Add Blackwell MLA forward (shape: d=192, dv=128) implementation in example_77 (#2472)

Author: dianzhangchen

examples/77_blackwell_fmha/77_blackwell_fmha.cu

@@ -853,7 +853,7 @@ struct FwdRunner {
       flops *= static_cast<double>(size<1>(problem_shape));
       flops *= static_cast<double>(size<3,1>(problem_shape));
     }
-    flops *= 4.0 * (std::is_same_v<ActiveMask, CausalMask> ? 0.5 : 1.0);
+    flops *= 4.0 * (std::is_same_v<ActiveMask, CausalMask<true>> || std::is_same_v<ActiveMask, CausalMask<false>> ? 0.5 : 1.0);
     flops *= static_cast<double>(size<2>(problem_shape));
     flops *= static_cast<double>(size<3,0>(problem_shape));
     double tflops_s = flops * 1e-12 /*tera*/ / (runtime_ms * 1e-3 /*ms*/);