Vm-dev May 2021

vm-dev@lists.squeakfoundation.org

21 participants
56 discussions

[OpenSmalltalk/opensmalltalk-vm] f7ea6a: Changed sqSocketListenOnPortBacklogSizeInterface()...
by Eliot Miranda 05 May '21

05 May '21

Branch: refs/heads/Cog Home: https://github.com/OpenSmalltalk/opensmalltalk-vm Commit: f7ea6a5bb050173b0d947598e7e97e5231743556 https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/f7ea6a5bb050173b0d… Author: John Dougan <jdougan(a)acm.org> Date: 2021-04-30 (Fri, 30 Apr 2021) Changed paths: M platforms/win32/plugins/SocketPlugin/sqWin32NewNet.c Log Message: ----------- Changed sqSocketListenOnPortBacklogSizeInterface() to match general behavior of Unix socket version. Previously was alway incorrectly binding UDP sockets to INADDR_ANY instead of the specified IPv4 address. Commit: 128d14994b4a8002bb006db1bf9a5b37544139c1 https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/128d14994b4a8002bb… Author: Eliot Miranda <eliot.miranda(a)gmail.com> Date: 2021-05-04 (Tue, 04 May 2021) Changed paths: M platforms/win32/plugins/SocketPlugin/sqWin32NewNet.c Log Message: ----------- Merge pull request #563 from jdougan/Cog Fix UDP binding in Windows sockets Compare: https://github.com/OpenSmalltalk/opensmalltalk-vm/compare/10aac272367b...12…

1 0

[OpenSmalltalk/opensmalltalk-vm] Incorrect become forwarder for objectSize: mirror primitive (#566)
by Nicolas Cellier 05 May '21

05 May '21

This has been reported in pharo fork as https://github.com/pharo-project/opensmalltalk-vm/issues/179 and also affects upstream opensmalltalk-vm ``` | result obj | result := OrderedCollection new. obj := { 1. 2. 3 }. result add: (thisContext objectSize: obj). "Should be 3" obj become: { 'a'. 'b'. }. result add: (thisContext objectSize: obj). "Should be 2, but is 0" Smalltalk garbageCollect. result add: (thisContext objectSize: obj). "Should be 2" result "an OrderedCollection(3 0 2)" ``` -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/OpenSmalltalk/opensmalltalk-vm/issues/566

2 1

[OpenSmalltalk/opensmalltalk-vm] 10aac2: CogVM source as per VMMaker.oscog-eem.2957
by Eliot Miranda 05 May '21

05 May '21

Branch: refs/heads/Cog Home: https://github.com/OpenSmalltalk/opensmalltalk-vm Commit: 10aac272367b89068af3730641240179269ce6a7 https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/10aac272367b89068a… Author: Eliot Miranda <eliot.miranda(a)gmail.com> Date: 2021-05-04 (Tue, 04 May 2021) Changed paths: M spur64src/vm/cointerp.c M spur64src/vm/cointerp.h M spur64src/vm/cointerpmt.c M spur64src/vm/cointerpmt.h M spur64src/vm/gcc3x-cointerp.c M spur64src/vm/gcc3x-cointerpmt.c M spurlowcode64src/vm/cointerp.c M spurlowcode64src/vm/cointerp.h M spurlowcode64src/vm/gcc3x-cointerp.c M spurlowcodesrc/vm/cointerp.c M spurlowcodesrc/vm/cointerp.h M spurlowcodesrc/vm/gcc3x-cointerp.c M spurlowcodestack64src/vm/gcc3x-interp.c M spurlowcodestack64src/vm/interp.c M spurlowcodestacksrc/vm/gcc3x-interp.c M spurlowcodestacksrc/vm/interp.c M spursista64src/vm/cointerp.c M spursista64src/vm/cointerp.h M spursista64src/vm/gcc3x-cointerp.c M spursistasrc/vm/cointerp.c M spursistasrc/vm/cointerp.h M spursistasrc/vm/gcc3x-cointerp.c M spursrc/vm/cointerp.c M spursrc/vm/cointerp.h M spursrc/vm/cointerpmt.c M spursrc/vm/cointerpmt.h M spursrc/vm/gcc3x-cointerp.c M spursrc/vm/gcc3x-cointerpmt.c M spurstack64src/vm/gcc3x-interp.c M spurstack64src/vm/interp.c M spurstack64src/vm/validImage.c M spurstacksrc/vm/gcc3x-interp.c M spurstacksrc/vm/interp.c M spurstacksrc/vm/validImage.c M src/plugins/CameraPlugin/CameraPlugin.c M src/vm/cointerp.c M src/vm/cointerp.h M src/vm/gcc3x-cointerp.c M stacksrc/vm/gcc3x-interp.c M stacksrc/vm/interp.c Log Message: ----------- CogVM source as per VMMaker.oscog-eem.2957 Fix https://github.com/OpenSmalltalk/opensmalltalk-vm/issues/566. The primitiveSize primtiive (#62) failed to check for its argument being a forwarder when the primtiive is used as a mirror primitive (num args > 0). As a minor simplification use methodReturnInteger: in it and primitiveContextSize, since both are varargs.

1 0

VM Maker: VMMaker.oscog-eem.2957.mcz
by commits＠source.squeak.org 04 May '21

04 May '21

Eliot Miranda uploaded a new version of VMMaker to project VM Maker: http://source.squeak.org/VMMaker/VMMaker.oscog-eem.2957.mcz ==================== Summary ==================== Name: VMMaker.oscog-eem.2957 Author: eem Time: 4 May 2021, 2:57:14.47864 pm UUID: 1905cffe-ff84-45bc-b423-79e7978ecfd7 Ancestors: VMMaker.oscog-eem.2956 Fix https://github.com/OpenSmalltalk/opensmalltalk-vm/issues/566. The primitiveSize primtiive (#62) failed to check for its argument being a forwarder when the primtiive is used as a mirror primitive (num args > 0). As a minro simplification use methodReturnInteger: in it and primitiveContextSize, sicne both are varargs. =============== Diff against VMMaker.oscog-eem.2956 =============== Item was changed: ----- Method: InterpreterPrimitives>>primitiveSize (in category 'indexing primitives') ----- primitiveSize <accessorDepth: 0> | rcvr hdr fmt fixedFields totalLength | rcvr := self stackTop. ((objectMemory isImmediate: rcvr) "Integers are not indexable" or: [hdr := objectMemory baseHeader: rcvr. + (fmt := objectMemory formatOfHeader: hdr) < objectMemory arrayFormat "This is not an indexable object" + or: [objectMemory isForwarderFormat: fmt]]) "And if Spur, this is a forwarder." - (fmt := objectMemory formatOfHeader: hdr) < 2]) "This is not an indexable object" ifTrue: [^self primitiveFailFor: PrimErrBadReceiver]. (fmt = objectMemory indexablePointersFormat and: [objectMemory isContextHeader: hdr]) ifTrue: [^self primitiveContextSize]. totalLength := objectMemory lengthOf: rcvr baseHeader: hdr format: fmt. fixedFields := objectMemory fixedFieldsOf: rcvr format: fmt length: totalLength. + self methodReturnInteger: totalLength - fixedFields! - self pop: argumentCount + 1 thenPush: (objectMemory integerObjectOf: totalLength - fixedFields)! Item was added: + ----- Method: ObjectMemory>>isForwarderFormat: (in category 'header access') ----- + isForwarderFormat: format + ^false! Item was added: + ----- Method: SpurMemoryManager>>isForwarderFormat: (in category 'header formats') ----- + isForwarderFormat: format + ^format = self forwardedFormat! Item was changed: ----- Method: StackInterpreterPrimitives>>primitiveContextSize (in category 'indexing primitives') ----- primitiveContextSize "Special version of primitiveSize for accessing contexts. Written to be varargs for use from mirror primitives." | rcvr sz hdr fmt totalLength fixedFields | <inline: false> rcvr := self stackTop. hdr := objectMemory baseHeader: rcvr. fmt := objectMemory formatOfHeader: hdr. totalLength := objectMemory lengthOf: rcvr baseHeader: hdr format: fmt. fixedFields := objectMemory fixedFieldsOf: rcvr format: fmt length: totalLength. (objectMemory isContextHeader: hdr) ifTrue: [self externalWriteBackHeadFramePointers. sz := self stackPointerForMaybeMarriedContext: rcvr] ifFalse: [sz := totalLength - fixedFields]. + self methodReturnInteger: sz! - self pop: argumentCount + 1 thenPush: (objectMemory integerObjectOf: sz)!

1 0

[OpenSmalltalk/opensmalltalk-vm] Update BitBlt support (primarily for 64-bit ARM) (#565)
by bavison 04 May '21

04 May '21

The accelerated BitBlt framework was initially targeted at the ARM11, running the AArch32 instruction set (which is the only one it fully supported). More recent ARMs run much faster, which has enabled more comprehensive testing via the BitBlt fuzz test framework (https://github.com/bavison/SqueakBitBltTest). This has detected a handful of bugs in both the AArch32-specific and the architecture-neutral parts of the fast BitBlt framework. First I address these. Next, I add a number of BitBlt fast paths written in platform-independent C. The 8-to-32bpp conversion routine is as fast as anything I could manage with hand-crafted AArch64 assembly. Others are useful as reference implementations for other architectures, or to fill in gaps in their abilities (for example, while I've introduced a class of fast paths for colour maps that only feature two distinct colours, I haven't retrospectively written any AArch32 fast paths for them, so the C fast path will be used for them on AArch32). The fast path that handles operations with scalar halftoning and 32bpp destination images is a bit of a special case, in that it acts to extend the capabilities of other fast paths. It thus accelerates both AArch32 and AArch64. The most significant commit, however, is the last one. This features a collection of fast paths implemented using inline AArch64 assembly, tuned for Cortex-A72 (as found in the Raspberry Pi 4). Based on the results of profiling, this has an emphasis on operations with a 32bpp destination image. Operations with any source depth, in conjunction with 22 of the possible combinationRules (including the common sourceWord, pixPaint and alphaBlend rules) should all be accelerated, providing you don't use little-endian pixel packing, vector halftoning, or non-standard colour map rules when converting from different colour depths. There are additional fast paths for alphaBlend for either a constant source colour, or a source image whose colour map only consists of two different colours (i.e. where the source image is effectively used as a 1bpp mask, despite being of a greater depth). You can view, comment on, or merge this pull request online at: https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/565 -- Commit Summary -- * Don't pass -m32 to GCC for ARM builds * Correct various "#if ENABLE_FAST_BLT" to "#ifdef" * Don't assume sourcePPW is valid on entry to copyBitsFallback * Fallback routines need extra help to detect intra-image operations * Remove invalid shortcut in rgbComponentAlphawith * Fix bug in 32-bit ARM fast paths * Fix buffer overflow bugs * Fix corruption bugs with wide 1bpp source images * Fix type of halftone array for 64-bit targets * Detect and add a new fast path flag for effective-1bpp colour maps * C fast path for 32bpp alphaBlend * C fast path for planar alphaBlend * C fast path for 8->32bpp conversion * C fast path for alphaBlend with 1bpp colour map and scalar halftone * Apply scalar halftoning to colour map entries instead for 32bpp destination * Enable fast blit code for AArch64 * AArch64 assembly optimisations -- File Changes -- M build.linux32ARMv6/squeak.cog.spur/build.assert/mvm (2) M build.linux32ARMv6/squeak.cog.spur/build.debug/mvm (2) M build.linux32ARMv6/squeak.cog.spur/build/mvm (2) M build.linux64ARMv8/squeak.cog.spur/build.assert/mvm (1) M build.linux64ARMv8/squeak.cog.spur/build.debug/mvm (1) M build.linux64ARMv8/squeak.cog.spur/build/mvm (1) A platforms/Cross/plugins/BitBltPlugin/BitBltArm64.c (2482) A platforms/Cross/plugins/BitBltPlugin/BitBltArm64.h (30) M platforms/Cross/plugins/BitBltPlugin/BitBltArmSimdAsm.hdr (2) M platforms/Cross/plugins/BitBltPlugin/BitBltArmSimdSourceWord.s (17) M platforms/Cross/plugins/BitBltPlugin/BitBltDispatch.c (126) M platforms/Cross/plugins/BitBltPlugin/BitBltDispatch.h (2) M platforms/Cross/plugins/BitBltPlugin/BitBltGeneric.c (247) M platforms/Cross/plugins/BitBltPlugin/BitBltInternal.h (32) M platforms/unix/config/configure (12) M platforms/unix/plugins/BitBltPlugin/acinclude.m4 (11) M src/plugins/BitBltPlugin/BitBltPlugin.c (23) -- Patch Links -- https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/565.patch https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/565.diff -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/565

2 1

ARMv8 BitBLT speedups provided by RPi/Ben Avison
by ken.dickey＠whidbey.com 02 May '21

02 May '21

Experience Report: FYI, I git cloned https://github.com/bavison/opensmalltalk-vm and built build.linux64ARMv8/squeak.cog.spur/build Build went fine (did not have to elide broken CameraPlugin or adjust config.h) and all seems to work fine for me (on aarch64 Chromebook). FYI, -KenD

2 1

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

Vm-dev May 2021