A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 23:15 UTC
[Bugfix] Fix CUDA version check for mma warning suppression (#5642)
Tyler Michael Smith committed
Jun 18, 2024 at 19:48 UTC
b23ce9203235488e080434108d3504d54b24e867
1 file changed
+2
-1
csrc/quantization/marlin/sparse/common/mma.h
+2
-1
@@ -17,6 +17,7 @@
17
18
#pragma once
19
#include "base.h"
20
+#include <cudaTypedefs.h>
21
22
namespace marlin_24 {
23
@@ -26,7 +27,7 @@ namespace marlin_24 {
27
// | Advisory: Modifier ‘.sp::ordered_metadata’ should be used on instruction
28
// | ‘mma’ instead of modifier ‘.sp’ as it is expected to have substantially
29
// | reduced performance on some future architectures
29
-#if defined CUDA_VERSION && CUDA_VERSION >= 12500
30
+#if defined CUDA_VERSION && CUDA_VERSION >= 12050
31
#define MMA_SP_INST \
32
"mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.f32.f16.f16.f32 "
33
#else