xe: jit: gemm: expand decomp cases, enforce fpmath

Author: kealan-barbieri

src/gpu/intel/jit/gemm/gen_gemm.hpp

@@ -73,9 +73,7 @@ struct gen_gemm_t : public gpu_gemm_t {
                             d->c_type(), f8_e5m2, f8_e4m3, f16, bf16, f32));
             wei_decomp_ = (utils::one_of(d->c_type(), f32, f16, bf16, f8_e5m2,
                                    f8_e4m3)
-                                  && utils::one_of(d->a_type(), u8, s8, s4, u4)
-                                  && utils::one_of(d->b_type(), f16, f32, bf16,
-                                          f8_e5m2, f8_e4m3))
+                                  && utils::one_of(d->a_type(), u8, s8, s4, u4))
                     && attr()->mayiconvert(d->a_type(), f32);
             dy_quant_enabled_
                     = (utils::one_of(d->c_type(), f32, f16, bf16)

src/gpu/intel/jit/gemm/gen_gemm_kernel.cpp

@@ -561,13 +561,23 @@ status_t gen_gemm_nocopy_kernel_desc_t::select_kernel(compute::gpu_arch_t arch,
     add_mode_matches(fpmath_bf16, [](Type dt) -> const char * {
         if (dt == Type::f32) { return "[SB]"; }
         if (dt.isInt8() || dt.isInt4()) return "[OB]";
+        return nullptr;
+    });
+
+    add_mode_matches(fpmath_bf16, [](Type dt) -> const char * {
+        if (dt.isInt8() || dt.isInt4()) return "B";
         if (dt.isF8()) return "B";
         return nullptr;
     });
 
     add_mode_matches(fpmath_f16, [](Type dt) -> const char * {
         if (dt == Type::f32) { return "[SH]"; }
         if (dt.isInt8() || dt.isInt4()) return "[OH]";
+        return nullptr;
+    });
+
+    add_mode_matches(fpmath_f16, [](Type dt) -> const char * {
+        if (dt.isInt8() || dt.isInt4()) return "H";
         if (dt.isF8()) return "H";
         return nullptr;
     });