1 files changed, 52 insertions, 88 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmax.ll b/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmax.ll
index 6218a5c82afc..b8a2476dc19b 100644
--- a/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmax.ll
+++ b/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmax.ll
@@ -16192,11 +16192,10 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__amdgpu_no_fine_grained_m
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v3, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v3.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v3, v5
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v3.h
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v3, v[0:1], v[5:6] th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
@@ -16314,7 +16313,6 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__amdgpu_no_fine_grained_m
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v4, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB54_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -16337,10 +16335,9 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__amdgpu_no_fine_grained_m
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v3, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v3, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v3.h
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v3, v5
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v3.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v3, v[0:1], v[5:6] glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -16352,7 +16349,6 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__amdgpu_no_fine_grained_m
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB54_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
@@ -16643,11 +16639,10 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_no
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v3, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v3.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v3, v5
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v3.h
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v3, v[0:1], v[5:6] offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
@@ -16765,7 +16760,6 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_no
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v4, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB55_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -16788,10 +16782,9 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_no
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v3, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v3, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v3.h
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v3, v5
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v3.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v3, v[0:1], v[5:6] offset:2044 glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -16803,7 +16796,6 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_no
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB55_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
@@ -17097,11 +17089,10 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v3, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v3.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v3, v5
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v3.h
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v3, v[0:1], v[5:6] offset:-2048 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
@@ -17229,7 +17220,6 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
 ; GFX11-TRUE16-NEXT:    flat_load_b32 v0, v[3:4]
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB56_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -17251,10 +17241,9 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v0, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v0, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v0, v5
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v0.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v0, v[3:4], v[5:6] glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -17266,7 +17255,6 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB56_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -17564,14 +17552,12 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__amdgpu_no_fine_grained_memory(
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
@@ -17687,7 +17673,6 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__amdgpu_no_fine_grained_memory(
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v4, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v5, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB57_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -17705,13 +17690,12 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__amdgpu_no_fine_grained_memory(
 ; GFX11-TRUE16-NEXT:    v_or_b32_e32 v10, 0x400000, v6
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -17724,7 +17708,6 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__amdgpu_no_fine_grained_memory(
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB57_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -18001,14 +17984,12 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine_
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
@@ -18124,7 +18105,6 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine_
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v4, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v5, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB58_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -18142,13 +18122,12 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine_
 ; GFX11-TRUE16-NEXT:    v_or_b32_e32 v10, 0x400000, v6
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -18161,7 +18140,6 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine_
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB58_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -18445,14 +18423,12 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:-2048 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
@@ -18577,7 +18553,6 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v5, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
 ; GFX11-TRUE16-NEXT:    flat_load_b32 v3, v[0:1]
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB59_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -18595,13 +18570,12 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX11-TRUE16-NEXT:    v_or_b32_e32 v10, 0x400000, v6
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -18614,7 +18588,6 @@ define void @flat_agent_atomic_fmax_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB59_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -18916,11 +18889,10 @@ define <2 x bfloat> @flat_system_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_n
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v3, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v3.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
 ; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v3, v5
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v3.h
 ; GFX12-TRUE16-NEXT:    global_wb scope:SCOPE_SYS
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v3, v[0:1], v[5:6] offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_SYS
@@ -19040,7 +19012,6 @@ define <2 x bfloat> @flat_system_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_n
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v4, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB60_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -19063,10 +19034,9 @@ define <2 x bfloat> @flat_system_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_n
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v3, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v3, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v5, v5
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v3.h
 ; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v5, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v5, 0xffff, v3, v5
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v5.l, v3.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v3, v[0:1], v[5:6] offset:2044 glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -19078,7 +19048,6 @@ define <2 x bfloat> @flat_system_atomic_fmax_ret_v2bf16__offset12b_pos__amdgpu_n
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB60_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
@@ -19369,14 +19338,12 @@ define void @flat_system_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX12-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX12-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
 ; GFX12-TRUE16-NEXT:    s_wait_alu 0xfffd
-; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX12-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX12-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX12-TRUE16-NEXT:    global_wb scope:SCOPE_SYS
 ; GFX12-TRUE16-NEXT:    s_wait_storecnt 0x0
 ; GFX12-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 th:TH_ATOMIC_RETURN scope:SCOPE_SYS
@@ -19494,7 +19461,6 @@ define void @flat_system_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v4, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v5, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB61_1: ; %atomicrmw.start
 ; GFX11-TRUE16-NEXT:    ; =>This Inner Loop Header: Depth=1
@@ -19512,13 +19478,12 @@ define void @flat_system_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine
 ; GFX11-TRUE16-NEXT:    v_or_b32_e32 v10, 0x400000, v6
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v7, v7, v2, 0x7fff
 ; GFX11-TRUE16-NEXT:    v_add3_u32 v8, v8, v6, 0x7fff
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v7, v9, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_cmp_u_f32_e32 vcc_lo, v6, v6
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v2.h
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v6, v8, v10, vcc_lo
-; GFX11-TRUE16-NEXT:    v_bfi_b32 v2, 0xffff, v2, v6
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e32 v2, v8, v10, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v7.h
 ; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-TRUE16-NEXT:    flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 glc
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -19531,7 +19496,6 @@ define void @flat_system_atomic_fmax_noret_v2bf16__offset12b_pos__amdgpu_no_fine
 ; GFX11-TRUE16-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_cbranch_execnz .LBB61_1
 ; GFX11-TRUE16-NEXT:  ; %bb.2: ; %atomicrmw.end
-; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x2
 ; GFX11-TRUE16-NEXT:    s_or_b32 exec_lo, exec_lo, s0
 ; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;