[PerfXLab] optimize fill performance

[bin913]

PR Category

[ Operator]

Type of Change

[ Performance Optimization]

Description

optimize fill.fill_scalar_ performance for fill

Issue

Progress

• Change is properly reviewed (1 reviewer required, 2 recommended).
• Change is responded to an issue.
• Change is fully covered by a UT.

Performance

Operator: fill_scalar_  Performance Test (dtype=torch.float16, mode=kernel,level=core)
Status       Torch Latency (ms)    Gems Latency (ms)         Gems Speedup          Size Detail
-----------------------------------------------------------------------------------------------
SUCCESS               0.653856            0.653792               1.000          [torch.Size([1073741824]), 3.14159]
SUCCESS               0.005008            0.004992               1.003          [torch.Size([64, 64]), 3.14159]
SUCCESS               0.015104            0.015296               0.987          [torch.Size([4096, 4096]), 3.14159]
SUCCESS               0.015328            0.015328               1.000          [torch.Size([64, 512, 512]), 3.14159]
SUCCESS               0.654080            0.654224               1.000          [torch.Size([1024, 1024, 1024]), 3.14159]


Operator: fill_scalar_  Performance Test (dtype=torch.float32, mode=kernel,level=core)
Status       Torch Latency (ms)    Gems Latency (ms)         Gems Speedup          Size Detail
-----------------------------------------------------------------------------------------------
SUCCESS               1.305952            1.303360               1.002          [torch.Size([1073741824]), 3.14159]
SUCCESS               0.005120            0.005024               1.019          [torch.Size([64, 64]), 3.14159]
SUCCESS               0.025280            0.025376               0.996          [torch.Size([4096, 4096]), 3.14159]
SUCCESS               0.025120            0.025216               0.996          [torch.Size([64, 512, 512]), 3.14159]
SUCCESS               1.306016            1.303264               1.002          [torch.Size([1024, 1024, 1024]), 3.14159]


Operator: fill_scalar_  Performance Test (dtype=torch.bfloat16, mode=kernel,level=core)
Status       Torch Latency (ms)    Gems Latency (ms)         Gems Speedup          Size Detail
-----------------------------------------------------------------------------------------------
SUCCESS               0.654176            0.654080               1.000          [torch.Size([1073741824]), 3.14159]
SUCCESS               0.004992            0.005152               0.969          [torch.Size([64, 64]), 3.14159]
SUCCESS               0.015232            0.015296               0.996          [torch.Size([4096, 4096]), 3.14159]
SUCCESS               0.015488            0.015328               1.010          [torch.Size([64, 512, 512]), 3.14159]
SUCCESS               0.653888            0.653904               1.000          [torch.Size([1024, 1024, 1024]), 3.14159]

[you-and-you]

Why is the FlagGems benchmark for fill_scalar_ commented out?