Errors when building on an MI250x server with ROCm 5.7 and PyTorch 2.2.1

[netw0rkf10w]

Describe the Bug

When installing on an MI250x server with ROCm 5.7 and PyTorch 2.2.1, I obtained the following errors:

 /home/user/code/apex/csrc/fused_dense_hip.hip:63:7: error: use of undeclared identifier 'CUBLAS_COMPUTE_64F'
        CUBLAS_COMPUTE_64F,
        ^
  /home/user/code/apex/csrc/fused_dense_hip.hip:101:7: error: use of undeclared identifier 'CUBLAS_COMPUTE_32F'
        CUBLAS_COMPUTE_32F,
        ^
  /home/user/code/apex/csrc/fused_dense_hip.hip:139:7: error: use of undeclared identifier 'CUBLAS_COMPUTE_16F'
        CUBLAS_COMPUTE_16F,
        ^
  3 errors generated when compiling for gfx90a.

Minimal Steps/Code to Reproduce the Bug

```bash git clone https://github.com/ROCm/apex.git cd apex git checkout release/1.2.0 export HCC_AMDGPU_TARGET=gfx90a export PYTORCH_ROCM_ARCH=gfx90a export GPU_ARCHS=gfx90a export MAX_JOBS=8 pip install -v --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./ --user ``` **Expected Behavior**

Environment

Collecting environment information...
PyTorch version: 2.2.1+rocm5.7
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 5.7.31921-d1770ee1b

OS: Red Hat Enterprise Linux release 8.8 (Ootpa) (x86_64)
GCC version: (GCC) 12.2.1 20221121 (Red Hat 12.2.1-7)
Clang version: 15.0.7 (Red Hat 15.0.7-1.module+el8.8.0+17939+b58878af)
CMake version: version 3.28.3
Libc version: glibc-2.28

Python version: 3.10.10 (main, Apr 14 2023, 19:33:04) [GCC 10.3.1 20210422 (Red Hat 10.3.1-1)] (64-bit runtime)
Python platform: Linux-4.18.0-477.10.1.el8_8.x86_64-x86_64-with-glibc2.28
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: AMD Instinct MI250X (gfx90a:sramecc+:xnack-)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: 5.7.31921
MIOpen runtime version: 2.20.0
Is XNNPACK available: True

CPU:
Architecture:        x86_64
CPU op-mode(s):      32-bit, 64-bit
Byte Order:          Little Endian
CPU(s):              128
On-line CPU(s) list: 0-127
Thread(s) per core:  2
Core(s) per socket:  64
Socket(s):           1
NUMA node(s):        4
Vendor ID:           AuthenticAMD
CPU family:          25
Model:               48
Model name:          AMD EPYC 7A53 64-Core Processor
Stepping:            1
CPU MHz:             2000.000
CPU max MHz:         3541.0149
CPU min MHz:         1500.0000
BogoMIPS:            3992.70
Virtualization:      AMD-V
L1d cache:           32K
L1i cache:           32K
L2 cache:            512K
L3 cache:            32768K
NUMA node0 CPU(s):   0-15,64-79
NUMA node1 CPU(s):   16-31,80-95
NUMA node2 CPU(s):   32-47,96-111
NUMA node3 CPU(s):   48-63,112-127
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 pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm

Versions of relevant libraries:
[pip3] numpy==1.23.5
[pip3] pytorch-triton-rocm==2.2.0
[pip3] torch==2.2.1+rocm5.7
[pip3] torchaudio==2.2.1+rocm5.7
[pip3] torching==0.0.1
[pip3] torchvision==0.17.1+rocm5.7
[conda] Could not collect

[etiennemlb]

Because they do not work in a develop branch, we end up getting an unfinished version when pulling the master. The officially supported pytorch<->rocm apex version are:	APEX Version	APEX branch	Torch Version
1.3.0 | master | 2.3
1.2.0 | release/1.2.0 | 2.2
1.1.0 | release/1.1.0 | 2.1
1.0.0 | release/1.0.0 | 2.0 and older

So be sure to do some kind of git clone -b release/1.2.0 .. if you are working with torch 2.2 which relies on ROCm 5.7.

[etiennemlb]

Also, a fix landed in d89d4bd - Correct the CUBLAS_COMPUTE type. You can find the hipify supported conversions (cuda->hip) here: https://rocm.docs.amd.com/projects/HIPIFY/en/latest/tables/CUBLAS_API_supported_by_HIP.html

[netw0rkf10w]

@etiennemlb Thanks a lot! As you can see, I already had git checkout release/1.2.0 in my commands, so I was using the right branch. I'm going to try again with https://github.com/ROCm/apex/commit/d89d4bd2bf15abda83113f2d0c846ff01fd90567.

[formiel]

Also, a fix landed in d89d4bd - Correct the CUBLAS_COMPUTE type. You can find the hipify supported conversions (cuda->hip) here: https://rocm.docs.amd.com/projects/HIPIFY/en/latest/tables/CUBLAS_API_supported_by_HIP.html

git checkout d89d4bd works for me! Thank you so much @etiennemlb @netw0rkf10w !

[etiennemlb]

@formiel good news, but still, I do not understand why the recipe given here: https://dci.dci-gitlab.cines.fr/webextranet/software_stack/libraries/index.html#apex

does not work for you. Of course, you need to adapt the rocm version to the apex version and to the torch version all these are interlinked.

[formiel]

@etiennemlb The recipe in the link you sent works for me with PyTorch 2.4.1 and ROCm 6.0.0, but not with PyTorch 2.2.2 and ROCm 5.7. It's strange, though, because I didn’t encounter this error before, even when installing in another virtual environment with the same PyTorch 2.2.2 and ROCm 5.7 setup. I'm not sure what has changed, but I’m glad the installation is successful now with the commit you shared. Thanks again for your help!