[AIEX] Simplify AIEClusterBaseAddress pass #237
Merged
+222
−325
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Collaborator
andcarminati
commented
Nov 15, 2024
andcarminati
commented
- Including a more generic chaining algorithm.
andcarminati
requested review from
abhinay-anubola,
abnikant,
gbossu,
khallouh,
konstantinschwarz,
martien-de-jong,
SagarMaheshwari99 and
stephenneuendorffer
as code owners
|
QoR results: Core Compute Cycle Count:
PM Size effect: -0.09% (basically unaffected). |
| for (auto RegSet : RegPtrUseMap) { | ||
|
|
||
| SmallVector<MachineInstr *, 8> &Instrs = RegSet.second; | ||
| // Chaining aceptance criteria. |
| return true; | ||
|
|
||
| // If the base reg is used in any of the successive MBBs, then we don't | ||
| // want to chain the corresponding ptr adds. Since that would introduce a |
There was a problem hiding this comment.
nit , since (since is a conjunctive of sub-sentences)
andcarminati
commented
Nov 18, 2024
| MINext->getOperand(2).getReg(), *MRI); | ||
|
|
||
| // Evaluate if we should restart the chain from the base | ||
| // pointer. This is necessary whenwe deal with unknonw offsets |
There was a problem hiding this comment.
nit: whenwe, unknonw
andcarminati
force-pushed
the
andreu.cluster.baseaddress.change
branch
from
8a8b895 to
ef3b592
Compare
gbossu
reviewed
Nov 19, 2024
| @@ -71,6 +71,10 @@ static cl::opt<unsigned> StackAddrSpace( | |||
| cl::desc("Specify the addrspace where the stack is allocated " | |||
| "(5: Bank A, 6: Bank B, 7: Bank C, 8: Bank D)")); | |||
|
|
|||
| static cl::opt<bool> EnableAddressChaining("aie-address-chaining", cl::Hidden, | |||
| cl::init(true), | |||
| cl::desc("Enable ptradd chaining.")); | |||
|
Regarding the |
gbossu
reviewed
Nov 21, 2024
| bool processBasicBlock(MachineBasicBlock &MBB, MachineRegisterInfo &MRI, | ||
| MachineIRBuilder &MIB, | ||
| MachineRegisterInfo *MRI = nullptr; | ||
| MachineDominatorTree *DT = nullptr; |
There was a problem hiding this comment.
Nit: Could those be const?
gbossu
reviewed
Nov 21, 2024
| // Get all candidates, i.e. groups of G_PTR_ADDs in the same | ||
| // basic block that shares the same input pointer. | ||
| void getChainingCandidates(MachineBasicBlock &MBB, | ||
| RegToPtrAddMap &RegPtrUseMap) { |
There was a problem hiding this comment.
How about changing this to RegToPtrAddMap getChainingCandidates(MachineBasicBlock &MBB)? This makes it clearer what comes out of this function?
gbossu
reviewed
Nov 21, 2024
| @@ -136,242 +117,152 @@ class AIEClusterBaseAddress : public MachineFunctionPass { | |||
|
|
|||
| StringRef getPassName() const override { return AIE_CLUSTER_BASE_ADDRESS; } | |||
|
|
|||
| using RegToPtrAddMap = std::map<Register, SmallVector<MachineInstr *, 8>>; | |||
gbossu
reviewed
Nov 21, 2024
| bool Changed = false; | ||
|
|
||
| // Get all G_PTR_ADDs that use the same pointer. | ||
| getChainingCandidates(MBB, RegPtrUseMap); |
There was a problem hiding this comment.
Super nit: I would rename this to e.g. collectPtrUses(), because we just collect registers used in ptr_adds, there is not extra processing.
gbossu
reviewed
Nov 21, 2024
| // case, we do not want to chain the addresses, because this would | ||
| // introduce a COPY that increases the pressure on PTR registers. | ||
| // Create chains, when profitable. | ||
| for (auto RegSet : RegPtrUseMap) { |
gbossu
reviewed
Nov 21, 2024
| } | ||
|
|
||
| // Find if a register is used in reachable MBBs. | ||
| bool isRegUsedInReachableMBBs(MachineBasicBlock *MBB, Register Reg) { |
There was a problem hiding this comment.
I would prefer to keep the previous code in that specific case: std::set<MachineBasicBlock *> ReachableMBBs = findReachableMBBs(MBB);. Reachability is different from dominance.
gbossu
reviewed
Nov 21, 2024
| auto Entry = ChainedPtrAdds.find(&MI); | ||
| if (Entry == ChainedPtrAdds.end()) | ||
| // Build a chain (or set of chains) of G_PTR_ADDs. We consider as | ||
| // chain a linear sequence of linked G_PTR_ADDs, tied no output and |
andcarminati
force-pushed
the
andreu.cluster.baseaddress.change
branch
from
ef3b592 to
8f2e71d
Compare
|
Hi @gbossu, all your comments were addressed. Thank you very much! |
* Including a more generic chaining algorithm.
andcarminati
force-pushed
the
andreu.cluster.baseaddress.change
branch
from
8f2e71d to
d4115b7
Compare
gbossu
approved these changes
Nov 22, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.