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 5, 12:15 UTC
[Bugfix] Correct mistake in chained comparison in static assert logic (#38699)
Signed-off-by: Kyle Mylonakis <kyle@protopia.ai>
Kyle Mylonakis committed
Apr 7, 2026 at 11:24 UTC
7b9de7c892b3b5d50942d245400b740767a40bd8
3 files changed
+3
-3
csrc/cpu/cpu_attn_vec.hpp
+1
-1
@@ -53,7 +53,7 @@ class TileGemm82 {
53
const int64_t ldb, const int64_t ldc,
54
const int32_t block_size, const int32_t dynamic_k_size,
55
const bool accum_c) {
56
- static_assert(0 < M <= 8);
56
+ static_assert(0 < M && M <= 8);
57
using load_vec_t = typename VecTypeTrait<kv_cache_t>::vec_t;
58
59
kv_cache_t* __restrict__ curr_b_0 = b_tile;
csrc/cpu/cpu_attn_vec16.hpp
+1
-1
@@ -68,7 +68,7 @@ class TileGemm161 {
68
const int64_t ldb, const int64_t ldc,
69
const int32_t block_size, const int32_t dynamic_k_size,
70
const bool accum_c) {
71
- static_assert(0 < M <= 16);
71
+ static_assert(0 < M && M <= 16);
72
using load_vec_t = typename VecTypeTrait<kv_cache_t>::vec_t;
73
74
kv_cache_t* __restrict__ curr_b_0 = b_tile;
csrc/cpu/micro_gemm/cpu_micro_gemm_vec.hpp
+1
-1
@@ -39,7 +39,7 @@ class TileGemm82 {
39
40
template <int32_t M>
41
static void gemm_micro(DEFINE_CPU_MICRO_GEMM_PARAMS) {
42
- static_assert(0 < M <= 8);
42
+ static_assert(0 < M && M <= 8);
43
using load_vec_t = typename cpu_utils::VecTypeTrait<scalar_t>::vec_t;
44
45
scalar_t* __restrict__ curr_b_0 = b_ptr;