matmul: x64: added support for bf16,f16 bias dt

Author: amakarev

src/cpu/x64/brgemm/brgemm.cpp

@@ -348,13 +348,15 @@ status_t brgemm_desc_set_postops(brgemm_desc_t *brg,
                                 data_type::f16)))
         return status::unimplemented;
     const auto bias_f8_e5m2_compatible
-            = one_of(dt_d, data_type::f32, data_type::f16, data_type::f8_e5m2)
+            = one_of(dt_d, data_type::f32, data_type::f16, data_type::bf16,
+                      data_type::f8_e5m2)
             && one_of(dt_bias, data_type::undef, data_type::f32, data_type::f16,
-                    data_type::f8_e5m2, data_type::f8_e4m3);
+                    data_type::bf16, data_type::f8_e5m2, data_type::f8_e4m3);
     const auto bias_f8_e4m3_compatible
-            = one_of(dt_d, data_type::f32, data_type::f16, data_type::f8_e4m3)
+            = one_of(dt_d, data_type::f32, data_type::f16, data_type::bf16,
+                      data_type::f8_e4m3)
             && one_of(dt_bias, data_type::undef, data_type::f32, data_type::f16,
-                    data_type::f8_e4m3, data_type::f8_e5m2);
+                    data_type::bf16, data_type::f8_e4m3, data_type::f8_e5m2);
     if (!IMPLICATION(brg->is_fp8,
                 bias_f8_e5m2_compatible || bias_f8_e4m3_compatible))
         return status::unimplemented;

src/cpu/x64/matmul/brgemm_matmul.cpp

@@ -70,7 +70,10 @@ status_t brgemm_matmul_t<isa>::pd_t::init(engine_t *engine) {
         const bool is_bia_dt_correct
                 = IMPLICATION(is_int8 == true,
                           one_of(bia_dt, f32, s32, s8, u8, bf16))
-                && IMPLICATION(!is_int8, one_of(bia_dt, f32, src_dt));
+                && IMPLICATION(
+                        is_f8 == true, one_of(bia_dt, f32, f16, bf16, src_dt))
+                && IMPLICATION(
+                        !(is_int8 || is_f8), one_of(bia_dt, f32, src_dt));
         return IMPLICATION(with_bias(), is_bia_dt_correct && is_bias_1xN());
     };
 

tests/benchdnn/inputs/matmul/test_matmul_fp8

@@ -4,7 +4,7 @@
 --dt=f8_e4m3:f8_e4m3:f32,f8_e4m3,f8_e5m2:f8_e5m2:f32,f8_e5m2
 --stag=ab,ba --wtag=ab,ba --dtag=ab
 --runtime_dims_masks=0,2:1,1:0,3:1
---bia_dt=undef,f32 --bia_mask=2
+--bia_dt=undef,f32,f16,bf16 --bia_mask=2
 
 --attr-scales=
 --attr-post-ops=
@@ -21,8 +21,8 @@
 
 --stag=ba --wtag=ab,ba --dtag=ab
 --runtime_dims_masks=3:1,3:3
---bia_dt=f8_e4m3,f8_e5m2 --bia_mask=1,2,3
---attr-scales=src:common:0.25+wei:common:0.5+dst:common:2.25
+--bia_dt=f8_e4m3,f8_e5m2,f16,bf16 --bia_mask=1,2,3
+--attr-scales=src:common:0.25+wei:common:0.5+dst:common:4
 --attr-post-ops=add:f32,sum+mul:s32:per_oc+linear:2:-1
 --batch=shapes_2d