tests: benchdnn: matmul: add decomp coverage to gpu

Author: kealan-barbieri

tests/benchdnn/inputs/matmul/test_matmul_gpu

@@ -95,6 +95,44 @@
 --attr-post-ops=sum+add:f32+add:u8:per_dim_01+mul:f32:per_dim_0+add:s8:per_tensor+add:f32:per_dim_01+linear:2:-1
 1x5x4x3x30x2:6x1x4x3x2x20
 
+### int8 wei decomp
+--reset
+--stag=ab,ba
+--dt=bf16:s8:bf16,bf16:u8:bf16
+--attr-scales=wei:common:2,wei:per_oc:bf16
+--attr-fpmath=bf16:true
+--batch=shapes_2d
+
+--reset
+--stag=ab,ba
+--dt=f16:s8:f16,f16:u8:f16
+--attr-scales=wei:common:2,wei:per_oc,wei:per_ocic:f16
+--attr-zero-points=,wei:common:2,wei:per_oc,wei:per_ocic:s8
+--attr-fpmath=f16:true
+--batch=shapes_2d_ci
+
+--stag=abc,acb
+--dt=bf16:s8:bf16,bf16:u8:bf16
+--attr-scales=wei:per_ocic:bf16:2x1
+--attr-zero-points=,wei:per_ocic:u8:4x1
+--attr-fpmath=bf16:true
+1x5x12:1x12x17
+3x5x12:3x12x17
+3x5x12:1x12x17
+
+### int4 wei decomp
+--reset
+--dt=bf16:s4:bf16,bf16:u4:bf16
+--stag=abc,acb
+--attr-scales=wei:common:2,wei:per_oc:bf16,wei:per_ocic:bf16:32x1
+# There is no support for a common u4/s4 zero point
+--attr-zero-points=wei:per_oc:u4,wei:per_ocic:s4,wei:per_ocic:s4:32x1
+--attr-fpmath=bf16:true
+7x6x32:7x32x64
+7x6x32:1x32x64
+3x6x96:3x96x64
+3x6x96:1x96x64
+
 # Test layers of some key and ext GPU DL Frameworks
 --reset
 --batch=option_set_fwks_key_gpu