Skip to content

Latest commit

 

History

10 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation

Sera

A Benchmark for GPU Soft Error Resilience Assessment. Sera is available for download at Sera.

Dataset files

  • kernel&instruction_with_num.json // Contains all SASS instructions and offset addresses for the GPU application.
  • test.csv // A test dataset for the GPU application that contains 20,000 fault samples.
  • train&val.csv // A training dataset for the GPU application that contains 10,000 fault samples.

Description of the GPU applications

  • Rodinia-v3.1. The GPU applications under this test suite are described in Rodinia.
  • PloyBench. The GPU applications under this test suite are described in PolyBench.

Description of the attributes in the dataset.

The description of the attributes in test.csv and train&val.csv

Attribute Description
kernel_name The name of the target GPU kernel
kernel_index An integer n representing the (n + 1) th dynamic instance of the target kernel, e.g., 0 indicates the first dynamic instance
sk_id The Index of the target kernel
static_instID The Index of the target instruction
opcode Operation type of the target instruction
instID An integer m representing the (m + 1) th dynamic instance of the target instruction
instr_count The number of instructions in the dynamic instance of the target kernel
regNo The number of the target register
maxregs The maximum number of registers for the target kernel
beforeVal The binary value in the target register before a soft error (i.e., a bit-flip fault) occurred
fip_pos The position in the target register where the bit flip occurred
fip_tp The type of bit flip, from 1 to 0 or 0 to 1
blockID The index of the thread block in which the thread executing the dynamic instance of the target instruction is located
TID The index of the thread that executes the dynamic instance of the target instruction
inject_result The fault injection result of the target soft error, 0 is Masked, 1 is SDC, 2 is DUE. The fault injection tool is NVBitFI

Note: The all attributes are associated with the target fault site.

Examples: Several fault samples from the BFS application.

kernel_name kernel_index sk_id dk_id static_instID opcode instID instr_count regNo maxregs beforeVal fip_pos fip_tp blockID blockTID TID inject_result
Kernel(Node*,int*,bool*,bool*,bool*,int*,int) 8 0 4 37 MOV 150977 251201 18 21 0x1 15 1 6 54 3126 0
Kernel2(bool*,bool*,bool*,bool*,int) 7 1 3 60 SHF 22744 69536 3 14 0x0 3 1 7 158 3742 2
Kernel(Node*,int*,bool*,bool*,bool*,int*,int) 10 0 5 37 MOV 62292 239309 18 21 0x1 0 0 6 252 3324 0
Kernel2(bool*,bool*,bool*,bool*,int) 9 1 4 64 LDG 38008 76900 0 14 0x0 24 1 7 61 3645 1
Kernel2(bool*,bool*,bool*,bool*,int) 9 1 4 56 S2R 9721 76900 3 14 0x99 9 1 1 153 665 1

About

A Benchmark for GPU Soft Error Resilience Assessment

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors