Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

SYCL: Reduce most of the compiler warnings #10748

Merged
merged 24 commits into from
Dec 13, 2024
Merged

SYCL: Reduce most of the compiler warnings #10748

Show file tree
Hide file tree
Changes from 9 commits
Commits
Show all changes
24 commits
Select commit Hold shift + click to select a range
3930184
Try to reduce some unused and typecast warnings
qnixsynapse Dec 9, 2024
a708dfc
Reduce compiler warnings step 2
qnixsynapse Dec 10, 2024
fb2e66e
add a newline at the end of the file
qnixsynapse Dec 10, 2024
32164aa
Initialize nreduce as size_t
qnixsynapse Dec 10, 2024
71d84a5
[SYCL] Remove pragma directives from mmq.cpp
qnixsynapse Dec 10, 2024
fe5afd4
SYCL: mmq add condition to prevent blocks_per_tile_x_row variable fro…
qnixsynapse Dec 10, 2024
9129362
SYCL softmax: Initialize nreduce as size_t
qnixsynapse Dec 10, 2024
4b5470f
ggml-sycl.cpp: fix some trailing whitespaces
qnixsynapse Dec 11, 2024
7dda9aa
SYCL: remove the unused variables instead of commenting it out
qnixsynapse Dec 11, 2024
cc7cd62
SYCL poo2d kernel: set NAN for invalid pooling op
qnixsynapse Dec 11, 2024
5a766c1
Merge branch 'master' into refactor
qnixsynapse Dec 11, 2024
274842d
SYCL gemm.hpp: remove pragma directives
qnixsynapse Dec 11, 2024
b0e27ad
SYCL gemm.hpp: use const cast to properly support dnnl::memory
qnixsynapse Dec 11, 2024
cb0daca
SYCL: wkv6 remove a comment
qnixsynapse Dec 11, 2024
8f123ae
SYCL: clean comments step 2
qnixsynapse Dec 11, 2024
39b4c47
SYCL: clean comments and variables step 3
qnixsynapse Dec 11, 2024
8dfac46
SYCL: Use GGML_UNUSED for unused variables
qnixsynapse Dec 12, 2024
90fe556
SYCL: remove extra empty lines and a comment
qnixsynapse Dec 12, 2024
46bcfe4
Remove TODO
qnixsynapse Dec 12, 2024
ffd7c1d
cleanup spaces
abhilash1910 Dec 12, 2024
ba661a4
add a stdout for unsupported op
abhilash1910 Dec 12, 2024
524acb4
use sycl printf over fprintf
abhilash1910 Dec 12, 2024
b828f4a
remove prints for CI
abhilash1910 Dec 12, 2024
6b0848c
SYCL ggml-sycl: pool2D use sycl::nan and remove if-else block
qnixsynapse Dec 12, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 2 additions & 11 deletions ggml/src/ggml-sycl/common.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,7 @@
//

#include "common.hpp"
#include "ggml-impl.h"

int get_current_device_id() {
return dpct::dev_mgr::instance().current_device_id();
Expand All @@ -28,11 +29,7 @@ void* ggml_sycl_host_malloc(size_t size) try {

if (err != 0) {
// clear the error
fprintf(
stderr,
"WARNING: failed to allocate %.2f MB of pinned memory: %s\n",
size / 1024.0 / 1024.0,
"syclGetErrorString is not supported");
GGML_LOG_ERROR("WARNING: failed to allocate %.2f MB of pinned memory: %s\n", size / 1024.0 / 1024.0, "syclGetErrorString is not supported");
return nullptr;
}

Expand Down Expand Up @@ -66,18 +63,12 @@ int64_t downsample_sycl_global_range(int64_t accumulate_block_num, int64_t block
void ggml_sycl_op_flatten(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
const ggml_tensor *src1, ggml_tensor *dst,
const ggml_sycl_op_flatten_t op) try {
const int64_t nrows0 = ggml_nrows(src0);

const bool use_src1 = src1 != nullptr;
const int64_t nrows1 = use_src1 ? ggml_nrows(src1) : 1;

GGML_ASSERT(!use_src1 || src1->backend != GGML_BACKEND_TYPE_GPU_SPLIT);
GGML_ASSERT( dst->backend != GGML_BACKEND_TYPE_GPU_SPLIT);

ggml_tensor_extra_gpu * src0_extra = (ggml_tensor_extra_gpu *) src0->extra;
ggml_tensor_extra_gpu * src1_extra = use_src1 ? (ggml_tensor_extra_gpu *) src1->extra : nullptr;
ggml_tensor_extra_gpu * dst_extra = (ggml_tensor_extra_gpu *) dst->extra;

// dd = data device
float * src0_ddf = (float *) src0->data;
float * src1_ddf = use_src1 ? (float *) src1->data : nullptr;
Expand Down
1 change: 1 addition & 0 deletions ggml/src/ggml-sycl/common.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -626,6 +626,7 @@ struct bin_bcast_sycl {
});
}
}
(void) ctx;
}
};

Expand Down
4 changes: 2 additions & 2 deletions ggml/src/ggml-sycl/concat.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,7 @@ static void concat_f32_dim1(const float *x, const float *y, float *dst,
// operation
int offset_dst = nidx + item_ct1.get_group(1) * ne0 +
item_ct1.get_group(0) * ne0 * item_ct1.get_group_range(1);
if (item_ct1.get_group(1) < ne01) { // src0
if (item_ct1.get_group(1) < (size_t) ne01) { // src0
int offset_src =
nidx + item_ct1.get_group(1) * ne0 + item_ct1.get_group(0) * ne0 * ne01;
dst[offset_dst] = x[offset_src];
Expand All @@ -70,7 +70,7 @@ static void concat_f32_dim2(const float *x, const float *y, float *dst,
// operation
int offset_dst = nidx + item_ct1.get_group(1) * ne0 +
item_ct1.get_group(0) * ne0 * item_ct1.get_group_range(1);
if (item_ct1.get_group(0) < ne02) { // src0
if (item_ct1.get_group(0) < (size_t) ne02) { // src0
int offset_src = nidx + item_ct1.get_group(1) * ne0 +
item_ct1.get_group(0) * ne0 * item_ct1.get_group_range(1);
dst[offset_dst] = x[offset_src];
Expand Down
2 changes: 1 addition & 1 deletion ggml/src/ggml-sycl/convert.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -424,7 +424,7 @@ static void convert_unary(const void * __restrict__ vx, dst_t * __restrict__ y,
const int64_t global_id = item_ct1.get_local_id(2) + work_group_size * item_ct1.get_group(2);

// make each work-item deal with more elements since sycl global range can not exceed max int
const src_t * x = (src_t *) vx;
const src_t * x = (const src_t *) vx;
for (int64_t i = global_id; i < k; i += work_group_size * item_ct1.get_group_range(2)) {
y[i] = x[i];
}
Expand Down
2 changes: 1 addition & 1 deletion ggml/src/ggml-sycl/dpct/helper.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1237,7 +1237,7 @@ namespace dpct

std::map<byte_t *, allocation>::iterator get_map_iterator(const void *ptr)
{
auto it = m_map.upper_bound((byte_t *)ptr);
auto it = m_map.upper_bound(const_cast<byte_t *>(reinterpret_cast<const byte_t *>(ptr)));
if (it == m_map.end())
{
// Not a virtual pointer.
Expand Down
25 changes: 22 additions & 3 deletions ggml/src/ggml-sycl/element_wise.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -237,7 +237,7 @@ void upscale_f32(const float *x, float *dst, const int nb00, const int nb01,
int i02 = i12 / sf2;
int i03 = i13 / sf3;

dst[index] = *(float *)((char *)x + i03 * nb03 + i02 * nb02 + i01 * nb01 + i00 * nb00);
dst[index] = *(const float *)((const char *)x + i03 * nb03 + i02 * nb02 + i01 * nb01 + i00 * nb00);
}

void pad_f32(const float *x, float *dst, const int ne0, const int ne00, const int ne01, const int ne02,
Expand All @@ -251,8 +251,7 @@ void pad_f32(const float *x, float *dst, const int ne0, const int ne00, const i
// operation
int offset_dst = nidx + item_ct1.get_group(1) * ne0 +
item_ct1.get_group(0) * ne0 * item_ct1.get_group_range(1);
if (nidx < ne00 && item_ct1.get_group(1) < ne01 &&
item_ct1.get_group(0) < ne02) {
if (nidx < ne00 && item_ct1.get_group(1) < (size_t) ne01 && item_ct1.get_group(0) < (size_t) ne02) {
int offset_src = nidx + item_ct1.get_group(1) * ne00 +
item_ct1.get_group(0) * ne00 * ne01;
dst[offset_dst] = x[offset_src];
Expand Down Expand Up @@ -523,6 +522,7 @@ inline void ggml_sycl_op_silu(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_gelu(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
Expand All @@ -538,6 +538,7 @@ inline void ggml_sycl_op_gelu(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}
inline void ggml_sycl_op_gelu_quick(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
const ggml_tensor *src1, ggml_tensor *dst,
Expand All @@ -553,6 +554,7 @@ inline void ggml_sycl_op_gelu_quick(ggml_backend_sycl_context & ctx, const ggml_
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_tanh(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
Expand All @@ -567,6 +569,7 @@ inline void ggml_sycl_op_tanh(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_relu(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
Expand All @@ -582,6 +585,7 @@ inline void ggml_sycl_op_relu(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_hardsigmoid(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -598,6 +602,7 @@ inline void ggml_sycl_op_hardsigmoid(ggml_backend_sycl_context & ctx, const ggml
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_hardswish(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -613,6 +618,7 @@ inline void ggml_sycl_op_hardswish(ggml_backend_sycl_context & ctx, const ggml_t
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_exp(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -628,6 +634,7 @@ inline void ggml_sycl_op_exp(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_log(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -643,6 +650,7 @@ inline void ggml_sycl_op_log(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_sigmoid(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -658,6 +666,7 @@ inline void ggml_sycl_op_sigmoid(ggml_backend_sycl_context & ctx, const ggml_ten
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_sqrt(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -673,6 +682,7 @@ inline void ggml_sycl_op_sqrt(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_sin(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -688,6 +698,7 @@ inline void ggml_sycl_op_sin(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_cos(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -703,6 +714,7 @@ inline void ggml_sycl_op_cos(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_step(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -718,6 +730,7 @@ inline void ggml_sycl_op_step(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_neg(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -733,6 +746,7 @@ inline void ggml_sycl_op_neg(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_leaky_relu(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -752,6 +766,7 @@ inline void ggml_sycl_op_leaky_relu(ggml_backend_sycl_context & ctx, const ggml_
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_sqr(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
Expand All @@ -767,6 +782,7 @@ inline void ggml_sycl_op_sqr(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_upscale(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
Expand All @@ -790,6 +806,7 @@ inline void ggml_sycl_op_upscale(ggml_backend_sycl_context & ctx, const ggml_ten
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_pad(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
Expand All @@ -808,6 +825,7 @@ inline void ggml_sycl_op_pad(ggml_backend_sycl_context & ctx, const ggml_tensor
(void) src1;
(void) dst;
(void) src1_dd;
(void) ctx;
}

inline void ggml_sycl_op_acc(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
Expand All @@ -828,6 +846,7 @@ inline void ggml_sycl_op_acc(ggml_backend_sycl_context & ctx, const ggml_tensor
acc_f32_sycl(src0_dd, src1_dd, dst_dd, ggml_nelements(dst), src1->ne[0], src1->ne[1], src1->ne[2], nb1, nb2, offset, main_stream);

(void) dst;
(void) ctx;
}

inline void ggml_sycl_op_add(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
Expand Down
5 changes: 5 additions & 0 deletions ggml/src/ggml-sycl/gemm.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,10 @@
#include "dnnl.hpp"
#include "dnnl_sycl.hpp"


// TODO: Remove this when needed
# pragma clang diagnostic push
# pragma clang diagnostic ignored "-Wcast-qual"
qnixsynapse marked this conversation as resolved.
Show resolved Hide resolved
class DnnlGemmWrapper {
public:
using dt = dnnl::memory::data_type;
Expand Down Expand Up @@ -96,6 +100,7 @@ class DnnlGemmWrapper {
}
};

# pragma clang diagnostic pop
#endif

#endif // GGML_SYCL_GEMM_HPP
Loading
Loading