llvm-project.git/llvm/lib/Analysis/InstructionPrecedenceTracking.cpp, branch main Unnamed repository; edit this file 'description' to name the repository. [IPT] Remove ipt.NumInstScanned statistic (#168515) 2025-11-19T16:14:50+00:00 Luke Lau luke@igalia.com 2025-11-19T16:14:50+00:00 f3d8a5cf5bc058328004bd0905ab4974f09fc131 The NumInstScanned statistic is non-determinstic across multiple identical invocations of LLVM, and leads to noise when trying to diff LLVM statistics with e.g. ./utils/tdiff.py in llvm-test-suite. My understanding is that it's non-deterministic because the users of IPT's hasSpecialInstructions/isPreceededBySpecialInstruction API aren't deterministic themselves. This PR removes it and fixes #157598. This is just a small quality-of-life improvement for the ./utils/tdiff.py workflow, but happy to leave the statistic in if others are using it. 
The NumInstScanned statistic is non-determinstic across multiple
identical invocations of LLVM, and leads to noise when trying to diff
LLVM statistics with e.g. ./utils/tdiff.py in llvm-test-suite.

My understanding is that it's non-deterministic because the users of
IPT's hasSpecialInstructions/isPreceededBySpecialInstruction API aren't
deterministic themselves.

This PR removes it and fixes #157598. This is just a small
quality-of-life improvement for the ./utils/tdiff.py workflow, but happy
to leave the statistic in if others are using it.
 [Analysis] Avoid repeated hash lookups (NFC) (#132584) 2025-03-23T14:36:19+00:00 Kazu Hirata kazu@google.com 2025-03-23T14:36:19+00:00 ce2c4ea658239771dec542249a77dcf2056d8b8a FirstSpecialInsts caches the first special instruction for each basic block. This patch "inlines" fill into getFirstSpecialInstruction, the sole caller, to eliminate repeated hash lookups. 
FirstSpecialInsts caches the first special instruction for each basic
block.  This patch "inlines" fill into getFirstSpecialInstruction, the
sole caller, to eliminate repeated hash lookups.
 [Analysis] Avoid repeated hash lookups (NFC) (#123159) 2025-01-16T16:45:27+00:00 Kazu Hirata kazu@google.com 2025-01-16T16:45:27+00:00 5fa989b034236ebf5a808dd47af50ab29d991a7d 






[llvm] Use *{Map,Set}::contains (NFC)
2023-03-16T01:06:32+00:00

Kazu Hirata
kazu@google.com

2023-03-16T01:06:32+00:00

398af9b43bbfbff291faac5bce524eca675d6caf












[Analysis] Use *{Set,Map}::contains (NFC)
2023-03-14T07:32:40+00:00

Kazu Hirata
kazu@google.com

2023-03-14T07:32:40+00:00

11efd1cb04d0e9115d38f2138fa7fa455e281e09












[Analysis] Qualify auto variables in for loops (NFC)
2022-07-17T06:26:34+00:00

Kazu Hirata
kazu@google.com

2022-07-17T06:26:34+00:00

601b3a13dea746e2b1e683bbbece41ae9de76c2b












Revert "[IPT] Restructure cache to allow lazy update following invalidation [NFC]"
2021-10-21T17:48:41+00:00

Arthur Eubanks
aeubanks@google.com

2021-10-21T17:48:41+00:00

3781a46c3c70a83b2a5312aa358512578e87b9ca

This reverts commit baea663a6e9bc52f80995d02bb8149934c825612.

Causes crashes, e.g. https://lab.llvm.org/buildbot/#/builders/77/builds/10715.





This reverts commit baea663a6e9bc52f80995d02bb8149934c825612.

Causes crashes, e.g. https://lab.llvm.org/buildbot/#/builders/77/builds/10715.







[IPT] Restructure cache to allow lazy update following invalidation [NFC]
2021-10-21T16:16:21+00:00

Philip Reames
listmail@philipreames.com

2021-10-21T16:15:08+00:00

baea663a6e9bc52f80995d02bb8149934c825612

This change restructures the cache used in IPT to point not to the first special instruction, but to the first instruction which *could* be special. That is, the cached reference is always equal to the first special, or comes before it in the block.

This avoids expensive block scans when we are removing special instructions from the beginning of the block. At the moment, this case is not heavily used, though it does trigger in GVN when doing CSE of calls. The main motivation was a change I'm no longer planning to move forward with, but the cache optimization seemed worthwhile as a minor perf win at low cost.

Differential Revision: https://reviews.llvm.org/D111768





This change restructures the cache used in IPT to point not to the first special instruction, but to the first instruction which *could* be special. That is, the cached reference is always equal to the first special, or comes before it in the block.

This avoids expensive block scans when we are removing special instructions from the beginning of the block. At the moment, this case is not heavily used, though it does trigger in GVN when doing CSE of calls. The main motivation was a change I'm no longer planning to move forward with, but the cache optimization seemed worthwhile as a minor perf win at low cost.

Differential Revision: https://reviews.llvm.org/D111768







[IPT] Add a statistic to track instructions scanned to answer queries
2021-10-08T17:59:35+00:00

Philip Reames
listmail@philipreames.com

2021-10-08T17:49:54+00:00

edf31b4db1be3acdac1204a1c232454af20569e4

I'm planning some changes to the invalidation mechanism here, and having a concrete mechanism to track progress is key.





I'm planning some changes to the invalidation mechanism here, and having a concrete mechanism to track progress is key.







[IPT] Narrow scope of removeInstruction invalidation [NFC]
2021-10-08T17:35:03+00:00

Philip Reames
listmail@philipreames.com

2021-10-08T17:32:36+00:00

b4498e6b8d7f55e0fe9e26a5f5d21ec1a78f6fb0

We only need to invalidate if the instruction being removed is the cached "first special instruction".  If the instruction is before that one, it can't (by assumption) be special.  If it is after that one, it wasn't the first.





We only need to invalidate if the instruction being removed is the cached "first special instruction".  If the instruction is before that one, it can't (by assumption) be special.  If it is after that one, it wasn't the first.