arm/vm/stubGenerator_arm.cpp

1043     int bytes = count * wordSize;
1046     __ ldp(regs[0], regs[1], Address(from, bytes, post_indexed));
1050       __ ldp(regs[i], regs[i+1], Address(from, -bytes + offset));
1054     assert (offset == bytes, "must be");
1061     int bytes = count * wordSize;
1064     __ stp(regs[0], regs[1], Address(to, bytes, post_indexed));
1068       __ stp(regs[i], regs[i+1], Address(to, -bytes + offset));
1072     assert (offset == bytes, "must be");
1080     int bytes = count * wordSize;
1090     __ ldp(regs[0], regs[1], Address(from, -bytes, pre_indexed));
1092     assert (offset == bytes, "must be");
1100     int bytes = count * wordSize;
1110     __ stp(regs[0], regs[1], Address(to, -bytes, pre_indexed));
1112     assert (offset == bytes, "must be");
1132   //      bytes_per_count: number of bytes for each unit of 'count'
1204       // LDM (32-bit ARM) / LDP (AArch64) copy of 'bytes_per_loop' bytes
1207       // PLD with 64 bytes cache line but the gain was not significant.
1258     BLOCK_COMMENT("Remaining bytes:");
1259     // still 0..bytes_per_loop-1 aligned bytes to copy, count already decreased by (at least) bytes_per_loop bytes
1329     __ ldmia(from, RegisterSet(R3, R6), writeback, ne); // copy 16 bytes
1333     __ ldmia(from, RegisterSet(R3, R4), writeback, ne); // copy 8 bytes
1338       __ ldr(R3, Address(from, 4, post_indexed), ne); // copy 4 bytes
1344       __ ldrh(R3, Address(from, 2, post_indexed), ne); // copy 2 bytes
1367   //      bytes_per_count: number of bytes for each unit of 'count'
1427       // LDM (32-bit ARM) / LDP (AArch64) copy of 'bytes_per_loop' bytes
1430       // PLD with 64 bytes cache line but the gain was not significant.
1476     BLOCK_COMMENT("Remaining bytes:");
1477     // still 0..bytes_per_loop-1 aligned bytes to copy, count already decreased by (at least) bytes_per_loop bytes
1547     __ ldmdb(end_from, RegisterSet(R3, R6), writeback, ne); // copy 16 bytes
1551     __ ldmdb(end_from, RegisterSet(R3, R4), writeback, ne); // copy 8 bytes
1556       __ ldr(R3, Address(end_from, -4, pre_indexed), ne); // copy 4 bytes
1562       __ ldrh(R3, Address(end_from, -2, pre_indexed), ne); // copy 2 bytes
1587   //      bytes_per_count: number of bytes for each unit of 'count'
1588   //      lsr_shift: shift applied to 'old' value to skipped already written bytes
1589   //      lsl_shift: shift applied to 'new' value to set the high bytes of the next write
1609   // This implementation may read more bytes than required.
1672     // read 32 bytes
1694     __ orr(R3, R3, AsmOperand(R4, lsl, lsl_shift)); // merged below low bytes of next val
1783     const int have_bytes = lsl_shift/BitsPerByte; // number of already read bytes in R3
1790       __ cmp_32(count, have_bytes/bytes_per_count); // do we have enough bytes to store?
1829       __ orr(R3, R3, AsmOperand(R4, lsl, lsl_shift), ne); // merged below low bytes of next val
1841       __ orr(R3, R3, AsmOperand(R4, lsl, lsl_shift), ne); // merged below low bytes of next val
1861       __ orr(R3, R3, AsmOperand(R4, lsl, lsl_shift), ne); // merged below low bytes of next val
1873       __ orr(R3, R3, AsmOperand(R4, lsl, lsl_shift), ne); // merged below low bytes of next val
1888       // Note: R3 might contain enough bytes ready to write (3 needed at most),
1899       __ strh(R3, Address(to, 2, post_indexed), ge); // two last bytes
1920   //      bytes_per_count: number of bytes for each unit of 'count'
1921   //      lsl_shift: shift applied to 'old' value to skipped already written bytes
1922   //      lsr_shift: shift applied to 'new' value to set the low bytes of the next write
1942   // This implementation may read more bytes than required.
2023     __ orr(R12, R12, AsmOperand(R10, lsr, lsr_shift)); // merged above high  bytes of prev val
2117     const int have_bytes = lsr_shift/BitsPerByte; // number of already read bytes in R12
2124       __ cmp_32(count, have_bytes/bytes_per_count); // do we have enough bytes to store?
2139         __ logical_shift_left(R12, R12, 4*BitsPerByte); // Promote remaining bytes to MSB
2150         __ logical_shift_left(R12, R12, 2*BitsPerByte); // Promote remaining bytes to MSB
2227         // avoid useless reading R10 when we already have 3 bytes ready in R12
2232       // Note: R12 contains enough bytes ready to write (3 needed at most)
2330   //   shifts 'from' and 'to' by the number of copied bytes in corresponding direction
2333   // Returns maximum number of bytes which may be copied.
2435   //     shifts 'to' by the number of written bytes (so that it becomes the bound of memory to be written)
2469     assert (offset == bytes_to_write * BitsPerByte, "all bytes must be copied");
2492   //     shifts 'to' by the number of copied bytes
2531   //     shifts 'to' by the number of copied bytes
2771     // aligned copy loop which require <=7 bytes for src alignment, and 8 words for aligned copy loop.
3923     __ vld1(V0, Address(from), MacroAssembler::VELEM_SIZE_8, 128); // get 16 bytes of input
4028     __ vld1(V0, Address(from), MacroAssembler::VELEM_SIZE_8, 128); // get 16 bytes of input