happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

refactor: enhance kernel argument setting with specialized templates

This commit is contained in:
chraac 2026-01-01 23:08:21 +08:00
parent 522ef487a1
commit 1fae16787e
1 changed files with 47 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

80
ggml/src/ggml-opencl/ggml-opencl.cpp
View File

@ -53,15 +53,6 @@
bool ggml_cl_compute_forward(ggml_backend_t backend, struct ggml_tensor * tensor);
template <typename... _TArgs> static inline void set_kernel_args(cl_kernel kernel, _TArgs... args) {
size_t index = 0;
(
[&] {
CL_CHECK(clSetKernelArg(kernel, index++, sizeof(args), &args));
}(),
...);
}
// See https://gmplib.org/~tege/divcnst-pldi94.pdf figure 4.1.
// Precompute mp (m' in the paper) and L such that division
// can be computed using a multiply (high 32b of 64b result)
@ -2687,6 +2678,49 @@ struct ggml_tensor_extra_cl {
}
};
template <typename _TData> struct ocl_kernel_arg_setter {};
template <> struct ocl_kernel_arg_setter<int> {
static size_t set_arg(cl_kernel kernel, size_t index, int arg) {
CL_CHECK(clSetKernelArg(kernel, index, sizeof(arg), &arg));
return index + 1;
}
};
template <> struct ocl_kernel_arg_setter<cl_ulong> {
static size_t set_arg(cl_kernel kernel, size_t index, cl_ulong arg) {
CL_CHECK(clSetKernelArg(kernel, index, sizeof(arg), &arg));
return index + 1;
}
};
template <> struct ocl_kernel_arg_setter<const ggml_tensor *> {
static size_t set_arg(cl_kernel kernel, size_t index, const ggml_tensor * t) {
ggml_tensor_extra_cl * extra = (ggml_tensor_extra_cl *) t->extra;
static_assert(std::is_same_v<decltype(extra->data_device), cl_mem>, "data_device type mismatch");
cl_ulong offset = extra->offset + t->view_offs;
CL_CHECK(clSetKernelArg(kernel, index, sizeof(cl_mem), &extra->data_device));
CL_CHECK(clSetKernelArg(kernel, index + 1, sizeof(cl_ulong), &offset));
return index + 2;
}
};
template <> struct ocl_kernel_arg_setter<ggml_tensor *> {
static size_t set_arg(cl_kernel kernel, size_t index, const ggml_tensor * t) {
return ocl_kernel_arg_setter<const ggml_tensor *>::set_arg(kernel, index, t);
}
};
template <typename... _TArgs> static inline void set_kernel_args(cl_kernel kernel, _TArgs... args) {
size_t index = 0;
(
[&] {
index = ocl_kernel_arg_setter<decltype(args)>::set_arg(kernel, index, args);
}(),
...);
}
// Additional tensor extra structs for quantized tensors.
// These tensors are loaded from files and should not be allocated in scratch --
// they should always be allocated from the pool. Hence, they do not have an
@ -5081,14 +5115,6 @@ static void ggml_cl_mul(ggml_backend_t backend, const ggml_tensor * src0, const
ggml_backend_opencl_context *backend_ctx = (ggml_backend_opencl_context *)backend->context;
ggml_tensor_extra_cl * extra0 = (ggml_tensor_extra_cl *)src0->extra;
ggml_tensor_extra_cl * extra1 = (ggml_tensor_extra_cl *)src1->extra;
ggml_tensor_extra_cl * extrad = (ggml_tensor_extra_cl *)dst->extra;
cl_ulong offset0 = extra0->offset + src0->view_offs;
cl_ulong offset1 = extra1->offset + src1->view_offs;
cl_ulong offsetd = extrad->offset + dst->view_offs;
bool bcast_row = false;
cl_kernel kernel;
@ -5107,16 +5133,7 @@ static void ggml_cl_mul(ggml_backend_t backend, const ggml_tensor * src0, const
kernel = backend_ctx->kernel_mul_row_f16;
}
set_kernel_args(
kernel,
extra0->data_device,
offset0,
extra1->data_device,
offset1,
extrad->data_device,
offsetd,
ne
);
set_kernel_args(kernel, src0, src1, dst, ne);
} else {
if (src0->type == GGML_TYPE_F32) {
kernel = backend_ctx->kernel_mul;
@ -5126,12 +5143,9 @@ static void ggml_cl_mul(ggml_backend_t backend, const ggml_tensor * src0, const
set_kernel_args(
kernel,
extra0->data_device,
offset0,
extra1->data_device,
offset1,
extrad->data_device,
offsetd,
src0,
src1,
dst,
ne00,
ne01,
ne02,