Slow native_call__to_dim_order_copy.out operation for fp16

🐛 Describe the bug

When running MobileNetV3 large, the native_call__to_dim_order_copy.out operation is used quite often. In the fp32 version of the model it is used 58 times, and in the fp16 version of the model it is used 116 times. I have found that the fp16 version of the operation is MUCH slower compared to the fp32 version.

Here were my commands to generate the .pte files:

weights = MobileNet_V3_Large_Weights.DEFAULT
model = mobilenet_v3_large(weights=weights)

# For fp32:
sample_inputs = (torch.randn(1, 3, 224, 224),)

# For fp16:
sample_inputs = (torch.randn(1, 3, 224, 224).half(),)
model.half()

model.eval()
exported_program = export(model, sample_inputs)

edge = to_edge(exported_program)

compile_options = {}
edge = edge.to_backend(VulkanPartitioner(compile_options))

exec_prog = edge.to_executorch()
with open("MobileNetV3_fp16_main.pte", "wb") as file:
    exec_prog.write_to_file(file)

I ran each .pte using the vulkan runner with --num_executions=100. Using ETDumps/ETRecords, I found that native_call__to_dim_order_copy.out takes on average 0.000960609 ms per operation in the fp32 version, and 2.09638 ms per operation in the fp16 version.

Versions

PyTorch version: 2.9.0.dev20250716+cpu
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.6 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
Clang version: Could not collect
CMake version: version 3.30.4
Libc version: glibc-2.31

Python version: 3.10.17 | packaged by conda-forge | (main, Apr 10 2025, 22:19:12) [GCC 13.3.0] (64-bit runtime)
Python platform: Linux-5.15.0-139-generic-x86_64-with-glibc2.31
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                         x86_64
CPU op-mode(s):                       32-bit, 64-bit
Byte Order:                           Little Endian
Address sizes:                        46 bits physical, 48 bits virtual
CPU(s):                               48
On-line CPU(s) list:                  0-47
Thread(s) per core:                   2
Core(s) per socket:                   24
Socket(s):                            1
NUMA node(s):                         1
Vendor ID:                            GenuineIntel
CPU family:                           6
Model:                                85
Model name:                           Intel(R) Xeon(R) Gold 5220R CPU @ 2.20GHz
Stepping:                             7
CPU MHz:                              2200.000
CPU max MHz:                          4000.0000
CPU min MHz:                          1000.0000
BogoMIPS:                             4400.00
Virtualization:                       VT-x
L1d cache:                            768 KiB
L1i cache:                            768 KiB
L2 cache:                             24 MiB
L3 cache:                             35.8 MiB
NUMA node0 CPU(s):                    0-47
Vulnerability Gather data sampling:   Mitigation; Microcode
Vulnerability Itlb multihit:          KVM: Mitigation: VMX disabled
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Mitigation; Enhanced IBRS
Vulnerability Spec rstack overflow:   Not affected
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Mitigation; TSX disabled
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req pku ospke avx512_vnni md_clear flush_l1d arch_capabilities

Versions of relevant libraries:
[pip3] executorch==0.8.0a0+44d24fa
[pip3] numpy==2.1.1
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu12==9.5.1.17
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] pytorch_tokenizers==0.1.0
[pip3] torch==2.9.0.dev20250716+cpu
[pip3] torchao==0.12.0+gitaee079503
[pip3] torchaudio==2.8.0.dev20250716+cpu
[pip3] torchdata==0.11.0
[pip3] torchexplorer==1.1.2
[pip3] torchsr==1.0.4
[pip3] torchtune==0.6.1
[pip3] torchview==0.2.7
[pip3] torchvision==0.24.0.dev20250716+cpu
[pip3] torchviz==0.0.3
[pip3] triton==3.3.1
[conda] executorch                0.8.0a0+44d24fa          pypi_0    pypi
[conda] numpy                     2.1.1                    pypi_0    pypi
[conda] nvidia-cublas-cu12        12.6.4.1                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.6.80                  pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.6.77                  pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.6.77                  pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.5.1.17                 pypi_0    pypi
[conda] nvidia-cufft-cu12         11.3.0.4                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.7.77                pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.7.1.2                 pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.5.4.2                 pypi_0    pypi
[conda] nvidia-cusparselt-cu12    0.6.3                    pypi_0    pypi
[conda] nvidia-nccl-cu12          2.26.2                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.6.85                  pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.6.77                  pypi_0    pypi
[conda] pytorch-tokenizers        0.1.0                    pypi_0    pypi
[conda] torch                     2.9.0.dev20250716+cpu          pypi_0    pypi
[conda] torchao                   0.12.0+gitaee079503          pypi_0    pypi
[conda] torchaudio                2.8.0.dev20250716+cpu          pypi_0    pypi
[conda] torchdata                 0.11.0                   pypi_0    pypi
[conda] torchexplorer             1.1.2                    pypi_0    pypi
[conda] torchsr                   1.0.4                    pypi_0    pypi
[conda] torchtune                 0.6.1                    pypi_0    pypi
[conda] torchview                 0.2.7                    pypi_0    pypi
[conda] torchvision               0.24.0.dev20250716+cpu          pypi_0    pypi
[conda] torchviz                  0.0.3                    pypi_0    pypi
[conda] triton                    3.3.1                    pypi_0    pypi

Reproduce the MobileNetV3 fp32 and fp16 exports from the issue, then run both .pte files with the Vulkan runner using --num_executions=100. Use ETDumps/ETRecords to compare native_call__to_dim_order_copy.out timings and determine where the fp16 slowdown occurs. Done means the reported fp16 performance problem is explained and addressed, with measurements showing the result.