Skip to content

[Issue]: MIOpenDriver crashes under specific conditions #3558

@menglcai

Description

@menglcai

Problem Description

Test on the Windows platform, MIOpen (rocm-libraries 7daf2d028, 12/29) crashes when running the following commands. However, on MI250 + MIOpen (version: 3.5.1) , all commands can successfully find the kernels.

MIOpenDriver convbfp16 -n 1 -c 512 --in_d 65 -H 72 -W 128 -k 512 --fil_d 5 -y 5 -x 5 --pad_d 0 -p 2 -q 2 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 512 -F 1 -t 1

MIOpenDriver convbfp16 -n 1 -c 512 --in_d 45 -H 88 -W 160 -k 512 --fil_d 5 -y 5 -x 5 --pad_d 0 -p 2 -q 2 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 512 -F 1 -t 1

MIOpenDriver convbfp16 -n 1 -c 64 --in_d 123 -H 352 -W 640 -k 64 --fil_d 3 -y 3 -x 3 --pad_d 0 -p 1 -q 1 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 1 -F 1 -t 1

MIOpenDriver convbfp16 -n 1 -c 64 --in_d 123 -H 352 -W 640 -k 32 --fil_d 3 -y 3 -x 3 --pad_d 0 -p 1 -q 1 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 1 -F 1 -t 1

MIOpenDriver convfp16 -n 1 -c 64 --in_d 123 -H 352 -W 640 -k 32 --fil_d 3 -y 3 -x 3 --pad_d 0 -p 1 -q 1 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 1 -F 1 -t 1
Image

Operating System

Windows 25H2 (26200.6584)

CPU

AMD RYZEN AI MAX+ 395 w

GPU

Radeon 8060S

ROCm Version

TheRock (rocm-sdk-libraries-gfx1151==7.11.0a20251229)

ROCm Component

No response

Steps to Reproduce

Run one of following commands and check crash report from EventVierwer.

MIOpenDriver convbfp16 -n 1 -c 512 --in_d 65 -H 72 -W 128 -k 512 --fil_d 5 -y 5 -x 5 --pad_d 0 -p 2 -q 2 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 512 -F 1 -t 1

MIOpenDriver convbfp16 -n 1 -c 512 --in_d 45 -H 88 -W 160 -k 512 --fil_d 5 -y 5 -x 5 --pad_d 0 -p 2 -q 2 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 512 -F 1 -t 1

MIOpenDriver convbfp16 -n 1 -c 64 --in_d 123 -H 352 -W 640 -k 64 --fil_d 3 -y 3 -x 3 --pad_d 0 -p 1 -q 1 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 1 -F 1 -t 1

MIOpenDriver convbfp16 -n 1 -c 64 --in_d 123 -H 352 -W 640 -k 32 --fil_d 3 -y 3 -x 3 --pad_d 0 -p 1 -q 1 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 1 -F 1 -t 1

MIOpenDriver convfp16 -n 1 -c 64 --in_d 123 -H 352 -W 640 -k 32 --fil_d 3 -y 3 -x 3 --pad_d 0 -p 1 -q 1 --conv_stride_d 1 -u 1 -v 1 --dilation_d 1 -l 1 -j 1 --spatial_dim 3 -m conv -g 1 -F 1 -t 1

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions