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

SYCL softmax: Initialize nreduce as size_t

Browse source
This commit is contained in:
Akarshan Biswas 2024-12-10 20:08:19 +05:30
parent fe5afd4a2d
commit 9129362c6f
No known key found for this signature in database
GPG key ID: 52A578A14B32134D
1 changed files with 5 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

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

@ -16,7 +16,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
const int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
const int nthreads = block_size;
const int nwarps = nthreads / WARP_SIZE;
int nreduce = nwarps / WARP_SIZE;
size_t nreduce = nwarps / WARP_SIZE;
float slope = 1.0f;
// ALiBi
@ -53,7 +53,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
if (block_size > WARP_SIZE) {
if (warp_id == 0) {
buf[lane_id] = -INFINITY;
for (size_t i = 1; i < (size_t) nreduce; i += 1) {
for (size_t i = 1; i < nreduce; i += 1) {
buf[lane_id + i * WARP_SIZE] = -INFINITY;
}
}
@ -64,7 +64,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
}
item_ct1.barrier(sycl::access::fence_space::local_space);
max_val = buf[lane_id];
for (size_t i = 1; i < (size_t) nreduce; i += 1) {
for (size_t i = 1; i < nreduce; i += 1) {
max_val = std::max(max_val, buf[lane_id + i * WARP_SIZE]);
}
max_val = warp_reduce_max(max_val, item_ct1);
@ -89,7 +89,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
item_ct1.barrier(sycl::access::fence_space::local_space);
if (warp_id == 0) {
buf[lane_id] = 0.f;
for (size_t i = 1; i < (size_t) nreduce; i += 1) {
for (size_t i = 1; i < nreduce; i += 1) {
buf[lane_id + i * WARP_SIZE] = 0.f;
}
}
@ -101,7 +101,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
item_ct1.barrier(sycl::access::fence_space::local_space);
tmp = buf[lane_id];
for (size_t i = 1; i < (size_t) nreduce; i += 1) {
for (size_t i = 1; i < nreduce; i += 1) {
tmp += buf[lane_id + i * WARP_SIZE];
}
tmp = warp_reduce_sum(tmp, item_ct1);