torch.uniform_() is single-threaded on CPU

[vincent-163]

🐛 Describe the bug

The uniform_ function is single-threaded and slow, causing nn.Linear to be slow during initialization. Since the first step to loading a model is to create an empty model, the creation of an empty LlamaModelForCausalLM alone costs 90 seconds while loading the state dict costs just a second. I expected that the creation of nn.Linear be as fast as torch.randn.

import torch
print(torch.__version__) # 2.2.2+cu118

x = torch.empty(40960, 40960)
x.uniform_(-1, 1) # Max CPU usage: 100%

Versions

Collecting environment information...
PyTorch version: 2.2.2+cu118
Is debug build: False
CUDA used to build PyTorch: 11.8
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 3.29.2
Libc version: glibc-2.35

Python version: 3.12.1 | packaged by Anaconda, Inc. | (main, Jan 19 2024, 15:51:05) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.15.0-102-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 11.8.89
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA GeForce RTX 4090
GPU 1: NVIDIA GeForce RTX 4090
GPU 2: NVIDIA GeForce RTX 4090
GPU 3: NVIDIA GeForce RTX 4090
GPU 4: NVIDIA GeForce RTX 4090
GPU 5: NVIDIA GeForce RTX 4090

Nvidia driver version: 545.23.08
cuDNN version: Could not collect
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:                      46 bits physical, 48 bits virtual
Byte Order:                         Little Endian
CPU(s):                             104
On-line CPU(s) list:                0-103
Vendor ID:                          GenuineIntel
Model name:                         Intel(R) Xeon(R) Gold 6230R CPU @ 2.10GHz
CPU family:                         6
Model:                              85
Thread(s) per core:                 2
Core(s) per socket:                 26
Socket(s):                          2
Stepping:                           7
CPU max MHz:                        4000.0000
CPU min MHz:                        1000.0000
BogoMIPS:                           4200.00
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 pku ospke avx512_vnni md_clear flush_l1d arch_capabilities
Virtualization:                     VT-x
L1d cache:                          1.6 MiB (52 instances)
L1i cache:                          1.6 MiB (52 instances)
L2 cache:                           52 MiB (52 instances)
L3 cache:                           71.5 MiB (2 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-25,52-77
NUMA node1 CPU(s):                  26-51,78-103
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 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 IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Mitigation; TSX disabled

Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] torch==2.2.2+cu118
[pip3] triton==2.3.0
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] torch                     2.2.2+cu118              pypi_0    pypi
[conda] triton                    2.3.0                    pypi_0    pypi

cc @msaroufim @fritzo @neerajprad @alicanb @nikitaved @albanD @mruberry @jbschlosser @walterddr @mikaylagawarecki @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10

[cpuhrsch]

This might be a fairly high impact change if the underlying implementation is easy to parallelize with parallel_for. I think the hard part might be setting up random number generation state across threads in a lock free way.

[mingfeima]

@CaoE