1 /*
  2  * Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
  3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  4  *
  5  * This code is free software; you can redistribute it and/or modify it
  6  * under the terms of the GNU General Public License version 2 only, as
  7  * published by the Free Software Foundation.
  8  *
  9  * This code is distributed in the hope that it will be useful, but WITHOUT
 10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 12  * version 2 for more details (a copy is included in the LICENSE file that
 13  * accompanied this code).
 14  *
 15  * You should have received a copy of the GNU General Public License version
 16  * 2 along with this work; if not, write to the Free Software Foundation,
 17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 18  *
 19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 20  * or visit www.oracle.com if you need additional information or have any
 21  * questions.
 22  *
 23  */
 24 
 25 #ifndef CPU_AARCH64_C2_MACROASSEMBLER_AARCH64_HPP
 26 #define CPU_AARCH64_C2_MACROASSEMBLER_AARCH64_HPP
 27 
 28 // C2_MacroAssembler contains high-level macros for C2
 29 
 30  private:
 31   // Return true if the phase output is in the scratch emit size mode.
 32   virtual bool in_scratch_emit_size() override;
 33 
 34   void neon_reduce_logical_helper(int opc, bool sf, Register Rd, Register Rn, Register Rm,
 35                                   enum shift_kind kind = Assembler::LSL, unsigned shift = 0);
 36 
 37   // Helper functions for min/max reduction operations
 38 
 39   void decode_minmax_reduction_opc(int opc, bool* is_min, bool* is_unsigned, Condition* cond);
 40 
 41   void neon_minmaxp(bool is_unsigned, bool is_min, FloatRegister dst,
 42                     SIMD_Arrangement size, FloatRegister src1, FloatRegister src2) {
 43     auto m = is_unsigned ? (is_min ? &Assembler::uminp : &Assembler::umaxp)
 44                          : (is_min ? &Assembler::sminp : &Assembler::smaxp);
 45     (this->*m)(dst, size, src1, src2);
 46   }
 47 
 48   // Typedefs used to disambiguate overloaded member functions.
 49   typedef void (Assembler::*neon_reduction2)
 50     (FloatRegister, SIMD_Arrangement, FloatRegister);
 51 
 52   void neon_minmaxv(bool is_unsigned, bool is_min, FloatRegister dst,
 53                     SIMD_Arrangement size, FloatRegister src) {
 54     auto m = is_unsigned ? (is_min ? (neon_reduction2)&Assembler::uminv
 55                                    : (neon_reduction2)&Assembler::umaxv)
 56                          : (is_min ? &Assembler::sminv
 57                                    : &Assembler::smaxv);
 58     (this->*m)(dst, size, src);
 59   }
 60 
 61   void sve_minmaxv(bool is_unsigned, bool is_min, FloatRegister dst,
 62                    SIMD_RegVariant size, PRegister pg, FloatRegister src) {
 63     auto m = is_unsigned ? (is_min ? &Assembler::sve_uminv : &Assembler::sve_umaxv)
 64                          : (is_min ? &Assembler::sve_sminv : &Assembler::sve_smaxv);
 65     (this->*m)(dst, size, pg, src);
 66   }
 67 
 68   void select_from_two_vectors_neon(FloatRegister dst, FloatRegister src1,
 69                                     FloatRegister src2, FloatRegister index,
 70                                     FloatRegister tmp, unsigned vector_length_in_bytes);
 71 
 72   void select_from_two_vectors_sve(FloatRegister dst, FloatRegister src1,
 73                                    FloatRegister src2, FloatRegister index,
 74                                    FloatRegister tmp, SIMD_RegVariant T,
 75                                    unsigned vector_length_in_bytes);
 76 
 77  public:
 78   using Assembler::sve_cpy;
 79 
 80   void entry_barrier();
 81 
 82   // jdk.internal.util.ArraysSupport.vectorizedHashCode
 83   address arrays_hashcode(Register ary, Register cnt, Register result, FloatRegister vdata0,
 84                           FloatRegister vdata1, FloatRegister vdata2, FloatRegister vdata3,
 85                           FloatRegister vmul0, FloatRegister vmul1, FloatRegister vmul2,
 86                           FloatRegister vmul3, FloatRegister vpow, FloatRegister vpowm,
 87                           BasicType eltype);
 88 
 89   // Code used by cmpFastLock and cmpFastUnlock mach instructions in .ad file.
 90   void fast_lock(Register object, Register box, Register t1, Register t2, Register t3);
 91   void fast_unlock(Register object, Register box, Register t1, Register t2, Register t3);
 92 
 93   void string_compare(Register str1, Register str2,
 94                       Register cnt1, Register cnt2, Register result,
 95                       Register tmp1, Register tmp2, FloatRegister vtmp1,
 96                       FloatRegister vtmp2, FloatRegister vtmp3,
 97                       PRegister pgtmp1, PRegister pgtmp2, int ae);
 98 
 99   void string_indexof(Register str1, Register str2,
100                       Register cnt1, Register cnt2,
101                       Register tmp1, Register tmp2,
102                       Register tmp3, Register tmp4,
103                       Register tmp5, Register tmp6,
104                       int int_cnt1, Register result, int ae);
105 
106   void string_indexof_char(Register str1, Register cnt1,
107                            Register ch, Register result,
108                            Register tmp1, Register tmp2, Register tmp3);
109 
110   void stringL_indexof_char(Register str1, Register cnt1,
111                             Register ch, Register result,
112                             Register tmp1, Register tmp2, Register tmp3);
113 
114   void string_indexof_char_sve(Register str1, Register cnt1,
115                                Register ch, Register result,
116                                FloatRegister ztmp1, FloatRegister ztmp2,
117                                PRegister pgtmp, PRegister ptmp, bool isL);
118 
119   // Compress the least significant bit of each byte to the rightmost and clear
120   // the higher garbage bits.
121   void bytemask_compress(Register dst);
122 
123   // Pack the value of each mask element in "src" into a long value in "dst", at most the
124   // first 64 lane elements. The input "src" is a vector of boolean represented as bytes
125   // with 0x00/0x01 as element values. Each lane value from "src" is packed into one bit in
126   // "dst".
127   void sve_vmask_tolong(Register dst, FloatRegister src, FloatRegister vtmp, int lane_cnt);
128 
129   void sve2_vmask_tolong(Register dst, FloatRegister src, FloatRegister vtmp1,
130                          FloatRegister vtmp2, int lane_cnt);
131 
132   // Unpack the mask, a long value in "src", into vector register "dst" with boolean type.
133   // Each bit in "src" is unpacked into one byte lane in "dst". Note that "dst" can support
134   // at most 64 lanes.
135   void sve_vmask_fromlong(FloatRegister dst, Register src, FloatRegister vtmp, int lane_cnt);
136 
137   // SIMD&FP comparison
138   void neon_compare(FloatRegister dst, BasicType bt, FloatRegister src1,
139                     FloatRegister src2, Condition cond, bool isQ);
140 
141   void neon_compare_zero(FloatRegister dst, BasicType bt, FloatRegister src,
142                          Condition cond, bool isQ);
143 
144   void sve_compare(PRegister pd, BasicType bt, PRegister pg,
145                    FloatRegister zn, FloatRegister zm, Condition cond);
146 
147   void sve_vmask_lasttrue(Register dst, BasicType bt, PRegister src, PRegister ptmp);
148 
149   // Vector cast
150   void neon_vector_extend(FloatRegister dst, BasicType dst_bt, unsigned dst_vlen_in_bytes,
151                           FloatRegister src, BasicType src_bt, bool is_unsigned = false);
152 
153   void neon_vector_narrow(FloatRegister dst, BasicType dst_bt,
154                           FloatRegister src, BasicType src_bt, unsigned src_vlen_in_bytes);
155 
156   void sve_vector_extend(FloatRegister dst, SIMD_RegVariant dst_size,
157                          FloatRegister src, SIMD_RegVariant src_size, bool is_unsigned = false);
158 
159   void sve_vector_narrow(FloatRegister dst, SIMD_RegVariant dst_size,
160                          FloatRegister src, SIMD_RegVariant src_size, FloatRegister tmp);
161 
162   void sve_vmaskcast_extend(PRegister dst, PRegister src,
163                             uint dst_element_length_in_bytes, uint src_element_lenght_in_bytes);
164 
165   void sve_vmaskcast_narrow(PRegister dst, PRegister src, PRegister ptmp,
166                             uint dst_element_length_in_bytes, uint src_element_lenght_in_bytes);
167 
168   // Vector reduction
169   void neon_reduce_add_integral(Register dst, BasicType bt,
170                                 Register isrc, FloatRegister vsrc,
171                                 unsigned vector_length_in_bytes, FloatRegister vtmp);
172 
173   void neon_reduce_mul_integral(Register dst, BasicType bt,
174                                 Register isrc, FloatRegister vsrc,
175                                 unsigned vector_length_in_bytes,
176                                 FloatRegister vtmp1, FloatRegister vtmp2);
177 
178   void neon_reduce_mul_fp(FloatRegister dst, BasicType bt,
179                           FloatRegister fsrc, FloatRegister vsrc,
180                           unsigned vector_length_in_bytes, FloatRegister vtmp);
181 
182   void neon_reduce_add_fp16(FloatRegister dst, FloatRegister fsrc, FloatRegister vsrc,
183                             unsigned vector_length_in_bytes, FloatRegister vtmp);
184 
185   void neon_reduce_logical(int opc, Register dst, BasicType bt, Register isrc,
186                            FloatRegister vsrc, unsigned vector_length_in_bytes);
187 
188   void neon_reduce_minmax_integral(int opc, Register dst, BasicType bt,
189                                    Register isrc, FloatRegister vsrc,
190                                    unsigned vector_length_in_bytes, FloatRegister vtmp);
191 
192   void sve_reduce_integral(int opc, Register dst, BasicType bt, Register src1,
193                            FloatRegister src2, PRegister pg, FloatRegister tmp);
194 
195   // Set elements of the dst predicate to true for lanes in the range of
196   // [0, lane_cnt), or to false otherwise. The input "lane_cnt" should be
197   // smaller than or equal to the supported max vector length of the basic
198   // type. Clobbers: rscratch1 and the rFlagsReg.
199   void sve_gen_mask_imm(PRegister dst, BasicType bt, uint32_t lane_cnt);
200 
201   // Extract a scalar element from an sve vector at position 'idx'.
202   // The input elements in src are expected to be of integral type.
203   void sve_extract_integral(Register dst, BasicType bt, FloatRegister src,
204                             int idx, FloatRegister vtmp);
205 
206   // java.lang.Math::round intrinsics
207   void vector_round_neon(FloatRegister dst, FloatRegister src, FloatRegister tmp1,
208                          FloatRegister tmp2, FloatRegister tmp3,
209                          SIMD_Arrangement T);
210   void vector_round_sve(FloatRegister dst, FloatRegister src, FloatRegister tmp1,
211                         FloatRegister tmp2, PRegister pgtmp,
212                         SIMD_RegVariant T);
213 
214   // Pack active elements of src, under the control of mask, into the
215   // lowest-numbered elements of dst. Any remaining elements of dst will
216   // be filled with zero.
217   void sve_compress_byte(FloatRegister dst, FloatRegister src, PRegister mask,
218                          FloatRegister vtmp1, FloatRegister vtmp2, FloatRegister vtmp3,
219                          PRegister ptmp, PRegister pgtmp, unsigned vector_length_in_bytes);
220 
221   void sve_compress_short(FloatRegister dst, FloatRegister src, PRegister mask,
222                           FloatRegister vzr, FloatRegister vtmp,
223                           PRegister pgtmp, unsigned vector_length_in_bytes);
224 
225   void neon_reverse_bits(FloatRegister dst, FloatRegister src, BasicType bt, bool isQ);
226 
227   void neon_reverse_bytes(FloatRegister dst, FloatRegister src, BasicType bt, bool isQ);
228 
229   void neon_rearrange_hsd(FloatRegister dst, FloatRegister src, FloatRegister shuffle,
230                           FloatRegister tmp, BasicType bt, bool isQ);
231   // java.lang.Math::signum intrinsics
232   void vector_signum_neon(FloatRegister dst, FloatRegister src, FloatRegister zero,
233                           FloatRegister one, SIMD_Arrangement T);
234 
235   void vector_signum_sve(FloatRegister dst, FloatRegister src, FloatRegister zero,
236                          FloatRegister one, FloatRegister vtmp, PRegister pgtmp, SIMD_RegVariant T);
237 
238   void verify_int_in_range(uint idx, const TypeInt* t, Register val, Register tmp);
239   void verify_long_in_range(uint idx, const TypeLong* t, Register val, Register tmp);
240 
241   void reconstruct_frame_pointer(Register rtmp);
242 
243   // Select from a table of two vectors
244   void select_from_two_vectors(FloatRegister dst, FloatRegister src1, FloatRegister src2,
245                                FloatRegister index, FloatRegister tmp, BasicType bt,
246                                unsigned vector_length_in_bytes);
247 
248   void vector_expand_neon(FloatRegister dst, FloatRegister src, FloatRegister mask,
249                           FloatRegister tmp1, FloatRegister tmp2, BasicType bt,
250                           int vector_length_in_bytes);
251   void vector_expand_sve(FloatRegister dst, FloatRegister src, PRegister pg,
252                          FloatRegister tmp1, FloatRegister tmp2, BasicType bt,
253                          int vector_length_in_bytes);
254 
255   void sve_cpy(FloatRegister dst, SIMD_RegVariant T, PRegister pg, int imm8,
256                bool isMerge);
257   int vector_iota_entry_index(BasicType bt);
258 #endif // CPU_AARCH64_C2_MACROASSEMBLER_AARCH64_HPP