Merge pull request #768 from JuliaGPU/tb/cuda

Update to CUDA 11.2 Update 2.

Author: maleadt

Artifacts.toml

@@ -151,35 +151,35 @@ lazy = true
 
 [[CUDA112]]
 arch = "powerpc64le"
-git-tree-sha1 = "ef3928da3f9b68a5213a93f91da0d27e32c01e50"
+git-tree-sha1 = "a0b84558b5a0c78f4d489f7b421ec4722899d6c0"
 libc = "glibc"
-os = "linux"
 lazy = true
+os = "linux"
 
     [[CUDA112.download]]
-    sha256 = "770235b69868b88e6db4efc30a8659e9708f3b432028e2032ba589cf2c3efaf8"
-    url = "https://github.com/JuliaBinaryWrappers/CUDA_jll.jl/releases/download/CUDA-v11.2.1+1/CUDA.v11.2.1.powerpc64le-linux-gnu.tar.gz"
+    sha256 = "23e3e4b0d0b325e223edd55e5b7dcc8d2a11722d0169b1dcd1a4a548e3f4d9c6"
+    url = "https://github.com/JuliaBinaryWrappers/CUDA_jll.jl/releases/download/CUDA-v11.2.2+0/CUDA.v11.2.2.powerpc64le-linux-gnu.tar.gz"
 
 [[CUDA112]]
 arch = "x86_64"
-git-tree-sha1 = "18f4e83091aec02d8229c2b009a45a5c22b47664"
+git-tree-sha1 = "53345f55f74b563606017c90cbafd05b61cd1768"
 libc = "glibc"
-os = "linux"
 lazy = true
+os = "linux"
 
     [[CUDA112.download]]
-    sha256 = "6da495c82fae19e0aae8691addc72829376547543324358f39e16835cb208e6e"
-    url = "https://github.com/JuliaBinaryWrappers/CUDA_jll.jl/releases/download/CUDA-v11.2.1+1/CUDA.v11.2.1.x86_64-linux-gnu.tar.gz"
+    sha256 = "a644bda2c4d7936d220ddbbe5719a2db5047659f23cc35b53bc4c9f7f6211711"
+    url = "https://github.com/JuliaBinaryWrappers/CUDA_jll.jl/releases/download/CUDA-v11.2.2+0/CUDA.v11.2.2.x86_64-linux-gnu.tar.gz"
 
 [[CUDA112]]
 arch = "x86_64"
-git-tree-sha1 = "4765905e93e1e93ca8d2eb52a1e8cec5de4627b1"
-os = "windows"
+git-tree-sha1 = "81069bd6e35b73f32622f78c73f5f32c3db4e47a"
 lazy = true
+os = "windows"
 
     [[CUDA112.download]]
-    sha256 = "6dc0ae6aab8b878864bf926fd9446c71f92f689e6115d6dcedc54ac492d30ea3"
-    url = "https://github.com/JuliaBinaryWrappers/CUDA_jll.jl/releases/download/CUDA-v11.2.1+1/CUDA.v11.2.1.x86_64-w64-mingw32.tar.gz"
+    sha256 = "36d22665027bca6eb9a292fd542344e8877b5089b2dacc29d0e334320d49e840"
+    url = "https://github.com/JuliaBinaryWrappers/CUDA_jll.jl/releases/download/CUDA-v11.2.2+0/CUDA.v11.2.2.x86_64-w64-mingw32.tar.gz"
 
 
 

deps/bindeps.jl

@@ -126,7 +126,7 @@ lazy_artifact(x) = @artifact_str(x)
 # NOTE: we don't use autogenerated JLLs, because we have multiple artifacts and need to
 #       decide at run time (i.e. not via package dependencies) which one to use.
 const cuda_artifacts = Dict(
-    (release=v"11.2", version=v"11.2.1",   preferred=true)  => ()->lazy_artifact("CUDA112"),
+    (release=v"11.2", version=v"11.2.2",   preferred=true)  => ()->lazy_artifact("CUDA112"),
     (release=v"11.1", version=v"11.1.1",   preferred=true)  => ()->lazy_artifact("CUDA111"),
     (release=v"11.0", version=v"11.0.3",   preferred=true)  => ()->lazy_artifact("CUDA110"),
     (release=v"10.2", version=v"10.2.89",  preferred=true)  => ()->lazy_artifact("CUDA102"),

deps/discovery.jl

@@ -257,6 +257,20 @@ const cuda_library_versions = Dict(
         "npp"       => v"11.3.2", #.139
         "nvjpeg"    => v"11.4.0", #.135
     ),
+    v"11.2.2" => Dict(
+        "cudart"    => v"11.2.152",
+        "cupti"     => "2020.3.1", # docs mention 11.2.152
+        "nvrtc"     => v"11.2.152",
+        "nvtx"      => v"11.2.152",
+        "nvvp"      => v"11.2.152",
+        "cublas"    => v"11.4.1", #.1043
+        "cufft"     => v"10.4.1", #.152
+        "curand"    => v"10.2.3", #.152
+        "cusolver"  => v"11.1.0", #.152
+        "cusparse"  => v"11.4.1", #.1152
+        "npp"       => v"11.3.2", #.152
+        "nvjpeg"    => v"11.4.0", #.152
+    ),
 )
 
 function cuda_library_version(library, toolkit_version)
@@ -301,6 +315,9 @@ const cuda_binary_versions = Dict(
     v"11.2.1" => Dict(
         "nvdisasm"  => v"11.2.135"
     ),
+    v"11.2.2" => Dict(
+        "nvdisasm"  => v"11.2.152"
+    ),
 )
 
 # simplified find_library/find_binary entry-points,

lib/cublas/libcublas_common.jl

@@ -3,7 +3,7 @@
 const CUBLAS_VER_MAJOR = 11
 const CUBLAS_VER_MINOR = 4
 const CUBLAS_VER_PATCH = 1
-const CUBLAS_VER_BUILD = 1026
+const CUBLAS_VER_BUILD = 1043
 const CUBLAS_VERSION = CUBLAS_VER_MAJOR * 1000 + CUBLAS_VER_MINOR * 100 + CUBLAS_VER_PATCH
 
 @cenum cublasStatus_t::UInt32 begin

lib/cufft/libcufft_common.jl

@@ -4,9 +4,9 @@
 
 const CUFFT_VER_MAJOR = 10
 const CUFFT_VER_MINOR = 4
-const CUFFT_VER_PATCH = 0
-const CUFFT_VER_BUILD = 135
-const CUFFT_VERSION = 10400
+const CUFFT_VER_PATCH = 1
+const CUFFT_VER_BUILD = 152
+const CUFFT_VERSION = 10401
 const MAX_CUFFT_ERROR = 0x11
 const CUFFT_FORWARD = -1
 const CUFFT_INVERSE = 1

lib/curand/libcurand_common.jl

@@ -3,7 +3,7 @@
 const CURAND_VER_MAJOR = 10
 const CURAND_VER_MINOR = 2
 const CURAND_VER_PATCH = 3
-const CURAND_VER_BUILD = 135
+const CURAND_VER_BUILD = 152
 const CURAND_VERSION = CURAND_VER_MAJOR * 1000 + CURAND_VER_MINOR * 100 + CURAND_VER_PATCH
 
 @cenum curandStatus::UInt32 begin

lib/cusolver/libcusolver_common.jl

@@ -3,7 +3,7 @@
 const CUSOLVER_VER_MAJOR = 11
 const CUSOLVER_VER_MINOR = 1
 const CUSOLVER_VER_PATCH = 0
-const CUSOLVER_VER_BUILD = 135
+const CUSOLVER_VER_BUILD = 152
 const CUSOLVER_VERSION = CUSOLVER_VER_MAJOR * 1000 + CUSOLVER_VER_MINOR * 100 + CUSOLVER_VER_PATCH
 const cusolverDnContext = Cvoid
 const cusolverDnHandle_t = Ptr{cusolverDnContext}

lib/cusparse/libcusparse_common.jl

@@ -2,8 +2,8 @@
 
 const CUSPARSE_VER_MAJOR = 11
 const CUSPARSE_VER_MINOR = 4
-const CUSPARSE_VER_PATCH = 0
-const CUSPARSE_VER_BUILD = 135
+const CUSPARSE_VER_PATCH = 1
+const CUSPARSE_VER_BUILD = 1152
 const CUSPARSE_VERSION = CUSPARSE_VER_MAJOR * 1000 + CUSPARSE_VER_MINOR * 100 + CUSPARSE_VER_PATCH
 
 # Skipping MacroDefinition: CUSPARSE_DEPRECATED ( new_func ) __attribute__ ( ( deprecated ( "please use " # new_func " instead" ) ) )

lib/nvml/libnvml.jl

@@ -1601,4 +1601,11 @@ end
     ccall((:nvmlDeviceSetTemperatureThreshold, libnvml()), nvmlReturn_t, (nvmlDevice_t, nvmlTemperatureThresholds_t, Ptr{Cint}), device, thresholdType, temp)
 end
 
+## Added in CUDA 11.2 Update 2
+
+@checked function nvmlDeviceCreateGpuInstanceWithPlacement(device, profileId, placement, gpuInstance)
+    initialize_api()
+    ccall((:nvmlDeviceCreateGpuInstanceWithPlacement, libnvml), nvmlReturn_t, (nvmlDevice_t, UInt32, Ptr{nvmlGpuInstancePlacement_t}, Ptr{nvmlGpuInstance_t}), device, profileId, placement, gpuInstance)
+end
+
 ##

lib/nvml/libnvml_common.jl

@@ -262,14 +262,16 @@ const NVML_GPU_INSTANCE_PROFILE_3_SLICE = 0x02
 const NVML_GPU_INSTANCE_PROFILE_4_SLICE = 0x03
 const NVML_GPU_INSTANCE_PROFILE_7_SLICE = 0x04
 const NVML_GPU_INSTANCE_PROFILE_8_SLICE = 0x05
-const NVML_GPU_INSTANCE_PROFILE_COUNT = 0x06
+const NVML_GPU_INSTANCE_PROFILE_6_SLICE = 0x06
+const NVML_GPU_INSTANCE_PROFILE_COUNT = 0x07
 const NVML_COMPUTE_INSTANCE_PROFILE_1_SLICE = 0x00
 const NVML_COMPUTE_INSTANCE_PROFILE_2_SLICE = 0x01
 const NVML_COMPUTE_INSTANCE_PROFILE_3_SLICE = 0x02
 const NVML_COMPUTE_INSTANCE_PROFILE_4_SLICE = 0x03
 const NVML_COMPUTE_INSTANCE_PROFILE_7_SLICE = 0x04
 const NVML_COMPUTE_INSTANCE_PROFILE_8_SLICE = 0x05
-const NVML_COMPUTE_INSTANCE_PROFILE_COUNT = 0x06
+const NVML_COMPUTE_INSTANCE_PROFILE_6_SLICE = 0x06
+const NVML_COMPUTE_INSTANCE_PROFILE_COUNT = 0x07
 const NVML_COMPUTE_INSTANCE_ENGINE_PROFILE_SHARED = 0x00
 const NVML_COMPUTE_INSTANCE_ENGINE_PROFILE_COUNT = 0x01
 const nvmlDevice_st = Cvoid
@@ -531,7 +533,17 @@ const nvmlEnableState_t = nvmlEnableState_enum
     NVML_BRAND_GRID = 4
     NVML_BRAND_GEFORCE = 5
     NVML_BRAND_TITAN = 6
-    NVML_BRAND_COUNT = 7
+    NVML_BRAND_NVIDIA_VAPPS = 7
+    NVML_BRAND_NVIDIA_VPC = 8
+    NVML_BRAND_NVIDIA_VCS = 9
+    NVML_BRAND_NVIDIA_VWS = 10
+    NVML_BRAND_NVIDIA_VGAMING = 11
+    NVML_BRAND_QUADRO_RTX = 12
+    NVML_BRAND_NVIDIA_RTX = 13
+    NVML_BRAND_NVIDIA = 14
+    NVML_BRAND_GEFORCE_RTX = 15
+    NVML_BRAND_TITAN_RTX = 16
+    NVML_BRAND_COUNT = 17
 end
 
 const nvmlBrandType_t = nvmlBrandType_enum