biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Fix CUDA kernel index data type in vllm/csrc/quantization/fused_kernels/layernorm_utils.cuh +10 (#15159)

Browse Source 
Signed-off-by: Lu Fang <lufang@fb.com>
Co-authored-by: Richard Barnes <rbarnes@meta.com>
This commit is contained in:
Lu Fang
2025-03-20 19:01:11 -07:00
committed by GitHub
parent 0cfe7d386d
commit d3ccbd6350
10 changed files with 124 additions and 124 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
csrc/quantization/gptq_allspark/allspark_repack.cu
View File

@@ -13,8 +13,8 @@ __global__ void __launch_bounds__(128)
const uint8_t* B, const FType* B_scale, const FType* B_zero,
uint8_t* B_result, FType* B_scale_result, FType* B_zero_result,
const int K, const int N, const int N_32align) {
const int lane_id = threadIdx.x % 32;
const int warp_id = threadIdx.x / 32;
const auto lane_id = threadIdx.x % 32;
const auto warp_id = threadIdx.x / 32;
if (blockIdx.x != gridDim.x - 1) {
// Load B
@@ -50,7 +50,7 @@ __global__ void __launch_bounds__(128)
}
// Store B
const int dst_row_base_idx = blockIdx.y * (128 / 4) + (lane_id / 8) * 8;
const auto dst_row_base_idx = blockIdx.y * (128 / 4) + (lane_id / 8) * 8;
const int dst_col_idx =
blockIdx.x * (64 * 4) + warp_id * 64 + (lane_id % 8) * 8;
for (int i = 0; i < 8; ++i) {
@@ -65,7 +65,7 @@ __global__ void __launch_bounds__(128)
} else {
// Load B_scale and B_zero
FType b_scale_reg, b_zero_reg;
int src_offset = blockIdx.y * 128 + threadIdx.x;
auto src_offset = blockIdx.y * 128 + threadIdx.x;
ldg16_cg_0(b_scale_reg, B_scale + src_offset, src_offset < N);
if (B_zero != nullptr)
ldg16_cg_0(b_zero_reg, B_zero + src_offset, src_offset < N);