birep birep
birep
/ fix_strix_cwsr.c
Created
October 1, 2026 00:16
Fix for hipStreamCreate OOM on Strix Halo + ROCm
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| /* | |
| * Fix for gfx1151 (Strix Halo) ROCm "out of memory" abort on hipStreamCreate: | |
| * | |
| * Issue: amdkfd enforces ctx_save_restore_area_size >= topo cwsr_size (0x124c000 / 19.2 MB). | |
| * hsakmt hardcodes 0xd4c000 (13.9 MB), causing AMDKFD_IOC_CREATE_QUEUE to return -EINVAL, | |
| * which ROCr surfaces to HIP as hipErrorOutOfMemory. | |
| * | |
| * Solution: Intercept mmap/ioctl to expand CWSR buffer sizes to match kernel requirements. | |
| * | |
| * Build: gcc -shared -fPIC -O3 fix_strix_cwsr.c -o libfix_strix_cwsr.so -ldl |
birep
/ libfix_strix_cwsr.so
Created
September 30, 2026 21:48
Fix for gfx1151 (Strix Halo) ROCm "out of memory" abort on hipStreamCreate
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| /* | |
| * Fix for gfx1151 (Strix Halo) ROCm "out of memory" abort on hipStreamCreate: | |
| * | |
| * Issue: amdkfd enforces ctx_save_restore_area_size >= topo cwsr_size (0x124c000 / 19.2 MB). | |
| * hsakmt hardcodes 0xd4c000 (13.9 MB), causing AMDKFD_IOC_CREATE_QUEUE to return -EINVAL, | |
| * which ROCr surfaces to HIP as hipErrorOutOfMemory. | |
| * | |
| * Solution: Intercept mmap/ioctl to expand CWSR buffer sizes to match kernel requirements. | |
| * | |
| * Build: gcc -shared -fPIC -O3 fix_strix_cwsr.c -o libfix_strix_cwsr.so -ldl |