Implement method calls using relative BL on ARM.
Store the linker patches with each CompiledMethod instead of
keeping them in CompilerDriver. Reorganize oat file creation
to apply the patches as we're writing the method code. Add
framework for platform-specific relative call patches in the
OatWriter. Implement relative call patches for ARM.
Change-Id: Ie2effb3d92b61ac8f356140eba09dc37d62290f8
diff --git a/compiler/dex/quick/arm/arm_lir.h b/compiler/dex/quick/arm/arm_lir.h
index b95789e..d935bc3 100644
--- a/compiler/dex/quick/arm/arm_lir.h
+++ b/compiler/dex/quick/arm/arm_lir.h
@@ -528,6 +528,7 @@
kThumb2Vldms, // vldms rd, <list>.
kThumb2Vstms, // vstms rd, <list>.
kThumb2BUncond, // b <label>.
+ kThumb2Bl, // bl with linker fixup. [11110] S imm10 [11] J1 [1] J2 imm11.
kThumb2MovImm16H, // similar to kThumb2MovImm16, but target high hw.
kThumb2AddPCR, // Thumb2 2-operand add with hard-coded PC target.
kThumb2Adr, // Special purpose encoding of ADR for switch tables.
diff --git a/compiler/dex/quick/arm/assemble_arm.cc b/compiler/dex/quick/arm/assemble_arm.cc
index dcec861..cf34948 100644
--- a/compiler/dex/quick/arm/assemble_arm.cc
+++ b/compiler/dex/quick/arm/assemble_arm.cc
@@ -968,6 +968,10 @@
kFmtOff24, -1, -1, kFmtUnused, -1, -1, kFmtUnused, -1, -1,
kFmtUnused, -1, -1, NO_OPERAND | IS_BRANCH,
"b", "!0t", 4, kFixupT2Branch),
+ ENCODING_MAP(kThumb2Bl, 0xf000d000,
+ kFmtOff24, -1, -1, kFmtUnused, -1, -1, kFmtUnused, -1, -1,
+ kFmtUnused, -1, -1, IS_UNARY_OP | IS_BRANCH | REG_DEF_LR | NEEDS_FIXUP,
+ "bl", "!0T", 4, kFixupLabel),
ENCODING_MAP(kThumb2MovImm16H, 0xf2c00000,
kFmtBitBlt, 11, 8, kFmtImm16, -1, -1, kFmtUnused, -1, -1,
kFmtUnused, -1, -1, IS_BINARY_OP | REG_DEF0 | REG_USE0,
diff --git a/compiler/dex/quick/arm/call_arm.cc b/compiler/dex/quick/arm/call_arm.cc
index f6588fe..b721e02 100644
--- a/compiler/dex/quick/arm/call_arm.cc
+++ b/compiler/dex/quick/arm/call_arm.cc
@@ -20,6 +20,8 @@
#include "codegen_arm.h"
#include "dex/quick/mir_to_lir-inl.h"
#include "gc/accounting/card_table.h"
+#include "mirror/art_method.h"
+#include "mirror/object_array-inl.h"
#include "entrypoints/quick/quick_entrypoints.h"
namespace art {
@@ -499,4 +501,117 @@
NewLIR1(kThumbBx, rs_rARM_LR.GetReg());
}
+static bool ArmUseRelativeCall(CompilationUnit* cu, const MethodReference& target_method) {
+ // Emit relative calls only within a dex file due to the limited range of the BL insn.
+ return cu->dex_file == target_method.dex_file;
+}
+
+/*
+ * Bit of a hack here - in the absence of a real scheduling pass,
+ * emit the next instruction in static & direct invoke sequences.
+ */
+static int ArmNextSDCallInsn(CompilationUnit* cu, CallInfo* info,
+ int state, const MethodReference& target_method,
+ uint32_t unused,
+ uintptr_t direct_code, uintptr_t direct_method,
+ InvokeType type) {
+ Mir2Lir* cg = static_cast<Mir2Lir*>(cu->cg.get());
+ if (direct_code != 0 && direct_method != 0) {
+ switch (state) {
+ case 0: // Get the current Method* [sets kArg0]
+ if (direct_code != static_cast<uintptr_t>(-1)) {
+ cg->LoadConstant(cg->TargetPtrReg(kInvokeTgt), direct_code);
+ } else if (ArmUseRelativeCall(cu, target_method)) {
+ // Defer to linker patch.
+ } else {
+ cg->LoadCodeAddress(target_method, type, kInvokeTgt);
+ }
+ if (direct_method != static_cast<uintptr_t>(-1)) {
+ cg->LoadConstant(cg->TargetReg(kArg0, kRef), direct_method);
+ } else {
+ cg->LoadMethodAddress(target_method, type, kArg0);
+ }
+ break;
+ default:
+ return -1;
+ }
+ } else {
+ RegStorage arg0_ref = cg->TargetReg(kArg0, kRef);
+ switch (state) {
+ case 0: // Get the current Method* [sets kArg0]
+ // TUNING: we can save a reg copy if Method* has been promoted.
+ cg->LoadCurrMethodDirect(arg0_ref);
+ break;
+ case 1: // Get method->dex_cache_resolved_methods_
+ cg->LoadRefDisp(arg0_ref,
+ mirror::ArtMethod::DexCacheResolvedMethodsOffset().Int32Value(),
+ arg0_ref,
+ kNotVolatile);
+ // Set up direct code if known.
+ if (direct_code != 0) {
+ if (direct_code != static_cast<uintptr_t>(-1)) {
+ cg->LoadConstant(cg->TargetPtrReg(kInvokeTgt), direct_code);
+ } else if (ArmUseRelativeCall(cu, target_method)) {
+ // Defer to linker patch.
+ } else {
+ CHECK_LT(target_method.dex_method_index, target_method.dex_file->NumMethodIds());
+ cg->LoadCodeAddress(target_method, type, kInvokeTgt);
+ }
+ }
+ break;
+ case 2: // Grab target method*
+ CHECK_EQ(cu->dex_file, target_method.dex_file);
+ cg->LoadRefDisp(arg0_ref,
+ mirror::ObjectArray<mirror::Object>::OffsetOfElement(
+ target_method.dex_method_index).Int32Value(),
+ arg0_ref,
+ kNotVolatile);
+ break;
+ case 3: // Grab the code from the method*
+ if (direct_code == 0) {
+ // kInvokeTgt := arg0_ref->entrypoint
+ cg->LoadWordDisp(arg0_ref,
+ mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value(),
+ cg->TargetPtrReg(kInvokeTgt));
+ }
+ break;
+ default:
+ return -1;
+ }
+ }
+ return state + 1;
+}
+
+NextCallInsn ArmMir2Lir::GetNextSDCallInsn() {
+ return ArmNextSDCallInsn;
+}
+
+LIR* ArmMir2Lir::CallWithLinkerFixup(const MethodReference& target_method, InvokeType type) {
+ // For ARM, just generate a relative BL instruction that will be filled in at 'link time'.
+ // If the target turns out to be too far, the linker will generate a thunk for dispatch.
+ int target_method_idx = target_method.dex_method_index;
+ const DexFile* target_dex_file = target_method.dex_file;
+
+ // Generate the call instruction and save index, dex_file, and type.
+ // NOTE: Method deduplication takes linker patches into account, so we can just pass 0
+ // as a placeholder for the offset.
+ LIR* call = RawLIR(current_dalvik_offset_, kThumb2Bl, 0,
+ target_method_idx, WrapPointer(const_cast<DexFile*>(target_dex_file)), type);
+ AppendLIR(call);
+ call_method_insns_.push_back(call);
+ return call;
+}
+
+LIR* ArmMir2Lir::GenCallInsn(const MirMethodLoweringInfo& method_info) {
+ LIR* call_insn;
+ if (method_info.FastPath() && ArmUseRelativeCall(cu_, method_info.GetTargetMethod()) &&
+ (method_info.GetSharpType() == kDirect || method_info.GetSharpType() == kStatic) &&
+ method_info.DirectCode() == static_cast<uintptr_t>(-1)) {
+ call_insn = CallWithLinkerFixup(method_info.GetTargetMethod(), method_info.GetSharpType());
+ } else {
+ call_insn = OpReg(kOpBlx, TargetPtrReg(kInvokeTgt));
+ }
+ return call_insn;
+}
+
} // namespace art
diff --git a/compiler/dex/quick/arm/codegen_arm.h b/compiler/dex/quick/arm/codegen_arm.h
index 84881e0..932dd87 100644
--- a/compiler/dex/quick/arm/codegen_arm.h
+++ b/compiler/dex/quick/arm/codegen_arm.h
@@ -20,6 +20,7 @@
#include "arm_lir.h"
#include "dex/compiler_internals.h"
#include "dex/quick/mir_to_lir.h"
+#include "utils/arena_containers.h"
namespace art {
@@ -185,6 +186,28 @@
return false; // Wide FPRs are formed by pairing.
}
+ NextCallInsn GetNextSDCallInsn() OVERRIDE;
+
+ /*
+ * @brief Generate a relative call to the method that will be patched at link time.
+ * @param target_method The MethodReference of the method to be invoked.
+ * @param type How the method will be invoked.
+ * @returns Call instruction
+ */
+ LIR* CallWithLinkerFixup(const MethodReference& target_method, InvokeType type);
+
+ /*
+ * @brief Generate the actual call insn based on the method info.
+ * @param method_info the lowering info for the method call.
+ * @returns Call instruction
+ */
+ LIR* GenCallInsn(const MirMethodLoweringInfo& method_info) OVERRIDE;
+
+ /*
+ * @brief Handle ARM specific literals.
+ */
+ void InstallLiteralPools() OVERRIDE;
+
LIR* InvokeTrampoline(OpKind op, RegStorage r_tgt, QuickEntrypointEnum trampoline) OVERRIDE;
size_t GetInstructionOffset(LIR* lir);
@@ -215,6 +238,8 @@
static constexpr ResourceMask GetRegMaskArm(RegStorage reg);
static constexpr ResourceMask EncodeArmRegList(int reg_list);
static constexpr ResourceMask EncodeArmRegFpcsList(int reg_list);
+
+ ArenaVector<LIR*> call_method_insns_;
};
} // namespace art
diff --git a/compiler/dex/quick/arm/target_arm.cc b/compiler/dex/quick/arm/target_arm.cc
index aaf4449..dd8f7fe 100644
--- a/compiler/dex/quick/arm/target_arm.cc
+++ b/compiler/dex/quick/arm/target_arm.cc
@@ -452,6 +452,11 @@
reinterpret_cast<uintptr_t>(base_addr) + lir->offset + 4 + (operand << 1),
lir->target);
break;
+ case 'T':
+ snprintf(tbuf, arraysize(tbuf), "%s", PrettyMethod(
+ static_cast<uint32_t>(lir->operands[1]),
+ *reinterpret_cast<const DexFile*>(UnwrapPointer(lir->operands[2]))).c_str());
+ break;
case 'u': {
int offset_1 = lir->operands[0];
int offset_2 = NEXT_LIR(lir)->operands[0];
@@ -551,7 +556,9 @@
}
ArmMir2Lir::ArmMir2Lir(CompilationUnit* cu, MIRGraph* mir_graph, ArenaAllocator* arena)
- : Mir2Lir(cu, mir_graph, arena) {
+ : Mir2Lir(cu, mir_graph, arena),
+ call_method_insns_(arena->Adapter()) {
+ call_method_insns_.reserve(100);
// Sanity check - make sure encoding map lines up.
for (int i = 0; i < kArmLast; i++) {
if (ArmMir2Lir::EncodingMap[i].opcode != i) {
@@ -823,4 +830,21 @@
return res;
}
+void ArmMir2Lir::InstallLiteralPools() {
+ // PC-relative calls to methods.
+ patches_.reserve(call_method_insns_.size());
+ for (LIR* p : call_method_insns_) {
+ DCHECK_EQ(p->opcode, kThumb2Bl);
+ uint32_t target_method_idx = p->operands[1];
+ const DexFile* target_dex_file =
+ reinterpret_cast<const DexFile*>(UnwrapPointer(p->operands[2]));
+
+ patches_.push_back(LinkerPatch::RelativeCodePatch(p->offset,
+ target_dex_file, target_method_idx));
+ }
+
+ // And do the normal processing.
+ Mir2Lir::InstallLiteralPools();
+}
+
} // namespace art