[Bug] [RISC-V RVV] sigmoid operator slower with vector extension

[yanyanyanggg]

Issue: [RISC-V RVV] sigmoid operator slower with vector extension

Description

The sigmoid activation function shows performance degradation with the RISC‑V Vector (RVV) extension, achieving only 0.703× the performance of the scalar implementation. This is unexpected for a common activation function that should benefit from vectorization.

Steps to Reproduce

1. Generate the sigmoid 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_sigmoid(params, set_dir=None, platform="rv"):
    data = relay.var("data",
                     shape=(params["batch"], params["channels"],
                            params["input_height"], params["input_width"]),
                     dtype=params["dtype"])
    sigmoid = relay.sigmoid(data)
    export_op(sigmoid, params["op_name"], [data], params, set_dir=set_dir)

Performance Data

• RV execution time: 19.811500 ms
• RVV execution time: 28.199300 ms
• Acceleration ratio (RV/RVV): 0.703 (RVV is ~1.4× 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 sigmoid activation on ~1.7M elements

Expected Behavior

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

Additional Context

• The sigmoid operation is applied elementwise to a tensor of ~1.7M elements.
• While the performance regression is less severe than for some other operators, it is still significant and indicates that the vectorized implementation of sigmoid may be using suboptimal instructions or inefficient vector length management.
• This is part of a pattern where multiple mathematical and activation functions (log, sqrt, sigmoid, etc.) show performance degradation with RVV, suggesting a potential issue with vector intrinsic mapping for transcendental functions.