@@ -77,7 +77,7 @@ define amdgpu_kernel void @copy_flat(ptr nocapture %d, ptr nocapture readonly %s
7777; GFX1250-NEXT: s_add_nc_u64 s[2:3], s[2:3], 16
7878; GFX1250-NEXT: s_cmp_lg_u32 s6, 0
7979; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
80- ; GFX1250-NEXT: flat_store_b128 v0, v[2:5], s[0:1]
80+ ; GFX1250-NEXT: flat_store_b128 v0, v[2:5], s[0:1] scope:SCOPE_SE
8181; GFX1250-NEXT: s_wait_xcnt 0x0
8282; GFX1250-NEXT: s_add_nc_u64 s[0:1], s[0:1], 16
8383; GFX1250-NEXT: s_cbranch_scc1 .LBB0_2
@@ -490,7 +490,7 @@ define amdgpu_kernel void @copy_flat_divergent(ptr nocapture %d, ptr nocapture r
490490; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
491491; GFX1250-NEXT: s_cmp_lg_u32 s0, 0
492492; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
493- ; GFX1250-NEXT: flat_store_b128 v[0:1], v[4:7]
493+ ; GFX1250-NEXT: flat_store_b128 v[0:1], v[4:7] scope:SCOPE_SE
494494; GFX1250-NEXT: s_wait_xcnt 0x0
495495; GFX1250-NEXT: v_add_nc_u64_e32 v[0:1], 16, v[0:1]
496496; GFX1250-NEXT: s_cbranch_scc1 .LBB4_2
0 commit comments