX 1 “s_register_operand” “w”) (match_operand:VDQX 2 “s_register_operand” “w”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VEXT))] “TARGET_NEON” { neon_const_bounds (operands[3], 0, GET_MODE_NUNITS (mode)); return “vext.<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2, %3”; } [(set_attr “type” “neon_ext”)] )
(define_insn “neon_vrev64” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (unspec:VDQ [(match_operand:VDQ 1 “s_register_operand” “w”)] UNSPEC_VREV64))] “TARGET_NEON” “vrev64.<V_sz_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_rev”)] )
(define_insn “neon_vrev32” [(set (match_operand:VX 0 “s_register_operand” “=w”) (unspec:VX [(match_operand:VX 1 “s_register_operand” “w”)] UNSPEC_VREV32))] “TARGET_NEON” “vrev32.<V_sz_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_rev”)] )
(define_insn “neon_vrev16” [(set (match_operand:VE 0 “s_register_operand” “=w”) (unspec:VE [(match_operand:VE 1 “s_register_operand” “w”)] UNSPEC_VREV16))] “TARGET_NEON” “vrev16.<V_sz_elem>\t%<V_reg>0, %<V_reg>1” [(set_attr “type” “neon_rev”)] )
; vbsl_* intrinsics may compile to any of vbsl/vbif/vbit depending on register ; allocation. For an intrinsic of form: ; rD = vbsl_* (rS, rN, rM) ; We can use any of: ; vbsl rS, rN, rM (if D = S) ; vbit rD, rN, rS (if D = M, so 1-bits in rS choose bits from rN, else rM) ; vbif rD, rM, rS (if D = N, so 0-bits in rS choose bits from rM, else rN)
(define_insn “neon_vbsl_internal” [(set (match_operand:VDQX 0 “s_register_operand” “=w,w,w”) (unspec:VDQX [(match_operand:VDQX 1 “s_register_operand” " 0,w,w") (match_operand:VDQX 2 “s_register_operand” " w,w,0") (match_operand:VDQX 3 “s_register_operand” " w,0,w")] UNSPEC_VBSL))] “TARGET_NEON” “@ vbsl\t%<V_reg>0, %<V_reg>2, %<V_reg>3 vbit\t%<V_reg>0, %<V_reg>2, %<V_reg>1 vbif\t%<V_reg>0, %<V_reg>3, %<V_reg>1” [(set_attr “type” “neon_bsl”)] )
(define_expand “neon_vbsl” [(set (match_operand:VDQX 0 “s_register_operand” "") (unspec:VDQX [(match_operand:<V_cmp_result> 1 “s_register_operand” "") (match_operand:VDQX 2 “s_register_operand” "") (match_operand:VDQX 3 “s_register_operand” "")] UNSPEC_VBSL))] “TARGET_NEON” { /* We can't alias operands together if they have different modes. */ operands[1] = gen_lowpart (mode, operands[1]); })
;; vshl, vrshl (define_insn “neon_v<shift_op>” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:VDQIX 2 “s_register_operand” “w”)] VSHL))] “TARGET_NEON” “v<shift_op>.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_shift_imm”)] )
;; vqshl, vqrshl (define_insn “neon_v<shift_op>” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:VDQIX 2 “s_register_operand” “w”)] VQSHL))] “TARGET_NEON” “v<shift_op>.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2” [(set_attr “type” “neon_sat_shift_imm”)] )
;; vshr_n, vrshr_n (define_insn “neon_v<shift_op>_n” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VSHR_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 1, neon_element_bits (mode) + 1); return “v<shift_op>.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %2”; } [(set_attr “type” “neon_shift_imm”)] )
;; vshrn_n, vrshrn_n (define_insn “neon_v<shift_op>_n” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VSHRN_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 1, neon_element_bits (mode) / 2 + 1); return “v<shift_op>.<V_if_elem>\t%P0, %q1, %2”; } [(set_attr “type” “neon_shift_imm_narrow_q”)] )
;; vqshrn_n, vqrshrn_n (define_insn “neon_v<shift_op>_n” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VQSHRN_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 1, neon_element_bits (mode) / 2 + 1); return “v<shift_op>.%#<V_sz_elem>\t%P0, %q1, %2”; } [(set_attr “type” “neon_sat_shift_imm_narrow_q”)] )
;; vqshrun_n, vqrshrun_n (define_insn “neon_v<shift_op>_n” [(set (match_operand:<V_narrow> 0 “s_register_operand” “=w”) (unspec:<V_narrow> [(match_operand:VN 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VQSHRUN_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 1, neon_element_bits (mode) / 2 + 1); return “v<shift_op>.<V_s_elem>\t%P0, %q1, %2”; } [(set_attr “type” “neon_sat_shift_imm_narrow_q”)] )
(define_insn “neon_vshl_n” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] UNSPEC_VSHL_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 0, neon_element_bits (mode)); return “vshl.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %2”; } [(set_attr “type” “neon_shift_imm”)] )
(define_insn “neon_vqshl__n” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VQSHL_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 0, neon_element_bits (mode)); return “vqshl.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>1, %2”; } [(set_attr “type” “neon_sat_shift_imm”)] )
(define_insn “neon_vqshlu_n” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] UNSPEC_VQSHLU_N))] “TARGET_NEON” { neon_const_bounds (operands[2], 0, neon_element_bits (mode)); return “vqshlu.<V_s_elem>\t%<V_reg>0, %<V_reg>1, %2”; } [(set_attr “type” “neon_sat_shift_imm”)] )
(define_insn “neon_vshll_n” [(set (match_operand:<V_widen> 0 “s_register_operand” “=w”) (unspec:<V_widen> [(match_operand:VW 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] VSHLL_N))] “TARGET_NEON” { /* The boundaries are: 0 < imm <= size. */ neon_const_bounds (operands[2], 0, neon_element_bits (mode) + 1); return “vshll.%#<V_sz_elem>\t%q0, %P1, %2”; } [(set_attr “type” “neon_shift_imm_long”)] )
;; vsra_n, vrsra_n (define_insn “neon_v<shift_op>_n” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “0”) (match_operand:VDQIX 2 “s_register_operand” “w”) (match_operand:SI 3 “immediate_operand” “i”)] VSRA_N))] “TARGET_NEON” { neon_const_bounds (operands[3], 1, neon_element_bits (mode) + 1); return “v<shift_op>.%#<V_sz_elem>\t%<V_reg>0, %<V_reg>2, %3”; } [(set_attr “type” “neon_shift_acc”)] )
(define_insn “neon_vsri_n” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “0”) (match_operand:VDQIX 2 “s_register_operand” “w”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VSRI))] “TARGET_NEON” { neon_const_bounds (operands[3], 1, neon_element_bits (mode) + 1); return “vsri.<V_sz_elem>\t%<V_reg>0, %<V_reg>2, %3”; } [(set_attr “type” “neon_shift_reg”)] )
(define_insn “neon_vsli_n” [(set (match_operand:VDQIX 0 “s_register_operand” “=w”) (unspec:VDQIX [(match_operand:VDQIX 1 “s_register_operand” “0”) (match_operand:VDQIX 2 “s_register_operand” “w”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VSLI))] “TARGET_NEON” { neon_const_bounds (operands[3], 0, neon_element_bits (mode)); return “vsli.<V_sz_elem>\t%<V_reg>0, %<V_reg>2, %3”; } [(set_attr “type” “neon_shift_reg”)] )
(define_insn “neon_vtbl1v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:V8QI 1 “s_register_operand” “w”) (match_operand:V8QI 2 “s_register_operand” “w”)] UNSPEC_VTBL))] “TARGET_NEON” “vtbl.8\t%P0, {%P1}, %P2” [(set_attr “type” “neon_tbl1”)] )
(define_insn “neon_vtbl2v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:TI 1 “s_register_operand” “w”) (match_operand:V8QI 2 “s_register_operand” “w”)] UNSPEC_VTBL))] “TARGET_NEON” { rtx ops[4]; int tabbase = REGNO (operands[1]);
ops[0] = operands[0]; ops[1] = gen_rtx_REG (V8QImode, tabbase); ops[2] = gen_rtx_REG (V8QImode, tabbase + 2); ops[3] = operands[2]; output_asm_insn (“vtbl.8\t%P0, {%P1, %P2}, %P3”, ops);
return ""; } [(set_attr “type” “neon_tbl2”)] )
(define_insn “neon_vtbl3v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:EI 1 “s_register_operand” “w”) (match_operand:V8QI 2 “s_register_operand” “w”)] UNSPEC_VTBL))] “TARGET_NEON” { rtx ops[5]; int tabbase = REGNO (operands[1]);
ops[0] = operands[0]; ops[1] = gen_rtx_REG (V8QImode, tabbase); ops[2] = gen_rtx_REG (V8QImode, tabbase + 2); ops[3] = gen_rtx_REG (V8QImode, tabbase + 4); ops[4] = operands[2]; output_asm_insn (“vtbl.8\t%P0, {%P1, %P2, %P3}, %P4”, ops);
return ""; } [(set_attr “type” “neon_tbl3”)] )
(define_insn “neon_vtbl4v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:OI 1 “s_register_operand” “w”) (match_operand:V8QI 2 “s_register_operand” “w”)] UNSPEC_VTBL))] “TARGET_NEON” { rtx ops[6]; int tabbase = REGNO (operands[1]);
ops[0] = operands[0]; ops[1] = gen_rtx_REG (V8QImode, tabbase); ops[2] = gen_rtx_REG (V8QImode, tabbase + 2); ops[3] = gen_rtx_REG (V8QImode, tabbase + 4); ops[4] = gen_rtx_REG (V8QImode, tabbase + 6); ops[5] = operands[2]; output_asm_insn (“vtbl.8\t%P0, {%P1, %P2, %P3, %P4}, %P5”, ops);
return ""; } [(set_attr “type” “neon_tbl4”)] )
;; These three are used by the vec_perm infrastructure for V16QImode. (define_insn_and_split “neon_vtbl1v16qi” [(set (match_operand:V16QI 0 “s_register_operand” “=&w”) (unspec:V16QI [(match_operand:V16QI 1 “s_register_operand” “w”) (match_operand:V16QI 2 “s_register_operand” “w”)] UNSPEC_VTBL))] “TARGET_NEON” “#” “&& reload_completed” [(const_int 0)] { rtx op0, op1, op2, part0, part2; unsigned ofs;
op0 = operands[0]; op1 = gen_lowpart (TImode, operands[1]); op2 = operands[2];
ofs = subreg_lowpart_offset (V8QImode, V16QImode); part0 = simplify_subreg (V8QImode, op0, V16QImode, ofs); part2 = simplify_subreg (V8QImode, op2, V16QImode, ofs); emit_insn (gen_neon_vtbl2v8qi (part0, op1, part2));
ofs = subreg_highpart_offset (V8QImode, V16QImode); part0 = simplify_subreg (V8QImode, op0, V16QImode, ofs); part2 = simplify_subreg (V8QImode, op2, V16QImode, ofs); emit_insn (gen_neon_vtbl2v8qi (part0, op1, part2)); DONE; } [(set_attr “type” “multiple”)] )
(define_insn_and_split “neon_vtbl2v16qi” [(set (match_operand:V16QI 0 “s_register_operand” “=&w”) (unspec:V16QI [(match_operand:OI 1 “s_register_operand” “w”) (match_operand:V16QI 2 “s_register_operand” “w”)] UNSPEC_VTBL))] “TARGET_NEON” “#” “&& reload_completed” [(const_int 0)] { rtx op0, op1, op2, part0, part2; unsigned ofs;
op0 = operands[0]; op1 = operands[1]; op2 = operands[2];
ofs = subreg_lowpart_offset (V8QImode, V16QImode); part0 = simplify_subreg (V8QImode, op0, V16QImode, ofs); part2 = simplify_subreg (V8QImode, op2, V16QImode, ofs); emit_insn (gen_neon_vtbl2v8qi (part0, op1, part2));
ofs = subreg_highpart_offset (V8QImode, V16QImode); part0 = simplify_subreg (V8QImode, op0, V16QImode, ofs); part2 = simplify_subreg (V8QImode, op2, V16QImode, ofs); emit_insn (gen_neon_vtbl2v8qi (part0, op1, part2)); DONE; } [(set_attr “type” “multiple”)] )
;; ??? Logically we should extend the regular neon_vcombine pattern to ;; handle quad-word input modes, producing octa-word output modes. But ;; that requires us to add support for octa-word vector modes in moves. ;; That seems overkill for this one use in vec_perm. (define_insn_and_split “neon_vcombinev16qi” [(set (match_operand:OI 0 “s_register_operand” “=w”) (unspec:OI [(match_operand:V16QI 1 “s_register_operand” “w”) (match_operand:V16QI 2 “s_register_operand” “w”)] UNSPEC_VCONCAT))] “TARGET_NEON” “#” “&& reload_completed” [(const_int 0)] { neon_split_vcombine (operands); DONE; } [(set_attr “type” “multiple”)] )
(define_insn “neon_vtbx1v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:V8QI 1 “s_register_operand” “0”) (match_operand:V8QI 2 “s_register_operand” “w”) (match_operand:V8QI 3 “s_register_operand” “w”)] UNSPEC_VTBX))] “TARGET_NEON” “vtbx.8\t%P0, {%P2}, %P3” [(set_attr “type” “neon_tbl1”)] )
(define_insn “neon_vtbx2v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:V8QI 1 “s_register_operand” “0”) (match_operand:TI 2 “s_register_operand” “w”) (match_operand:V8QI 3 “s_register_operand” “w”)] UNSPEC_VTBX))] “TARGET_NEON” { rtx ops[4]; int tabbase = REGNO (operands[2]);
ops[0] = operands[0]; ops[1] = gen_rtx_REG (V8QImode, tabbase); ops[2] = gen_rtx_REG (V8QImode, tabbase + 2); ops[3] = operands[3]; output_asm_insn (“vtbx.8\t%P0, {%P1, %P2}, %P3”, ops);
return ""; } [(set_attr “type” “neon_tbl2”)] )
(define_insn “neon_vtbx3v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:V8QI 1 “s_register_operand” “0”) (match_operand:EI 2 “s_register_operand” “w”) (match_operand:V8QI 3 “s_register_operand” “w”)] UNSPEC_VTBX))] “TARGET_NEON” { rtx ops[5]; int tabbase = REGNO (operands[2]);
ops[0] = operands[0]; ops[1] = gen_rtx_REG (V8QImode, tabbase); ops[2] = gen_rtx_REG (V8QImode, tabbase + 2); ops[3] = gen_rtx_REG (V8QImode, tabbase + 4); ops[4] = operands[3]; output_asm_insn (“vtbx.8\t%P0, {%P1, %P2, %P3}, %P4”, ops);
return ""; } [(set_attr “type” “neon_tbl3”)] )
(define_insn “neon_vtbx4v8qi” [(set (match_operand:V8QI 0 “s_register_operand” “=w”) (unspec:V8QI [(match_operand:V8QI 1 “s_register_operand” “0”) (match_operand:OI 2 “s_register_operand” “w”) (match_operand:V8QI 3 “s_register_operand” “w”)] UNSPEC_VTBX))] “TARGET_NEON” { rtx ops[6]; int tabbase = REGNO (operands[2]);
ops[0] = operands[0]; ops[1] = gen_rtx_REG (V8QImode, tabbase); ops[2] = gen_rtx_REG (V8QImode, tabbase + 2); ops[3] = gen_rtx_REG (V8QImode, tabbase + 4); ops[4] = gen_rtx_REG (V8QImode, tabbase + 6); ops[5] = operands[3]; output_asm_insn (“vtbx.8\t%P0, {%P1, %P2, %P3, %P4}, %P5”, ops);
return ""; } [(set_attr “type” “neon_tbl4”)] )
(define_expand “neon_vtrn_internal” [(parallel [(set (match_operand:VDQW 0 “s_register_operand” "") (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” "") (match_operand:VDQW 2 “s_register_operand” "")] UNSPEC_VTRN1)) (set (match_operand:VDQW 3 “s_register_operand” "") (unspec:VDQW [(match_dup 1) (match_dup 2)] UNSPEC_VTRN2))])] “TARGET_NEON” "" )
;; Note: Different operand numbering to handle tied registers correctly. (define_insn “*neon_vtrn_insn” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” “0”) (match_operand:VDQW 3 “s_register_operand” “2”)] UNSPEC_VTRN1)) (set (match_operand:VDQW 2 “s_register_operand” “=w”) (unspec:VDQW [(match_dup 1) (match_dup 3)] UNSPEC_VTRN2))] “TARGET_NEON” “vtrn.<V_sz_elem>\t%<V_reg>0, %<V_reg>2” [(set_attr “type” “neon_permute”)] )
(define_expand “neon_vzip_internal” [(parallel [(set (match_operand:VDQW 0 “s_register_operand” "") (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” "") (match_operand:VDQW 2 “s_register_operand” "")] UNSPEC_VZIP1)) (set (match_operand:VDQW 3 “s_register_operand” "") (unspec:VDQW [(match_dup 1) (match_dup 2)] UNSPEC_VZIP2))])] “TARGET_NEON” "" )
;; Note: Different operand numbering to handle tied registers correctly. (define_insn “*neon_vzip_insn” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” “0”) (match_operand:VDQW 3 “s_register_operand” “2”)] UNSPEC_VZIP1)) (set (match_operand:VDQW 2 “s_register_operand” “=w”) (unspec:VDQW [(match_dup 1) (match_dup 3)] UNSPEC_VZIP2))] “TARGET_NEON” “vzip.<V_sz_elem>\t%<V_reg>0, %<V_reg>2” [(set_attr “type” “neon_zip”)] )
(define_expand “neon_vuzp_internal” [(parallel [(set (match_operand:VDQW 0 “s_register_operand” "") (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” "") (match_operand:VDQW 2 “s_register_operand” "")] UNSPEC_VUZP1)) (set (match_operand:VDQW 3 “s_register_operand” "") (unspec:VDQW [(match_dup 1) (match_dup 2)] UNSPEC_VUZP2))])] “TARGET_NEON” "" )
;; Note: Different operand numbering to handle tied registers correctly. (define_insn “*neon_vuzp_insn” [(set (match_operand:VDQW 0 “s_register_operand” “=w”) (unspec:VDQW [(match_operand:VDQW 1 “s_register_operand” “0”) (match_operand:VDQW 3 “s_register_operand” “2”)] UNSPEC_VUZP1)) (set (match_operand:VDQW 2 “s_register_operand” “=w”) (unspec:VDQW [(match_dup 1) (match_dup 3)] UNSPEC_VUZP2))] “TARGET_NEON” “vuzp.<V_sz_elem>\t%<V_reg>0, %<V_reg>2” [(set_attr “type” “neon_zip”)] )
(define_expand “neon_vreinterpretv8qi” [(match_operand:V8QI 0 “s_register_operand” "") (match_operand:VDX 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv4hi” [(match_operand:V4HI 0 “s_register_operand” "") (match_operand:VDX 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv2si” [(match_operand:V2SI 0 “s_register_operand” "") (match_operand:VDX 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv2sf” [(match_operand:V2SF 0 “s_register_operand” "") (match_operand:VDX 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretdi” [(match_operand:DI 0 “s_register_operand” "") (match_operand:VDX 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretti” [(match_operand:TI 0 “s_register_operand” "") (match_operand:VQXMOV 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv16qi” [(match_operand:V16QI 0 “s_register_operand” "") (match_operand:VQXMOV 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv8hi” [(match_operand:V8HI 0 “s_register_operand” "") (match_operand:VQXMOV 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv4si” [(match_operand:V4SI 0 “s_register_operand” "") (match_operand:VQXMOV 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv4sf” [(match_operand:V4SF 0 “s_register_operand” "") (match_operand:VQXMOV 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “neon_vreinterpretv2di” [(match_operand:V2DI 0 “s_register_operand” "") (match_operand:VQXMOV 1 “s_register_operand” "")] “TARGET_NEON” { neon_reinterpret (operands[0], operands[1]); DONE; })
(define_expand “vec_load_lanes” [(set (match_operand:VDQX 0 “s_register_operand”) (unspec:VDQX [(match_operand:VDQX 1 “neon_struct_operand”)] UNSPEC_VLD1))] “TARGET_NEON”)
(define_insn “neon_vld1” [(set (match_operand:VDQX 0 “s_register_operand” “=w”) (unspec:VDQX [(match_operand:VDQX 1 “neon_struct_operand” “Um”)] UNSPEC_VLD1))] “TARGET_NEON” “vld1.<V_sz_elem>\t%h0, %A1” [(set_attr “type” “neon_load1_1reg”)] )
(define_insn “neon_vld1_lane” [(set (match_operand:VDX 0 “s_register_operand” “=w”) (unspec:VDX [(match_operand:<V_elem> 1 “neon_struct_operand” “Um”) (match_operand:VDX 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VLD1_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); if (lane < 0 || lane >= max) error (“lane out of range”); if (max == 1) return “vld1.<V_sz_elem>\t%P0, %A1”; else return “vld1.<V_sz_elem>\t{%P0[%c3]}, %A1”; } [(set_attr “type” “neon_load1_one_lane”)] )
(define_insn “neon_vld1_lane” [(set (match_operand:VQX 0 “s_register_operand” “=w”) (unspec:VQX [(match_operand:<V_elem> 1 “neon_struct_operand” “Um”) (match_operand:VQX 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”)] UNSPEC_VLD1_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[0]); if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; operands[3] = GEN_INT (lane); } operands[0] = gen_rtx_REG (<V_HALF>mode, regno); if (max == 2) return “vld1.<V_sz_elem>\t%P0, %A1”; else return “vld1.<V_sz_elem>\t{%P0[%c3]}, %A1”; } [(set_attr “type” “neon_load1_one_lane”)] )
(define_insn “neon_vld1_dup” [(set (match_operand:VD 0 “s_register_operand” “=w”) (vec_duplicate:VD (match_operand:<V_elem> 1 “neon_struct_operand” “Um”)))] “TARGET_NEON” “vld1.<V_sz_elem>\t{%P0[]}, %A1” [(set_attr “type” “neon_load1_all_lanes”)] )
;; Special case for DImode. Treat it exactly like a simple load. (define_expand “neon_vld1_dupdi” [(set (match_operand:DI 0 “s_register_operand” "") (unspec:DI [(match_operand:DI 1 “neon_struct_operand” "")] UNSPEC_VLD1))] “TARGET_NEON” "" )
(define_insn “neon_vld1_dup” [(set (match_operand:VQ 0 “s_register_operand” “=w”) (vec_duplicate:VQ (match_operand:<V_elem> 1 “neon_struct_operand” “Um”)))] “TARGET_NEON” { return “vld1.<V_sz_elem>\t{%e0[], %f0[]}, %A1”; } [(set_attr “type” “neon_load1_all_lanes”)] )
(define_insn_and_split “neon_vld1_dupv2di” [(set (match_operand:V2DI 0 “s_register_operand” “=w”) (vec_duplicate:V2DI (match_operand:DI 1 “neon_struct_operand” “Um”)))] “TARGET_NEON” “#” “&& reload_completed” [(const_int 0)] { rtx tmprtx = gen_lowpart (DImode, operands[0]); emit_insn (gen_neon_vld1_dupdi (tmprtx, operands[1])); emit_move_insn (gen_highpart (DImode, operands[0]), tmprtx ); DONE; } [(set_attr “length” “8”) (set_attr “type” “neon_load1_all_lanes_q”)] )
(define_expand “vec_store_lanes” [(set (match_operand:VDQX 0 “neon_struct_operand”) (unspec:VDQX [(match_operand:VDQX 1 “s_register_operand”)] UNSPEC_VST1))] “TARGET_NEON”)
(define_insn “neon_vst1” [(set (match_operand:VDQX 0 “neon_struct_operand” “=Um”) (unspec:VDQX [(match_operand:VDQX 1 “s_register_operand” “w”)] UNSPEC_VST1))] “TARGET_NEON” “vst1.<V_sz_elem>\t%h1, %A0” [(set_attr “type” “neon_store1_1reg”)])
(define_insn “neon_vst1_lane” [(set (match_operand:<V_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_elem> [(match_operand:VDX 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] UNSPEC_VST1_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); if (lane < 0 || lane >= max) error (“lane out of range”); if (max == 1) return “vst1.<V_sz_elem>\t{%P1}, %A0”; else return “vst1.<V_sz_elem>\t{%P1[%c2]}, %A0”; } [(set_attr “type” “neon_store1_one_lane”)] )
(define_insn “neon_vst1_lane” [(set (match_operand:<V_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_elem> [(match_operand:VQX 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”)] UNSPEC_VST1_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[1]); if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; operands[2] = GEN_INT (lane); } operands[1] = gen_rtx_REG (<V_HALF>mode, regno); if (max == 2) return “vst1.<V_sz_elem>\t{%P1}, %A0”; else return “vst1.<V_sz_elem>\t{%P1[%c2]}, %A0”; } [(set_attr “type” “neon_store1_one_lane”)] )
(define_expand “vec_load_lanesti” [(set (match_operand:TI 0 “s_register_operand”) (unspec:TI [(match_operand:TI 1 “neon_struct_operand”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD2))] “TARGET_NEON”)
(define_insn “neon_vld2” [(set (match_operand:TI 0 “s_register_operand” “=w”) (unspec:TI [(match_operand:TI 1 “neon_struct_operand” “Um”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD2))] “TARGET_NEON” { if (<V_sz_elem> == 64) return “vld1.64\t%h0, %A1”; else return “vld2.<V_sz_elem>\t%h0, %A1”; } [(set (attr “type”) (if_then_else (eq (const_string “<V_sz_elem>”) (const_string “64”)) (const_string “neon_load1_2reg”) (const_string “neon_load2_2reg”)))] )
(define_expand “vec_load_lanesoi” [(set (match_operand:OI 0 “s_register_operand”) (unspec:OI [(match_operand:OI 1 “neon_struct_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD2))] “TARGET_NEON”)
(define_insn “neon_vld2” [(set (match_operand:OI 0 “s_register_operand” “=w”) (unspec:OI [(match_operand:OI 1 “neon_struct_operand” “Um”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD2))] “TARGET_NEON” “vld2.<V_sz_elem>\t%h0, %A1” [(set_attr “type” “neon_load2_2reg_q”)])
(define_insn “neon_vld2_lane” [(set (match_operand:TI 0 “s_register_operand” “=w”) (unspec:TI [(match_operand:<V_two_elem> 1 “neon_struct_operand” “Um”) (match_operand:TI 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”) (unspec:VD [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD2_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[0]); rtx ops[4]; if (lane < 0 || lane >= max) error (“lane out of range”); ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 2); ops[2] = operands[1]; ops[3] = operands[3]; output_asm_insn (“vld2.<V_sz_elem>\t{%P0[%c3], %P1[%c3]}, %A2”, ops); return ""; } [(set_attr “type” “neon_load2_one_lane”)] )
(define_insn “neon_vld2_lane” [(set (match_operand:OI 0 “s_register_operand” “=w”) (unspec:OI [(match_operand:<V_two_elem> 1 “neon_struct_operand” “Um”) (match_operand:OI 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”) (unspec:VMQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD2_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[0]); rtx ops[4]; if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; } ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 4); ops[2] = operands[1]; ops[3] = GEN_INT (lane); output_asm_insn (“vld2.<V_sz_elem>\t{%P0[%c3], %P1[%c3]}, %A2”, ops); return ""; } [(set_attr “type” “neon_load2_one_lane”)] )
(define_insn “neon_vld2_dup” [(set (match_operand:TI 0 “s_register_operand” “=w”) (unspec:TI [(match_operand:<V_two_elem> 1 “neon_struct_operand” “Um”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD2_DUP))] “TARGET_NEON” { if (GET_MODE_NUNITS (mode) > 1) return “vld2.<V_sz_elem>\t{%e0[], %f0[]}, %A1”; else return “vld1.<V_sz_elem>\t%h0, %A1”; } [(set (attr “type”) (if_then_else (gt (const_string “<V_mode_nunits>”) (const_string “1”)) (const_string “neon_load2_all_lanes”) (const_string “neon_load1_1reg”)))] )
(define_expand “vec_store_lanesti” [(set (match_operand:TI 0 “neon_struct_operand”) (unspec:TI [(match_operand:TI 1 “s_register_operand”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST2))] “TARGET_NEON”)
(define_insn “neon_vst2” [(set (match_operand:TI 0 “neon_struct_operand” “=Um”) (unspec:TI [(match_operand:TI 1 “s_register_operand” “w”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST2))] “TARGET_NEON” { if (<V_sz_elem> == 64) return “vst1.64\t%h1, %A0”; else return “vst2.<V_sz_elem>\t%h1, %A0”; } [(set (attr “type”) (if_then_else (eq (const_string “<V_sz_elem>”) (const_string “64”)) (const_string “neon_store1_2reg”) (const_string “neon_store2_one_lane”)))] )
(define_expand “vec_store_lanesoi” [(set (match_operand:OI 0 “neon_struct_operand”) (unspec:OI [(match_operand:OI 1 “s_register_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST2))] “TARGET_NEON”)
(define_insn “neon_vst2” [(set (match_operand:OI 0 “neon_struct_operand” “=Um”) (unspec:OI [(match_operand:OI 1 “s_register_operand” “w”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST2))] “TARGET_NEON” “vst2.<V_sz_elem>\t%h1, %A0” [(set_attr “type” “neon_store2_4reg”)] )
(define_insn “neon_vst2_lane” [(set (match_operand:<V_two_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_two_elem> [(match_operand:TI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”) (unspec:VD [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST2_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[1]); rtx ops[4]; if (lane < 0 || lane >= max) error (“lane out of range”); ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 2); ops[3] = operands[2]; output_asm_insn (“vst2.<V_sz_elem>\t{%P1[%c3], %P2[%c3]}, %A0”, ops); return ""; } [(set_attr “type” “neon_store2_one_lane”)] )
(define_insn “neon_vst2_lane” [(set (match_operand:<V_two_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_two_elem> [(match_operand:OI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”) (unspec:VMQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST2_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[1]); rtx ops[4]; if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; } ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = GEN_INT (lane); output_asm_insn (“vst2.<V_sz_elem>\t{%P1[%c3], %P2[%c3]}, %A0”, ops); return ""; } [(set_attr “type” “neon_store2_one_lane”)] )
(define_expand “vec_load_lanesei” [(set (match_operand:EI 0 “s_register_operand”) (unspec:EI [(match_operand:EI 1 “neon_struct_operand”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD3))] “TARGET_NEON”)
(define_insn “neon_vld3” [(set (match_operand:EI 0 “s_register_operand” “=w”) (unspec:EI [(match_operand:EI 1 “neon_struct_operand” “Um”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD3))] “TARGET_NEON” { if (<V_sz_elem> == 64) return “vld1.64\t%h0, %A1”; else return “vld3.<V_sz_elem>\t%h0, %A1”; } [(set (attr “type”) (if_then_else (eq (const_string “<V_sz_elem>”) (const_string “64”)) (const_string “neon_load1_3reg”) (const_string “neon_load3_3reg”)))] )
(define_expand “vec_load_lanesci” [(match_operand:CI 0 “s_register_operand”) (match_operand:CI 1 “neon_struct_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { emit_insn (gen_neon_vld3 (operands[0], operands[1])); DONE; })
(define_expand “neon_vld3” [(match_operand:CI 0 “s_register_operand”) (match_operand:CI 1 “neon_struct_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { rtx mem;
mem = adjust_address (operands[1], EImode, 0); emit_insn (gen_neon_vld3qa (operands[0], mem)); mem = adjust_address (mem, EImode, GET_MODE_SIZE (EImode)); emit_insn (gen_neon_vld3qb (operands[0], mem, operands[0])); DONE; })
(define_insn “neon_vld3qa” [(set (match_operand:CI 0 “s_register_operand” “=w”) (unspec:CI [(match_operand:EI 1 “neon_struct_operand” “Um”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD3A))] “TARGET_NEON” { int regno = REGNO (operands[0]); rtx ops[4]; ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 4); ops[2] = gen_rtx_REG (DImode, regno + 8); ops[3] = operands[1]; output_asm_insn (“vld3.<V_sz_elem>\t{%P0, %P1, %P2}, %A3”, ops); return ""; } [(set_attr “type” “neon_load3_3reg”)] )
(define_insn “neon_vld3qb” [(set (match_operand:CI 0 “s_register_operand” “=w”) (unspec:CI [(match_operand:EI 1 “neon_struct_operand” “Um”) (match_operand:CI 2 “s_register_operand” “0”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD3B))] “TARGET_NEON” { int regno = REGNO (operands[0]); rtx ops[4]; ops[0] = gen_rtx_REG (DImode, regno + 2); ops[1] = gen_rtx_REG (DImode, regno + 6); ops[2] = gen_rtx_REG (DImode, regno + 10); ops[3] = operands[1]; output_asm_insn (“vld3.<V_sz_elem>\t{%P0, %P1, %P2}, %A3”, ops); return ""; } [(set_attr “type” “neon_load3_3reg”)] )
(define_insn “neon_vld3_lane” [(set (match_operand:EI 0 “s_register_operand” “=w”) (unspec:EI [(match_operand:<V_three_elem> 1 “neon_struct_operand” “Um”) (match_operand:EI 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”) (unspec:VD [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD3_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[0]); rtx ops[5]; if (lane < 0 || lane >= max) error (“lane out of range”); ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 2); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = operands[1]; ops[4] = operands[3]; output_asm_insn (“vld3.<V_sz_elem>\t{%P0[%c4], %P1[%c4], %P2[%c4]}, %3”, ops); return ""; } [(set_attr “type” “neon_load3_one_lane”)] )
(define_insn “neon_vld3_lane” [(set (match_operand:CI 0 “s_register_operand” “=w”) (unspec:CI [(match_operand:<V_three_elem> 1 “neon_struct_operand” “Um”) (match_operand:CI 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”) (unspec:VMQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD3_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[0]); rtx ops[5]; if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; } ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 4); ops[2] = gen_rtx_REG (DImode, regno + 8); ops[3] = operands[1]; ops[4] = GEN_INT (lane); output_asm_insn (“vld3.<V_sz_elem>\t{%P0[%c4], %P1[%c4], %P2[%c4]}, %3”, ops); return ""; } [(set_attr “type” “neon_load3_one_lane”)] )
(define_insn “neon_vld3_dup” [(set (match_operand:EI 0 “s_register_operand” “=w”) (unspec:EI [(match_operand:<V_three_elem> 1 “neon_struct_operand” “Um”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD3_DUP))] “TARGET_NEON” { if (GET_MODE_NUNITS (mode) > 1) { int regno = REGNO (operands[0]); rtx ops[4]; ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 2); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = operands[1]; output_asm_insn (“vld3.<V_sz_elem>\t{%P0[], %P1[], %P2[]}, %3”, ops); return ""; } else return “vld1.<V_sz_elem>\t%h0, %A1”; } [(set (attr “type”) (if_then_else (gt (const_string “<V_mode_nunits>”) (const_string “1”)) (const_string “neon_load3_all_lanes”) (const_string “neon_load1_1reg”)))])
(define_expand “vec_store_lanesei” [(set (match_operand:EI 0 “neon_struct_operand”) (unspec:EI [(match_operand:EI 1 “s_register_operand”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST3))] “TARGET_NEON”)
(define_insn “neon_vst3” [(set (match_operand:EI 0 “neon_struct_operand” “=Um”) (unspec:EI [(match_operand:EI 1 “s_register_operand” “w”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST3))] “TARGET_NEON” { if (<V_sz_elem> == 64) return “vst1.64\t%h1, %A0”; else return “vst3.<V_sz_elem>\t%h1, %A0”; } [(set (attr “type”) (if_then_else (eq (const_string “<V_sz_elem>”) (const_string “64”)) (const_string “neon_store1_3reg”) (const_string “neon_store3_one_lane”)))])
(define_expand “vec_store_lanesci” [(match_operand:CI 0 “neon_struct_operand”) (match_operand:CI 1 “s_register_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { emit_insn (gen_neon_vst3 (operands[0], operands[1])); DONE; })
(define_expand “neon_vst3” [(match_operand:CI 0 “neon_struct_operand”) (match_operand:CI 1 “s_register_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { rtx mem;
mem = adjust_address (operands[0], EImode, 0); emit_insn (gen_neon_vst3qa (mem, operands[1])); mem = adjust_address (mem, EImode, GET_MODE_SIZE (EImode)); emit_insn (gen_neon_vst3qb (mem, operands[1])); DONE; })
(define_insn “neon_vst3qa” [(set (match_operand:EI 0 “neon_struct_operand” “=Um”) (unspec:EI [(match_operand:CI 1 “s_register_operand” “w”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST3A))] “TARGET_NEON” { int regno = REGNO (operands[1]); rtx ops[4]; ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = gen_rtx_REG (DImode, regno + 8); output_asm_insn (“vst3.<V_sz_elem>\t{%P1, %P2, %P3}, %A0”, ops); return ""; } [(set_attr “type” “neon_store3_3reg”)] )
(define_insn “neon_vst3qb” [(set (match_operand:EI 0 “neon_struct_operand” “=Um”) (unspec:EI [(match_operand:CI 1 “s_register_operand” “w”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST3B))] “TARGET_NEON” { int regno = REGNO (operands[1]); rtx ops[4]; ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno + 2); ops[2] = gen_rtx_REG (DImode, regno + 6); ops[3] = gen_rtx_REG (DImode, regno + 10); output_asm_insn (“vst3.<V_sz_elem>\t{%P1, %P2, %P3}, %A0”, ops); return ""; } [(set_attr “type” “neon_store3_3reg”)] )
(define_insn “neon_vst3_lane” [(set (match_operand:<V_three_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_three_elem> [(match_operand:EI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”) (unspec:VD [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST3_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[1]); rtx ops[5]; if (lane < 0 || lane >= max) error (“lane out of range”); ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 2); ops[3] = gen_rtx_REG (DImode, regno + 4); ops[4] = operands[2]; output_asm_insn (“vst3.<V_sz_elem>\t{%P1[%c4], %P2[%c4], %P3[%c4]}, %0”, ops); return ""; } [(set_attr “type” “neon_store3_one_lane”)] )
(define_insn “neon_vst3_lane” [(set (match_operand:<V_three_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_three_elem> [(match_operand:CI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”) (unspec:VMQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST3_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[1]); rtx ops[5]; if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; } ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = gen_rtx_REG (DImode, regno + 8); ops[4] = GEN_INT (lane); output_asm_insn (“vst3.<V_sz_elem>\t{%P1[%c4], %P2[%c4], %P3[%c4]}, %0”, ops); return ""; } [(set_attr “type” “neon_store3_one_lane”)] )
(define_expand “vec_load_lanesoi” [(set (match_operand:OI 0 “s_register_operand”) (unspec:OI [(match_operand:OI 1 “neon_struct_operand”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD4))] “TARGET_NEON”)
(define_insn “neon_vld4” [(set (match_operand:OI 0 “s_register_operand” “=w”) (unspec:OI [(match_operand:OI 1 “neon_struct_operand” “Um”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD4))] “TARGET_NEON” { if (<V_sz_elem> == 64) return “vld1.64\t%h0, %A1”; else return “vld4.<V_sz_elem>\t%h0, %A1”; } [(set (attr “type”) (if_then_else (eq (const_string “<V_sz_elem>”) (const_string “64”)) (const_string “neon_load1_4reg”) (const_string “neon_load4_4reg”)))] )
(define_expand “vec_load_lanesxi” [(match_operand:XI 0 “s_register_operand”) (match_operand:XI 1 “neon_struct_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { emit_insn (gen_neon_vld4 (operands[0], operands[1])); DONE; })
(define_expand “neon_vld4” [(match_operand:XI 0 “s_register_operand”) (match_operand:XI 1 “neon_struct_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { rtx mem;
mem = adjust_address (operands[1], OImode, 0); emit_insn (gen_neon_vld4qa (operands[0], mem)); mem = adjust_address (mem, OImode, GET_MODE_SIZE (OImode)); emit_insn (gen_neon_vld4qb (operands[0], mem, operands[0])); DONE; })
(define_insn “neon_vld4qa” [(set (match_operand:XI 0 “s_register_operand” “=w”) (unspec:XI [(match_operand:OI 1 “neon_struct_operand” “Um”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD4A))] “TARGET_NEON” { int regno = REGNO (operands[0]); rtx ops[5]; ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 4); ops[2] = gen_rtx_REG (DImode, regno + 8); ops[3] = gen_rtx_REG (DImode, regno + 12); ops[4] = operands[1]; output_asm_insn (“vld4.<V_sz_elem>\t{%P0, %P1, %P2, %P3}, %A4”, ops); return ""; } [(set_attr “type” “neon_load4_4reg”)] )
(define_insn “neon_vld4qb” [(set (match_operand:XI 0 “s_register_operand” “=w”) (unspec:XI [(match_operand:OI 1 “neon_struct_operand” “Um”) (match_operand:XI 2 “s_register_operand” “0”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD4B))] “TARGET_NEON” { int regno = REGNO (operands[0]); rtx ops[5]; ops[0] = gen_rtx_REG (DImode, regno + 2); ops[1] = gen_rtx_REG (DImode, regno + 6); ops[2] = gen_rtx_REG (DImode, regno + 10); ops[3] = gen_rtx_REG (DImode, regno + 14); ops[4] = operands[1]; output_asm_insn (“vld4.<V_sz_elem>\t{%P0, %P1, %P2, %P3}, %A4”, ops); return ""; } [(set_attr “type” “neon_load4_4reg”)] )
(define_insn “neon_vld4_lane” [(set (match_operand:OI 0 “s_register_operand” “=w”) (unspec:OI [(match_operand:<V_four_elem> 1 “neon_struct_operand” “Um”) (match_operand:OI 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”) (unspec:VD [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD4_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[0]); rtx ops[6]; if (lane < 0 || lane >= max) error (“lane out of range”); ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 2); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = gen_rtx_REG (DImode, regno + 6); ops[4] = operands[1]; ops[5] = operands[3]; output_asm_insn (“vld4.<V_sz_elem>\t{%P0[%c5], %P1[%c5], %P2[%c5], %P3[%c5]}, %A4”, ops); return ""; } [(set_attr “type” “neon_load4_one_lane”)] )
(define_insn “neon_vld4_lane” [(set (match_operand:XI 0 “s_register_operand” “=w”) (unspec:XI [(match_operand:<V_four_elem> 1 “neon_struct_operand” “Um”) (match_operand:XI 2 “s_register_operand” “0”) (match_operand:SI 3 “immediate_operand” “i”) (unspec:VMQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD4_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[3]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[0]); rtx ops[6]; if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; } ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 4); ops[2] = gen_rtx_REG (DImode, regno + 8); ops[3] = gen_rtx_REG (DImode, regno + 12); ops[4] = operands[1]; ops[5] = GEN_INT (lane); output_asm_insn (“vld4.<V_sz_elem>\t{%P0[%c5], %P1[%c5], %P2[%c5], %P3[%c5]}, %A4”, ops); return ""; } [(set_attr “type” “neon_load4_one_lane”)] )
(define_insn “neon_vld4_dup” [(set (match_operand:OI 0 “s_register_operand” “=w”) (unspec:OI [(match_operand:<V_four_elem> 1 “neon_struct_operand” “Um”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VLD4_DUP))] “TARGET_NEON” { if (GET_MODE_NUNITS (mode) > 1) { int regno = REGNO (operands[0]); rtx ops[5]; ops[0] = gen_rtx_REG (DImode, regno); ops[1] = gen_rtx_REG (DImode, regno + 2); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = gen_rtx_REG (DImode, regno + 6); ops[4] = operands[1]; output_asm_insn (“vld4.<V_sz_elem>\t{%P0[], %P1[], %P2[], %P3[]}, %A4”, ops); return ""; } else return “vld1.<V_sz_elem>\t%h0, %A1”; } [(set (attr “type”) (if_then_else (gt (const_string “<V_mode_nunits>”) (const_string “1”)) (const_string “neon_load4_all_lanes”) (const_string “neon_load1_1reg”)))] )
(define_expand “vec_store_lanesoi” [(set (match_operand:OI 0 “neon_struct_operand”) (unspec:OI [(match_operand:OI 1 “s_register_operand”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST4))] “TARGET_NEON”)
(define_insn “neon_vst4” [(set (match_operand:OI 0 “neon_struct_operand” “=Um”) (unspec:OI [(match_operand:OI 1 “s_register_operand” “w”) (unspec:VDX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST4))] “TARGET_NEON” { if (<V_sz_elem> == 64) return “vst1.64\t%h1, %A0”; else return “vst4.<V_sz_elem>\t%h1, %A0”; } [(set (attr “type”) (if_then_else (eq (const_string “<V_sz_elem>”) (const_string “64”)) (const_string “neon_store1_4reg”) (const_string “neon_store4_4reg”)))] )
(define_expand “vec_store_lanesxi” [(match_operand:XI 0 “neon_struct_operand”) (match_operand:XI 1 “s_register_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { emit_insn (gen_neon_vst4 (operands[0], operands[1])); DONE; })
(define_expand “neon_vst4” [(match_operand:XI 0 “neon_struct_operand”) (match_operand:XI 1 “s_register_operand”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] “TARGET_NEON” { rtx mem;
mem = adjust_address (operands[0], OImode, 0); emit_insn (gen_neon_vst4qa (mem, operands[1])); mem = adjust_address (mem, OImode, GET_MODE_SIZE (OImode)); emit_insn (gen_neon_vst4qb (mem, operands[1])); DONE; })
(define_insn “neon_vst4qa” [(set (match_operand:OI 0 “neon_struct_operand” “=Um”) (unspec:OI [(match_operand:XI 1 “s_register_operand” “w”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST4A))] “TARGET_NEON” { int regno = REGNO (operands[1]); rtx ops[5]; ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = gen_rtx_REG (DImode, regno + 8); ops[4] = gen_rtx_REG (DImode, regno + 12); output_asm_insn (“vst4.<V_sz_elem>\t{%P1, %P2, %P3, %P4}, %A0”, ops); return ""; } [(set_attr “type” “neon_store4_4reg”)] )
(define_insn “neon_vst4qb” [(set (match_operand:OI 0 “neon_struct_operand” “=Um”) (unspec:OI [(match_operand:XI 1 “s_register_operand” “w”) (unspec:VQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST4B))] “TARGET_NEON” { int regno = REGNO (operands[1]); rtx ops[5]; ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno + 2); ops[2] = gen_rtx_REG (DImode, regno + 6); ops[3] = gen_rtx_REG (DImode, regno + 10); ops[4] = gen_rtx_REG (DImode, regno + 14); output_asm_insn (“vst4.<V_sz_elem>\t{%P1, %P2, %P3, %P4}, %A0”, ops); return ""; } [(set_attr “type” “neon_store4_4reg”)] )
(define_insn “neon_vst4_lane” [(set (match_operand:<V_four_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_four_elem> [(match_operand:OI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”) (unspec:VD [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST4_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[1]); rtx ops[6]; if (lane < 0 || lane >= max) error (“lane out of range”); ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 2); ops[3] = gen_rtx_REG (DImode, regno + 4); ops[4] = gen_rtx_REG (DImode, regno + 6); ops[5] = operands[2]; output_asm_insn (“vst4.<V_sz_elem>\t{%P1[%c5], %P2[%c5], %P3[%c5], %P4[%c5]}, %A0”, ops); return ""; } [(set_attr “type” “neon_store4_one_lane”)] )
(define_insn “neon_vst4_lane” [(set (match_operand:<V_four_elem> 0 “neon_struct_operand” “=Um”) (unspec:<V_four_elem> [(match_operand:XI 1 “s_register_operand” “w”) (match_operand:SI 2 “immediate_operand” “i”) (unspec:VMQ [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] UNSPEC_VST4_LANE))] “TARGET_NEON” { HOST_WIDE_INT lane = INTVAL (operands[2]); HOST_WIDE_INT max = GET_MODE_NUNITS (mode); int regno = REGNO (operands[1]); rtx ops[6]; if (lane < 0 || lane >= max) error (“lane out of range”); else if (lane >= max / 2) { lane -= max / 2; regno += 2; } ops[0] = operands[0]; ops[1] = gen_rtx_REG (DImode, regno); ops[2] = gen_rtx_REG (DImode, regno + 4); ops[3] = gen_rtx_REG (DImode, regno + 8); ops[4] = gen_rtx_REG (DImode, regno + 12); ops[5] = GEN_INT (lane); output_asm_insn (“vst4.<V_sz_elem>\t{%P1[%c5], %P2[%c5], %P3[%c5], %P4[%c5]}, %A0”, ops); return ""; } [(set_attr “type” “neon_store4_4reg”)] )
(define_insn “neon_vec_unpacklo” [(set (match_operand:<V_unpack> 0 “register_operand” “=w”) (SE:<V_unpack> (vec_select:<V_HALF> (match_operand:VU 1 “register_operand” “w”) (match_operand:VU 2 “vect_par_constant_low” ""))))] “TARGET_NEON && !BYTES_BIG_ENDIAN” “vmovl.<V_sz_elem> %q0, %e1” [(set_attr “type” “neon_shift_imm_long”)] )
(define_insn “neon_vec_unpackhi” [(set (match_operand:<V_unpack> 0 “register_operand” “=w”) (SE:<V_unpack> (vec_select:<V_HALF> (match_operand:VU 1 “register_operand” “w”) (match_operand:VU 2 “vect_par_constant_high” ""))))] “TARGET_NEON && !BYTES_BIG_ENDIAN” “vmovl.<V_sz_elem> %q0, %f1” [(set_attr “type” “neon_shift_imm_long”)] )
(define_expand “vec_unpackhi” [(match_operand:<V_unpack> 0 “register_operand” "") (SE:<V_unpack> (match_operand:VU 1 “register_operand”))] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtvec v = rtvec_alloc (<V_mode_nunits>/2) ; rtx t1; int i; for (i = 0; i < (<V_mode_nunits>/2); i++) RTVEC_ELT (v, i) = GEN_INT ((<V_mode_nunits>/2) + i);
t1 = gen_rtx_PARALLEL (mode, v); emit_insn (gen_neon_vec_unpackhi (operands[0], operands[1], t1)); DONE; } )
(define_expand “vec_unpacklo” [(match_operand:<V_unpack> 0 “register_operand” "") (SE:<V_unpack> (match_operand:VU 1 “register_operand” ""))] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtvec v = rtvec_alloc (<V_mode_nunits>/2) ; rtx t1; int i; for (i = 0; i < (<V_mode_nunits>/2) ; i++) RTVEC_ELT (v, i) = GEN_INT (i); t1 = gen_rtx_PARALLEL (mode, v); emit_insn (gen_neon_vec_unpacklo (operands[0], operands[1], t1)); DONE; } )
(define_insn “neon_vec_mult_lo_” [(set (match_operand:<V_unpack> 0 “register_operand” “=w”) (mult:<V_unpack> (SE:<V_unpack> (vec_select:<V_HALF> (match_operand:VU 1 “register_operand” “w”) (match_operand:VU 2 “vect_par_constant_low” ""))) (SE:<V_unpack> (vec_select:<V_HALF> (match_operand:VU 3 “register_operand” “w”) (match_dup 2)))))] “TARGET_NEON && !BYTES_BIG_ENDIAN” “vmull.<V_sz_elem> %q0, %e1, %e3” [(set_attr “type” “neon_mul_<V_elem_ch>_long”)] )
(define_expand “vec_widen_mult_lo_” [(match_operand:<V_unpack> 0 “register_operand” "") (SE:<V_unpack> (match_operand:VU 1 “register_operand” "")) (SE:<V_unpack> (match_operand:VU 2 “register_operand” ""))] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtvec v = rtvec_alloc (<V_mode_nunits>/2) ; rtx t1; int i; for (i = 0; i < (<V_mode_nunits>/2) ; i++) RTVEC_ELT (v, i) = GEN_INT (i); t1 = gen_rtx_PARALLEL (mode, v);
emit_insn (gen_neon_vec_mult_lo_ (operands[0], operands[1], t1, operands[2])); DONE; } )
(define_insn “neon_vec_mult_hi_” [(set (match_operand:<V_unpack> 0 “register_operand” “=w”) (mult:<V_unpack> (SE:<V_unpack> (vec_select:<V_HALF> (match_operand:VU 1 “register_operand” “w”) (match_operand:VU 2 “vect_par_constant_high” ""))) (SE:<V_unpack> (vec_select:<V_HALF> (match_operand:VU 3 “register_operand” “w”) (match_dup 2)))))] “TARGET_NEON && !BYTES_BIG_ENDIAN” “vmull.<V_sz_elem> %q0, %f1, %f3” [(set_attr “type” “neon_mul_<V_elem_ch>_long”)] )
(define_expand “vec_widen_mult_hi_” [(match_operand:<V_unpack> 0 “register_operand” "") (SE:<V_unpack> (match_operand:VU 1 “register_operand” "")) (SE:<V_unpack> (match_operand:VU 2 “register_operand” ""))] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtvec v = rtvec_alloc (<V_mode_nunits>/2) ; rtx t1; int i; for (i = 0; i < (<V_mode_nunits>/2) ; i++) RTVEC_ELT (v, i) = GEN_INT (<V_mode_nunits>/2 + i); t1 = gen_rtx_PARALLEL (mode, v);
emit_insn (gen_neon_vec_mult_hi_ (operands[0], operands[1], t1, operands[2])); DONE;
} )
(define_insn “neon_vec_shiftl_” [(set (match_operand:<V_widen> 0 “register_operand” “=w”) (SE:<V_widen> (ashift:VW (match_operand:VW 1 “register_operand” “w”) (match_operand:<V_innermode> 2 “const_neon_scalar_shift_amount_operand” ""))))] “TARGET_NEON” { return “vshll.<V_sz_elem> %q0, %P1, %2”; } [(set_attr “type” “neon_shift_imm_long”)] )
(define_expand “vec_widen_shiftl_lo_” [(match_operand:<V_unpack> 0 “register_operand” "") (SE:<V_unpack> (match_operand:VU 1 “register_operand” "")) (match_operand:SI 2 “immediate_operand” “i”)] “TARGET_NEON && !BYTES_BIG_ENDIAN” { emit_insn (gen_neon_vec_shiftl_<V_half> (operands[0], simplify_gen_subreg (<V_HALF>mode, operands[1], mode, 0), operands[2])); DONE; } )
(define_expand “vec_widen_shiftl_hi_” [(match_operand:<V_unpack> 0 “register_operand” "") (SE:<V_unpack> (match_operand:VU 1 “register_operand” "")) (match_operand:SI 2 “immediate_operand” “i”)] “TARGET_NEON && !BYTES_BIG_ENDIAN” { emit_insn (gen_neon_vec_shiftl_<V_half> (operands[0], simplify_gen_subreg (<V_HALF>mode, operands[1], mode, GET_MODE_SIZE (<V_HALF>mode)), operands[2])); DONE; } )
;; Vectorize for non-neon-quad case (define_insn “neon_unpack_” [(set (match_operand:<V_widen> 0 “register_operand” “=w”) (SE:<V_widen> (match_operand:VDI 1 “register_operand” “w”)))] “TARGET_NEON” “vmovl.<V_sz_elem> %q0, %P1” [(set_attr “type” “neon_move”)] )
(define_expand “vec_unpacklo” [(match_operand:<V_double_width> 0 “register_operand” "") (SE:<V_double_width>(match_operand:VDI 1 “register_operand”))] “TARGET_NEON” { rtx tmpreg = gen_reg_rtx (<V_widen>mode); emit_insn (gen_neon_unpack_ (tmpreg, operands[1])); emit_insn (gen_neon_vget_low<V_widen_l> (operands[0], tmpreg));
DONE; } )
(define_expand “vec_unpackhi” [(match_operand:<V_double_width> 0 “register_operand” "") (SE:<V_double_width>(match_operand:VDI 1 “register_operand”))] “TARGET_NEON” { rtx tmpreg = gen_reg_rtx (<V_widen>mode); emit_insn (gen_neon_unpack_ (tmpreg, operands[1])); emit_insn (gen_neon_vget_high<V_widen_l> (operands[0], tmpreg));
DONE; } )
(define_insn “neon_vec_mult_” [(set (match_operand:<V_widen> 0 “register_operand” “=w”) (mult:<V_widen> (SE:<V_widen> (match_operand:VDI 1 “register_operand” “w”)) (SE:<V_widen> (match_operand:VDI 2 “register_operand” “w”))))] “TARGET_NEON” “vmull.<V_sz_elem> %q0, %P1, %P2” [(set_attr “type” “neon_mul_<V_elem_ch>_long”)] )
(define_expand “vec_widen_mult_hi_” [(match_operand:<V_double_width> 0 “register_operand” "") (SE:<V_double_width> (match_operand:VDI 1 “register_operand” "")) (SE:<V_double_width> (match_operand:VDI 2 “register_operand” ""))] “TARGET_NEON” { rtx tmpreg = gen_reg_rtx (<V_widen>mode); emit_insn (gen_neon_vec_mult_ (tmpreg, operands[1], operands[2])); emit_insn (gen_neon_vget_high<V_widen_l> (operands[0], tmpreg));
DONE;
} )
(define_expand “vec_widen_mult_lo_” [(match_operand:<V_double_width> 0 “register_operand” "") (SE:<V_double_width> (match_operand:VDI 1 “register_operand” "")) (SE:<V_double_width> (match_operand:VDI 2 “register_operand” ""))] “TARGET_NEON” { rtx tmpreg = gen_reg_rtx (<V_widen>mode); emit_insn (gen_neon_vec_mult_ (tmpreg, operands[1], operands[2])); emit_insn (gen_neon_vget_low<V_widen_l> (operands[0], tmpreg));
DONE;
} )
(define_expand “vec_widen_shiftl_hi_” [(match_operand:<V_double_width> 0 “register_operand” "") (SE:<V_double_width> (match_operand:VDI 1 “register_operand” "")) (match_operand:SI 2 “immediate_operand” “i”)] “TARGET_NEON” { rtx tmpreg = gen_reg_rtx (<V_widen>mode); emit_insn (gen_neon_vec_shiftl_ (tmpreg, operands[1], operands[2])); emit_insn (gen_neon_vget_high<V_widen_l> (operands[0], tmpreg));
DONE; } )
(define_expand “vec_widen_shiftl_lo_” [(match_operand:<V_double_width> 0 “register_operand” "") (SE:<V_double_width> (match_operand:VDI 1 “register_operand” "")) (match_operand:SI 2 “immediate_operand” “i”)] “TARGET_NEON” { rtx tmpreg = gen_reg_rtx (<V_widen>mode); emit_insn (gen_neon_vec_shiftl_ (tmpreg, operands[1], operands[2])); emit_insn (gen_neon_vget_low<V_widen_l> (operands[0], tmpreg));
DONE; } )
; FIXME: These instruction patterns can't be used safely in big-endian mode ; because the ordering of vector elements in Q registers is different from what ; the semantics of the instructions require.
(define_insn “vec_pack_trunc_” [(set (match_operand:<V_narrow_pack> 0 “register_operand” “=&w”) (vec_concat:<V_narrow_pack> (truncate:<V_narrow> (match_operand:VN 1 “register_operand” “w”)) (truncate:<V_narrow> (match_operand:VN 2 “register_operand” “w”))))] “TARGET_NEON && !BYTES_BIG_ENDIAN” “vmovn.i<V_sz_elem>\t%e0, %q1;vmovn.i<V_sz_elem>\t%f0, %q2” [(set_attr “type” “multiple”) (set_attr “length” “8”)] )
;; For the non-quad case. (define_insn “neon_vec_pack_trunc_” [(set (match_operand:<V_narrow> 0 “register_operand” “=w”) (truncate:<V_narrow> (match_operand:VN 1 “register_operand” “w”)))] “TARGET_NEON && !BYTES_BIG_ENDIAN” “vmovn.i<V_sz_elem>\t%P0, %q1” [(set_attr “type” “neon_move_narrow_q”)] )
(define_expand “vec_pack_trunc_” [(match_operand:<V_narrow_pack> 0 “register_operand” "") (match_operand:VSHFT 1 “register_operand” "") (match_operand:VSHFT 2 “register_operand”)] “TARGET_NEON && !BYTES_BIG_ENDIAN” { rtx tempreg = gen_reg_rtx (<V_DOUBLE>mode);
emit_insn (gen_move_lo_quad_<V_double> (tempreg, operands[1])); emit_insn (gen_move_hi_quad_<V_double> (tempreg, operands[2])); emit_insn (gen_neon_vec_pack_trunc_<V_double> (operands[0], tempreg)); DONE; })
(define_insn “neon_vabd_2” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (abs:VDQ (minus:VDQ (match_operand:VDQ 1 “s_register_operand” “w”) (match_operand:VDQ 2 “s_register_operand” “w”))))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” “vabd.<V_s_elem> %<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (ne (symbol_ref “<Is_float_mode>”) (const_int 0)) (const_string “neon_fp_abd_s”) (const_string “neon_abd”)))] )
(define_insn “neon_vabd_3” [(set (match_operand:VDQ 0 “s_register_operand” “=w”) (abs:VDQ (unspec:VDQ [(match_operand:VDQ 1 “s_register_operand” “w”) (match_operand:VDQ 2 “s_register_operand” “w”)] UNSPEC_VSUB)))] “TARGET_NEON && (!<Is_float_mode> || flag_unsafe_math_optimizations)” “vabd.<V_if_elem> %<V_reg>0, %<V_reg>1, %<V_reg>2” [(set (attr “type”) (if_then_else (ne (symbol_ref “<Is_float_mode>”) (const_int 0)) (const_string “neon_fp_abd_s”) (const_string “neon_abd”)))] )
;; Copy from core-to-neon regs, then extend, not vice-versa
(define_split [(set (match_operand:DI 0 “s_register_operand” "") (sign_extend:DI (match_operand:SI 1 “s_register_operand” "")))] “TARGET_NEON && reload_completed && IS_VFP_REGNUM (REGNO (operands[0]))” [(set (match_dup 2) (vec_duplicate:V2SI (match_dup 1))) (set (match_dup 0) (ashiftrt:DI (match_dup 0) (const_int 32)))] { operands[2] = gen_rtx_REG (V2SImode, REGNO (operands[0])); })
(define_split [(set (match_operand:DI 0 “s_register_operand” "") (sign_extend:DI (match_operand:HI 1 “s_register_operand” "")))] “TARGET_NEON && reload_completed && IS_VFP_REGNUM (REGNO (operands[0]))” [(set (match_dup 2) (vec_duplicate:V4HI (match_dup 1))) (set (match_dup 0) (ashiftrt:DI (match_dup 0) (const_int 48)))] { operands[2] = gen_rtx_REG (V4HImode, REGNO (operands[0])); })
(define_split [(set (match_operand:DI 0 “s_register_operand” "") (sign_extend:DI (match_operand:QI 1 “s_register_operand” "")))] “TARGET_NEON && reload_completed && IS_VFP_REGNUM (REGNO (operands[0]))” [(set (match_dup 2) (vec_duplicate:V8QI (match_dup 1))) (set (match_dup 0) (ashiftrt:DI (match_dup 0) (const_int 56)))] { operands[2] = gen_rtx_REG (V8QImode, REGNO (operands[0])); })
(define_split [(set (match_operand:DI 0 “s_register_operand” "") (zero_extend:DI (match_operand:SI 1 “s_register_operand” "")))] “TARGET_NEON && reload_completed && IS_VFP_REGNUM (REGNO (operands[0]))” [(set (match_dup 2) (vec_duplicate:V2SI (match_dup 1))) (set (match_dup 0) (lshiftrt:DI (match_dup 0) (const_int 32)))] { operands[2] = gen_rtx_REG (V2SImode, REGNO (operands[0])); })
(define_split [(set (match_operand:DI 0 “s_register_operand” "") (zero_extend:DI (match_operand:HI 1 “s_register_operand” "")))] “TARGET_NEON && reload_completed && IS_VFP_REGNUM (REGNO (operands[0]))” [(set (match_dup 2) (vec_duplicate:V4HI (match_dup 1))) (set (match_dup 0) (lshiftrt:DI (match_dup 0) (const_int 48)))] { operands[2] = gen_rtx_REG (V4HImode, REGNO (operands[0])); })
(define_split [(set (match_operand:DI 0 “s_register_operand” "") (zero_extend:DI (match_operand:QI 1 “s_register_operand” "")))] “TARGET_NEON && reload_completed && IS_VFP_REGNUM (REGNO (operands[0]))” [(set (match_dup 2) (vec_duplicate:V8QI (match_dup 1))) (set (match_dup 0) (lshiftrt:DI (match_dup 0) (const_int 56)))] { operands[2] = gen_rtx_REG (V8QImode, REGNO (operands[0])); })