compilation failure for simple f32 matmul using the cuda backend

[ziereis]

What happened?

compiling this example:

func.func @matmul(%arg0: tensor<4096x4096xf32>, %arg1: tensor<4096x4096xf32>) -> tensor<4096x4096xf32> {
    %0 = tensor.empty() : tensor<4096x4096xf32>
    %cst = arith.constant 0.000000e+00 : f32
    %1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<4096x4096xf32>) -> tensor<4096x4096xf32>
    %2 = linalg.matmul ins(%arg0, %arg1 : tensor<4096x4096xf32>, tensor<4096x4096xf32>) outs(%1 : tensor<4096x4096xf32>) -> tensor<4096x4096xf32>
    return %2 : tensor<4096x4096xf32>
}

fails with:

example.mlir:5:10: error: 'func.func' op uses 229376 bytes of shared memory; exceeded the limit of 166912 bytes

Steps to reproduce your issue

iree-compile --iree-hal-target-device=cuda --iree-cuda-target=sm_86

What component(s) does this issue relate to?

Compiler

Version information

f27feff

Additional context

No response