- ghc-commits - Haskell.org

[Git][ghc/ghc][wip/mangoiv/ci-stages] 2 commits: testsuite: flush more often when running testsuite
by Magnus (＠MangoIV) 17 Aug '26

17 Aug '26

Magnus pushed to branch wip/mangoiv/ci-stages at Glasgow Haskell Compiler / GHC Commits: f3b5ca14 by mangoiv at 2026-08-17T12:29:32+02:00 testsuite: flush more often when running testsuite - - - - - b44db2b0 by mangoiv at 2026-08-17T14:01:48+02:00 Allow metric decreases on windows Metric Decrease: LinkableUsage01 MultiLayerModulesTH_OneShot T10547 T12234 T12425 T13035 T13253-spj T13820 T18140 T18730 T18923 T9198 TcPlugin_RewritePerf mhu-perf - - - - - 1 changed file: - testsuite/driver/runtests.py Changes: ===================================== testsuite/driver/runtests.py ===================================== @@ -353,6 +353,7 @@ except Exception as e: print('Failed to detect CPU features: ', e) sys.stdout.flush() +sys.stderr.flush() if config.local: tempdir = '' @@ -479,6 +480,9 @@ for name in config.only: # .T file errors. pass +sys.stdout.flush() +sys.stderr.flush() + if config.list_broken: print('') print('Broken tests:') @@ -506,12 +510,17 @@ else: # wait for parallel tests to finish asyncio.run(run_parallelTests()) + sys.stdout.flush() + sys.stderr.flush() + # Run the following tests purely sequential async def run_aloneTests(): for oneTest in aloneTests: if stopping(): break await oneTest(None) + sys.stdout.flush() + sys.stderr.flush() asyncio.run(run_aloneTests()) @@ -520,6 +529,7 @@ else: # flush everything before we continue sys.stdout.flush() + sys.stderr.flush() # Dump metrics data. print("\nPerformance Metrics (test environment: {}):\n".format(config.test_env)) View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/03e3bc4ecb964373785da11442c1c9… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/03e3bc4ecb964373785da11442c1c9… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/marge_bot_batch_merge_job] 5 commits: Avoid wasteful allocations in mkTyConAppCo
by Marge Bot (＠marge-bot) 17 Aug '26

17 Aug '26

Marge Bot pushed to branch wip/marge_bot_batch_merge_job at Glasgow Haskell Compiler / GHC Commits: 1aac7095 by sheaf at 2026-08-16T04:37:16-04:00 Avoid wasteful allocations in mkTyConAppCo The idiom "traverse isReflCo_maybe" followed by "map fst" used in 'GHC.Core.Coercion.mkTyConAppCo' was allocating a lot of waste. This commit uses 'GHC.Data.Unboxed.traverseMaybeUB' to avoid all these unnecessary intermediate allocations. In a quick microbenchmark for 'mkTyConAppCo', this change resulted in: - refl case (all argument coercions are reflexive): - -60% runtime - -80% allocations - non-refl case: - from 0% to -12% runtime (depending on which argument is non-refl) - from 0% to -70% allocations ( -- '' -- ) Fixes #27648 ------------------------- Metric Decrease: FamAppCachePerf SimplCastPerf T12425 T15703 T26426 T3064 T9872a T9872b T9872b_defer T9872c T9872d T5321Fun T9020 T9630 TcPlugin_RewritePerf Metric Increase: LinkableUsage02 ------------------------- - - - - - 3915e982 by Alan Zimmerman at 2026-08-16T04:37:58-04:00 EPA: Remove al_trailing from AnnList It was not being used - - - - - 999a903b by Andreas Klebinger at 2026-08-17T07:51:38-04:00 testsuite: Use sigkill in process009. SIGHUB might have been blocked by a (grand)*parent of the test. In such cases the test would fail as the python process would simply exist gracefully instead of committing to a premature end. By using SIGKILL we can rely on the signal not being blocked, fixing #27578 in the process. - - - - - f8b3c1ce by Andreas Klebinger at 2026-08-17T07:51:40-04:00 Simplify comparison in DFM.hs Fixes #27669 - - - - - 2667cef6 by Alan Zimmerman at 2026-08-17T07:51:40-04:00 TTG: Add extension points to BooleanFormula They are currently unused, but will be used for exact print annotations next, allowing us to get rid of LocatedBF / SrcSpanAnnBF - - - - - 41 changed files: - compiler/GHC/Core/Class.hs - compiler/GHC/Core/Coercion.hs - compiler/GHC/CoreToIface.hs - compiler/GHC/Data/BooleanFormula.hs - compiler/GHC/Data/Unboxed.hs - compiler/GHC/Data/Word64Map/Internal.hs - compiler/GHC/Data/Word64Map/Lazy.hs - compiler/GHC/Data/Word64Map/Strict.hs - compiler/GHC/Data/Word64Map/Strict/Internal.hs - compiler/GHC/Iface/Ext/Ast.hs - compiler/GHC/Iface/Syntax.hs - compiler/GHC/IfaceToCore.hs - compiler/GHC/Parser.y - compiler/GHC/Parser/Annotation.hs - compiler/GHC/Parser/PostProcess.hs - compiler/GHC/Tc/TyCl/Class.hs - compiler/GHC/Types/Unique/DFM.hs - compiler/Language/Haskell/Syntax/BooleanFormula.hs - compiler/Language/Haskell/Syntax/Extension.hs - testsuite/tests/count-deps/CountDepsParser.stdout - testsuite/tests/ghc-api/exactprint/T22919.stderr - testsuite/tests/ghc-api/exactprint/ZeroWidthSemi.stderr - testsuite/tests/module/mod185.stderr - testsuite/tests/parser/should_compile/DumpParsedAst.stderr - testsuite/tests/parser/should_compile/DumpParsedAstComments.stderr - testsuite/tests/parser/should_compile/DumpRenamedAst.stderr - testsuite/tests/parser/should_compile/DumpSemis.stderr - testsuite/tests/parser/should_compile/KindSigs.stderr - testsuite/tests/parser/should_compile/T15279.stderr - testsuite/tests/parser/should_compile/T20718.stderr - testsuite/tests/parser/should_compile/T20846.stderr - testsuite/tests/printer/Test20297.stdout - testsuite/tests/printer/Test24533.stdout - testsuite/tests/process/process009.hs - testsuite/tests/process/process009.stdout - utils/check-exact/ExactPrint.hs - utils/check-exact/Main.hs - utils/check-exact/Transform.hs - utils/haddock/haddock-api/src/Haddock/Backends/Xhtml/Decl.hs - utils/haddock/haddock-api/src/Haddock/Interface/Rename.hs - utils/haddock/haddock-api/src/Haddock/Types.hs The diff was not included because it is too large. View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/71e972714b2923fff551d4e35e6360… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/71e972714b2923fff551d4e35e6360… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/sjakobi/T27602] testsuite: Fold Baseline into CommitMetric
by Simon Jakobi (＠sjakobi) 17 Aug '26

17 Aug '26

Simon Jakobi pushed to branch wip/sjakobi/T27602 at Glasgow Haskell Compiler / GHC Commits: 311048aa by Simon Jakobi at 2026-08-17T13:25:44+02:00 testsuite: Fold Baseline into CommitMetric A Baseline was just a CommitMetric plus the commit it came from, built by copying fields across. Since get_commit_metric already knows that commit, record it on CommitMetric itself and drop Baseline. This also collapses both branches of find_baseline into plain returns. Assisted-by: Claude Fable 5 - - - - - 2 changed files: - testsuite/driver/perf_notes.py - testsuite/driver/testglobals.py Changes: ===================================== testsuite/driver/perf_notes.py ===================================== @@ -84,17 +84,12 @@ PerfStat = NamedTuple('PerfStat', [('test_env', TestEnv), ('value', float)]) # A test's metric recovered from a commit's git note: the raw sample values -# recorded there, and a PerfStat whose value is their mean. +# recorded there, and a PerfStat whose value is their mean. Serves as the +# baseline when comparing a test run against an earlier commit. class CommitMetric(NamedTuple): - perfStat: PerfStat - samples: List[float] - -# A baseline recovered form stored metrics. -class Baseline(NamedTuple): perfStat: PerfStat commit: GitHash - # The raw samples the baseline value was averaged over. - samples: List[float] = [] + samples: List[float] # The type of exceptions which are thrown when computing the current stat value # fails. @@ -486,7 +481,7 @@ _commit_metric_cache: Dict[Tuple[NoteNamespace, GitHash], # instead when looking for ci results) # metric: str - test metric # way: str - test way -# returns: the Baseline or None if no metric was found within +# returns: the baseline CommitMetric or None if no metric was found within # BaselineSearchDepth commits and since the last expected change # (ignoring any expected change in the given commit). def baseline_metric(commit: GitHash, @@ -495,7 +490,7 @@ def baseline_metric(commit: GitHash, metric: MetricName, way: WayName, baseline_ref: Optional[GitRef] - ) -> Optional[Baseline]: + ) -> Optional[CommitMetric]: # For performance reasons (in order to avoid calling commit_hash), we assert # commit is already a commit hash. assert is_commit_hash(commit) @@ -511,22 +506,16 @@ def baseline_metric(commit: GitHash, # Searches through previous commits trying local then ci for each commit in. def find_baseline(namespace: NoteNamespace, test_env: TestEnv - ) -> Optional[Baseline]: + ) -> Optional[CommitMetric]: if baseline_commit is not None: - current_metric = get_commit_metric(namespace, baseline_commit, test_env, name, metric, way) - if current_metric is not None: - return Baseline(current_metric.perfStat, baseline_commit, - current_metric.samples) - else: - return None + return get_commit_metric(namespace, baseline_commit, test_env, name, metric, way) for depth, current_commit in list(enumerate(commit_hashes)): if current_commit == commit: continue # Check for a metric on this commit. current_metric = get_commit_metric(namespace, current_commit, test_env, name, metric, way) if current_metric is not None: - return Baseline(current_metric.perfStat, current_commit, - current_metric.samples) + return current_metric # Stop if there is an expected change at this commit. In that case # metrics on ancestor commits will not be a valid baseline. @@ -538,7 +527,7 @@ def baseline_metric(commit: GitHash, # Test environment to use when comparing against CI namespace ci_test_env = best_fit_ci_test_env() - baseline = find_baseline(LocalNamespace, test_env) # type: Optional[Baseline] + baseline = find_baseline(LocalNamespace, test_env) # type: Optional[CommitMetric] if baseline is None and ci_test_env is not None: baseline = find_baseline(CiNamespace, ci_test_env) @@ -608,6 +597,7 @@ def get_commit_metric(gitNoteRef, currentCacheKey[3], currentCacheKey[2], sum(currentValues) / len(currentValues)), + commit, currentValues) # Save metrics to the cache. @@ -621,7 +611,7 @@ def format_samples(samples: List[float]) -> str: return ', '.join(format_sample(s) for s in samples) def check_stats_change(actual: PerfStat, - baseline: Baseline, + baseline: CommitMetric, acceptance_window: MetricAcceptanceWindow, allowed_perf_changes: Dict[TestName, List[AllowedPerfChange]] = {}, force_print = False @@ -631,8 +621,8 @@ def check_stats_change(actual: PerfStat, Parameters: actual: the PerfStat with actual value - baseline: the expected Baseline value (this should generally be derived - from baseline_metric()) + baseline: the CommitMetric to compare against (this should generally be + derived from baseline_metric()) acceptance_window: allowed deviation of the actual value from the expected value. allowed_perf_changes: allowed changes in stats. This is a dictionary as ===================================== testsuite/driver/testglobals.py ===================================== @@ -4,7 +4,7 @@ from my_typing import * from pathlib import Path -from perf_notes import MetricChange, PerfStat, Baseline, GitRef +from perf_notes import MetricChange, PerfStat, CommitMetric, GitRef from datetime import datetime # ----------------------------------------------------------------------------- @@ -312,7 +312,7 @@ class TestResult: PerfMetric = NamedTuple('PerfMetric', [('change', MetricChange), ('stat', PerfStat), - ('baseline', Optional[Baseline]) ]) + ('baseline', Optional[CommitMetric]) ]) class TestRun: def __init__(self) -> None: View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/311048aa5dc176fb36dc6eb1ed0a90f… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/311048aa5dc176fb36dc6eb1ed0a90f… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/sjakobi/T25233] 2 commits: X86 NCG: handle single-bit literal masks on the left
by Simon Jakobi (＠sjakobi) 17 Aug '26

17 Aug '26

Simon Jakobi pushed to branch wip/sjakobi/T25233 at Glasgow Haskell Compiler / GHC Commits: 0116eca3 by Simon Jakobi at 2026-08-17T12:27:43+02:00 X86 NCG: handle single-bit literal masks on the left Constant folding canonicalizes constants to the right of commutable operations, so these patterns cannot arise from Haskell code — but hand-written Cmm can reach the NCG unfolded. The variable-index bit-test equations already match both operand orders. Assisted-by: Claude Fable 5 - - - - - 235b8823 by Simon Jakobi at 2026-08-17T12:27:54+02:00 X86 NCG: fix overclaim in Note [Bit-test instructions] The Note claimed that an and/or/xor with an imm32 is "at least as good" as a bit-test instruction. That overlooks code size: for bit indices 7..30 the btr/bts/btc encoding is two bytes shorter. Spell out the actual trade-off (equal latency, better throughput, at worst two bytes of code size) instead. Assisted-by: Claude Fable 5 - - - - - 4 changed files: - compiler/GHC/CmmToAsm/X86/CodeGen.hs - + testsuite/tests/codeGen/should_gen_asm/T25233b.asm - + testsuite/tests/codeGen/should_gen_asm/T25233b.cmm - testsuite/tests/codeGen/should_gen_asm/all.T Changes: ===================================== compiler/GHC/CmmToAsm/X86/CodeGen.hs ===================================== @@ -1480,6 +1480,23 @@ getRegister' platform is32Bit (CmmMachOp (MO_Xor w) [x, CmmLit lit@(CmmInt m _)] , not (is32BitLit platform lit) = genBitTestImmCode (intFormat w) BTC x i +-- Mirrored versions with the literal mask on the left. Constant folding +-- canonicalizes constants to the right (see GHC.Cmm.Opt.cmmMachOpFoldM), so +-- these only fire on Cmm that reaches the NCG unfolded, e.g. hand-written +-- .cmm code. +getRegister' platform is32Bit (CmmMachOp (MO_And w) [CmmLit lit@(CmmInt m _), x]) + | Just i <- clearBitLit_maybe w m, bitTestOpWidthOK is32Bit w + , not (is32BitLit platform lit) + = genBitTestImmCode (intFormat w) BTR x i +getRegister' platform is32Bit (CmmMachOp (MO_Or w) [CmmLit lit@(CmmInt m _), x]) + | Just i <- setBitLit_maybe w m, bitTestOpWidthOK is32Bit w + , not (is32BitLit platform lit) + = genBitTestImmCode (intFormat w) BTS x i +getRegister' platform is32Bit (CmmMachOp (MO_Xor w) [CmmLit lit@(CmmInt m _), x]) + | Just i <- setBitLit_maybe w m, bitTestOpWidthOK is32Bit w + , not (is32BitLit platform lit) + = genBitTestImmCode (intFormat w) BTC x i + getRegister' platform is32Bit (CmmMachOp mop [x, y]) = do -- dyadic MachOps sse4_1 <- sse4_1Enabled sse4_2 <- sse4_2Enabled @@ -5938,11 +5955,14 @@ in any case undefined). The bit-offset operand of these instructions must be an immediate or a register. When the bit index is a literal, no shift reaches the NCG: constant folding has already turned the whole mask into a literal. If that -mask fits in an imm32, an ordinary and/or/xor with an immediate is at least -as good; but a W64 mask touching the upper bits, e.g. ~(1 << 40), would have -to be moved into a register first. For such masks we recognise the folded -literal itself (exactly one bit clear resp. set) and emit btr/bts/btc with -an immediate bit offset. +mask fits in an imm32, we keep the ordinary and/or/xor with an immediate: +it has the same latency and better throughput (more execution ports) than +the bit-test instructions, and at worst two bytes of extra code size for bit +indices 7..30. +But a W64 mask touching the upper bits, e.g. ~(1 << 40), would have to be moved +into a register first. For such masks we recognise the folded literal itself +(exactly one bit clear resp. set) and emit btr/bts/btc with an immediate +bit offset. We restrict the pattern to W32 and native-width W64: the instructions do not exist at width 8, and sub-word Cmm operations at W8/W16 are rare enough that ===================================== testsuite/tests/codeGen/should_gen_asm/T25233b.asm ===================================== @@ -0,0 +1,3 @@ +btrq $40, +btsq $40, +btcq $40, ===================================== testsuite/tests/codeGen/should_gen_asm/T25233b.cmm ===================================== @@ -0,0 +1,17 @@ +#include "Cmm.h" + +// Single-bit literal masks written on the left of the operator. Constant +// folding canonicalizes literals to the right, so these shapes only reach +// the NCG from hand-written Cmm like this (#25233). + +clearBit40 (W_ x) { + return ((0xFFFFFEFFFFFFFFFF :: bits64) & x); +} + +setBit40 (W_ x) { + return ((0x10000000000 :: bits64) | x); +} + +complementBit40 (W_ x) { + return ((0x10000000000 :: bits64) ^ x); +} ===================================== testsuite/tests/codeGen/should_gen_asm/all.T ===================================== @@ -25,6 +25,8 @@ test('avx512-word64-minmax', [unless(arch('x86_64'), skip), when(unregisterised(), skip)], compile_grep_asm, ['hs', True, '-mavx512vl']) test('T25233', [unless(arch('x86_64'), skip), when(unregisterised(), skip)], compile_grep_asm, ['hs', True, '-O']) +test('T25233b', [unless(arch('x86_64'), skip), + when(unregisterised(), skip)], compile_grep_asm, ['cmm', True, '']) is_aarch64_codegen = [ unless(arch('aarch64'), skip), when(unregisterised(), skip), View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/6963717182f0070c6c041f71a28121… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/6963717182f0070c6c041f71a28121… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/sjakobi/T27619-liveness-fixpoint-formats] 2 commits: NCG: iterate the liveness fixpoint until register formats converge
by Simon Jakobi (＠sjakobi) 17 Aug '26

17 Aug '26

Simon Jakobi pushed to branch wip/sjakobi/T27619-liveness-fixpoint-formats at Glasgow Haskell Compiler / GHC Commits: 103d8ba4 by Simon Jakobi at 2026-08-17T11:02:25+02:00 NCG: iterate the liveness fixpoint until register formats converge The convergence test of the per-SCC liveness fixpoint in GHC.CmmToAsm.Reg.Liveness compared BlockMap Regs values with the key-only Eq of UniqSet, which ignores the Format attached to each register. Formats grow across iterations (one control-flow edge per iteration), so the fixpoint could stop while formats were still growing, recording a vector register live at a too-narrow format and handing the register allocator an under-sized spill/reload width (#27619). The test also compared the entire accumulated block map on every iteration, although only the current SCC's entries can change, making it quadratic in procedure size (#27437). Compare only the SCC's own entries, using the new format-aware equalRegs. See the new Note [Convergence of the liveness fixpoint]. Also remove the Eq instance of Regs: inherited from UniqSet, it compared the register uniques only, silently ignoring the formats. The convergence test was its sole user. Fixes #27619. Fixes #27437. Assisted-by: Claude Fable 5 - - - - - 8d5058d9 by Simon Jakobi at 2026-08-17T11:02:25+02:00 NCG: compare register formats by width, remove Ord Format maxRegWithFormat, minusCoveredRegs and shrinkingRegs compared formats with Format's derived Ord, which orders by constructor, not by width: FF32 > II64, and VecFormat 16 FmtInt8 (16 bytes) > VecFormat 8 FmtDouble (64 bytes). This is harmless while all vector vregs are 128-bit, but becomes unsound in the (FmtBwd2) direction of Note [Register formats in liveness analysis] once wider vectors land in the NCG. Noted in #27619. The comparisons now go by formatToWidth, except the liveness join maxRegWithFormat, which needs a total order to keep the liveness fixpoint convergent and uses the new width-major compareFormat. See Note [Convergence of the liveness fixpoint] in GHC.CmmToAsm.Reg.Liveness. Also remove the Ord instances of Format and ScalarFormat, so no future call site can mistake the derived order for a width order. Their only other user was a width test in GHC.CmmToAsm.PPC.CodeGen, now expressed with formatToWidth. Assisted-by: Claude Fable 5 - - - - - 7 changed files: - compiler/GHC/CmmToAsm/Format.hs - compiler/GHC/CmmToAsm/PPC/CodeGen.hs - compiler/GHC/CmmToAsm/Reg/Liveness.hs - compiler/GHC/CmmToAsm/Reg/Regs.hs - compiler/GHC/Types/Unique/FM.hs - + testsuite/tests/regalloc/T27619.hs - testsuite/tests/regalloc/all.T Changes: ===================================== compiler/GHC/CmmToAsm/Format.hs ===================================== @@ -23,6 +23,7 @@ module GHC.CmmToAsm.Format ( vecFormat, isVecFormat, cmmTypeFormat, + compareFormat, formatToWidth, scalarWidth, formatInBytes, @@ -40,11 +41,13 @@ where import GHC.Prelude +import Data.Semigroup ( (<>) ) + import GHC.Cmm import GHC.Platform.Reg ( Reg(..), RealReg, VirtualReg ) import GHC.Types.Unique ( Uniquable(..) ) import GHC.Types.Unique.Set -import GHC.Utils.Outputable +import GHC.Utils.Outputable hiding ( (<>) ) import GHC.Utils.Panic {- Note [GHC's data format representations] @@ -92,7 +95,9 @@ data Format | FF64 | VecFormat !Length -- ^ number of elements (always at least 2) !ScalarFormat -- ^ format of each element - deriving (Show, Eq, Ord) + deriving (Show, Eq) + -- No Ord: compare via 'formatToWidth', or use 'compareFormat' where a + -- total order is needed. pattern IntegerFormat :: Format pattern IntegerFormat <- ( isIntegerFormat -> True ) @@ -117,7 +122,7 @@ data ScalarFormat | FmtInt64 | FmtFloat | FmtDouble - deriving (Show, Eq, Ord) + deriving (Show, Eq) scalarFormatFormat :: ScalarFormat -> Format scalarFormatFormat = \case @@ -248,6 +253,33 @@ scalarWidth = \case formatInBytes :: Format -> Int formatInBytes = widthInBytes . formatToWidth +-- | Total order on formats: by width, with an arbitrary but fixed tiebreak +-- between distinct formats of the same width. +-- +-- See Note [Convergence of the liveness fixpoint] in GHC.CmmToAsm.Reg.Liveness. +compareFormat :: Format -> Format -> Ordering +compareFormat f1 f2 = + compare (formatToWidth f1) (formatToWidth f2) <> compare (tag f1) (tag f2) + where + tag :: Format -> (Int, Length) + tag = \case + II8 -> (0, 0) + II16 -> (1, 0) + II32 -> (2, 0) + II64 -> (3, 0) + FF32 -> (4, 0) + FF64 -> (5, 0) + VecFormat l s -> (6 + scalarTag s, l) + + scalarTag :: ScalarFormat -> Int + scalarTag = \case + FmtInt8 -> 0 + FmtInt16 -> 1 + FmtInt32 -> 2 + FmtInt64 -> 3 + FmtFloat -> 4 + FmtDouble -> 5 + -------------------------------------------------------------------------------- -- | A typed virtual register: a virtual register, together with the specific ===================================== compiler/GHC/CmmToAsm/PPC/CodeGen.hs ===================================== @@ -484,7 +484,7 @@ getRegister' _ _ (CmmMachOp (MO_SS_Conv src tgt) [CmmLoad mem pk _]) , src < tgt = do let format = cmmTypeFormat pk -- lwa is DS-form. See Note [Power instruction format] - let form = if format >= II32 then DS else D + let form = if formatToWidth format >= W32 then DS else D Amode addr addr_code <- getAmode form mem let code dst = assert (format == intFormat src) $ addr_code `snocOL` LA format dst addr ===================================== compiler/GHC/CmmToAsm/Reg/Liveness.hs ===================================== @@ -879,7 +879,7 @@ computeLiveness platform sccs , ppr sccs']) livenessSCCs - :: Instruction instr + :: forall instr. Instruction instr => Platform -> BlockMap Regs -> [SCC (LiveBasicBlock instr)] -- accum @@ -897,37 +897,50 @@ livenessSCCs platform blockmap done (AcyclicSCC block : sccs) livenessSCCs platform blockmap done (CyclicSCC blocks : sccs) = livenessSCCs platform blockmap' (CyclicSCC blocks':done) sccs - where (blockmap', blocks') - = iterateUntilUnchanged linearLiveness equalBlockMaps - blockmap blocks + where (blockmap', blocks') = fixpoint blockmap - iterateUntilUnchanged - :: (a -> b -> (a,c)) -> (a -> a -> Bool) - -> a -> b - -> (a,c) - - iterateUntilUnchanged f eq aa b = go aa + -- See Note [Convergence of the liveness fixpoint] + fixpoint :: BlockMap Regs -> (BlockMap Regs, [LiveBasicBlock instr]) + fixpoint bm + | all unchanged blocks = (bm', blocks'') + | otherwise = fixpoint bm' where - go a = if eq a a' then ac else go a' - where - ac@(a', _) = f a b - - linearLiveness - :: Instruction instr - => BlockMap Regs -> [LiveBasicBlock instr] - -> (BlockMap Regs, [LiveBasicBlock instr]) - - linearLiveness = mapAccumL (livenessBlock platform) - - -- probably the least efficient way to compare two - -- BlockMaps for equality. - equalBlockMaps :: BlockMap Regs -> BlockMap Regs -> Bool - equalBlockMaps a b - = a' == b' - where a' = mapToList a - b' = mapToList b - -- See Note [Unique Determinism and code generation] - + (bm', blocks'') = mapAccumL (livenessBlock platform) bm blocks + + unchanged :: LiveBasicBlock instr -> Bool + unchanged block = + case (mapLookup bid bm, mapLookup bid bm') of + (Just old, Just new) -> old `equalRegs` new + (Nothing, _ ) -> False -- first iteration + (Just _, Nothing ) -> False -- can't happen + where bid = blockId block + +{- Note [Convergence of the liveness fixpoint] +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +For a cyclic SCC, we iterate 'livenessBlock' over the SCC's blocks until the +recorded entry liveness stops changing. Two subtleties in the convergence +test: + +* It must compare register formats, not just sets of live registers – + hence the format-aware 'equalRegs'. A block's live-in set joins the + successors' entries with 'unionRegsMaxFmt', so a wide format may take one + iteration per control-flow edge to propagate backwards around a loop. The + register sets themselves are typically complete after the first iteration. + If we stopped as soon as the sets stabilise, a register could be recorded + at a narrower format than the reads it flows into. That would violate + property (FmtBwd1) of Note [Register formats in liveness analysis], which + the register allocator relies on for spill and reload widths. See #27619. + +* It is sufficient to compare the entries of the SCC's own blocks. No other + entries can change: 'livenessBlock' inserts only the block it processes, + and successor SCCs are already final because SCCs are processed in reverse + dependency order. Comparing the whole accumulated block map would make the + fixpoint quadratic in procedure size (#27437). + +To ensure termination, 'maxRegWithFormat' uses 'compareFormat' which defines a +total order on formats. Comparing the widths only would introduce the risk of +an infinite loop where each iteration swaps two formats of the same width. +-} -- | Annotate a basic block with register liveness information. ===================================== compiler/GHC/CmmToAsm/Reg/Regs.hs ===================================== @@ -13,16 +13,19 @@ module GHC.CmmToAsm.Reg.Regs ( shrinkingRegs, mapRegs, elemRegs, lookupReg, + equalRegs, ) where import GHC.Prelude import GHC.Platform.Reg ( Reg ) -import GHC.CmmToAsm.Format ( Format, RegWithFormat(..), isVecFormat ) +import GHC.CmmToAsm.Format ( Format, RegWithFormat(..), isVecFormat, + compareFormat, formatToWidth ) import GHC.Utils.Outputable ( Outputable ) import GHC.Types.Unique ( Uniquable(..) ) +import GHC.Types.Unique.FM ( equalUFMBy ) import GHC.Types.Unique.Set import Data.Coerce ( coerce ) @@ -33,15 +36,16 @@ import Data.Coerce ( coerce ) -- register liveness analysis. See Note [Register formats in liveness analysis] -- in GHC.CmmToAsm.Reg.Liveness. newtype Regs = Regs { getRegs :: UniqSet RegWithFormat } - deriving newtype (Eq, Outputable) + deriving newtype (Outputable) maxRegWithFormat :: RegWithFormat -> RegWithFormat -> RegWithFormat maxRegWithFormat r1@(RegWithFormat _ fmt1) r2@(RegWithFormat _ fmt2) - = if fmt1 >= fmt2 - then r1 - else r2 - -- Re-using one of the arguments avoids allocating a new 'RegWithFormat', - -- compared with returning 'RegWithFormat r1 (max fmt1 fmt2)'. + | LT <- compareFormat fmt1 fmt2 = r2 + | otherwise = r1 + -- See Note [Convergence of the liveness fixpoint] in + -- GHC.CmmToAsm.Reg.Liveness. + -- + -- Re-using one of the arguments avoids allocating a new 'RegWithFormat'. noRegs :: Regs noRegs = Regs emptyUniqSet @@ -64,7 +68,7 @@ minusCoveredRegs = coerce $ minusUniqSet_C f where f :: RegWithFormat -> RegWithFormat -> Maybe RegWithFormat f r1@(RegWithFormat _ fmt1) (RegWithFormat _ fmt2) = - if fmt2 >= fmt1 + if formatToWidth fmt2 >= formatToWidth fmt1 || not ( isVecFormat fmt1 ) -- See Wrinkle [Don't allow scalar partial writes] @@ -97,7 +101,7 @@ shrinkingRegs = coerce $ minusUniqSet_C f where f :: RegWithFormat -> RegWithFormat -> Maybe RegWithFormat f (RegWithFormat _ fmt1) r2@(RegWithFormat _ fmt2) - | fmt2 < fmt1 + | formatToWidth fmt2 < formatToWidth fmt1 = Just r2 | otherwise = Nothing @@ -117,3 +121,10 @@ elemRegs r (Regs live) = elemUniqSet_Directly (getUnique r) live lookupReg :: Reg -> Regs -> Maybe Format lookupReg r (Regs live) = regWithFormat_format <$> lookupUniqSet_Directly live (getUnique r) + +-- | Do the two sets contain the same registers, at the same formats? +equalRegs :: Regs -> Regs -> Bool +equalRegs (Regs a) (Regs b) = equalUFMBy sameFormat (getUniqSet a) (getUniqSet b) + where + -- Registers with equal uniques are equal, so only compare the formats. + sameFormat (RegWithFormat _ fmt1) (RegWithFormat _ fmt2) = fmt1 == fmt2 ===================================== compiler/GHC/Types/Unique/FM.hs ===================================== @@ -67,6 +67,7 @@ module GHC.Types.Unique.FM ( strictIntersectUFM_C, disjointUFM, equalKeysUFM, + equalUFMBy, diffUFM, nonDetStrictFoldUFM, nonDetFoldUFM, nonDetStrictFoldUFM_DirectlyM, nonDetFoldWithKeyUFM, @@ -590,7 +591,12 @@ unsafeCastUFMKey (UFM m) = UFM m -- Determines whether two 'UniqFM's contain the same keys. equalKeysUFM :: UniqFM key a -> UniqFM key b -> Bool -equalKeysUFM (UFM m1) (UFM m2) = liftEq (\_ _ -> True) m1 m2 +equalKeysUFM = equalUFMBy (\_ _ -> True) + +-- | Determines whether two 'UniqFM's contain the same keys, with values +-- that agree according to the given predicate. +equalUFMBy :: (a -> b -> Bool) -> UniqFM key a -> UniqFM key b -> Bool +equalUFMBy eq (UFM m1) (UFM m2) = liftEq eq m1 m2 -- | An edit on type @a@, relating an element of a container (like an entry in a -- map or a line in a file) before and after. ===================================== testsuite/tests/regalloc/T27619.hs ===================================== @@ -0,0 +1,27 @@ +{-# LANGUAGE MagicHash, UnboxedTuples #-} + +-- The native code for this loop is a three-block cycle H -> X -> W -> H: +-- +-- H, X: read v at FF64 (lane-0 extracts) +-- W: reads v at F64x2 (the full unpack) +-- +-- v is loop-invariant, so its live format on entry to every block in the +-- cycle must be F64x2. -fno-cse keeps the two syntactically identical +-- lane-0 extracts from being merged. +module T27619 where + +import GHC.Exts + +loop :: Int# -> DoubleX2# -> Double# -> Double# +loop i v acc = + case unpackDoubleX2# v of + (# a1, _ #) -> + if isTrue# (a1 <## int2Double# i) + then acc + else case unpackDoubleX2# v of + (# a2, _ #) -> + if isTrue# (a2 *## 2.0## <## int2Double# i) + then acc *## 2.0## + else case unpackDoubleX2# v of + (# x, y #) -> loop (i -# 1#) v (acc +## (x *## y)) +{-# NOINLINE loop #-} ===================================== testsuite/tests/regalloc/all.T ===================================== @@ -6,3 +6,14 @@ test('regalloc_unit_tests', [ignore_stderr, only_ways(['normal'])], extra_run_opts('"' + config.libdir + '"') ], compile_and_run, ['-package ghc']) + +# The liveness fixpoint must iterate until the register formats converge, not +# just the sets of live registers (#27619). A vector register live at +# VecFormat 2 FmtDouble around the loop must not be recorded at FF64. +test('T27619', + [ unless(arch('x86_64'), skip), + when(not have_ncg(), skip), + only_ways(['normal']), + grep_errmsg(r'%vV128_\S+ :: FF64') ], + compile, + ['-O -fno-cse -ddump-asm-liveness']) View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/10e6330d9c54a295946b9aebb05a6d… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/10e6330d9c54a295946b9aebb05a6d… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/sjakobi/T27602] 3 commits: testsuite: Show baseline sample count and range in perf failures
by Simon Jakobi (＠sjakobi) 17 Aug '26

17 Aug '26

Simon Jakobi pushed to branch wip/sjakobi/T27602 at Glasgow Haskell Compiler / GHC Commits: 736fcdfb by Simon Jakobi at 2026-08-17T10:25:36+02:00 testsuite: Show baseline sample count and range in perf failures A perf baseline is the mean of all samples recorded for a commit, and it prints as a single number, hiding how far the samples spread. When the spread is wide, this can indicate an unstable metric that isn't actually useful as a signal for the perf tests. For example, in #27602, T27336's peak_megabytes_allocated baseline showed as 757 when the underlying samples were 605 and 909. When the baseline is averaged from more than one sample, say so in the failure output: the one-line stat-failure reason shows the sample range, and the detail block lists the raw samples. Single-sample baselines print exactly as before. Context: #27602 Assisted-by: Claude Fable 5 - - - - - 29b1d269 by Simon Jakobi at 2026-08-17T10:30:11+02:00 ci: Clarify comment on pushing perf notes after failures Context: #27602 Assisted-by: Claude Fable 5 - - - - - 9fc0bcaf by Simon Jakobi at 2026-08-17T10:30:11+02:00 DEMO: Tighten T3064 residency tolerance (do not merge) Assisted-by: Claude Fable 5 - - - - - 3 changed files: - .gitlab/ci.sh - testsuite/driver/perf_notes.py - testsuite/tests/perf/compiler/all.T Changes: ===================================== .gitlab/ci.sh ===================================== @@ -1120,9 +1120,10 @@ case ${1:-help} in setup) setup && cleanup_submodules ;; configure) time_it "configure" configure ;; build_hadrian) time_it "build" build_hadrian ;; - # N.B. Always push notes, even if the build fails. This is okay to do as the - # testsuite driver doesn't record notes for tests that fail due to - # correctness. + # N.B. Always push notes, even if the build fails. Metrics from runs failing + # a perf stat check are deliberately recorded too — discarding them would + # bias the baseline towards whichever sample came first. Only correctness + # failures record nothing. test_hadrian) fetch_perf_notes res=0 ===================================== testsuite/driver/perf_notes.py ===================================== @@ -83,9 +83,18 @@ PerfStat = NamedTuple('PerfStat', [('test_env', TestEnv), ('metric', MetricName), ('value', float)]) +# A test's metric recovered from a commit's git note: the raw sample values +# recorded there, and a PerfStat whose value is their mean. +class CommitMetric(NamedTuple): + perfStat: PerfStat + samples: List[float] + # A baseline recovered form stored metrics. -Baseline = NamedTuple('Baseline', [('perfStat', PerfStat), - ('commit', GitHash)]) +class Baseline(NamedTuple): + perfStat: PerfStat + commit: GitHash + # The raw samples the baseline value was averaged over. + samples: List[float] = [] # The type of exceptions which are thrown when computing the current stat value # fails. @@ -460,10 +469,10 @@ def get_allowed_changes(baseline_ref: Optional[GitRef]) -> Dict[TestName, List[A else: return get_allowed_perf_changes() -# Cache of baseline values. This is a dict of dicts indexed on: -# (useCiNamespace, commit) -> (test_env, test, metric, way) -> baseline -# (bool , str ) -> (str , str , str , str) -> float -_commit_metric_cache = {} # type: ignore +# Cache of commit metrics. +_commit_metric_cache: Dict[Tuple[NoteNamespace, GitHash], + Dict[Tuple[TestEnv, TestName, MetricName, WayName], + CommitMetric]] = {} # Get the baseline of a test at a given commit. This is the expected value # *before* the commit is applied (i.e. on the parent commit). @@ -506,7 +515,8 @@ def baseline_metric(commit: GitHash, if baseline_commit is not None: current_metric = get_commit_metric(namespace, baseline_commit, test_env, name, metric, way) if current_metric is not None: - return Baseline(current_metric, baseline_commit) + return Baseline(current_metric.perfStat, baseline_commit, + current_metric.samples) else: return None @@ -515,7 +525,8 @@ def baseline_metric(commit: GitHash, # Check for a metric on this commit. current_metric = get_commit_metric(namespace, current_commit, test_env, name, metric, way) if current_metric is not None: - return Baseline(current_metric, current_commit) + return Baseline(current_metric.perfStat, current_commit, + current_metric.samples) # Stop if there is an expected change at this commit. In that case # metrics on ancestor commits will not be a valid baseline. @@ -545,23 +556,23 @@ def get_commit_metric_value_str_or_none(gitNoteRef, result = get_commit_metric(gitNoteRef, commit, test_env, name, metric, way) if result is None: return None - return str(result.value) + return str(result.perfStat.value) -# gets the average commit metric from git notes. +# gets the commit metric (average and raw samples) from git notes. # gitNoteRef: git notes ref space e.g. "perf" or "ci/perf" # ref: git commit # test_env: test environment # name: test name # metric: test metric # way: test way -# returns: PerfStat | None if stats don't exist for the given input +# returns: CommitMetric | None if stats don't exist for the given input def get_commit_metric(gitNoteRef, ref: Union[GitRef, GitHash], test_env: TestEnv, name: TestName, metric: MetricName, way: WayName - ) -> Optional[PerfStat]: + ) -> Optional[CommitMetric]: global _commit_metric_cache assert test_env != None commit = commit_hash(ref) @@ -573,9 +584,9 @@ def get_commit_metric(gitNoteRef, return _commit_metric_cache[cacheKeyA].get(cacheKeyB) # Cache miss. - # Calculate baselines from the current commit's git note. + # Calculate metrics from the current commit's git note. # Note that the git note may contain data for other tests. All tests' - # baselines will be collected and cached for future use. + # metrics will be collected and cached for future use. allCommitMetrics = get_perf_stats(ref, gitNoteRef) # Collect recorded values by cacheKeyB. @@ -586,19 +597,28 @@ def get_commit_metric(gitNoteRef, currentValues = values_by_cache_key_b.setdefault(currentCacheKey, []) currentValues.append(float(perfStat.value)) - # Calculate and baseline (average of values) by cacheKeyB. - baseline_by_cache_key_b = {} + # Calculate the metric (average of values, plus the values themselves) + # by cacheKeyB. + metric_by_cache_key_b = {} for currentCacheKey, currentValues in values_by_cache_key_b.items(): - baseline_by_cache_key_b[currentCacheKey] = PerfStat( \ - currentCacheKey[0], - currentCacheKey[1], - currentCacheKey[3], - currentCacheKey[2], - sum(currentValues) / len(currentValues)) + metric_by_cache_key_b[currentCacheKey] = CommitMetric( + PerfStat( + currentCacheKey[0], + currentCacheKey[1], + currentCacheKey[3], + currentCacheKey[2], + sum(currentValues) / len(currentValues)), + currentValues) + + # Save metrics to the cache. + _commit_metric_cache[cacheKeyA] = metric_by_cache_key_b + return metric_by_cache_key_b.get(cacheKeyB) - # Save baselines to the cache. - _commit_metric_cache[cacheKeyA] = baseline_by_cache_key_b - return baseline_by_cache_key_b.get(cacheKeyB) +def format_sample(s: float) -> str: + return str(int(s)) if s == int(s) else str(s) + +def format_samples(samples: List[float]) -> str: + return ', '.join(format_sample(s) for s in samples) def check_stats_change(actual: PerfStat, baseline: Baseline, @@ -654,9 +674,17 @@ def check_stats_change(actual: PerfStat, ' baseline @ %s' % baseline.commit print(actual.metric, error + ':') dev = 100.0 if expected_val == 0 else round(((float(actual.value) * 100) / int(expected_val)) - 100, 1) + # Show the sample spread so unreliable baselines become visible (#27602). + if len(baseline.samples) > 1: + samples_note = ('; baseline is mean of %d samples spanning %s..%s' + % (len(baseline.samples), + format_sample(min(baseline.samples)), + format_sample(max(baseline.samples)))) + else: + samples_note = '' change_line = (f'{actual.metric} {change.value} from {baseline.perfStat.test_env} ' f'baseline @ {baseline.commit[:7]}: {expected_val} -> {actual.value} ' - f'({dev:+g}%, allowed {acceptance_window.describe()})') + f'({dev:+g}%, allowed {acceptance_window.describe()}{samples_note})') result = failBecause('stat ' + change_line, tag='stat') if not change_allowed or force_print: @@ -666,6 +694,10 @@ def check_stats_change(actual: PerfStat, print(descr, str(val).rjust(length), extra) display(' Expected ' + full_name + ' ' + actual.metric + ':', expected_val, acceptance_window.describe()) + if len(baseline.samples) > 1: + display(' Samples ' + full_name + ' ' + actual.metric + ':', + len(baseline.samples), + '(' + format_samples(baseline.samples) + ')') display(' Lower bound ' + full_name + ' ' + actual.metric + ':', lowerBound, '') display(' Upper bound ' + full_name + ' ' + actual.metric + ':', upperBound, '') display(' Actual ' + full_name + ' ' + actual.metric + ':', actual.value, '') @@ -866,7 +898,7 @@ def main() -> None: # HEAD~2 21234 21234 # HEAD~3 20000 20000 def strMetric(x): - return '{:.2f}'.format(x.value) if x != None else "" + return '{:.2f}'.format(x.perfStat.value) if x != None else "" # Data is in column major format, so transpose and pass to print_table. T = TypeVar('T') def transpose(xss: List[List[T]]) -> List[List[T]]: ===================================== testsuite/tests/perf/compiler/all.T ===================================== @@ -48,7 +48,9 @@ test('T4801', ['']) test('T3064', - [collect_compiler_residency(20), + [# Demo-only: tightened from 20 so the bimodal max_bytes_used + # baseline on aarch64-linux-deb13-validate fails. + collect_compiler_residency(1), collect_compiler_runtime(2), only_ways(['normal']), ], View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/324435b2d2df03e43ffbb8059a2850… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/324435b2d2df03e43ffbb8059a2850… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/mangoiv/ci-stages] configure: fix sed script and subst tooldir script
by Magnus (＠MangoIV) 17 Aug '26

17 Aug '26

Magnus pushed to branch wip/mangoiv/ci-stages at Glasgow Haskell Compiler / GHC Commits: 03e3bc4e by mangoiv at 2026-08-17T09:38:33+02:00 configure: fix sed script and subst tooldir script Fixes #27672 - - - - - 2 changed files: - distrib/configure.ac.in - m4/subst_tooldir.m4 Changes: ===================================== distrib/configure.ac.in ===================================== @@ -103,7 +103,7 @@ AC_ARG_ENABLE(distro-toolchain, AC_SUBST([EnableDistroToolchain]) if test "$HostOS" = "mingw32" -a "$EnableDistroToolchain" = "NO"; then - FP_SETUP_WINDOWS_TOOLCHAIN([$hardtop/mingw/], [\$\$topdir/../mingw/]) + FP_SETUP_WINDOWS_TOOLCHAIN([$hardtop/mingw/], [\$topdir/../mingw/]) fi ===================================== m4/subst_tooldir.m4 ===================================== @@ -41,5 +41,5 @@ dnl ghc-toolchain. # See Note [How we configure the bundled windows toolchain] AC_DEFUN([SUBST_TOOLDIR], [ - sed -i.bkp $1 's%'"$mingw_prefix"'%'"$mingw_install_prefix"'%g' + sed -i.bkp 's%'"$mingw_prefix"'%'"$mingw_install_prefix"'%g' $1 ]) View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/03e3bc4ecb964373785da11442c1c9c… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/03e3bc4ecb964373785da11442c1c9c… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/mangoiv/ci-stages] configure: fix sed script and subst tooldir script
by Magnus (＠MangoIV) 17 Aug '26

17 Aug '26

Magnus pushed to branch wip/mangoiv/ci-stages at Glasgow Haskell Compiler / GHC Commits: 1ffe3ba7 by mangoiv at 2026-08-17T09:38:13+02:00 configure: fix sed script and subst tooldir script - - - - - 2 changed files: - distrib/configure.ac.in - m4/subst_tooldir.m4 Changes: ===================================== distrib/configure.ac.in ===================================== @@ -103,7 +103,7 @@ AC_ARG_ENABLE(distro-toolchain, AC_SUBST([EnableDistroToolchain]) if test "$HostOS" = "mingw32" -a "$EnableDistroToolchain" = "NO"; then - FP_SETUP_WINDOWS_TOOLCHAIN([$hardtop/mingw/], [\$\$topdir/../mingw/]) + FP_SETUP_WINDOWS_TOOLCHAIN([$hardtop/mingw/], [\$topdir/../mingw/]) fi ===================================== m4/subst_tooldir.m4 ===================================== @@ -41,5 +41,5 @@ dnl ghc-toolchain. # See Note [How we configure the bundled windows toolchain] AC_DEFUN([SUBST_TOOLDIR], [ - sed -i.bkp $1 's%'"$mingw_prefix"'%'"$mingw_install_prefix"'%g' + sed -i.bkp 's%'"$mingw_prefix"'%'"$mingw_install_prefix"'%g' $1 ]) View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/1ffe3ba77e71f10a47ba5847be1ec54… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/1ffe3ba77e71f10a47ba5847be1ec54… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc][wip/sjakobi/T27602] 3 commits: testsuite: Show baseline sample count and spread in perf failures
by Simon Jakobi (＠sjakobi) 17 Aug '26

17 Aug '26

Simon Jakobi pushed to branch wip/sjakobi/T27602 at Glasgow Haskell Compiler / GHC Commits: 9dd81712 by Simon Jakobi at 2026-08-16T21:19:49+02:00 testsuite: Show baseline sample count and spread in perf failures A perf baseline is the mean of all samples recorded for a commit, so a single outlier can silently corrupt it. Previously, the failure output gave no hint about such outliers: the baseline printed as one number. In #27602, T27336's peak_megabytes_allocated baseline showed as 757 when the underlying samples were 605 and 909. When the baseline is averaged from more than one sample, say so in the failure message and list the samples, both in the one-line stat-failure reason and in the detail block. Single-sample baselines print exactly as before. Context: #27602 Assisted-by: Claude Fable 5 - - - - - 0ea57bad by Simon Jakobi at 2026-08-16T21:19:49+02:00 ci: Clarify comment on pushing perf notes after failures Context: #27602 Assisted-by: Claude Fable 5 - - - - - 324435b2 by Simon Jakobi at 2026-08-16T21:19:49+02:00 DEMO: Tighten T3064 residency tolerance (do not merge) Assisted-by: Claude Fable 5 - - - - - 3 changed files: - .gitlab/ci.sh - testsuite/driver/perf_notes.py - testsuite/tests/perf/compiler/all.T Changes: ===================================== .gitlab/ci.sh ===================================== @@ -1120,9 +1120,10 @@ case ${1:-help} in setup) setup && cleanup_submodules ;; configure) time_it "configure" configure ;; build_hadrian) time_it "build" build_hadrian ;; - # N.B. Always push notes, even if the build fails. This is okay to do as the - # testsuite driver doesn't record notes for tests that fail due to - # correctness. + # N.B. Always push notes, even if the build fails. Metrics from runs failing + # a perf stat check are deliberately recorded too — discarding them would + # bias the baseline towards whichever sample came first. Only correctness + # failures record nothing. test_hadrian) fetch_perf_notes res=0 ===================================== testsuite/driver/perf_notes.py ===================================== @@ -83,9 +83,18 @@ PerfStat = NamedTuple('PerfStat', [('test_env', TestEnv), ('metric', MetricName), ('value', float)]) +# A test's metric recovered from a commit's git note: the raw sample values +# recorded there, and a PerfStat whose value is their mean. +class CommitMetric(NamedTuple): + perfStat: PerfStat + samples: List[float] + # A baseline recovered form stored metrics. -Baseline = NamedTuple('Baseline', [('perfStat', PerfStat), - ('commit', GitHash)]) +class Baseline(NamedTuple): + perfStat: PerfStat + commit: GitHash + # The raw samples the baseline value was averaged over. + samples: List[float] = [] # The type of exceptions which are thrown when computing the current stat value # fails. @@ -460,10 +469,10 @@ def get_allowed_changes(baseline_ref: Optional[GitRef]) -> Dict[TestName, List[A else: return get_allowed_perf_changes() -# Cache of baseline values. This is a dict of dicts indexed on: -# (useCiNamespace, commit) -> (test_env, test, metric, way) -> baseline -# (bool , str ) -> (str , str , str , str) -> float -_commit_metric_cache = {} # type: ignore +# Cache of commit metrics. +_commit_metric_cache: Dict[Tuple[NoteNamespace, GitHash], + Dict[Tuple[TestEnv, TestName, MetricName, WayName], + CommitMetric]] = {} # Get the baseline of a test at a given commit. This is the expected value # *before* the commit is applied (i.e. on the parent commit). @@ -506,7 +515,8 @@ def baseline_metric(commit: GitHash, if baseline_commit is not None: current_metric = get_commit_metric(namespace, baseline_commit, test_env, name, metric, way) if current_metric is not None: - return Baseline(current_metric, baseline_commit) + return Baseline(current_metric.perfStat, baseline_commit, + current_metric.samples) else: return None @@ -515,7 +525,8 @@ def baseline_metric(commit: GitHash, # Check for a metric on this commit. current_metric = get_commit_metric(namespace, current_commit, test_env, name, metric, way) if current_metric is not None: - return Baseline(current_metric, current_commit) + return Baseline(current_metric.perfStat, current_commit, + current_metric.samples) # Stop if there is an expected change at this commit. In that case # metrics on ancestor commits will not be a valid baseline. @@ -545,23 +556,23 @@ def get_commit_metric_value_str_or_none(gitNoteRef, result = get_commit_metric(gitNoteRef, commit, test_env, name, metric, way) if result is None: return None - return str(result.value) + return str(result.perfStat.value) -# gets the average commit metric from git notes. +# gets the commit metric (average and raw samples) from git notes. # gitNoteRef: git notes ref space e.g. "perf" or "ci/perf" # ref: git commit # test_env: test environment # name: test name # metric: test metric # way: test way -# returns: PerfStat | None if stats don't exist for the given input +# returns: CommitMetric | None if stats don't exist for the given input def get_commit_metric(gitNoteRef, ref: Union[GitRef, GitHash], test_env: TestEnv, name: TestName, metric: MetricName, way: WayName - ) -> Optional[PerfStat]: + ) -> Optional[CommitMetric]: global _commit_metric_cache assert test_env != None commit = commit_hash(ref) @@ -573,9 +584,9 @@ def get_commit_metric(gitNoteRef, return _commit_metric_cache[cacheKeyA].get(cacheKeyB) # Cache miss. - # Calculate baselines from the current commit's git note. + # Calculate metrics from the current commit's git note. # Note that the git note may contain data for other tests. All tests' - # baselines will be collected and cached for future use. + # metrics will be collected and cached for future use. allCommitMetrics = get_perf_stats(ref, gitNoteRef) # Collect recorded values by cacheKeyB. @@ -586,19 +597,25 @@ def get_commit_metric(gitNoteRef, currentValues = values_by_cache_key_b.setdefault(currentCacheKey, []) currentValues.append(float(perfStat.value)) - # Calculate and baseline (average of values) by cacheKeyB. - baseline_by_cache_key_b = {} + # Calculate the metric (average of values, plus the values themselves) + # by cacheKeyB. + metric_by_cache_key_b = {} for currentCacheKey, currentValues in values_by_cache_key_b.items(): - baseline_by_cache_key_b[currentCacheKey] = PerfStat( \ - currentCacheKey[0], - currentCacheKey[1], - currentCacheKey[3], - currentCacheKey[2], - sum(currentValues) / len(currentValues)) - - # Save baselines to the cache. - _commit_metric_cache[cacheKeyA] = baseline_by_cache_key_b - return baseline_by_cache_key_b.get(cacheKeyB) + metric_by_cache_key_b[currentCacheKey] = CommitMetric( + PerfStat( + currentCacheKey[0], + currentCacheKey[1], + currentCacheKey[3], + currentCacheKey[2], + sum(currentValues) / len(currentValues)), + currentValues) + + # Save metrics to the cache. + _commit_metric_cache[cacheKeyA] = metric_by_cache_key_b + return metric_by_cache_key_b.get(cacheKeyB) + +def format_samples(samples: List[float]) -> str: + return ', '.join(str(int(s)) if s == int(s) else str(s) for s in samples) def check_stats_change(actual: PerfStat, baseline: Baseline, @@ -654,9 +671,16 @@ def check_stats_change(actual: PerfStat, ' baseline @ %s' % baseline.commit print(actual.metric, error + ':') dev = 100.0 if expected_val == 0 else round(((float(actual.value) * 100) / int(expected_val)) - 100, 1) + # Show the samples so outliers become visible (#27602). + if len(baseline.samples) > 1: + samples_note = ('; baseline is mean of %d samples: %s' + % (len(baseline.samples), + format_samples(baseline.samples))) + else: + samples_note = '' change_line = (f'{actual.metric} {change.value} from {baseline.perfStat.test_env} ' f'baseline @ {baseline.commit[:7]}: {expected_val} -> {actual.value} ' - f'({dev:+g}%, allowed {acceptance_window.describe()})') + f'({dev:+g}%, allowed {acceptance_window.describe()}{samples_note})') result = failBecause('stat ' + change_line, tag='stat') if not change_allowed or force_print: @@ -666,6 +690,10 @@ def check_stats_change(actual: PerfStat, print(descr, str(val).rjust(length), extra) display(' Expected ' + full_name + ' ' + actual.metric + ':', expected_val, acceptance_window.describe()) + if len(baseline.samples) > 1: + display(' Samples ' + full_name + ' ' + actual.metric + ':', + len(baseline.samples), + '(' + format_samples(baseline.samples) + ')') display(' Lower bound ' + full_name + ' ' + actual.metric + ':', lowerBound, '') display(' Upper bound ' + full_name + ' ' + actual.metric + ':', upperBound, '') display(' Actual ' + full_name + ' ' + actual.metric + ':', actual.value, '') @@ -866,7 +894,7 @@ def main() -> None: # HEAD~2 21234 21234 # HEAD~3 20000 20000 def strMetric(x): - return '{:.2f}'.format(x.value) if x != None else "" + return '{:.2f}'.format(x.perfStat.value) if x != None else "" # Data is in column major format, so transpose and pass to print_table. T = TypeVar('T') def transpose(xss: List[List[T]]) -> List[List[T]]: ===================================== testsuite/tests/perf/compiler/all.T ===================================== @@ -48,7 +48,9 @@ test('T4801', ['']) test('T3064', - [collect_compiler_residency(20), + [# Demo-only: tightened from 20 so the bimodal max_bytes_used + # baseline on aarch64-linux-deb13-validate fails. + collect_compiler_residency(1), collect_compiler_runtime(2), only_ways(['normal']), ], View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/89a9867d67346af26f4f4e8f4caae5… -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/89a9867d67346af26f4f4e8f4caae5… You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0

[Git][ghc/ghc] Pushed new branch wip/az/epa-tidy-locatedxxx-17
by Alan Zimmerman (＠alanz) 17 Aug '26

17 Aug '26

Alan Zimmerman pushed new branch wip/az/epa-tidy-locatedxxx-17 at Glasgow Haskell Compiler / GHC -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/tree/wip/az/epa-tidy-locatedxxx-17 You're receiving this email because of your account on gitlab.haskell.org. Manage all notifications: https://gitlab.haskell.org/-/profile/notifications | Help: https://gitlab.haskell.org/help

1 0