[Bug] [RISC-V RVV] cos operator shows slight performance degradation

[yanyanyanggg]

Issue: [RISC-V RVV] cos operator shows slight performance degradation

Description

The cosine operator shows minor performance degradation with the RISC‑V Vector (RVV) extension, achieving 0.981× the performance of the scalar implementation. While the regression is small, it still indicates room for optimization in vectorized trigonometric functions.

Steps to Reproduce

1. Generate the cos operator with the following configuration:

params = {
    "dtype": "float32",
    "batch": 14,
    "channels": 23,
    "input_height": 67,
    "input_width": 99
}

2. Export the operator to two targets:

   • RV target (scalar, without vector extension):

     llvm -mtriple=riscv64-linux-gnu -mcpu=generic-rv64 -mabi=lp64d -mattr=+64bit,+m,+a,+f,+d,+c
     

   • RVV target (with vector extension):

     llvm -mtriple=riscv64-linux-gnu -mcpu=generic-rv64 -mabi=lp64d -mattr=+64bit,+m,+a,+f,+d,+c,+v
     

3. Run performance measurement on both targets.

Operator definition code:

def export_cos(params, set_dir=None, platform="rv"):
    data = relay.var("data",
                     shape=(params["batch"], params["channels"],
                            params["input_height"], params["input_width"]),
                     dtype=params["dtype"])
    cos_op = relay.cos(data)
    export_op(cos_op, params["op_name"], [data], params, set_dir=set_dir)

Performance Data

• RV execution time: 15.894500 ms
• RVV execution time: 16.210500 ms
• Acceleration ratio (RV/RVV): 0.981 (RVV is ~1.02× slower)

Environment Information

• TVM version: 0.19.0
• LLVM version: [Please provide: llvm-config --version]
• Hardware: Spacemit K1‑X bit‑brick board
• CPU: Spacemit X60 (8 cores, 1.6 GHz)
• ISA: rv64imafdcv (with vector extensions)
• Memory: 7.6 GB
• OS: Bianbu 2.2, Linux kernel 6.6.63
• Operation: Elementwise cosine on ~1.7M elements

Expected Behavior

RVV vectorization should provide a performance improvement over the scalar RV baseline for trigonometric functions like cosine.

Additional Context

• The cos operation is applied elementwise to a tensor of ~1.7M elements.
• While the performance regression is minimal compared to other operators, it still shows that vectorization is not providing the expected speedup. This suggests that even for operations that are computationally intensive, the current RVV vectorization may not be optimal.
• This issue is part of a broader pattern where all tested operators (including sum, log, relu, bias_add, sqrt, floor, round, avg_pool2d, sigmoid, softmax, negative, max_pool2d, and cos) show performance degradation with RVV, indicating a potential systemic issue in TVM's RVV code generation or optimization.