8327964: Simplify BigInteger.implMultiplyToLen intrinsic

Reviewed-by: mdoerr, amitkumar, kvn, fyang
azvegint · May 27, 2024 · ed81a47 · ed81a47
1 parent 08face8
commit ed81a47
Show file tree

Hide file tree

Showing 17 changed files with 64 additions and 135 deletions.
diff --git a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
@@ -3879,7 +3879,7 @@ void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
  * r2: y
  * r3: ylen
  * r4:  z
- * r5: zlen
+ * r5: tmp0
  * r10: tmp1
  * r11: tmp2
  * 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,
-                                     Register z, Register zlen,
+                                     Register z, Register tmp0,
                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
                                      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 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 carry = tmp5;
   const Register product  = xlen;
-  const Register x_xstart = zlen;  // reuse register
+  const Register x_xstart = tmp0;
 
   // First Loop.
   //
@@ -3919,9 +3919,9 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Regi
   //  z[xstart] = (int)carry;
   //
 
-  movw(idx, ylen);      // idx = ylen;
-  movw(kdx, zlen);      // kdx = xlen+ylen;
-  mov(carry, zr);       // carry = 0;
+  movw(idx, ylen);       // idx = ylen;
+  addw(kdx, xlen, ylen); // kdx = xlen+ylen;
+  mov(carry, zr);        // carry = 0;
 
   Label L_done;
 

diff --git a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
@@ -1510,7 +1510,7 @@ class MacroAssembler: public Assembler {
   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
 public:
   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);
   void mul_add(Register out, Register in, Register offs, Register len, Register k);
   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,

diff --git a/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp b/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
@@ -4645,7 +4645,6 @@ class StubGenerator: public StubCodeGenerator {
    *    c_rarg2   - y address
    *    c_rarg3   - y length
    *    c_rarg4   - z address
-   *    c_rarg5   - z length
    */
   address generate_multiplyToLen() {
     __ align(CodeEntryAlignment);
@@ -4657,8 +4656,8 @@ class StubGenerator: public StubCodeGenerator {
     const Register y     = r2;
     const Register ylen  = r3;
     const Register z     = r4;
-    const Register zlen  = r5;
 
+    const Register tmp0  = r5;
     const Register tmp1  = r10;
     const Register tmp2  = r11;
     const Register tmp3  = r12;
@@ -4669,7 +4668,7 @@ class StubGenerator: public StubCodeGenerator {
 
     BLOCK_COMMENT("Entry:");
     __ 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
     __ ret(lr);
 
@@ -4687,10 +4686,10 @@ class StubGenerator: public StubCodeGenerator {
     const Register x     = r0;
     const Register xlen  = r1;
     const Register z     = r2;
-    const Register zlen  = r3;
     const Register y     = r4; // == x
     const Register ylen  = r5; // == xlen
 
+    const Register tmp0  = r3;
     const Register tmp1  = r10;
     const Register tmp2  = r11;
     const Register tmp3  = r12;
@@ -4705,7 +4704,7 @@ class StubGenerator: public StubCodeGenerator {
     __ push(spilled_regs, sp);
     __ mov(y, x);
     __ 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);
     __ leave();
     __ ret(lr);

diff --git a/src/hotspot/cpu/ppc/macroAssembler_ppc.cpp b/src/hotspot/cpu/ppc/macroAssembler_ppc.cpp
@@ -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.
  * 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,
                                      Register y, Register ylen,
-                                     Register z, Register zlen,
+                                     Register z,
                                      Register tmp1, Register tmp2,
                                      Register tmp3, Register tmp4,
                                      Register tmp5, Register tmp6,
@@ -3909,11 +3909,11 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
 
   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);
-  assert_different_registers(x, xlen, y, ylen, z, zlen,
+  assert_different_registers(x, xlen, y, ylen, z,
                              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);
 
   const Register idx = tmp1;
@@ -3941,7 +3941,7 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
   //  z[xstart] = (int)carry;
 
   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
 
   Label L_done;

diff --git a/src/hotspot/cpu/ppc/macroAssembler_ppc.hpp b/src/hotspot/cpu/ppc/macroAssembler_ppc.hpp
@@ -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.
  * 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);
   void multiply_to_len(Register x, Register xlen,
                        Register y, Register ylen,
-                       Register z, Register zlen,
+                       Register z,
                        Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
                        Register tmp6, Register tmp7, Register tmp8, Register tmp9, Register tmp10,
                        Register tmp11, Register tmp12, Register tmp13);

diff --git a/src/hotspot/cpu/ppc/stubGenerator_ppc.cpp b/src/hotspot/cpu/ppc/stubGenerator_ppc.cpp
@@ -3204,7 +3204,6 @@ class StubGenerator: public StubCodeGenerator {
   //    R5 - y address
   //    R6 - y length
   //    R7 - z address
-  //    R8 - z length
   //
   address generate_multiplyToLen() {
 
@@ -3217,7 +3216,6 @@ class StubGenerator: public StubCodeGenerator {
     const Register y     = R5;
     const Register ylen  = R6;
     const Register z     = R7;
-    const Register zlen  = R8;
 
     const Register tmp1  = R2; // TOC not used.
     const Register tmp2  = R9;
@@ -3240,7 +3238,6 @@ class StubGenerator: public StubCodeGenerator {
     // C2 does not respect int to long conversion for stub calls.
     __ clrldi(xlen, xlen, 32);
     __ clrldi(ylen, ylen, 32);
-    __ clrldi(zlen, zlen, 32);
 
     // Save non-volatile regs (frameless).
     int current_offs = 8;
@@ -3253,7 +3250,7 @@ class StubGenerator: public StubCodeGenerator {
     __ std(R30, -current_offs, R1_SP); current_offs += 8;
     __ 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);
 
     // Restore non-volatile regs.

diff --git a/src/hotspot/cpu/riscv/macroAssembler_riscv.cpp b/src/hotspot/cpu/riscv/macroAssembler_riscv.cpp
@@ -4120,7 +4120,7 @@ void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
  * x12: y
  * x13: ylen
  * x14: z
- * x15: zlen
+ * x15: tmp0
  * x16: tmp1
  * x17: tmp2
  * x7:  tmp3
@@ -4130,10 +4130,10 @@ void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
  * x31: tmp7
  */
 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 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 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 carry = tmp5;
   const Register product = xlen;
-  const Register x_xstart = zlen; // reuse register
+  const Register x_xstart = tmp0;
 
-  mv(idx, ylen); // idx = ylen;
-  mv(kdx, zlen); // kdx = xlen+ylen;
-  mv(carry, zr); // carry = 0;
+  mv(idx, ylen);         // idx = ylen;
+  addw(kdx, xlen, ylen); // kdx = xlen+ylen;
+  mv(carry, zr);         // carry = 0;
 
   Label L_multiply_64_x_64_loop, L_done;
 

diff --git a/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp b/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp
@@ -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) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -1287,7 +1287,7 @@ class MacroAssembler: public Assembler {
                                Register tmp, Register tmp3, Register tmp4,
                                Register tmp6, Register product_hi);
   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 tmp5, Register tmp6, Register product_hi);
 #endif

diff --git a/src/hotspot/cpu/riscv/stubGenerator_riscv.cpp b/src/hotspot/cpu/riscv/stubGenerator_riscv.cpp
@@ -2840,7 +2840,6 @@ class StubGenerator: public StubCodeGenerator {
    *    c_rarg2   - y address
    *    c_rarg3   - y length
    *    c_rarg4   - z address
-   *    c_rarg5   - z length
    */
   address generate_multiplyToLen()
   {
@@ -2853,8 +2852,8 @@ class StubGenerator: public StubCodeGenerator {
     const Register y     = x12;
     const Register ylen  = x13;
     const Register z     = x14;
-    const Register zlen  = x15;
 
+    const Register tmp0  = x15;
     const Register tmp1  = x16;
     const Register tmp2  = x17;
     const Register tmp3  = x7;
@@ -2865,7 +2864,7 @@ class StubGenerator: public StubCodeGenerator {
 
     BLOCK_COMMENT("Entry:");
     __ 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
     __ ret();
 
@@ -2881,10 +2880,10 @@ class StubGenerator: public StubCodeGenerator {
     const Register x     = x10;
     const Register xlen  = x11;
     const Register z     = x12;
-    const Register zlen  = x13;
     const Register y     = x14; // == x
     const Register ylen  = x15; // == xlen
 
+    const Register tmp0  = x13; // zlen, unused
     const Register tmp1  = x16;
     const Register tmp2  = x17;
     const Register tmp3  = x7;
@@ -2897,7 +2896,7 @@ class StubGenerator: public StubCodeGenerator {
     __ enter();
     __ mv(y, x);
     __ 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();
     __ ret();
 

diff --git a/src/hotspot/cpu/s390/macroAssembler_s390.cpp b/src/hotspot/cpu/s390/macroAssembler_s390.cpp
@@ -5281,9 +5281,6 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
 
   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 kdx = tmp2;
   const Register xstart = tmp3;
@@ -5308,7 +5305,7 @@ void MacroAssembler::multiply_to_len(Register x, Register xlen,
   //
 
   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
 
   Label L_done;

diff --git a/src/hotspot/cpu/s390/stubGenerator_s390.cpp b/src/hotspot/cpu/s390/stubGenerator_s390.cpp
@@ -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.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -2981,7 +2981,6 @@ class StubGenerator: public StubCodeGenerator {
   //   Z_ARG3    - y address
   //   Z_ARG4    - y length
   //   Z_ARG5    - z address
-  //   160[Z_SP] - z length
   address generate_multiplyToLen() {
     __ align(CodeEntryAlignment);
     StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
@@ -2993,8 +2992,6 @@ class StubGenerator: public StubCodeGenerator {
     const Register y    = Z_ARG3;
     const Register ylen = Z_ARG4;
     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().
     const Register tmp1 = Z_tmp_1;