https://github.com/iskandr/honeycomb

https://github.com/iskandr/honeycomb

Last synced: 23 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/iskandr/honeycomb
• Owner: iskandr
• Created: 2013-06-19T17:23:28.000Z (over 11 years ago)
• Default Branch: master
• Last Pushed: 2013-06-19T19:53:51.000Z (over 11 years ago)
• Last Synced: 2024-10-15T20:25:35.313Z (2 months ago)
• Language: Python
• Homepage: https://github.com
• Size: 165 KB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
HoneyComb

=========

### Data Parallel Bytecode for Simpler GPU Programming

Examples:

---------

Adding two vectors:

```

  n = DIM a0 0

  PARFOR n

    offset = MUL_SCALAR idx.x vecsize

    v0 = LOAD a0 offset 

    v1 = LOAD a1 offset

    IADD v1 v0 

    STORE v1 a1 s0 

```

Naive matrix multiply: 

```

  n = DIM a0 0 

  m = DIM a1 1

  PARFOR n, m  

    k = DIM a0 1 

    FOR_OFFSETS(k, offset)   

      v0 = LOAD_ROW a0 idx.x offset

      v1 = LOAD_COL a1 idx.y offset

      IMUL v1 v0 

      total = SUM v1 

      STORE_SCALAR_2D total a2 idx.x idx.y 

```

Tree-structured parallel sum of a 1D vector using an outermost loop:

```

  b = CONST 1

  WHILE b

    n_full = DIM a0 0 

    n_small = DIV n vecsize 

    a1 = ALLOC n_small

    PARFOR n_full BY vecsize  

      v0 = LOAD a0 idx.x 

      total = SUM v0 

      i = DIV_SCALAR idx.x vecsize 

      STORE_SCALAR total a1 i 

    b = GT n_small 1

    a0 = a1 

  res =  LOAD_SCALAR a0 0 

  RETURN res 

```

Or, more pleasantly, using a global reduce: 

```

  n = DIM a0 0

  REDUCE n

    v0 = LOAD a0 idx.x 

    total = SUM v0 

    return total 

```