--reset

--dt=f8_e5m2,\
     f8_e4m3,\
     f8_e4m3:f8_e4m3:f32,\
     f8_e5m2:f8_e5m2:f32,\
     f8_e4m3:f8_e5m2:f8_e4m3,\
     f8_e4m3:f8_e5m2:f8_e5m2,\
     f8_e5m2:f8_e4m3:f8_e5m2,\
     f8_e5m2:f8_e4m3:f8_e4m3
--bia_dt=undef,f32,bf16
--beta=0,1
--attr-post-ops=,sum:2,relu
--brgemm-attr=,use_uker:1+use_interleave_stores:1,use_uker:0+use_interleave_stores:1
--batch=option_set_int8 # fp8 uses int8 blocking for AMX

# FP8 Emulation: using no interleaved stores changes the size of the temporary data buffer
--brgemm-attr=use_uker:1+use_interleave_stores:0,use_uker:0+use_interleave_stores:0
--bs=1
--batch=shapes_2d_no_tail_int8
--batch=shapes_2d_tail_n_int8

# f16 inputs with f8 output -- requires native f8 ISA (avx10_2+)
# Add tests for matmul with pre-converted f8->f16
--reset
--dt=f16:f16:f8_e5m2,\
     f16:f16:f8_e4m3
--bia_dt=undef,f32,f8_e5m2,f8_e4m3
--beta=0,1
--attr-post-ops=,sum:2,relu
--attr-scales=,dst:common:5
--batch=option_set_bf16
