NTT#340

Merged

preda merged 116 commits intopreda:masterfrom

gwoltman:master

Dec 13, 2025

Commits on Apr 7, 2025

Minor optimization. Compute weights while waiting for carries to be ready.
gwoltman
committed
Fixed bug were default TAIL_KERNELS in C code did mot match default in OpenCL code.
gwoltman
committed

Commits on Apr 9, 2025

Eliminated BCAST=1 and UNROLL_W=3 hack. Replaced with a 3 digit FFT specification.
gwoltman
committed

Commits on Jun 8, 2025

Changed FFT spec from a single-digit code (0,1,2,3) to a 3-digit WMH code.
gwoltman
committed

Commits on Jun 10, 2025

Added -tune code to time all important -use options!
gwoltman
committed

Commits on Jun 14, 2025

Added tune code for IN_WG,IN_SIZEX,OUT_WG,OUT_SIZEX
gwoltman
committed

Commits on Jun 23, 2025

Commits on Jun 30, 2025

Fixed memory allocation bug with MIDDLE=4, PAD=512
gwoltman
committed

Commits on Jul 5, 2025

Changed results.txt to results-N.txt -- the same naming scheme as worktodo files.
gwoltman
committed

Commits on Jul 9, 2025

Fixed bug where an AMD-only FFT variant (000) was chosen whenever no useful FFT spec could be found in tune.txt
gwoltman
committed

Commits on Jul 11, 2025

Commits on Aug 5, 2025

Commits on Aug 12, 2025

Initialize all varibles before initializing the thread. Prevents a race condition were launched thread accesses uninitialized variables.
gwoltman
committed

Commits on Sep 14, 2025

Commits on Sep 17, 2025

Improved GF31 and GF61 math
gwoltman
committed

Commits on Sep 18, 2025

Minor tweaks and cleanup in carryutil. Some alternate GF61 math implementations that are not faster (at least on TitanV).
gwoltman
committed

Commits on Sep 23, 2025

Commits on Sep 24, 2025

Commits on Sep 25, 2025

Fixed assert
gwoltman
committed

Commits on Sep 30, 2025

Fixed optimization bug in GF31*GF61 NTT where BPW was low (less than 23)
gwoltman
committed

Commits on Oct 3, 2025

Fixed typo bug in FP32 code
gwoltman
committed

Commits on Oct 5, 2025

Commits on Oct 6, 2025

Fixed typo in -workers 2 suggestion
gwoltman
committed

Commits on Oct 8, 2025

Don't output -log and -workers tune suggestions if they are already set.
gwoltman
committed

Commits on Oct 10, 2025

Added quick=n to -tune parameters for config.txt options testing. n ranges from 1 (slowest,more accurate) to 10 (fastest,least accurate).
gwoltman
committed

Commits on Oct 12, 2025

Changed the way sleep time is computed when Queue is full. A shorter wait for NTTs and hybrid FFTs since they execute kernels to do a squaring.
gwoltman
committed

Commits on Oct 13, 2025

Commits on Oct 17, 2025

Commits on Oct 18, 2025

Commits on Oct 19, 2025

Return correct type from mad32
gwoltman
committed

Commits on Oct 20, 2025

Commits on Oct 24, 2025

Implemented i96 as three 32-bit quantities with PTX asm.
gwoltman
committed

Commits on Oct 25, 2025

Ignore first call to setSquareTime. First timings are inaccurate due to substantial startup costs.
gwoltman
committed

Commits on Oct 27, 2025

Commits on Oct 30, 2025

Changed the wording in some -tune messages
gwoltman
committed

Commits on Nov 1, 2025

Reduce default exponent for config tuning FP32. MaxExp depends on variant,TAIL_TRIGS32,TABMUL_CHAIN32 settings.
gwoltman
committed

Commits on Nov 2, 2025

Fixed AMD asm problem handling BPW between 31 and 32
gwoltman
committed

Commits on Nov 6, 2025

Changed from executing smallest exponent in worktodo.txt to requiring command line argument -smallest
gwoltman
committed

Commits on Nov 7, 2025

Commits on Nov 8, 2025

Commits on Nov 9, 2025

Commits on Nov 10, 2025

Check for needed builtins in variant 0 FFTs
gwoltman
committed

Commits on Nov 15, 2025

Commits on Nov 16, 2025

Commits on Nov 18, 2025

Commits on Nov 19, 2025

Commits on Dec 10, 2025

Commits on Dec 12, 2025