René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"[BUG] Elementwise value assignment silently fails after in-place array downsize using `rows()`","articleBody":"In some scenarios, it seems that, after downsizing an array using an in-place call to `rows()` (ie, overwrite an array with only subset of its current rows), elementwise value assignment silently fails (ie, the element's value does not change and there are no compile or runtime errors).\r\n\r\nDescription\r\n===========\r\n* Did you build ArrayFire yourself or did you use the official installers: **Built myself.**\r\n* Which backend is experiencing this issue? **CUDA.**\r\n* Do you have a workaround? **No.**\r\n* Can the bug be reproduced reliably on your system? **Yes.**\r\n* A clear and concise description of what you expected to happen. **Expected elementwise value assignment to succeed (eg, in example below, `Drows(0,0)` to have value `1234`). Given that that fails, expected a compile or runtime error.**\r\n* Run your executable with AF_TRACE=all and AF_PRINT_ERRORS=1 environment variables set:\r\n```\r\n# AF_TRACE=all AF_PRINT_ERRORS=1 aftest\r\n[platform][1708117940][9998] [ /tmp/arrayfire/src/backend/common/DependencyModule.cpp:102 ] Attempting to load: libforge.so\r\n[platform][1708117940][9998] [ /tmp/arrayfire/src/backend/common/DependencyModule.cpp:107 ] Unable to open forge\r\n[platform][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:497 ] CUDA Driver supports up to CUDA 12.3.0 ArrayFire CUDA Runtime 11.3.0\r\n[platform][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:478 ] CUDA driver version(12.3.0) not part of the CudaToDriverVersion array. Please create an issue or a pull request on the ArrayFire repository to update the CudaToDriverVersion variable with this version of the CUDA runtime.\r\n\r\n[platform][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:566 ] Found 1 CUDA devices\r\n[platform][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:588 ] Found device: NVIDIA RTX A3000 Laptop GPU (sm_86) (5.80 GB | ~12187.5 GFLOPs | 32 SMs)\r\n[platform][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:652 ] AF_CUDA_DEFAULT_DEVICE: \r\n[platform][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:670 ] Default device: 0(NVIDIA RTX A3000 Laptop GPU)\r\n[mem][1708117942][9998] [ /tmp/arrayfire/src/backend/common/DefaultMemoryManager.cpp:127 ] memory[0].max_bytes: 4.8 GB\r\n[mem][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/memory.cpp:155 ] nativeAlloc:    1 KB 0x7fe876800000\r\n[jit][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {14966320269747309860 : loaded from /root/.arrayfire/KER14966320269747309860_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::range\u003cfloat\u003e: Blocks: [1, 1, 1] Threads: [32, 8, 1] Shared Memory: 0\r\nDrows\r\n[5 5 1 1]\r\n[mem][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/memory.cpp:155 ] nativeAlloc:    1 KB 0x7fe876800400\r\n[jit][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {291210446400920389   : loaded from /root/.arrayfire/KER291210446400920389_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::transpose\u003cfloat,false,false\u003e: Blocks: [1, 1, 1] Threads: [32, 8, 1] Shared Memory: 0\r\n    0.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000 \r\n    4.0000     4.0000     4.0000     4.0000     4.0000 \r\n\r\n[jit][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {8447298384643760287  : loaded from /root/.arrayfire/KER8447298384643760287_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::memCopy\u003cfloat\u003e: Blocks: [1, 5, 1] Threads: [32, 1, 1] Shared Memory: 0\r\n[jit][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {8641389271879371835  : loaded from /root/.arrayfire/KER8641389271879371835_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/jit.cpp:512 ] Launching : Dims: [1,1,1,1] Blocks: [1 1 1] Threads: [128 1 1] threads: 128\r\nDrows\r\n[4 5 1 1]\r\n[kernel][1708117942][9998] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::transpose\u003cfloat,false,false\u003e: Blocks: [1, 1, 1] Threads: [32, 8, 1] Shared Memory: 0\r\n    0.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000\r\n```\r\n\r\nReproducible Code and/or Steps\r\n------------------------------\r\nProgram/output (note the `(0,0)` element of the last print):\r\n```\r\n#include \u003carrayfire.h\u003e\r\n\r\nusing namespace af;\r\n\r\nint main(int argc, char **argv)\r\n{\r\n    array Drows = range(dim4(5,5));\r\n    af_print(Drows);\r\n\r\n    Drows = Drows.rows(0,3);\r\n    Drows(0,0) = 1234;\r\n    af_print(Drows);\r\n\r\n    return 0;\r\n}\r\n\r\n\r\n# aftest\r\nDrows\r\n[5 5 1 1]\r\n    0.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000 \r\n    4.0000     4.0000     4.0000     4.0000     4.0000 \r\n\r\nDrows\r\n[4 5 1 1]\r\n    0.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000\r\n```\r\n\r\nInterestingly, initializing the array as a copy of an existing array yields the expected behavior: \r\n```\r\n#include \u003carrayfire.h\u003e\r\n\r\nusing namespace af;\r\n\r\nint main(int argc, char **argv)\r\n{\r\n    array D = range(dim4(5,5));\r\n    array Drows = D;\r\n    af_print(Drows);\r\n\r\n    Drows = Drows.rows(0,3);\r\n    Drows(0,0) = 1234;\r\n    af_print(Drows);\r\n\r\n    return 0;\r\n}\r\n\r\n\r\n# aftest\r\nDrows\r\n[5 5 1 1]\r\n    0.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000 \r\n    4.0000     4.0000     4.0000     4.0000     4.0000 \r\n\r\nDrows\r\n[4 5 1 1]\r\n 1234.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000\r\n```\r\n\r\n**EDIT**: Additionally, here is the full debugging output for this case that succeeds (I note an additional `[mem]` line before the final `af_print`):\r\n```\r\n# AF_TRACE=all AF_PRINT_ERRORS=1 aftest\r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/common/DependencyModule.cpp:102 ] Attempting to load: libforge.so\r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/common/DependencyModule.cpp:107 ] Unable to open forge\r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:497 ] CUDA Driver supports up to CUDA 12.3.0 ArrayFire CUDA Runtime 11.3.0\r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:478 ] CUDA driver version(12.3.0) not part of the CudaToDriverVersion array. Please create an issue or a pull request on the ArrayFire repository to update the CudaToDriverVersion variable with this version of the CUDA runtime.\r\n\r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:566 ] Found 1 CUDA devices\r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:588 ] Found device: NVIDIA RTX A3000 Laptop GPU (sm_86) (5.80 GB | ~12187.5 GFLOPs | 32 SMs)\r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:652 ] AF_CUDA_DEFAULT_DEVICE: \r\n[platform][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/device_manager.cpp:670 ] Default device: 0(NVIDIA RTX A3000 Laptop GPU)\r\nArrayFire v3.9.0 (CUDA, 64-bit Linux, build b59a1ae53)\r\nPlatform: CUDA Runtime 11.3, Driver: 545.23.08\r\n[0] NVIDIA RTX A3000 Laptop GPU, 5938 MB, CUDA Compute 8.6\r\n[mem][1708119537][10870] [ /tmp/arrayfire/src/backend/common/DefaultMemoryManager.cpp:127 ] memory[0].max_bytes: 4.8 GB\r\n[mem][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/memory.cpp:155 ] nativeAlloc:    1 KB 0x7fe1c8800000\r\n[jit][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {14966320269747309860 : loaded from /root/.arrayfire/KER14966320269747309860_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::range\u003cfloat\u003e: Blocks: [1, 1, 1] Threads: [32, 8, 1] Shared Memory: 0\r\nDrows\r\n[5 5 1 1]\r\n[mem][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/memory.cpp:155 ] nativeAlloc:    1 KB 0x7fe1c8800400\r\n[jit][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {291210446400920389   : loaded from /root/.arrayfire/KER291210446400920389_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::transpose\u003cfloat,false,false\u003e: Blocks: [1, 1, 1] Threads: [32, 8, 1] Shared Memory: 0\r\n    0.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000 \r\n    4.0000     4.0000     4.0000     4.0000     4.0000 \r\n\r\n[jit][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {8447298384643760287  : loaded from /root/.arrayfire/KER8447298384643760287_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::memCopy\u003cfloat\u003e: Blocks: [1, 5, 1] Threads: [32, 1, 1] Shared Memory: 0\r\n[jit][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/compile_module.cpp:472 ] {8641389271879371835  : loaded from /root/.arrayfire/KER8641389271879371835_CU_86_AF_39.bin for NVIDIA RTX A3000 Laptop GPU }\r\n[kernel][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/jit.cpp:512 ] Launching : Dims: [1,1,1,1] Blocks: [1 1 1] Threads: [128 1 1] threads: 128\r\nDrows\r\n[4 5 1 1]\r\n[mem][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/memory.cpp:155 ] nativeAlloc:    1 KB 0x7fe1c8800800\r\n[kernel][1708119537][10870] [ /tmp/arrayfire/src/backend/cuda/Kernel.hpp:37 ] Launching arrayfire::cuda::transpose\u003cfloat,false,false\u003e: Blocks: [1, 1, 1] Threads: [32, 8, 1] Shared Memory: 0\r\n 1234.0000     0.0000     0.0000     0.0000     0.0000 \r\n    1.0000     1.0000     1.0000     1.0000     1.0000 \r\n    2.0000     2.0000     2.0000     2.0000     2.0000 \r\n    3.0000     3.0000     3.0000     3.0000     3.0000\r\n```\r\n\r\nSystem Information\r\n------------------\r\nPlease provide the following information:\r\n1. ArrayFire version: **3.9.0**\r\n2. Devices installed on the system **NVIDIA RTX A3000 Laptop GPU**\r\n3. Output from the af::info() function if applicable.\r\n```\r\nArrayFire v3.9.0 (CUDA, 64-bit Linux, build b59a1ae53)\r\nPlatform: CUDA Runtime 11.3, Driver: 545.23.08\r\n[0] NVIDIA RTX A3000 Laptop GPU, 5938 MB, CUDA Compute 8.6\r\n```\r\n4. Output from the following scripts:\r\nLinux output:\r\n```\r\n# bash afbugreport.sh \r\nNo LSB modules are available.\r\nDistributor ID:\tUbuntu\r\nDescription:\tUbuntu 20.04.6 LTS\r\nRelease:\t20.04\r\nCodename:\tfocal\r\nname, memory.total [MiB], driver_version\r\nNVIDIA RTX A3000 Laptop GPU, 6144 MiB, 545.23.08\r\nrocm-smi not found.\r\nclinfo not found.\r\n```\r\n\r\nChecklist\r\n---------\r\n\r\n- [x] Using the latest available ArrayFire release\r\n- [x] GPU drivers are up to date\r\n","author":{"url":"https://github.com/warnellg","@type":"Person","name":"warnellg"},"datePublished":"2024-02-16T21:36:39.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":1},"url":"https://github.com/3534/arrayfire/issues/3534"}