Forum Replies Created

Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
    Posts
  • bakul_retired
    Member
    January 2, 2018 at 8:05 pm
    Post count: 4
    in reply to: Security on SSOOO processors is HARD #3136

    I think even mill would be susceptible to Rowhammer.

    Though I should mention that the current bug being discussed (we are speculating that it is about speculative execution across protection boundary) likely won’t be an issue for mill.

    • This reply was modified 6 years, 8 months ago by  bakul_retired.
  • bakul_retired
    Member
    December 11, 2017 at 3:49 pm
    Post count: 4
    in reply to: Support for multidimensional arrays #3114

    What I was wondering about is whether Mill can optimize each “step”. So for example, a dope vector for 3D array x, declared as

    int32_t x[a:b, c:d, e:f]; // I am "extending" C here to allow arbitrary upper/lower bounds, and a:b is half open

    may be

    typedef struct { int base, elemsize, size; } Dope;

Dope dv[] = { {a, s1, s0}, {c, s2, s1}, {e, s3, s2} };
// s3 = 4, s2 = s3*(f-e), s1 = s2*(d-c), s0 = s1*(b-a)

    Given an expression x[i,j,k] you’d have to do a bunch of multiplies and additions or subtractions and comparisons. I think you should be able to do computation for each dim. in parallel before a final add.

  • Author
    Posts
Viewing 2 posts - 1 through 2 (of 2 total)