mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-11-15 20:06:46 +00:00
57810cdac4
labels are generated bracketing each call (not just invokes). This is used to generate entries in the exception table required by the C++ personality. However it gets in the way of tail-merging. This patch solves the problem by no longer placing labels around ordinary calls. Instead we generate entries in the exception table that cover every instruction in the function that wasn't covered by an invoke range (the range given by the labels around the invoke). As an optimization, such entries are only generated for parts of the function that contain a call, since for the moment those are the only instructions that can throw an exception [1]. As a happy consequence, we now get a smaller exception table, since the same region can cover many calls. While there, I also implemented folding of invoke ranges - successive ranges are merged when safe to do so. Finally, if a selector contains only a cleanup, there's a special shorthand for it - place a 0 in the call-site entry. I implemented this while there. As a result, the exception table output (excluding filters) is now optimal - it cannot be made smaller [2]. The problem with throw filters is that folding them optimally is hard, and the benefit of folding them is minimal. [1] I tested that having trapping instructions (eg divide by zero) in such a region doesn't cause trouble. [2] It could be made smaller with the help of higher layers, eg by having branch folding reorder basic blocks ending in invokes with the same landing pad so they follow each other. I don't know if this is worth doing. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@41718 91177308-0d34-0410-b5e6-96231b3b80d8 |
||
---|---|---|
.. | ||
SelectionDAG | ||
AsmPrinter.cpp | ||
BranchFolding.cpp | ||
DwarfWriter.cpp | ||
ELFWriter.cpp | ||
ELFWriter.h | ||
IfConversion.cpp | ||
IntrinsicLowering.cpp | ||
LiveInterval.cpp | ||
LiveIntervalAnalysis.cpp | ||
LiveVariables.cpp | ||
LLVMTargetMachine.cpp | ||
LowerSubregs.cpp | ||
MachineBasicBlock.cpp | ||
MachineFunction.cpp | ||
MachineInstr.cpp | ||
MachineModuleInfo.cpp | ||
MachinePassRegistry.cpp | ||
MachOWriter.cpp | ||
MachOWriter.h | ||
Makefile | ||
Passes.cpp | ||
PHIElimination.cpp | ||
PhysRegTracker.h | ||
PostRASchedulerList.cpp | ||
PrologEpilogInserter.cpp | ||
README.txt | ||
RegAllocBigBlock.cpp | ||
RegAllocLinearScan.cpp | ||
RegAllocLocal.cpp | ||
RegAllocSimple.cpp | ||
RegisterScavenging.cpp | ||
SimpleRegisterCoalescing.cpp | ||
TwoAddressInstructionPass.cpp | ||
UnreachableBlockElim.cpp | ||
VirtRegMap.cpp | ||
VirtRegMap.h |
//===---------------------------------------------------------------------===// Common register allocation / spilling problem: mul lr, r4, lr str lr, [sp, #+52] ldr lr, [r1, #+32] sxth r3, r3 ldr r4, [sp, #+52] mla r4, r3, lr, r4 can be: mul lr, r4, lr mov r4, lr str lr, [sp, #+52] ldr lr, [r1, #+32] sxth r3, r3 mla r4, r3, lr, r4 and then "merge" mul and mov: mul r4, r4, lr str lr, [sp, #+52] ldr lr, [r1, #+32] sxth r3, r3 mla r4, r3, lr, r4 It also increase the likelyhood the store may become dead. //===---------------------------------------------------------------------===// I think we should have a "hasSideEffects" flag (which is automatically set for stuff that "isLoad" "isCall" etc), and the remat pass should eventually be able to remat any instruction that has no side effects, if it can handle it and if profitable. For now, I'd suggest having the remat stuff work like this: 1. I need to spill/reload this thing. 2. Check to see if it has side effects. 3. Check to see if it is simple enough: e.g. it only has one register destination and no register input. 4. If so, clone the instruction, do the xform, etc. Advantages of this are: 1. the .td file describes the behavior of the instructions, not the way the algorithm should work. 2. as remat gets smarter in the future, we shouldn't have to be changing the .td files. 3. it is easier to explain what the flag means in the .td file, because you don't have to pull in the explanation of how the current remat algo works. Some potential added complexities: 1. Some instructions have to be glued to it's predecessor or successor. All of the PC relative instructions and condition code setting instruction. We could mark them as hasSideEffects, but that's not quite right. PC relative loads from constantpools can be remat'ed, for example. But it requires more than just cloning the instruction. Some instructions can be remat'ed but it expands to more than one instruction. But allocator will have to make a decision. 4. As stated in 3, not as simple as cloning in some cases. The target will have to decide how to remat it. For example, an ARM 2-piece constant generation instruction is remat'ed as a load from constantpool. //===---------------------------------------------------------------------===// bb27 ... ... %reg1037 = ADDri %reg1039, 1 %reg1038 = ADDrs %reg1032, %reg1039, %NOREG, 10 Successors according to CFG: 0x8b03bf0 (#5) bb76 (0x8b03bf0, LLVM BB @0x8b032d0, ID#5): Predecessors according to CFG: 0x8b0c5f0 (#3) 0x8b0a7c0 (#4) %reg1039 = PHI %reg1070, mbb<bb76.outer,0x8b0c5f0>, %reg1037, mbb<bb27,0x8b0a7c0> Note ADDri is not a two-address instruction. However, its result %reg1037 is an operand of the PHI node in bb76 and its operand %reg1039 is the result of the PHI node. We should treat it as a two-address code and make sure the ADDri is scheduled after any node that reads %reg1039. //===---------------------------------------------------------------------===// Use local info (i.e. register scavenger) to assign it a free register to allow reuse: ldr r3, [sp, #+4] add r3, r3, #3 ldr r2, [sp, #+8] add r2, r2, #2 ldr r1, [sp, #+4] <== add r1, r1, #1 ldr r0, [sp, #+4] add r0, r0, #2 //===---------------------------------------------------------------------===// LLVM aggressively lift CSE out of loop. Sometimes this can be negative side- effects: R1 = X + 4 R2 = X + 7 R3 = X + 15 loop: load [i + R1] ... load [i + R2] ... load [i + R3] Suppose there is high register pressure, R1, R2, R3, can be spilled. We need to implement proper re-materialization to handle this: R1 = X + 4 R2 = X + 7 R3 = X + 15 loop: R1 = X + 4 @ re-materialized load [i + R1] ... R2 = X + 7 @ re-materialized load [i + R2] ... R3 = X + 15 @ re-materialized load [i + R3] Furthermore, with re-association, we can enable sharing: R1 = X + 4 R2 = X + 7 R3 = X + 15 loop: T = i + X load [T + 4] ... load [T + 7] ... load [T + 15] //===---------------------------------------------------------------------===//