llama.cpp
ggml : use dynamic thread scheduling for matrix multiplication
#6915
Merged
Overview
Commits
27
Changes
View On GitHub
Commits
  • Just reordering some structs.
    kunnis committed 1 year ago
  • Adding in the calls to mm_pause
    kunnis committed 1 year ago
  • Passing around the state
    kunnis committed 1 year ago
  • Renaming and moving a bunch of variables around.
    kunnis committed 1 year ago
  • Extracting the logic to it's own function.
    kunnis committed 1 year ago
  • Moving some variable definitions into the chunk function.
    kunnis committed 1 year ago
  • Moving some variables around
    kunnis committed 1 year ago
  • moving src1_cont inside
    kunnis committed 1 year ago
  • Moving row_size
    kunnis committed 1 year ago
  • adding the current_chunk
    kunnis committed 1 year ago
  • Reorg the code.
    kunnis committed 1 year ago
  • Formatting to match the orig patch
    kunnis committed 1 year ago
  • starting to setup the chunking variables
    kunnis committed 1 year ago
  • Starting the buildup of the loop
    kunnis committed 1 year ago
  • The yield shouldn't be necessary.
    kunnis committed 1 year ago
  • adding the looping structure based on the chunk configuration.
    kunnis committed 1 year ago
  • Add in the re-chunking code.
    kunnis committed 1 year ago
  • Making it much more likely to rechunk.
    kunnis committed 1 year ago
  • disable resizing if numa is enabled.
    kunnis committed 1 year ago
  • Updating comments with what we've learned.
    kunnis committed 1 year ago
  • Fix formatting
    kunnis committed 1 year ago
  • Couple more formatting fixes.
    kunnis committed 1 year ago
  • More style fixes.
    kunnis committed 1 year ago
  • Fix Warnings
    kunnis committed 1 year ago
  • Going with unused because there's conditional logic that needs it.
    kunnis committed 1 year ago
  • Update ggml.c
    slaren committed 1 year ago
  • Update ggml.c
    slaren committed 1 year ago
FAQ Terms Privacy Refunds Impressum
Loading