[Git][ghc/ghc][wip/T23109] 50 commits: gitlab-ci: Don't use .full-ci to run test-primops

Tue Feb 11 16:33:25 UTC 2025

Simon Peyton Jones pushed to branch wip/T23109 at Glasgow Haskell Compiler / GHC

Commits:
9a59b026 by Ben Gamari at 2025-02-04T10:00:18-05:00
gitlab-ci: Don't use .full-ci to run test-primops

test-primops depends upon the existence of validate jobs, yet these do
not exist in the context of nightly jobs, which .full-ci includes.

- - - - -
7cc08550 by Ben Gamari at 2025-02-04T18:34:49-05:00
CorePrep: Name `sat` binders more descriptively

- - - - -
fb40981d by Ben Gamari at 2025-02-04T18:35:26-05:00
ghc-toolchain: Parse i686 triples

This is a moniker used for later 32-bit x86 implementations
(Pentium Pro and later).

Fixes #25691.

- - - - -
02794411 by Cheng Shao at 2025-02-04T18:36:03-05:00
compiler: remove unused assembleOneBCO function

This patch removes the unused assembleOneBCO function from the
bytecode assembler.

- - - - -
db19c8a9 by Matthew Pickering at 2025-02-05T23:16:50-05:00
perf: Replace uses of genericLength with strictGenericLength

genericLength is a recursive function and marked NOINLINE. It is not
going to specialise. In profiles, it can be seen that 3% of total compilation
time when computing bytecode is spend calling this non-specialised
function.

In addition, we can simplify `addListToSS` to avoid traversing the input
list twice and also allocating an intermediate list (after the call to
reverse).

Overall these changes reduce the time spend in 'assembleBCOs' from 5.61s
to 3.88s. Allocations drop from 8GB to 5.3G.

Fixes #25706

- - - - -
5622a14a by Matthew Pickering at 2025-02-05T23:17:27-05:00
perf: nameToCLabel: Directly manipulate ByteString rather than going via strings

`nameToCLabel` is called from `lookupHsSymbol` many times during
bytecode linking. We can save a lot of allocations and time by directly
manipulating the bytestrings rather than going via intermediate lists.

Before: 2GB allocation, 1.11s
After: 260MB allocation, 375ms

Fixes #25719

-------------------------
Metric Decrease:
    MultiLayerModulesTH_OneShot
-------------------------

- - - - -
66c7f656 by Matthew Pickering at 2025-02-06T17:15:25-05:00
interpreter: Fix INTERP_STATS profiling code

The profiling code had slightly bitrotted since the last time it was
used. This just fixes things so that if you toggle the INTERP_STATS
macro then it just works and prints out the stats.

Fixes #25695

- - - - -
f71c2835 by Matthew Pickering at 2025-02-06T17:16:02-05:00
perf: Key the interpreter symbol cache by Name rather than FastString

Profiles showed that about 0.2s was being spend constructing the keys
before looking up values in the old symbol cache.

The performance of this codepath is critical as it translates directly
to a delay when a user evaluates a function like `main` in the
interpreter.

Therefore we implement a solution which keys the cache(s) by `Name`
rather than the symbol directly, so the cache can be consulted before
the symbol is constructed.

Fixes #25731

- - - - -
8f8d3a90 by Ben Gamari at 2025-02-08T01:17:28-05:00
base: Label threads forked by IO operations

Addresses part of #25452.