SYCL SET operator optimized for F32 tensors

[GittyBurstein]

SYCL F32 SET Operator Implementation

This PR implements the SET operator for F32 in SYCL. This is the first implementation of SET in SYCL.

All tests for F32 passed successfully, ensuring correctness. The implementation preserves the existing library structure and follows GGML conventions.

Performance Improvements

• I ran an extensive set of benchmarks, and the changes lead to significant performance improvements.
• The GPU clearly outperforms the CPU.
• For small matrices, performance improved by more than 1.5x compared to the previous approach.
• Larger matrices were not fully benchmarked due to GPU memory limits, but improvements are expected there as well.

Changed Files

• ggml/src/ggml-sycl/ggml-sycl.cpp
• ggml/src/ggml-sycl/set.cpp
• ggml/src/ggml-sycl/set.hpp

Technical Details

• Uses SYCL parallel_for to perform element-wise SET operations.
• Supports multi-dimensional tensors (up to 4D) with contiguous memory layouts.
• Handles both in-place and out-of-place operations.
• Optimized for GPU, leveraging thread-level parallelism.
• Inline function set_f32 converts a linear index to multi-dimensional indices for accurate copying.

Example

inline void set_f32(
    const float* src, float* dst,
    const int64_t ne0, const int64_t ne1,
    const int64_t ne2, const int64_t ne3,
    const int64_t nb[3], const int64_t src_nb[3],
    const int64_t offset_elem,
    const nd_item<1>& item) 
{
    const size_t idx = item.get_global_id(0);
    const size_t total = ne0 * ne1 * ne2 * ne3;
    if (idx >= total) return;

    const size_t i3 = idx / (ne2 * ne1 * ne0);
    const size_t rem = idx % (ne2 * ne1 * ne0);
    const size_t i2 = rem / (ne1 * ne0);
    const size_t rem2 = rem % (ne1 * ne0);
    const size_t i1 = rem2 / ne0;
    const size_t i0 = rem2 % ne0;

    dst[i0 + i1*nb[0] + i2*nb[1] + i3*nb[2] + offset_elem] =
        src[i0 + i1*src_nb[0] + i2*src_nb[1] + i3*src_nb[2]];
}