< prev index next >
src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp
Print this page
rev 54670 : Port of valuetypes to aarch64
@@ -24,10 +24,11 @@
*/
#include "precompiled.hpp"
#include "asm/macroAssembler.hpp"
#include "asm/macroAssembler.inline.hpp"
+#include "classfile/symbolTable.hpp"
#include "code/debugInfoRec.hpp"
#include "code/icBuffer.hpp"
#include "code/vtableStubs.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interp_masm.hpp"
@@ -287,10 +288,11 @@
assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
// fall through
case T_OBJECT:
case T_ARRAY:
case T_ADDRESS:
+ case T_VALUETYPE:
if (int_args < Argument::n_int_register_parameters_j) {
regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
} else {
regs[i].set2(VMRegImpl::stack2reg(stk_args));
stk_args += 2;
@@ -320,10 +322,94 @@
}
return align_up(stk_args, 2);
}
+
+// const uint SharedRuntime::java_return_convention_max_int = Argument::n_int_register_parameters_j+1;
+const uint SharedRuntime::java_return_convention_max_int = 6;
+const uint SharedRuntime::java_return_convention_max_float = Argument::n_float_register_parameters_j;
+
+int SharedRuntime::java_return_convention(const BasicType *sig_bt, VMRegPair *regs, int total_args_passed) {
+
+ // Create the mapping between argument positions and
+ // registers.
+ // r1, r2 used to address klasses and states, exclude it from return convention to avoid colision
+
+ static const Register INT_ArgReg[java_return_convention_max_int] = {
+ r0 /* j_rarg7 */, j_rarg6, j_rarg5, j_rarg4, j_rarg3, j_rarg2
+ };
+
+ static const FloatRegister FP_ArgReg[java_return_convention_max_float] = {
+ j_farg0, j_farg1, j_farg2, j_farg3, j_farg4, j_farg5, j_farg6, j_farg7
+ };
+
+ uint int_args = 0;
+ uint fp_args = 0;
+
+ for (int i = 0; i < total_args_passed; i++) {
+ switch (sig_bt[i]) {
+ case T_BOOLEAN:
+ case T_CHAR:
+ case T_BYTE:
+ case T_SHORT:
+ case T_INT:
+ if (int_args < Argument::n_int_register_parameters_j) {
+ regs[i].set1(INT_ArgReg[int_args]->as_VMReg());
+ int_args ++;
+ } else {
+ // Should we have gurantee here?
+ return -1;
+ }
+ break;
+ case T_VOID:
+ // halves of T_LONG or T_DOUBLE
+ assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
+ regs[i].set_bad();
+ break;
+ case T_LONG:
+ assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
+ // fall through
+ case T_OBJECT:
+ case T_ARRAY:
+ case T_ADDRESS:
+ // Should T_METADATA be added to java_calling_convention as well ?
+ case T_METADATA:
+ case T_VALUETYPE:
+ if (int_args < Argument::n_int_register_parameters_j) {
+ regs[i].set2(INT_ArgReg[int_args]->as_VMReg());
+ int_args ++;
+ } else {
+ return -1;
+ }
+ break;
+ case T_FLOAT:
+ if (fp_args < Argument::n_float_register_parameters_j) {
+ regs[i].set1(FP_ArgReg[fp_args]->as_VMReg());
+ fp_args ++;
+ } else {
+ return -1;
+ }
+ break;
+ case T_DOUBLE:
+ assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
+ if (fp_args < Argument::n_float_register_parameters_j) {
+ regs[i].set2(FP_ArgReg[fp_args]->as_VMReg());
+ fp_args ++;
+ } else {
+ return -1;
+ }
+ break;
+ default:
+ ShouldNotReachHere();
+ break;
+ }
+ }
+
+ return int_args + fp_args;
+}
+
// Patch the callers callsite with entry to compiled code if it exists.
static void patch_callers_callsite(MacroAssembler *masm) {
Label L;
__ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
__ cbz(rscratch1, L);
@@ -350,148 +436,351 @@
// restore sp
__ leave();
__ bind(L);
}
+// For each value type argument, sig includes the list of fields of
+// the value type. This utility function computes the number of
+// arguments for the call if value types are passed by reference (the
+// calling convention the interpreter expects).
+static int compute_total_args_passed_int(const GrowableArray<SigEntry>* sig_extended) {
+ int total_args_passed = 0;
+ if (ValueTypePassFieldsAsArgs) {
+ for (int i = 0; i < sig_extended->length(); i++) {
+ BasicType bt = sig_extended->at(i)._bt;
+ if (SigEntry::is_reserved_entry(sig_extended, i)) {
+ // Ignore reserved entry
+ } else if (bt == T_VALUETYPE) {
+ // In sig_extended, a value type argument starts with:
+ // T_VALUETYPE, followed by the types of the fields of the
+ // value type and T_VOID to mark the end of the value
+ // type. Value types are flattened so, for instance, in the
+ // case of a value type with an int field and a value type
+ // field that itself has 2 fields, an int and a long:
+ // T_VALUETYPE T_INT T_VALUETYPE T_INT T_LONG T_VOID (second
+ // slot for the T_LONG) T_VOID (inner T_VALUETYPE) T_VOID
+ // (outer T_VALUETYPE)
+ total_args_passed++;
+ int vt = 1;
+ do {
+ i++;
+ BasicType bt = sig_extended->at(i)._bt;
+ BasicType prev_bt = sig_extended->at(i-1)._bt;
+ if (bt == T_VALUETYPE) {
+ vt++;
+ } else if (bt == T_VOID &&
+ prev_bt != T_LONG &&
+ prev_bt != T_DOUBLE) {
+ vt--;
+ }
+ } while (vt != 0);
+ } else {
+ total_args_passed++;
+ }
+ }
+ } else {
+ total_args_passed = sig_extended->length();
+ }
+
+ return total_args_passed;
+}
+
+
+static void gen_c2i_adapter_helper(MacroAssembler* masm, BasicType bt, const VMRegPair& reg_pair, int extraspace, const Address& to) {
+
+ assert(bt != T_VALUETYPE || !ValueTypePassFieldsAsArgs, "no value type here");
+
+ // Say 4 args:
+ // i st_off
+ // 0 32 T_LONG
+ // 1 24 T_VOID
+ // 2 16 T_OBJECT
+ // 3 8 T_BOOL
+ // - 0 return address
+ //
+ // However to make thing extra confusing. Because we can fit a long/double in
+ // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
+ // leaves one slot empty and only stores to a single slot. In this case the
+ // slot that is occupied is the T_VOID slot. See I said it was confusing.
+
+ // int next_off = st_off - Interpreter::stackElementSize;
+
+ VMReg r_1 = reg_pair.first();
+ VMReg r_2 = reg_pair.second();
+
+ if (!r_1->is_valid()) {
+ assert(!r_2->is_valid(), "");
+ return;
+ }
+
+ if (r_1->is_stack()) {
+ // memory to memory use rscratch1
+ // DMS CHECK: words_pushed is always 0 ?
+ // int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace + words_pushed * wordSize);
+ int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace);
+ if (!r_2->is_valid()) {
+ // sign extend??
+ __ ldrw(rscratch1, Address(sp, ld_off));
+ __ str(rscratch1, to);
+
+ } else {
+ __ ldr(rscratch1, Address(sp, ld_off));
+ __ str(rscratch1, to);
+ }
+ } else if (r_1->is_Register()) {
+ Register r = r_1->as_Register();
+ __ str(r, to);
+
+ // DMS CHECK: removed redundant if
+ //if (!r_2->is_valid()) {
+ // // must be only an int (or less ) so move only 32bits to slot
+ // // why not sign extend??
+ // __ str(r, to);
+ // } else {
+ // __ str(r, to);
+ // }
+ } else {
+ assert(r_1->is_FloatRegister(), "");
+ if (!r_2->is_valid()) {
+ // only a float use just part of the slot
+ __ strs(r_1->as_FloatRegister(), to);
+ } else {
+ __ strd(r_1->as_FloatRegister(), to);
+ }
+ }
+}
+
static void gen_c2i_adapter(MacroAssembler *masm,
- int total_args_passed,
- int comp_args_on_stack,
- const BasicType *sig_bt,
+ const GrowableArray<SigEntry>* sig_extended,
const VMRegPair *regs,
- Label& skip_fixup) {
+ Label& skip_fixup,
+ address start,
+ OopMapSet* oop_maps,
+ int& frame_complete,
+ int& frame_size_in_words,
+ bool alloc_value_receiver) {
+
// Before we get into the guts of the C2I adapter, see if we should be here
// at all. We've come from compiled code and are attempting to jump to the
// interpreter, which means the caller made a static call to get here
// (vcalls always get a compiled target if there is one). Check for a
// compiled target. If there is one, we need to patch the caller's call.
patch_callers_callsite(masm);
__ bind(skip_fixup);
+ bool has_value_argument = false;
+ if (ValueTypePassFieldsAsArgs) {
+ // Is there a value type argument?
+ for (int i = 0; i < sig_extended->length() && !has_value_argument; i++) {
+ has_value_argument = (sig_extended->at(i)._bt == T_VALUETYPE);
+ }
+ if (has_value_argument) {
+ // There is at least a value type argument: we're coming from
+ // compiled code so we have no buffers to back the value
+ // types. Allocate the buffers here with a runtime call.
+ OopMap* map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
+
+ frame_complete = __ offset();
+ address the_pc = __ pc();
+
+ __ set_last_Java_frame(noreg, noreg, the_pc, rscratch1);
+
+ __ mov(c_rarg0, rthread);
+ __ mov(c_rarg1, r1);
+ __ mov(c_rarg2, (int64_t)alloc_value_receiver);
+
+ __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::allocate_value_types)));
+ __ blrt(rscratch1, 3, 0, 1);
+
+ oop_maps->add_gc_map((int)(__ pc() - start), map);
+ __ reset_last_Java_frame(false);
+
+ RegisterSaver::restore_live_registers(masm);
+
+ Label no_exception;
+ __ ldr(r0, Address(rthread, Thread::pending_exception_offset()));
+ __ cbz(r0, no_exception);
+
+ __ str(zr, Address(rthread, JavaThread::vm_result_offset()));
+ __ ldr(r0, Address(rthread, Thread::pending_exception_offset()));
+ __ b(RuntimeAddress(StubRoutines::forward_exception_entry()));
+
+ __ bind(no_exception);
+
+ // We get an array of objects from the runtime call
+ __ get_vm_result(r10, rthread);
+ __ get_vm_result_2(r1, rthread); // TODO: required to keep the callee Method live?
+ }
+ }
+
int words_pushed = 0;
// Since all args are passed on the stack, total_args_passed *
// Interpreter::stackElementSize is the space we need.
+ int total_args_passed = compute_total_args_passed_int(sig_extended);
int extraspace = total_args_passed * Interpreter::stackElementSize;
__ mov(r13, sp);
// stack is aligned, keep it that way
- extraspace = align_up(extraspace, 2*wordSize);
-
+ extraspace = align_up(extraspace, 2 * wordSize);
if (extraspace)
__ sub(sp, sp, extraspace);
// Now write the args into the outgoing interpreter space
- for (int i = 0; i < total_args_passed; i++) {
- if (sig_bt[i] == T_VOID) {
- assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
+
+ int ignored = 0, next_vt_arg = 0, next_arg_int = 0;
+ bool has_oop_field = false;
+
+ for (int next_arg_comp = 0; next_arg_comp < total_args_passed; next_arg_comp++) {
+ BasicType bt = sig_extended->at(next_arg_comp)._bt;
+ // offset to start parameters
+ int st_off = (total_args_passed - next_arg_int - 1) * Interpreter::stackElementSize;
+
+ if (!ValueTypePassFieldsAsArgs || bt != T_VALUETYPE) {
+ if (SigEntry::is_reserved_entry(sig_extended, next_arg_comp)) {
+ continue; // Ignore reserved entry
+ }
+
+ if (bt == T_VOID) {
+ assert(next_arg_comp > 0 && (sig_extended->at(next_arg_comp - 1)._bt == T_LONG || sig_extended->at(next_arg_comp - 1)._bt == T_DOUBLE), "missing half");
+ next_arg_int ++;
continue;
}
- // offset to start parameters
- int st_off = (total_args_passed - i - 1) * Interpreter::stackElementSize;
int next_off = st_off - Interpreter::stackElementSize;
+ int offset = (bt == T_LONG || bt == T_DOUBLE) ? next_off : st_off;
- // Say 4 args:
- // i st_off
- // 0 32 T_LONG
- // 1 24 T_VOID
- // 2 16 T_OBJECT
- // 3 8 T_BOOL
- // - 0 return address
- //
- // However to make thing extra confusing. Because we can fit a long/double in
- // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
- // leaves one slot empty and only stores to a single slot. In this case the
- // slot that is occupied is the T_VOID slot. See I said it was confusing.
+ gen_c2i_adapter_helper(masm, bt, regs[next_arg_comp], extraspace, Address(sp, offset));
+ next_arg_int ++;
+
+ } else {
+ ignored++;
+ // get the buffer from the just allocated pool of buffers
+ int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + next_vt_arg * type2aelembytes(T_VALUETYPE);
+ // DMS CHECK: Is r11 correct register here and below?
+ __ load_heap_oop(r11, Address(r10, index));
+ next_vt_arg++;
+ next_arg_int++;
+ int vt = 1;
+ // write fields we get from compiled code in registers/stack
+ // slots to the buffer: we know we are done with that value type
+ // argument when we hit the T_VOID that acts as an end of value
+ // type delimiter for this value type. Value types are flattened
+ // so we might encounter embedded value types. Each entry in
+ // sig_extended contains a field offset in the buffer.
+ do {
+ next_arg_comp++;
+ BasicType bt = sig_extended->at(next_arg_comp)._bt;
+ BasicType prev_bt = sig_extended->at(next_arg_comp - 1)._bt;
+ if (bt == T_VALUETYPE) {
+ vt++;
+ ignored++;
+ } else if (bt == T_VOID && prev_bt != T_LONG && prev_bt != T_DOUBLE) {
+ vt--;
+ ignored++;
+ } else if (SigEntry::is_reserved_entry(sig_extended, next_arg_comp)) {
+ // Ignore reserved entry
+ } else {
+ int off = sig_extended->at(next_arg_comp)._offset;
+ assert(off > 0, "offset in object should be positive");
+ gen_c2i_adapter_helper(masm, bt, regs[next_arg_comp - ignored], extraspace, Address(r11, off));
+ }
+ } while (vt != 0);
+ // pass the buffer to the interpreter
+ __ str(r11, Address(sp, st_off));
+ }
+
+ }
- VMReg r_1 = regs[i].first();
- VMReg r_2 = regs[i].second();
+// If a value type was allocated and initialized, apply post barrier to all oop fields
+ if (has_value_argument && has_oop_field) {
+ __ push(r13); // save senderSP
+ __ push(r1); // save callee
+ // Allocate argument register save area
+ if (frame::arg_reg_save_area_bytes != 0) {
+ __ sub(sp, sp, frame::arg_reg_save_area_bytes);
+ }
+ __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::apply_post_barriers), rthread, r10);
+ // De-allocate argument register save area
+ if (frame::arg_reg_save_area_bytes != 0) {
+ __ add(sp, sp, frame::arg_reg_save_area_bytes);
+ }
+ __ pop(r1); // restore callee
+ __ pop(r13); // restore sender SP
+ }
+
+
+ __ mov(esp, sp); // Interp expects args on caller's expression stack
+
+ __ ldr(rscratch1, Address(rmethod, in_bytes(Method::interpreter_entry_offset())));
+ __ br(rscratch1);
+}
+
+
+static void gen_i2c_adapter_helper(MacroAssembler* masm, BasicType bt, const VMRegPair& reg_pair, const Address& from) {
+
+ assert(bt != T_VALUETYPE || !ValueTypePassFieldsAsArgs, "no value type here");
+
+ VMReg r_1 = reg_pair.first();
+ VMReg r_2 = reg_pair.second();
if (!r_1->is_valid()) {
assert(!r_2->is_valid(), "");
- continue;
+ return;
}
+
if (r_1->is_stack()) {
- // memory to memory use rscratch1
- int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size
- + extraspace
- + words_pushed * wordSize);
+ // Convert stack slot to an SP offset (+ wordSize to account for return address )
+ int st_off = r_1->reg2stack() * VMRegImpl::stack_slot_size;
if (!r_2->is_valid()) {
- // sign extend??
- __ ldrw(rscratch1, Address(sp, ld_off));
- __ str(rscratch1, Address(sp, st_off));
-
+ // sign extend???
+ __ ldrsw(rscratch2, from);
+ __ str(rscratch2, Address(sp, st_off));
} else {
+ //
+ // We are using two optoregs. This can be either T_OBJECT,
+ // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
+ // two slots but only uses one for thr T_LONG or T_DOUBLE case
+ // So we must adjust where to pick up the data to match the
+ // interpreter.
+ //
+ // Interpreter local[n] == MSW, local[n+1] == LSW however locals
+ // are accessed as negative so LSW is at LOW address
- __ ldr(rscratch1, Address(sp, ld_off));
-
- // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
- // T_DOUBLE and T_LONG use two slots in the interpreter
- if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
- // ld_off == LSW, ld_off+wordSize == MSW
- // st_off == MSW, next_off == LSW
- __ str(rscratch1, Address(sp, next_off));
-#ifdef ASSERT
- // Overwrite the unused slot with known junk
- __ mov(rscratch1, 0xdeadffffdeadaaaaul);
- __ str(rscratch1, Address(sp, st_off));
-#endif /* ASSERT */
- } else {
- __ str(rscratch1, Address(sp, st_off));
- }
+ // ld_off is MSW so get LSW
+ __ ldr(rscratch2, from);
+ // st_off is LSW (i.e. reg.first())
+ __ str(rscratch2, Address(sp, st_off));
}
- } else if (r_1->is_Register()) {
+ } else if (r_1->is_Register()) { // Register argument
Register r = r_1->as_Register();
- if (!r_2->is_valid()) {
- // must be only an int (or less ) so move only 32bits to slot
- // why not sign extend??
- __ str(r, Address(sp, st_off));
- } else {
- // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
- // T_DOUBLE and T_LONG use two slots in the interpreter
- if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
- // long/double in gpr
-#ifdef ASSERT
- // Overwrite the unused slot with known junk
- __ mov(rscratch1, 0xdeadffffdeadaaabul);
- __ str(rscratch1, Address(sp, st_off));
-#endif /* ASSERT */
- __ str(r, Address(sp, next_off));
+ if (r_2->is_valid()) {
+ //
+ // We are using two VMRegs. This can be either T_OBJECT,
+ // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
+ // two slots but only uses one for thr T_LONG or T_DOUBLE case
+ // So we must adjust where to pick up the data to match the
+ // interpreter.
+ // this can be a misaligned move
+ __ ldr(r, from);
} else {
- __ str(r, Address(sp, st_off));
- }
+ // sign extend and use a full word?
+ __ ldrw(r, from);
}
} else {
- assert(r_1->is_FloatRegister(), "");
if (!r_2->is_valid()) {
- // only a float use just part of the slot
- __ strs(r_1->as_FloatRegister(), Address(sp, st_off));
+ __ ldrs(r_1->as_FloatRegister(), from);
} else {
-#ifdef ASSERT
- // Overwrite the unused slot with known junk
- __ mov(rscratch1, 0xdeadffffdeadaaacul);
- __ str(rscratch1, Address(sp, st_off));
-#endif /* ASSERT */
- __ strd(r_1->as_FloatRegister(), Address(sp, next_off));
+ __ ldrd(r_1->as_FloatRegister(), from);
}
}
- }
-
- __ mov(esp, sp); // Interp expects args on caller's expression stack
-
- __ ldr(rscratch1, Address(rmethod, in_bytes(Method::interpreter_entry_offset())));
- __ br(rscratch1);
}
+void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm, int comp_args_on_stack, const GrowableArray<SigEntry>* sig, const VMRegPair *regs) {
-void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
- int total_args_passed,
- int comp_args_on_stack,
- const BasicType *sig_bt,
- const VMRegPair *regs) {
// Note: r13 contains the senderSP on entry. We must preserve it since
// we may do a i2c -> c2i transition if we lose a race where compiled
// code goes non-entrant while we get args ready.
@@ -547,11 +836,11 @@
__ block_comment("} verify_i2ce ");
#endif
}
// Cut-out for having no stack args.
- int comp_words_on_stack = align_up(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
+ int comp_words_on_stack = align_up(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize) >> LogBytesPerWord;
if (comp_args_on_stack) {
__ sub(rscratch1, sp, comp_words_on_stack * wordSize);
__ andr(sp, rscratch1, -16);
}
@@ -569,86 +858,78 @@
__ str(zr, Address(rthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
__ bind(no_alternative_target);
}
#endif // INCLUDE_JVMCI
+ int total_args_passed = compute_total_args_passed_int(sig);
+
+ int ignored = 0, next_arg_int = 0;
+
// Now generate the shuffle code.
- for (int i = 0; i < total_args_passed; i++) {
- if (sig_bt[i] == T_VOID) {
- assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
- continue;
- }
+ for (int next_arg_comp = 0; next_arg_comp < total_args_passed; next_arg_comp++) {
- // Pick up 0, 1 or 2 words from SP+offset.
+ assert(ignored <= next_arg_comp, "shouldn't skip over more slots than there are arguments");
+ assert(next_arg_int <= total_args_passed, "more arguments from the interpreter than expected?");
- assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
- "scrambled load targets?");
+ BasicType bt = sig->at(next_arg_comp)._bt;
+ int ld_off = (total_args_passed - next_arg_int - 1) * Interpreter::stackElementSize;
+ // Pick up 0, 1 or 2 words from SP+offset.
+ assert(!regs[next_arg_comp].second()->is_valid() || regs[next_arg_comp].first()->next() == regs[next_arg_comp].second(), "scrambled load targets?");
+ //
+ if (!ValueTypePassFieldsAsArgs || bt != T_VALUETYPE) {
// Load in argument order going down.
- int ld_off = (total_args_passed - i - 1)*Interpreter::stackElementSize;
// Point to interpreter value (vs. tag)
- int next_off = ld_off - Interpreter::stackElementSize;
- //
- //
- //
- VMReg r_1 = regs[i].first();
- VMReg r_2 = regs[i].second();
- if (!r_1->is_valid()) {
- assert(!r_2->is_valid(), "");
- continue;
+ if (SigEntry::is_reserved_entry(sig, next_arg_comp)) {
+ continue; // Ignore reserved entry
}
- if (r_1->is_stack()) {
- // Convert stack slot to an SP offset (+ wordSize to account for return address )
- int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size;
- if (!r_2->is_valid()) {
- // sign extend???
- __ ldrsw(rscratch2, Address(esp, ld_off));
- __ str(rscratch2, Address(sp, st_off));
- } else {
- //
- // We are using two optoregs. This can be either T_OBJECT,
- // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
- // two slots but only uses one for thr T_LONG or T_DOUBLE case
- // So we must adjust where to pick up the data to match the
- // interpreter.
- //
- // Interpreter local[n] == MSW, local[n+1] == LSW however locals
- // are accessed as negative so LSW is at LOW address
- // ld_off is MSW so get LSW
- const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
- next_off : ld_off;
- __ ldr(rscratch2, Address(esp, offset));
- // st_off is LSW (i.e. reg.first())
- __ str(rscratch2, Address(sp, st_off));
+ if (bt == T_VOID) {
+ assert(next_arg_comp > 0 && (sig->at(next_arg_comp - 1)._bt == T_LONG || sig->at(next_arg_comp - 1)._bt == T_DOUBLE), "missing half");
+ next_arg_int++;
+ continue;
}
- } else if (r_1->is_Register()) { // Register argument
- Register r = r_1->as_Register();
- if (r_2->is_valid()) {
- //
- // We are using two VMRegs. This can be either T_OBJECT,
- // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
- // two slots but only uses one for thr T_LONG or T_DOUBLE case
- // So we must adjust where to pick up the data to match the
- // interpreter.
- const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
- next_off : ld_off;
+ int next_off = ld_off - Interpreter::stackElementSize;
+ int offset = (bt == T_LONG || bt == T_DOUBLE) ? next_off : ld_off;
+
+ gen_i2c_adapter_helper(masm, bt, regs[next_arg_comp - ignored], Address(esp, offset));
- // this can be a misaligned move
- __ ldr(r, Address(esp, offset));
- } else {
- // sign extend and use a full word?
- __ ldrw(r, Address(esp, ld_off));
- }
+ next_arg_int++;
} else {
- if (!r_2->is_valid()) {
- __ ldrs(r_1->as_FloatRegister(), Address(esp, ld_off));
- } else {
- __ ldrd(r_1->as_FloatRegister(), Address(esp, next_off));
- }
+ next_arg_int++;
+ ignored++;
+ // get the buffer for that value type
+ __ ldr(r10, Address(sp, ld_off));
+ int vt = 1;
+ // load fields to registers/stack slots from the buffer: we know
+ // we are done with that value type argument when we hit the
+ // T_VOID that acts as an end of value type delimiter for this
+ // value type. Value types are flattened so we might encounter
+ // embedded value types. Each entry in sig_extended contains a
+ // field offset in the buffer.
+ do {
+ next_arg_comp++;
+ BasicType bt = sig->at(next_arg_comp)._bt;
+ BasicType prev_bt = sig->at(next_arg_comp - 1)._bt;
+ BasicType bt_int = sig->at(next_arg_int)._bt;
+
+ if (bt == T_VALUETYPE) {
+ vt++;
+ ignored++;
+ } else if (bt == T_VOID && prev_bt != T_LONG && prev_bt != T_DOUBLE) {
+ vt--;
+ ignored++;
+ } else if (SigEntry::is_reserved_entry(sig, next_arg_comp)) {
+ // Ignore reserved entry
+ } else {
+ int off = sig->at(next_arg_comp)._offset;
+ assert(off > 0, "offset in object should be positive");
+ gen_i2c_adapter_helper(masm, bt, regs[next_arg_comp - ignored], Address(r10, off));
}
+ } while (vt != 0);
}
+ } // for
// 6243940 We might end up in handle_wrong_method if
// the callee is deoptimized as we race thru here. If that
// happens we don't want to take a safepoint because the
// caller frame will look interpreted and arguments are now
@@ -657,11 +938,10 @@
// we try and find the callee by normal means a safepoint
// is possible. So we stash the desired callee in the thread
// and the vm will find there should this case occur.
__ str(rmethod, Address(rthread, JavaThread::callee_target_offset()));
-
__ br(rscratch1);
}
#ifdef BUILTIN_SIM
static void generate_i2c_adapter_name(char *result, int total_args_passed, const BasicType *sig_bt)
@@ -728,31 +1008,22 @@
}
#endif
// ---------------------------------------------------------------
AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
- int total_args_passed,
int comp_args_on_stack,
- const BasicType *sig_bt,
- const VMRegPair *regs,
- AdapterFingerPrint* fingerprint) {
+ int comp_args_on_stack_cc,
+ const GrowableArray<SigEntry>* sig,
+ const VMRegPair* regs,
+ const GrowableArray<SigEntry>* sig_cc,
+ const VMRegPair* regs_cc,
+ const GrowableArray<SigEntry>* sig_cc_ro,
+ const VMRegPair* regs_cc_ro,
+ AdapterFingerPrint* fingerprint,
+ AdapterBlob*& new_adapter) {
address i2c_entry = __ pc();
-#ifdef BUILTIN_SIM
- char *name = NULL;
- AArch64Simulator *sim = NULL;
- size_t len = 65536;
- if (NotifySimulator) {
- name = NEW_C_HEAP_ARRAY(char, len, mtInternal);
- }
-
- if (name) {
- generate_i2c_adapter_name(name, total_args_passed, sig_bt);
- sim = AArch64Simulator::get_current(UseSimulatorCache, DisableBCCheck);
- sim->notifyCompile(name, i2c_entry);
- }
-#endif
- gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
+ gen_i2c_adapter(masm, comp_args_on_stack_cc, sig_cc, regs_cc);
address c2i_unverified_entry = __ pc();
Label skip_fixup;
Label ok;
@@ -787,25 +1058,42 @@
__ cbz(rscratch1, skip_fixup);
__ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
__ block_comment("} c2i_unverified_entry");
}
- address c2i_entry = __ pc();
-#ifdef BUILTIN_SIM
- if (name) {
- name[0] = 'c';
- name[2] = 'i';
- sim->notifyCompile(name, c2i_entry);
- FREE_C_HEAP_ARRAY(char, name, mtInternal);
+ OopMapSet* oop_maps = new OopMapSet();
+ int frame_complete = CodeOffsets::frame_never_safe;
+ int frame_size_in_words = 0;
+
+ // Scalarized c2i adapter with non-scalarized receiver (i.e., don't pack receiver)
+ address c2i_value_ro_entry = __ pc();
+ if (regs_cc != regs_cc_ro) {
+ Label unused;
+ gen_c2i_adapter(masm, sig_cc_ro, regs_cc_ro, skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, false);
+ skip_fixup = unused;
}
-#endif
- gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
+ // Scalarized c2i adapter
+ address c2i_entry = __ pc();
+ gen_c2i_adapter(masm, sig_cc, regs_cc, skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, true);
+
+ // Non-scalarized c2i adapter
+ address c2i_value_entry = c2i_entry;
+ if (regs != regs_cc) {
+ c2i_value_entry = __ pc();
+ Label unused;
+ gen_c2i_adapter(masm, sig, regs, unused, i2c_entry, oop_maps, frame_complete, frame_size_in_words, false);
+ }
__ flush();
- return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry);
+
+ // The c2i adapter might safepoint and trigger a GC. The caller must make sure that
+ // the GC knows about the location of oop argument locations passed to the c2i adapter.
+ bool caller_must_gc_arguments = (regs != regs_cc);
+ new_adapter = AdapterBlob::create(masm->code(), frame_complete, frame_size_in_words, oop_maps, caller_must_gc_arguments);
+ return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_value_entry, c2i_value_ro_entry, c2i_unverified_entry);
}
int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
VMRegPair *regs,
VMRegPair *regs2,
@@ -3194,5 +3482,115 @@
// Set exception blob
_exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
}
#endif // COMPILER2_OR_JVMCI
+
+BufferedValueTypeBlob* SharedRuntime::generate_buffered_value_type_adapter(const ValueKlass* vk) {
+ BufferBlob* buf = BufferBlob::create("value types pack/unpack", 16 * K);
+ CodeBuffer buffer(buf);
+ short buffer_locs[20];
+ buffer.insts()->initialize_shared_locs((relocInfo*)buffer_locs,
+ sizeof(buffer_locs)/sizeof(relocInfo));
+
+ MacroAssembler _masm(&buffer);
+ MacroAssembler* masm = &_masm;
+
+ const Array<SigEntry>* sig_vk = vk->extended_sig();
+ const Array<VMRegPair>* regs = vk->return_regs();
+
+ int pack_fields_off = __ offset();
+
+ int j = 1;
+ for (int i = 0; i < sig_vk->length(); i++) {
+ BasicType bt = sig_vk->at(i)._bt;
+ if (bt == T_VALUETYPE) {
+ continue;
+ }
+ if (bt == T_VOID) {
+ if (sig_vk->at(i-1)._bt == T_LONG ||
+ sig_vk->at(i-1)._bt == T_DOUBLE) {
+ j++;
+ }
+ continue;
+ }
+ int off = sig_vk->at(i)._offset;
+ VMRegPair pair = regs->at(j);
+ VMReg r_1 = pair.first();
+ VMReg r_2 = pair.second();
+ Address to(r0, off);
+ if (bt == T_FLOAT) {
+ __ strs(r_1->as_FloatRegister(), to);
+ } else if (bt == T_DOUBLE) {
+ __ strd(r_1->as_FloatRegister(), to);
+ } else if (bt == T_OBJECT || bt == T_ARRAY) {
+ Register val = r_1->as_Register();
+ assert_different_registers(r0, val);
+ // We don't need barriers because the destination is a newly allocated object.
+ // Also, we cannot use store_heap_oop(to, val) because it uses r8 as tmp.
+ if (UseCompressedOops) {
+ __ encode_heap_oop(val);
+ __ str(val, to);
+ } else {
+ __ str(val, to);
+ }
+ } else {
+ assert(is_java_primitive(bt), "unexpected basic type");
+ assert_different_registers(r0, r_1->as_Register());
+ size_t size_in_bytes = type2aelembytes(bt);
+ __ store_sized_value(to, r_1->as_Register(), size_in_bytes);
+ }
+ j++;
+ }
+ assert(j == regs->length(), "missed a field?");
+
+ __ ret(lr);
+
+ int unpack_fields_off = __ offset();
+
+ j = 1;
+ for (int i = 0; i < sig_vk->length(); i++) {
+ BasicType bt = sig_vk->at(i)._bt;
+ if (bt == T_VALUETYPE) {
+ continue;
+ }
+ if (bt == T_VOID) {
+ if (sig_vk->at(i-1)._bt == T_LONG ||
+ sig_vk->at(i-1)._bt == T_DOUBLE) {
+ j++;
+ }
+ continue;
+ }
+ int off = sig_vk->at(i)._offset;
+ VMRegPair pair = regs->at(j);
+ VMReg r_1 = pair.first();
+ VMReg r_2 = pair.second();
+ Address from(r0, off);
+ if (bt == T_FLOAT) {
+ __ ldrs(r_1->as_FloatRegister(), from);
+ } else if (bt == T_DOUBLE) {
+ __ ldrd(r_1->as_FloatRegister(), from);
+ } else if (bt == T_OBJECT || bt == T_ARRAY) {
+ assert_different_registers(r0, r_1->as_Register());
+ __ load_heap_oop(r_1->as_Register(), from);
+ } else {
+ assert(is_java_primitive(bt), "unexpected basic type");
+ assert_different_registers(r0, r_1->as_Register());
+
+ size_t size_in_bytes = type2aelembytes(bt);
+ __ load_sized_value(r_1->as_Register(), from, size_in_bytes, bt != T_CHAR && bt != T_BOOLEAN);
+ }
+ j++;
+ }
+ assert(j == regs->length(), "missed a field?");
+
+ if (StressValueTypeReturnedAsFields) {
+ __ load_klass(r0, r0);
+ __ orr(r0, r0, 1);
+ }
+
+ __ ret(lr);
+
+ __ flush();
+
+ return BufferedValueTypeBlob::create(&buffer, pack_fields_off, unpack_fields_off);
+}
< prev index next >