Review CUB `util.ptx` for CCCL 2.x #3342

fbusato · 2025-01-10T20:36:00Z

#3289

Description

Add deprecation warnings to:
- CUDA special register usage
- BFI, IADD3, PRMT, BAR, FMUL_RZ, FFMA_RZ, ThreadTrap because never used
Replace PTX shf/shl (detail namespace) with standard shift operations, which could be beneficial for NVVM optimizations

cub/cub/agent/agent_radix_sort_histogram.cuh

bernhardmgruber

Except for the shift and a typo, LGTM

cub/cub/warp/specializations/warp_reduce_shfl.cuh

cub/test/test_util.h

cub/cub/agent/agent_batch_memcpy.cuh

github-actions · 2025-01-11T03:43:54Z

🟨 CI finished in 3h 04m: Pass: 97%/78 | Total: 2d 06h | Avg: 41m 41s | Max: 1h 37m | Hits: 216%/12368

🟨 cub: Pass: 94%/38 | Total: 1d 09h | Avg: 53m 05s | Max: 1h 37m | Hits: 219%/3108

🔍 cpu: amd64 🔍
  🔍 amd64              Pass:  94%/36  | Total:  1d 07h | Avg: 52m 55s | Max:  1h 37m | Hits: 219%/3108  
  🟩 arm64              Pass: 100%/2   | Total:  1h 52m | Avg: 56m 10s | Max: 57m 13s
🔍 ctk: 12.6 🔍
  🟩 12.0               Pass: 100%/5   | Total:  4h 53m | Avg: 58m 38s | Max:  1h 07m | Hits: 220%/777   
  🟩 12.5               Pass: 100%/2   | Total:  2h 19m | Avg:  1h 09m | Max:  1h 10m
  🔍 12.6               Pass:  93%/31  | Total:  1d 02h | Avg: 51m 07s | Max:  1h 37m | Hits: 219%/2331  
🔍 cudacxx: nvcc12.6 🔍
  🟩 ClangCUDA18        Pass: 100%/2   | Total:  1h 57m | Avg: 58m 34s | Max: 58m 58s
  🟩 nvcc12.0           Pass: 100%/5   | Total:  4h 53m | Avg: 58m 38s | Max:  1h 07m | Hits: 220%/777   
  🟩 nvcc12.5           Pass: 100%/2   | Total:  2h 19m | Avg:  1h 09m | Max:  1h 10m
  🔍 nvcc12.6           Pass:  93%/29  | Total:  1d 00h | Avg: 50m 36s | Max:  1h 37m | Hits: 219%/2331  
🔍 cudacxx_family: nvcc 🔍
  🟩 ClangCUDA          Pass: 100%/2   | Total:  1h 57m | Avg: 58m 34s | Max: 58m 58s
  🔍 nvcc               Pass:  94%/36  | Total:  1d 07h | Avg: 52m 47s | Max:  1h 37m | Hits: 219%/3108  
🔍 gpu: v100 🔍
  🟩 h100               Pass: 100%/2   | Total: 43m 32s | Avg: 21m 46s | Max: 27m 24s
  🔍 v100               Pass:  94%/36  | Total:  1d 08h | Avg: 54m 50s | Max:  1h 37m | Hits: 219%/3108  
🚨 jobs: TestGPU 🚨
  🟩 Build              Pass: 100%/31  | Total:  1d 05h | Avg: 57m 10s | Max:  1h 12m | Hits: 219%/3108  
  🟩 DeviceLaunch       Pass: 100%/1   | Total: 27m 21s | Avg: 27m 21s | Max: 27m 21s
  🟩 GraphCapture       Pass: 100%/1   | Total: 19m 12s | Avg: 19m 12s | Max: 19m 12s
  🟩 HostLaunch         Pass: 100%/3   | Total:  2h 26m | Avg: 48m 45s | Max:  1h 37m
  🔥 TestGPU            Pass:   0%/2   | Total: 52m 23s | Avg: 26m 11s | Max: 28m 20s
🔍 std: 20 🔍
  🟩 17                 Pass: 100%/14  | Total: 14h 12m | Avg:  1h 00m | Max:  1h 12m | Hits: 220%/2331  
  🔍 20                 Pass:  91%/24  | Total: 19h 24m | Avg: 48m 32s | Max:  1h 37m | Hits: 216%/777   
🟨 cxx
  🟩 Clang14            Pass: 100%/4   | Total:  3h 41m | Avg: 55m 20s | Max: 57m 15s
  🟩 Clang15            Pass: 100%/1   | Total: 59m 04s | Avg: 59m 04s | Max: 59m 04s
  🟩 Clang16            Pass: 100%/1   | Total: 53m 08s | Avg: 53m 08s | Max: 53m 08s
  🟩 Clang17            Pass: 100%/1   | Total: 55m 47s | Avg: 55m 47s | Max: 55m 47s
  🟨 Clang18            Pass:  85%/7   | Total:  5h 37m | Avg: 48m 11s | Max: 58m 58s
  🟩 GCC7               Pass: 100%/2   | Total:  1h 50m | Avg: 55m 28s | Max: 56m 58s
  🟩 GCC8               Pass: 100%/1   | Total: 58m 41s | Avg: 58m 41s | Max: 58m 41s
  🟩 GCC9               Pass: 100%/2   | Total:  2h 00m | Avg:  1h 00m | Max:  1h 00m
  🟩 GCC10              Pass: 100%/1   | Total: 55m 24s | Avg: 55m 24s | Max: 55m 24s
  🟩 GCC11              Pass: 100%/1   | Total: 55m 59s | Avg: 55m 59s | Max: 55m 59s
  🟩 GCC12              Pass: 100%/3   | Total:  1h 46m | Avg: 35m 25s | Max:  1h 02m
  🟨 GCC13              Pass:  87%/8   | Total:  6h 09m | Avg: 46m 10s | Max:  1h 37m
  🟩 MSVC14.29          Pass: 100%/2   | Total:  2h 13m | Avg:  1h 06m | Max:  1h 07m | Hits: 220%/1554  
  🟩 MSVC14.39          Pass: 100%/2   | Total:  2h 21m | Avg:  1h 10m | Max:  1h 12m | Hits: 218%/1554  
  🟩 NVHPC24.7          Pass: 100%/2   | Total:  2h 19m | Avg:  1h 09m | Max:  1h 10m
🟨 cxx_family
  🟨 Clang              Pass:  92%/14  | Total: 12h 06m | Avg: 51m 54s | Max: 59m 04s
  🟨 GCC                Pass:  94%/18  | Total: 14h 37m | Avg: 48m 43s | Max:  1h 37m
  🟩 MSVC               Pass: 100%/4   | Total:  4h 34m | Avg:  1h 08m | Max:  1h 12m | Hits: 219%/3108  
  🟩 NVHPC              Pass: 100%/2   | Total:  2h 19m | Avg:  1h 09m | Max:  1h 10m
🟩 sm
  🟩 90                 Pass: 100%/2   | Total: 43m 32s | Avg: 21m 46s | Max: 27m 24s
  🟩 90a                Pass: 100%/1   | Total: 23m 17s | Avg: 23m 17s | Max: 23m 17s

🟩 thrust: Pass: 100%/37 | Total: 19h 52m | Avg: 32m 13s | Max: 1h 06m | Hits: 215%/9260

🟩 cmake_options
  🟩 -DTHRUST_DISPATCH_TYPE=Force32bit Pass: 100%/2   | Total: 37m 27s | Avg: 18m 43s | Max: 25m 30s
🟩 cpu
  🟩 amd64              Pass: 100%/35  | Total: 18h 55m | Avg: 32m 25s | Max:  1h 06m | Hits: 215%/9260  
  🟩 arm64              Pass: 100%/2   | Total: 57m 14s | Avg: 28m 37s | Max: 30m 32s
🟩 ctk
  🟩 12.0               Pass: 100%/5   | Total:  3h 04m | Avg: 36m 55s | Max: 59m 04s | Hits: 178%/1852  
  🟩 12.5               Pass: 100%/2   | Total:  1h 48m | Avg: 54m 29s | Max: 54m 31s
  🟩 12.6               Pass: 100%/30  | Total: 14h 58m | Avg: 29m 57s | Max:  1h 06m | Hits: 225%/7408  
🟩 cudacxx
  🟩 ClangCUDA18        Pass: 100%/2   | Total: 50m 51s | Avg: 25m 25s | Max: 25m 53s
  🟩 nvcc12.0           Pass: 100%/5   | Total:  3h 04m | Avg: 36m 55s | Max: 59m 04s | Hits: 178%/1852  
  🟩 nvcc12.5           Pass: 100%/2   | Total:  1h 48m | Avg: 54m 29s | Max: 54m 31s
  🟩 nvcc12.6           Pass: 100%/28  | Total: 14h 07m | Avg: 30m 16s | Max:  1h 06m | Hits: 225%/7408  
🟩 cudacxx_family
  🟩 ClangCUDA          Pass: 100%/2   | Total: 50m 51s | Avg: 25m 25s | Max: 25m 53s
  🟩 nvcc               Pass: 100%/35  | Total: 19h 01m | Avg: 32m 36s | Max:  1h 06m | Hits: 215%/9260  
🟩 cxx
  🟩 Clang14            Pass: 100%/4   | Total:  1h 58m | Avg: 29m 40s | Max: 31m 54s
  🟩 Clang15            Pass: 100%/1   | Total: 33m 44s | Avg: 33m 44s | Max: 33m 44s
  🟩 Clang16            Pass: 100%/1   | Total: 29m 35s | Avg: 29m 35s | Max: 29m 35s
  🟩 Clang17            Pass: 100%/1   | Total: 29m 37s | Avg: 29m 37s | Max: 29m 37s
  🟩 Clang18            Pass: 100%/7   | Total:  2h 46m | Avg: 23m 48s | Max: 30m 51s
  🟩 GCC7               Pass: 100%/2   | Total:  1h 05m | Avg: 32m 33s | Max: 32m 37s
  🟩 GCC8               Pass: 100%/1   | Total: 29m 35s | Avg: 29m 35s | Max: 29m 35s
  🟩 GCC9               Pass: 100%/2   | Total:  1h 05m | Avg: 32m 43s | Max: 33m 31s
  🟩 GCC10              Pass: 100%/1   | Total: 33m 05s | Avg: 33m 05s | Max: 33m 05s
  🟩 GCC11              Pass: 100%/1   | Total: 31m 10s | Avg: 31m 10s | Max: 31m 10s
  🟩 GCC12              Pass: 100%/1   | Total: 33m 09s | Avg: 33m 09s | Max: 33m 09s
  🟩 GCC13              Pass: 100%/8   | Total:  2h 52m | Avg: 21m 36s | Max: 34m 11s
  🟩 MSVC14.29          Pass: 100%/2   | Total:  1h 53m | Avg: 56m 36s | Max: 59m 04s | Hits: 178%/3704  
  🟩 MSVC14.39          Pass: 100%/3   | Total:  2h 41m | Avg: 53m 47s | Max:  1h 06m | Hits: 240%/5556  
  🟩 NVHPC24.7          Pass: 100%/2   | Total:  1h 48m | Avg: 54m 29s | Max: 54m 31s
🟩 cxx_family
  🟩 Clang              Pass: 100%/14  | Total:  6h 18m | Avg: 27m 01s | Max: 33m 44s
  🟩 GCC                Pass: 100%/16  | Total:  7h 10m | Avg: 26m 54s | Max: 34m 11s
  🟩 MSVC               Pass: 100%/5   | Total:  4h 34m | Avg: 54m 54s | Max:  1h 06m | Hits: 215%/9260  
  🟩 NVHPC              Pass: 100%/2   | Total:  1h 48m | Avg: 54m 29s | Max: 54m 31s
🟩 gpu
  🟩 v100               Pass: 100%/37  | Total: 19h 52m | Avg: 32m 13s | Max:  1h 06m | Hits: 215%/9260  
🟩 jobs
  🟩 Build              Pass: 100%/31  | Total: 18h 12m | Avg: 35m 14s | Max:  1h 06m | Hits: 178%/7408  
  🟩 TestCPU            Pass: 100%/3   | Total: 51m 59s | Avg: 17m 19s | Max: 36m 43s | Hits: 365%/1852  
  🟩 TestGPU            Pass: 100%/3   | Total: 47m 45s | Avg: 15m 55s | Max: 22m 17s
🟩 sm
  🟩 90a                Pass: 100%/1   | Total: 18m 18s | Avg: 18m 18s | Max: 18m 18s
🟩 std
  🟩 17                 Pass: 100%/14  | Total:  8h 52m | Avg: 38m 02s | Max: 59m 04s | Hits: 178%/5556  
  🟩 20                 Pass: 100%/21  | Total: 10h 22m | Avg: 29m 38s | Max:  1h 06m | Hits: 272%/3704

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 13m 25s | Avg: 6m 42s | Max: 11m 15s

🟩 cpu
  🟩 amd64              Pass: 100%/2   | Total: 13m 25s | Avg:  6m 42s | Max: 11m 15s
🟩 ctk
  🟩 12.6               Pass: 100%/2   | Total: 13m 25s | Avg:  6m 42s | Max: 11m 15s
🟩 cudacxx
  🟩 nvcc12.6           Pass: 100%/2   | Total: 13m 25s | Avg:  6m 42s | Max: 11m 15s
🟩 cudacxx_family
  🟩 nvcc               Pass: 100%/2   | Total: 13m 25s | Avg:  6m 42s | Max: 11m 15s
🟩 cxx
  🟩 GCC13              Pass: 100%/2   | Total: 13m 25s | Avg:  6m 42s | Max: 11m 15s
🟩 cxx_family
  🟩 GCC                Pass: 100%/2   | Total: 13m 25s | Avg:  6m 42s | Max: 11m 15s
🟩 gpu
  🟩 v100               Pass: 100%/2   | Total: 13m 25s | Avg:  6m 42s | Max: 11m 15s
🟩 jobs
  🟩 Build              Pass: 100%/1   | Total:  2m 10s | Avg:  2m 10s | Max:  2m 10s
  🟩 Test               Pass: 100%/1   | Total: 11m 15s | Avg: 11m 15s | Max: 11m 15s

🟩 python: Pass: 100%/1 | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s

🟩 cpu
  🟩 amd64              Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s
🟩 ctk
  🟩 12.6               Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s
🟩 cudacxx
  🟩 nvcc12.6           Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s
🟩 cudacxx_family
  🟩 nvcc               Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s
🟩 cxx
  🟩 GCC13              Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s
🟩 cxx_family
  🟩 GCC                Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s
🟩 gpu
  🟩 v100               Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s
🟩 jobs
  🟩 Test               Pass: 100%/1   | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s

👃 Inspect Changes

Modifications in project?

	Project
	CCCL Infrastructure
	libcu++
+/-	CUB
	Thrust
	CUDA Experimental
	python
	CCCL C Parallel Library
	Catch2Helper

Modifications in project or dependencies?

	Project
	CCCL Infrastructure
	libcu++
+/-	CUB
+/-	Thrust
	CUDA Experimental
+/-	python
+/-	CCCL C Parallel Library
+/-	Catch2Helper

🏃‍ Runner counts (total jobs: 78)

#	Runner
53	`linux-amd64-cpu16`
11	`linux-amd64-gpu-v100-latest-1`
9	`windows-amd64-cpu16`
4	`linux-arm64-cpu16`
1	`linux-amd64-gpu-h100-latest-1-testing`

github-actions · 2025-01-14T04:47:52Z

🟨 CI finished in 3h 35m: Pass: 97%/78 | Total: 2d 04h | Avg: 40m 43s | Max: 1h 15m | Hits: 189%/12340

🟨 cub: Pass: 94%/38 | Total: 1d 08h | Avg: 51m 25s | Max: 1h 15m | Hits: 112%/3120

🔍 cpu: amd64 🔍
  🔍 amd64              Pass:  94%/36  | Total:  1d 06h | Avg: 51m 01s | Max:  1h 15m | Hits: 112%/3120  
  🟩 arm64              Pass: 100%/2   | Total:  1h 57m | Avg: 58m 42s | Max: 58m 43s
🔍 ctk: 12.6 🔍
  🟩 12.0               Pass: 100%/5   | Total:  4h 59m | Avg: 59m 48s | Max:  1h 09m | Hits: 113%/780   
  🟩 12.5               Pass: 100%/2   | Total:  2h 11m | Avg:  1h 05m | Max:  1h 05m
  🔍 12.6               Pass:  93%/31  | Total:  1d 01h | Avg: 49m 10s | Max:  1h 15m | Hits: 112%/2340  
🔍 cudacxx: nvcc12.6 🔍
  🟩 ClangCUDA18        Pass: 100%/2   | Total:  2h 05m | Avg:  1h 02m | Max:  1h 04m
  🟩 nvcc12.0           Pass: 100%/5   | Total:  4h 59m | Avg: 59m 48s | Max:  1h 09m | Hits: 113%/780   
  🟩 nvcc12.5           Pass: 100%/2   | Total:  2h 11m | Avg:  1h 05m | Max:  1h 05m
  🔍 nvcc12.6           Pass:  93%/29  | Total: 23h 18m | Avg: 48m 13s | Max:  1h 15m | Hits: 112%/2340  
🔍 cudacxx_family: nvcc 🔍
  🟩 ClangCUDA          Pass: 100%/2   | Total:  2h 05m | Avg:  1h 02m | Max:  1h 04m
  🔍 nvcc               Pass:  94%/36  | Total:  1d 06h | Avg: 50m 47s | Max:  1h 15m | Hits: 112%/3120  
🔍 gpu: v100 🔍
  🟩 h100               Pass: 100%/2   | Total: 43m 22s | Avg: 21m 41s | Max: 27m 18s
  🔍 v100               Pass:  94%/36  | Total:  1d 07h | Avg: 53m 05s | Max:  1h 15m | Hits: 112%/3120  
🚨 jobs: TestGPU 🚨
  🟩 Build              Pass: 100%/31  | Total:  1d 06h | Avg: 58m 56s | Max:  1h 15m | Hits: 112%/3120  
  🟩 DeviceLaunch       Pass: 100%/1   | Total: 19m 07s | Avg: 19m 07s | Max: 19m 07s
  🟩 GraphCapture       Pass: 100%/1   | Total: 16m 19s | Avg: 16m 19s | Max: 16m 19s
  🟩 HostLaunch         Pass: 100%/3   | Total: 53m 08s | Avg: 17m 42s | Max: 18m 44s
  🔥 TestGPU            Pass:   0%/2   | Total: 38m 26s | Avg: 19m 13s | Max: 20m 20s
🔍 std: 20 🔍
  🟩 17                 Pass: 100%/14  | Total: 14h 11m | Avg:  1h 00m | Max:  1h 12m | Hits: 113%/2340  
  🔍 20                 Pass:  91%/24  | Total: 18h 23m | Avg: 45m 57s | Max:  1h 15m | Hits: 110%/780   
🟨 cxx
  🟩 Clang14            Pass: 100%/4   | Total:  3h 57m | Avg: 59m 16s | Max:  1h 02m
  🟩 Clang15            Pass: 100%/1   | Total:  1h 02m | Avg:  1h 02m | Max:  1h 02m
  🟩 Clang16            Pass: 100%/1   | Total: 59m 45s | Avg: 59m 45s | Max: 59m 45s
  🟩 Clang17            Pass: 100%/1   | Total:  1h 00m | Avg:  1h 00m | Max:  1h 00m
  🟨 Clang18            Pass:  85%/7   | Total:  5h 37m | Avg: 48m 13s | Max:  1h 04m
  🟩 GCC7               Pass: 100%/2   | Total:  1h 58m | Avg: 59m 07s | Max: 59m 19s
  🟩 GCC8               Pass: 100%/1   | Total: 54m 32s | Avg: 54m 32s | Max: 54m 32s
  🟩 GCC9               Pass: 100%/2   | Total:  1h 53m | Avg: 56m 50s | Max: 58m 50s
  🟩 GCC10              Pass: 100%/1   | Total: 59m 10s | Avg: 59m 10s | Max: 59m 10s
  🟩 GCC11              Pass: 100%/1   | Total:  1h 01m | Avg:  1h 01m | Max:  1h 01m
  🟩 GCC12              Pass: 100%/3   | Total:  1h 44m | Avg: 34m 48s | Max:  1h 01m
  🟨 GCC13              Pass:  87%/8   | Total:  4h 29m | Avg: 33m 42s | Max: 58m 43s
  🟩 MSVC14.29          Pass: 100%/2   | Total:  2h 21m | Avg:  1h 10m | Max:  1h 12m | Hits: 113%/1560  
  🟩 MSVC14.39          Pass: 100%/2   | Total:  2h 23m | Avg:  1h 11m | Max:  1h 15m | Hits: 111%/1560  
  🟩 NVHPC24.7          Pass: 100%/2   | Total:  2h 11m | Avg:  1h 05m | Max:  1h 05m
🟨 cxx_family
  🟨 Clang              Pass:  92%/14  | Total: 12h 37m | Avg: 54m 05s | Max:  1h 04m
  🟨 GCC                Pass:  94%/18  | Total: 13h 00m | Avg: 43m 23s | Max:  1h 01m
  🟩 MSVC               Pass: 100%/4   | Total:  4h 45m | Avg:  1h 11m | Max:  1h 15m | Hits: 112%/3120  
  🟩 NVHPC              Pass: 100%/2   | Total:  2h 11m | Avg:  1h 05m | Max:  1h 05m
🟩 sm
  🟩 90                 Pass: 100%/2   | Total: 43m 22s | Avg: 21m 41s | Max: 27m 18s
  🟩 90a                Pass: 100%/1   | Total: 26m 18s | Avg: 26m 18s | Max: 26m 18s

🟩 thrust: Pass: 100%/37 | Total: 19h 46m | Avg: 32m 04s | Max: 1h 00m | Hits: 215%/9220

🟩 cmake_options
  🟩 -DTHRUST_DISPATCH_TYPE=Force32bit Pass: 100%/2   | Total: 39m 39s | Avg: 19m 49s | Max: 28m 04s
🟩 cpu
  🟩 amd64              Pass: 100%/35  | Total: 18h 49m | Avg: 32m 16s | Max:  1h 00m | Hits: 215%/9220  
  🟩 arm64              Pass: 100%/2   | Total: 57m 05s | Avg: 28m 32s | Max: 29m 54s
🟩 ctk
  🟩 12.0               Pass: 100%/5   | Total:  3h 06m | Avg: 37m 17s | Max: 56m 27s | Hits: 177%/1844  
  🟩 12.5               Pass: 100%/2   | Total:  1h 45m | Avg: 52m 54s | Max: 54m 46s
  🟩 12.6               Pass: 100%/30  | Total: 14h 54m | Avg: 29m 49s | Max:  1h 00m | Hits: 224%/7376  
🟩 cudacxx
  🟩 ClangCUDA18        Pass: 100%/2   | Total: 53m 30s | Avg: 26m 45s | Max: 27m 25s
  🟩 nvcc12.0           Pass: 100%/5   | Total:  3h 06m | Avg: 37m 17s | Max: 56m 27s | Hits: 177%/1844  
  🟩 nvcc12.5           Pass: 100%/2   | Total:  1h 45m | Avg: 52m 54s | Max: 54m 46s
  🟩 nvcc12.6           Pass: 100%/28  | Total: 14h 01m | Avg: 30m 02s | Max:  1h 00m | Hits: 224%/7376  
🟩 cudacxx_family
  🟩 ClangCUDA          Pass: 100%/2   | Total: 53m 30s | Avg: 26m 45s | Max: 27m 25s
  🟩 nvcc               Pass: 100%/35  | Total: 18h 53m | Avg: 32m 22s | Max:  1h 00m | Hits: 215%/9220  
🟩 cxx
  🟩 Clang14            Pass: 100%/4   | Total:  2h 04m | Avg: 31m 05s | Max: 33m 12s
  🟩 Clang15            Pass: 100%/1   | Total: 32m 17s | Avg: 32m 17s | Max: 32m 17s
  🟩 Clang16            Pass: 100%/1   | Total: 30m 09s | Avg: 30m 09s | Max: 30m 09s
  🟩 Clang17            Pass: 100%/1   | Total: 32m 42s | Avg: 32m 42s | Max: 32m 42s
  🟩 Clang18            Pass: 100%/7   | Total:  2h 43m | Avg: 23m 17s | Max: 31m 24s
  🟩 GCC7               Pass: 100%/2   | Total:  1h 02m | Avg: 31m 07s | Max: 31m 30s
  🟩 GCC8               Pass: 100%/1   | Total: 31m 25s | Avg: 31m 25s | Max: 31m 25s
  🟩 GCC9               Pass: 100%/2   | Total:  1h 05m | Avg: 32m 52s | Max: 33m 27s
  🟩 GCC10              Pass: 100%/1   | Total: 34m 35s | Avg: 34m 35s | Max: 34m 35s
  🟩 GCC11              Pass: 100%/1   | Total: 32m 23s | Avg: 32m 23s | Max: 32m 23s
  🟩 GCC12              Pass: 100%/1   | Total: 33m 45s | Avg: 33m 45s | Max: 33m 45s
  🟩 GCC13              Pass: 100%/8   | Total:  2h 54m | Avg: 21m 46s | Max: 36m 34s
  🟩 MSVC14.29          Pass: 100%/2   | Total:  1h 51m | Avg: 55m 37s | Max: 56m 27s | Hits: 177%/3688  
  🟩 MSVC14.39          Pass: 100%/3   | Total:  2h 32m | Avg: 50m 58s | Max:  1h 00m | Hits: 240%/5532  
  🟩 NVHPC24.7          Pass: 100%/2   | Total:  1h 45m | Avg: 52m 54s | Max: 54m 46s
🟩 cxx_family
  🟩 Clang              Pass: 100%/14  | Total:  6h 22m | Avg: 27m 19s | Max: 33m 12s
  🟩 GCC                Pass: 100%/16  | Total:  7h 14m | Avg: 27m 08s | Max: 36m 34s
  🟩 MSVC               Pass: 100%/5   | Total:  4h 24m | Avg: 52m 50s | Max:  1h 00m | Hits: 215%/9220  
  🟩 NVHPC              Pass: 100%/2   | Total:  1h 45m | Avg: 52m 54s | Max: 54m 46s
🟩 gpu
  🟩 v100               Pass: 100%/37  | Total: 19h 46m | Avg: 32m 04s | Max:  1h 00m | Hits: 215%/9220  
🟩 jobs
  🟩 Build              Pass: 100%/31  | Total: 18h 16m | Avg: 35m 22s | Max:  1h 00m | Hits: 177%/7376  
  🟩 TestCPU            Pass: 100%/3   | Total: 50m 58s | Avg: 16m 59s | Max: 36m 20s | Hits: 365%/1844  
  🟩 TestGPU            Pass: 100%/3   | Total: 39m 05s | Avg: 13m 01s | Max: 14m 28s
🟩 sm
  🟩 90a                Pass: 100%/1   | Total: 17m 38s | Avg: 17m 38s | Max: 17m 38s
🟩 std
  🟩 17                 Pass: 100%/14  | Total:  8h 49m | Avg: 37m 48s | Max: 56m 27s | Hits: 177%/5532  
  🟩 20                 Pass: 100%/21  | Total: 10h 17m | Avg: 29m 25s | Max:  1h 00m | Hits: 271%/3688

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 10m 00s | Avg: 5m 00s | Max: 7m 49s

🟩 cpu
  🟩 amd64              Pass: 100%/2   | Total: 10m 00s | Avg:  5m 00s | Max:  7m 49s
🟩 ctk
  🟩 12.6               Pass: 100%/2   | Total: 10m 00s | Avg:  5m 00s | Max:  7m 49s
🟩 cudacxx
  🟩 nvcc12.6           Pass: 100%/2   | Total: 10m 00s | Avg:  5m 00s | Max:  7m 49s
🟩 cudacxx_family
  🟩 nvcc               Pass: 100%/2   | Total: 10m 00s | Avg:  5m 00s | Max:  7m 49s
🟩 cxx
  🟩 GCC13              Pass: 100%/2   | Total: 10m 00s | Avg:  5m 00s | Max:  7m 49s
🟩 cxx_family
  🟩 GCC                Pass: 100%/2   | Total: 10m 00s | Avg:  5m 00s | Max:  7m 49s
🟩 gpu
  🟩 v100               Pass: 100%/2   | Total: 10m 00s | Avg:  5m 00s | Max:  7m 49s
🟩 jobs
  🟩 Build              Pass: 100%/1   | Total:  2m 11s | Avg:  2m 11s | Max:  2m 11s
  🟩 Test               Pass: 100%/1   | Total:  7m 49s | Avg:  7m 49s | Max:  7m 49s

🟩 python: Pass: 100%/1 | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s

🟩 cpu
  🟩 amd64              Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s
🟩 ctk
  🟩 12.6               Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s
🟩 cudacxx
  🟩 nvcc12.6           Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s
🟩 cudacxx_family
  🟩 nvcc               Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s
🟩 cxx
  🟩 GCC13              Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s
🟩 cxx_family
  🟩 GCC                Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s
🟩 gpu
  🟩 v100               Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s
🟩 jobs
  🟩 Test               Pass: 100%/1   | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s

👃 Inspect Changes

Modifications in project?

	Project
	CCCL Infrastructure
	libcu++
+/-	CUB
+/-	Thrust
	CUDA Experimental
	python
	CCCL C Parallel Library
	Catch2Helper

Modifications in project or dependencies?

	Project
	CCCL Infrastructure
	libcu++
+/-	CUB
+/-	Thrust
	CUDA Experimental
+/-	python
+/-	CCCL C Parallel Library
+/-	Catch2Helper

🏃‍ Runner counts (total jobs: 78)

#	Runner
53	`linux-amd64-cpu16`
11	`linux-amd64-gpu-v100-latest-1`
9	`windows-amd64-cpu16`
4	`linux-arm64-cpu16`
1	`linux-amd64-gpu-h100-latest-1-testing`

copy-pr-bot · 2025-01-14T18:21:37Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

Signed-off-by: fbusato <[email protected]>

github-actions · 2025-01-14T21:39:42Z

🟩 CI finished in 2h 36m: Pass: 100%/78 | Total: 2d 05h | Avg: 41m 26s | Max: 1h 16m | Hits: 187%/12340

🟩 cub: Pass: 100%/38 | Total: 1d 08h | Avg: 51m 40s | Max: 1h 16m | Hits: 110%/3120

🟩 cpu
  🟩 amd64              Pass: 100%/36  | Total:  1d 06h | Avg: 51m 14s | Max:  1h 16m | Hits: 110%/3120  
  🟩 arm64              Pass: 100%/2   | Total:  1h 58m | Avg: 59m 16s | Max:  1h 00m
🟩 ctk
  🟩 12.0               Pass: 100%/5   | Total:  4h 53m | Avg: 58m 46s | Max:  1h 02m | Hits: 110%/780   
  🟩 12.5               Pass: 100%/2   | Total:  2h 14m | Avg:  1h 07m | Max:  1h 08m
  🟩 12.6               Pass: 100%/31  | Total:  1d 01h | Avg: 49m 30s | Max:  1h 16m | Hits: 110%/2340  
🟩 cudacxx
  🟩 ClangCUDA18        Pass: 100%/2   | Total:  1h 55m | Avg: 57m 48s | Max: 58m 44s
  🟩 nvcc12.0           Pass: 100%/5   | Total:  4h 53m | Avg: 58m 46s | Max:  1h 02m | Hits: 110%/780   
  🟩 nvcc12.5           Pass: 100%/2   | Total:  2h 14m | Avg:  1h 07m | Max:  1h 08m
  🟩 nvcc12.6           Pass: 100%/29  | Total: 23h 39m | Avg: 48m 56s | Max:  1h 16m | Hits: 110%/2340  
🟩 cudacxx_family
  🟩 ClangCUDA          Pass: 100%/2   | Total:  1h 55m | Avg: 57m 48s | Max: 58m 44s
  🟩 nvcc               Pass: 100%/36  | Total:  1d 06h | Avg: 51m 19s | Max:  1h 16m | Hits: 110%/3120  
🟩 cxx
  🟩 Clang14            Pass: 100%/4   | Total:  3h 48m | Avg: 57m 14s | Max: 58m 46s
  🟩 Clang15            Pass: 100%/1   | Total: 59m 38s | Avg: 59m 38s | Max: 59m 38s
  🟩 Clang16            Pass: 100%/1   | Total: 55m 20s | Avg: 55m 20s | Max: 55m 20s
  🟩 Clang17            Pass: 100%/1   | Total: 55m 59s | Avg: 55m 59s | Max: 55m 59s
  🟩 Clang18            Pass: 100%/7   | Total:  5h 52m | Avg: 50m 24s | Max:  1h 02m
  🟩 GCC7               Pass: 100%/2   | Total:  1h 58m | Avg: 59m 03s | Max: 59m 32s
  🟩 GCC8               Pass: 100%/1   | Total: 54m 01s | Avg: 54m 01s | Max: 54m 01s
  🟩 GCC9               Pass: 100%/2   | Total:  1h 54m | Avg: 57m 17s | Max: 58m 36s
  🟩 GCC10              Pass: 100%/1   | Total:  1h 04m | Avg:  1h 04m | Max:  1h 04m
  🟩 GCC11              Pass: 100%/1   | Total: 55m 17s | Avg: 55m 17s | Max: 55m 17s
  🟩 GCC12              Pass: 100%/3   | Total:  1h 48m | Avg: 36m 09s | Max:  1h 01m
  🟩 GCC13              Pass: 100%/8   | Total:  4h 46m | Avg: 35m 46s | Max:  1h 02m
  🟩 MSVC14.29          Pass: 100%/2   | Total:  2h 11m | Avg:  1h 05m | Max:  1h 09m | Hits: 112%/1560  
  🟩 MSVC14.39          Pass: 100%/2   | Total:  2h 24m | Avg:  1h 12m | Max:  1h 16m | Hits: 108%/1560  
  🟩 NVHPC24.7          Pass: 100%/2   | Total:  2h 14m | Avg:  1h 07m | Max:  1h 08m
🟩 cxx_family
  🟩 Clang              Pass: 100%/14  | Total: 12h 32m | Avg: 53m 46s | Max:  1h 02m
  🟩 GCC                Pass: 100%/18  | Total: 13h 20m | Avg: 44m 28s | Max:  1h 04m
  🟩 MSVC               Pass: 100%/4   | Total:  4h 35m | Avg:  1h 08m | Max:  1h 16m | Hits: 110%/3120  
  🟩 NVHPC              Pass: 100%/2   | Total:  2h 14m | Avg:  1h 07m | Max:  1h 08m
🟩 gpu
  🟩 h100               Pass: 100%/2   | Total: 46m 32s | Avg: 23m 16s | Max: 26m 55s
  🟩 v100               Pass: 100%/36  | Total:  1d 07h | Avg: 53m 14s | Max:  1h 16m | Hits: 110%/3120  
🟩 jobs
  🟩 Build              Pass: 100%/31  | Total:  1d 06h | Avg: 58m 06s | Max:  1h 16m | Hits: 110%/3120  
  🟩 DeviceLaunch       Pass: 100%/1   | Total: 19m 58s | Avg: 19m 58s | Max: 19m 58s
  🟩 GraphCapture       Pass: 100%/1   | Total: 16m 12s | Avg: 16m 12s | Max: 16m 12s
  🟩 HostLaunch         Pass: 100%/3   | Total:  1h 14m | Avg: 24m 47s | Max: 33m 03s
  🟩 TestGPU            Pass: 100%/2   | Total: 51m 34s | Avg: 25m 47s | Max: 28m 18s
🟩 sm
  🟩 90                 Pass: 100%/2   | Total: 46m 32s | Avg: 23m 16s | Max: 26m 55s
  🟩 90a                Pass: 100%/1   | Total: 26m 52s | Avg: 26m 52s | Max: 26m 52s
🟩 std
  🟩 17                 Pass: 100%/14  | Total: 14h 01m | Avg:  1h 00m | Max:  1h 09m | Hits: 111%/2340  
  🟩 20                 Pass: 100%/24  | Total: 18h 41m | Avg: 46m 44s | Max:  1h 16m | Hits: 107%/780

🟩 thrust: Pass: 100%/37 | Total: 20h 30m | Avg: 33m 15s | Max: 1h 06m | Hits: 213%/9220

🟩 cmake_options
  🟩 -DTHRUST_DISPATCH_TYPE=Force32bit Pass: 100%/2   | Total: 37m 49s | Avg: 18m 54s | Max: 24m 49s
🟩 cpu
  🟩 amd64              Pass: 100%/35  | Total: 19h 32m | Avg: 33m 30s | Max:  1h 06m | Hits: 213%/9220  
  🟩 arm64              Pass: 100%/2   | Total: 58m 09s | Avg: 29m 04s | Max: 30m 56s
🟩 ctk
  🟩 12.0               Pass: 100%/5   | Total:  3h 04m | Avg: 36m 59s | Max:  1h 01m | Hits: 174%/1844  
  🟩 12.5               Pass: 100%/2   | Total:  1h 53m | Avg: 56m 59s | Max: 59m 33s
  🟩 12.6               Pass: 100%/30  | Total: 15h 31m | Avg: 31m 03s | Max:  1h 06m | Hits: 222%/7376  
🟩 cudacxx
  🟩 ClangCUDA18        Pass: 100%/2   | Total: 56m 09s | Avg: 28m 04s | Max: 29m 00s
  🟩 nvcc12.0           Pass: 100%/5   | Total:  3h 04m | Avg: 36m 59s | Max:  1h 01m | Hits: 174%/1844  
  🟩 nvcc12.5           Pass: 100%/2   | Total:  1h 53m | Avg: 56m 59s | Max: 59m 33s
  🟩 nvcc12.6           Pass: 100%/28  | Total: 14h 35m | Avg: 31m 16s | Max:  1h 06m | Hits: 222%/7376  
🟩 cudacxx_family
  🟩 ClangCUDA          Pass: 100%/2   | Total: 56m 09s | Avg: 28m 04s | Max: 29m 00s
  🟩 nvcc               Pass: 100%/35  | Total: 19h 34m | Avg: 33m 33s | Max:  1h 06m | Hits: 213%/9220  
🟩 cxx
  🟩 Clang14            Pass: 100%/4   | Total:  2h 07m | Avg: 31m 50s | Max: 33m 06s
  🟩 Clang15            Pass: 100%/1   | Total: 30m 52s | Avg: 30m 52s | Max: 30m 52s
  🟩 Clang16            Pass: 100%/1   | Total: 30m 47s | Avg: 30m 47s | Max: 30m 47s
  🟩 Clang17            Pass: 100%/1   | Total: 30m 15s | Avg: 30m 15s | Max: 30m 15s
  🟩 Clang18            Pass: 100%/7   | Total:  3h 05m | Avg: 26m 27s | Max: 33m 28s
  🟩 GCC7               Pass: 100%/2   | Total:  1h 05m | Avg: 32m 37s | Max: 34m 07s
  🟩 GCC8               Pass: 100%/1   | Total: 31m 08s | Avg: 31m 08s | Max: 31m 08s
  🟩 GCC9               Pass: 100%/2   | Total:  1h 04m | Avg: 32m 09s | Max: 34m 13s
  🟩 GCC10              Pass: 100%/1   | Total: 31m 29s | Avg: 31m 29s | Max: 31m 29s
  🟩 GCC11              Pass: 100%/1   | Total: 34m 36s | Avg: 34m 36s | Max: 34m 36s
  🟩 GCC12              Pass: 100%/1   | Total: 33m 17s | Avg: 33m 17s | Max: 33m 17s
  🟩 GCC13              Pass: 100%/8   | Total:  2h 52m | Avg: 21m 31s | Max: 32m 44s
  🟩 MSVC14.29          Pass: 100%/2   | Total:  2h 03m | Avg:  1h 01m | Max:  1h 01m | Hits: 175%/3688  
  🟩 MSVC14.39          Pass: 100%/3   | Total:  2h 36m | Avg: 52m 18s | Max:  1h 06m | Hits: 238%/5532  
  🟩 NVHPC24.7          Pass: 100%/2   | Total:  1h 53m | Avg: 56m 59s | Max: 59m 33s
🟩 cxx_family
  🟩 Clang              Pass: 100%/14  | Total:  6h 44m | Avg: 28m 53s | Max: 33m 28s
  🟩 GCC                Pass: 100%/16  | Total:  7h 12m | Avg: 27m 00s | Max: 34m 36s
  🟩 MSVC               Pass: 100%/5   | Total:  4h 40m | Avg: 56m 01s | Max:  1h 06m | Hits: 213%/9220  
  🟩 NVHPC              Pass: 100%/2   | Total:  1h 53m | Avg: 56m 59s | Max: 59m 33s
🟩 gpu
  🟩 v100               Pass: 100%/37  | Total: 20h 30m | Avg: 33m 15s | Max:  1h 06m | Hits: 213%/9220  
🟩 jobs
  🟩 Build              Pass: 100%/31  | Total: 18h 42m | Avg: 36m 11s | Max:  1h 06m | Hits: 175%/7376  
  🟩 TestCPU            Pass: 100%/3   | Total: 49m 37s | Avg: 16m 32s | Max: 34m 58s | Hits: 365%/1844  
  🟩 TestGPU            Pass: 100%/3   | Total: 59m 01s | Avg: 19m 40s | Max: 33m 28s
🟩 sm
  🟩 90a                Pass: 100%/1   | Total: 18m 50s | Avg: 18m 50s | Max: 18m 50s
🟩 std
  🟩 17                 Pass: 100%/14  | Total:  9h 14m | Avg: 39m 37s | Max:  1h 01m | Hits: 175%/5532  
  🟩 20                 Pass: 100%/21  | Total: 10h 38m | Avg: 30m 23s | Max:  1h 06m | Hits: 269%/3688

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 9m 25s | Avg: 4m 42s | Max: 7m 16s

🟩 cpu
  🟩 amd64              Pass: 100%/2   | Total:  9m 25s | Avg:  4m 42s | Max:  7m 16s
🟩 ctk
  🟩 12.6               Pass: 100%/2   | Total:  9m 25s | Avg:  4m 42s | Max:  7m 16s
🟩 cudacxx
  🟩 nvcc12.6           Pass: 100%/2   | Total:  9m 25s | Avg:  4m 42s | Max:  7m 16s
🟩 cudacxx_family
  🟩 nvcc               Pass: 100%/2   | Total:  9m 25s | Avg:  4m 42s | Max:  7m 16s
🟩 cxx
  🟩 GCC13              Pass: 100%/2   | Total:  9m 25s | Avg:  4m 42s | Max:  7m 16s
🟩 cxx_family
  🟩 GCC                Pass: 100%/2   | Total:  9m 25s | Avg:  4m 42s | Max:  7m 16s
🟩 gpu
  🟩 v100               Pass: 100%/2   | Total:  9m 25s | Avg:  4m 42s | Max:  7m 16s
🟩 jobs
  🟩 Build              Pass: 100%/1   | Total:  2m 09s | Avg:  2m 09s | Max:  2m 09s
  🟩 Test               Pass: 100%/1   | Total:  7m 16s | Avg:  7m 16s | Max:  7m 16s

🟩 python: Pass: 100%/1 | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s

🟩 cpu
  🟩 amd64              Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s
🟩 ctk
  🟩 12.6               Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s
🟩 cudacxx
  🟩 nvcc12.6           Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s
🟩 cudacxx_family
  🟩 nvcc               Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s
🟩 cxx
  🟩 GCC13              Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s
🟩 cxx_family
  🟩 GCC                Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s
🟩 gpu
  🟩 v100               Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s
🟩 jobs
  🟩 Test               Pass: 100%/1   | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s

👃 Inspect Changes

Modifications in project?

	Project
	CCCL Infrastructure
	libcu++
+/-	CUB
+/-	Thrust
	CUDA Experimental
	python
	CCCL C Parallel Library
	Catch2Helper

Modifications in project or dependencies?

	Project
	CCCL Infrastructure
	libcu++
+/-	CUB
+/-	Thrust
	CUDA Experimental
+/-	python
+/-	CCCL C Parallel Library
+/-	Catch2Helper

🏃‍ Runner counts (total jobs: 78)

#	Runner
53	`linux-amd64-cpu16`
11	`linux-amd64-gpu-v100-latest-1`
9	`windows-amd64-cpu16`
4	`linux-arm64-cpu16`
1	`linux-amd64-gpu-h100-latest-1-testing`

github-actions · 2025-01-15T00:39:35Z

Git push to origin failed for branch/2.8.x with exitcode 128

Co-authored-by: Federico Busato <[email protected]>

implement `add_sat` split `signed`/`unsigned` implementation, improve implementation for MSVC improve device `add_sat` implementation add `add_sat` test improve generic `add_sat` implementation for signed types implement `sub_sat` allow more msvc intrinsics on x86 add op tests partially implement `mul_sat` implement `div_sat` and `saturate_cast` add `saturate_cast` test simplify `div_sat` test Deprectate C++11 and C++14 for libcu++ (#3173) * Deprectate C++11 and C++14 for libcu++ Co-authored-by: Bernhard Manfred Gruber <[email protected]> Implement `abs` and `div` from `cstdlib` (#3153) * implement integer abs functions * improve tests, fix constexpr support * just use the our implementation * implement `cuda::std::div` * prefer host's `div_t` like types * provide `cuda::std::abs` overloads for floats * allow fp abs for NVRTC * silence msvc's warning about conversion from floating point to integral Fix missing radix sort policies (#3174) Fixes NVBug 5009941 Introduces new `DeviceReduce::Arg{Min,Max}` interface with two output iterators (#3148) * introduces new arg{min,max} interface with two output iterators * adds fp inf tests * fixes docs * improves code example * fixes exec space specifier * trying to fix deprecation warning for more compilers * inlines unzip operator * trying to fix deprecation warning for nvhpc * integrates supression fixes in diagnostics * pre-ctk 11.5 deprecation suppression * fixes icc * fix for pre-ctk11.5 * cleans up deprecation suppression * cleanup Extend tuning documentation (#3179) Add codespell pre-commit hook, fix typos in CCCL (#3168) * Add codespell pre-commit hook * Automatic changes from codespell. * Manual changes. Fix parameter space for TUNE_LOAD in scan benchmark (#3176) fix various old compiler checks (#3178) implement C++26 `std::projected` (#3175) Fix pre-commit config for codespell and remaining typos (#3182) Massive cleanup of our config (#3155) Fix UB in atomics with automatic storage (#2586) * Adds specialized local cuda atomics and injects them into most atomics paths. Co-authored-by: Georgy Evtushenko <[email protected]> Co-authored-by: gonzalobg <[email protected]> * Allow CUDA 12.2 to keep perf, this addresses earlier comments in #478 * Remove extraneous double brackets in unformatted code. * Merge unsafe atomic logic into `__cuda_is_local`. * Use `const_cast` for type conversions in cuda_local.h * Fix build issues from interface changes * Fix missing __nanosleep on sm70- * Guard __isLocal from NVHPC * Use PTX instead of running nothing from NVHPC * fixup /s/nvrtc/nvhpc * Fixup missing CUDA ifdef surrounding device code * Fix codegen * Bypass some sort of compiler bug on GCC7 * Apply suggestions from code review * Use unsafe automatic storage atomics in codegen tests --------- Co-authored-by: Georgy Evtushenko <[email protected]> Co-authored-by: gonzalobg <[email protected]> Co-authored-by: Michael Schellenberger Costa <[email protected]> Refactor the source code layout for `cuda.parallel` (#3177) * Refactor the source layout for cuda.parallel * Add copyright * Address review feedback * Don't import anything into `experimental` namespace * fix import --------- Co-authored-by: Ashwin Srinath <[email protected]> new type-erased memory resources (#2824) s/_LIBCUDACXX_DECLSPEC_EMPTY_BASES/_CCCL_DECLSPEC_EMPTY_BASES/g (#3186) Document address stability of `thrust::transform` (#3181) * Do not document _LIBCUDACXX_MARK_CAN_COPY_ARGUMENTS * Reformat and fix UnaryFunction/BinaryFunction in transform docs * Mention transform can use proclaim_copyable_arguments * Document cuda::proclaims_copyable_arguments better * Deprecate depending on transform functor argument addresses Fixes: #3053 turn off cuda version check for clangd (#3194) [STF] jacobi example based on parallel_for (#3187) * Simple jacobi example with parallel for and reductions * clang-format * remove useless capture list fixes pre-nv_diag suppression issues (#3189) Prefer c2h::type_name over c2h::demangle (#3195) Fix memcpy_async* tests (#3197) * memcpy_async_tx: Fix bug in test Two bugs, one of which occurs in practice: 1. There is a missing fence.proxy.space::global between the writes to global memory and the memcpy_async_tx. (Occurs in practice) 2. The end of the kernel should be fenced with `__syncthreads()`, because the barrier is invalidated in the destructor. If other threads are still waiting on it, there will be UB. (Has not yet manifested itself) * cp_async_bulk_tensor: Pre-emptively fence more in test Add type annotations and mypy checks for `cuda.parallel` (#3180) * Refactor the source layout for cuda.parallel * Add initial type annotations * Update pre-commit config * More typing * Fix bad merge * Fix TYPE_CHECKING and numpy annotations * typing bindings.py correctly * Address review feedback --------- Co-authored-by: Ashwin Srinath <[email protected]> Fix rendering of cuda.parallel docs (#3192) * Fix pre-commit config for codespell and remaining typos * Fix rendering of docs for cuda.parallel --------- Co-authored-by: Ashwin Srinath <[email protected]> Enable PDL for DeviceMergeSortBlockSortKernel (#3199) The kernel already contains a call to _CCCL_PDL_GRID_DEPENDENCY_SYNC. This commit enables PDL when launching the kernel. Adds support for large `num_items` to `DeviceReduce::{ArgMin,ArgMax}` (#2647) * adds benchmarks for reduce::arg{min,max} * preliminary streaming arg-extremum reduction * fixes implicit conversion * uses streaming dispatch class * changes arg benches to use new streaming reduce * streaming arg-extrema reduction * fixes style * fixes compilation failures * cleanups * adds rst style comments * declare vars const and use clamp * consolidates argmin argmax benchmarks * fixes thrust usage * drops offset type in arg-extrema benchmarks * fixes clang cuda * exec space macros * switch to signed global offset type for slightly better perf * clarifies documentation * applies minor benchmark style changes from review comments * fixes interface documentation and comments * list-init accumulating output op * improves style, comments, and tests * cleans up aggregate init * renames dispatch class usage in benchmarks * fixes merge conflicts * addresses review comments * addresses review comments * fixes assertion * removes superseded implementation * changes large problem tests to use new interface * removes obsolete tests for deprecated interface Fixes for Python 3.7 docs environment (#3206) Co-authored-by: Ashwin Srinath <[email protected]> Adds support for large number of items to `DeviceTransform` (#3172) * moves large problem test helper to common file * adds support for large num items to device transform * adds tests for large number of items to device interface * fixes format * addresses review comments cp_async_bulk: Fix test (#3198) * memcpy_async_tx: Fix bug in test Two bugs, one of which occurs in practice: 1. There is a missing fence.proxy.space::global between the writes to global memory and the memcpy_async_tx. (Occurs in practice) 2. The end of the kernel should be fenced with `__syncthreads()`, because the barrier is invalidated in the destructor. If other threads are still waiting on it, there will be UB. (Has not yet manifested itself) * cp_async_bulk_tensor: Pre-emptively fence more in test * cp_async_bulk: Fix test The global memory pointer could be misaligned. cudax fixes for msvc 14.41 (#3200) avoid instantiating class templates in `is_same` implementation when possible (#3203) Fix: make launchers a CUB detail; make kernel source functions hidden. (#3209) * Fix: make launchers a CUB detail; make kernel source functions hidden. * [pre-commit.ci] auto code formatting * Address review comments, fix which macro gets fixed. help the ranges concepts recognize standard contiguous iterators in c++14/17 (#3202) unify macros and cmake options that control the suppression of deprecation warnings (#3220) * unify macros and cmake options that control the suppression of deprecation warnings * suppress nvcc warning #186 in thrust header tests * suppress c++ dialect deprecation warnings in libcudacxx header tests Fx thread-reduce performance regression (#3225) cuda.parallel: In-memory caching of build objects (#3216) * Define __eq__ and __hash__ for Iterators * Define cache_with_key utility and use it to cache Reduce objects * Add tests for caching Reduce objects * Tighten up types * Updates to support 3.7 * Address review feedback * Introduce IteratorKind to hold iterator type information * Use the .kind to generate an abi_name * Remove __eq__ and __hash__ methods from IteratorBase * Move helper function * Formatting * Don't unpack tuple in cache key --------- Co-authored-by: Ashwin Srinath <[email protected]> Just enough ranges for c++14 `span` (#3211) use generalized concepts portability macros to simplify the `range` concept (#3217) fixes some issues in the concepts portability macros and then re-implements the `range` concept with `_CCCL_REQUIRES_EXPR` Use Ruff to sort imports (#3230) * Update pyproject.tomls for import sorting * Update files after running pre-commit * Move ruff config to pyproject.toml --------- Co-authored-by: Ashwin Srinath <[email protected]> fix tuning_scan sm90 config issue (#3236) Co-authored-by: Shijie Chen <[email protected]> [STF] Logical token (#3196) * Split the implementation of the void interface into the definition of the interface, and its implementations on streams and graphs. * Add missing files * Check if a task implementation can match a prototype where the void_interface arguments are ignored * Implement ctx.abstract_logical_data() which relies on a void data interface * Illustrate how to use abstract handles in local contexts * Introduce an is_void_interface() virtual method in the data interface to potentially optimize some stages * Small improvements in the examples * Do not try to allocate or move void data * Do not use I as a variable * fix linkage error * rename abtract_logical_data into logical_token * Document logical token * fix spelling error * fix sphinx error * reflect name changes * use meaningful variable names * simplify logical_token implementation because writeback is already disabled * add a unit test for token elision * implement token elision in host_launch * Remove unused type * Implement helpers to check if a function can be invoked from a tuple, or from a tuple where we removed tokens * Much simpler is_tuple_invocable_with_filtered implementation * Fix buggy test * Factorize code * Document that we can ignore tokens for task and host_launch * Documentation for logical data freeze Fix ReduceByKey tuning (#3240) Fix RLE tuning (#3239) cuda.parallel: Forbid non-contiguous arrays as inputs (or outputs) (#3233) * Forbid non-contiguous arrays as inputs (or outputs) * Implement a more robust way to check for contiguity * Don't bother if cublas unavailable * Fix how we check for zero-element arrays * sort imports --------- Co-authored-by: Ashwin Srinath <[email protected]> expands support for more offset types in segmented benchmark (#3231) Add escape hatches to the cmake configuration of the header tests so that we can tests deprecated compilers / dialects (#3253) * Add escape hatches to the cmake configuration of the header tests so that we can tests deprecated compilers / dialects * Do not add option twice ptx: Add add_instruction.py (#3190) This file helps create the necessary structure for new PTX instructions. Co-authored-by: Allard Hendriksen <[email protected]> Bump main to 2.9.0. (#3247) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Drop cub::Mutex (#3251) Fixes: #3250 Remove legacy macros from CUB util_arch.cuh (#3257) Fixes: #3256 Remove thrust::[unary|binary]_traits (#3260) Fixes: #3259 Architecture and OS identification macros (#3237) Bump main to 3.0.0. (#3265) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Drop thrust not1 and not2 (#3264) Fixes: #3263 CCCL Internal macro documentation (#3238) Deprecate GridBarrier and GridBarrierLifetime (#3258) Fixes: #1389 Require at least gcc7 (#3268) Fixes: #3267 Drop thrust::[unary|binary]_function (#3274) Fixes: #3273 Drop ICC from CI (#3277) [STF] Corruption of the capture list of an extended lambda with a parallel_for construct on a host execution place (#3270) * Add a test to reproduce a bug observed with parallel_for on a host place * clang-format * use _CCCL_ASSERT * Attempt to debug * do not create a tuple with a universal reference that is out of scope when we use it, use an lvalue instead * fix lambda expression * clang-format Enable thrust::identity test for non-MSVC (#3281) This seems to be an oversight when the test was added Co-authored-by: Michael Schellenberger Costa <[email protected]> Enable PDL in triple chevron launch (#3282) It seems PDL was disabled by accident when _THRUST_HAS_PDL was renamed to _CCCL_HAS_PDL during the review introducing the feature. Disambiguate line continuations and macro continuations in <nv/target> (#3244) Drop VS 2017 from CI (#3287) Fixes: #3286 Drop ICC support in code (#3279) * Drop ICC from code Fixes: #3278 Co-authored-by: Michael Schellenberger Costa <[email protected]> Make CUB NVRTC commandline arguments come from a cmake template (#3292) Propose the same components (thrust, cub, libc++, cudax, cuda.parallel,...) in the bug report template than in the feature request template (#3295) Use process isolation instead of default hyper-v for Windows. (#3294) Try improving build times by using process isolation instead of hyper-v Co-authored-by: Michael Schellenberger Costa <[email protected]> [pre-commit.ci] pre-commit autoupdate (#3248) * [pre-commit.ci] pre-commit autoupdate updates: - [github.com/pre-commit/mirrors-clang-format: v18.1.8 → v19.1.6](https://github.com/pre-commit/mirrors-clang-format/compare/v18.1.8...v19.1.6) - [github.com/astral-sh/ruff-pre-commit: v0.8.3 → v0.8.6](https://github.com/astral-sh/ruff-pre-commit/compare/v0.8.3...v0.8.6) - [github.com/pre-commit/mirrors-mypy: v1.13.0 → v1.14.1](https://github.com/pre-commit/mirrors-mypy/compare/v1.13.0...v1.14.1) Co-authored-by: Michael Schellenberger Costa <[email protected]> Drop Thrust legacy arch macros (#3298) Which were disabled and could be re-enabled using THRUST_PROVIDE_LEGACY_ARCH_MACROS Drop Thrust's compiler_fence.h (#3300) Drop CTK 11.x from CI (#3275) * Add cuda12.0-gcc7 devcontainer * Move MSVC2017 jobs to CTK 12.6 Those is the only combination where rapidsai has devcontainers * Add /Zc:__cplusplus for the libcudacxx tests * Only add excape hatch for affected CTKs * Workaround missing cudaLaunchKernelEx on MSVC cudaLaunchKernelEx requires C++11, but unfortunately <cuda_runtime.h> checks this using the __cplusplus macro, which is reported wrongly for MSVC. CTK 12.3 fixed this by additionally detecting _MSV_VER. As a workaround, we provide our own copy of cudaLaunchKernelEx when it is not available from the CTK. * Workaround nvcc+MSVC issue * Regenerate devcontainers Fixes: #3249 Co-authored-by: Michael Schellenberger Costa <[email protected]> Drop CUB's util_compiler.cuh (#3302) All contained macros were deprecated Update packman and repo_docs versions (#3293) Co-authored-by: Ashwin Srinath <[email protected]> Drop Thrust's deprecated compiler macros (#3301) Drop CUB_RUNTIME_ENABLED and __THRUST_HAS_CUDART__ (#3305) Adds support for large number of items to `DevicePartition::If` with the `ThreeWayPartition` overload (#2506) * adds support for large number of items to three-way partition * adapts interface to use choose_signed_offset_t * integrates applicable feedback from device-select pr * changes behavior for empty problems * unifies grid constant macro * fixes kernel template specialization mismatch * integrates _CCCL_GRID_CONSTANT changes * resolve merge conflicts * fixes checks in test * fixes test verification * improves tests * makes few improvements to streaming dispatch * improves code comment on test * fixes unrelated compiler error * minor style improvements Refactor scan tunings (#3262) Require C++17 for compiling Thrust and CUB (#3255) * Issue an unsuppressable warning when compiling with < C++17 * Remove C++11/14 presets * Remove CCCL_IGNORE_DEPRECATED_CPP_DIALECT from headers * Remove [CUB|THRUST|TCT]_IGNORE_DEPRECATED_CPP_[11|14] * Remove CUB_ENABLE_DIALECT_CPP[11|14] * Update CI runs * Remove C++11/14 CI runs for CUB and Thrust * Raise compiler minimum versions for C++17 * Update ReadMe * Drop Thrust's cpp14_required.h * Add escape hatch for C++17 removal Fixes: #3252 Implement `views::empty` (#3254) * Disable pair conversion of subrange with clang in C++17 * Fix namespace views * Implement `views::empty` This implements `std::ranges::views::empty`, see https://en.cppreference.com/w/cpp/ranges/empty_view Refactor `limits` and `climits` (#3221) * implement builtins for huge val, nan and nans * change `INFINITY` and `NAN` implementation for NVRTC cuda.parallel: Add documentation for the current iterators along with examples and tests (#3311) * Add tests demonstrating usage of different iterators * Update documentation of reduce_into by merging import code snippet with the rest of the example * Add documentation for current iterators * Run pre-commit checks and update accordingly * Fix comments to refer to the proper lines in the code snippets in the docs Drop clang<14 from CI, update devcontainers. (#3309) Co-authored-by: Bernhard Manfred Gruber <[email protected]> [STF] Cleanup task dependencies object constructors (#3291) * Define tag types for access modes * - Rework how we build task_dep objects based on access mode tags - pack_state is now responsible for using a const_cast for read only data * Greatly simplify the previous attempt : do not define new types, but use integral constants based on the enums * It seems the const_cast was not necessarily so we can simplify it and not even do some dispatch based on access modes Disable test with a gcc-14 regression (#3297) Deprecate Thrust's cpp_compatibility.h macros (#3299) Remove dropped function objects from docs (#3319) Document `NV_TARGET` macros (#3313) [STF] Define ctx.pick_stream() which was missing for the unified context (#3326) * Define ctx.pick_stream() which was missing for the unified context * clang-format Deprecate cub::IterateThreadStore (#3337) Drop CUB's BinaryFlip operator (#3332) Deprecate cub::Swap (#3333) Clarify transform output can overlap input (#3323) Drop CUB APIs with a debug_synchronous parameter (#3330) Fixes: #3329 Drop CUB's util_compiler.cuh for real (#3340) PR #3302 planned to drop the file, but only dropped its content. This was an oversight. So let's drop the entire file. Drop cub::ValueCache (#3346) limits offset types for merge sort (#3328) Drop CDPv1 (#3344) Fixes: #3341 Drop thrust::void_t (#3362) Use cuda::std::addressof in Thrust (#3363) Fix all_of documentation for empty ranges (#3358) all_of always returns true on an empty range. [STF] Do not keep track of dangling events in a CUDA graph backend (#3327) * Unlike the CUDA stream backend, nodes in a CUDA graph are necessarily done when the CUDA graph completes. Therefore keeping track of "dangling events" is a waste of time and resources. * replace can_ignore_dangling_events by track_dangling_events which leads to more readable code * When not storing the dangling events, we must still perform the deinit operations that were producing these events ! Extract scan kernels into NVRTC-compilable header (#3334) * Extract scan kernels into NVRTC-compilable header * Update cub/cub/device/dispatch/dispatch_scan.cuh Co-authored-by: Georgii Evtushenko <[email protected]> --------- Co-authored-by: Ashwin Srinath <[email protected]> Co-authored-by: Georgii Evtushenko <[email protected]> Drop deprecated aliases in Thrust functional (#3272) Fixes: #3271 Drop cub::DivideAndRoundUp (#3347) Use cuda::std::min/max in Thrust (#3364) Implement `cuda::std::numeric_limits` for `__half` and `__nv_bfloat16` (#3361) * implement `cuda::std::numeric_limits` for `__half` and `__nv_bfloat16` Cleanup util_arch (#2773) Deprecate thrust::null_type (#3367) Deprecate cub::DeviceSpmv (#3320) Fixes: #896 Improves `DeviceSegmentedSort` test run time for large number of items and segments (#3246) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * fixes spelling * adds tests for large number of segments * fixes narrowing conversion in tests * addresses review comments * fixes includes Compile basic infra test with C++17 (#3377) Adds support for large number of items and large number of segments to `DeviceSegmentedSort` (#3308) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * addresses review comments * introduces segment offset type * adds tests for large number of segments * adds support for large number of segments * drops segment offset type * fixes thrust namespace * removes about-to-be-deprecated cub iterators * no exec specifier on defaulted ctor * fixes gcc7 linker error * uses local_segment_index_t throughout * determine offset type based on type returned by segment iterator begin/end iterators * minor style improvements Exit with error when RAPIDS CI fails. (#3385) cuda.parallel: Support structured types as algorithm inputs (#3218) * Introduce gpu_struct decorator and typing * Enable `reduce` to accept arrays of structs as inputs * Add test for reducing arrays-of-struct * Update documentation * Use a numpy array rather than ctypes object * Change zeros -> empty for output array and temp storage * Add a TODO for typing GpuStruct * Documentation udpates * Remove test_reduce_struct_type from test_reduce.py * Revert to `to_cccl_value()` accepting ndarray + GpuStruct * Bump copyrights --------- Co-authored-by: Ashwin Srinath <[email protected]> Deprecate thrust::async (#3324) Fixes: #100 Review/Deprecate CUB `util.ptx` for CCCL 2.x (#3342) Fix broken `_CCCL_BUILTIN_ASSUME` macro (#3314) * add compiler-specific path * fix device code path * add _CCC_ASSUME Deprecate thrust::numeric_limits (#3366) Replace `typedef` with `using` in libcu++ (#3368) Deprecate thrust::optional (#3307) Fixes: #3306 Upgrade to Catch2 3.8 (#3310) Fixes: #1724 refactor `<cuda/std/cstdint>` (#3325) Co-authored-by: Bernhard Manfred Gruber <[email protected]> Update CODEOWNERS (#3331) * Update CODEOWNERS * Update CODEOWNERS * Update CODEOWNERS * [pre-commit.ci] auto code formatting --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Fix sign-compare warning (#3408) Implement more cmath functions to be usable on host and device (#3382) * Implement more cmath functions to be usable on host and device * Implement math roots functions * Implement exponential functions Redefine and deprecate thrust::remove_cvref (#3394) * Redefine and deprecate thrust::remove_cvref Co-authored-by: Michael Schellenberger Costa <[email protected]> Fix assert definition for NVHPC due to constexpr issues (#3418) NVHPC cannot decide at compile time where the code would run so _CCCL_ASSERT within a constexpr function breaks it. Fix this by always using the host definition which should also work on device. Fixes #3411 Extend CUB reduce benchmarks (#3401) * Rename max.cu to custom.cu, since it uses a custom operator * Extend types covered my min.cu to all fundamental types * Add some notes on how to collect tuning parameters Fixes: #3283 Update upload-pages-artifact to v3 (#3423) * Update upload-pages-artifact to v3 * Empty commit --------- Co-authored-by: Ashwin Srinath <[email protected]> Replace and deprecate thrust::cuda_cub::terminate (#3421) `std::linalg` accessors and `transposed_layout` (#2962) Add round up/down to multiple (#3234) [FEA]: Introduce Python module with CCCL headers (#3201) * Add cccl/python/cuda_cccl directory and use from cuda_parallel, cuda_cooperative * Run `copy_cccl_headers_to_aude_include()` before `setup()` * Create python/cuda_cccl/cuda/_include/__init__.py, then simply import cuda._include to find the include path. * Add cuda.cccl._version exactly as for cuda.cooperative and cuda.parallel * Bug fix: cuda/_include only exists after shutil.copytree() ran. * Use `f"cuda-cccl @ file://{cccl_path}/python/cuda_cccl"` in setup.py * Remove CustomBuildCommand, CustomWheelBuild in cuda_parallel/setup.py (they are equivalent to the default functions) * Replace := operator (needs Python 3.8+) * Fix oversights: remove `pip3 install ./cuda_cccl` lines from README.md * Restore original README.md: `pip3 install -e` now works on first pass. * cuda_cccl/README.md: FOR INTERNAL USE ONLY * Remove `$pymajor.$pyminor.` prefix in cuda_cccl _version.py (as suggested under https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894035917) Command used: ci/update_version.sh 2 8 0 * Modernize pyproject.toml, setup.py Trigger for this change: * https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894043178 * https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894044996 * Install CCCL headers under cuda.cccl.include Trigger for this change: * https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894048562 Unexpected accidental discovery: cuda.cooperative unit tests pass without CCCL headers entirely. * Factor out cuda_cccl/cuda/cccl/include_paths.py * Reuse cuda_cccl/cuda/cccl/include_paths.py from cuda_cooperative * Add missing Copyright notice. * Add missing __init__.py (cuda.cccl) * Add `"cuda.cccl"` to `autodoc.mock_imports` * Move cuda.cccl.include_paths into function where it is used. (Attempt to resolve Build and Verify Docs failure.) * Add # TODO: move this to a module-level import * Modernize cuda_cooperative/pyproject.toml, setup.py * Convert cuda_cooperative to use hatchling as build backend. * Revert "Convert cuda_cooperative to use hatchling as build backend." This reverts commit 61637d608da06fcf6851ef6197f88b5e7dbc3bbe. * Move numpy from [build-system] requires -> [project] dependencies * Move pyproject.toml [project] dependencies -> setup.py install_requires, to be able to use CCCL_PATH * Remove copy_license() and use license_files=["../../LICENSE"] instead. * Further modernize cuda_cccl/setup.py to use pathlib * Trivial simplifications in cuda_cccl/pyproject.toml * Further simplify cuda_cccl/pyproject.toml, setup.py: remove inconsequential code * Make cuda_cooperative/pyproject.toml more similar to cuda_cccl/pyproject.toml * Add taplo-pre-commit to .pre-commit-config.yaml * taplo-pre-commit auto-fixes * Use pathlib in cuda_cooperative/setup.py * CCCL_PYTHON_PATH in cuda_cooperative/setup.py * Modernize cuda_parallel/pyproject.toml, setup.py * Use pathlib in cuda_parallel/setup.py * Add `# TOML lint & format` comment. * Replace MANIFEST.in with `[tool.setuptools.package-data]` section in pyproject.toml * Use pathlib in cuda/cccl/include_paths.py * pre-commit autoupdate (EXCEPT clang-format, which was manually restored) * Fixes after git merge main * Resolve warning: AttributeError: '_Reduce' object has no attribute 'build_result' ``` =========================================================================== warnings summary =========================================================================== tests/test_reduce.py::test_reduce_non_contiguous /home/coder/cccl/python/devenv/lib/python3.12/site-packages/_pytest/unraisableexception.py:85: PytestUnraisableExceptionWarning: Exception ignored in: <function _Reduce.__del__ at 0x7bf123139080> Traceback (most recent call last): File "/home/coder/cccl/python/cuda_parallel/cuda/parallel/experimental/algorithms/reduce.py", line 132, in __del__ bindings.cccl_device_reduce_cleanup(ctypes.byref(self.build_result)) ^^^^^^^^^^^^^^^^^ AttributeError: '_Reduce' object has no attribute 'build_result' warnings.warn(pytest.PytestUnraisableExceptionWarning(msg)) -- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html ============================================================= 1 passed, 93 deselected, 1 warning in 0.44s ============================================================== ``` * Move `copy_cccl_headers_to_cuda_cccl_include()` functionality to `class CustomBuildPy` * Introduce cuda_cooperative/constraints.txt * Also add cuda_parallel/constraints.txt * Add `--constraint constraints.txt` in ci/test_python.sh * Update Copyright dates * Switch to https://github.com/ComPWA/taplo-pre-commit (the other repo has been archived by the owner on Jul 1, 2024) For completeness: The other repo took a long time to install into the pre-commit cache; so long it lead to timeouts in the CCCL CI. * Remove unused cuda_parallel jinja2 dependency (noticed by chance). * Remove constraints.txt files, advertise running `pip install cuda-cccl` first instead. * Make cuda_cooperative, cuda_parallel testing completely independent. * Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc] * Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Fix sign-compare warning (#3408) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Revert "Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc]" This reverts commit ea33a218ed77a075156cd1b332047202adb25aa2. Error message: https://github.com/NVIDIA/cccl/pull/3201#issuecomment-2594012971 * Try using A100 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Also show cuda-cooperative site-packages, cuda-parallel site-packages (after pip install) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Try using l4 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Restore original ci/matrix.yaml [skip-rapids] * Use for loop in test_python.sh to avoid code duplication. * Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci] * Comment out taplo-lint in pre-commit config [skip-rapids][skip-matx][skip-docs][skip-vdc] * Revert "Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci]" This reverts commit ec206fd8b50a6a293e00a5825b579e125010b13d. * Implement suggestion by @shwina (https://github.com/NVIDIA/cccl/pull/3201#pullrequestreview-2556918460) * Address feedback by @leofang --------- Co-authored-by: Bernhard Manfred Gruber <[email protected]> cuda.parallel: Add optional stream argument to reduce_into() (#3348) * Add optional stream argument to reduce_into() * Add tests to check for reduce_into() stream behavior * Move protocol related utils to separate file and rework __cuda_stream__ error messages * Fix synchronization issue in stream test and add one more invalid stream test case * Rename cuda stream validation function after removing leading underscore * Unpack values from __cuda_stream__ instead of indexing * Fix linting errors * Handle TypeError when unpacking invalid __cuda_stream__ return * Use stream to allocate cupy memory in new stream test Upgrade to actions/deploy-pages@v4 (from v2), as suggested by @leofang (#3434) Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ (#3419) * Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ Fixes #3404 move to c++17, finalize device optimization fix msvc compilation, update tests Deprectate C++11 and C++14 for libcu++ (#3173) * Deprectate C++11 and C++14 for libcu++ Co-authored-by: Bernhard Manfred Gruber <[email protected]> Implement `abs` and `div` from `cstdlib` (#3153) * implement integer abs functions * improve tests, fix constexpr support * just use the our implementation * implement `cuda::std::div` * prefer host's `div_t` like types * provide `cuda::std::abs` overloads for floats * allow fp abs for NVRTC * silence msvc's warning about conversion from floating point to integral Fix missing radix sort policies (#3174) Fixes NVBug 5009941 Introduces new `DeviceReduce::Arg{Min,Max}` interface with two output iterators (#3148) * introduces new arg{min,max} interface with two output iterators * adds fp inf tests * fixes docs * improves code example * fixes exec space specifier * trying to fix deprecation warning for more compilers * inlines unzip operator * trying to fix deprecation warning for nvhpc * integrates supression fixes in diagnostics * pre-ctk 11.5 deprecation suppression * fixes icc * fix for pre-ctk11.5 * cleans up deprecation suppression * cleanup Extend tuning documentation (#3179) Add codespell pre-commit hook, fix typos in CCCL (#3168) * Add codespell pre-commit hook * Automatic changes from codespell. * Manual changes. Fix parameter space for TUNE_LOAD in scan benchmark (#3176) fix various old compiler checks (#3178) implement C++26 `std::projected` (#3175) Fix pre-commit config for codespell and remaining typos (#3182) Massive cleanup of our config (#3155) Fix UB in atomics with automatic storage (#2586) * Adds specialized local cuda atomics and injects them into most atomics paths. Co-authored-by: Georgy Evtushenko <[email protected]> Co-authored-by: gonzalobg <[email protected]> * Allow CUDA 12.2 to keep perf, this addresses earlier comments in #478 * Remove extraneous double brackets in unformatted code. * Merge unsafe atomic logic into `__cuda_is_local`. * Use `const_cast` for type conversions in cuda_local.h * Fix build issues from interface changes * Fix missing __nanosleep on sm70- * Guard __isLocal from NVHPC * Use PTX instead of running nothing from NVHPC * fixup /s/nvrtc/nvhpc * Fixup missing CUDA ifdef surrounding device code * Fix codegen * Bypass some sort of compiler bug on GCC7 * Apply suggestions from code review * Use unsafe automatic storage atomics in codegen tests --------- Co-authored-by: Georgy Evtushenko <[email protected]> Co-authored-by: gonzalobg <[email protected]> Co-authored-by: Michael Schellenberger Costa <[email protected]> Refactor the source code layout for `cuda.parallel` (#3177) * Refactor the source layout for cuda.parallel * Add copyright * Address review feedback * Don't import anything into `experimental` namespace * fix import --------- Co-authored-by: Ashwin Srinath <[email protected]> new type-erased memory resources (#2824) s/_LIBCUDACXX_DECLSPEC_EMPTY_BASES/_CCCL_DECLSPEC_EMPTY_BASES/g (#3186) Document address stability of `thrust::transform` (#3181) * Do not document _LIBCUDACXX_MARK_CAN_COPY_ARGUMENTS * Reformat and fix UnaryFunction/BinaryFunction in transform docs * Mention transform can use proclaim_copyable_arguments * Document cuda::proclaims_copyable_arguments better * Deprecate depending on transform functor argument addresses Fixes: #3053 turn off cuda version check for clangd (#3194) [STF] jacobi example based on parallel_for (#3187) * Simple jacobi example with parallel for and reductions * clang-format * remove useless capture list fixes pre-nv_diag suppression issues (#3189) Prefer c2h::type_name over c2h::demangle (#3195) Fix memcpy_async* tests (#3197) * memcpy_async_tx: Fix bug in test Two bugs, one of which occurs in practice: 1. There is a missing fence.proxy.space::global between the writes to global memory and the memcpy_async_tx. (Occurs in practice) 2. The end of the kernel should be fenced with `__syncthreads()`, because the barrier is invalidated in the destructor. If other threads are still waiting on it, there will be UB. (Has not yet manifested itself) * cp_async_bulk_tensor: Pre-emptively fence more in test Add type annotations and mypy checks for `cuda.parallel` (#3180) * Refactor the source layout for cuda.parallel * Add initial type annotations * Update pre-commit config * More typing * Fix bad merge * Fix TYPE_CHECKING and numpy annotations * typing bindings.py correctly * Address review feedback --------- Co-authored-by: Ashwin Srinath <[email protected]> Fix rendering of cuda.parallel docs (#3192) * Fix pre-commit config for codespell and remaining typos * Fix rendering of docs for cuda.parallel --------- Co-authored-by: Ashwin Srinath <[email protected]> Enable PDL for DeviceMergeSortBlockSortKernel (#3199) The kernel already contains a call to _CCCL_PDL_GRID_DEPENDENCY_SYNC. This commit enables PDL when launching the kernel. Adds support for large `num_items` to `DeviceReduce::{ArgMin,ArgMax}` (#2647) * adds benchmarks for reduce::arg{min,max} * preliminary streaming arg-extremum reduction * fixes implicit conversion * uses streaming dispatch class * changes arg benches to use new streaming reduce * streaming arg-extrema reduction * fixes style * fixes compilation failures * cleanups * adds rst style comments * declare vars const and use clamp * consolidates argmin argmax benchmarks * fixes thrust usage * drops offset type in arg-extrema benchmarks * fixes clang cuda * exec space macros * switch to signed global offset type for slightly better perf * clarifies documentation * applies minor benchmark style changes from review comments * fixes interface documentation and comments * list-init accumulating output op * improves style, comments, and tests * cleans up aggregate init * renames dispatch class usage in benchmarks * fixes merge conflicts * addresses review comments * addresses review comments * fixes assertion * removes superseded implementation * changes large problem tests to use new interface * removes obsolete tests for deprecated interface Fixes for Python 3.7 docs environment (#3206) Co-authored-by: Ashwin Srinath <[email protected]> Adds support for large number of items to `DeviceTransform` (#3172) * moves large problem test helper to common file * adds support for large num items to device transform * adds tests for large number of items to device interface * fixes format * addresses review comments cp_async_bulk: Fix test (#3198) * memcpy_async_tx: Fix bug in test Two bugs, one of which occurs in practice: 1. There is a missing fence.proxy.space::global between the writes to global memory and the memcpy_async_tx. (Occurs in practice) 2. The end of the kernel should be fenced with `__syncthreads()`, because the barrier is invalidated in the destructor. If other threads are still waiting on it, there will be UB. (Has not yet manifested itself) * cp_async_bulk_tensor: Pre-emptively fence more in test * cp_async_bulk: Fix test The global memory pointer could be misaligned. cudax fixes for msvc 14.41 (#3200) avoid instantiating class templates in `is_same` implementation when possible (#3203) Fix: make launchers a CUB detail; make kernel source functions hidden. (#3209) * Fix: make launchers a CUB detail; make kernel source functions hidden. * [pre-commit.ci] auto code formatting * Address review comments, fix which macro gets fixed. help the ranges concepts recognize standard contiguous iterators in c++14/17 (#3202) unify macros and cmake options that control the suppression of deprecation warnings (#3220) * unify macros and cmake options that control the suppression of deprecation warnings * suppress nvcc warning #186 in thrust header tests * suppress c++ dialect deprecation warnings in libcudacxx header tests Fx thread-reduce performance regression (#3225) cuda.parallel: In-memory caching of build objects (#3216) * Define __eq__ and __hash__ for Iterators * Define cache_with_key utility and use it to cache Reduce objects * Add tests for caching Reduce objects * Tighten up types * Updates to support 3.7 * Address review feedback * Introduce IteratorKind to hold iterator type information * Use the .kind to generate an abi_name * Remove __eq__ and __hash__ methods from IteratorBase * Move helper function * Formatting * Don't unpack tuple in cache key --------- Co-authored-by: Ashwin Srinath <[email protected]> Just enough ranges for c++14 `span` (#3211) use generalized concepts portability macros to simplify the `range` concept (#3217) fixes some issues in the concepts portability macros and then re-implements the `range` concept with `_CCCL_REQUIRES_EXPR` Use Ruff to sort imports (#3230) * Update pyproject.tomls for import sorting * Update files after running pre-commit * Move ruff config to pyproject.toml --------- Co-authored-by: Ashwin Srinath <[email protected]> fix tuning_scan sm90 config issue (#3236) Co-authored-by: Shijie Chen <[email protected]> [STF] Logical token (#3196) * Split the implementation of the void interface into the definition of the interface, and its implementations on streams and graphs. * Add missing files * Check if a task implementation can match a prototype where the void_interface arguments are ignored * Implement ctx.abstract_logical_data() which relies on a void data interface * Illustrate how to use abstract handles in local contexts * Introduce an is_void_interface() virtual method in the data interface to potentially optimize some stages * Small improvements in the examples * Do not try to allocate or move void data * Do not use I as a variable * fix linkage error * rename abtract_logical_data into logical_token * Document logical token * fix spelling error * fix sphinx error * reflect name changes * use meaningful variable names * simplify logical_token implementation because writeback is already disabled * add a unit test for token elision * implement token elision in host_launch * Remove unused type * Implement helpers to check if a function can be invoked from a tuple, or from a tuple where we removed tokens * Much simpler is_tuple_invocable_with_filtered implementation * Fix buggy test * Factorize code * Document that we can ignore tokens for task and host_launch * Documentation for logical data freeze Fix ReduceByKey tuning (#3240) Fix RLE tuning (#3239) cuda.parallel: Forbid non-contiguous arrays as inputs (or outputs) (#3233) * Forbid non-contiguous arrays as inputs (or outputs) * Implement a more robust way to check for contiguity * Don't bother if cublas unavailable * Fix how we check for zero-element arrays * sort imports --------- Co-authored-by: Ashwin Srinath <[email protected]> expands support for more offset types in segmented benchmark (#3231) Add escape hatches to the cmake configuration of the header tests so that we can tests deprecated compilers / dialects (#3253) * Add escape hatches to the cmake configuration of the header tests so that we can tests deprecated compilers / dialects * Do not add option twice ptx: Add add_instruction.py (#3190) This file helps create the necessary structure for new PTX instructions. Co-authored-by: Allard Hendriksen <[email protected]> Bump main to 2.9.0. (#3247) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Drop cub::Mutex (#3251) Fixes: #3250 Remove legacy macros from CUB util_arch.cuh (#3257) Fixes: #3256 Remove thrust::[unary|binary]_traits (#3260) Fixes: #3259 Architecture and OS identification macros (#3237) Bump main to 3.0.0. (#3265) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Drop thrust not1 and not2 (#3264) Fixes: #3263 CCCL Internal macro documentation (#3238) Deprecate GridBarrier and GridBarrierLifetime (#3258) Fixes: #1389 Require at least gcc7 (#3268) Fixes: #3267 Drop thrust::[unary|binary]_function (#3274) Fixes: #3273 Drop ICC from CI (#3277) [STF] Corruption of the capture list of an extended lambda with a parallel_for construct on a host execution place (#3270) * Add a test to reproduce a bug observed with parallel_for on a host place * clang-format * use _CCCL_ASSERT * Attempt to debug * do not create a tuple with a universal reference that is out of scope when we use it, use an lvalue instead * fix lambda expression * clang-format Enable thrust::identity test for non-MSVC (#3281) This seems to be an oversight when the test was added Co-authored-by: Michael Schellenberger Costa <[email protected]> Enable PDL in triple chevron launch (#3282) It seems PDL was disabled by accident when _THRUST_HAS_PDL was renamed to _CCCL_HAS_PDL during the review introducing the feature. Disambiguate line continuations and macro continuations in <nv/target> (#3244) Drop VS 2017 from CI (#3287) Fixes: #3286 Drop ICC support in code (#3279) * Drop ICC from code Fixes: #3278 Co-authored-by: Michael Schellenberger Costa <[email protected]> Make CUB NVRTC commandline arguments come from a cmake template (#3292) Propose the same components (thrust, cub, libc++, cudax, cuda.parallel,...) in the bug report template than in the feature request template (#3295) Use process isolation instead of default hyper-v for Windows. (#3294) Try improving build times by using process isolation instead of hyper-v Co-authored-by: Michael Schellenberger Costa <[email protected]> [pre-commit.ci] pre-commit autoupdate (#3248) * [pre-commit.ci] pre-commit autoupdate updates: - [github.com/pre-commit/mirrors-clang-format: v18.1.8 → v19.1.6](https://github.com/pre-commit/mirrors-clang-format/compare/v18.1.8...v19.1.6) - [github.com/astral-sh/ruff-pre-commit: v0.8.3 → v0.8.6](https://github.com/astral-sh/ruff-pre-commit/compare/v0.8.3...v0.8.6) - [github.com/pre-commit/mirrors-mypy: v1.13.0 → v1.14.1](https://github.com/pre-commit/mirrors-mypy/compare/v1.13.0...v1.14.1) Co-authored-by: Michael Schellenberger Costa <[email protected]> Drop Thrust legacy arch macros (#3298) Which were disabled and could be re-enabled using THRUST_PROVIDE_LEGACY_ARCH_MACROS Drop Thrust's compiler_fence.h (#3300) Drop CTK 11.x from CI (#3275) * Add cuda12.0-gcc7 devcontainer * Move MSVC2017 jobs to CTK 12.6 Those is the only combination where rapidsai has devcontainers * Add /Zc:__cplusplus for the libcudacxx tests * Only add excape hatch for affected CTKs * Workaround missing cudaLaunchKernelEx on MSVC cudaLaunchKernelEx requires C++11, but unfortunately <cuda_runtime.h> checks this using the __cplusplus macro, which is reported wrongly for MSVC. CTK 12.3 fixed this by additionally detecting _MSV_VER. As a workaround, we provide our own copy of cudaLaunchKernelEx when it is not available from the CTK. * Workaround nvcc+MSVC issue * Regenerate devcontainers Fixes: #3249 Co-authored-by: Michael Schellenberger Costa <[email protected]> Update packman and repo_docs versions (#3293) Co-authored-by: Ashwin Srinath <[email protected]> Drop Thrust's deprecated compiler macros (#3301) Drop CUB_RUNTIME_ENABLED and __THRUST_HAS_CUDART__ (#3305) Adds support for large number of items to `DevicePartition::If` with the `ThreeWayPartition` overload (#2506) * adds support for large number of items to three-way partition * adapts interface to use choose_signed_offset_t * integrates applicable feedback from device-select pr * changes behavior for empty problems * unifies grid constant macro * fixes kernel template specialization mismatch * integrates _CCCL_GRID_CONSTANT changes * resolve merge conflicts * fixes checks in test * fixes test verification * improves tests * makes few improvements to streaming dispatch * improves code comment on test * fixes unrelated compiler error * minor style improvements Refactor scan tunings (#3262) Require C++17 for compiling Thrust and CUB (#3255) * Issue an unsuppressable warning when compiling with < C++17 * Remove C++11/14 presets * Remove CCCL_IGNORE_DEPRECATED_CPP_DIALECT from headers * Remove [CUB|THRUST|TCT]_IGNORE_DEPRECATED_CPP_[11|14] * Remove CUB_ENABLE_DIALECT_CPP[11|14] * Update CI runs * Remove C++11/14 CI runs for CUB and Thrust * Raise compiler minimum versions for C++17 * Update ReadMe * Drop Thrust's cpp14_required.h * Add escape hatch for C++17 removal Fixes: #3252 Implement `views::empty` (#3254) * Disable pair conversion of subrange with clang in C++17 * Fix namespace views * Implement `views::empty` This implements `std::ranges::views::empty`, see https://en.cppreference.com/w/cpp/ranges/empty_view Refactor `limits` and `climits` (#3221) * implement builtins for huge val, nan and nans * change `INFINITY` and `NAN` implementation for NVRTC cuda.parallel: Add documentation for the current iterators along with examples and tests (#3311) * Add tests demonstrating usage of different iterators * Update documentation of reduce_into by merging import code snippet with the rest of the example * Add documentation for current iterators * Run pre-commit checks and update accordingly * Fix comments to refer to the proper lines in the code snippets in the docs Drop clang<14 from CI, update devcontainers. (#3309) Co-authored-by: Bernhard Manfred Gruber <[email protected]> [STF] Cleanup task dependencies object constructors (#3291) * Define tag types for access modes * - Rework how we build task_dep objects based on access mode tags - pack_state is now responsible for using a const_cast for read only data * Greatly simplify the previous attempt : do not define new types, but use integral constants based on the enums * It seems the const_cast was not necessarily so we can simplify it and not even do some dispatch based on access modes Disable test with a gcc-14 regression (#3297) Deprecate Thrust's cpp_compatibility.h macros (#3299) Remove dropped function objects from docs (#3319) Document `NV_TARGET` macros (#3313) [STF] Define ctx.pick_stream() which was missing for the unified context (#3326) * Define ctx.pick_stream() which was missing for the unified context * clang-format Deprecate cub::IterateThreadStore (#3337) Drop CUB's BinaryFlip operator (#3332) Deprecate cub::Swap (#3333) Clarify transform output can overlap input (#3323) Drop CUB APIs with a debug_synchronous parameter (#3330) Fixes: #3329 Drop CUB's util_compiler.cuh for real (#3340) PR #3302 planned to drop the file, but only dropped its content. This was an oversight. So let's drop the entire file. Drop cub::ValueCache (#3346) limits offset types for merge sort (#3328) Drop CDPv1 (#3344) Fixes: #3341 Drop thrust::void_t (#3362) Use cuda::std::addressof in Thrust (#3363) Fix all_of documentation for empty ranges (#3358) all_of always returns true on an empty range. [STF] Do not keep track of dangling events in a CUDA graph backend (#3327) * Unlike the CUDA stream backend, nodes in a CUDA graph are necessarily done when the CUDA graph completes. Therefore keeping track of "dangling events" is a waste of time and resources. * replace can_ignore_dangling_events by track_dangling_events which leads to more readable code * When not storing the dangling events, we must still perform the deinit operations that were producing these events ! Extract scan kernels into NVRTC-compilable header (#3334) * Extract scan kernels into NVRTC-compilable header * Update cub/cub/device/dispatch/dispatch_scan.cuh Co-authored-by: Georgii Evtushenko <[email protected]> --------- Co-authored-by: Ashwin Srinath <[email protected]> Co-authored-by: Georgii Evtushenko <[email protected]> Drop deprecated aliases in Thrust functional (#3272) Fixes: #3271 Drop cub::DivideAndRoundUp (#3347) Use cuda::std::min/max in Thrust (#3364) Implement `cuda::std::numeric_limits` for `__half` and `__nv_bfloat16` (#3361) * implement `cuda::std::numeric_limits` for `__half` and `__nv_bfloat16` Cleanup util_arch (#2773) Deprecate thrust::null_type (#3367) Deprecate cub::DeviceSpmv (#3320) Fixes: #896 Improves `DeviceSegmentedSort` test run time for large number of items and segments (#3246) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * fixes spelling * adds tests for large number of segments * fixes narrowing conversion in tests * addresses review comments * fixes includes Compile basic infra test with C++17 (#3377) Adds support for large number of items and large number of segments to `DeviceSegmentedSort` (#3308) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * addresses review comments * introduces segment offset type * adds tests for large number of segments * adds support for large number of segments * drops segment offset type * fixes thrust namespace * removes about-to-be-deprecated cub iterators * no exec specifier on defaulted ctor * fixes gcc7 linker error * uses local_segment_index_t throughout * determine offset type based on type returned by segment iterator begin/end iterators * minor style improvements Exit with error when RAPIDS CI fails. (#3385) cuda.parallel: Support structured types as algorithm inputs (#3218) * Introduce gpu_struct decorator and typing * Enable `reduce` to accept arrays of structs as inputs * Add test for reducing arrays-of-struct * Update documentation * Use a numpy array rather than ctypes object * Change zeros -> empty for output array and temp storage * Add a TODO for typing GpuStruct * Documentation udpates * Remove test_reduce_struct_type from test_reduce.py * Revert to `to_cccl_value()` accepting ndarray + GpuStruct * Bump copyrights --------- Co-authored-by: Ashwin Srinath <[email protected]> Deprecate thrust::async (#3324) Fixes: #100 Review/Deprecate CUB `util.ptx` for CCCL 2.x (#3342) Fix broken `_CCCL_BUILTIN_ASSUME` macro (#3314) * add compiler-specific path * fix device code path * add _CCC_ASSUME Deprecate thrust::numeric_limits (#3366) Replace `typedef` with `using` in libcu++ (#3368) Deprecate thrust::optional (#3307) Fixes: #3306 Upgrade to Catch2 3.8 (#3310) Fixes: #1724 refactor `<cuda/std/cstdint>` (#3325) Co-authored-by: Bernhard Manfred Gruber <[email protected]> Update CODEOWNERS (#3331) * Update CODEOWNERS * Update CODEOWNERS * Update CODEOWNERS * [pre-commit.ci] auto code formatting --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Fix sign-compare warning (#3408) Implement more cmath functions to be usable on host and device (#3382) * Implement more cmath functions to be usable on host and device * Implement math roots functions * Implement exponential functions Redefine and deprecate thrust::remove_cvref (#3394) * Redefine and deprecate thrust::remove_cvref Co-authored-by: Michael Schellenberger Costa <[email protected]> Fix assert definition for NVHPC due to constexpr issues (#3418) NVHPC cannot decide at compile time where the code would run so _CCCL_ASSERT within a constexpr function breaks it. Fix this by always using the host definition which should also work on device. Fixes #3411 Extend CUB reduce benchmarks (#3401) * Rename max.cu to custom.cu, since it uses a custom operator * Extend types covered my min.cu to all fundamental types * Add some notes on how to collect tuning parameters Fixes: #3283 Update upload-pages-artifact to v3 (#3423) * Update upload-pages-artifact to v3 * Empty commit --------- Co-authored-by: Ashwin Srinath <[email protected]> Replace and deprecate thrust::cuda_cub::terminate (#3421) `std::linalg` accessors and `transposed_layout` (#2962) Add round up/down to multiple (#3234) [FEA]: Introduce Python module with CCCL headers (#3201) * Add cccl/python/cuda_cccl directory and use from cuda_parallel, cuda_cooperative * Run `copy_cccl_headers_to_aude_include()` before `setup()` * Create python/cuda_cccl/cuda/_include/__init__.py, then simply import cuda._include to find the include path. * Add cuda.cccl._version exactly as for cuda.cooperative and cuda.parallel * Bug fix: cuda/_include only exists after shutil.copytree() ran. * Use `f"cuda-cccl @ file://{cccl_path}/python/cuda_cccl"` in setup.py * Remove CustomBuildCommand, CustomWheelBuild in cuda_parallel/setup.py (they are equivalent to the default functions) * Replace := operator (needs Python 3.8+) * Fix oversights: remove `pip3 install ./cuda_cccl` lines from README.md * Restore original README.md: `pip3 install -e` now works on first pass. * cuda_cccl/README.md: FOR INTERNAL USE ONLY * Remove `$pymajor.$pyminor.` prefix in cuda_cccl _version.py (as suggested under https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894035917) Command used: ci/update_version.sh 2 8 0 * Modernize pyproject.toml, setup.py Trigger for this change: * https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894043178 * https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894044996 * Install CCCL headers under cuda.cccl.include Trigger for this change: * https://github.com/NVIDIA/cccl/pull/3201#discussion_r1894048562 Unexpected accidental discovery: cuda.cooperative unit tests pass without CCCL headers entirely. * Factor out cuda_cccl/cuda/cccl/include_paths.py * Reuse cuda_cccl/cuda/cccl/include_paths.py from cuda_cooperative * Add missing Copyright notice. * Add missing __init__.py (cuda.cccl) * Add `"cuda.cccl"` to `autodoc.mock_imports` * Move cuda.cccl.include_paths into function where it is used. (Attempt to resolve Build and Verify Docs failure.) * Add # TODO: move this to a module-level import * Modernize cuda_cooperative/pyproject.toml, setup.py * Convert cuda_cooperative to use hatchling as build backend. * Revert "Convert cuda_cooperative to use hatchling as build backend." This reverts commit 61637d608da06fcf6851ef6197f88b5e7dbc3bbe. * Move numpy from [build-system] requires -> [project] dependencies * Move pyproject.toml [project] dependencies -> setup.py install_requires, to be able to use CCCL_PATH * Remove copy_license() and use license_files=["../../LICENSE"] instead. * Further modernize cuda_cccl/setup.py to use pathlib * Trivial simplifications in cuda_cccl/pyproject.toml * Further simplify cuda_cccl/pyproject.toml, setup.py: remove inconsequential code * Make cuda_cooperative/pyproject.toml more similar to cuda_cccl/pyproject.toml * Add taplo-pre-commit to .pre-commit-config.yaml * taplo-pre-commit auto-fixes * Use pathlib in cuda_cooperative/setup.py * CCCL_PYTHON_PATH in cuda_cooperative/setup.py * Modernize cuda_parallel/pyproject.toml, setup.py * Use pathlib in cuda_parallel/setup.py * Add `# TOML lint & format` comment. * Replace MANIFEST.in with `[tool.setuptools.package-data]` section in pyproject.toml * Use pathlib in cuda/cccl/include_paths.py * pre-commit autoupdate (EXCEPT clang-format, which was manually restored) * Fixes after git merge main * Resolve warning: AttributeError: '_Reduce' object has no attribute 'build_result' ``` =========================================================================== warnings summary =========================================================================== tests/test_reduce.py::test_reduce_non_contiguous /home/coder/cccl/python/devenv/lib/python3.12/site-packages/_pytest/unraisableexception.py:85: PytestUnraisableExceptionWarning: Exception ignored in: <function _Reduce.__del__ at 0x7bf123139080> Traceback (most recent call last): File "/home/coder/cccl/python/cuda_parallel/cuda/parallel/experimental/algorithms/reduce.py", line 132, in __del__ bindings.cccl_device_reduce_cleanup(ctypes.byref(self.build_result)) ^^^^^^^^^^^^^^^^^ AttributeError: '_Reduce' object has no attribute 'build_result' warnings.warn(pytest.PytestUnraisableExceptionWarning(msg)) -- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html ============================================================= 1 passed, 93 deselected, 1 warning in 0.44s ============================================================== ``` * Move `copy_cccl_headers_to_cuda_cccl_include()` functionality to `class CustomBuildPy` * Introduce cuda_cooperative/constraints.txt * Also add cuda_parallel/constraints.txt * Add `--constraint constraints.txt` in ci/test_python.sh * Update Copyright dates * Switch to https://github.com/ComPWA/taplo-pre-commit (the other repo has been archived by the owner on Jul 1, 2024) For completeness: The other repo took a long time to install into the pre-commit cache; so long it lead to timeouts in the CCCL CI. * Remove unused cuda_parallel jinja2 dependency (noticed by chance). * Remove constraints.txt files, advertise running `pip install cuda-cccl` first instead. * Make cuda_cooperative, cuda_parallel testing completely independent. * Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc] * Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Fix sign-compare warning (#3408) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Revert "Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc]" This reverts commit ea33a218ed77a075156cd1b332047202adb25aa2. Error message: https://github.com/NVIDIA/cccl/pull/3201#issuecomment-2594012971 * Try using A100 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Also show cuda-cooperative site-packages, cuda-parallel site-packages (after pip install) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Try using l4 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Restore original ci/matrix.yaml [skip-rapids] * Use for loop in test_python.sh to avoid code duplication. * Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci] * Comment out taplo-lint in pre-commit config [skip-rapids][skip-matx][skip-docs][skip-vdc] * Revert "Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci]" This reverts commit ec206fd8b50a6a293e00a5825b579e125010b13d. * Implement suggestion by @shwina (https://github.com/NVIDIA/cccl/pull/3201#pullrequestreview-2556918460) * Address feedback by @leofang --------- Co-authored-by: Bernhard Manfred Gruber <[email protected]> cuda.parallel: Add optional stream argument to reduce_into() (#3348) * Add optional stream argument to reduce_into() * Add tests to check for reduce_into() stream behavior * Move protocol related utils to separate file and rework __cuda_stream__ error messages * Fix synchronization issue in stream test and add one more invalid stream test case * Rename cuda stream validation function after removing leading underscore * Unpack values from __cuda_stream__ instead of indexing * Fix linting errors * Handle TypeError when unpacking invalid __cuda_stream__ return * Use stream to allocate cupy memory in new stream test Upgrade to actions/deploy-pages@v4 (from v2), as suggested by @leofang (#3434) Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ (#3419) * Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ Fixes #3404 Fix CI issues (#3443) update docs fix review restrict allowed types replace constexpr implementations with generic optimize `__is_arithmetic_integral`

@shwina

update docs update docs add `memcmp`, `memmove` and `memchr` implementations implement tests Use cuda::std::min/max in Thrust (NVIDIA#3364) Implement `cuda::std::numeric_limits` for `__half` and `__nv_bfloat16` (NVIDIA#3361) * implement `cuda::std::numeric_limits` for `__half` and `__nv_bfloat16` Cleanup util_arch (NVIDIA#2773) Deprecate thrust::null_type (NVIDIA#3367) Deprecate cub::DeviceSpmv (NVIDIA#3320) Fixes: NVIDIA#896 Improves `DeviceSegmentedSort` test run time for large number of items and segments (NVIDIA#3246) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * fixes spelling * adds tests for large number of segments * fixes narrowing conversion in tests * addresses review comments * fixes includes Compile basic infra test with C++17 (NVIDIA#3377) Adds support for large number of items and large number of segments to `DeviceSegmentedSort` (NVIDIA#3308) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * addresses review comments * introduces segment offset type * adds tests for large number of segments * adds support for large number of segments * drops segment offset type * fixes thrust namespace * removes about-to-be-deprecated cub iterators * no exec specifier on defaulted ctor * fixes gcc7 linker error * uses local_segment_index_t throughout * determine offset type based on type returned by segment iterator begin/end iterators * minor style improvements Exit with error when RAPIDS CI fails. (NVIDIA#3385) cuda.parallel: Support structured types as algorithm inputs (NVIDIA#3218) * Introduce gpu_struct decorator and typing * Enable `reduce` to accept arrays of structs as inputs * Add test for reducing arrays-of-struct * Update documentation * Use a numpy array rather than ctypes object * Change zeros -> empty for output array and temp storage * Add a TODO for typing GpuStruct * Documentation udpates * Remove test_reduce_struct_type from test_reduce.py * Revert to `to_cccl_value()` accepting ndarray + GpuStruct * Bump copyrights --------- Co-authored-by: Ashwin Srinath <[email protected]> Deprecate thrust::async (NVIDIA#3324) Fixes: NVIDIA#100 Review/Deprecate CUB `util.ptx` for CCCL 2.x (NVIDIA#3342) Fix broken `_CCCL_BUILTIN_ASSUME` macro (NVIDIA#3314) * add compiler-specific path * fix device code path * add _CCC_ASSUME Deprecate thrust::numeric_limits (NVIDIA#3366) Replace `typedef` with `using` in libcu++ (NVIDIA#3368) Deprecate thrust::optional (NVIDIA#3307) Fixes: NVIDIA#3306 Upgrade to Catch2 3.8 (NVIDIA#3310) Fixes: NVIDIA#1724 refactor `<cuda/std/cstdint>` (NVIDIA#3325) Co-authored-by: Bernhard Manfred Gruber <[email protected]> Update CODEOWNERS (NVIDIA#3331) * Update CODEOWNERS * Update CODEOWNERS * Update CODEOWNERS * [pre-commit.ci] auto code formatting --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Fix sign-compare warning (NVIDIA#3408) Implement more cmath functions to be usable on host and device (NVIDIA#3382) * Implement more cmath functions to be usable on host and device * Implement math roots functions * Implement exponential functions Redefine and deprecate thrust::remove_cvref (NVIDIA#3394) * Redefine and deprecate thrust::remove_cvref Co-authored-by: Michael Schellenberger Costa <[email protected]> Fix assert definition for NVHPC due to constexpr issues (NVIDIA#3418) NVHPC cannot decide at compile time where the code would run so _CCCL_ASSERT within a constexpr function breaks it. Fix this by always using the host definition which should also work on device. Fixes NVIDIA#3411 Extend CUB reduce benchmarks (NVIDIA#3401) * Rename max.cu to custom.cu, since it uses a custom operator * Extend types covered my min.cu to all fundamental types * Add some notes on how to collect tuning parameters Fixes: NVIDIA#3283 Update upload-pages-artifact to v3 (NVIDIA#3423) * Update upload-pages-artifact to v3 * Empty commit --------- Co-authored-by: Ashwin Srinath <[email protected]> Replace and deprecate thrust::cuda_cub::terminate (NVIDIA#3421) `std::linalg` accessors and `transposed_layout` (NVIDIA#2962) Add round up/down to multiple (NVIDIA#3234) [FEA]: Introduce Python module with CCCL headers (NVIDIA#3201) * Add cccl/python/cuda_cccl directory and use from cuda_parallel, cuda_cooperative * Run `copy_cccl_headers_to_aude_include()` before `setup()` * Create python/cuda_cccl/cuda/_include/__init__.py, then simply import cuda._include to find the include path. * Add cuda.cccl._version exactly as for cuda.cooperative and cuda.parallel * Bug fix: cuda/_include only exists after shutil.copytree() ran. * Use `f"cuda-cccl @ file://{cccl_path}/python/cuda_cccl"` in setup.py * Remove CustomBuildCommand, CustomWheelBuild in cuda_parallel/setup.py (they are equivalent to the default functions) * Replace := operator (needs Python 3.8+) * Fix oversights: remove `pip3 install ./cuda_cccl` lines from README.md * Restore original README.md: `pip3 install -e` now works on first pass. * cuda_cccl/README.md: FOR INTERNAL USE ONLY * Remove `$pymajor.$pyminor.` prefix in cuda_cccl _version.py (as suggested under NVIDIA#3201 (comment)) Command used: ci/update_version.sh 2 8 0 * Modernize pyproject.toml, setup.py Trigger for this change: * NVIDIA#3201 (comment) * NVIDIA#3201 (comment) * Install CCCL headers under cuda.cccl.include Trigger for this change: * NVIDIA#3201 (comment) Unexpected accidental discovery: cuda.cooperative unit tests pass without CCCL headers entirely. * Factor out cuda_cccl/cuda/cccl/include_paths.py * Reuse cuda_cccl/cuda/cccl/include_paths.py from cuda_cooperative * Add missing Copyright notice. * Add missing __init__.py (cuda.cccl) * Add `"cuda.cccl"` to `autodoc.mock_imports` * Move cuda.cccl.include_paths into function where it is used. (Attempt to resolve Build and Verify Docs failure.) * Add # TODO: move this to a module-level import * Modernize cuda_cooperative/pyproject.toml, setup.py * Convert cuda_cooperative to use hatchling as build backend. * Revert "Convert cuda_cooperative to use hatchling as build backend." This reverts commit 61637d6. * Move numpy from [build-system] requires -> [project] dependencies * Move pyproject.toml [project] dependencies -> setup.py install_requires, to be able to use CCCL_PATH * Remove copy_license() and use license_files=["../../LICENSE"] instead. * Further modernize cuda_cccl/setup.py to use pathlib * Trivial simplifications in cuda_cccl/pyproject.toml * Further simplify cuda_cccl/pyproject.toml, setup.py: remove inconsequential code * Make cuda_cooperative/pyproject.toml more similar to cuda_cccl/pyproject.toml * Add taplo-pre-commit to .pre-commit-config.yaml * taplo-pre-commit auto-fixes * Use pathlib in cuda_cooperative/setup.py * CCCL_PYTHON_PATH in cuda_cooperative/setup.py * Modernize cuda_parallel/pyproject.toml, setup.py * Use pathlib in cuda_parallel/setup.py * Add `# TOML lint & format` comment. * Replace MANIFEST.in with `[tool.setuptools.package-data]` section in pyproject.toml * Use pathlib in cuda/cccl/include_paths.py * pre-commit autoupdate (EXCEPT clang-format, which was manually restored) * Fixes after git merge main * Resolve warning: AttributeError: '_Reduce' object has no attribute 'build_result' ``` =========================================================================== warnings summary =========================================================================== tests/test_reduce.py::test_reduce_non_contiguous /home/coder/cccl/python/devenv/lib/python3.12/site-packages/_pytest/unraisableexception.py:85: PytestUnraisableExceptionWarning: Exception ignored in: <function _Reduce.__del__ at 0x7bf123139080> Traceback (most recent call last): File "/home/coder/cccl/python/cuda_parallel/cuda/parallel/experimental/algorithms/reduce.py", line 132, in __del__ bindings.cccl_device_reduce_cleanup(ctypes.byref(self.build_result)) ^^^^^^^^^^^^^^^^^ AttributeError: '_Reduce' object has no attribute 'build_result' warnings.warn(pytest.PytestUnraisableExceptionWarning(msg)) -- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html ============================================================= 1 passed, 93 deselected, 1 warning in 0.44s ============================================================== ``` * Move `copy_cccl_headers_to_cuda_cccl_include()` functionality to `class CustomBuildPy` * Introduce cuda_cooperative/constraints.txt * Also add cuda_parallel/constraints.txt * Add `--constraint constraints.txt` in ci/test_python.sh * Update Copyright dates * Switch to https://github.com/ComPWA/taplo-pre-commit (the other repo has been archived by the owner on Jul 1, 2024) For completeness: The other repo took a long time to install into the pre-commit cache; so long it lead to timeouts in the CCCL CI. * Remove unused cuda_parallel jinja2 dependency (noticed by chance). * Remove constraints.txt files, advertise running `pip install cuda-cccl` first instead. * Make cuda_cooperative, cuda_parallel testing completely independent. * Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc] * Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Fix sign-compare warning (NVIDIA#3408) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Revert "Try using another runner (because V100 runners seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc]" This reverts commit ea33a21. Error message: NVIDIA#3201 (comment) * Try using A100 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Also show cuda-cooperative site-packages, cuda-parallel site-packages (after pip install) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Try using l4 runner (because V100 runners still seem to be stuck) [skip-rapids][skip-matx][skip-docs][skip-vdc] * Restore original ci/matrix.yaml [skip-rapids] * Use for loop in test_python.sh to avoid code duplication. * Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci] * Comment out taplo-lint in pre-commit config [skip-rapids][skip-matx][skip-docs][skip-vdc] * Revert "Run only test_python.sh [skip-rapids][skip-matx][skip-docs][skip-vdc][skip pre-commit.ci]" This reverts commit ec206fd. * Implement suggestion by @shwina (NVIDIA#3201 (review)) * Address feedback by @leofang --------- Co-authored-by: Bernhard Manfred Gruber <[email protected]> cuda.parallel: Add optional stream argument to reduce_into() (NVIDIA#3348) * Add optional stream argument to reduce_into() * Add tests to check for reduce_into() stream behavior * Move protocol related utils to separate file and rework __cuda_stream__ error messages * Fix synchronization issue in stream test and add one more invalid stream test case * Rename cuda stream validation function after removing leading underscore * Unpack values from __cuda_stream__ instead of indexing * Fix linting errors * Handle TypeError when unpacking invalid __cuda_stream__ return * Use stream to allocate cupy memory in new stream test Upgrade to actions/deploy-pages@v4 (from v2), as suggested by @leofang (NVIDIA#3434) Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ (NVIDIA#3419) * Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ Fixes NVIDIA#3404 Fix CI issues (NVIDIA#3443) Remove deprecated `cub::min` (NVIDIA#3450) * Remove deprecated `cuda::{min,max}` * Drop unused `thrust::remove_cvref` file Fix typo in builtin (NVIDIA#3451) Moves agents to `detail::<algorithm_name>` namespace (NVIDIA#3435) uses unsigned offset types in thrust's scan dispatch (NVIDIA#3436) Default transform_iterator's copy ctor (NVIDIA#3395) Fixes: NVIDIA#2393 Turn C++ dialect warning into error (NVIDIA#3453) Uses unsigned offset types in thrust's sort algorithm calling into `DispatchMergeSort` (NVIDIA#3437) * uses thrust's dynamic dispatch for merge_sort * [pre-commit.ci] auto code formatting --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Refactor allocator handling of contiguous_storage (NVIDIA#3050) Co-authored-by: Michael Schellenberger Costa <[email protected]> Drop thrust::detail::integer_traits (NVIDIA#3391) Add cuda::is_floating_point supporting half and bfloat (NVIDIA#3379) Co-authored-by: Michael Schellenberger Costa <[email protected]> Improve docs of std headers (NVIDIA#3416) Drop C++11 and C++14 support for all of cccl (NVIDIA#3417) * Drop C++11 and C++14 support for all of cccl --------- Co-authored-by: Bernhard Manfred Gruber <[email protected]> Deprecate a few CUB macros (NVIDIA#3456) Deprecate thrust universal iterator categories (NVIDIA#3461) Fix launch args order (NVIDIA#3465) Add `--extended-lambda` to the list of removed clangd flags (NVIDIA#3432) add `_CCCL_HAS_NVFP8` macro (NVIDIA#3429) Add `_CCCL_BUILTIN_PREFETCH` (NVIDIA#3433) Drop universal iterator categories (NVIDIA#3474) Ensure that headers in `<cuda/*>` can be build with a C++ only compiler (NVIDIA#3472) Specialize __is_extended_floating_point for FP8 types (NVIDIA#3470) Also ensure that we actually can enable FP8 due to FP16 and BF16 requirements Co-authored-by: Michael Schellenberger Costa <[email protected]> Moves CUB kernel entry points to a detail namespace (NVIDIA#3468) * moves emptykernel to detail ns * second batch * third batch * fourth batch * fixes cuda parallel * concatenates nested namespaces Deprecate block/warp algo specializations (NVIDIA#3455) Fixes: NVIDIA#3409 Refactor CUB's util_debug (NVIDIA#3345)

Cleanup util_arch (NVIDIA#2773) Improves `DeviceSegmentedSort` test run time for large number of items and segments (NVIDIA#3246) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * fixes spelling * adds tests for large number of segments * fixes narrowing conversion in tests * addresses review comments * fixes includes Adds support for large number of items and large number of segments to `DeviceSegmentedSort` (NVIDIA#3308) * fixes segment offset generation * switches to analytical verification * switches to analytical verification for pairs * addresses review comments * introduces segment offset type * adds tests for large number of segments * adds support for large number of segments * drops segment offset type * fixes thrust namespace * removes about-to-be-deprecated cub iterators * no exec specifier on defaulted ctor * fixes gcc7 linker error * uses local_segment_index_t throughout * determine offset type based on type returned by segment iterator begin/end iterators * minor style improvements cuda.parallel: Support structured types as algorithm inputs (NVIDIA#3218) * Introduce gpu_struct decorator and typing * Enable `reduce` to accept arrays of structs as inputs * Add test for reducing arrays-of-struct * Update documentation * Use a numpy array rather than ctypes object * Change zeros -> empty for output array and temp storage * Add a TODO for typing GpuStruct * Documentation udpates * Remove test_reduce_struct_type from test_reduce.py * Revert to `to_cccl_value()` accepting ndarray + GpuStruct * Bump copyrights --------- Co-authored-by: Ashwin Srinath <[email protected]> Deprecate thrust::async (NVIDIA#3324) Fixes: NVIDIA#100 Review/Deprecate CUB `util.ptx` for CCCL 2.x (NVIDIA#3342) Deprecate thrust::numeric_limits (NVIDIA#3366) Upgrade to Catch2 3.8 (NVIDIA#3310) Fixes: NVIDIA#1724 Fix sign-compare warning (NVIDIA#3408) Implement more cmath functions to be usable on host and device (NVIDIA#3382) * Implement more cmath functions to be usable on host and device * Implement math roots functions * Implement exponential functions Redefine and deprecate thrust::remove_cvref (NVIDIA#3394) * Redefine and deprecate thrust::remove_cvref Co-authored-by: Michael Schellenberger Costa <[email protected]> cuda.parallel: Add optional stream argument to reduce_into() (NVIDIA#3348) * Add optional stream argument to reduce_into() * Add tests to check for reduce_into() stream behavior * Move protocol related utils to separate file and rework __cuda_stream__ error messages * Fix synchronization issue in stream test and add one more invalid stream test case * Rename cuda stream validation function after removing leading underscore * Unpack values from __cuda_stream__ instead of indexing * Fix linting errors * Handle TypeError when unpacking invalid __cuda_stream__ return * Use stream to allocate cupy memory in new stream test Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ (NVIDIA#3419) * Deprecate `cub::{min, max}` and replace internal uses with those from libcu++ Fixes NVIDIA#3404 Remove deprecated `cub::min` (NVIDIA#3450) * Remove deprecated `cuda::{min,max}` * Drop unused `thrust::remove_cvref` file Fix typo in builtin (NVIDIA#3451) Moves agents to `detail::<algorithm_name>` namespace (NVIDIA#3435) Drop thrust::detail::integer_traits (NVIDIA#3391) Add cuda::is_floating_point supporting half and bfloat (NVIDIA#3379) Co-authored-by: Michael Schellenberger Costa <[email protected]> add `_CCCL_HAS_NVFP8` macro (NVIDIA#3429) Specialize __is_extended_floating_point for FP8 types (NVIDIA#3470) Also ensure that we actually can enable FP8 due to FP16 and BF16 requirements Co-authored-by: Michael Schellenberger Costa <[email protected]> Moves CUB kernel entry points to a detail namespace (NVIDIA#3468) * moves emptykernel to detail ns * second batch * third batch * fourth batch * fixes cuda parallel * concatenates nested namespaces Deprecate block/warp algo specializations (NVIDIA#3455) Fixes: NVIDIA#3409 fix documentation

fbusato added 2 commits January 10, 2025 20:24

add deprecate warning

e961e07

try to remove LogicShiftLeft and LogicShiftRight

b3d2324

fbusato requested review from a team as code owners January 10, 2025 20:36

fbusato requested review from bernhardmgruber and wmaxey January 10, 2025 20:36

replace cub special registers usage

284ff7e

bernhardmgruber reviewed Jan 10, 2025

View reviewed changes

cub/cub/agent/agent_radix_sort_histogram.cuh Outdated Show resolved Hide resolved

fbusato and others added 3 commits January 10, 2025 23:34

deprecate unused functions

e7af385

fix laneid()

c1084d4

Merge branch 'main' into review-cub-util-ptx

691b271

bernhardmgruber reviewed Jan 10, 2025

View reviewed changes

cub/cub/warp/specializations/warp_reduce_shfl.cuh Outdated Show resolved Hide resolved

cub/test/test_util.h Outdated Show resolved Hide resolved

cub/cub/agent/agent_batch_memcpy.cuh Outdated Show resolved Hide resolved

fbusato and others added 2 commits January 11, 2025 00:37

fix lanemask_gt()

a88ef76

Merge branch 'main' into review-cub-util-ptx

70d105c

fbusato self-assigned this Jan 11, 2025

fbusato and others added 4 commits January 13, 2025 20:32

use cuda::std::terminate()

68a115f

typo fix

43b2b58

replace SHR_ADD with shift-add

0af70fd

Merge branch 'main' into review-cub-util-ptx

8d44adb

fbusato force-pushed the review-cub-util-ptx branch from 96cf032 to 8d44adb Compare January 13, 2025 23:07

fbusato added 2 commits January 13, 2025 23:29

replace voting instructions

a2fff97

replace cta sync instructions

73dabf6

fbusato requested a review from a team as a code owner January 14, 2025 00:50

fbusato requested a review from elstehle January 14, 2025 00:50

replace shuffle idx

3cf7cb5

fbusato force-pushed the review-cub-util-ptx branch from ee86833 to 3cf9263 Compare January 14, 2025 18:41

recover LogicShiftLeft and LogicShiftRight

e60f288

Signed-off-by: fbusato <[email protected]>

deprecate shfl functions

ffe65ca

Signed-off-by: fbusato <[email protected]>

fbusato force-pushed the review-cub-util-ptx branch from 3cf9263 to ffe65ca Compare January 14, 2025 19:01

Merge branch 'main' into review-cub-util-ptx

046f1e5

fbusato requested a review from bernhardmgruber January 14, 2025 19:51

bernhardmgruber approved these changes Jan 15, 2025

View reviewed changes

bernhardmgruber added cub For all items related to CUB backport branch/2.8.x labels Jan 15, 2025

fbusato merged commit 43fb061 into NVIDIA:main Jan 15, 2025
94 checks passed

bernhardmgruber pushed a commit to bernhardmgruber/cccl that referenced this pull request Jan 15, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (NVIDIA#3342)

baecc8e

bernhardmgruber linked an issue Jan 15, 2025 that may be closed by this pull request

Review and deprecate features from CUB util_ptx.cuh #3289

Closed

fbusato deleted the review-cub-util-ptx branch January 15, 2025 01:29

miscco pushed a commit that referenced this pull request Jan 15, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (#3342) (#3389)

9092760

Co-authored-by: Federico Busato <[email protected]>

shwina pushed a commit to shwina/cccl that referenced this pull request Jan 16, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (NVIDIA#3342)

e4a7ed3

fbusato mentioned this pull request Jan 16, 2025

Drop deprecated features from CUB util_ptx #3430

Closed

davebayer pushed a commit to davebayer/cccl that referenced this pull request Jan 18, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (NVIDIA#3342)

8369468

davebayer pushed a commit to davebayer/cccl that referenced this pull request Jan 18, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (NVIDIA#3342)

330e976

bernhardmgruber linked an issue Jan 21, 2025 that may be closed by this pull request

Drop deprecated features from CUB util_ptx #3430

Closed

davebayer pushed a commit to davebayer/cccl that referenced this pull request Jan 22, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (NVIDIA#3342)

5648bd7

davebayer pushed a commit to davebayer/cccl that referenced this pull request Jan 22, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (NVIDIA#3342)

0b8aa3c

davebayer pushed a commit to davebayer/cccl that referenced this pull request Jan 23, 2025

Review/Deprecate CUB util.ptx for CCCL 2.x (NVIDIA#3342)

4e5378d

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Review CUB `util.ptx` for CCCL 2.x #3342

Review CUB `util.ptx` for CCCL 2.x #3342

fbusato commented Jan 10, 2025 •

edited

Loading

bernhardmgruber left a comment •

edited

Loading

github-actions bot commented Jan 11, 2025

🟨 cub: Pass: 94%/38 | Total: 1d 09h | Avg: 53m 05s | Max: 1h 37m | Hits: 219%/3108

🟩 thrust: Pass: 100%/37 | Total: 19h 52m | Avg: 32m 13s | Max: 1h 06m | Hits: 215%/9260

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 13m 25s | Avg: 6m 42s | Max: 11m 15s

🟩 python: Pass: 100%/1 | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s

👃 Inspect Changes

Modifications in project?

Modifications in project or dependencies?

🏃‍ Runner counts (total jobs: 78)

github-actions bot commented Jan 14, 2025

🟨 cub: Pass: 94%/38 | Total: 1d 08h | Avg: 51m 25s | Max: 1h 15m | Hits: 112%/3120

🟩 thrust: Pass: 100%/37 | Total: 19h 46m | Avg: 32m 04s | Max: 1h 00m | Hits: 215%/9220

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 10m 00s | Avg: 5m 00s | Max: 7m 49s

🟩 python: Pass: 100%/1 | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s

👃 Inspect Changes

Modifications in project?

Modifications in project or dependencies?

🏃‍ Runner counts (total jobs: 78)

copy-pr-bot bot commented Jan 14, 2025

github-actions bot commented Jan 14, 2025

🟩 cub: Pass: 100%/38 | Total: 1d 08h | Avg: 51m 40s | Max: 1h 16m | Hits: 110%/3120

🟩 thrust: Pass: 100%/37 | Total: 20h 30m | Avg: 33m 15s | Max: 1h 06m | Hits: 213%/9220

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 9m 25s | Avg: 4m 42s | Max: 7m 16s

🟩 python: Pass: 100%/1 | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s

👃 Inspect Changes

Modifications in project?

Modifications in project or dependencies?

🏃‍ Runner counts (total jobs: 78)

github-actions bot commented Jan 15, 2025

Review CUB util.ptx for CCCL 2.x #3342

Review CUB util.ptx for CCCL 2.x #3342

Conversation

fbusato commented Jan 10, 2025 • edited Loading

Description

bernhardmgruber left a comment • edited Loading

Choose a reason for hiding this comment

github-actions bot commented Jan 11, 2025

🟨 cub: Pass: 94%/38 | Total: 1d 09h | Avg: 53m 05s | Max: 1h 37m | Hits: 219%/3108

🟩 thrust: Pass: 100%/37 | Total: 19h 52m | Avg: 32m 13s | Max: 1h 06m | Hits: 215%/9260

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 13m 25s | Avg: 6m 42s | Max: 11m 15s

🟩 python: Pass: 100%/1 | Total: 28m 25s | Avg: 28m 25s | Max: 28m 25s

👃 Inspect Changes

Modifications in project?

Modifications in project or dependencies?

🏃‍ Runner counts (total jobs: 78)

github-actions bot commented Jan 14, 2025

🟨 cub: Pass: 94%/38 | Total: 1d 08h | Avg: 51m 25s | Max: 1h 15m | Hits: 112%/3120

🟩 thrust: Pass: 100%/37 | Total: 19h 46m | Avg: 32m 04s | Max: 1h 00m | Hits: 215%/9220

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 10m 00s | Avg: 5m 00s | Max: 7m 49s

🟩 python: Pass: 100%/1 | Total: 25m 18s | Avg: 25m 18s | Max: 25m 18s

👃 Inspect Changes

Modifications in project?

Modifications in project or dependencies?

🏃‍ Runner counts (total jobs: 78)

copy-pr-bot bot commented Jan 14, 2025

github-actions bot commented Jan 14, 2025

🟩 cub: Pass: 100%/38 | Total: 1d 08h | Avg: 51m 40s | Max: 1h 16m | Hits: 110%/3120

🟩 thrust: Pass: 100%/37 | Total: 20h 30m | Avg: 33m 15s | Max: 1h 06m | Hits: 213%/9220

🟩 cccl_c_parallel: Pass: 100%/2 | Total: 9m 25s | Avg: 4m 42s | Max: 7m 16s

🟩 python: Pass: 100%/1 | Total: 29m 05s | Avg: 29m 05s | Max: 29m 05s

👃 Inspect Changes

Modifications in project?

Modifications in project or dependencies?

🏃‍ Runner counts (total jobs: 78)

github-actions bot commented Jan 15, 2025

Review CUB `util.ptx` for CCCL 2.x #3342

Review CUB `util.ptx` for CCCL 2.x #3342

fbusato commented Jan 10, 2025 •

edited

Loading

bernhardmgruber left a comment •

edited

Loading