[Issue]: hipblas status not supported on certain shape pytorch matmul

[ZJLi2013]

Problem Description

    ray_directions = torch.matmul(K, C).squeeze(-1)
    ray_directions = ray_directions.reshape(*B, 3) 

[HIPBLAS DEBUG] Original intrinsics shape: torch.Size([1, 98, 378, 504, 3, 3]), dtype: torch.float32
[HIPBLAS DEBUG] Original coordinates shape: torch.Size([1, 98, 378, 504, 3]), dtype: torch.float32
[HIPBLAS DEBUG] Flattened K shape: torch.Size([18670176, 3, 3]), dtype: torch.float32
[HIPBLAS DEBUG] Flattened C shape: torch.Size([18670176, 3, 1]), dtype: torch.float32
[HIPBLAS DEBUG] Batch dimensions B: torch.Size([1, 98, 378, 504]), Total N: 18670176

    ray_directions = torch.matmul(K, C).squeeze(-1)  # (N, 3) - uses 2D GEMM
                     ^^^^^^^^^^^^^^^^^^
RuntimeError: CUDA error: HIPBLAS_STATUS_NOT_SUPPORTED when calling `HIPBLAS_STATUS_NOT_SUPPORTED`

any suggestions ?

Operating System

ubuntu 22.04

CPU

amd

GPU

mi300

ROCm Version

rocm6.4.3

ROCm Component

No response

Steps to Reproduce

No response

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

No response

Additional Information

No response