Smalltalk › Squeak › Squeak VM

Re: [OpenSmalltalk/opensmalltalk-vm] Lowcode fixup (#111)

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

2 messages Options

Eliot Miranda-2

Re: [OpenSmalltalk/opensmalltalk-vm] Lowcode fixup (#111)

Hi Ronie,

On Wed, Jan 11, 2017 at 12:12 AM, Ronie Salgado <[hidden email]> wrote:

Fixing Lowcode to work in 32 bits Linux, OS X and Windows.

This pull request includes the adaptation of normal OpenSmalltalk VM building scripts to build some Pharo VMs with the Lowcode inline primitives in Linux and OS X. In Windows I failed to build a Pharo VM due to the dependencies, so I tested using a Squeak VM with the Lowcode instructions and a Pharo image ( http://ronie.cl/files/images/Pharo-lowcode.zip ).

What is changed:

More build scripts

Fixing the LowcodeFFI callout interpreter trampolines for building in OS X and Windows

Some missing memory accessors.

What is working with these changes:

All WodenMath tests are passing. WodenMath uses the Lowcode type system for generating Lowcode instructions at compile time.

UnifiedFFI callouts using Lowcode (Need to install the Lowcode uFFI backend by calling LowcodeFFICalloutAPI install, otherwise the SqueakFFI backend will be used).

Woot!

What is not working(yet):

UnifiedFFI callbacks using Lowcode. Crashing immediately in Windows. In Linux, does not crash all of the time, which seems to be some difference between the Interpreter and the Cogit

Can you describe what a Locoed callback is in relation to Alien callback machinery? Does the Locoed callback come in through thunkEntry or by another route?

and below...

You can view, comment on, or merge this pull request online at:

https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/111

Commit Summary

I am adding some building scripts for Pharo with Lowcode.

Newer version of the Lowcode callout trampoline.

Building scripts for a pharo VM with Lowcode.

Missing premissions from the latest commit.

Oops

I added the building scripts for a Pharo Lowcode VM in OS X.

Oops, missing permissions from the last commit.

Adding the generated Lowcode sources.

Build scripts for the Lowcode Pharo VM in Windows.

More building scripts for the Windows Lowcode VM.

Adding missing accessors for Lowcode.

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Trying to fix GCC crash when building a Lowcode VM in Windows.

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Newer version of the Lowcode sources.

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Minor changes to the Lowcode build scripts for Windows.

Merge branch 'Cog' into LowcodeFixup

File Changes

A build.linux32x86/pharo.cog.lowcode/build.assert.itimerheartbeat/mvm (37)

The build directly names break the naming convention. Since the VMs below are Spur perhaps we should call them pharo.cog.spur.lowcode? e.g. when I build immutability VMs I call the directory e.g. squeak.cog.spur.immutability. So maybe we can use some convention like dialect.[sista|cog|stack].[spur|v3].[option1+...optionN]. Or, if lowcode always implies Sista support then pharo.lowcode.spur would work.

[snip]

_,,,^..^,,,_

best, Eliot

Ronie Salgado

Re: [OpenSmalltalk/opensmalltalk-vm] Lowcode fixup (#111)

Hi Eliot,

Can you describe what a Locoed callback is in relation to Alien callback machinery? Does the Locoed callback come in through thunkEntry or by another route?

I am using the standard Alien Callback machinery. I have not started to debug completely this problem, but I have my suspicious: The full sequence is something like this:

- Smalltalk method with a ffi callout

- Argument marshalling using Lowcode

- Actual callout: Switch from the Smalltalk Stack to the C stack via a special trampoline. The trampoline is used for storing the return instruction pointer. Probably I am also adding a reverse trampoline. This is the part where I have my suspicious

- Some C code that invokes the callback in some time

- The smalltalk callback

During the rest of the week I will try to fix this problem. However, my priority right now is in supporting 64 bits with Lowcode.

The build directly names break the naming convention. Since the VMs below are Spur perhaps we should call them pharo.cog.spur.lowcode? e.g. when I build immutability VMs I call the directory e.g. squeak.cog.spur.immutability. So maybe we can use some convention like dialect.[sista|cog|stack].[spur|v3].[option1+...optionN]. Or, if lowcode always implies Sista support then pharo.lowcode.spur would work.

Fixed. pharo.cog.spur.lowcode . The actual Sista Cogit is not mandatory for supporting Lowcode. Only StackToRegisterMappingCogit with the LowcodeVM and the SistaVM flags is required. Probably, I should try to remove the requirement on the SistaVM option.

Best regards,

Ronie

2017-01-11 13:36 GMT-03:00 Eliot Miranda <[hidden email]>:

Hi Ronie,

On Wed, Jan 11, 2017 at 12:12 AM, Ronie Salgado <[hidden email]> wrote:
Fixing Lowcode to work in 32 bits Linux, OS X and Windows.

This pull request includes the adaptation of normal OpenSmalltalk VM building scripts to build some Pharo VMs with the Lowcode inline primitives in Linux and OS X. In Windows I failed to build a Pharo VM due to the dependencies, so I tested using a Squeak VM with the Lowcode instructions and a Pharo image ( http://ronie.cl/files/images/Pharo-lowcode.zip ).

What is changed:

More build scripts

Fixing the LowcodeFFI callout interpreter trampolines for building in OS X and Windows

Some missing memory accessors.

What is working with these changes:

All WodenMath tests are passing. WodenMath uses the Lowcode type system for generating Lowcode instructions at compile time.

UnifiedFFI callouts using Lowcode (Need to install the Lowcode uFFI backend by calling LowcodeFFICalloutAPI install, otherwise the SqueakFFI backend will be used).
Woot!

What is not working(yet):

UnifiedFFI callbacks using Lowcode. Crashing immediately in Windows. In Linux, does not crash all of the time, which seems to be some difference between the Interpreter and the Cogit

Can you describe what a Locoed callback is in relation to Alien callback machinery? Does the Locoed callback come in through thunkEntry or by another route?

and below...

You can view, comment on, or merge this pull request online at:

https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/111

Commit Summary

I am adding some building scripts for Pharo with Lowcode.

Newer version of the Lowcode callout trampoline.

Building scripts for a pharo VM with Lowcode.

Missing premissions from the latest commit.

Oops

I added the building scripts for a Pharo Lowcode VM in OS X.

Oops, missing permissions from the last commit.

Adding the generated Lowcode sources.

Build scripts for the Lowcode Pharo VM in Windows.

More building scripts for the Windows Lowcode VM.

Adding missing accessors for Lowcode.

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Trying to fix GCC crash when building a Lowcode VM in Windows.

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Newer version of the Lowcode sources.

Merge branch 'LowcodeFixup' of github.com:ronsaldo/opensmalltalk-vm into LowcodeFixup

Minor changes to the Lowcode build scripts for Windows.

Merge branch 'Cog' into LowcodeFixup

File Changes

A build.linux32x86/pharo.cog.lowcode/build.assert.itimerheartbeat/mvm (37)

The build directly names break the naming convention. Since the VMs below are Spur perhaps we should call them pharo.cog.spur.lowcode? e.g. when I build immutability VMs I call the directory e.g. squeak.cog.spur.immutability. So maybe we can use some convention like dialect.[sista|cog|stack].[spur|v3].[option1+...optionN]. Or, if lowcode always implies Sista support then pharo.lowcode.spur would work.

[snip]

_,,,^..^,,,_
best, Eliot