Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(465)

Side by Side Diff: runtime/vm/flow_graph_compiler_arm64.cc

Issue 295243005: Adds more SIMD instructions to arm64. (Closed) Base URL: http://dart.googlecode.com/svn/branches/bleeding_edge/dart/
Patch Set: Created 6 years, 7 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch | Annotate | Revision Log
« no previous file with comments | « runtime/vm/disassembler_arm64.cc ('k') | runtime/vm/intermediate_language_arm.cc » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 // Copyright (c) 2014, the Dart project authors. Please see the AUTHORS file 1 // Copyright (c) 2014, the Dart project authors. Please see the AUTHORS file
2 // for details. All rights reserved. Use of this source code is governed by a 2 // for details. All rights reserved. Use of this source code is governed by a
3 // BSD-style license that can be found in the LICENSE file. 3 // BSD-style license that can be found in the LICENSE file.
4 4
5 #include "vm/globals.h" // Needed here to get TARGET_ARCH_ARM64. 5 #include "vm/globals.h" // Needed here to get TARGET_ARCH_ARM64.
6 #if defined(TARGET_ARCH_ARM64) 6 #if defined(TARGET_ARCH_ARM64)
7 7
8 #include "vm/flow_graph_compiler.h" 8 #include "vm/flow_graph_compiler.h"
9 9
10 #include "vm/ast_printer.h" 10 #include "vm/ast_printer.h"
(...skipping 1422 matching lines...) Expand 10 before | Expand all | Expand 10 after
1433 void FlowGraphCompiler::SaveLiveRegisters(LocationSummary* locs) { 1433 void FlowGraphCompiler::SaveLiveRegisters(LocationSummary* locs) {
1434 // TODO(vegorov): consider saving only caller save (volatile) registers. 1434 // TODO(vegorov): consider saving only caller save (volatile) registers.
1435 const intptr_t fpu_regs_count = locs->live_registers()->FpuRegisterCount(); 1435 const intptr_t fpu_regs_count = locs->live_registers()->FpuRegisterCount();
1436 if (fpu_regs_count > 0) { 1436 if (fpu_regs_count > 0) {
1437 // Store fpu registers with the lowest register number at the lowest 1437 // Store fpu registers with the lowest register number at the lowest
1438 // address. 1438 // address.
1439 for (intptr_t reg_idx = kNumberOfVRegisters - 1; 1439 for (intptr_t reg_idx = kNumberOfVRegisters - 1;
1440 reg_idx >= 0; --reg_idx) { 1440 reg_idx >= 0; --reg_idx) {
1441 VRegister fpu_reg = static_cast<VRegister>(reg_idx); 1441 VRegister fpu_reg = static_cast<VRegister>(reg_idx);
1442 if (locs->live_registers()->ContainsFpuRegister(fpu_reg)) { 1442 if (locs->live_registers()->ContainsFpuRegister(fpu_reg)) {
1443 // TODO(zra): Save the whole V register. 1443 __ PushQuad(fpu_reg);
1444 __ PushDouble(fpu_reg);
1445 } 1444 }
1446 } 1445 }
1447 } 1446 }
1448 1447
1449 // Store general purpose registers with the highest register number at the 1448 // Store general purpose registers with the highest register number at the
1450 // lowest address. 1449 // lowest address.
1451 for (intptr_t reg_idx = 0; reg_idx < kNumberOfCpuRegisters; ++reg_idx) { 1450 for (intptr_t reg_idx = 0; reg_idx < kNumberOfCpuRegisters; ++reg_idx) {
1452 Register reg = static_cast<Register>(reg_idx); 1451 Register reg = static_cast<Register>(reg_idx);
1453 if (locs->live_registers()->ContainsRegister(reg)) { 1452 if (locs->live_registers()->ContainsRegister(reg)) {
1454 __ Push(reg); 1453 __ Push(reg);
(...skipping 11 matching lines...) Expand all
1466 __ Pop(reg); 1465 __ Pop(reg);
1467 } 1466 }
1468 } 1467 }
1469 1468
1470 const intptr_t fpu_regs_count = locs->live_registers()->FpuRegisterCount(); 1469 const intptr_t fpu_regs_count = locs->live_registers()->FpuRegisterCount();
1471 if (fpu_regs_count > 0) { 1470 if (fpu_regs_count > 0) {
1472 // Fpu registers have the lowest register number at the lowest address. 1471 // Fpu registers have the lowest register number at the lowest address.
1473 for (intptr_t reg_idx = 0; reg_idx < kNumberOfVRegisters; ++reg_idx) { 1472 for (intptr_t reg_idx = 0; reg_idx < kNumberOfVRegisters; ++reg_idx) {
1474 VRegister fpu_reg = static_cast<VRegister>(reg_idx); 1473 VRegister fpu_reg = static_cast<VRegister>(reg_idx);
1475 if (locs->live_registers()->ContainsFpuRegister(fpu_reg)) { 1474 if (locs->live_registers()->ContainsFpuRegister(fpu_reg)) {
1476 // TODO(zra): Restore the whole V register. 1475 __ PopQuad(fpu_reg);
1477 __ PopDouble(fpu_reg);
1478 } 1476 }
1479 } 1477 }
1480 } 1478 }
1481 } 1479 }
1482 1480
1483 1481
1484 void FlowGraphCompiler::EmitTestAndCall(const ICData& ic_data, 1482 void FlowGraphCompiler::EmitTestAndCall(const ICData& ic_data,
1485 Register class_id_reg, 1483 Register class_id_reg,
1486 intptr_t argument_count, 1484 intptr_t argument_count,
1487 const Array& argument_names, 1485 const Array& argument_names,
(...skipping 100 matching lines...) Expand 10 before | Expand all | Expand 10 after
1588 __ LoadFromOffset(destination.reg(), FP, source_offset, PP); 1586 __ LoadFromOffset(destination.reg(), FP, source_offset, PP);
1589 } else { 1587 } else {
1590 ASSERT(destination.IsStackSlot()); 1588 ASSERT(destination.IsStackSlot());
1591 const intptr_t source_offset = source.ToStackSlotOffset(); 1589 const intptr_t source_offset = source.ToStackSlotOffset();
1592 const intptr_t dest_offset = destination.ToStackSlotOffset(); 1590 const intptr_t dest_offset = destination.ToStackSlotOffset();
1593 __ LoadFromOffset(TMP, FP, source_offset, PP); 1591 __ LoadFromOffset(TMP, FP, source_offset, PP);
1594 __ StoreToOffset(TMP, FP, dest_offset, PP); 1592 __ StoreToOffset(TMP, FP, dest_offset, PP);
1595 } 1593 }
1596 } else if (source.IsFpuRegister()) { 1594 } else if (source.IsFpuRegister()) {
1597 if (destination.IsFpuRegister()) { 1595 if (destination.IsFpuRegister()) {
1598 __ fmovdd(destination.fpu_reg(), source.fpu_reg()); 1596 __ vmov(destination.fpu_reg(), source.fpu_reg());
1599 } else { 1597 } else {
1600 if (destination.IsDoubleStackSlot()) { 1598 if (destination.IsDoubleStackSlot()) {
1601 const intptr_t dest_offset = destination.ToStackSlotOffset(); 1599 const intptr_t dest_offset = destination.ToStackSlotOffset();
1602 VRegister src = source.fpu_reg(); 1600 VRegister src = source.fpu_reg();
1603 __ StoreDToOffset(src, FP, dest_offset, PP); 1601 __ StoreDToOffset(src, FP, dest_offset, PP);
1604 } else { 1602 } else {
1605 ASSERT(destination.IsQuadStackSlot()); 1603 ASSERT(destination.IsQuadStackSlot());
1606 UNIMPLEMENTED(); 1604 const intptr_t dest_offset = destination.ToStackSlotOffset();
1605 __ StoreQToOffset(source.fpu_reg(), FP, dest_offset, PP);
1607 } 1606 }
1608 } 1607 }
1609 } else if (source.IsDoubleStackSlot()) { 1608 } else if (source.IsDoubleStackSlot()) {
1610 if (destination.IsFpuRegister()) { 1609 if (destination.IsFpuRegister()) {
1611 const intptr_t dest_offset = source.ToStackSlotOffset(); 1610 const intptr_t dest_offset = source.ToStackSlotOffset();
1612 const VRegister dst = destination.fpu_reg(); 1611 const VRegister dst = destination.fpu_reg();
1613 __ LoadDFromOffset(dst, FP, dest_offset, PP); 1612 __ LoadDFromOffset(dst, FP, dest_offset, PP);
1614 } else { 1613 } else {
1615 ASSERT(destination.IsDoubleStackSlot()); 1614 ASSERT(destination.IsDoubleStackSlot());
1616 const intptr_t source_offset = source.ToStackSlotOffset(); 1615 const intptr_t source_offset = source.ToStackSlotOffset();
1617 const intptr_t dest_offset = destination.ToStackSlotOffset(); 1616 const intptr_t dest_offset = destination.ToStackSlotOffset();
1618 __ LoadDFromOffset(VTMP, FP, source_offset, PP); 1617 __ LoadDFromOffset(VTMP, FP, source_offset, PP);
1619 __ StoreDToOffset(VTMP, FP, dest_offset, PP); 1618 __ StoreDToOffset(VTMP, FP, dest_offset, PP);
1620 } 1619 }
1621 } else if (source.IsQuadStackSlot()) { 1620 } else if (source.IsQuadStackSlot()) {
1622 UNIMPLEMENTED(); 1621 if (destination.IsFpuRegister()) {
1622 const intptr_t dest_offset = source.ToStackSlotOffset();
1623 __ LoadQFromOffset(destination.fpu_reg(), FP, dest_offset, PP);
1624 } else {
1625 ASSERT(destination.IsQuadStackSlot());
1626 const intptr_t source_offset = source.ToStackSlotOffset();
1627 const intptr_t dest_offset = destination.ToStackSlotOffset();
1628 __ LoadQFromOffset(VTMP, FP, source_offset, PP);
1629 __ StoreQToOffset(VTMP, FP, dest_offset, PP);
1630 }
1623 } else { 1631 } else {
1624 ASSERT(source.IsConstant()); 1632 ASSERT(source.IsConstant());
1625 const Object& constant = source.constant(); 1633 const Object& constant = source.constant();
1626 if (destination.IsRegister()) { 1634 if (destination.IsRegister()) {
1627 __ LoadObject(destination.reg(), constant, PP); 1635 __ LoadObject(destination.reg(), constant, PP);
1628 } else if (destination.IsFpuRegister()) { 1636 } else if (destination.IsFpuRegister()) {
1629 const VRegister dst = destination.fpu_reg(); 1637 const VRegister dst = destination.fpu_reg();
1630 __ LoadObject(TMP, constant, PP); 1638 __ LoadObject(TMP, constant, PP);
1631 __ LoadDFieldFromOffset(dst, TMP, Double::value_offset(), PP); 1639 __ LoadDFieldFromOffset(dst, TMP, Double::value_offset(), PP);
1632 } else if (destination.IsDoubleStackSlot()) { 1640 } else if (destination.IsDoubleStackSlot()) {
(...skipping 144 matching lines...) Expand 10 before | Expand all | Expand 10 after
1777 void ParallelMoveResolver::RestoreFpuScratch(FpuRegister reg) { 1785 void ParallelMoveResolver::RestoreFpuScratch(FpuRegister reg) {
1778 UNIMPLEMENTED(); 1786 UNIMPLEMENTED();
1779 } 1787 }
1780 1788
1781 1789
1782 #undef __ 1790 #undef __
1783 1791
1784 } // namespace dart 1792 } // namespace dart
1785 1793
1786 #endif // defined TARGET_ARCH_ARM64 1794 #endif // defined TARGET_ARCH_ARM64
OLDNEW
« no previous file with comments | « runtime/vm/disassembler_arm64.cc ('k') | runtime/vm/intermediate_language_arm.cc » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698