kfdtest: consolidate LoopIsa

1, Initialize the registers before using them is the best practice.
Though the use case here doesn't care whether the registers are
initialized or not, some emulators complain the "read_before_write"
behavior. Initialize the registers used to silence these complaints.

2, Update s_wait stuff for gfx12.

Change-Id: I462b2b0b5017dd2876a5954169d3b6b2f1c2a75b
Signed-off-by: Lang Yu <lang.yu@amd.com>

libhsakmt/tests/kfdtest/src/ShaderStore.cpp

@@ -691,6 +691,14 @@ const char *LoopIsa =
     R"(
         s_movk_i32    s0, 0x0008
         s_movk_i32    s1, 0x00ff
+        s_mov_b32     s4, 0
+        s_mov_b32     s5, 0
+        s_mov_b32     s6, 0
+        s_mov_b32     s7, 0
+        s_mov_b32     s12, 0
+        s_mov_b32     s13, 0
+        s_mov_b32     s14, 0
+        s_mov_b32     s15, 0
         v_mov_b32     v0, 0
         v_mov_b32     v1, 0
         v_mov_b32     v2, 0
@@ -717,7 +725,12 @@ const char *LoopIsa =
         s_cbranch_scc1  END_OF_PGM
         v_add_f32     v0, 2.0, v0
         v_cvt_f32_i32 v17, s1
-        s_waitcnt     lgkmcnt(0)
+        .if (.amdgcn.gfx_generation_number >= 12)
+            s_wait_dscnt     0
+            s_wait_kmcnt     0
+        .else
+            s_waitcnt lgkmcnt(0)
+        .endif
         v_add_f32     v18, s8, v17
         v_add_f32     v19, s9, v17
         v_add_f32     v20, s10, v17