[Vm-dev] VM Maker: VMMaker.oscog-eem.3311.mcz

[Marcel Taeumel]

Hi Eliot (eem), hi Tom (WoC) --

Note that implicit int conversions (-Wint-conversion) are errors (!) these days.

For example in SpurSegmentManager >> #segmentContainingObj:, the expression

seg := segments at: mid.


will not (always) compile because the types "int" and "SpurSegmentInfo*" must match or be casted explicitely.

Here is an exemplary compiler error (MSYS2, clang 15.0.5):

...
../../../src/spur32.cog/gcc3x-cointerp.c:61910:7: error: assigning to 'sqInt' (aka 'int') from incompatible type 'SpurSegmentInfo'
                seg = GIV(segments)[mid];
                    ^ ~~~~~~~~~~~~~~~~~~
...

Best,
Marcel
Am 16.03.2023 20:42:00 schrieb commits at source.squeak.org <commits at source.squeak.org>:

Eliot Miranda uploaded a new version of VMMaker to project VM Maker:
http://source.squeak.org/VMMaker/VMMaker.oscog-eem.3311.mcz

==================== Summary ====================

Name: VMMaker.oscog-eem.3311
Author: eem
Time: 16 March 2023, 12:40:54.390562 pm
UUID: 635091b8-f80a-43b2-817b-14723e9d73c2
Ancestors: VMMaker.oscog-eem.3310

Implement SpurSegmentManager>>segmentContainingObj: via binary search for the benefit of the IGC.
Make stack page printing less confusing (include the baseAddress).
Include a benchmark primitive to run the scavenge (we have garbageCollectMost, but the benchmark answers the time taken as measured by the high-res timer).
Cull some unused simulation methods.

=============== Diff against VMMaker.oscog-eem.3310 ===============

Item was added:
+ ----- Method: CoInterpreterPrimitives>>primitiveBenchmarkScavenge (in category 'benchmark primitives') -----
+ primitiveBenchmarkScavenge
+
+
+ self primitiveReturnTimeTakenFor:
+ [objectMemory hasSpurMemoryManagerAPI
+ ifTrue: [objectMemory scavengingGC]
+ ifFalse: [objectMemory incrementalGC].
+ self integerObjectOf: (objectMemory bytesLeft: false)]!

Item was changed:
----- Method: CogStackPageSurrogate>>printOn: (in category 'printing') -----
printOn: aStream
super printOn: aStream.
+ aStream space.
+ self baseAddress printOn: aStream base: 16.
+ aStream nextPut: $@.
+ aStream space; print: address; nextPut: $/.
- aStream nextPut: $@; print: address; nextPut: $/.
address printOn: aStream base: 16!

Item was removed:
- ----- Method: CogVMSimulator>>sqMemoryExtraBytesLeft: (in category 'memory access') -----
- sqMemoryExtraBytesLeft: includingSwap
- ^0!

Item was removed:
- ----- Method: InterpreterSimulator>>sqMemoryExtraBytesLeft: (in category 'memory access') -----
- sqMemoryExtraBytesLeft: includingSwap
- ^0!

Item was changed:
----- Method: SpurSegmentManager>>segmentContainingObj: (in category 'accessing') -----
segmentContainingObj: objOop
+ "Answer the segment containing an object. This is mostly for assert checking, but
+ variations on the incremental GC may use it in anger. Binary search is (of course)
+ marginally slower than linear search for a single segment (e.g. in a 720k object heap,
+ 67.1ms vs 61.3ms, or 9.5% slower to derive the segment containing every old space
+ entity), but usefully faster for many segments (e.g. 92.7ms vs 116ms, or 20% faster
+ in the same heap extended with enough large arrays to require 11 segments; and this
+ is pessimal; there are fewer objects at high addresses since the large arrays are there)."


+ | high low mid seg |
+ low := 0. mid := numSegments // 2. high := numSegments - 1.
+ [seg := segments at: mid.
+ (self oop: objOop isGreaterThanOrEqualTo: seg segStart)
+ ifTrue:
+ [mid = high
+ ifTrue:
+ [^(self oop: objOop isLessThan: seg segLimit) ifTrue:
+ [seg]]
+ ifFalse:
+ [low := mid.
+ mid := mid + high + 1 // 2]]
+ ifFalse:
+ [high := mid - 1.
+ mid := low + mid // 2].
+ low <= high] whileTrue.
- numSegments - 1 to: 0 by: -1 do:
- [:i|
- objOop >= (segments at: i) segStart ifTrue:
- [^self addressOf: (segments at: i)]].
^nil!

Item was removed:
- ----- Method: StackInterpreterSimulator>>sqMemoryExtraBytesLeft: (in category 'memory access') -----
- sqMemoryExtraBytesLeft: includingSwap
- ^0!

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20230321/f470e65a/attachment-0001.html>