xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 05:04:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCm] MI300A compile targets deprecation (#13560)

Browse Source
This commit is contained in:
Gregory Shtrasberg 2025-02-20 02:05:00 -05:00 committed by GitHub
parent 041e294716
commit 0023cd2b9d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 4 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CMakeLists.txt
View File

@ -34,7 +34,7 @@ set(PYTHON_SUPPORTED_VERSIONS "3.9" "3.10" "3.11" "3.12")
set(CUDA_SUPPORTED_ARCHS "7.0;7.2;7.5;8.0;8.6;8.7;8.9;9.0")
# Supported AMD GPU architectures.
set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101")
set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101")
#
# Supported/expected torch versions for CUDA/ROCm.

3
csrc/quantization/fp8/amd/hip_float8_impl.h
View File

@ -1,7 +1,6 @@
#pragma once
#if defined(__HIPCC__) && \
(defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
#if defined(__HIPCC__) && defined(__gfx942__)
#define __HIP__MI300__
#endif

3
csrc/rocm/attention.cu
View File

@ -24,8 +24,7 @@
#include "../attention/dtype_fp8.cuh"
#include "../quantization/fp8/amd/quant_utils.cuh"
#if defined(__HIPCC__) && (defined(__gfx90a__) || defined(__gfx940__) || \
defined(__gfx941__) || defined(__gfx942__))
#if defined(__HIPCC__) && (defined(__gfx90a__) || defined(__gfx942__))
#define __HIP__MI300_MI250__
#endif

3
vllm/attention/backends/rocm_flash_attn.py
View File

@ -25,8 +25,7 @@ logger = init_logger(__name__)
_PARTITION_SIZE_ROCM = 512
_GPU_ARCH = torch.cuda.get_device_properties("cuda").gcnArchName
_ON_NAVI = "gfx1" in _GPU_ARCH
_ON_MI250_MI300 = any(arch in _GPU_ARCH
for arch in ["gfx90a", "gfx940", "gfx941", "gfx942"])
_ON_MI250_MI300 = any(arch in _GPU_ARCH for arch in ["gfx90a", "gfx942"])
class ROCmFlashAttentionBackend(AttentionBackend):