danielkrause/colibri-strix
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

expert dot: AVX-512 int4→float accumulator, runtime-switchable (I4_ACC512), +7% CPU-heavy decode, more accurate than AVX2 order (#95)

Browse Source 
AVX-512F/BW build path only; non-AVX-512 builds bit-identical. Qualified on Xeon Silver 4510: max rel-err 2-6× lower than scalar oracle order, ppl 5.99 vs 5.98 (0.24%), runtime escape hatch I4_ACC512=0. Clears the #80/#94 numerical bar.
This commit is contained in:
ZacharyZcR
2026-07-13 21:24:25 +08:00
committed by GitHub
parent 97c756a064
commit 4b1d0e3a57
3 changed files with 146 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
c/Makefile
+4 -1
View File
@@ -66,7 +66,7 @@ CUDA_ARCH ?= native
NVCCFLAGS ?= -O3 -std=c++17 -arch=$(CUDA_ARCH) -Xcompiler=-Wall,-Wextra
PYTHON ?= python3
CUDA_OBJ =
TEST_BINS = tests/test_json$(EXE) tests/test_st$(EXE) tests/test_tier$(EXE) tests/test_grammar$(EXE) tests/test_decode_batch$(EXE) tests/test_idot$(EXE)
TEST_BINS = tests/test_json$(EXE) tests/test_st$(EXE) tests/test_tier$(EXE) tests/test_grammar$(EXE) tests/test_decode_batch$(EXE) tests/test_idot$(EXE) tests/test_i4_acc512$(EXE)
ifeq ($(CUDA),1)
ifeq ($(UNAME_S),Darwin)
$(error CUDA=1 is supported only on Linux)
@@ -151,6 +151,9 @@ tests/test_decode_batch$(EXE): tests/test_decode_batch.c decode_batch.h
tests/test_idot$(EXE): tests/test_idot.c glm.c st.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h
$(CC) $(CFLAGS) $< -o $@ $(LDFLAGS)
tests/test_i4_acc512$(EXE): tests/test_i4_acc512.c
$(CC) $(CFLAGS) $< -o $@ $(LDFLAGS)
test-c: $(TEST_BINS)
@for test in $(TEST_BINS); do ./$$test || exit 1; done