[Build] Add support for Ascend950 chip (#7151)

### What this PR does / why we need it?
This PR adds support for the Ascend950 chip. This includes:
- Updating build scripts (`CMakeLists.txt` and `setup.py`) to recognize
the Ascend950 chip and set appropriate compilation flags.
- Disabling a set of custom operators that are not yet supported on the
Ascend950 hardware target.
- Performing a codebase-wide refactoring of `pipe_barrier()` calls to
the namespaced `AscendC::PipeBarrier<>()` for improved code consistency
and adherence to the latest API standards.

Ascend950DT e2e passed (Qwen3-32B-MXFP8) and CI passed
- vLLM version: v0.16.0
- vLLM main:
4034c3d32e
---------
Signed-off-by: linfeng-yuan <1102311262@qq.com>
This commit is contained in:
linfeng-yuan
2026-03-12 10:25:51 +08:00
committed by GitHub
parent da01a74009
commit 5f3826b093
24 changed files with 246 additions and 227 deletions

View File

@@ -267,7 +267,10 @@ def enable_custom_op():
# There are some customed operators which aren't implemented
# with batch invariant in vllm-ascend, we need to disable them.
if vllm_is_batch_invariant():
# FIXME(linfeng): Currently custom op compilation and execution are partially available
# in ASCEND950 chip, we temporarily disable all custom ops. Please refer to
# https://github.com/vllm-project/vllm-ascend/issues/7157 for latest update about custom op.
if vllm_is_batch_invariant() or get_ascend_device_type() == AscendDeviceType.A5:
_CUSTOM_OP_ENABLED = False
return _CUSTOM_OP_ENABLED