cbalint13 commented on PR #18182: URL: https://github.com/apache/tvm/pull/18182#issuecomment-3214574657
@fzi-peccia , Tested with tuning a [resnet18](https://huggingface.co/frgfm/resnet18/blob/main/model.onnx) model Here is the TVM program and results after 5000 trials: [rvv-resnet18-mstune-rpc-2025Aug22.tar.gz](https://github.com/user-attachments/files/21939266/rvv-resnet18-mstune-rpc-2025Aug22.tar.gz) --- ### Tests In a rpc setup, I used the provided ```tvm-rvv-tune.py``` script. * There was trial proposals for tensorization: ``` $ cat workdir/logs/*.log | grep Tensorizing | awk '{print $NF}' | sort -u rvv_float32_multivmul_8_16_m8 rvv_float32_multivmul_8_32_m8 rvv_float32_multivmul_8_4_m8 rvv_float32_multivmul_8_64_m8 rvv_float32_multivmul_8_8_m8 rvv_float32_vmacc_1_16_m8 rvv_float32_vmacc_1_32_m8 rvv_float32_vmacc_1_4_m8 rvv_float32_vmacc_1_64_m8 rvv_float32_vmacc_1_8_m8 rvv_float32_vmul_1_16_m8 rvv_float32_vmul_1_32_m8 rvv_float32_vmul_1_4_m8 rvv_float32_vmul_1_64_m8 rvv_float32_vmul_1_8_m8 ``` * The post analytics of all entries on IR level: ``` $ ./msch-database-tir-parse.py Parsed #5000 records No tensorized schedules found. ``` This needs investigation. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
