diff --git a/src/singeli/src/slash.singeli b/src/singeli/src/slash.singeli index dcebe0a6..9062b1ad 100644 --- a/src/singeli/src/slash.singeli +++ b/src/singeli/src/slash.singeli @@ -225,7 +225,7 @@ fn slash{c==1, T & hasarch{if (width{T}>=32) 'AVX512F' else 'AVX512VBMI2'}}(w:*u # The compress-store instruction performs very poorly on Zen4, # and is also a lot worse than the following on Tiger Lake # emitT{void, 'mask_compressstoreu', r, m, x} - cs := cast_i{I,promote{i64,1}<>6 v := emitT{V, 'mask_compress', x, m, x} emitT{void, 'mask_storeu', r, to_mask{cs}, v}