Vm-dev January 2018

vm-dev@lists.squeakfoundation.org

29 participants
255 discussions

ASM questions for tight loops
by Clément Bera 06 Jan '18

06 Jan '18

Hi Eliot, (sorry long mail but I think you will enjoy this discussion; read it) Looking deeper into tight loops performance to get some boost out of Sista, I have questions and remarks. First thing is when I compare 3 versions or array copies (code at the bottom of mail). Version [1] : copyArray : written in plain Smalltalk Version [2] : fastCopyArray : sista generated code (only unchecked operations) Version [3] : primCopyArray : simply call the machine code primitive For small arrays everything is fine, sista code is good. But for large arrays I get the following results (here a copy of 1000 elements): 1) 85k/second 2) 390k/second 3) 2,000k/second So the sista version is ~5x compared to pure smalltalk code, but ~5x slower to the primitive. My guess is that the reason for this is that the hand written primitive is written with a 7 instructions tight loop (2 additions, 1 read, 1 write, 2 jumps, 1 comp), while the sista generated code requires 6 temp read, 1 temp write and a couple cheap instructions in addition. We discussed earlier than register allocation might bring a 20% performance boost, but it seems tight loops are quite important and in this case the performance difference is around 5x (500%). So big question is that should I had more macro-instructions (let's call them stubs) in the sista extended bytecode set such as bytesEquals which bulkCompares the bytes of 2 objects, since performance difference is bigger than expected, or should I wait for better register allocation algorithm ? I believe hand-written stub will always be faster than sista generated code (since a couple things can't be expressed at the bytecode level), but based on multiple benchmarks I think if the better register allocator generates code with the same number of memory read-write (including temp access on stack) the difference should be less than 20% performance difference as we expected. It's not clear how many stubs would be needed. I have a prototype of array copy, other things such as indexOf or fillObject are needed. Maybe I can add only 3-4 for now to show massive speed-ups like we did with bytesEquals. Let's note that we could try to share the stubs with normal primitives (primStringReplace with the arraycopy stub for example) to avoid too much maintenance cost. The problem is that normal primitives have complexity with all the checks at the beginning, then a generic case (copy without knowing anything about the operands) while sista stubs have no checks, but complexity is in generating better code when some operands are constants. So it's not really the same thing. Second question is about RTL code for tight loops. I looked into V8 and it seems that for tight loop they try to use backward conditional instead of forward conditional + backward unconditional. What do you think will be the difference if I do it for array copy (the copy is guaranteed not to be empty at this point, example below Original-New) ? I wonder if conditional back jumps are not slower than conditional forward jumps or something like that (I am always confused with branch performance, and the performance is different depending on caches so small benchs may not be relevant). What's your take on this ? And then the question is that should I be able to generate that kind of loop from the sista back-end (I don't really like it, I prefer to limit the number of back-jumps else it makes some things harder) ? Else I can do it in the stubs for most critical performance things and ignore it in other cases. *Original* instr := cogit CmpR: startReg R: stopReg. jumpFinished := cogit JumpBelow: 0. cogit MoveXwr: repStartReg R: replReg R: TempReg. cogit MoveR: TempReg Xwr: startReg R: arrayReg. cogit AddCq: 1 R: startReg. cogit AddCq: 1 R: repStartReg. cogit Jump: instr. jumpFinished jmpTarget: (jumpEmpty jmpTarget: cogit genPrimReturn). Total 7 instructions in the loop *New* instr := cogit MoveXwr: repStartReg R: replReg R: TempReg. cogit MoveR: TempReg Xwr: startReg R: arrayReg. cogit AddCq: 1 R: startReg. cogit AddCq: 1 R: repStartReg. cogit CmpR: startReg R: stopReg. cogit JumpGreaterOrEqual: instr. jumpEmpty jmpTarget: cogit genPrimReturn. Total 6 instructions in the loop But a conditional back-jump. NB : Eliot I tried incrementing the pointer instead of the index and use a 0 displacement read/writes and it made no performance difference on Intel. Saves a register in the loop but we don't need them. It makes a difference in the inlined stub however since other registers can be used for other things in the method where it's inlined. *[1]* Smalltalk code 1 to: (y - x + 1) do: [ :i | array2 at: y2 + i put: (array at: x + i) ]. *[2] *(sista bytecodes of the tight loop) 32 <45> pushTemp: 5 33 <46> pushTemp: 6 34 <F8 F3 87> smiLessOrEqual: 37 <EF 1B> jumpFalse: 66 39 <43> pushTemp: 3 40 <45> pushTemp: 5 41 <40> pushTemp: 0 42 <45> pushTemp: 5 43 <F8 10 88> pointerAt: 46 <F8 B8 8B> pointerAt:put: 49 <D8> pop 50 <45> pushTemp: 5 51 <E1 00 E8 01> pushConstant: 1 55 <F8 D0 87> smiAdd: 58 <D5> popIntoTemp: 5 59 <E1 FF E8 DE> pushConstant: -34 63 <F8 70 97> backjumpNoInterrupt *[3]* primitive call array2 replaceFrom: x to: y with: array startingAt: y2 Current tight loop RTL: instr := cogit CmpR: startReg R: stopReg. jumpFinished := cogit JumpBelow: 0. cogit MoveXwr: repStartReg R: replReg R: TempReg. cogit MoveR: TempReg Xwr: startReg R: arrayReg. cogit AddCq: 1 R: startReg. cogit AddCq: 1 R: repStartReg. cogit Jump: instr. jumpFinished jmpTarget: (jumpEmpty jmpTarget: cogit genPrimReturn). -- Clément Béra https://clementbera.wordpress.com/ Bâtiment B 40, avenue Halley 59650 Villeneuve d'Ascq

3 4

VM Maker: VMMaker.oscog-eem.2310.mcz
by commits＠source.squeak.org 06 Jan '18

06 Jan '18

Eliot Miranda uploaded a new version of VMMaker to project VM Maker: http://source.squeak.org/VMMaker/VMMaker.oscog-eem.2310.mcz ==================== Summary ==================== Name: VMMaker.oscog-eem.2310 Author: eem Time: 5 January 2018, 6:52:19.253773 pm UUID: ecb88457-93a0-47c1-9d8a-69a576c283b3 Ancestors: VMMaker.oscog-eem.2309 Simulation: Fix charsOf: so that longPrintOop:/shortPrint: prints the full 8-bit character range, not the 7 bit character range. Provide accessors for breakCount and nuke the unused (and composable runWithBreakCount:). Make sure trhat click step in the StackInterpreter updates the bytecCount display. =============== Diff against VMMaker.oscog-eem.2309 =============== Item was added: + ----- Method: CogVMSimulator>>breakCount (in category 'accessing') ----- + breakCount + ^breakCount! Item was removed: - ----- Method: CogVMSimulator>>charsOfLong: (in category 'debug support') ----- - charsOfLong: long - ^self subclassResponsibility! Item was changed: ----- Method: CogVMSimulator>>incrementByteCount (in category 'interpreter shell') ----- incrementByteCount (byteCount := byteCount + 1) = breakCount ifTrue: [self doOrDefer: [self changed: #byteCountText; changed: #composeAll]. + self halt: 'breakCount reached']. - self halt]. byteCount \\ 1000 = 0 ifTrue: [self doOrDefer: [self changed: #byteCountText; changed: #composeAll]. self forceInterruptCheck]! Item was removed: - ----- Method: CogVMSimulator>>runWithBreakCount: (in category 'testing') ----- - runWithBreakCount: theBreakCount - "Just run, halting when byteCount is reached" - quitBlock := [displayView ifNotNil: - [displayView containingWindow ifNotNil: - [:topWindow| - ((World submorphs includes: topWindow) - and: [UIManager default confirm: 'close?']) ifTrue: - [topWindow delete]]]. - ^self close]. - breakCount := theBreakCount. - self initStackPages. - self loadInitialContext. - self initialEnterSmalltalkExecutive! Item was added: + ----- Method: CogVMSimulator>>setBreakCount: (in category 'accessing') ----- + setBreakCount: anInteger + breakCount := anInteger! Item was removed: - ----- Method: CogVMSimulatorLSB>>charsOfLong: (in category 'debug support') ----- - charsOfLong: long - ^ (1 to: objectMemory wordSize) collect: - [:i | ((long digitAt: i) between: 14 and: 126) - ifTrue: [(long digitAt: i) asCharacter] - ifFalse: [$?]]! Item was removed: - ----- Method: InterpreterSimulator>>runWithBreakCount: (in category 'testing') ----- - runWithBreakCount: breakCount - self internalizeIPandSP. - self fetchNextBytecode. - [true] whileTrue: - [self dispatchOn: currentBytecode in: BytecodeTable. - (byteCount := byteCount + 1) = breakCount ifTrue: - [self halt]]. - localIP := localIP - 1. - "undo the pre-increment of IP before returning" - self externalizeIPandSP! Item was added: + ----- Method: StackInterpreter>>charsOfLong: (in category 'debug support') ----- + charsOfLong: long + "N.B. This will need to be overridden in big-endian simulator subclasses." + <doNotGenerate> + ^ (1 to: objectMemory wordSize) collect: + [:i | | d | + (d := long digitAt: i) >= 14 + ifTrue: [d asCharacter] + ifFalse: [$?]]! Item was added: + ----- Method: StackInterpreterSimulator>>breakCount (in category 'accessing') ----- + breakCount + ^breakCount! Item was removed: - ----- Method: StackInterpreterSimulator>>charsOfLong: (in category 'debug support') ----- - charsOfLong: long - ^self subclassResponsibility! Item was changed: ----- Method: StackInterpreterSimulator>>incrementByteCount (in category 'interpreter shell') ----- incrementByteCount (byteCount := byteCount + 1) = breakCount ifTrue: [self doOrDefer: [self changed: #byteCountText]. + self halt: 'breakCount reached']. - self halt]. byteCount \\ 1000 = 0 ifTrue: [self doOrDefer: [self changed: #byteCountText]. self forceInterruptCheck]! Item was removed: - ----- Method: StackInterpreterSimulator>>runWithBreakCount: (in category 'testing') ----- - runWithBreakCount: theBreakCount - "Just run, halting when byteCount is reached" - quitBlock := [displayView ifNotNil: - [displayView containingWindow ifNotNil: - [:topWindow| - ((World submorphs includes: topWindow) - and: [UIManager default confirm: 'close?']) ifTrue: - [topWindow delete]]]. - ^self close]. - breakCount := theBreakCount. - self initStackPages. - self loadInitialContext. - self internalizeIPandSP. - self fetchNextBytecode. - [true] whileTrue: - [self assertValidExecutionPointers. - self dispatchOn: currentBytecode in: BytecodeTable. - self incrementByteCount]. - localIP := localIP - 1. - "undo the pre-increment of IP before returning" - self externalizeIPandSP! Item was added: + ----- Method: StackInterpreterSimulator>>setBreakCount: (in category 'accessing') ----- + setBreakCount: anInteger + breakCount := anInteger! Item was changed: ----- Method: StackInterpreterSimulator>>setClickStepBreakBlock (in category 'UI') ----- setClickStepBreakBlock "Set the break block to present a confirmer, breaking if true, and restoring the previous break block. If an open debugger on the receiver can be found, proceed it." <doNotGenerate> | previousAtEachStepBlock | (atEachStepBlock isNil or: [atEachStepBlock method ~~ thisContext method]) ifTrue: [previousAtEachStepBlock := atEachStepBlock. atEachStepBlock := [previousAtEachStepBlock value. + self changed: #byteCountText. (UIManager confirm: 'step?') ifFalse: [atEachStepBlock := previousAtEachStepBlock. self halt]]]. (World submorphs detect: [:m| m model class == Debugger and: [(m model interruptedProcess suspendedContext findContextSuchThat: [:ctxt| ctxt receiver == self and: [ctxt selector == #run]]) notNil]] ifNone: []) ifNotNil: [:debuggerWindow| WorldState addDeferredUIMessage: [debuggerWindow model proceed]]! Item was removed: - ----- Method: StackInterpreterSimulatorLSB>>charsOfLong: (in category 'debug support') ----- - charsOfLong: long - ^ (1 to: objectMemory wordSize) collect: - [:i | ((long digitAt: i) between: 14 and: 126) - ifTrue: [(long digitAt: i) asCharacter] - ifFalse: [$?]]!

1 0

[OpenSmalltalk/opensmalltalk-vm] Fix improper commenting (#188)
by Stuart Cassoff 05 Jan '18

05 Jan '18

It looks like the original intent was to disable ac3 on line 119 like on line 95, and afterwards there was a decision to drop file extensions altogether. You can view, comment on, or merge this pull request online at: https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/188 -- Commit Summary -- * Fix improper commenting -- File Changes -- M platforms/Cross/plugins/Mpeg3Plugin/libmpeg/libmpeg3.c (15) -- Patch Links -- https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/188.patch https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/188.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/188

2 2

[OpenSmalltalk/opensmalltalk-vm] 0b70c2: Fix improper commenting
by GitHub 05 Jan '18

05 Jan '18

Branch: refs/heads/Cog Home: https://github.com/OpenSmalltalk/opensmalltalk-vm Commit: 0b70c2ca8a47ffa09ddccb42e7dc894ab080c6e4 https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/0b70c2ca8a47ffa09d… Author: Stuart Cassoff <aryler(a)users.noreply.github.com> Date: 2018-01-04 (Thu, 04 Jan 2018) Changed paths: M platforms/Cross/plugins/Mpeg3Plugin/libmpeg/libmpeg3.c Log Message: ----------- Fix improper commenting It looks like the original intent was to disable ac3 on line 119 like on line 95, and afterwards there was a decision to drop file extensions altogether. Commit: 0c7c4d395f5c78e015f1528bc70391fb4da8a7eb https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/0c7c4d395f5c78e015… Author: Eliot Miranda <eliot.miranda(a)gmail.com> Date: 2018-01-05 (Fri, 05 Jan 2018) Changed paths: M platforms/Cross/plugins/Mpeg3Plugin/libmpeg/libmpeg3.c Log Message: ----------- Merge pull request #188 from aryler/aryler-crazycomments Fix improper commenting Compare: https://github.com/OpenSmalltalk/opensmalltalk-vm/compare/2be6dc8df271...0c…

1 0

VM Maker: VMMaker.oscog-eem.2308.mcz
by commits＠source.squeak.org 05 Jan '18

05 Jan '18

Eliot Miranda uploaded a new version of VMMaker to project VM Maker: http://source.squeak.org/VMMaker/VMMaker.oscog-eem.2308.mcz ==================== Summary ==================== Name: VMMaker.oscog-eem.2308 Author: eem Time: 5 January 2018, 12:19:53.562181 am UUID: 6c41bece-0d81-4a69-a677-bf6e8f275674 Ancestors: VMMaker.oscog-eem.2307 Slang: Make TParseNode>>comment; do what it implies and accept a string argument, instead of insistign clients know the internal contract is a sequence of comment strings. Tweak the recent inlining comment change; no need to comment if what's being inlined is a named constant. =============== Diff against VMMaker.oscog-eem.2307 =============== Item was changed: ----- Method: TMethod>>tryToInlineMethodExpressionsIn: (in category 'inlining') ----- tryToInlineMethodExpressionsIn: aCodeGen "Expand any (complete) inline methods sent by this method as receivers or parameters. Answer if anything was inlined." | sendsToInline | sendsToInline := Dictionary new: 100. aCodeGen pushScope: declarations while: [parseTree nodesDo: [:node| (self inlineableFunctionCall: node in: aCodeGen) ifTrue: [(self inlineFunctionCall: node in: aCodeGen) ifNotNil: [:replacement| + (replacement isConstant + and: [replacement isDefine not + and: [replacement value isNumber + and: [replacement comment isNil]]]) ifTrue: - (replacement isConstant and: [replacement value isNumber and: [replacement comment isNil]]) ifTrue: [replacement comment: node selector]. sendsToInline at: node put: replacement]]] unless: "Don't inline the arguments to asserts to keep the asserts readable" [:node| node isSend and: [node selector == #cCode:inSmalltalk: or: [aCodeGen isAssertSelector: node selector]]]]. sendsToInline isEmpty ifTrue: [^false]. self replaceNodesIn: sendsToInline. ^true! Item was changed: ----- Method: TParseNode>>comment: (in category 'accessing') ----- comment: aComment + "Ugh,. comment is a sequence of strings." + comment := aComment isString ifTrue: [{aComment}] ifFalse: [aComment]! - - comment := aComment !

2 1

VM Maker: VMMaker.oscog-eem.2309.mcz
by commits＠source.squeak.org 05 Jan '18

05 Jan '18

Eliot Miranda uploaded a new version of VMMaker to project VM Maker: http://source.squeak.org/VMMaker/VMMaker.oscog-eem.2309.mcz ==================== Summary ==================== Name: VMMaker.oscog-eem.2309 Author: eem Time: 5 January 2018, 12:28:49.355435 am UUID: 122d5ac0-b546-4f56-a020-8e8988fc5c10 Ancestors: VMMaker.oscog-eem.2307 Tweak the recent inlining commenting change; don't bother to comment if what's being inlined is a named constant. =============== Diff against VMMaker.oscog-eem.2307 =============== Item was changed: ----- Method: TMethod>>tryToInlineMethodExpressionsIn: (in category 'inlining') ----- tryToInlineMethodExpressionsIn: aCodeGen "Expand any (complete) inline methods sent by this method as receivers or parameters. Answer if anything was inlined." | sendsToInline | sendsToInline := Dictionary new: 100. aCodeGen pushScope: declarations while: [parseTree nodesDo: [:node| (self inlineableFunctionCall: node in: aCodeGen) ifTrue: [(self inlineFunctionCall: node in: aCodeGen) ifNotNil: [:replacement| + (replacement isConstant + and: [replacement isDefine not + and: [replacement value isNumber + and: [replacement comment isNil]]]) ifTrue: - (replacement isConstant and: [replacement value isNumber and: [replacement comment isNil]]) ifTrue: [replacement comment: node selector]. sendsToInline at: node put: replacement]]] unless: "Don't inline the arguments to asserts to keep the asserts readable" [:node| node isSend and: [node selector == #cCode:inSmalltalk: or: [aCodeGen isAssertSelector: node selector]]]]. sendsToInline isEmpty ifTrue: [^false]. self replaceNodesIn: sendsToInline. ^true!

1 0

VM Maker: VMMaker.oscog-eem.2307.mcz
by commits＠source.squeak.org 05 Jan '18

05 Jan '18

Eliot Miranda uploaded a new version of VMMaker to project VM Maker: http://source.squeak.org/VMMaker/VMMaker.oscog-eem.2307.mcz ==================== Summary ==================== Name: VMMaker.oscog-eem.2307 Author: eem Time: 5 January 2018, 12:07:17.508019 am UUID: bede75bd-3b6a-49d4-b6a8-14325d3573c7 Ancestors: VMMaker.oscog-eem.2306 Refactor preDeclareInterpreterProxyOn: to extract collecting the InterpreterProxy interface to its own method to simplify consistency checking. To make the checking more correct use a concrete Sour class for the referenceObjectMemoryClass. Fix several InterpreterProxy interface methods that don't specify a return type. =============== Diff against VMMaker.oscog-eem.2306 =============== Item was changed: ----- Method: InterpreterProxy>>become:with: (in category 'other') ----- become: array1 with: array2 + <returnTypeC: #sqInt> array1 elementsExchangeIdentityWith: array2! Item was changed: ----- Method: InterpreterProxy>>copyBits (in category 'BitBlt support') ----- copyBits + <returnTypeC: #sqInt> + bb copyBits! - bb copyBits.! Item was changed: ----- Method: InterpreterProxy>>copyBitsFrom:to:at: (in category 'BitBlt support') ----- copyBitsFrom: leftX to: rightX at: yValue + <returnTypeC: #sqInt> bb copyBitsFrom: leftX to: rightX at: yValue.! Item was changed: ----- Method: InterpreterProxy>>fullDisplayUpdate (in category 'other') ----- fullDisplayUpdate + <returnTypeC: #sqInt> Display display! Item was changed: ----- Method: InterpreterProxy>>loadBitBltFrom: (in category 'BitBlt support') ----- loadBitBltFrom: bbOop + <returnTypeC: #sqInt> + bb := bbOop! - bb := bbOop.! Item was changed: ----- Method: InterpreterProxy>>reestablishContextPriorToCallback: (in category 'callback support') ----- reestablishContextPriorToCallback: callbackContext "callbackContext is an activation of invokeCallback:stack:registers:jmpbuf:. Its sender is the interpreter's state prior to the callback. Reestablish that state." + <returnTypeC: #sqInt> self notYetImplemented! Item was changed: ----- Method: InterpreterProxy>>sendInvokeCallback:Stack:Registers:Jmpbuf: (in category 'callback support') ----- sendInvokeCallback: thunkPtr Stack: stackPtr Registers: regsPtr Jmpbuf: jmpBufPtr "Send the 4 argument callback message invokeCallback:stack:registers:jmpbuf: to Alien class with the supplied args. The arguments are raw C addresses and are converted to integer objects on the way." + <returnTypeC: #sqInt> self notYetImplemented! Item was changed: ----- Method: InterpreterProxy>>sendInvokeCallbackContext: (in category 'callback support') ----- sendInvokeCallbackContext: vmCallbackContext <var: #vmCallbackContext type: #'VMCallbackContext *'> "Send the calllback message to Alien class with the supplied arg(s). Use either the 1 arg invokeCallbackContext: or the 4 arg invokeCallback:stack:registers:jmpbuf: message, depending on what selector is installed in the specialObjectsArray. Note that if invoking the legacy invokeCallback:stack:registers:jmpbuf: we pass the vmCallbackContext as the jmpbuf argument (see reestablishContextPriorToCallback:). The arguments are raw C addresses and are converted to integer objects on the way." + <returnTypeC: #sqInt> self notYetImplemented! Item was changed: ----- Method: InterpreterProxy>>stringForCString: (in category 'testing') ----- stringForCString: aCString "Answer a ByteString object containing the bytes (possibly UTF-8?) in the null-terminated C string aCString." + <returnTypeC: #sqInt> <var: #aCString type: #'char *'> self notYetImplemented! Item was changed: ----- Method: VMMaker class>>generateVMPlugins (in category 'configurations') ----- generateVMPlugins ^VMMaker generatePluginsTo: (FileDirectory default pathFromURI: self sourceTree, '/src') options: #() platformDir: (FileDirectory default pathFromURI: self sourceTree, '/platforms') including:#( ADPCMCodecPlugin AsynchFilePlugin BalloonEnginePlugin B3DAcceleratorPlugin B3DEnginePlugin BMPReadWriterPlugin BitBltSimulation BochsIA32Plugin BochsX64Plugin + CameraPlugin CroquetPlugin DeflatePlugin DropPlugin + "Cryptography Plugins:" DESPlugin DSAPlugin MD5Plugin SHA256Plugin + "FT2Plugin" FFTPlugin FileCopyPlugin FilePlugin FileAttributesPlugin FloatArrayPlugin FloatMathPlugin - CameraPlugin CroquetPlugin DSAPlugin DeflatePlugin DropPlugin - "FT2Plugin" FFTPlugin FileCopyPlugin FilePlugin FloatArrayPlugin FloatMathPlugin GeniePlugin GdbARMPlugin HostWindowPlugin IA32ABIPlugin ImmX11Plugin InternetConfigPlugin JPEGReadWriter2Plugin JPEGReaderPlugin JoystickTabletPlugin KlattSynthesizerPlugin LargeIntegersPlugin LocalePlugin MIDIPlugin MacMenubarPlugin Matrix2x3Plugin MiscPrimitivePlugin Mpeg3Plugin QuicktimePlugin RePlugin ScratchPlugin SecurityPlugin SerialPlugin SocketPlugin SoundCodecPlugin SoundGenerationPlugin SoundPlugin SqueakSSLPlugin StarSqueakPlugin ThreadedFFIPlugin ThreadedARMFFIPlugin ThreadedIA32FFIPlugin ThreadedX64SysVFFIPlugin ThreadedX64Win64FFIPlugin UnicodePlugin UnixAioPlugin UUIDPlugin UnixOSProcessPlugin Win32OSProcessPlugin VMProfileLinuxSupportPlugin VMProfileMacSupportPlugin WeDoPlugin XDisplayControlPlugin)! Item was added: + ----- Method: VMPluginCodeGenerator>>collectAndCheckInterpreterProxyInterfaceFor:verbose: (in category 'utilities') ----- + collectAndCheckInterpreterProxyInterfaceFor: selectors verbose: beVerbose + "self new + collectAndCheckInterpreterProxyInterfaceFor: (InterpreterProxy selectors reject: [:s| #(initialize private) includes: (InterpreterProxy whichCategoryIncludesSelector: s)]) + verbose: true" + | interpreterClass objectMemoryClass | + interpreterClass := self referenceInterpreterClass. + objectMemoryClass := self referenceObjectMemoryClass. + ^selectors collect: + [:selector| | reference actual | + reference := self compileToTMethodSelector: selector + in: ((interpreterClass whichClassIncludesSelector: selector) ifNil: + [(objectMemoryClass whichClassIncludesSelector: selector) ifNil: + [InterpreterProxy]]). + actual := self compileToTMethodSelector: selector in: InterpreterProxy. + { actual. reference } do: + [:tMethod| + tMethod recordDeclarationsIn: self. + tMethod returnType ifNil: + [tMethod inferReturnTypeIn: self. + tMethod returnType ifNil: + [tMethod returnType: #sqInt]]]. + (reference returnType ~= actual returnType + or: [(1 to: reference args size) anySatisfy: + [:i| (reference typeFor: (reference args at: i) in: self) + ~= (actual typeFor: (actual args at: i) in: self)]]) ifTrue: + [self logger + nextPutAll: 'warning, signature of InterpreterProxy>>'; + nextPutAll: selector; + nextPutAll: ' does not match reference implementation.'; + cr. + beVerbose ifTrue: + [self logger nextPutAll: 'reference:'; tab. + reference emitCFunctionPrototype: self logger generator: self. + self logger nextPutAll: 'actual:'; tab; tab. + actual emitCFunctionPrototype: self logger generator: self]. + self logger flush"; halt: selector"]. + actual]! Item was changed: ----- Method: VMPluginCodeGenerator>>preDeclareInterpreterProxyOn: (in category 'C code generator') ----- preDeclareInterpreterProxyOn: aStream "Put the necessary #defines needed before interpreterProxy. Basically internal plugins use the VM's interpreterProxy variable and external plugins use their own. Override to keep local copies of all functions in external prims, and link directly in internal plugins." "| pcc | pcc := self new. (InterpreterProxy selectors reject: [:s| #(initialize private) includes: (InterpreterProxy whichCategoryIncludesSelector: s)]) do: [:s| pcc noteUsedPluginFunction: s]. pcc preDeclareInterpreterProxyOn: Transcript. Transcript flush" + | pluginFuncs | - | pluginFuncs interpreterClass objectMemoryClass | self notePluginFunctionsUsedByMacros. (pluginFuncs := self pluginFunctionsToClone) isEmpty ifTrue: [^super preDeclareInterpreterProxyOn: aStream]. (pluginFuncs includesAnyOf: self selectorsThatAreGeneratedAsMacros) ifTrue: [self preDeclareMacrosForFastClassCheckingOn: aStream]. pluginFuncs := pluginFuncs copyWithoutAll: self selectorsThatAreGeneratedAsMacros. pluginFuncs isEmpty ifTrue: [^self]. + pluginFuncs := self collectAndCheckInterpreterProxyInterfaceFor: pluginFuncs verbose: false. - interpreterClass := self referenceInterpreterClass. - objectMemoryClass := self referenceObjectMemoryClass. - pluginFuncs := pluginFuncs collect: - [:selector| | reference actual | - reference := self compileToTMethodSelector: selector - in: ((interpreterClass whichClassIncludesSelector: selector) ifNil: - [(objectMemoryClass whichClassIncludesSelector: selector) ifNil: - [InterpreterProxy]]). - actual := self compileToTMethodSelector: selector in: InterpreterProxy. - { actual. reference } do: - [:tMethod| - tMethod recordDeclarationsIn: self. - tMethod returnType ifNil: - [tMethod inferReturnTypeIn: self]]. - (reference returnType ~= actual returnType - or: [(1 to: reference args size) anySatisfy: - [:i| (reference typeFor: (reference args at: i) in: self) - ~= (actual typeFor: (actual args at: i) in: self)]]) ifTrue: - [self logger - nextPutAll: 'warning, signature of InterpreterProxy>>'; - nextPutAll: selector; - nextPutAll: ' does not match reference implementation.'; - cr]. - actual]. aStream cr; nextPutAll: '#if !!defined(SQUEAK_BUILTIN_PLUGIN)'; cr. pluginFuncs do: [:tMethod| | functionName | functionName := self cFunctionNameFor: tMethod selector. aStream nextPutAll: ((String streamContents: [:s| tMethod static: true; emitCFunctionPrototype: s generator: self]) copyReplaceAll: functionName with: '(*', functionName, ')' tokenish: [:ch| ch = $_ or: [ch isAlphaNumeric]])]. aStream nextPutAll: '#else /* !!defined(SQUEAK_BUILTIN_PLUGIN) */'; cr. pluginFuncs do: [:tMethod| self withGuardAgainstDefinitionOf: tMethod selector on: aStream do: [self withOptionalVerbiageFor: tMethod selector on: aStream do: [tMethod static: false; export: false; emitCFunctionPrototype: aStream generator: self] ifOptionalDo: [aStream nextPutAll: '# define '. (TSendNode new setSelector: tMethod selector receiver: (TVariableNode new setName: 'interpreterProxy') arguments: (tMethod args collect: [:a| TVariableNode new setName: a])) emitCCodeAsArgumentOn: aStream level: 0 generator: self. aStream nextPutAll: ' 0'; cr]]]. aStream nextPutAll: 'extern'; cr; nextPutAll: '#endif'; cr! Item was changed: ----- Method: VMPluginCodeGenerator>>referenceObjectMemoryClass (in category 'C code generator') ----- referenceObjectMemoryClass "Define the class from which to take methods to define the interpreter proxy imports." + ^(Smalltalk classNamed: #Spur32BitMemoryManager) ifNil: - ^(Smalltalk classNamed: #SpurMemoryManager) ifNil: [(Smalltalk classNamed: #NewObjectMemory) ifNil: [ObjectMemory]]!

1 0

[OpenSmalltalk/opensmalltalk-vm] Let Win64 ABI use an unwind-free setjmp (#189)
by Nicolas Cellier 05 Jan '18

05 Jan '18

You can view, comment on, or merge this pull request online at: https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/189 -- Commit Summary -- * Let Win64 ABI use an unwind-free setjmp -- File Changes -- M platforms/Cross/plugins/IA32ABI/ia32abi.h (12) -- Patch Links -- https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/189.patch https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/189.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/189

2 3

[OpenSmalltalk/opensmalltalk-vm] 55dcd6: Let Win64 ABI use an unwind-free setjmp
by GitHub 05 Jan '18

05 Jan '18

Branch: refs/heads/Cog Home: https://github.com/OpenSmalltalk/opensmalltalk-vm Commit: 55dcd605448d0cf305365982dcab31e69c58cb7d https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/55dcd605448d0cf305… Author: Nicolas Cellier <nicolas.cellier.aka.nice(a)gmail.com> Date: 2018-01-04 (Thu, 04 Jan 2018) Changed paths: M platforms/Cross/plugins/IA32ABI/ia32abi.h Log Message: ----------- Let Win64 ABI use an unwind-free setjmp Commit: 00f06826bef6cf2f0bac56d41d573b9d4eac4970 https://github.com/OpenSmalltalk/opensmalltalk-vm/commit/00f06826bef6cf2f0b… Author: Eliot Miranda <eliot.miranda(a)gmail.com> Date: 2018-01-04 (Thu, 04 Jan 2018) Changed paths: M platforms/Cross/plugins/IA32ABI/ia32abi.h Log Message: ----------- Merge pull request #189 from OpenSmalltalk/Win64_ABI_use_unwind_free_setjmp Let Win64 ABI use an unwind-free setjmp Compare: https://github.com/OpenSmalltalk/opensmalltalk-vm/compare/3059a3de8c6d...00…

1 0

[OpenSmalltalk/opensmalltalk-vm]
by GitHub 04 Jan '18

04 Jan '18

Branch: refs/heads/Windows_vm_doesnt_require_libssl Home: https://github.com/OpenSmalltalk/opensmalltalk-vm

1 0

← Newer
1
...
19
20
21
22
23
24
25
26
Older →

Jump to page:

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

Vm-dev January 2018