Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Jan 27, 2014

• further reduction of local memory in keccak.cu

  cbuchner1

  committed
  9c9dd26

  Copy full SHA for 9c9dd26

• further reduction of local memory use in keccak code

  cbuchner1

  committed
  dc68582

  Copy full SHA for dc68582

• already using much less local memory: generic update() functions for hash states were replaced with hardcoded variants for specific data sizes and "leftover" counts.

  cbuchner1

  committed
  c552be5

  Copy full SHA for c552be5

• Get rid of the uint8_t types in the keccak functions.

  cbuchner1

  committed
  5ffff12

  Copy full SHA for 5ffff12

• Manual inlining and simplification of the PBKDF2 functions. Replace uint8_t with uint32_t. uint8_t types currently remain in the keccak_ functions

  cbuchner1

  committed
  518e87c

  Copy full SHA for 518e87c

Commits on Jan 26, 2014

• Use pinned host memory buffers and asynchronous memory copies in scrypt-jane -H 2 GPU keccak implementation. This is more efficient. It also lowers CPU utilization of the cudaminer process a lot.

  cbuchner1

  committed
  345bc56

  Copy full SHA for 345bc56

• fix one unnecessary (and incorrect) D2H copy when -H 2 mode is used with scrypt-jane

  cbuchner1

  committed
  f0c143f

  Copy full SHA for f0c143f

• keccak now runs on the GPU (-H 2) but the code is still very unoptimized!

  cbuchner1

  committed
  cf86187

  Copy full SHA for cf86187

• keccak.cu now contains the PBKDF2 and Keccak hashing code. But it's very much untested and probably slow at this point (too much local memory spillage of variables)

  cbuchner1

  committed
  bb328ef

  Copy full SHA for bb328ef

• add a .cu file for the keccak kernel. It's not functional yet.

  cbuchner1

  committed
  a6e12c3

  Copy full SHA for a6e12c3

Commits on Jan 25, 2014

• move PBKDF2 and Keccak from "code" folder into scrypt-jane.cpp module and strip out unneeded parts.

  cbuchner1

  committed
  07ebf9d

  Copy full SHA for 07ebf9d

• moving the SHA256 GPU code into a separate .cu file (speeds up compilation of salsa_kernel.cu a lot)

  cbuchner1

  committed
  5b66f55

  Copy full SHA for 5b66f55

• add RadioactiveCoin parameters. How many more coins with stupid names do we need?

  cbuchner1

  committed
  7ccf723

  Copy full SHA for 7ccf723

• Add a couple of new coins. I've looked up the Nfactor parameters in the github repos of various wallets.

  cbuchner1

  committed
  691d0c9

  Copy full SHA for 691d0c9

Commits on Jan 24, 2014

• ...and Linux wants <ctype.h> included...

  

  Christian Buchner

  committed
  9352d5c

  Copy full SHA for 9352d5c

Commits on Jan 23, 2014

• Add some code to specify GPUs by name (ideal when swapping out cards often as the CUDA device IDs change). Fermi kernels are now again built as Compute 1.0 code (lowering temps and fixing a perform…

  Show description for cc3140a
  cbuchner1

  committed
  cc3140a

  Copy full SHA for cc3140a

Commits on Jan 22, 2014

• add the nVidia submitted kernels to the Windows project files as well

  cbuchner1

  committed
  295d239

  Copy full SHA for 295d239

• launch configs with Y and Z did not validate properly (only autotune was working)

  

  Christian Buchner

  authored and

  Christian Buchner

  committed
  2baa971

  Copy full SHA for 2baa971

• add some kernels submitted by nVidia as "Y" and "Z" kernels (for testing). Y is compute_30, Z is compute_35. Both are good for scrypt mostly.

  

  Christian Buchner

  authored and

  Christian Buchner

  committed
  0c2bbbc

  Copy full SHA for 0c2bbbc

Commits on Jan 20, 2014

• computeGold() function for Salsa20/8 CPU validation wasn't properly working for N != 1024 yet. Fixed.

  cbuchner1

  committed
  142261b

  Copy full SHA for 142261b

• fix a problem with scratchpad being allocated with a hardcoded size. This made vertcoin CPU validation crash.

  cbuchner1

  committed
  5d95569

  Copy full SHA for 5d95569

• add AppleCoin [APC]

  cbuchner1

  committed
  54a81c2

  Copy full SHA for 54a81c2

• oops, I forgot that the test_kernel requires a dynamic shared memory assignment when I copied code from the Kepler kernel. This lead to a crash

  

  Christian Buchner

  authored and

  Christian Buchner

  committed
  e244a62

  Copy full SHA for e244a62

• add MicroCoin parameters

  

  Christian Buchner

  authored and

  Christian Buchner

  committed
  203de62

  Copy full SHA for 203de62

• make gcc happy (ambiguous argument to min function)

  

  Christian Buchner

  committed
  b0d178c

  Copy full SHA for b0d178c

• Make some changes to improve reliability and speed of autotune. Bump up version indicator to 2014-01-20

  cbuchner1

  committed
  c1689e6

  Copy full SHA for c1689e6

Commits on Jan 19, 2014

• test_kernel (Fermi) also gets the lookup_gap treatment...

  cbuchner1

  committed
  fba1561

  Copy full SHA for fba1561

• attempting lookup gap also for Titan kernel (so far untested)

  cbuchner1

  committed
  5657ac6

  Copy full SHA for 5657ac6

Commits on Jan 18, 2014

• checking in a lookup-gap enabled Kepler kernel. When the gap is disabled, a faster kernel is used.

  cbuchner1

  committed
  d5d1cbd

  Copy full SHA for d5d1cbd

• fix a problem when parsing --algo=scrypt:2048, add infrastructure for lookup-gap support (without actually providing the support in the kernels at the moment)

  cbuchner1

  committed
  04561c2

  Copy full SHA for 04561c2

Commits on Jan 17, 2014

• missing curly braces can be a bitch...

  cbuchner1

  committed
  7a32ea6

  Copy full SHA for 7a32ea6

• Change hardcoded warp limit of 1024 to a #define (currently 4096) to accomodate changed warp definition for Dave's kernels)

  cbuchner1

  committed
  cf673ae

  Copy full SHA for cf673ae

• breaking change! warp size for David Andersen's kernels was reduced from 128 threads to 32 threads (to match CUDA definition of a warp). This changes kernel launch configurations! But memory alloca…

  Show description for a82ebc3
  cbuchner1

  committed
  a82ebc3

  Copy full SHA for a82ebc3

Commits on Jan 16, 2014

• replace expensive modular arithmetics with non-constant values with logical AND (assuming that we pass power of 2 constants, of course)

  cbuchner1

  committed
  9348d1a

  Copy full SHA for 9348d1a

• Radical simplification of memory access patterns in write/read/read_xoe_keys_direct functions in Dave Andersen's kernels. This should lower power consumption.

  cbuchner1

  committed
  c4c67b5

  Copy full SHA for c4c67b5

Pagination

Previous

Next