python: ../../../lib/Dialect/TritonGPU/Transforms/AccelerateMatmul.cpp:36: int mlir::triton::gpu::(anonymous namespace)::getMMAVersionSafe(int, DotOp): Assertion `false && "computeCapability not supported"' failed.

[ruralharry]

Describe the bug

我在使用5080的运行Ktransformers的时候，出现这种报错

python: ../../../lib/Dialect/TritonGPU/Transforms/AccelerateMatmul.cpp:36: int mlir::triton::gpu::(anonymous namespace)::getMMAVersionSafe(int, DotOp): Assertion `false && "computeCapability not supported"' failed.

经查验，问题出现在这里
// Get the highest version supported for the hardware and the dot.
static int getMMAVersionSafe(int computeCapability, DotOp op) {
// List supported mma version in order of preference.
SmallVector versionsSupported;
if (computeCapability < 75) {
versionsSupported = {1};
} else if (computeCapability < 90) {
versionsSupported = {2};
} else if (computeCapability < 100) {
versionsSupported = {3, 2};
} else if (computeCapability < 110) {
versionsSupported = {5, 2};
} else {
assert(false && "computeCapability not supported");
}
for (int baseVersion : versionsSupported) {
if (supportMMA(op, baseVersion))
return baseVersion;
if (baseVersion == 3)
op.emitRemark() << "Warning: can't use MMA V3 for the dot op";
}
return 0;
}

能否更改computeCapability < 110
改为computeCapability 支持120以及最新的显卡能力
谢谢！

Environment details

triton版本: 3.2.0
GPU：Nvidia RTX 5080
Python版本：3.11.11
PyTorch版本: 2.7.0.dev20250302+cu128, CUDA可用: True
Ktransformers版本: 0.2.2rc1

[Jokeren]

能否更改computeCapability < 110
改为computeCapability 支持120以及最新的显卡能力
谢谢！

This is not correct, RTX 5080 doesn't support wgmma and tcgen. The correct implementation would be fall back to versionsSupported = {2}

[Jokeren]

Also I would recommend you to translate the issue in English