cturan/llama.cpp

Retire the ggml_mul_mat() branch for transposed src0 (#500)

* Retire the ggml_mul_mat() for transposed src0

- It can always be made contiguous with ggml_cpy()
- The code is now simplified
- The results are deterministic in respect to num threads

* SIMD-ify dequantize_row_q4_0() for ARM_NEON (#502)

* Attempt to SIMD-ify dequantize_row_q4_0() for ARM_NEON

* Fix dequantization - forgot to interleave the quants

Georgi Gerganov 2 anni fa

parent

502a400192

commit

ecbe466a36

1 ha cambiato i file con 166 aggiunte e 559 eliminazioni

Visualizzazione separata Mostra Diff Stats

				
					File diff suppressed because it is too large
					
						+ 166
						
						- 559
					
ggml.c

Retire the ggml_mul_mat() branch for transposed src0 (#500)

File diff suppressed because it is too large + 166 - 559 ggml.c

Some files were not shown because too many files changed in this diff

File diff suppressed because it is too large
+ 166 - 559
ggml.c