Task03 Амир Батыров ITMO

[c5xheavy]

Локальный вывод

$ ./main_matrix_transpose 1
Found 3 GPUs in 0.0466199 sec (OpenCL: 0.0275665 sec, Vulkan: 0.0190034 sec)
Available devices:
  Device #0: API: Vulkan. iGPU. Intel(R) Arc(tm) Graphics (MTL). Free memory: 5406/7750 Mb.
  Device #1: API: OpenCL. CPU. Intel(R) Core(TM) Ultra 5 125H. Intel(R) Corporation. Total memory: 15501 Mb.
  Device #2: API: Vulkan. CPU. llvmpipe (LLVM 20.1.2, 256 bits). Free memory: 15501/15501 Mb.
Using device #1: API: OpenCL. CPU. Intel(R) Core(TM) Ultra 5 125H. Intel(R) Corporation. Total memory: 15501 Mb.
Using OpenCL API...
Matrix size: rows=H=8192 x cols=W=16384 (512 MB)
______________________________________________________
Evaluating algorithm #1/2: 01 naive transpose (non-coalesced)
Kernels compilation done in 0.0794389 seconds
algorithm times (in seconds) - 10 values (min=0.268004 10%=0.281873 median=0.293419 90%=0.408015 max=0.408015)
median effective algorithm bandwidth: 3.40809 GB/s
______________________________________________________
Evaluating algorithm #2/2: 02 transpose via local memory (coalesced)
Kernels compilation done in 0.031347 seconds
algorithm times (in seconds) - 10 values (min=0.0461301 10%=0.0461855 median=0.0465298 90%=0.0790783 max=0.0790783)
median effective algorithm bandwidth: 21.4916 GB/s


$ ./main_matrix_multiply 1
Found 3 GPUs in 0.0602803 sec (OpenCL: 0.0295763 sec, Vulkan: 0.0306527 sec)
Available devices:
  Device #0: API: Vulkan. iGPU. Intel(R) Arc(tm) Graphics (MTL). Free memory: 5456/7750 Mb.
  Device #1: API: OpenCL. CPU. Intel(R) Core(TM) Ultra 5 125H. Intel(R) Corporation. Total memory: 15501 Mb.
  Device #2: API: Vulkan. CPU. llvmpipe (LLVM 20.1.2, 256 bits). Free memory: 15501/15501 Mb.
Using device #1: API: OpenCL. CPU. Intel(R) Core(TM) Ultra 5 125H. Intel(R) Corporation. Total memory: 15501 Mb.
Using OpenCL API...
C = A x B, matrices size: C (rows=H=2048 x cols=W=4096) = A (rows=H=2048 x cols=K=1024) x B (rows=K=1024 x cols=W=4096)
matrices data size: A - 8 MB, B - 16 MB, C - 16 MB
______________________________________________________
Evaluating algorithm #1/3: CPU with OpenMP
algorithm times (in seconds) - 1 values (min=8.46722 10%=8.46722 median=8.46722 90%=8.46722 max=8.46722)
algorithm GFlops: 2.02799 GFlops
algorithm effective memory bandwidth: 0.00645873 GB/s
______________________________________________________
Evaluating algorithm #2/3: 01 naive
Kernels compilation done in 0.0982374 seconds
algorithm times (in seconds) - 10 values (min=0.795038 10%=0.798513 median=0.845341 90%=1.18228 max=1.18228)
algorithm GFlops: 20.3131 GFlops
algorithm effective memory bandwidth: 0.0646928 GB/s
relative differences with CPU: 8388608 values (min=0 10%=0 median=2.21073e-07 90%=1.12363e-06 max=2.77294)
median relative difference with CPU: 2.21073e-07
99% percentile relative difference with CPU: 1.09303e-05
______________________________________________________
Evaluating algorithm #3/3: 02 using local memory
Kernels compilation done in 0.0551597 seconds
algorithm times (in seconds) - 10 values (min=0.591612 10%=0.627633 median=0.661366 90%=0.754793 max=0.754793)
algorithm GFlops: 25.9637 GFlops
algorithm effective memory bandwidth: 0.0826887 GB/s
relative differences with CPU: 8388608 values (min=0 10%=0 median=2.21073e-07 90%=1.12363e-06 max=2.77294)
median relative difference with CPU: 2.21073e-07
99% percentile relative difference with CPU: 1.09303e-05

Вывод Github CI

Run ./main_matrix_transpose 0
Found 2 GPUs in 0.0531353 sec (CUDA: 8.5881e-05 sec, OpenCL: 0.0247909 sec, Vulkan: 0.0282084 sec)
Available devices:
  Device #0: API: OpenCL. CPU. AMD EPYC 7763 64-Core Processor                . Intel(R) Corporation. Total memory: 15990 Mb.
  Device #1: API: Vulkan. CPU. llvmpipe (LLVM 20.1.2, 256 bits). Free memory: 15990/15990 Mb.
Using device #0: API: OpenCL. CPU. AMD EPYC 7763 64-Core Processor                . Intel(R) Corporation. Total memory: 15990 Mb.
Using OpenCL API...
Matrix size: rows=H=8192 x cols=W=16384 (512 MB)
______________________________________________________
Evaluating algorithm #1/2: 01 naive transpose (non-coalesced)
Kernels compilation done in 0.124981 seconds
algorithm times (in seconds) - 10 values (min=0.501865 10%=0.531583 median=0.574656 90%=0.678069 max=0.678069)
median effective algorithm bandwidth: 1.74017 GB/s
______________________________________________________
Evaluating algorithm #2/2: 02 transpose via local memory (coalesced)
Kernels compilation done in 0.0448901 seconds
algorithm times (in seconds) - 10 values (min=0.169914 10%=0.170557 median=0.170789 90%=0.216634 max=0.216634)
median effective algorithm bandwidth: 5.85519 GB/s

Run ./main_matrix_multiply 0
Found 2 GPUs in 0.0521385 sec (CUDA: 0.00010718 sec, OpenCL: 0.02417 sec, Vulkan: 0.0278133 sec)
Available devices:
  Device #0: API: OpenCL. CPU. AMD EPYC 7763 64-Core Processor                . Intel(R) Corporation. Total memory: 15990 Mb.
  Device #1: API: Vulkan. CPU. llvmpipe (LLVM 20.1.2, 256 bits). Free memory: 15990/15990 Mb.
Using device #0: API: OpenCL. CPU. AMD EPYC 7763 64-Core Processor                . Intel(R) Corporation. Total memory: 15990 Mb.
Using OpenCL API...
C = A x B, matrices size: C (rows=H=2048 x cols=W=4096) = A (rows=H=2048 x cols=K=1024) x B (rows=K=1024 x cols=W=4096)
matrices data size: A - 8 MB, B - 16 MB, C - 16 MB
______________________________________________________
Evaluating algorithm #1/3: CPU with OpenMP
algorithm times (in seconds) - 1 values (min=17.4311 10%=17.4311 median=17.4311 90%=17.4311 max=17.4311)
algorithm GFlops: 0.985106 GFlops
algorithm effective memory bandwidth: 0.00313735 GB/s
______________________________________________________
Evaluating algorithm #2/3: 01 naive
Kernels compilation done in 0.128409 seconds
algorithm times (in seconds) - 10 values (min=1.61385 10%=1.67209 median=1.67886 90%=1.78617 max=1.78617)
algorithm GFlops: 10.2281 GFlops
algorithm effective memory bandwidth: 0.0325742 GB/s
relative differences with CPU: 8388608 values (min=0 10%=0 median=2.21073e-07 90%=1.12363e-06 max=2.77294)
median relative difference with CPU: 2.21073e-07
99% percentile relative difference with CPU: 1.09303e-05
______________________________________________________
Evaluating algorithm #3/3: 02 using local memory
Kernels compilation done in 0.0707914 seconds
algorithm times (in seconds) - 10 values (min=2.79739 10%=2.79834 median=2.80095 90%=2.87392 max=2.87392)
algorithm GFlops: 6.13059 GFlops
algorithm effective memory bandwidth: 0.0195246 GB/s
relative differences with CPU: 8388608 values (min=0 10%=0 median=2.21073e-07 90%=1.12363e-06 max=2.77294)
median relative difference with CPU: 2.21073e-07
99% percentile relative difference with CPU: 1.09303e-05

[GPUcourseBOT]

✅ Результаты тестирования PR #1047

Логи тестирования (нажмите чтобы развернуть)

=== СТАТУС: Успешно выполнены программы: main_matrix_transpose, main_matrix_multiply ===
=== main_matrix_transpose stdout (exit code: -11 (segfault после выполнения)) ===
Found 1 GPUs in 8.72174 sec (CUDA: 0.11048 sec, OpenCL: 0.8089 sec, Vulkan: 7.80229 sec)
Available devices:
Device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using OpenCL API...
Matrix size: rows=H=8192 x cols=W=16384 (512 MB)
______________________________________________________
Evaluating algorithm #1/2: 01 naive transpose (non-coalesced)
Kernels compilation done in 3.46203 seconds
algorithm times (in seconds) - 10 values (min=0.028699 10%=0.0297832 median=0.0298146 90%=3.4951 max=3.4951)
median effective algorithm bandwidth: 33.5407 GB/s
______________________________________________________
Evaluating algorithm #2/2: 02 transpose via local memory (coalesced)
Kernels compilation done in 0.0991758 seconds
algorithm times (in seconds) - 10 values (min=0.00847714 10%=0.00847844 median=0.00849441 90%=0.107754 max=0.107754)
median effective algorithm bandwidth: 117.724 GB/s
=== main_matrix_multiply stdout (exit code: -11 (segfault после выполнения)) ===
Found 1 GPUs in 0.320729 sec (CUDA: 0.125026 sec, OpenCL: 0.0377321 sec, Vulkan: 0.157902 sec)
Available devices:
Device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using OpenCL API...
C = A x B, matrices size: C (rows=H=2048 x cols=W=4096) = A (rows=H=2048 x cols=K=1024) x B (rows=K=1024 x cols=W=4096)
matrices data size: A - 8 MB, B - 16 MB, C - 16 MB
______________________________________________________
Evaluating algorithm #1/3: CPU with OpenMP
algorithm times (in seconds) - 1 values (min=11.9457 10%=11.9457 median=11.9457 90%=11.9457 max=11.9457)
algorithm GFlops: 1.43746 GFlops
algorithm effective memory bandwidth: 0.00457799 GB/s
______________________________________________________
Evaluating algorithm #2/3: 01 naive
Kernels compilation done in 0.107665 seconds
algorithm times (in seconds) - 10 values (min=0.061319 10%=0.0617409 median=0.0631913 90%=0.172276 max=0.172276)
algorithm GFlops: 271.738 GFlops
algorithm effective memory bandwidth: 0.865428 GB/s
relative differences with CPU: 8388608 values (min=0 10%=0 median=2.21073e-07 90%=1.12363e-06 max=2.77294)
median relative difference with CPU: 2.21073e-07
99% percentile relative difference with CPU: 1.09303e-05
______________________________________________________
Evaluating algorithm #3/3: 02 using local memory
Kernels compilation done in 0.111219 seconds
algorithm times (in seconds) - 10 values (min=0.0288905 10%=0.0315843 median=0.031748 90%=0.136915 max=0.136915)
algorithm GFlops: 540.869 GFlops
algorithm effective memory bandwidth: 1.72255 GB/s
relative differences with CPU: 8388608 values (min=0 10%=0 median=2.21073e-07 90%=1.12363e-06 max=2.77294)
median relative difference with CPU: 2.21073e-07
99% percentile relative difference with CPU: 1.09303e-05

Посмотреть полные логи