Gahow Wang
d77f921a12
- Naive GEMM kernel: one thread per output element (F32 + BF16) - Tiled GEMM kernel: 32x32 shared memory tiles (F32 + BF16) - cuBLAS wrapper: cublasGemmEx with row-major trick - GemmBackend enum for runtime backend selection - CublasContext RAII handle - Made error::check public for cross-crate use - 17 GEMM tests: small/medium/rect sizes, all backends, F32+BF16 - Cross-backend consistency verified (naive vs tiled vs cuBLAS) - All 44 tests pass across all crates Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
13 lines
236 B
TOML
13 lines
236 B
TOML
[package]
|
|
name = "xserv-kernels"
|
|
version.workspace = true
|
|
edition.workspace = true
|
|
|
|
[build-dependencies]
|
|
cc = "1"
|
|
|
|
[dependencies]
|
|
xserv-cuda = { path = "../xserv-cuda" }
|
|
xserv-tensor = { path = "../xserv-tensor" }
|
|
half.workspace = true
|