SYCL: Reduce most of the compiler warnings (#10748)
* Try to reduce some unused and typecast warnings * Reduce compiler warnings step 2 * add a newline at the end of the file * Initialize nreduce as size_t * [SYCL] Remove pragma directives from mmq.cpp * SYCL: mmq add condition to prevent blocks_per_tile_x_row variable from becoming 0 * SYCL softmax: Initialize nreduce as size_t * ggml-sycl.cpp: fix some trailing whitespaces * SYCL: remove the unused variables instead of commenting it out * SYCL poo2d kernel: set NAN for invalid pooling op * SYCL gemm.hpp: remove pragma directives * SYCL gemm.hpp: use const cast to properly support dnnl::memory * SYCL: wkv6 remove a comment * SYCL: clean comments step 2 * SYCL: clean comments and variables step 3 * SYCL: Use GGML_UNUSED for unused variables * SYCL: remove extra empty lines and a comment * Remove TODO * cleanup spaces * add a stdout for unsupported op * use sycl printf over fprintf * remove prints for CI * SYCL ggml-sycl: pool2D use sycl::nan and remove if-else block --------- Co-authored-by: Abhilash Majumder <30946547+abhilash1910@users.noreply.github.com>
This commit is contained in:
Akarshan Biswas
GitHub
parent
d583cd03f6
commit
83ed24a97b
17 changed files with 205 additions and 187 deletions
|
|
@ -47,7 +47,7 @@ static ggml_sycl_device_info ggml_sycl_init() {
|
|||
|
||||
info.device_count = dpct::dev_mgr::instance().device_count();
|
||||
if (info.device_count == 0) {
|
||||
GGML_LOG_ERROR("%s: failed to initialize " GGML_SYCL_NAME ": %s\n", __func__);
|
||||
GGML_LOG_ERROR("%s: failed to initialize: %s\n", GGML_SYCL_NAME, __func__);
|
||||
return info;
|
||||
}
|
||||
|
||||
|
|
@ -64,7 +64,7 @@ static ggml_sycl_device_info ggml_sycl_init() {
|
|||
#else
|
||||
GGML_LOG_INFO("%s: SYCL_USE_XMX: no\n", __func__);
|
||||
#endif
|
||||
GGML_LOG_INFO("%s: found %d " GGML_SYCL_NAME " devices:\n", __func__, info.device_count);
|
||||
GGML_LOG_INFO("%s: found %d %s devices:\n", __func__, info.device_count, GGML_SYCL_NAME);
|
||||
|
||||
for (int i = 0; i < info.device_count; ++i) {
|
||||
info.devices[i].vmm = 0;
|
||||
|
|
@ -137,7 +137,6 @@ void ggml_backend_sycl_print_sycl_devices() {
|
|||
|
||||
for (int id = 0; id < device_count; ++id) {
|
||||
sycl::device device = dpct::dev_mgr::instance().get_device(id);
|
||||
sycl::backend backend = device.get_backend();
|
||||
std::string backend_type = get_device_backend_and_type(device);
|
||||
int type_id = DeviceNums[backend_type]++;
|
||||
std::stringstream device_type;
|
||||
|
|
@ -420,13 +419,11 @@ ggml_backend_sycl_buffer_cpy_tensor(ggml_backend_buffer_t buffer,
|
|||
return true;
|
||||
}
|
||||
return false;
|
||||
GGML_UNUSED(buffer);
|
||||
} catch (const sycl::exception & exc) {
|
||||
std::cerr << exc.what() << "Exception caught at file:" << __FILE__ << ", line:" << __LINE__ << std::endl;
|
||||
std::exit(1);
|
||||
}
|
||||
catch (sycl::exception const &exc) {
|
||||
std::cerr << exc.what() << "Exception caught at file:" << __FILE__
|
||||
<< ", line:" << __LINE__ << std::endl;
|
||||
std::exit(1);
|
||||
}
|
||||
|
||||
|
||||
static void ggml_backend_sycl_buffer_clear(ggml_backend_buffer_t buffer,
|
||||
uint8_t value) try {
|
||||
|
|
@ -1092,10 +1089,7 @@ struct ggml_sycl_pool_leg : public ggml_sycl_pool {
|
|||
ggml_sycl_buffer buffer_pool[MAX_SYCL_BUFFERS] = {};
|
||||
size_t pool_size = 0;
|
||||
|
||||
explicit ggml_sycl_pool_leg(queue_ptr qptr_, int device_) :
|
||||
qptr(qptr_),
|
||||
device(device_) {
|
||||
}
|
||||
explicit ggml_sycl_pool_leg(queue_ptr qptr_, int device_) : device(device_), qptr(qptr_) {}
|
||||
|
||||
~ggml_sycl_pool_leg() {
|
||||
for (int i = 0; i < MAX_SYCL_BUFFERS; ++i) {
|
||||
|
|
@ -1238,7 +1232,7 @@ static void quantize_q8_1(const float * __restrict__ x, void * __restrict__ vy,
|
|||
zeros[i] = 0.f;
|
||||
qzeros[i] = 0;
|
||||
}
|
||||
const TC xi = ix < kx ? *(TC *)&x[iy * kx + ix] : zeros;
|
||||
const TC xi = ix < kx ? *(const TC *)&x[iy * kx + ix] : zeros;
|
||||
float sum = xi[0];
|
||||
float amax = sycl::fabs(xi[0]);
|
||||
#pragma unroll
|
||||
|
|
@ -1799,6 +1793,9 @@ static void pool2d_nchw_kernel(
|
|||
switch (op) {
|
||||
case GGML_OP_POOL_AVG: res = 0; break;
|
||||
case GGML_OP_POOL_MAX: res = -FLT_MAX; break;
|
||||
default:
|
||||
res = (To) sycl::nan(uint32_t(0));
|
||||
break;
|
||||
}
|
||||
|
||||
for (int i = bh; i < eh; i += 1) {
|
||||
|
|
@ -1817,6 +1814,9 @@ static void pool2d_nchw_kernel(
|
|||
switch (op) {
|
||||
case GGML_OP_POOL_AVG: res += (cur / (kh * kw)); break;
|
||||
case GGML_OP_POOL_MAX: res = sycl::max(res, (To)cur); break;
|
||||
default:
|
||||
res = (To) sycl::nan(uint32_t(0));
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -1855,7 +1855,8 @@ static void get_rows_sycl(ggml_backend_sycl_context & ctx, const ggml_tensor *sr
|
|||
s3, nb01, nb02, nb03, s10, s11, s12, item_ct1);
|
||||
});
|
||||
|
||||
(void) dst;
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
template <typename src0_t>
|
||||
|
|
@ -1893,10 +1894,10 @@ static void get_rows_sycl_float(ggml_backend_sycl_context & ctx, const ggml_tens
|
|||
});
|
||||
}
|
||||
|
||||
(void) dst;
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
|
||||
static void quantize_row_q8_1_sycl(const float *x, void *vy, const int kx,
|
||||
const int ky, const int kx_padded,
|
||||
queue_ptr stream) {
|
||||
|
|
@ -2464,8 +2465,8 @@ static void ggml_sycl_op_repeat(ggml_backend_sycl_context & ctx, const ggml_tens
|
|||
|
||||
ggml_sycl_op_bin_bcast<bin_bcast_sycl<op_repeat>>(ctx, dst, src0, dst, nullptr, src0_d, dst_d, main_stream);
|
||||
|
||||
(void) src1;
|
||||
(void) src1_d;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(src1_d);
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -2484,17 +2485,18 @@ inline void ggml_sycl_op_mul_mat_sycl(
|
|||
const int64_t ne00 = src0->ne[0];
|
||||
const int64_t ne10 = src1->ne[0];
|
||||
|
||||
const int64_t ne0 = dst->ne[0];
|
||||
|
||||
const int64_t row_diff = row_high - row_low;
|
||||
|
||||
int id;
|
||||
SYCL_CHECK(
|
||||
CHECK_TRY_ERROR(id = get_current_device_id()));
|
||||
|
||||
#if !GGML_SYCL_DNNL
|
||||
const int64_t ne0 = dst->ne[0];
|
||||
// the main device has a larger memory buffer to hold the results from all GPUs
|
||||
// ldc == nrows of the matrix that cuBLAS writes into
|
||||
int ldc = id == ctx.device ? ne0 : row_diff;
|
||||
#endif
|
||||
|
||||
#ifdef GGML_SYCL_F16
|
||||
bool use_fp16 = true; // TODO(Yu) SYCL capability check
|
||||
|
|
@ -2531,9 +2533,9 @@ inline void ggml_sycl_op_mul_mat_sycl(
|
|||
: src1_as_f16.get();
|
||||
ggml_sycl_pool_alloc<sycl::half> dst_f16(ctx.pool(), row_diff * src1_ncols);
|
||||
|
||||
const sycl::half alpha_f16 = 1.0f;
|
||||
const sycl::half beta_f16 = 0.0f;
|
||||
#if !GGML_SYCL_DNNL
|
||||
const sycl::half alpha_f16 = 1.0f;
|
||||
const sycl::half beta_f16 = 0.0f;
|
||||
SYCL_CHECK(CHECK_TRY_ERROR(dpct::gemm(
|
||||
*stream, oneapi::mkl::transpose::trans,
|
||||
oneapi::mkl::transpose::nontrans, row_diff, src1_ncols, ne10,
|
||||
|
|
@ -2570,9 +2572,9 @@ inline void ggml_sycl_op_mul_mat_sycl(
|
|||
const float * src0_ddf_i = src0->type == GGML_TYPE_F32 ? (const float *) src0_dd_i : src0_ddq_as_f32.get();
|
||||
const float * src1_ddf1_i = src1->type == GGML_TYPE_F32 ? (const float *) src1_ddf_i : src1_ddq_as_f32.get();
|
||||
|
||||
const float alpha = 1.0f;
|
||||
const float beta = 0.0f;
|
||||
#if !GGML_SYCL_DNNL
|
||||
const float alpha = 1.0f;
|
||||
const float beta = 0.0f;
|
||||
# ifdef GGML_SYCL_NVIDIA
|
||||
SYCL_CHECK(CHECK_TRY_ERROR(oneapi::mkl::blas::column_major::gemm(
|
||||
oneapi::mkl::backend_selector<oneapi::mkl::backend::cublas>{ *stream }, oneapi::mkl::transpose::trans,
|
||||
|
|
@ -2590,9 +2592,9 @@ inline void ggml_sycl_op_mul_mat_sycl(
|
|||
src0_ddf_i, DnnlGemmWrapper::to_dt<float>(), dst_dd_i, DnnlGemmWrapper::to_dt<float>());
|
||||
#endif
|
||||
}
|
||||
(void) dst;
|
||||
(void) src1_ddq_i;
|
||||
(void) src1_padded_row_size;
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_ddq_i);
|
||||
GGML_UNUSED(src1_padded_row_size);
|
||||
}
|
||||
catch (sycl::exception const &exc) {
|
||||
std::cerr << exc.what() << "Exception caught at file:" << __FILE__
|
||||
|
|
@ -2638,8 +2640,9 @@ static void ggml_sycl_op_pool2d(ggml_backend_sycl_context & ctx, const ggml_tens
|
|||
item_ct1);
|
||||
});
|
||||
|
||||
(void) src1;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
inline void ggml_sycl_op_sum(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
|
||||
|
|
@ -2654,9 +2657,10 @@ inline void ggml_sycl_op_sum(ggml_backend_sycl_context & ctx, const ggml_tensor
|
|||
|
||||
sum_rows_f32_sycl(src0_dd, dst_dd, ne, 1, main_stream);
|
||||
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
inline void ggml_sycl_op_sum_rows(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
|
||||
|
|
@ -2673,9 +2677,10 @@ inline void ggml_sycl_op_sum_rows(ggml_backend_sycl_context & ctx, const ggml_te
|
|||
|
||||
sum_rows_f32_sycl(src0_dd, dst_dd, ncols, nrows, main_stream);
|
||||
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
inline void ggml_sycl_op_argsort(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
|
||||
|
|
@ -2694,9 +2699,10 @@ inline void ggml_sycl_op_argsort(ggml_backend_sycl_context & ctx, const ggml_ten
|
|||
|
||||
argsort_f32_i32_sycl(src0_dd, (int *)dst_dd, ncols, nrows, order, main_stream);
|
||||
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
inline void ggml_sycl_op_argmax(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
|
||||
|
|
@ -2713,9 +2719,10 @@ inline void ggml_sycl_op_argmax(ggml_backend_sycl_context & ctx, const ggml_tens
|
|||
|
||||
argmax_f32_i32_sycl(src0_dd, (int *)dst_dd, ncols, nrows, main_stream);
|
||||
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
inline void ggml_sycl_op_diag_mask_inf(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
|
||||
|
|
@ -2735,9 +2742,10 @@ inline void ggml_sycl_op_diag_mask_inf(ggml_backend_sycl_context & ctx, const gg
|
|||
|
||||
diag_mask_inf_f32_sycl(src0_dd, dst_dd, ne00, nrows0, ne01, n_past, main_stream);
|
||||
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
inline void ggml_sycl_op_scale(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
|
||||
|
|
@ -2758,9 +2766,10 @@ inline void ggml_sycl_op_scale(ggml_backend_sycl_context & ctx, const ggml_tenso
|
|||
*/
|
||||
SYCL_CHECK(0);
|
||||
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
inline void ggml_sycl_op_clamp(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
|
||||
|
|
@ -2783,9 +2792,10 @@ inline void ggml_sycl_op_clamp(ggml_backend_sycl_context & ctx, const ggml_tenso
|
|||
*/
|
||||
SYCL_CHECK(0);
|
||||
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
(void) src1_dd;
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(src1_dd);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
static void ggml_sycl_set_peer_access(const int n_tokens, int main_device) {
|
||||
|
|
@ -2862,7 +2872,6 @@ static void ggml_sycl_op_mul_mat(ggml_backend_sycl_context & ctx, const ggml_ten
|
|||
|
||||
ggml_tensor_extra_gpu * src0_extra = (ggml_tensor_extra_gpu *) src0->extra;
|
||||
ggml_tensor_extra_gpu * src1_extra = (ggml_tensor_extra_gpu *) src1->extra;
|
||||
ggml_tensor_extra_gpu * dst_extra = (ggml_tensor_extra_gpu *) dst->extra;
|
||||
|
||||
const bool src0_is_contiguous = ggml_is_contiguous(src0);
|
||||
const bool src1_is_contiguous = ggml_is_contiguous(src1);
|
||||
|
|
@ -3289,7 +3298,6 @@ static void ggml_sycl_mul_mat_batched_sycl(ggml_backend_sycl_context & ctx,
|
|||
|
||||
GGML_TENSOR_BINARY_OP_LOCALS
|
||||
|
||||
const int64_t ne_dst = ggml_nelements(dst);
|
||||
|
||||
SYCL_CHECK(ggml_sycl_set_device(ctx.device));
|
||||
queue_ptr main_stream = ctx.stream();;
|
||||
|
|
@ -3397,6 +3405,7 @@ catch (sycl::exception const &exc) {
|
|||
|
||||
inline bool ggml_sycl_supports_mmq(enum ggml_type type) {
|
||||
// TODO: accuracy issues in MMQ
|
||||
GGML_UNUSED(type);
|
||||
return false;
|
||||
}
|
||||
|
||||
|
|
@ -3772,7 +3781,7 @@ static void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor *sr
|
|||
GGML_ABORT("fatal error");
|
||||
}
|
||||
|
||||
(void) dst;
|
||||
GGML_UNUSED(dst);
|
||||
}
|
||||
catch (sycl::exception const &exc) {
|
||||
std::cerr << exc.what() << "Exception caught at file:" << __FILE__
|
||||
|
|
@ -3783,7 +3792,7 @@ catch (sycl::exception const &exc) {
|
|||
static void ggml_sycl_dup(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
|
||||
// TODO: why do we pass dst as src1 here?
|
||||
ggml_sycl_cpy(ctx, src0, dst, nullptr);
|
||||
(void) src1;
|
||||
GGML_UNUSED(src1);
|
||||
}
|
||||
|
||||
static void ggml_sycl_diag_mask_inf(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
|
||||
|
|
@ -3828,13 +3837,16 @@ static void ggml_sycl_argmax(ggml_backend_sycl_context & ctx, const ggml_tensor
|
|||
}
|
||||
|
||||
static void ggml_sycl_nop(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
|
||||
(void) src0;
|
||||
(void) src1;
|
||||
(void) dst;
|
||||
GGML_UNUSED(src0);
|
||||
GGML_UNUSED(src1);
|
||||
GGML_UNUSED(dst);
|
||||
GGML_UNUSED(ctx);
|
||||
}
|
||||
|
||||
void ggml_sycl_set_main_device(const int main_device) try {
|
||||
if (dpct::get_current_device_id() == main_device) return;
|
||||
if (dpct::get_current_device_id() == static_cast<unsigned int> (main_device)) {
|
||||
return;
|
||||
}
|
||||
check_allow_gpu_index(main_device);
|
||||
dpct::select_device(main_device);
|
||||
|
||||
|
|
@ -4202,6 +4214,7 @@ try
|
|||
{
|
||||
ggml_backend_sycl_context *sycl_ctx =
|
||||
(ggml_backend_sycl_context *)backend->context;
|
||||
|
||||
sycl::event *sycl_event = static_cast<sycl::event *>(event->context);
|
||||
|
||||
const queue_ptr &stream = sycl_ctx->stream(sycl_ctx->device, 0);
|
||||
|
|
@ -4216,7 +4229,7 @@ catch (sycl::exception const &exc)
|
|||
}
|
||||
|
||||
static void ggml_backend_sycl_event_wait(ggml_backend_t backend, ggml_backend_event_t event) try {
|
||||
ggml_backend_sycl_context* sycl_ctx = static_cast<ggml_backend_sycl_context*>(backend->context);
|
||||
|
||||
sycl::event* sycl_event = static_cast<sycl::event*>(event->context);
|
||||
|
||||
if (ggml_backend_is_sycl(backend)) {
|
||||
|
|
@ -4624,6 +4637,7 @@ static void *ggml_backend_sycl_reg_get_proc_address(ggml_backend_reg_t reg, cons
|
|||
// SYCL doesn't support registering host memory, left here for reference
|
||||
// "ggml_backend_register_host_buffer"
|
||||
// "ggml_backend_unregister_host_buffer"
|
||||
GGML_UNUSED(name);
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue