Difference in CPU and CUDA Decode Output Values May Be Reduced with CSC Function

[pjs102793]

🐛 Describe the bug

@fmassa

First of all, I want to express my gratitude to the team for creating and maintaining this amazing open-source project. It’s a fantastic tool, and I deeply appreciate all the effort that has gone into its development.

The solution to this issue was inspired by this discussion.

Currently, the Tensor value difference between CPU and CUDA decoding seems to be quite significant. If a model used for training or inference is sensitive to color, this discrepancy could cause potential issues. This behavior was unexpected, as the TorchAudio decode pipeline I tested did not exhibit such large differences, even when decoding with CUDA.

Below is the test code I used:

import torch
from torchcodec.decoders import VideoDecoder

decoder_cpu = VideoDecoder("tmp.mp4", device="cpu")
value_cpu = decoder_cpu[0].float() / 255

decoder_cuda = VideoDecoder("tmp.mp4", device="cuda")
value_cuda = decoder_cuda[0].float() / 255

if isinstance(value_cuda, torch.Tensor):
    value_cuda = value_cuda.cpu()

if isinstance(value_cpu, torch.Tensor) and isinstance(value_cuda, torch.Tensor):
    difference = torch.abs(value_cpu - value_cuda)
    print("Difference:", difference)
    print("Max difference:", torch.max(difference).item())
    print("Mean difference:", torch.mean(difference).item())
else:
    print("Values are equal:", value_cpu == value_cuda)

Output

# Output
$ Max difference: 0.1490195393562317
$ Mean difference: 0.048768386244773865

I replaced the NPP function nppiNV12ToRGB_709HDTV_8u_P2C3R with nppiNV12ToRGB_709CSC_8u_P2C3R in convertAVFrameToDecodedOutputOnCuda located in src/torchcodec/decoders/_core/CudaDevice.cpp., I got the following results:

# Output
$ Max difference: 0.007843196392059326
$ Mean difference: 0.0015894902171567082

After simply replacing the function, all pytest cases passed successfully. This demonstrates that using the CSC function allows CUDA to produce more accurate colors.

Versions

PyTorch version: 2.5.1+cu124
Is debug build: False
CUDA used to build PyTorch: 12.4
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.24.0
Libc version: glibc-2.35

Python version: 3.10.12 (main, Jul 29 2024, 16:56:48) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.10.223-211.872.amzn2.x86_64-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.6.68
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA L4
Nvidia driver version: 535.183.06
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.4.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.4.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.4.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.4.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.4.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.4.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.4.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.4.0
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
Address sizes: 48 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 4
On-line CPU(s) list: 0-3
Vendor ID: AuthenticAMD
Model name: AMD EPYC 7R13 Processor
CPU family: 25
Model: 1
Thread(s) per core: 2
Core(s) per socket: 2
Socket(s): 1
Stepping: 1
BogoMIPS: 5299.99
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch topoext invpcid_single ssbd ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 clzero xsaveerptr rdpru wbnoinvd arat npt nrip_save vaes vpclmulqdq rdpid
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 64 KiB (2 instances)
L1i cache: 64 KiB (2 instances)
L2 cache: 1 MiB (2 instances)
L3 cache: 8 MiB (1 instance)
NUMA node(s): 1
NUMA node0 CPU(s): 0-3
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; safe RET, no microcode
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; Retpolines, IBPB conditional, IBRS_FW, STIBP always-on, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected

Versions of relevant libraries:
[pip3] numpy==1.23.5
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] torch==2.5.1
[pip3] torchaudio==2.5.1
[pip3] torchcodec==0.0.4a0
[pip3] torchvision==0.20.1
[pip3] triton==3.1.0
[conda] Could not collect

[NicolasHug]

Thank you so much for chasing this up @pjs102793 !

I opened #413 to see how low we can lower the tolerance in our tests when using. Let me run more experiments later today.