Search results for

“translate scheme”

6,658 results found

Post

Replies

Boosts

Views

Activity

C program posix_spawn diskutil fails with error -69877
Hello, I am programming a CLI tool to partition USB disks. I am calling diskutil to do the work, but I am hitting issues with permissions, it seems. Here is a trial run of the same command running diskutil directly on the terminal vs running from my code: Calling diskutil directly (works as expected) % /usr/sbin/diskutil partitionDisk /dev/disk2 MBR Free Space gap 2048S fat32 f-fix 100353S Free Space tail 0 Started partitioning on disk2 Unmounting disk Creating the partition map Waiting for partitions to activate Formatting disk2s1 as MS-DOS (FAT32) with name f-fix 512 bytes per physical sector /dev/rdisk2s1: 98784 sectors in 98784 FAT32 clusters (512 bytes/cluster) bps=512 spc=1 res=32 nft=2 mid=0xf8 spt=32 hds=16 hid=2079 drv=0x80 bsec=100360 bspf=772 rdcl=2 infs=1 bkbs=6 Mounting disk Finished partitioning on disk2 /dev/disk2 (disk image): #: TYPE NAME SIZE IDENTIFIER 0: FDisk_partition_scheme +104.9 MB disk2 1: DOS_FAT_32 F-FIX 51.4 MB disk2s1 Calling diskutil programmatically (error -69877) % sudo ./f-fi
1
0
251
May ’25
Shortcuts Services doesn't work properly in Books & iMessages
The built-in Books and iMessages on the latest macOS can not handle Shortcuts properly. If Books (no matter the Home scheme or the reading scheme) or iMessages is the current focused application, Shortcuts doesn't work. Once I move out and focus app turns to Finder or any other app, Shortcuts works properly. An exception is that when I pin the shortcut in the Menu Bar, the Menu Bar one works, while the one in the application's menu doesn't work. I have no idea why this would happen. Could it be part of privilege control or something?
0
0
179
May ’25
Reply to Can someone review this log and tell me if suspcious
Dealing with same. Have attempted to sanitize, have quarantined all devices, changed router, changed devices x3, pre and post paid, changed device manufacturer, transfer provider, cancel cable, hell we even cut up and threw away all chip cards. After doing all the above in tandem was able to get a week or so of ”normalcy”. Have become much more educated over last 90 days for sure. All the above fits my description, however I believe that they are targeting crypto wallets. Agree that it’s a spy or malware, can’t agree enough that it’s more personal than financially motivated. Apple (especially Apple Community) act as though their first born will cease to function by admitting that a high school kid with a gaming computer can log into github and easily take ownership of any device. Swear you gotta be important and it takes millions of dollars to get into iPhones. So happy I found this thread, especially with an expert who’s at least willing to admit it’s possible. I’ve just started reading (and unzipping/translating
May ’25
Suspicious CloudKit Telemetry Data
Starting 20th March 2025, I see an increase in bandwidth and latency for one of my CloudKit projects. I'm using NSPersistentCloudKitContainer to synchronise my data. I haven't changed any CloudKit scheme during that time but shipped an update. Since then, I reverted some changes from that update, which could have led to changes in the sync behaviour. Is anyone else seeing any issues? I would love to file a DTS and use one of my credits for that, but unfortunately, I can't because I cannot reproduce it with a demo project because I cannot travel back in time and check if it also has an increase in metrics during that time. Maybe an Apple engineer can green-light me filing a DTS request, please.
0
0
158
Apr ’25
Reply to mac 开发 com.apple.security.application-groups 问题
Sadly, I don’t read Chinese, so I’m working from a machine translation of your post. I’m not entirely sure what’s going on here but, in general, Mach IPC should be able to work with any service name that’s prefixed by a valid app group ID. Now, the definition of valid app group ID has been in flux recently. I talked about that evolution in App Groups: macOS vs iOS: Working Towards Harmony. My advice here varies based on whether this is new code or code that you’ve already shipped and need to maintain compatibility with. For new code, use an iOS-style app group and have that as your Mach service name. For example, group.eskimo1.test with a Mach service name of group.eskimo1.test.my-service. Then make sure that your use of that app group is authorised by a provisioning profile. For existing code, it depends on whether this Mach IPC is entirely within your product: If this IPC is inside your product — for example, you’re communicating between two subcomponents of the same app — then you can use an iOS-s
Apr ’25
Denmark's biggest mobile payment service is skirting IAP rules
Hello I recently discovered that Denmark's (and soon the rest of the Nordic countries) biggest mobile payment service - MobilePay - is seemingly getting away with selling digital products in their app without using IAP. I am confused how this is possible, as I spent a lot of time making sure I understand the rules -- and I'm quite confident I do, so the only conclusion I can reach from this is that this is a violation that has not been discovered. If it isn't, I would like an explanation. So, what am I talking about? First; MobilePay is an app that lets you send money to whoever, essentially a Danish version of CashApp. You attach a credit card, and you can transfer funds using people's phone numbers. Nothing about this is problematic. There are some fees involved, which is also fine. But. You can select to wrap your money in digital wrapping paper, which can be unwrapped by the recipient to reveal the amount you transferred. This is, no matter how you spin it, a digital product. It adds about 1 USD to the tr
2
0
135
Apr ’25
Cannot Find UI to Add Core Data Database Indexes in Xcode 16.2
Hi everyone, I'm trying to add standard, non-unique database indexes to my Core Data entities for performance optimization (e.g., indexing Date or String attributes used in predicates and sort descriptors). I'm using Xcode 16.2 on macOS Sequoia 15.1. My problem is that I cannot find the expected UI element in the Core Data model editor (.xcdatamodeld) to configure these database indexes. What I Understand / Expect: I know the old Indexed checkbox on the Attribute Inspector is deprecated/gone. My understanding from recent documentation and tutorials is that database indexing (separate from Spotlight indexing) should be configured in the Entity Inspector (when the Entity itself is selected), within a section titled Indexes (usually located below Constraints). This Indexes section should allow adding individual or compound indexes that translate to SQL CREATE INDEX commands, distinct from uniqueness constraints. What I'm Experiencing: When I select an Entity in the model editor, the Indexes section is c
0
0
96
Apr ’25
Request for Rosetta: support optionally faster x87 emulation (via some env variable similar to AVX) like Rosettax87 project..
Hi, Since recently there is interest in having faster x87 translation speeds than Rosetta offers.. mainly some old PC games getting stuck at less than 5fps using Wine that uses Rosetta..( some world of warcraft game for ex.).. so main case right now, is games using old fmodex library versions (dll or statically)that uses heavy x87 instructions for audio processing , and such games not being threaded ,stalls the render threead, which is the same thread.. Luckily there is a hack, see: https://github.com/Lifeisawful/rosettax87 ”This is an experimental project that modifies Apple's Rosetta technology to use less precise but significantly faster x87 instruction handlers. The benchmarks show approximately 4-5x performance improvement for x87 floating-point operations.” but limitations are: 1)it runs only on specific Mac version (15.4.1) due to searching some fixed offsets in current rosetta library that may change with mac updates.. 2)requires to run two binaries (a server and the launcher program).. 3)cur
3
0
318
Apr ’25
How to Complete the Apple Developer Program Enrollment
How to Complete the Apple Developer Program Enrollment I want to register an Apple Developer account. I've communicated many times with Apple technical Advisors via email but haven't received effective solutions. The most recent reply was: 经核实,我们已经将你的请求升级到相关运营团队,如果有进一步的消息,我们会尽快通知你。 Here's the English translation: After verification, we have escalated your request to the relevant operations team. If there are any updates, we will notify you as soon as possible. Almost every time I get the same response, but there's been no progress. Could other developers in the forum give me some advice? How should I communicate to successfully complete the registration? Additional background information: I previously submitted my documents and passed the review, but encountered issues at the payment stage when using a credit card. I asked the advisor whether alternative payment methods were possible if I couldn't use a credit card—I just wanted to understand the options. However, my application was directly revoked
1
0
87
Apr ’25
Errors compiling C++ code for x86_64
I have a project (that uses pre-compiled headers) that uses different compiler options for SOME files. For those files I have this is in my CMakelists.txt: if(CMAKE_C_COMPILER_ID MATCHES GNU|Clang) set_source_files_properties(${AVX_Files} PROPERTIES COMPILE_OPTIONS -mavx;-mavx2;-mfma;-mssse3;-msse4.2) set_source_files_properties(avx_simd_check.cpp PROPERTIES COMPILE_OPTIONS -mxsave) endif() When I build for ARM, it all works :) When I try to build for X86_64, I get the following error for avx_simd_check.cpp: error: current translation unit is compiled with the target feature '+xsave' but the AST file was not 1 error generated. and for all the other files in question: error: current translation unit is compiled with the target feature '+avx' but the AST file was not error: current translation unit is compiled with the target feature '+avx2' but the AST file was not error: current translation unit is compiled with the target feature '+fma' but the AST file was not error: curr
3
0
185
Apr ’25
Reply to Using handleExternalEvents scene modifier to route external events to the correct scene
Thank you for your post. When utilizing Universal Links to initiate a call to your application, the delegate that will be invoked in SwiftUI is: func scene(_ scene: UIScene, willConnectTo session: UISceneSession, options connectionOptions: UIScene.ConnectionOptions) { You can see that and how it works in the link: https://developer.apple.com/documentation/xcode/supporting-universal-links-in-your-app In case you are using deep linking to call your app, how have you registered the link in the info.plist? Scene Matching Rules: The handlesExternalEvents(matching:) modifier is used to specify which external events a scene should respond to. Ensure Correct URL Schemes: Double-check that the URL schemes you're trying to handle match exactly with what your app is set up to receive. For deep links like Companion://, ensure that this scheme is properly registered in your Info.plist under the CFBundleURLTypes key.
xml
 Review Scene Initialization: Your URLHandlerScene is initialized with a tem
Topic: UI Frameworks SubTopic: SwiftUI Tags:
Apr ’25
How to detect iPad trackpad touch-down (indirectPointer) to immediately stop coasting animation
Hello, I have a custom 3D object viewer on iOS that lets users spin the model using the touchscreen or a trackpad and supports coasting (momentum spinning). I need to stop the coasting animation as soon as the user touches down, but I can only immediately detect touches on the screen itself - on the trackpad I can't get an immediate notification of the touches. So far I’ve tried: State.began on my UIPanGestureRecognizer. It only fires after a small movement on both touchscreen and trackpad. .possible on the pan gesture; this state never occurs during the gesture cycle. UIApplicationSupportsIndirectInputEvents = YES in Info.plist; it didn’t make touchesBegan fire for indirectPointer touches. Since UITableView (and other UIScrollView subclasses) clearly detect trackpad “touch-down” to cancel scrolling, there must be a way to receive that event. Does anyone know how to catch the initial trackpad contact—before any movement—on an indirect input device? Below is a minimal code snippet demonstrating the issue. On t
Topic: UI Frameworks SubTopic: UIKit Tags:
2
0
115
Apr ’25
Reply to 关于本地回环地址调节优先级的问题
Sadly, I don’t read Chinese so I’m basing my response on a machine translation of your question. It seems like: You’re on the Mac. You have a product that communicates using TCP/IP over the loopback interface. You’d like to use a packet trace to monitor that communication. If so, you can do that using standard packet tracing techniques. The trick is to specify the loopback interface (lo0) when you start the trace. Here’s how I tested this today: On macOS 15.4, in one Terminal window, I started nc to listen on port 12345 of 127.0.0.1: % nc -l 127.0.0.1 12345 In another window, I told nc to connect to that listener: % nc 127.0.0.1 12345 In a third window, I ran tcpdump like so: % sudo tcpdump -n -i lo0 port 12345 I typed some text into the window from step 2. It generated traffic on the loopback interface which was displayed by tcpdump. Share and Enjoy — Quinn “The Eskimo!” @ Developer Technical Support @ Apple let myEmail = eskimo + 1 + @ + apple.com
Apr ’25
Reply to Why slower with larger threadgroup memory?
The answer would be usually, no. The basic idea underlying modern GPUs is they try to get around memory latency by having A LOT of parallelism. The cores are usually designed to execute several workgroups at the same time. Some hardware (not sure about Apple Silicon, but I knew several CUDA machines that have this) wants you to have several thread groups AND several simd groups/warps in each threadgroup to reach peak efficiency. My understanding is, you even proved this yourself - you got 250GFlops when asking for 32kB and 400GFlops when asking for 8kB of threadgroup memory. That memory request translates into I want only one threadgroup to run on each core vs I am ok with up to 4 thread groups running on each core. If I were you, I would immediately try with 4kb and 16kb, so with up to 8 and up to 2 thread groups, then check performance.
Apr ’25
C program posix_spawn diskutil fails with error -69877
Hello, I am programming a CLI tool to partition USB disks. I am calling diskutil to do the work, but I am hitting issues with permissions, it seems. Here is a trial run of the same command running diskutil directly on the terminal vs running from my code: Calling diskutil directly (works as expected) % /usr/sbin/diskutil partitionDisk /dev/disk2 MBR Free Space gap 2048S fat32 f-fix 100353S Free Space tail 0 Started partitioning on disk2 Unmounting disk Creating the partition map Waiting for partitions to activate Formatting disk2s1 as MS-DOS (FAT32) with name f-fix 512 bytes per physical sector /dev/rdisk2s1: 98784 sectors in 98784 FAT32 clusters (512 bytes/cluster) bps=512 spc=1 res=32 nft=2 mid=0xf8 spt=32 hds=16 hid=2079 drv=0x80 bsec=100360 bspf=772 rdcl=2 infs=1 bkbs=6 Mounting disk Finished partitioning on disk2 /dev/disk2 (disk image): #: TYPE NAME SIZE IDENTIFIER 0: FDisk_partition_scheme +104.9 MB disk2 1: DOS_FAT_32 F-FIX 51.4 MB disk2s1 Calling diskutil programmatically (error -69877) % sudo ./f-fi
Replies
1
Boosts
0
Views
251
Activity
May ’25
Reply to App Crash Only When Downloaded From TestFlight
Try using the same scheme used for local build for Testflight too and see if any issues. Thanks, VoxelBusters Team
Replies
Boosts
Views
Activity
May ’25
Shortcuts Services doesn't work properly in Books & iMessages
The built-in Books and iMessages on the latest macOS can not handle Shortcuts properly. If Books (no matter the Home scheme or the reading scheme) or iMessages is the current focused application, Shortcuts doesn't work. Once I move out and focus app turns to Finder or any other app, Shortcuts works properly. An exception is that when I pin the shortcut in the Menu Bar, the Menu Bar one works, while the one in the application's menu doesn't work. I have no idea why this would happen. Could it be part of privilege control or something?
Replies
0
Boosts
0
Views
179
Activity
May ’25
Reply to Can someone review this log and tell me if suspcious
Dealing with same. Have attempted to sanitize, have quarantined all devices, changed router, changed devices x3, pre and post paid, changed device manufacturer, transfer provider, cancel cable, hell we even cut up and threw away all chip cards. After doing all the above in tandem was able to get a week or so of ”normalcy”. Have become much more educated over last 90 days for sure. All the above fits my description, however I believe that they are targeting crypto wallets. Agree that it’s a spy or malware, can’t agree enough that it’s more personal than financially motivated. Apple (especially Apple Community) act as though their first born will cease to function by admitting that a high school kid with a gaming computer can log into github and easily take ownership of any device. Swear you gotta be important and it takes millions of dollars to get into iPhones. So happy I found this thread, especially with an expert who’s at least willing to admit it’s possible. I’ve just started reading (and unzipping/translating
Replies
Boosts
Views
Activity
May ’25
Suspicious CloudKit Telemetry Data
Starting 20th March 2025, I see an increase in bandwidth and latency for one of my CloudKit projects. I'm using NSPersistentCloudKitContainer to synchronise my data. I haven't changed any CloudKit scheme during that time but shipped an update. Since then, I reverted some changes from that update, which could have led to changes in the sync behaviour. Is anyone else seeing any issues? I would love to file a DTS and use one of my credits for that, but unfortunately, I can't because I cannot reproduce it with a demo project because I cannot travel back in time and check if it also has an increase in metrics during that time. Maybe an Apple engineer can green-light me filing a DTS request, please.
Replies
0
Boosts
0
Views
158
Activity
Apr ’25
Reply to mac 开发 com.apple.security.application-groups 问题
Sadly, I don’t read Chinese, so I’m working from a machine translation of your post. I’m not entirely sure what’s going on here but, in general, Mach IPC should be able to work with any service name that’s prefixed by a valid app group ID. Now, the definition of valid app group ID has been in flux recently. I talked about that evolution in App Groups: macOS vs iOS: Working Towards Harmony. My advice here varies based on whether this is new code or code that you’ve already shipped and need to maintain compatibility with. For new code, use an iOS-style app group and have that as your Mach service name. For example, group.eskimo1.test with a Mach service name of group.eskimo1.test.my-service. Then make sure that your use of that app group is authorised by a provisioning profile. For existing code, it depends on whether this Mach IPC is entirely within your product: If this IPC is inside your product — for example, you’re communicating between two subcomponents of the same app — then you can use an iOS-s
Replies
Boosts
Views
Activity
Apr ’25
Denmark's biggest mobile payment service is skirting IAP rules
Hello I recently discovered that Denmark's (and soon the rest of the Nordic countries) biggest mobile payment service - MobilePay - is seemingly getting away with selling digital products in their app without using IAP. I am confused how this is possible, as I spent a lot of time making sure I understand the rules -- and I'm quite confident I do, so the only conclusion I can reach from this is that this is a violation that has not been discovered. If it isn't, I would like an explanation. So, what am I talking about? First; MobilePay is an app that lets you send money to whoever, essentially a Danish version of CashApp. You attach a credit card, and you can transfer funds using people's phone numbers. Nothing about this is problematic. There are some fees involved, which is also fine. But. You can select to wrap your money in digital wrapping paper, which can be unwrapped by the recipient to reveal the amount you transferred. This is, no matter how you spin it, a digital product. It adds about 1 USD to the tr
Replies
2
Boosts
0
Views
135
Activity
Apr ’25
Cannot Find UI to Add Core Data Database Indexes in Xcode 16.2
Hi everyone, I'm trying to add standard, non-unique database indexes to my Core Data entities for performance optimization (e.g., indexing Date or String attributes used in predicates and sort descriptors). I'm using Xcode 16.2 on macOS Sequoia 15.1. My problem is that I cannot find the expected UI element in the Core Data model editor (.xcdatamodeld) to configure these database indexes. What I Understand / Expect: I know the old Indexed checkbox on the Attribute Inspector is deprecated/gone. My understanding from recent documentation and tutorials is that database indexing (separate from Spotlight indexing) should be configured in the Entity Inspector (when the Entity itself is selected), within a section titled Indexes (usually located below Constraints). This Indexes section should allow adding individual or compound indexes that translate to SQL CREATE INDEX commands, distinct from uniqueness constraints. What I'm Experiencing: When I select an Entity in the model editor, the Indexes section is c
Replies
0
Boosts
0
Views
96
Activity
Apr ’25
Request for Rosetta: support optionally faster x87 emulation (via some env variable similar to AVX) like Rosettax87 project..
Hi, Since recently there is interest in having faster x87 translation speeds than Rosetta offers.. mainly some old PC games getting stuck at less than 5fps using Wine that uses Rosetta..( some world of warcraft game for ex.).. so main case right now, is games using old fmodex library versions (dll or statically)that uses heavy x87 instructions for audio processing , and such games not being threaded ,stalls the render threead, which is the same thread.. Luckily there is a hack, see: https://github.com/Lifeisawful/rosettax87 ”This is an experimental project that modifies Apple's Rosetta technology to use less precise but significantly faster x87 instruction handlers. The benchmarks show approximately 4-5x performance improvement for x87 floating-point operations.” but limitations are: 1)it runs only on specific Mac version (15.4.1) due to searching some fixed offsets in current rosetta library that may change with mac updates.. 2)requires to run two binaries (a server and the launcher program).. 3)cur
Replies
3
Boosts
0
Views
318
Activity
Apr ’25
How to Complete the Apple Developer Program Enrollment
How to Complete the Apple Developer Program Enrollment I want to register an Apple Developer account. I've communicated many times with Apple technical Advisors via email but haven't received effective solutions. The most recent reply was: 经核实,我们已经将你的请求升级到相关运营团队,如果有进一步的消息,我们会尽快通知你。 Here's the English translation: After verification, we have escalated your request to the relevant operations team. If there are any updates, we will notify you as soon as possible. Almost every time I get the same response, but there's been no progress. Could other developers in the forum give me some advice? How should I communicate to successfully complete the registration? Additional background information: I previously submitted my documents and passed the review, but encountered issues at the payment stage when using a credit card. I asked the advisor whether alternative payment methods were possible if I couldn't use a credit card—I just wanted to understand the options. However, my application was directly revoked
Replies
1
Boosts
0
Views
87
Activity
Apr ’25
Errors compiling C++ code for x86_64
I have a project (that uses pre-compiled headers) that uses different compiler options for SOME files. For those files I have this is in my CMakelists.txt: if(CMAKE_C_COMPILER_ID MATCHES GNU|Clang) set_source_files_properties(${AVX_Files} PROPERTIES COMPILE_OPTIONS -mavx;-mavx2;-mfma;-mssse3;-msse4.2) set_source_files_properties(avx_simd_check.cpp PROPERTIES COMPILE_OPTIONS -mxsave) endif() When I build for ARM, it all works :) When I try to build for X86_64, I get the following error for avx_simd_check.cpp: error: current translation unit is compiled with the target feature '+xsave' but the AST file was not 1 error generated. and for all the other files in question: error: current translation unit is compiled with the target feature '+avx' but the AST file was not error: current translation unit is compiled with the target feature '+avx2' but the AST file was not error: current translation unit is compiled with the target feature '+fma' but the AST file was not error: curr
Replies
3
Boosts
0
Views
185
Activity
Apr ’25
Reply to Using handleExternalEvents scene modifier to route external events to the correct scene
Thank you for your post. When utilizing Universal Links to initiate a call to your application, the delegate that will be invoked in SwiftUI is: func scene(_ scene: UIScene, willConnectTo session: UISceneSession, options connectionOptions: UIScene.ConnectionOptions) { You can see that and how it works in the link: https://developer.apple.com/documentation/xcode/supporting-universal-links-in-your-app In case you are using deep linking to call your app, how have you registered the link in the info.plist? Scene Matching Rules: The handlesExternalEvents(matching:) modifier is used to specify which external events a scene should respond to. Ensure Correct URL Schemes: Double-check that the URL schemes you're trying to handle match exactly with what your app is set up to receive. For deep links like Companion://, ensure that this scheme is properly registered in your Info.plist under the CFBundleURLTypes key.
xml
 Review Scene Initialization: Your URLHandlerScene is initialized with a tem
Topic: UI Frameworks SubTopic: SwiftUI Tags:
Replies
Boosts
Views
Activity
Apr ’25
How to detect iPad trackpad touch-down (indirectPointer) to immediately stop coasting animation
Hello, I have a custom 3D object viewer on iOS that lets users spin the model using the touchscreen or a trackpad and supports coasting (momentum spinning). I need to stop the coasting animation as soon as the user touches down, but I can only immediately detect touches on the screen itself - on the trackpad I can't get an immediate notification of the touches. So far I’ve tried: State.began on my UIPanGestureRecognizer. It only fires after a small movement on both touchscreen and trackpad. .possible on the pan gesture; this state never occurs during the gesture cycle. UIApplicationSupportsIndirectInputEvents = YES in Info.plist; it didn’t make touchesBegan fire for indirectPointer touches. Since UITableView (and other UIScrollView subclasses) clearly detect trackpad “touch-down” to cancel scrolling, there must be a way to receive that event. Does anyone know how to catch the initial trackpad contact—before any movement—on an indirect input device? Below is a minimal code snippet demonstrating the issue. On t
Topic: UI Frameworks SubTopic: UIKit Tags:
Replies
2
Boosts
0
Views
115
Activity
Apr ’25
Reply to 关于本地回环地址调节优先级的问题
Sadly, I don’t read Chinese so I’m basing my response on a machine translation of your question. It seems like: You’re on the Mac. You have a product that communicates using TCP/IP over the loopback interface. You’d like to use a packet trace to monitor that communication. If so, you can do that using standard packet tracing techniques. The trick is to specify the loopback interface (lo0) when you start the trace. Here’s how I tested this today: On macOS 15.4, in one Terminal window, I started nc to listen on port 12345 of 127.0.0.1: % nc -l 127.0.0.1 12345 In another window, I told nc to connect to that listener: % nc 127.0.0.1 12345 In a third window, I ran tcpdump like so: % sudo tcpdump -n -i lo0 port 12345 I typed some text into the window from step 2. It generated traffic on the loopback interface which was displayed by tcpdump. Share and Enjoy — Quinn “The Eskimo!” @ Developer Technical Support @ Apple let myEmail = eskimo + 1 + @ + apple.com
Replies
Boosts
Views
Activity
Apr ’25
Reply to Why slower with larger threadgroup memory?
The answer would be usually, no. The basic idea underlying modern GPUs is they try to get around memory latency by having A LOT of parallelism. The cores are usually designed to execute several workgroups at the same time. Some hardware (not sure about Apple Silicon, but I knew several CUDA machines that have this) wants you to have several thread groups AND several simd groups/warps in each threadgroup to reach peak efficiency. My understanding is, you even proved this yourself - you got 250GFlops when asking for 32kB and 400GFlops when asking for 8kB of threadgroup memory. That memory request translates into I want only one threadgroup to run on each core vs I am ok with up to 4 thread groups running on each core. If I were you, I would immediately try with 4kb and 16kb, so with up to 8 and up to 2 thread groups, then check performance.
Replies
Boosts
Views
Activity
Apr ’25