Download as PDF, PPTX
![Af_Array [1] pry(main)> a = ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-14-2048.jpg)
![Af_Array [1] pry(main)> a = ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-15-2048.jpg)
![Af_Array [1] pry(main)> a = ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-16-2048.jpg)
![Af_Array [1] pry(main)> a = ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-17-2048.jpg)
![[2] pry(main)> b = a + a No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 2.0000 6.0000 4.0000 8.0000 => #<ArrayFire::Af_Array:0x000000020625c8>](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-18-2048.jpg)
![[1] pry(main)> left = ArrayFire::Af_Array.new 2 , [3,3] , [1, 4, 6, 4, 11 , 2 ,-5, 8, 10] No Name Array [3 3 1 1] 1.0000 4.0000 -5.0000 4.0000 11.0000 8.0000 6.0000 2.0000 10.0000 => #<ArrayFire::Af_Array:0x000000014e56c8> [2] pry(main)> right = ArrayFire::Af_Array.new 2 , [3,2] , [1, 0, 8, 10, -11, 8] No Name Array [3 2 1 1] 1.0000 10.0000 0.0000 -11.0000 8.0000 8.0000 => #<ArrayFire::Af_Array:0x00000001591db0>](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-19-2048.jpg)
![[3] pry(main)> result = ArrayFire::BLAS.matmul(left, right, :AF_MAT_NONE, :AF_MAT_NONE) No Name Array [3 2 1 1] -39.0000 -74.0000 68.0000 -17.0000 86.0000 118.0000 => #<ArrayFire::Af_Array:0x000000016136f8>](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-20-2048.jpg)
![VALUE arf_init(int argc, VALUE* argv, VALUE self) { afstruct* afarray; Data_Get_Struct(self, afstruct, afarray); dim_t ndims = (dim_t)NUM2LONG(argv[0]); dim_t* dimensions = (dim_t*)malloc(ndims * sizeof(dim_t)); dim_t count = 1; for (size_t index = 0; index < ndims; index++) { dimensions[index] = (dim_t)NUM2LONG(RARRAY_AREF(argv[1], index)); count *= dimensions[index]; } double* host_array = (double*)malloc(count * sizeof(double)); for (size_t index = 0; index < count; index++) { host_array[index] = (double)NUM2DBL(RARRAY_AREF(argv[2], index)); } af_create_array(&afarray->carray, host_array, ndims, dimensions, f64); return self; }](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-21-2048.jpg)
![vadd_kernel_src = <<-EOS extern "C" { __global__ void matSum(int *a, int *b, int *c) { int tid = blockIdx.x; if (tid < 100) c[tid] = a[tid] + b[tid]; } } EOS f = compile(vadd_kernel_src) RbCUDA::Driver.run_kernel(f.path)](https://image.slidesharecdn.com/rubykaigi-180602164557/75/High-Performance-GPU-computing-with-Ruby-Rubykaigi-2018-40-2048.jpg)
Uploaded byPrasun Anand
High Performance GPU computing with Ruby, Rubykaigi 2018
Prasun Anand presented on using GPUs for high performance computing in Ruby. He discussed how he has implemented GPU libraries like ArrayFire and RbCUDA to enable fast matrix multiplication and neural network computations. Benchmark results showed his GPU libraries in Ruby could achieve speedups of over 100,000x compared to non-GPU Ruby matrix libraries for large problems. He thanked various contributors and organizations for their support in bringing GPU computing capabilities to the Ruby ecosystem.
High Performance GPU computing with Ruby, Rubykaigi 2018
- 1. High Performance GPU Computingwith Ruby Prasun Anand
- 3. About me ● ModakAnalytics ● Genenetwork project ● SciRuby Contributor ● Google Summer of Code 2016, 2017 ● Ruby Grant 2017 ● Fukuoka Ruby Award 2018 ● Projects: ○ JRuby port of NMatrix ○ ArrayFire gem ○ RbCUDA
- 4. Data is thenew Oil!
- 5. Highlights Modak Analytics ishelping implement one of the largest Life Sciences Platform in the world. Platform Details 2100 Structured data sources 500k Tables 1350 Unstructured sources 1.3 Billion Files 1200 Data Nodes 6 Petabytes Usable information • 1000+ clinical trials being standardized to CDISC (SDTM) model for cross-study analysis, placebo baseline etc. • Single integrated data platform comprising of compound, activity results, assay protocol and project information • “Like Minded” data has been grounded into Data Domains by business areas. E.g. Clinical, Assay, Gene, Regulatory etc • Around 17+ solutions have been developed and deployed for business Awarded at the prestigious ‘Strata Data Conference 2017’ for building this platform in record time
- 6. Governed Data Lakeapproach AUTOMATED DATA DISCOVERY • Modak is providing end-to-end service for the platform including Automated Ingestion, Curation, and innovative Solutions • Modak is also providing 24*7 support for the massive platform AUTOMATED DATA INGESTION Data Spider Postgres SQL serverOracle MySQL Structured Data SAS Data Sets Unstructured Data File shares SharePointDocumentum BOTS FOUNDATION LAYER Ingested Raw Data Data Tagging Data Masking Data cleansing Data lineage Data profiling Augmented Data Mapping/ Standardization Data Fingerprinting A replica of the Data is ingested into the Integration Layer INTEGRATION LAYER SOLUTIONS LAYER Data Analytics SEMANTICLAYER Visulaisation Dashboards and Reports MetaData Catalog (KOSH) Flow controller Streamsets Pipelines are generated automatically Data Governance Data Security System / Application Management SOURCE DATA Originators of data and serve as “authoring” systems to support business processes Optimized for computing and distribution of data Optimized for strategic BI product development Optimized for Business Users Optimized for Analysts, Data scientists GWAS
- 7.
- 8.
- 9.
- 10.
- 11. GPU Computing isnot easy !
- 12.
- 14. Af_Array [1] pry(main)> a= ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>
- 15. Af_Array [1] pry(main)> a= ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>
- 16. Af_Array [1] pry(main)> a= ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>
- 17. Af_Array [1] pry(main)> a= ArrayFire::Af_Array.new 2, [2,2],[1,2,3,4] No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 1.0000 3.0000 2.0000 4.0000 => #<ArrayFire::Af_Array:0x000000020aeab8>
- 18. [2] pry(main)> b= a + a No Name Array [2 2 1 1] Offsets: [0 0 0 0] Strides: [1 2 4 4] 2.0000 6.0000 4.0000 8.0000 => #<ArrayFire::Af_Array:0x000000020625c8>
- 19. [1] pry(main)> left= ArrayFire::Af_Array.new 2 , [3,3] , [1, 4, 6, 4, 11 , 2 ,-5, 8, 10] No Name Array [3 3 1 1] 1.0000 4.0000 -5.0000 4.0000 11.0000 8.0000 6.0000 2.0000 10.0000 => #<ArrayFire::Af_Array:0x000000014e56c8> [2] pry(main)> right = ArrayFire::Af_Array.new 2 , [3,2] , [1, 0, 8, 10, -11, 8] No Name Array [3 2 1 1] 1.0000 10.0000 0.0000 -11.0000 8.0000 8.0000 => #<ArrayFire::Af_Array:0x00000001591db0>
- 20. [3] pry(main)> result= ArrayFire::BLAS.matmul(left, right, :AF_MAT_NONE, :AF_MAT_NONE) No Name Array [3 2 1 1] -39.0000 -74.0000 68.0000 -17.0000 86.0000 118.0000 => #<ArrayFire::Af_Array:0x000000016136f8>
- 21. VALUE arf_init(int argc,VALUE* argv, VALUE self) { afstruct* afarray; Data_Get_Struct(self, afstruct, afarray); dim_t ndims = (dim_t)NUM2LONG(argv[0]); dim_t* dimensions = (dim_t*)malloc(ndims * sizeof(dim_t)); dim_t count = 1; for (size_t index = 0; index < ndims; index++) { dimensions[index] = (dim_t)NUM2LONG(RARRAY_AREF(argv[1], index)); count *= dimensions[index]; } double* host_array = (double*)malloc(count * sizeof(double)); for (size_t index = 0; index < count; index++) { host_array[index] = (double)NUM2DBL(RARRAY_AREF(argv[2], index)); } af_create_array(&afarray->carray, host_array, ndims, dimensions, f64); return self; }
- 22. static VALUE arf_matmul(VALUEself, VALUE left_val, VALUE right_val, VALUE left_prop_val, VALUE right_prop_val){ afstruct* left; afstruct* right; afstruct* result = ALLOC(afstruct); Data_Get_Struct(left_val, afstruct, left); Data_Get_Struct(right_val, afstruct, right); af_mat_prop left_mat_prop = arf_mat_type_from_rbsymbol(left_prop_val); af_mat_prop right_mat_prop = arf_mat_type_from_rbsymbol(right_prop_val); af_matmul(&result->carray, left->carray, right->carray, left_mat_prop, right_mat_prop); return Data_Wrap_Struct(CLASS_OF(left_val), NULL, arf_free, result); }
- 23. BLAS functionalities ● Matmult ●Transpose LAPACK functionalities ● Det ● Inverse ● Norm ● Qr ● Cholesky ● Svd ● lu
- 24. Statistics ● Mean ● Median ●Variance
- 25. Benchmarks ● AMD FX8350 octacore processor ● Nvidia GTX 750Ti GPU ● Double dtype
- 27. 10 X Faster thanNMatrix-Ruby-Lapack
- 31. 10,000 X Faster thanNMatrix-Ruby
- 35. 100,000 X Faster thanNMatrix-Ruby-BLAS
- 38.
- 39. GPU Array ● Genericpointer used to handle an array of elements on the GPU. ● Memory copying from CPU to GPU and vice-versa. ● Interfaced with NMatrix and NArray
- 40. vadd_kernel_src = <<-EOS extern"C" { __global__ void matSum(int *a, int *b, int *c) { int tid = blockIdx.x; if (tid < 100) c[tid] = a[tid] + b[tid]; } } EOS f = compile(vadd_kernel_src) RbCUDA::Driver.run_kernel(f.path)
- 41. ● CuBLAS ● CuSolver ●CuRand
- 42. Benchmarks ● AMD FX8350 octacore processor ● Nvidia GTX 750Ti GPU ● Double dtype
- 44. 1,000,000 X Faster thanNMatrix-Ruby-BLAS
- 46.
- 48. Future Work ● ImageProcessing APIs and Indexers ● Multiple dtypes ● RbCUDA is under development.
- 49.
- 50. Acknowledgements 1. Pjotr Prins 2.Pradeep Garigipati 3. Kenta Murata 4. Ruby Science Foundation 5. Ruby Association 6. Modak Analytics
- 51. Thanks! Github: prasunanand Twitter: @prasun_anand Blog:prasunanand.com
- 52.