archive/jdk
1
0
Fork 0
mirror of https://github.com/openjdk/jdk.git synced 2025-08-27 14:54:52 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

8327964: Simplify BigInteger.implMultiplyToLen intrinsic

Browse source 
Reviewed-by: mdoerr, amitkumar, kvn, fyang
This commit is contained in:
Yudi Zheng 2024-05-27 14:25:13 +00:00 • committed by Martin Doerr
parent 08face8c4c
commit ed81a478e1
17 changed files with 64 additions and 135 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
View file

@ -3879,7 +3879,7 @@ void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
* r2: y * r2: y
* r3: ylen * r3: ylen
* r4: z * r4: z
* r5: zlen * r5: tmp0
* r10: tmp1 * r10: tmp1
* r11: tmp2 * r11: tmp2
* r12: tmp3 * r12: tmp3
@ -3890,11 +3890,11 @@ void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
* *
*/ */
void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
Register z, Register zlen, Register z, Register tmp0,
Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp1, Register tmp2, Register tmp3, Register tmp4,
Register tmp5, Register tmp6, Register product_hi) { Register tmp5, Register tmp6, Register product_hi) {
assert_different_registers(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6); assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
const Register idx = tmp1; const Register idx = tmp1;
const Register kdx = tmp2; const Register kdx = tmp2;
@ -3903,7 +3903,7 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
const Register y_idx = tmp4; const Register y_idx = tmp4;
const Register carry = tmp5; const Register carry = tmp5;
const Register product = xlen; const Register product = xlen;
const Register x_xstart = zlen; // reuse register const Register x_xstart = tmp0;
// First Loop. // First Loop.
// //
@ -3919,9 +3919,9 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
// z[xstart] = (int)carry; // z[xstart] = (int)carry;
// //
movw(idx, ylen); // idx = ylen; movw(idx, ylen); // idx = ylen;
movw(kdx, zlen); // kdx = xlen+ylen; addw(kdx, xlen, ylen); // kdx = xlen+ylen;
mov(carry, zr); // carry = 0; mov(carry, zr); // carry = 0;
Label L_done; Label L_done;

2
src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
View file

@ -1510,7 +1510,7 @@ public:
void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state); void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
public: public:
void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp0, Register tmp1, Register tmp2, Register tmp3,
Register tmp4, Register tmp5, Register tmp6, Register tmp7); Register tmp4, Register tmp5, Register tmp6, Register tmp7);
void mul_add(Register out, Register in, Register offs, Register len, Register k); void mul_add(Register out, Register in, Register offs, Register len, Register k);
void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi, void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,

9
src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
View file

@ -4645,7 +4645,6 @@ class StubGenerator: public StubCodeGenerator {
* c_rarg2 - y address * c_rarg2 - y address
* c_rarg3 - y length * c_rarg3 - y length
* c_rarg4 - z address * c_rarg4 - z address
* c_rarg5 - z length
*/ */
address generate_multiplyToLen() { address generate_multiplyToLen() {
__ align(CodeEntryAlignment); __ align(CodeEntryAlignment);
@ -4657,8 +4656,8 @@ class StubGenerator: public StubCodeGenerator {
const Register y = r2; const Register y = r2;
const Register ylen = r3; const Register ylen = r3;
const Register z = r4; const Register z = r4;
const Register zlen = r5;
const Register tmp0 = r5;
const Register tmp1 = r10; const Register tmp1 = r10;
const Register tmp2 = r11; const Register tmp2 = r11;
const Register tmp3 = r12; const Register tmp3 = r12;
@ -4669,7 +4668,7 @@ class StubGenerator: public StubCodeGenerator {
BLOCK_COMMENT("Entry:"); BLOCK_COMMENT("Entry:");
__ enter(); // required for proper stackwalking of RuntimeStub frame __ enter(); // required for proper stackwalking of RuntimeStub frame
__ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7); __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
__ leave(); // required for proper stackwalking of RuntimeStub frame __ leave(); // required for proper stackwalking of RuntimeStub frame
__ ret(lr); __ ret(lr);
@ -4687,10 +4686,10 @@ class StubGenerator: public StubCodeGenerator {
const Register x = r0; const Register x = r0;
const Register xlen = r1; const Register xlen = r1;
const Register z = r2; const Register z = r2;
const Register zlen = r3;
const Register y = r4; // == x const Register y = r4; // == x
const Register ylen = r5; // == xlen const Register ylen = r5; // == xlen
const Register tmp0 = r3;
const Register tmp1 = r10; const Register tmp1 = r10;
const Register tmp2 = r11; const Register tmp2 = r11;
const Register tmp3 = r12; const Register tmp3 = r12;
@ -4705,7 +4704,7 @@ class StubGenerator: public StubCodeGenerator {
__ push(spilled_regs, sp); __ push(spilled_regs, sp);
__ mov(y, x); __ mov(y, x);
__ mov(ylen, xlen); __ mov(ylen, xlen);
__ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7); __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
__ pop(spilled_regs, sp); __ pop(spilled_regs, sp);
__ leave(); __ leave();
__ ret(lr); __ ret(lr);

12
src/hotspot/cpu/ppc/macroAssembler_ppc.cpp
View file

@ -1,5 +1,5 @@
/* /*
* Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012, 2023 SAP SE. All rights reserved. * Copyright (c) 2012, 2023 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* *
@ -3898,7 +3898,7 @@ void MacroAssembler::muladd(Register out, Register in,
void MacroAssembler::multiply_to_len(Register x, Register xlen, void MacroAssembler::multiply_to_len(Register x, Register xlen,
Register y, Register ylen, Register y, Register ylen,
Register z, Register zlen, Register z,
Register tmp1, Register tmp2, Register tmp1, Register tmp2,
Register tmp3, Register tmp4, Register tmp3, Register tmp4,
Register tmp5, Register tmp6, Register tmp5, Register tmp6,
@ -3909,11 +3909,11 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
ShortBranchVerifier sbv(this); ShortBranchVerifier sbv(this);
assert_different_registers(x, xlen, y, ylen, z, zlen, assert_different_registers(x, xlen, y, ylen, z,
tmp1, tmp2, tmp3, tmp4, tmp5, tmp6); tmp1, tmp2, tmp3, tmp4, tmp5, tmp6);
assert_different_registers(x, xlen, y, ylen, z, zlen, assert_different_registers(x, xlen, y, ylen, z,
tmp1, tmp2, tmp3, tmp4, tmp5, tmp7); tmp1, tmp2, tmp3, tmp4, tmp5, tmp7);
assert_different_registers(x, xlen, y, ylen, z, zlen, assert_different_registers(x, xlen, y, ylen, z,
tmp1, tmp2, tmp3, tmp4, tmp5, tmp8); tmp1, tmp2, tmp3, tmp4, tmp5, tmp8);
const Register idx = tmp1; const Register idx = tmp1;
@ -3941,7 +3941,7 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
// z[xstart] = (int)carry; // z[xstart] = (int)carry;
mr_if_needed(idx, ylen); // idx = ylen mr_if_needed(idx, ylen); // idx = ylen
mr_if_needed(kdx, zlen); // kdx = xlen + ylen add(kdx, xlen, ylen); // kdx = xlen + ylen
li(carry, 0); // carry = 0 li(carry, 0); // carry = 0
Label L_done; Label L_done;

4
src/hotspot/cpu/ppc/macroAssembler_ppc.hpp
View file

@ -1,5 +1,5 @@
/* /*
* Copyright (c) 2002, 2023, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2002, 2024, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012, 2023 SAP SE. All rights reserved. * Copyright (c) 2012, 2023 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* *
@ -784,7 +784,7 @@ class MacroAssembler: public Assembler {
Register tmp1, Register tmp2, Register carry); Register tmp1, Register tmp2, Register carry);
void multiply_to_len(Register x, Register xlen, void multiply_to_len(Register x, Register xlen,
Register y, Register ylen, Register y, Register ylen,
Register z, Register zlen, Register z,
Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
Register tmp6, Register tmp7, Register tmp8, Register tmp9, Register tmp10, Register tmp6, Register tmp7, Register tmp8, Register tmp9, Register tmp10,
Register tmp11, Register tmp12, Register tmp13); Register tmp11, Register tmp12, Register tmp13);

5
src/hotspot/cpu/ppc/stubGenerator_ppc.cpp
View file

@ -3204,7 +3204,6 @@ class StubGenerator: public StubCodeGenerator {
// R5 - y address // R5 - y address
// R6 - y length // R6 - y length
// R7 - z address // R7 - z address
// R8 - z length
// //
address generate_multiplyToLen() { address generate_multiplyToLen() {
@ -3217,7 +3216,6 @@ class StubGenerator: public StubCodeGenerator {
const Register y = R5; const Register y = R5;
const Register ylen = R6; const Register ylen = R6;
const Register z = R7; const Register z = R7;
const Register zlen = R8;
const Register tmp1 = R2; // TOC not used. const Register tmp1 = R2; // TOC not used.
const Register tmp2 = R9; const Register tmp2 = R9;
@ -3240,7 +3238,6 @@ class StubGenerator: public StubCodeGenerator {
// C2 does not respect int to long conversion for stub calls. // C2 does not respect int to long conversion for stub calls.
__ clrldi(xlen, xlen, 32); __ clrldi(xlen, xlen, 32);
__ clrldi(ylen, ylen, 32); __ clrldi(ylen, ylen, 32);
__ clrldi(zlen, zlen, 32);
// Save non-volatile regs (frameless). // Save non-volatile regs (frameless).
int current_offs = 8; int current_offs = 8;
@ -3253,7 +3250,7 @@ class StubGenerator: public StubCodeGenerator {
__ std(R30, -current_offs, R1_SP); current_offs += 8; __ std(R30, -current_offs, R1_SP); current_offs += 8;
__ std(R31, -current_offs, R1_SP); __ std(R31, -current_offs, R1_SP);
__ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, __ multiply_to_len(x, xlen, y, ylen, z, tmp1, tmp2, tmp3, tmp4, tmp5,
tmp6, tmp7, tmp8, tmp9, tmp10, tmp11, tmp12, tmp13); tmp6, tmp7, tmp8, tmp9, tmp10, tmp11, tmp12, tmp13);
// Restore non-volatile regs. // Restore non-volatile regs.

14
src/hotspot/cpu/riscv/macroAssembler_riscv.cpp
View file

@ -4120,7 +4120,7 @@ void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
* x12: y * x12: y
* x13: ylen * x13: ylen
* x14: z * x14: z
* x15: zlen * x15: tmp0
* x16: tmp1 * x16: tmp1
* x17: tmp2 * x17: tmp2
* x7: tmp3 * x7: tmp3
@ -4130,10 +4130,10 @@ void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
* x31: tmp7 * x31: tmp7
*/ */
void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
Register z, Register zlen, Register z, Register tmp0,
Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp1, Register tmp2, Register tmp3, Register tmp4,
Register tmp5, Register tmp6, Register product_hi) { Register tmp5, Register tmp6, Register product_hi) {
assert_different_registers(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6); assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6);
const Register idx = tmp1; const Register idx = tmp1;
const Register kdx = tmp2; const Register kdx = tmp2;
@ -4142,11 +4142,11 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
const Register y_idx = tmp4; const Register y_idx = tmp4;
const Register carry = tmp5; const Register carry = tmp5;
const Register product = xlen; const Register product = xlen;
const Register x_xstart = zlen; // reuse register const Register x_xstart = tmp0;
mv(idx, ylen); // idx = ylen; mv(idx, ylen); // idx = ylen;
mv(kdx, zlen); // kdx = xlen+ylen; addw(kdx, xlen, ylen); // kdx = xlen+ylen;
mv(carry, zr); // carry = 0; mv(carry, zr); // carry = 0;
Label L_multiply_64_x_64_loop, L_done; Label L_multiply_64_x_64_loop, L_done;

4
src/hotspot/cpu/riscv/macroAssembler_riscv.hpp
View file

@ -1,5 +1,5 @@
/* /*
* Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved. * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
* Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved. * Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@ -1287,7 +1287,7 @@ public:
Register tmp, Register tmp3, Register tmp4, Register tmp, Register tmp3, Register tmp4,
Register tmp6, Register product_hi); Register tmp6, Register product_hi);
void multiply_to_len(Register x, Register xlen, Register y, Register ylen, void multiply_to_len(Register x, Register xlen, Register y, Register ylen,
Register z, Register zlen, Register z, Register tmp0,
Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp1, Register tmp2, Register tmp3, Register tmp4,
Register tmp5, Register tmp6, Register product_hi); Register tmp5, Register tmp6, Register product_hi);
#endif #endif

9
src/hotspot/cpu/riscv/stubGenerator_riscv.cpp
View file

@ -2840,7 +2840,6 @@ class StubGenerator: public StubCodeGenerator {
* c_rarg2 - y address * c_rarg2 - y address
* c_rarg3 - y length * c_rarg3 - y length
* c_rarg4 - z address * c_rarg4 - z address
* c_rarg5 - z length
*/ */
address generate_multiplyToLen() address generate_multiplyToLen()
{ {
@ -2853,8 +2852,8 @@ class StubGenerator: public StubCodeGenerator {
const Register y = x12; const Register y = x12;
const Register ylen = x13; const Register ylen = x13;
const Register z = x14; const Register z = x14;
const Register zlen = x15;
const Register tmp0 = x15;
const Register tmp1 = x16; const Register tmp1 = x16;
const Register tmp2 = x17; const Register tmp2 = x17;
const Register tmp3 = x7; const Register tmp3 = x7;
@ -2865,7 +2864,7 @@ class StubGenerator: public StubCodeGenerator {
BLOCK_COMMENT("Entry:"); BLOCK_COMMENT("Entry:");
__ enter(); // required for proper stackwalking of RuntimeStub frame __ enter(); // required for proper stackwalking of RuntimeStub frame
__ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7); __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
__ leave(); // required for proper stackwalking of RuntimeStub frame __ leave(); // required for proper stackwalking of RuntimeStub frame
__ ret(); __ ret();
@ -2881,10 +2880,10 @@ class StubGenerator: public StubCodeGenerator {
const Register x = x10; const Register x = x10;
const Register xlen = x11; const Register xlen = x11;
const Register z = x12; const Register z = x12;
const Register zlen = x13;
const Register y = x14; // == x const Register y = x14; // == x
const Register ylen = x15; // == xlen const Register ylen = x15; // == xlen
const Register tmp0 = x13; // zlen, unused
const Register tmp1 = x16; const Register tmp1 = x16;
const Register tmp2 = x17; const Register tmp2 = x17;
const Register tmp3 = x7; const Register tmp3 = x7;
@ -2897,7 +2896,7 @@ class StubGenerator: public StubCodeGenerator {
__ enter(); __ enter();
__ mv(y, x); __ mv(y, x);
__ mv(ylen, xlen); __ mv(ylen, xlen);
__ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7); __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
__ leave(); __ leave();
__ ret(); __ ret();

5
src/hotspot/cpu/s390/macroAssembler_s390.cpp
View file

@ -5281,9 +5281,6 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
z_stmg(Z_R7, Z_R13, _z_abi(gpr7), Z_SP); z_stmg(Z_R7, Z_R13, _z_abi(gpr7), Z_SP);
// In openJdk, we store the argument as 32-bit value to slot.
Address zlen(Z_SP, _z_abi(remaining_cargs)); // Int in long on big endian.
const Register idx = tmp1; const Register idx = tmp1;
const Register kdx = tmp2; const Register kdx = tmp2;
const Register xstart = tmp3; const Register xstart = tmp3;
@ -5308,7 +5305,7 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
// //
lgr_if_needed(idx, ylen); // idx = ylen lgr_if_needed(idx, ylen); // idx = ylen
z_llgf(kdx, zlen); // C2 does not respect int to long conversion for stub calls, thus load zero-extended. z_agrk(kdx, xlen, ylen); // kdx = xlen + ylen
clear_reg(carry); // carry = 0 clear_reg(carry); // carry = 0
Label L_done; Label L_done;

5
src/hotspot/cpu/s390/stubGenerator_s390.cpp
View file

@ -1,5 +1,5 @@
/* /*
* Copyright (c) 2016, 2023, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2016, 2023 SAP SE. All rights reserved. * Copyright (c) 2016, 2023 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* *
@ -2981,7 +2981,6 @@ class StubGenerator: public StubCodeGenerator {
// Z_ARG3 - y address // Z_ARG3 - y address
// Z_ARG4 - y length // Z_ARG4 - y length
// Z_ARG5 - z address // Z_ARG5 - z address
// 160[Z_SP] - z length
address generate_multiplyToLen() { address generate_multiplyToLen() {
__ align(CodeEntryAlignment); __ align(CodeEntryAlignment);
StubCodeMark mark(this, "StubRoutines", "multiplyToLen"); StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
@ -2993,8 +2992,6 @@ class StubGenerator: public StubCodeGenerator {
const Register y = Z_ARG3; const Register y = Z_ARG3;
const Register ylen = Z_ARG4; const Register ylen = Z_ARG4;
const Register z = Z_ARG5; const Register z = Z_ARG5;
// zlen is passed on the stack:
// Address zlen(Z_SP, _z_abi(remaining_cargs));
// Next registers will be saved on stack in multiply_to_len(). // Next registers will be saved on stack in multiply_to_len().
const Register tmp1 = Z_tmp_1; const Register tmp1 = Z_tmp_1;

18
src/hotspot/cpu/x86/macroAssembler_x86.cpp
View file

@ -6983,7 +6983,7 @@ void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
* rsi: y * rsi: y
* rcx: ylen * rcx: ylen
* r8: z * r8: z
* r11: zlen * r11: tmp0
* r12: tmp1 * r12: tmp1
* r13: tmp2 * r13: tmp2
* r14: tmp3 * r14: tmp3
@ -6991,11 +6991,12 @@ void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
* rbx: tmp5 * rbx: tmp5
* *
*/ */
void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen, void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) { Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
ShortBranchVerifier sbv(this); ShortBranchVerifier sbv(this);
assert_different_registers(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx); assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
push(tmp0);
push(tmp1); push(tmp1);
push(tmp2); push(tmp2);
push(tmp3); push(tmp3);
@ -7003,7 +7004,6 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
push(tmp5); push(tmp5);
push(xlen); push(xlen);
push(zlen);
const Register idx = tmp1; const Register idx = tmp1;
const Register kdx = tmp2; const Register kdx = tmp2;
@ -7012,7 +7012,7 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
const Register y_idx = tmp4; const Register y_idx = tmp4;
const Register carry = tmp5; const Register carry = tmp5;
const Register product = xlen; const Register product = xlen;
const Register x_xstart = zlen; // reuse register const Register x_xstart = tmp0;
// First Loop. // First Loop.
// //
@ -7028,9 +7028,9 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
// z[xstart] = (int)carry; // z[xstart] = (int)carry;
// //
movl(idx, ylen); // idx = ylen; movl(idx, ylen); // idx = ylen;
movl(kdx, zlen); // kdx = xlen+ylen; lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
xorq(carry, carry); // carry = 0; xorq(carry, carry); // carry = 0;
Label L_done; Label L_done;
@ -7134,7 +7134,6 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
bind(L_done); bind(L_done);
pop(zlen);
pop(xlen); pop(xlen);
pop(tmp5); pop(tmp5);
@ -7142,6 +7141,7 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
pop(tmp3); pop(tmp3);
pop(tmp2); pop(tmp2);
pop(tmp1); pop(tmp1);
pop(tmp0);
} }
void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale, void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,

2
src/hotspot/cpu/x86/macroAssembler_x86.hpp
View file

@ -2018,7 +2018,7 @@ public:
Register yz_idx, Register idx, Register jdx, Register yz_idx, Register idx, Register jdx,
Register carry, Register product, Register carry, Register product,
Register carry2); Register carry2);
void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen, void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5); Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3, void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3,
Register tmp4, Register tmp5, Register rdxReg, Register raxReg); Register tmp4, Register tmp5, Register rdxReg, Register raxReg);

14
src/hotspot/cpu/x86/stubGenerator_x86_64.cpp
View file

@ -3044,10 +3044,8 @@ address StubGenerator::generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
* c_rarg3 - y length * c_rarg3 - y length
* not Win64 * not Win64
* c_rarg4 - z address * c_rarg4 - z address
* c_rarg5 - z length
* Win64 * Win64
* rsp+40 - z address * rsp+40 - z address
* rsp+48 - z length
*/ */
address StubGenerator::generate_multiplyToLen() { address StubGenerator::generate_multiplyToLen() {
__ align(CodeEntryAlignment); __ align(CodeEntryAlignment);
@ -3061,9 +3059,9 @@ address StubGenerator::generate_multiplyToLen() {
const Register y = rsi; const Register y = rsi;
const Register ylen = rcx; const Register ylen = rcx;
const Register z = r8; const Register z = r8;
const Register zlen = r11;
// Next registers will be saved on stack in multiply_to_len(). // Next registers will be saved on stack in multiply_to_len().
const Register tmp0 = r11;
const Register tmp1 = r12; const Register tmp1 = r12;
const Register tmp2 = r13; const Register tmp2 = r13;
const Register tmp3 = r14; const Register tmp3 = r14;
@ -3073,21 +3071,17 @@ address StubGenerator::generate_multiplyToLen() {
BLOCK_COMMENT("Entry:"); BLOCK_COMMENT("Entry:");
__ enter(); // required for proper stackwalking of RuntimeStub frame __ enter(); // required for proper stackwalking of RuntimeStub frame
#ifndef _WIN64
__ movptr(zlen, r9); // Save r9 in r11 - zlen
#endif
setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
// ylen => rcx, z => r8, zlen => r11 // ylen => rcx, z => r8
// r9 and r10 may be used to save non-volatile registers // r9 and r10 may be used to save non-volatile registers
#ifdef _WIN64 #ifdef _WIN64
// last 2 arguments (#4, #5) are on stack on Win64 // last argument (#4) is on stack on Win64
__ movptr(z, Address(rsp, 6 * wordSize)); __ movptr(z, Address(rsp, 6 * wordSize));
__ movptr(zlen, Address(rsp, 7 * wordSize));
#endif #endif
__ movptr(xlen, rsi); __ movptr(xlen, rsi);
__ movptr(y, rdx); __ movptr(y, rdx);
__ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5); __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
restore_arg_regs(); restore_arg_regs();

72
src/hotspot/share/opto/library_call.cpp
View file

@ -5978,71 +5978,17 @@ bool LibraryCallKit::inline_multiplyToLen() {
return false; return false;
} }
// Set the original stack and the reexecute bit for the interpreter to reexecute Node* x_start = array_element_address(x, intcon(0), x_elem);
// the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens Node* y_start = array_element_address(y, intcon(0), y_elem);
// on the return from z array allocation in runtime. // 'x_start' points to x array + scaled xlen
{ PreserveReexecuteState preexecs(this); // 'y_start' points to y array + scaled ylen
jvms()->set_should_reexecute(true);
Node* x_start = array_element_address(x, intcon(0), x_elem); Node* z_start = array_element_address(z, intcon(0), T_INT);
Node* y_start = array_element_address(y, intcon(0), y_elem);
// 'x_start' points to x array + scaled xlen
// 'y_start' points to y array + scaled ylen
// Allocate the result array Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
Node* zlen = _gvn.transform(new AddINode(xlen, ylen)); OptoRuntime::multiplyToLen_Type(),
ciKlass* klass = ciTypeArrayKlass::make(T_INT); stubAddr, stubName, TypePtr::BOTTOM,
Node* klass_node = makecon(TypeKlassPtr::make(klass)); x_start, xlen, y_start, ylen, z_start);
IdealKit ideal(this);
#define __ ideal.
Node* one = __ ConI(1);
Node* zero = __ ConI(0);
IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
__ set(need_alloc, zero);
__ set(z_alloc, z);
__ if_then(z, BoolTest::eq, null()); {
__ increment (need_alloc, one);
} __ else_(); {
// Update graphKit memory and control from IdealKit.
sync_kit(ideal);
Node* cast = new CastPPNode(control(), z, TypePtr::NOTNULL);
_gvn.set_type(cast, cast->bottom_type());
C->record_for_igvn(cast);
Node* zlen_arg = load_array_length(cast);
// Update IdealKit memory and control from graphKit.
__ sync_kit(this);
__ if_then(zlen_arg, BoolTest::lt, zlen); {
__ increment (need_alloc, one);
} __ end_if();
} __ end_if();
__ if_then(__ value(need_alloc), BoolTest::ne, zero); {
// Update graphKit memory and control from IdealKit.
sync_kit(ideal);
Node * narr = new_array(klass_node, zlen, 1);
// Update IdealKit memory and control from graphKit.
__ sync_kit(this);
__ set(z_alloc, narr);
} __ end_if();
sync_kit(ideal);
z = __ value(z_alloc);
// Can't use TypeAryPtr::INTS which uses Bottom offset.
_gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
// Final sync IdealKit and GraphKit.
final_sync(ideal);
#undef __
Node* z_start = array_element_address(z, intcon(0), T_INT);
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::multiplyToLen_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
x_start, xlen, y_start, ylen, z_start, zlen);
} // original reexecute is set back here
C->set_has_split_ifs(true); // Has chance for split-if optimization C->set_has_split_ifs(true); // Has chance for split-if optimization
set_result(z); set_result(z);

3
src/hotspot/share/opto/runtime.cpp
View file

@ -1150,7 +1150,7 @@ const TypeFunc* OptoRuntime::digestBase_implCompressMB_Type(bool is_sha3) {
const TypeFunc* OptoRuntime::multiplyToLen_Type() { const TypeFunc* OptoRuntime::multiplyToLen_Type() {
// create input type (domain) // create input type (domain)
int num_args = 6; int num_args = 5;
int argcnt = num_args; int argcnt = num_args;
const Type** fields = TypeTuple::fields(argcnt); const Type** fields = TypeTuple::fields(argcnt);
int argp = TypeFunc::Parms; int argp = TypeFunc::Parms;
@ -1159,7 +1159,6 @@ const TypeFunc* OptoRuntime::multiplyToLen_Type() {
fields[argp++] = TypePtr::NOTNULL; // y fields[argp++] = TypePtr::NOTNULL; // y
fields[argp++] = TypeInt::INT; // ylen fields[argp++] = TypeInt::INT; // ylen
fields[argp++] = TypePtr::NOTNULL; // z fields[argp++] = TypePtr::NOTNULL; // z
fields[argp++] = TypeInt::INT; // zlen
assert(argp == TypeFunc::Parms+argcnt, "correct decoding"); assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields); const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);

7
src/java.base/share/classes/java/math/BigInteger.java
View file

@ -1831,6 +1831,10 @@ public class BigInteger extends Number implements Comparable<BigInteger> {
private static int[] multiplyToLen(int[] x, int xlen, int[] y, int ylen, int[] z) { private static int[] multiplyToLen(int[] x, int xlen, int[] y, int ylen, int[] z) {
multiplyToLenCheck(x, xlen); multiplyToLenCheck(x, xlen);
multiplyToLenCheck(y, ylen); multiplyToLenCheck(y, ylen);
if (z == null || z.length < (xlen + ylen))
z = new int[xlen + ylen];
return implMultiplyToLen(x, xlen, y, ylen, z); return implMultiplyToLen(x, xlen, y, ylen, z);
} }
@ -1839,9 +1843,6 @@ public class BigInteger extends Number implements Comparable<BigInteger> {
int xstart = xlen - 1; int xstart = xlen - 1;
int ystart = ylen - 1; int ystart = ylen - 1;
if (z == null || z.length < (xlen+ ylen))
z = new int[xlen+ylen];
long carry = 0; long carry = 0;
for (int j=ystart, k=ystart+1+xstart; j >= 0; j--, k--) { for (int j=ystart, k=ystart+1+xstart; j >= 0; j--, k--) {
long product = (y[j] & LONG_MASK) * long product = (y[j] & LONG_MASK) *