vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix the unalign size

Browse source
This commit is contained in:
luoyu-intel 2024-07-15 15:15:41 +08:00
parent d9b89eb308
commit 127d62fa06
1 changed files with 23 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

10
ggml/src/ggml-sycl/dmmv.cpp
View file

@ -112,6 +112,7 @@ static void dequantize_mul_mat_vec_q4_0(const void * __restrict__ vx, const dflo
int constexpr Unroll = 2;
const int iqs = tid; // x quant index
int ncols_pad = ncols - ncols % (WarpK * Unroll);
int ncols_pad1 = ncols - ncols % (WarpK * 1);
int i = 0;
for (; i < ncols_pad; i += WarpK * Unroll) {
#pragma unroll
@ -140,14 +141,16 @@ static void dequantize_mul_mat_vec_q4_0(const void * __restrict__ vx, const dflo
}
}
}
for (; i < ncols_pad; i += WarpK * 1) {
if (i + WarpK <= ncols_pad1)
{
for (; i < ncols_pad1; i += WarpK * 1) {
#pragma unroll
for (int iu = 0; iu < 1; iu++)
{
const int iybs = i + tid * QK4_0 + iu * WarpK; // y block start index
const int ib = (row * ncols + i) / QK4_0 + tid + iu * WARP_SIZE; // x block index
const dfloat d = *(sycl::half *)((char *)x + ncols * nrows / 2 + ib * 2);
sycl::vec<uint8_t, QK4_0 / 2> tmp_qs = *(sycl::vec<uint8_t, QK4_0 / 2>*)((char *)x + ib * QK4_0 / 2);
const dfloat d = *(sycl::half*)((char*)x + ncols * nrows / 2 + ib * 2);
sycl::vec<uint8_t, QK4_0 / 2> tmp_qs = *(sycl::vec<uint8_t, QK4_0 / 2>*)((char*)x + ib * QK4_0 / 2);
int constexpr KUnroll = 1;
#pragma unroll
for (int ir = 0; ir < QK4_0 / 2; ir += KUnroll)
@ -167,6 +170,7 @@ static void dequantize_mul_mat_vec_q4_0(const void * __restrict__ vx, const dflo
}
}
}
}
#if 1
for (; i < ncols; i += QK4_0) {
const int iybs = i; // y block start index