| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch adds eight new primops that fuse a multiplication and an
addition or subtraction:
- `{fmadd,fmsub,fnmadd,fnmsub}{Float,Double}#`
fmadd x y z is x * y + z, computed with a single rounding step.
This patch implements code generation for these primops in the following
backends:
- X86, AArch64 and PowerPC NCG,
- LLVM
- C
WASM uses the C implementation. The primops are unsupported in the
JavaScript backend.
The following constant folding rules are also provided:
- compute a * b + c when a, b, c are all literals,
- x * y + 0 ==> x * y,
- ±1 * y + z ==> z ± y and x * ±1 + z ==> z ± x.
NB: the constant folding rules incorrectly handle signed zero.
This is a known limitation with GHC's floating-point constant folding
rules (#21227), which we hope to resolve in the future.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This MR changes some simple optimizations and is a first step in re-architecting
the JS backend pipeline to add the optimizer. In particular it:
- removes simple peep hole optimizations from `GHC.StgToJS.Printer` and removes that module
- adds module `GHC.JS.Optimizer`
- defines the same peep hole opts that were removed only now they are `Syntax -> Syntax` transformations rather than `Syntax -> JS code` optimizations
- hooks the optimizer into code gen
- adds FuncStat and ForStat constructors to the backend.
Working Ticket:
- #22736
Related MRs:
- MR !10142
- MR !10000
-------------------------
Metric Decrease:
CoOpt_Read
ManyAlternatives
PmSeriesS
PmSeriesT
PmSeriesV
T10421
T12707
T13253
T13253-spj
T15164
T17516
T18140
T18282
T18698a
T18698b
T18923
T1969
T19695
T20049
T3064
T5321FD
T5321Fun
T783
T9198
T9233
T9630
-------------------------
|
| |
|
|
| |
As requested by @treeowl in CLC#139.
|
| |
|
|
| |
fixes #23278
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
While fixing these I've also changed the way we store addresses into
ByteArray#. Addr# are composed of two parts: a JavaScript array and an
offset (32-bit number).
Suppose we want to store an Addr# in a ByteArray# foo at offset i.
Before this patch, we were storing both fields as a tuple in the "arr"
array field:
foo.arr[i] = [addr_arr, addr_offset];
Now we only store the array part in the "arr" field and the offset
directly in the array:
foo.dv.setInt32(i, addr_offset):
foo.arr[i] = addr_arr;
It avoids wasting space for the tuple.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* For ByteArray-based bounds-checking, the JavaScript backend must use the
`len` field, instead of the inbuild JavaScript `length` field.
* Range-based operations must also check both the start and end of the range
for bounds
* All indicies are valid for ranges of size zero, since they are essentially no-ops
* For cases of ByteArray accesses (e.g. read as Int), the end index is
(i * sizeof(type) + sizeof(type) - 1), while the previous implementation
uses (i + sizeof(type) - 1). In the Int32 example, this is (i * 4 + 3)
* IndexByteArrayOp_Word8As* primitives use byte array indicies (unlike
the previous point), but now check both start and end indicies
* Byte array copies now check if the arrays are the same by identity and
then if the ranges overlap.
|
| | |
|
| |
|
|
|
|
|
|
|
| |
#22364)
Carry the actual type of an expression through the PreStgRhs and into GenStgRhs
for use in later stages. Currently this is used in the JavaScript backend to fix
some tests from the above mentioned issues: EtaExpandLevPoly, RepPolyWrappedVar2,
T13822, T14749.
|
| | |
|
| |
|
|
|
|
| |
- Use dedicated list functions
- Make cloneBndrs and cloneRecIdBndrs monadic
- Fix invalid haddock comments in libraries/base
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* copyMutableByteArrayNonOverlapping#
* copyAddrToAddr#
* copyAddrToAddrNonOverlapping#
* setAddrRange#
The implementations of copyBytes, moveBytes, and fillBytes
in base:Foreign.Marshal.Utils now use these new primops,
which can cause us to work a bit harder generating code for them,
resulting in the metric increase in T21839c observed by CI on
some architectures. But in exchange, we get better code!
Metric Increase:
T21839c
|
| |
|
|
| |
Follow on to MR!10142 in pursuit of #22736
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In pursuit of #22426. The driver and unit state are major contributors.
This commit also bumps the haddock submodule to reflect the API changes in
UniqMap.
-------------------------
Metric Decrease:
MultiComponentModules
MultiComponentModulesRecomp
T10421
T10547
T12150
T12234
T12425
T13035
T16875
T18140
T18304
T18698a
T18698b
T18923
T20049
T5837
T6048
T9198
-------------------------
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit:
Splits JExpr and JStat into two nearly identical DSLs:
- GHC.JS.Syntax is the JMacro based DSL without unsaturation, i.e., a
value cannot be unsaturated, or, a value of this DSL is a witness that a
value of GHC.JS.Unsat has been saturated
- GHC.JS.Unsat is the JMacro DSL from GHCJS with Unsaturation.
Then all binary and outputable instances are changed to use
GHC.JS.Syntax.
This moves us closer to closing out #22736 and #22352. See #22736 for
roadmap.
-------------------------
Metric Increase:
CoOpt_Read
LargeRecord
ManyAlternatives
PmSeriesS
PmSeriesT
PmSeriesV
T10421
T10858
T11195
T11374
T11822
T12227
T12707
T13035
T13253
T13253-spj
T13379
T14683
T15164
T15703
T16577
T17096
T17516
T17836
T18140
T18282
T18304
T18478
T18698a
T18698b
T18923
T1969
T19695
T20049
T21839c
T3064
T4801
T5321FD
T5321Fun
T5631
T5642
T783
T9198
T9233
T9630
TcPlugin_RewritePerf
WWRec
-------------------------
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch moves the field-based logic for disambiguating record updates
to the renamer. The type-directed logic, scheduled for removal, remains
in the typechecker.
To do this properly (and fix the myriad of bugs surrounding the treatment
of duplicate record fields), we took the following main steps:
1. Create GREInfo, a renamer-level equivalent to TyThing which stores
information pertinent to the renamer.
This allows us to uniformly treat imported and local Names in the
renamer, as described in Note [GREInfo].
2. Remove GreName. Instead of a GlobalRdrElt storing GreNames, which
distinguished between normal names and field names, we now store
simple Names in GlobalRdrElt, along with the new GREInfo information
which allows us to recover the FieldLabel for record fields.
3. Add namespacing for record fields, within the OccNames themselves.
This allows us to remove the mangling of duplicate field selectors.
This change ensures we don't print mangled names to the user in
error messages, and allows us to handle duplicate record fields
in Template Haskell.
4. Move record disambiguation to the renamer, and operate on the
level of data constructors instead, to handle #21443.
The error message text for ambiguous record updates has also been
changed to reflect that type-directed disambiguation is on the way
out.
(3) means that OccEnv is now a bit more complex: we first key on the
textual name, which gives an inner map keyed on NameSpace:
OccEnv a ~ FastStringEnv (UniqFM NameSpace a)
Note that this change, along with (2), both increase the memory residency
of GlobalRdrEnv = OccEnv [GlobalRdrElt], which causes a few tests to
regress somewhat in compile-time allocation.
Even though (3) simplified a lot of code (in particular the treatment of
field selectors within Template Haskell and in error messages), it came
with one important wrinkle: in the situation of
-- M.hs-boot
module M where { data A; foo :: A -> Int }
-- M.hs
module M where { data A = MkA { foo :: Int } }
we have that M.hs-boot exports a variable foo, which is supposed to match
with the record field foo that M exports. To solve this issue, we add a
new impedance-matching binding to M
foo{var} = foo{fld}
This mimics the logic that existed already for impedance-binding DFunIds,
but getting it right was a bit tricky.
See Note [Record field impedance matching] in GHC.Tc.Module.
We also needed to be careful to avoid introducing space leaks in GHCi.
So we dehydrate the GlobalRdrEnv before storing it anywhere, e.g. in
ModIface. This means stubbing out all the GREInfo fields, with the
function forceGlobalRdrEnv.
When we read it back in, we rehydrate with rehydrateGlobalRdrEnv.
This robustly avoids any space leaks caused by retaining old type
environments.
Fixes #13352 #14848 #17381 #17551 #19664 #21443 #21444 #21720 #21898 #21946 #21959 #22125 #22160 #23010 #23062 #23063
Updates haddock submodule
-------------------------
Metric Increase:
MultiComponentModules
MultiLayerModules
MultiLayerModulesDefsGhci
MultiLayerModulesNoCode
T13701
T14697
hard_hole_fits
-------------------------
|
| | |
|
| |
|
|
|
|
| |
The code wasn't taking into account some kind of overlap.
cgrun070 has been extended to test the missing case.
|
| |
|
|
|
|
| |
Don't use BigInt for wordAdd2, mulWord32, and timesInt32.
Co-authored-by: Matthew Craven <5086-clyring@users.noreply.gitlab.haskell.org>
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch fixes #22745 and #15205, which are about GHC's
failure to discard unnecessary superclass selections that
yield coercions. See
GHC.Core.Utils Note [exprOkForSpeculation and type classes]
The main changes are:
* Write new Note [NON-BOTTOM_DICTS invariant] in GHC.Core, and
refer to it
* Define new function isTerminatingType, to identify those
guaranteed-terminating dictionary types.
* exprOkForSpeculation has a new (very simple) case for ClassOpId
* ClassOpId has a new field that says if the return type is
an unlifted type, or a terminating type.
This was surprisingly tricky to get right. In particular note
that unlifted types are not terminating types; you can write an
expression of unlifted type, that diverges. Not so for dictionaries
(or, more precisely, for the dictionaries that GHC constructs).
Metric Decrease:
LargeRecord
|
| | |
|
| |
|
|
|
|
|
|
|
| |
- Remove unused uniques and hs-boot declarations
- Fix types of seq and unsafeCoerce#
- Remove FastString/String roundtrip in JS
- Use TTG to enforce totality
- Remove enumeration in Heap/Inspect; the 'otherwise' clause
serves the primitive types well.
|
| |
|
|
|
|
|
| |
The `-outputdir` option wasn't correctly handled with the JS backend
because the same code path was used to handle both objects produced by
the JS backend and foreign .js files. Now we clearly distinguish the
two in the pipeline, fixing the bug.
|
|
|
Add JS backend adapted from the GHCJS project by Luite Stegeman.
Some features haven't been ported or implemented yet. Tests for these
features have been disabled with an associated gitlab ticket.
Bump array submodule
Work funded by IOG.
Co-authored-by: Jeffrey Young <jeffrey.young@iohk.io>
Co-authored-by: Luite Stegeman <stegeman@gmail.com>
Co-authored-by: Josh Meredith <joshmeredith2008@gmail.com>
|
