[Bug]: Inductor fails to fuse pointwise ops with sequence parallelism + async TP #27699
Description
Your current environment
The output of python collect_env.py
Collecting environment information...
uv is set
==============================
System Info
==============================
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 4.1.0
Libc version : glibc-2.35
==============================
PyTorch Info
==============================
PyTorch version : 2.9.0+cu128
Is debug build : False
CUDA used to build PyTorch : 12.8
ROCM used to build PyTorch : N/A
==============================
Python Environment
==============================
Python version : 3.10.12 (main, Jan 17 2025, 14:35:34) [GCC 11.4.0] (64-bit runtime)
Python platform : Linux-5.15.0-113-generic-x86_64-with-glibc2.35
==============================
CUDA / GPU Info
==============================
Is CUDA available : True
CUDA runtime version : 12.8.61
CUDA_MODULE_LOADING set to :
GPU models and configuration :
GPU 0: NVIDIA H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3
GPU 2: NVIDIA H100 80GB HBM3
GPU 3: NVIDIA H100 80GB HBM3
GPU 4: NVIDIA H100 80GB HBM3
GPU 5: NVIDIA H100 80GB HBM3
GPU 6: NVIDIA H100 80GB HBM3
GPU 7: NVIDIA H100 80GB HBM3
Nvidia driver version : 570.86.10
cuDNN version : Could not collect
HIP runtime version : N/A
MIOpen runtime version : N/A
Is XNNPACK available : True
==============================
CPU Info
==============================
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 52 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 384
On-line CPU(s) list: 0-383
Vendor ID: AuthenticAMD
Model name: AMD EPYC 9654 96-Core Processor
CPU family: 25
Model: 17
Thread(s) per core: 2
Core(s) per socket: 96
Socket(s): 2
Stepping: 1
Frequency boost: enabled
CPU max MHz: 3707.8120
CPU min MHz: 1500.0000
BogoMIPS: 4799.98
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 rapl 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 skinit 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 avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d
Virtualization: AMD-V
L1d cache: 6 MiB (192 instances)
L1i cache: 6 MiB (192 instances)
L2 cache: 192 MiB (192 instances)
L3 cache: 768 MiB (24 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-95,192-287
NUMA node1 CPU(s): 96-191,288-383
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 Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; safe RET
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; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
==============================
Versions of relevant libraries
==============================
[pip3] efficientnet-pytorch==0.7.1
[pip3] flashinfer-python==0.4.1
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.15.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-cutlass-dsl==4.2.1
[pip3] nvidia-ml-py==13.580.82
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvshmem-cu12==3.3.20
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] open-clip-torch==2.32.0
[pip3] pytorch-lightning==2.5.2
[pip3] pyzmq==27.1.0
[pip3] segmentation-models-pytorch==0.4.0
[pip3] sentence-transformers==3.2.1
[pip3] terratorch==1.0.2
[pip3] torch==2.9.0+cu128
[pip3] torchaudio==2.9.0+cu128
[pip3] torchgeo==0.6.2
[pip3] torchmetrics==1.7.4
[pip3] torchvision==0.24.0+cu128
[pip3] transformers==4.56.2
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==3.5.0
[pip3] tritonclient==2.51.0
[pip3] vector-quantize-pytorch==1.21.2
[conda] Could not collect
==============================
vLLM Info
==============================
ROCM Version : Could not collect
vLLM Version : 0.11.0rc2.dev353+g11064728b.d20251012 (git sha: 11064728b, date: 20251012)
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 NODE NODE PIX NODE SYS SYS SYS SYS 0-95,192-287 0 N/A
GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 NODE NODE NODE PIX SYS SYS SYS SYS 0-95,192-287 0 N/A
GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 NODE PIX NODE NODE SYS SYS SYS SYS 0-95,192-287 0 N/A
GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 PIX NODE NODE NODE SYS SYS SYS SYS 0-95,192-287 0 N/A
GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 SYS SYS SYS SYS NODE NODE PIX NODE 96-191,288-383 1 N/A
GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 SYS SYS SYS SYS NODE NODE NODE PIX 96-191,288-383 1 N/A
GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 SYS SYS SYS SYS NODE PIX NODE NODE 96-191,288-383 1 N/A
GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X SYS SYS SYS SYS PIX NODE NODE NODE 96-191,288-383 1 N/A
NIC0 NODE NODE NODE PIX SYS SYS SYS SYS X NODE NODE NODE SYS SYS SYS SYS
NIC1 NODE NODE PIX NODE SYS SYS SYS SYS NODE X NODE NODE SYS SYS SYS SYS
NIC2 PIX NODE NODE NODE SYS SYS SYS SYS NODE NODE X NODE SYS SYS SYS SYS
NIC3 NODE PIX NODE NODE SYS SYS SYS SYS NODE NODE NODE X SYS SYS SYS SYS
NIC4 SYS SYS SYS SYS NODE NODE NODE PIX SYS SYS SYS SYS X NODE NODE NODE
NIC5 SYS SYS SYS SYS NODE NODE PIX NODE SYS SYS SYS SYS NODE X NODE NODE
NIC6 SYS SYS SYS SYS PIX NODE NODE NODE SYS SYS SYS SYS NODE NODE X NODE
NIC7 SYS SYS SYS SYS NODE PIX NODE NODE SYS SYS SYS SYS NODE NODE NODE X
Legend:
X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks
NIC Legend:
NIC0: mlx5_0
NIC1: mlx5_1
NIC2: mlx5_2
NIC3: mlx5_3
NIC4: mlx5_4
NIC5: mlx5_5
NIC6: mlx5_6
NIC7: mlx5_7
==============================
Environment Variables
==============================
LD_LIBRARY_PATH=/usr/local/cuda-12.8/lib64
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
🐛 Describe the bug
When using sequence parallelism, we transform a sequence of all_reduce -> rms_norm -> quant into reduce_scatter -> rms_norm -> quant_fp8 -> all_gather. Async TP then fuses the 2 communication ops onto the GEMMs surrounding this pattern. Whether we're using SP or not, we expect the sequence of rms_norm -> quant_fp8 to produce a single fused Triton kernel. That is the case without SP + Async TP but with those two passes, Inductor generates two separate kernels for some reason.
This requires #27126. Repro command:
python examples/offline_inference/basic/generate.py --model redhatai/meta-llama-3.1-70B-Instruct-FP8 --tensor-parallel-size=4 --kv-cache-dtype=fp8 --no-enable-prefix-caching --gpu_memory_utilization=0.8 -O.pass_config='{"enable_async_tp":true,"enable_noop":true}' -O.use_inductor_graph_partition=True -O.cudagraph_mode=NONE --load-format=dummy
Inductor output code without SP + Async TP:
def partition_1(...):
...
# Topologically Sorted Source Nodes: [view_7, unified_attention_with_output, to_10, reciprocal_2, mul_15, clamp_2, to_11, cutlass_scaled_mm_1], Original ATen: [aten.view, aten._to_copy, aten.reciprocal, aten.mul, aten.clamp, _C.cutlass_scaled_mm]
triton_poi_fused__to_copy_clamp_cutlass_scaled_mm_mul_reciprocal_view_3_xnumel = 2048*s72
stream0 = get_raw_stream(0)
triton_poi_fused__to_copy_clamp_cutlass_scaled_mm_mul_reciprocal_view_3.run(buf10, arg10_1, buf15, triton_poi_fused__to_copy_clamp_cutlass_scaled_mm_mul_reciprocal_view_3_xnumel, stream=stream0)
# Topologically Sorted Source Nodes: [view_7, unified_attention_with_output, to_10, reciprocal_2, mul_15, clamp_2, to_11, cutlass_scaled_mm_1], Original ATen: [aten.view, aten._to_copy, aten.reciprocal, aten.mul, aten.clamp, _C.cutlass_scaled_mm]
torch.ops._C.cutlass_scaled_mm.default(buf14, buf15, arg11_1, arg10_1, arg12_1, None)
del arg10_1
del arg11_1
del arg12_1
buf18 = empty_strided_cuda((s72, 14336), (14336, 1), torch.bfloat16)
# Topologically Sorted Source Nodes: [all_reduce_1], Original ATen: [vllm.all_reduce]
buf19 = torch.ops.vllm.all_reduce.default(buf14, 'tp:0')
buf20 = buf19
assert_size_stride(buf20, (s72, 8192), (8192, 1), 'torch.ops.vllm.all_reduce.default')
assert_alignment(buf20, 16, 'torch.ops.vllm.all_reduce.default')
del buf19
buf22 = empty_strided_cuda((s72, 8192), (8192, 1), torch.float8_e4m3fn)
# Topologically Sorted Source Nodes: [to_12, add_4, pow_2, mean_1, add_5, rsqrt_1, mul_16, to_14, mul_17, to_15, reciprocal_3, mul_18, clamp_3, to_16, cutlass_scaled_mm_2], Original ATen: [aten._to_copy, aten.add, aten.pow, aten.mean, aten.rsqrt, aten.mul, aten.reciprocal, aten.clamp, _C.cutlass_scaled_mm]
stream0 = get_raw_stream(0)
triton_red_fused__to_copy_add_clamp_cutlass_scaled_mm_mean_mul_pow_reciprocal_rsqrt_4.run(buf20, buf3, arg13_1, arg14_1, buf22, s72, 8192, stream=stream0)
del arg13_1
# Topologically Sorted Source Nodes: [to_12, add_4, pow_2, mean_1, add_5, rsqrt_1, mul_16, to_14, mul_17, to_15, reciprocal_3, mul_18, clamp_3, to_16, cutlass_scaled_mm_2], Original ATen: [aten._to_copy, aten.add, aten.pow, aten.mean, aten.rsqrt, aten.mul, aten.reciprocal, aten.clamp, _C.cutlass_scaled_mm]
torch.ops._C.cutlass_scaled_mm.default(buf18, buf22, arg15_1, arg14_1, arg16_1, None)
Inductor output code with SP + Async TP:
def partition_1(...):
...
# Topologically Sorted Source Nodes: [view_7, unified_attention_with_output, to_10, reciprocal_2, mul_15, clamp_2, to_11], Original ATen: [aten.view, aten._to_copy, aten.reciprocal, aten.mul, aten.clamp, vllm.patched_fused_scaled_matmul_reduce_scatter]
triton_poi_fused__to_copy_clamp_mul_patched_fused_scaled_matmul_reduce_scatter_reciprocal_view_5_xnumel = 2048*s72
stream0 = get_raw_stream(0)
triton_poi_fused__to_copy_clamp_mul_patched_fused_scaled_matmul_reduce_scatter_reciprocal_view_5.run(buf10, arg10_1, buf14, triton_poi_fused__to_copy_clamp_mul_patched_fused_scaled_matmul_reduce_scatter_reciprocal_view_5_xnumel, stream=stream0)
# Topologically Sorted Source Nodes: [view_7, unified_attention_with_output, to_10, reciprocal_2, mul_15, clamp_2, to_11], Original ATen: [aten.view, aten._to_copy, aten.reciprocal, aten.mul, aten.clamp, vllm.patched_fused_scaled_matmul_reduce_scatter]
buf15 = torch.ops.vllm.patched_fused_scaled_matmul_reduce_scatter.default(buf14, arg11_1, arg10_1, arg12_1, 'avg', 0, 0, '3', [s72, 8192], None, None, torch.bfloat16, use_fast_accum=False)
del arg10_1
del arg11_1
del arg12_1
buf16 = buf15
assert_size_stride(buf16, (s72 // 4, 8192), (8192, 1), 'torch.ops.vllm.patched_fused_scaled_matmul_reduce_scatter.default')
assert_alignment(buf16, 16, 'torch.ops.vllm.patched_fused_scaled_matmul_reduce_scatter.default')
del buf15
buf17 = empty_strided_cuda((s72 // 4, 1), (1, s72 // 4), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: [aten._to_copy, aten.add, aten.pow, aten.mean]
triton_red_fused__to_copy_add_mean_pow_6_xnumel = s72 // 4
stream0 = get_raw_stream(0)
triton_red_fused__to_copy_add_mean_pow_6.run(buf16, buf2, buf17, triton_red_fused__to_copy_add_mean_pow_6_xnumel, 8192, stream=stream0)
buf18 = empty_strided_cuda((s72 // 4, 8192), (8192, 1), torch.float8_e4m3fn)
# Topologically Sorted Source Nodes: [], Original ATen: [aten._to_copy, aten.add, aten.pow, aten.mean, aten.rsqrt, aten.mul, aten.reciprocal, aten.clamp, symm_mem.fused_all_gather_scaled_matmul]
triton_poi_fused__to_copy_add_clamp_fused_all_gather_scaled_matmul_mean_mul_pow_reciprocal_rsqrt_7_xnumel = 8192*(s72 // 4)
stream0 = get_raw_stream(0)
triton_poi_fused__to_copy_add_clamp_fused_all_gather_scaled_matmul_mean_mul_pow_reciprocal_rsqrt_7.run(buf16, buf2, buf17, arg13_1, arg14_1, buf18, triton_poi_fused__to_copy_add_clamp_fused_all_gather_scaled_matmul_mean_mul_pow_reciprocal_rsqrt_7_xnumel, stream=stream0)
del arg13_1
# Topologically Sorted Source Nodes: [], Original ATen: [aten._to_copy, aten.add, aten.pow, aten.mean, aten.rsqrt, aten.mul, aten.reciprocal, aten.clamp, symm_mem.fused_all_gather_scaled_matmul]
buf19 = torch.ops.symm_mem.fused_all_gather_scaled_matmul.default(buf18, [arg15_1], arg14_1, [arg16_1], 0, '3', [None], [None], [15], [False])
Before submitting a new issue...
- Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.