Skip to content

WMMA BFloat16 (BF16) #1425

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Draft
wants to merge 2 commits into
base: master
Choose a base branch
from
Draft

WMMA BFloat16 (BF16) #1425

wants to merge 2 commits into from

Conversation

carstenbauer
Copy link
Member

Sister PR of #1419 which tries to add TensorFloat32 (TF32) support.

While TF32 has higher priority for me, I thought that, in the spirit of "while I'm at it", it would make sense to add BF16 support simultaneously.

cc: @HenriDeh, @thomasfaingnaert

@carstenbauer
Copy link
Member Author

carstenbauer commented Mar 3, 2022 •
edited
Loading

First test kernel:

function kernel_wmma_bf16_lowlevel(a_dev, b_dev, c_dev, d_dev)                                                                                   
    a_frag = WMMA.llvm_wmma_load_a_col_m16n16k16_global_stride_bf16(pointer(a_dev), 16)                                                          
    b_frag = WMMA.llvm_wmma_load_b_col_m16n16k16_global_stride_bf16(pointer(b_dev), 16)                                                          
    c_frag = WMMA.llvm_wmma_load_c_col_m16n16k16_global_stride_f32(pointer(c_dev), 16)                                                           
                                                                                                                                                
    d_frag = WMMA.llvm_wmma_mma_col_col_m16n16k16_bf16(a_frag, b_frag, c_frag)                                                                   
                                                                                                                                                
    WMMA.llvm_wmma_store_d_col_m16n16k16_global_stride_f32(pointer(d_dev), d_frag, 16)                                                           
    return nothing                                                                                                                               
end                                                                                                                                              
                                                                                                                                                        
function call_kernel()                                                                                                                           
    m = n = k = 16                                                                                                                               
    dtype_a = dtype_b = CUDA.BFloat16                                                                                                            
    dtype_c = dtype_d = Float32                                                                                                                  
                                                                                                                                                
    d_a = CUDA.rand(dtype_a, m, k)                                                                                                               
    d_b = CUDA.rand(dtype_b, k, n)                                                                                                               
    d_c = CUDA.rand(dtype_c, m, n)                                                                                                               
    d_d = CUDA.zeros(dtype_d, m, n)                                                                                                              
                                                                                                                                                
    CUDA.@sync @cuda kernel_wmma_bf16_lowlevel(d_a, d_b, d_c, d_d)                                                                               
    return nothing                                                                                                                               
end

Current output (on Julia 1.8-beta1):

julia> call_kernel()                                                                                                                                    
ERROR: CUDA error: an illegal memory access was encountered (code 700, ERROR_ILLEGAL_ADDRESS)                                                           
Stacktrace:                                                                                                                                             
 [1] throw_api_error(res::CUDA.cudaError_enum)                                                                                                          
   @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91                                                                   
 [2] isdone                                                                                                                                             
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:109 [inlined]                                                            
 [3] nonblocking_synchronize                                                                                                                            
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:139 [inlined]                                                            
 [4] nonblocking_synchronize                                                                                                                            
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/context.jl:325 [inlined]                                                           
 [5] device_synchronize()                                                                                                                               
   @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/context.jl:319                                                                
 [6] top-level scope                                                                                                                                    
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/initialization.jl:54                                                                       
                                                                                                                                                        
caused by: CUDA error: an illegal memory access was encountered (code 700, ERROR_ILLEGAL_ADDRESS)  
Stacktrace:                                                                                                                                     [86/791]
  [1] throw_api_error(res::CUDA.cudaError_enum)                                                                                                         
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91                                                                  
  [2] isdone                                                                                                                                            
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:109 [inlined]                                                           
  [3] nonblocking_synchronize                                                                                                                           
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:139 [inlined]                                                           
  [4] synchronize(stream::CuStream; blocking::Nothing)                                                                                                  
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:128                                                                
  [5] synchronize                                                                                                                                       
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:122 [inlined]                                                           
  [6] macro expansion                                                                                                                                   
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:186 [inlined]                                                                     
  [7] macro expansion                                                                                                                                   
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/error.jl:67 [inlined]                                                              
  [8] curandCreateGenerator(typ::CUDA.CURAND.curandRngType)                                                                                             
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/wrappers.jl:5                                                          
  [9] CUDA.CURAND.RNG(typ::CUDA.CURAND.curandRngType; stream::CuStream)
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/random.jl:13
 [10] RNG (repeats 2 times)
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/random.jl:13 [inlined]
 [11] #167
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:38 [inlined]
 [12] (::CUDA.APIUtils.var"#8#11"{CUDA.CURAND.var"#167#173", CUDA.APIUtils.HandleCache{CuContext, CUDA.CURAND.RNG}, CuContext})()
    @ CUDA.APIUtils /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/cache.jl:24
 [13] lock(f::CUDA.APIUtils.var"#8#11"{CUDA.CURAND.var"#167#173", CUDA.APIUtils.HandleCache{CuContext, CUDA.CURAND.RNG}, CuContext}, l::ReentrantLock)
    @ Base ./lock.jl:185
 [14] (::CUDA.APIUtils.var"#check_cache#9"{CUDA.APIUtils.HandleCache{CuContext, CUDA.CURAND.RNG}, CuContext})(f::CUDA.CURAND.var"#167#173")
    @ CUDA.APIUtils /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/cache.jl:22
 [15] pop!
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/cache.jl:46 [inlined]
 [16] (::CUDA.CURAND.var"#new_state#172")(cuda::NamedTuple{(:device, :context, :stream, :math_mode, :math_precision), Tuple{CuDevice, CuContext, CuStrea
m, CUDA.MathMode, Symbol}})
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:37
 [17] #170
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:51 [inlined]
 [18] get!(default::CUDA.CURAND.var"#170#176"{CUDA.CURAND.var"#new_state#172", NamedTuple{(:device, :context, :stream, :math_mode, :math_precision), Tu$
le{CuDevice, CuContext, CuStream, CUDA.MathMode, Symbol}}}, h::Dict{CuContext, NamedTuple{(:rng,), Tuple{CUDA.CURAND.RNG}}}, key::CuContext)
    @ Base ./dict.jl:481
 [19] default_rng()
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:50
 [20] curand_rng
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/random.jl:229 [inlined]
 [21] rand(T::Type{Float32}, dim1::Int64, dims::Int64)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/random.jl:249
 [22] call_kernel()
    @ Main ./REPL[2]:8
 [23] top-level scope
    @ REPL[3]:1
 [24] top-level scope
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/initialization.jl:52

@carstenbauer
Copy link
Member Author

I can't add the "cuda kernels" and "enhancement" labels (being a member of JuliaGPU is apparently not enough).

@carstenbauer
Copy link
Member Author

Comment: I'm not sure what the fragment type (map_ptx_to_jl_frag) and size (map_frag_sizes) should be...

@carstenbauer carstenbauer mentioned this pull request Mar 3, 2022
Draft
@carstenbauer
Copy link
Member Author

After latest commit (fixing the fragment sizes) I (still) get

julia> call_kernel()
ERROR: CUDA error: an illegal memory access was encountered (code 700, ERROR_ILLEGAL_ADDRESS)
Stacktrace:
 [1] throw_api_error(res::CUDA.cudaError_enum)
   @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91
 [2] isdone
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:109 [inlined]
 [3] nonblocking_synchronize
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:139 [inlined]
 [4] nonblocking_synchronize
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/context.jl:325 [inlined]
 [5] device_synchronize()
   @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/context.jl:319
 [6] top-level scope
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/initialization.jl:54

caused by: CUDA error: an illegal memory access was encountered (code 700, ERROR_ILLEGAL_ADDRESS)
Stacktrace:
  [1] throw_api_error(res::CUDA.cudaError_enum)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91
  [2] isdone
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:109 [inlined]
  [3] nonblocking_synchronize
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:139 [inlined]
  [4] synchronize(stream::CuStream; blocking::Nothing)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:128
  [5] synchronize
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:122 [inlined]
  [6] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:186 [inlined]
  [7] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/error.jl:67 [inlined]
  [8] curandCreateGenerator(typ::CUDA.CURAND.curandRngType)
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/wrappers.jl:5
  [9] CUDA.CURAND.RNG(typ::CUDA.CURAND.curandRngType; stream::CuStream)
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/random.jl:13
 [10] RNG (repeats 2 times)
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/random.jl:13 [inlined]
 [11] #167
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:38 [inlined]
 [12] (::CUDA.APIUtils.var"#8#11"{CUDA.CURAND.var"#167#173", CUDA.APIUtils.HandleCache{CuContext, CUDA.CURAND.RNG}, CuContext})()
julia> 
WARNING: Error while freeing DeviceBuffer(512 bytes at 0x0000000a02000200):
CUDA.CuError(code=CUDA.cudaError_enum(0x000002bc), meta=nothing)

Stacktrace:
  [1] throw_api_error(res::CUDA.cudaError_enum)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91
  [2] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:101 [inlined]
  [3] cuMemFreeAsync(dptr::CUDA.Mem.DeviceBuffer, hStream::CuStream)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/call.jl:26
  [4] #free#2
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/memory.jl:97 [inlined]
  [5] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:58 [inlined]
  [6] macro expansion
    @ ./timing.jl:359 [inlined]
  [7] #actual_free#189
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:57 [inlined]
  [8] #_free#207
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:323 [inlined]
  [9] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:288 [inlined]
 [10] macro expansion
    @ ./timing.jl:359 [inlined]
 [11] #free#206
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:287 [inlined]
 [12] #212
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/array.jl:79 [inlined]
 [13] context!(f::CUDA.var"#212#213"{CuArray{Float16, 2, CUDA.Mem.DeviceBuffer}, CuStream}, ctx::CuContext; skip_destroyed::Bool)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/state.jl:164
 [14] unsafe_free!(xs::CuArray{Float16, 2, CUDA.Mem.DeviceBuffer}, stream::CuStream)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/array.jl:78
 [15] unsafe_finalize!(xs::CuArray{Float16, 2, CUDA.Mem.DeviceBuffer})
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/array.jl:99
error in running finalizer: CUDA.CuError(code=CUDA.cudaError_enum(0x000002bc), meta=nothing)
throw_api_error at /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91
macro expansion at /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:101 [inlined]
cuModuleUnload at /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/call.jl:26
#27 at /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/module.jl:82 [inlined]
#context!#63 at /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/state.jl:164
unknown function (ip: 0x7fa4a50fa68b)
_jl_invoke at /buildworker/worker/package_linux64/build/src/gf.c:2340 [inlined]
ijl_apply_generic at /buildworker/worker/package_linux64/build/src/gf.c:2522
context!##kw at /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/state.jl:161
unsafe_unload! at /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/module.jl:81
unknown function (ip: 0x7fa4a50fa212)
_jl_invoke at /buildworker/worker/package_linux64/build/src/gf.c:2340 [inlined]
ijl_apply_generic at /buildworker/worker/package_linux64/build/src/gf.c:2522
jl_apply at /buildworker/worker/package_linux64/build/src/julia.h:1825 [inlined]
run_finalizer at /buildworker/worker/package_linux64/build/src/gc.c:280
jl_gc_run_finalizers_in_list at /buildworker/worker/package_linux64/build/src/gc.c:367
run_finalizers at /buildworker/worker/package_linux64/build/src/gc.c:396
ijl_atexit_hook at /buildworker/worker/package_linux64/build/src/init.c:236
jl_repl_entrypoint at /buildworker/worker/package_linux64/build/src/jlapi.c:707
main at julia-beta (unknown line)
__libc_start_main at /lib64/libc.so.6 (unknown line)
unknown function (ip: 0x400808)
WARNING: Error while freeing DeviceBuffer(512 bytes at 0x0000000a02000000):
CUDA.CuError(code=CUDA.cudaError_enum(0x000002bc), meta=nothing)

Stacktrace:
  [1] throw_api_error(res::CUDA.cudaError_enum)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91
  [2] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:101 [inlined]
  [3] cuMemFreeAsync(dptr::CUDA.Mem.DeviceBuffer, hStream::CuStream)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/call.jl:26
  [4] #free#2
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/memory.jl:97 [inlined]
  [5] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:58 [inlined]
  [6] macro expansion
    @ ./timing.jl:359 [inlined]
  [7] #actual_free#189
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:57 [inlined]
  [8] #_free#207
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:323 [inlined]
  [9] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:288 [inlined]
 [10] macro expansion
    @ ./timing.jl:359 [inlined]
 [11] #free#206
julia> call_kernel()
ERROR: CUDA error: an illegal memory access was encountered (code 700, ERROR_ILLEGAL_ADDRESS)
Stacktrace:
 [1] throw_api_error(res::CUDA.cudaError_enum)
   @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91
 [2] isdone
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:109 [inlined]
 [3] nonblocking_synchronize
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:139 [inlined]
 [4] nonblocking_synchronize
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/context.jl:325 [inlined]
 [5] device_synchronize()
   @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/context.jl:319
 [6] top-level scope
   @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/initialization.jl:54

caused by: CUDA error: an illegal memory access was encountered (code 700, ERROR_ILLEGAL_ADDRESS)
Stacktrace:
  [1] throw_api_error(res::CUDA.cudaError_enum)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/error.jl:91
  [2] isdone
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:109 [inlined]
  [3] nonblocking_synchronize
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:139 [inlined]
  [4] synchronize(stream::CuStream; blocking::Nothing)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:128
  [5] synchronize
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/cudadrv/stream.jl:122 [inlined]
  [6] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/pool.jl:186 [inlined]
  [7] macro expansion
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/error.jl:67 [inlined]
  [8] curandCreateGenerator(typ::CUDA.CURAND.curandRngType)
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/wrappers.jl:5
  [9] CUDA.CURAND.RNG(typ::CUDA.CURAND.curandRngType; stream::CuStream)
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/random.jl:13
 [10] RNG (repeats 2 times)
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/random.jl:13 [inlined]
 [11] #167
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:38 [inlined]
 [12] (::CUDA.APIUtils.var"#8#11"{CUDA.CURAND.var"#167#173", CUDA.APIUtils.HandleCache{CuContext, CUDA.CURAND.RNG}, CuContext})()
    @ CUDA.APIUtils /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/cache.jl:24
 [13] lock(f::CUDA.APIUtils.var"#8#11"{CUDA.CURAND.var"#167#173", CUDA.APIUtils.HandleCache{CuContext, CUDA.CURAND.RNG}, CuContext}, l::ReentrantLock)
    @ Base ./lock.jl:185
 [14] (::CUDA.APIUtils.var"#check_cache#9"{CUDA.APIUtils.HandleCache{CuContext, CUDA.CURAND.RNG}, CuContext})(f::CUDA.CURAND.var"#167#173")
    @ CUDA.APIUtils /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/cache.jl:22
 [15] pop!
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/utils/cache.jl:46 [inlined]
 [16] (::CUDA.CURAND.var"#new_state#172")(cuda::NamedTuple{(:device, :context, :stream, :math_mode, :math_precision), Tuple{CuDevice, CuContext, CuStream, CUDA.MathMode, Symbol}})
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:37
 [17] #170
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:51 [inlined]
 [18] get!(default::CUDA.CURAND.var"#170#176"{CUDA.CURAND.var"#new_state#172", NamedTuple{(:device, :context, :stream, :math_mode, :math_precision), Tuple{CuDevice, CuContext, CuStream, CUDA.MathMode, Symbol}}}, h::Dict{CuContext, NamedTuple{(:rng,), Tuple{CUDA.CURAND.RNG}}}, key::CuContext)
    @ Base ./dict.jl:481
 [19] default_rng()
    @ CUDA.CURAND /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/lib/curand/CURAND.jl:50
 [20] curand_rng
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/random.jl:229 [inlined]
 [21] rand(T::Type{Float32}, dim1::Int64, dims::Int64)
    @ CUDA /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/random.jl:249
 [22] call_kernel()
    @ Main ./REPL[3]:8
 [23] top-level scope
    @ REPL[4]:1
 [24] top-level scope
    @ /scratch/pc2-mitarbeiter/bauerc/devel/CUDA_bfloat16/src/initialization.jl:52

@carstenbauer carstenbauer mentioned this pull request Mar 4, 2022
Closed
@maleadt maleadt force-pushed the master branch 15 times, most recently from 5d585c4 to c850163 Compare December 20, 2024 08:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@carstenbauer