On x86_64-linux, I managed to Ion-compile both ort-wasm-simd-threaded.wasm and ort-wasm-simd-threaded.jsep.wasm to completion. The latter is larger and took about 10 minutes and around 4GB of memory. It contains some very large functions, the largest of which is 1233871 wasm bytecodes, producing 388722 LIRs in 132855 basic blocks. This takes the allocator a long time to process (several minutes), but it doesn't loop. I imagine it will take about twice as long on an ARM64 platform, since ARM64 has about twice as many integer registers to search through.
Bug 1916442 Comment 4 Edit History
Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.
On x86_64-linux, I managed to Ion-compile both ort-wasm-simd-threaded.wasm and ort-wasm-simd-threaded.jsep.wasm to completion. The latter is larger and took about 10 minutes and around 4GB of memory. It contains some very large functions, the largest of which is 1233871 wasm bytecode bytes, producing 388722 LIRs in 132855 basic blocks. This takes the allocator a long time to process (several minutes), but it doesn't loop. I imagine it will take about twice as long on an ARM64 platform, since ARM64 has about twice as many integer registers to search through.