improvements for version 0.2.0

Author: decarvalhojunior-fh

Project.toml

@@ -1,7 +1,7 @@
 name = "PlatformAware"
 uuid = "e7c50b67-2c03-471e-9cf2-69e515d86ecf"
 authors = ["Francisco Heron de Carvalho Junior <heron@dc.ufc.br> and contributors"]
-version = "0.1.0"
+version = "0.2.0"
 
 [deps]
 Artifacts = "56f22d72-fd6d-98f1-02f0-08ddc0907c33"

src/awareness.jl

@@ -40,7 +40,7 @@ global processor_dict = merge(processor_dict_amd, processor_dict_intel)
 global accelerator_dict = merge(accelerator_dict_intel, accelerator_dict_amd, accelerator_dict_nvidia)
 
 function get_info_dict(idtype)
-   command = `sudo lshw -xml -C $idtype`    
+   command = `sudo lshw -xml -quiet -C $idtype`    
    xmlinfo = read(command, String)
    xml_dict(xmlinfo)
 end
@@ -400,11 +400,40 @@ end
 end
 =#
 
+function collectProcessorFeaturesDefault()
+
+   processor_features = Dict()
+
+   processor_features["processor_count"] = 1
+   processor_features["processor_core_count"] = 1
+   processor_features["processor_core_threads_count"] = 1
+   processor_features["processor_core_clock"] = "unset"
+   processor_features["processor_simd"] = "unset"
+   processor_features["processor_core_L1_size"] = "unset"
+   processor_features["processor_core_L2_size"] = "unset"
+   processor_features["processor_L3_size"] = "unset"
+   processor_features["processor_manufacturer"] = "unset"
+   processor_features["processor_tdp"] = "unset"
+   processor_features["processor"] = "unset"
+
+   return processor_features
+
+end
+
 # using CpuId (safe)
 
 function identifyProcessor()
-
-   collectProcessorFeatures_CpuId() 
+   try      
+      processor_features = collectProcessorFeatures_CpuId() 
+      println(stderr, "ok")
+      return processor_features
+   catch error
+      println(stderr, "fail.")
+      println(stderr, "=> Error: detecting processor features failed. Using default features.")
+      println(stderr, "          You can setup processor features manually.")
+      println(stderr, error)
+      return collectProcessorFeaturesDefault()
+   end
 
 #=
    l = Vector()
@@ -464,13 +493,40 @@ function collectAcceleratorFeatures(l)
    return i == 2 ? accelerator_features["1"] : accelerator_features
 end
 
+function collectAcceleratorFeaturesDefault()
+
+      default_features = Dict()
+
+      default_features["accelerator_count"] = 0
+      default_features["accelerator"] = "unset"
+      default_features["accelerator_type"] = "unset"
+      default_features["accelerator_manufacturer"] = "unset"
+      default_features["accelerator_api"] = "unset"
+      default_features["accelerator_architecture"] = "unset"
+      default_features["accelerator_memorysize"] = "unset"
+      default_features["accelerator_tdp"] = "unset"
+
+    return default_features
+end
+
 function identifyAccelerator()
-   l = Vector()
-   for d in identifyComponent("display")
-      push!(l,d["vendor"] * " " * d["product"])
-   end
+   try
+      
+      l = Vector()
+      for d in identifyComponent("display")
+         push!(l,d["vendor"] * " " * d["product"])
+      end
 
-   collectAcceleratorFeatures(l)
+      accelerator_features = collectAcceleratorFeatures(l)
+      println(stderr, "ok")
+      return accelerator_features
+   catch error
+      println(stderr, "fail.")
+      println(stderr, "=> Error: detecting accelerator features failed. Using default features.")
+      println(stderr, "          You can setup accelerator features manually.")
+      println(stderr, error)
+      return collectAcceleratorFeaturesDefault()
+   end
 end
 
 #=
@@ -580,11 +636,15 @@ function identifyMemory()
             end
          end
       end
-
-      return collectMemoryFeatures(d1)
+      
+      memory_features = collectMemoryFeatures(d1)
+      println(stderr, "ok")
+      return memory_features
 
    catch error
-      println(stderr, "Error fetching memory info. Loading default values.")
+      println(stderr, "fail.")
+      println(stderr, "=> Error: detecting primary memory features failed. Using default features.")
+      println(stderr, "          You can setup primary memory features manually.")
       println(stderr,error)
       return collectMemoryFeaturesDefault()
    end
@@ -638,7 +698,9 @@ function identifyStorage()
          end
       end
    catch error
-      println(stderr, "Error fetching storage info. Loading default values.")
+      println(stderr, "fail.")
+      println(stderr, "=> Error: detection of storage features failed. Using default features.")
+      println(stderr, "          You can setup storage features manually.")
       println(stderr, error)
 
       # default
@@ -656,7 +718,9 @@ end
 # TODO
 function identityInterconnection()
 
-   println("NOTE: The identification of interconnection features is not yet implemented. Using defaults.")
+   println(stderr, "fail.")
+   println(stderr, "=> Note: detection of interconnection features (cluster computing) not yet implemented. Using default features. You can setup manually.")
+   println(stderr, "         You can setup interconnection features manually.")
 
    interconnection_features = Dict()
 
@@ -673,14 +737,16 @@ end
 # TODO
 function identifyNode()
 
-   println("NOTE: The identification of node features is not yet implemented. Using defaults.")
+   println(stderr, "fail.")
+   println(stderr, "=> Note: detection of node features (cluster and cloud computing) not yet implemented. Using default features.")
+   println(stderr, "         You can setup node features manually.")
 
    interconnection_features = Dict()
 
    interconnection_features["node_count"] = 1
    interconnection_features["node_provider"] = "OnPremise"
    interconnection_features["node_virtual"] = "No"
-   interconnection_features["node_virtual"] = "Yes"
+   interconnection_features["node_dedicated"] = "No"
    interconnection_features["node_machinefamily"] = "unset"
    interconnection_features["node_machinetype"] = "unset"
    interconnection_features["node_machinesize"] = "unset"
@@ -716,12 +782,12 @@ end
 
 function setup()
 
-   print("indentifying node... "); node_features = identifyNode(); println(stderr, "ok")
-   print("indentifying processor... "); processor_features = identifyProcessor(); println(stderr, "ok")
-   print("indentifying accelerator... "); accelerator_features = identifyAccelerator(); println(stderr, "ok")
-   print("indentifying memory... "); memory_features = identifyMemory(); println(stderr, "ok")
-   print("indentifying storage... "); storage_features = identifyStorage(); println(stderr, "ok")
-   print("indentifying interconnection... "); interconnection_features = identityInterconnection(); println(stderr, "ok")
+   print("identifying node... "); node_features = identifyNode()
+   print("identifying processor... "); processor_features = identifyProcessor()
+   print("identifying accelerator... "); accelerator_features = identifyAccelerator()
+   print("identifying memory... "); memory_features = identifyMemory()
+   print("identifying storage... "); storage_features = identifyStorage()
+   print("identifying interconnection... "); interconnection_features = identityInterconnection()
 
    platform_features = Dict()
 
@@ -749,12 +815,12 @@ function setup()
       println("Platform.toml will be searched in the following locations:")
       println("  1) A file path pointed by a PLATFORM_DESCRIPTION environment variable;")
       println("  2) The current directory;")
-      println("  3) The /etc/conf directory.")
+      println("  3) The /etc directory.")
    else
       TOML.print(stdout, platform_features)
       println(stderr)
-      println(stderr, "Platform description file already exists in the current folder (Platform.toml)")
-      println(stderr, "You must erase or move it before to create a new one.")
+      println(stderr, "Platform description file already exists in the current folder (Platform.toml).")
+      println(stderr, "You must delete or move it before creating a new one.")
    end
 
 end

src/features.jl

@@ -86,12 +86,11 @@ function readPlatormDescription()
                 contents
             catch
                 println(stderr,"The platform description file (Platform.toml) was not found.")
-                println(stderr,"Using default platform settings (calling only default kernels).")
-                println(stderr,"A Platform.toml file may be created by calling PlatformAware.setup(). Do it !")
+                println(stderr,"Using default platform features (calling default kernels).")
+                println(stderr,"A Platform.toml file may be created by calling PlatformAware.setup()")
 
                 io = joinpath(artifact"default_platform_description", "DefaultPlatform.toml")                
-                contents = read(io,String)
-                contents
+                read(io,String)
             end
          end
 

src/platforms/general.jl

@@ -24,7 +24,6 @@ abstract type MachineSize end
 
 abstract type Locale end
 
-
 # manufacturer
 
 abstract type Manufacturer end
@@ -54,7 +53,6 @@ abstract type FPGA <: AcceleratorType end
 
 abstract type MIC <: AcceleratorType end
 
-
 #interconnection
 abstract type InterconnectionTopology end
 abstract type Interconnection end
@@ -64,8 +62,6 @@ abstract type Interconnection end
 abstract type StorageType end   
 abstract type StorageInterface end   
 
-
-
 # memory system
 
 abstract type MemoryType end

src/platforms/nvidia/db-accelerators.NVIDIA.csv

@@ -45,4 +45,4 @@ NVIDIA;Tesla,A30,NVIDIATesla_A30,GPU,NVIDIA,CUDA_8_0,Ampere,24G,165,1,GA100
 NVIDIA;Tesla,A2,NVIDIATesla_A2,GPU,NVIDIA,CUDA_8_6,Ampere,24G,60,1,GA107
 NVIDIA;Tesla,H100,NVIDIATesla_H100,GPU,NVIDIA,CUDA_9_0,Hopper,80G,350,1,GH100
 NVIDIA;Tesla,H100,NVIDIATesla_H100,GPU,NVIDIA,CUDA_8_0,Hopper,80G,700,1,GH100
-NVIDIA;GeForce,940MX,NVIDIAGeForce_940MX,GPU,NVIDIA,CUDA_5_0,Maxwell,4G,23,1,GM108M
+NVIDIA;GeForce,940MX,NVIDIAGeForce_940MX,GPU,NVIDIA,CUDA_5_0,Maxwell,4G,23,1,GM108M

src/platforms/xilinx/xilinx.jl

@@ -4,4 +4,6 @@
 
 abstract type Xilinx <: Manufacturer end; export Xilinx
 
-abstract type UltrascalePlus_HBM_FPGA <: AcceleratorType end; export UltrascalePlus_HBM_FPGA
+abstract type UltrascalePlus_HBM_FPGA <: AcceleratorType end; export UltrascalePlus_HBM_FPGA
+
+#TODO 

src/quantifiers/atleast.jl

@@ -136,28 +136,4 @@ abstract type AtLeast512E <: AtLeast256E end   # 2^69
 
 # ...
 
-abstract type AtLeastInf <: AtLeast512E end    # ∞
-
-
-
-
-
-
-
-#=
-quantifiers = Integer[]
-
-function AtLeast(n::Integer)
-    idx = findfirst(x -> x > n, quantifiers)
-    if (isnothing(idx))
-        idx = size(l) + 1
-        push!(quantifiers,n)
-    elseif 
-        splice!(quantifiers,idx,[n,quantifiers[idx]])
-    end
-    if (idx-1 > 0)
-        super_n = quantifiers[idx-1]
-
-    eval(Meta.parse)
-end
-=#
+abstract type AtLeastInf <: AtLeast512E end    # ∞

test/basics.jl

@@ -27,7 +27,7 @@
     # specify platform-aware implementations
     @platform aware function kernel({accelerator_count::(@atleast 1)},
                                     x,y,args...; z=1, kwargs...)
-        println(z,": kernel for 1 accelerators")
+        println(z,": kernel for 1 accelerators of unspecified kind")
     end
     @platform aware function kernel({accelerator_count::(@atleast 1),
                                      accelerator_manufacturer::NVIDIA},
@@ -37,13 +37,13 @@
     @platform aware function kernel({node_count::(@atleast 32),
                                      processor::IntelCore_i7_7500U},
                                     x,y,args...; z=3, kwargs...)
-        println(z,": kernel for 1 NVIDIA accelerators")
+        println(z,": kernel optimized to the features of clusters with at least 32 nodes with Intel(R) Core(TM) i7-7500U processors")
     end
     @platform aware function kernel({accelerator_count::(@just 4),
                                      accelerator_manufacturer::NVIDIA,
                                      accelerator_architecture::Turing},
                                     x,y,args...; z=4, kwargs...)
-        println(z,": kernel for 4 accelerators of NVIDIA's Turing architecture")
+        println(z,": kernel for exactly 4 accelerators of NVIDIA's Turing architecture")
     end
     @platform aware function kernel({node_count::(@between 8 16),
                                      node_memory_size::(@atleast 16G),
@@ -53,16 +53,28 @@
                                      interconnection_bandwidth::(@atleast 128G)                                     
                                     },
                                     x,y,args...; z=5, kwargs...)
-        println(z,": kernel for a cluster of 8-16 nodes having at least 2 processors each wih at least 8 cores ...")
+        println(z,": kernel tuned for a cluster of 8 to 16 nodes having at least 2 processors with at least 8 cores each,")
+        println(z,": connected through an intereconnection having at most 32us latency and at least 128Gbs bandwidth.")
     end
-    @platform aware function kernel({accelerator_type::FPGA,
+    @platform aware function kernel({accelerator_count::(@atleast 1),
+                                     accelerator_type::FPGA,
                                      accelerator_memorysize::(@atleast 16G),
                                      processor_simd::AVX512,
                                      node_memory_size::(@atleast 256G)
                                     },
                                     x,y,args...; z=6, kwargs...)
-        println(z,": kernel for a UltrascalePlus+ HBM FPGA")
+        println(z,": kernel for a platform equipped with a FPGA accelerator with at least 16GB of memory,")
+        println(z,": a processor with AVX512 SIMD support, and 256GB of primary memory.")
     end
 
     kernel(0,1,2,3;z=10,kwargs=0)
+
+    PlatformAware.setplatform!(:accelerator_count, @just 1)
+    PlatformAware.setplatform!(:accelerator_manufacturer, NVIDIA)
+
+    @platform default function kernel(x,y,args...; z=0, kwargs...)
+        println(z,": default implementation of kernel_example:")
+    end
+
+    kernel(1,2,3,4;z=11,kwargs=0)
 end