diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll | 1207 |
1 files changed, 534 insertions, 673 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll index f7a1fb3..316ba85 100644 --- a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll +++ b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll @@ -37,10 +37,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX942-NEXT: v_mov_b32_e32 v1, v0 ; GFX942-NEXT: v_mov_b32_e32 v0, s16 ; GFX942-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s6, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[4:5], 0 ; GFX942-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX942-NEXT: v_mov_b32_e32 v3, s6 +; GFX942-NEXT: v_mov_b32_e32 v3, s16 ; GFX942-NEXT: .LBB0_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -49,7 +48,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX942-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX942-NEXT: v_mov_b64_e32 v[0:1], v[4:5] ; GFX942-NEXT: buffer_wbl2 sc1 -; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -88,10 +87,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX90A-NEXT: v_mov_b32_e32 v1, v0 ; GFX90A-NEXT: v_mov_b32_e32 v0, s20 ; GFX90A-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX90A-NEXT: v_mov_b32_e32 v3, s6 +; GFX90A-NEXT: v_mov_b32_e32 v3, s20 ; GFX90A-NEXT: .LBB0_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -99,7 +97,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX90A-NEXT: v_max_f32_e32 v0, v5, v5 ; GFX90A-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[4:5] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -116,10 +114,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX908-NEXT: v_mov_b32_e32 v1, v0 ; GFX908-NEXT: v_mov_b32_e32 v0, s20 ; GFX908-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s6, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 ; GFX908-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX908-NEXT: v_mov_b32_e32 v3, s6 +; GFX908-NEXT: v_mov_b32_e32 v3, s20 ; GFX908-NEXT: .LBB0_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -128,7 +125,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX908-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX908-NEXT: v_mov_b32_e32 v0, v4 ; GFX908-NEXT: v_mov_b32_e32 v1, v5 -; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -145,10 +142,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX8-NEXT: v_mov_b32_e32 v1, v0 ; GFX8-NEXT: v_mov_b32_e32 v0, s20 ; GFX8-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s6, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 ; GFX8-NEXT: v_mul_f32_e32 v2, 1.0, v1 -; GFX8-NEXT: v_mov_b32_e32 v3, s6 +; GFX8-NEXT: v_mov_b32_e32 v3, s20 ; GFX8-NEXT: .LBB0_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -157,7 +153,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX8-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX8-NEXT: v_mov_b32_e32 v0, v4 ; GFX8-NEXT: v_mov_b32_e32 v1, v5 -; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -212,10 +208,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset__amdgpu_no_fine_ ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX942-NEXT: v_mov_b32_e32 v1, s16 ; GFX942-NEXT: buffer_load_dword v1, v1, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s6, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[4:5], 0 ; GFX942-NEXT: v_max_f32_e32 v2, v0, v0 -; GFX942-NEXT: v_mov_b32_e32 v3, s6 +; GFX942-NEXT: v_mov_b32_e32 v3, s16 ; GFX942-NEXT: .LBB1_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -223,7 +218,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset__amdgpu_no_fine_ ; GFX942-NEXT: v_max_f32_e32 v0, v0, v2 ; GFX942-NEXT: v_mov_b64_e32 v[4:5], v[0:1] ; GFX942-NEXT: buffer_wbl2 sc1 -; GFX942-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v1 @@ -262,17 +257,16 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset__amdgpu_no_fine_ ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_mov_b32_e32 v1, s20 ; GFX90A-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_max_f32_e32 v2, v0, v0 -; GFX90A-NEXT: v_mov_b32_e32 v3, s6 +; GFX90A-NEXT: v_mov_b32_e32 v3, s20 ; GFX90A-NEXT: .LBB1_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_max_f32_e32 v0, v1, v1 ; GFX90A-NEXT: v_max_f32_e32 v0, v0, v2 ; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[0:1], v[0:1] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v1 @@ -289,10 +283,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset__amdgpu_no_fine_ ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX908-NEXT: v_mov_b32_e32 v1, s20 ; GFX908-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s6, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 ; GFX908-NEXT: v_max_f32_e32 v2, v0, v0 -; GFX908-NEXT: v_mov_b32_e32 v3, s6 +; GFX908-NEXT: v_mov_b32_e32 v3, s20 ; GFX908-NEXT: .LBB1_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -300,7 +293,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_max_f32_e32 v0, v0, v2 ; GFX908-NEXT: v_mov_b32_e32 v5, v1 ; GFX908-NEXT: v_mov_b32_e32 v4, v0 -; GFX908-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v1 @@ -317,10 +310,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset__amdgpu_no_fine_ ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-NEXT: v_mov_b32_e32 v1, s20 ; GFX8-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s6, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 ; GFX8-NEXT: v_mul_f32_e32 v2, 1.0, v0 -; GFX8-NEXT: v_mov_b32_e32 v3, s6 +; GFX8-NEXT: v_mov_b32_e32 v3, s20 ; GFX8-NEXT: .LBB1_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -328,7 +320,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f32__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_max_f32_e32 v0, v0, v2 ; GFX8-NEXT: v_mov_b32_e32 v5, v1 ; GFX8-NEXT: v_mov_b32_e32 v4, v0 -; GFX8-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v1 @@ -402,7 +394,6 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX942-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX942: ; %bb.0: ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX942-NEXT: v_add_u32_e32 v8, 0x400, v4 ; GFX942-NEXT: s_mov_b64 s[2:3], exec ; GFX942-NEXT: .LBB2_1: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: v_readfirstlane_b32 s4, v0 @@ -414,22 +405,21 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX942-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], v[2:3] ; GFX942-NEXT: s_and_b64 s[0:1], vcc, s[0:1] ; GFX942-NEXT: s_and_saveexec_b64 s[0:1], s[0:1] -; GFX942-NEXT: buffer_load_dword v7, v4, s[4:7], 0 offen offset:1024 -; GFX942-NEXT: ; implicit-def: $vgpr4 +; GFX942-NEXT: buffer_load_dword v9, v4, s[4:7], 0 offen offset:1024 ; GFX942-NEXT: s_xor_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB2_1 ; GFX942-NEXT: ; %bb.2: ; GFX942-NEXT: s_mov_b64 exec, s[2:3] ; GFX942-NEXT: s_mov_b64 s[2:3], 0 -; GFX942-NEXT: v_max_f32_e32 v9, v5, v5 +; GFX942-NEXT: v_max_f32_e32 v5, v5, v5 ; GFX942-NEXT: .LBB2_3: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Loop Header: Depth=1 ; GFX942-NEXT: ; Child Loop BB2_4 Depth 2 ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: v_max_f32_e32 v4, v7, v7 -; GFX942-NEXT: v_max_f32_e32 v6, v4, v9 +; GFX942-NEXT: v_max_f32_e32 v6, v9, v9 +; GFX942-NEXT: v_max_f32_e32 v8, v6, v5 ; GFX942-NEXT: s_mov_b64 s[8:9], exec -; GFX942-NEXT: v_mov_b64_e32 v[4:5], v[6:7] +; GFX942-NEXT: v_mov_b64_e32 v[6:7], v[8:9] ; GFX942-NEXT: buffer_wbl2 sc1 ; GFX942-NEXT: .LBB2_4: ; Parent Loop BB2_3 Depth=1 ; GFX942-NEXT: ; => This Inner Loop Header: Depth=2 @@ -443,21 +433,21 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX942-NEXT: s_and_b64 s[0:1], vcc, s[0:1] ; GFX942-NEXT: s_and_saveexec_b64 s[0:1], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: buffer_atomic_cmpswap v[4:5], v8, s[4:7], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[4:7], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_xor_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB2_4 ; GFX942-NEXT: ; %bb.5: ; in Loop: Header=BB2_3 Depth=1 ; GFX942-NEXT: s_mov_b64 exec, s[8:9] ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v7 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v6, v9 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v7, v4 +; GFX942-NEXT: v_mov_b32_e32 v9, v6 ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB2_3 ; GFX942-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX942-NEXT: s_or_b64 exec, exec, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v0, v4 +; GFX942-NEXT: v_mov_b32_e32 v0, v6 ; GFX942-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgpu_no_fine_grained_memory: @@ -522,7 +512,6 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_add_u32_e32 v8, 0x400, v4 ; GFX90A-NEXT: s_mov_b64 s[6:7], exec ; GFX90A-NEXT: .LBB2_1: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: v_readfirstlane_b32 s8, v0 @@ -534,22 +523,21 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX90A-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX90A-NEXT: s_nop 0 -; GFX90A-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 -; GFX90A-NEXT: ; implicit-def: $vgpr4 +; GFX90A-NEXT: buffer_load_dword v9, v4, s[8:11], 0 offen offset:1024 ; GFX90A-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB2_1 ; GFX90A-NEXT: ; %bb.2: ; GFX90A-NEXT: s_mov_b64 exec, s[6:7] ; GFX90A-NEXT: s_mov_b64 s[6:7], 0 -; GFX90A-NEXT: v_max_f32_e32 v9, v5, v5 +; GFX90A-NEXT: v_max_f32_e32 v5, v5, v5 ; GFX90A-NEXT: .LBB2_3: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Loop Header: Depth=1 ; GFX90A-NEXT: ; Child Loop BB2_4 Depth 2 ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_max_f32_e32 v4, v7, v7 -; GFX90A-NEXT: v_max_f32_e32 v6, v4, v9 +; GFX90A-NEXT: v_max_f32_e32 v6, v9, v9 +; GFX90A-NEXT: v_max_f32_e32 v8, v6, v5 ; GFX90A-NEXT: s_mov_b64 s[12:13], exec -; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[6:7], v[6:7] op_sel:[0,1] +; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[8:9], v[8:9] op_sel:[0,1] ; GFX90A-NEXT: .LBB2_4: ; Parent Loop BB2_3 Depth=1 ; GFX90A-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX90A-NEXT: v_readfirstlane_b32 s8, v0 @@ -561,27 +549,26 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX90A-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: buffer_atomic_cmpswap v[4:5], v8, s[8:11], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[8:11], 0 offen offset:1024 glc ; GFX90A-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB2_4 ; GFX90A-NEXT: ; %bb.5: ; in Loop: Header=BB2_3 Depth=1 ; GFX90A-NEXT: s_mov_b64 exec, s[12:13] ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v7 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v6, v9 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v7, v4 +; GFX90A-NEXT: v_mov_b32_e32 v9, v6 ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB2_3 ; GFX90A-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX90A-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v0, v4 +; GFX90A-NEXT: v_mov_b32_e32 v0, v6 ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX908: ; %bb.0: ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX908-NEXT: v_add_u32_e32 v7, 0x400, v4 ; GFX908-NEXT: s_mov_b64 s[6:7], exec ; GFX908-NEXT: .LBB2_1: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: v_readfirstlane_b32 s8, v0 @@ -593,8 +580,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX908-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_nop 0 -; GFX908-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX908-NEXT: ; implicit-def: $vgpr4 +; GFX908-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB2_1 ; GFX908-NEXT: ; %bb.2: @@ -605,11 +591,11 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX908-NEXT: ; =>This Loop Header: Depth=1 ; GFX908-NEXT: ; Child Loop BB2_4 Depth 2 ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: v_max_f32_e32 v4, v6, v6 -; GFX908-NEXT: v_max_f32_e32 v5, v4, v8 -; GFX908-NEXT: v_mov_b32_e32 v4, v5 -; GFX908-NEXT: s_mov_b64 s[12:13], exec +; GFX908-NEXT: v_max_f32_e32 v5, v7, v7 +; GFX908-NEXT: v_max_f32_e32 v6, v5, v8 ; GFX908-NEXT: v_mov_b32_e32 v5, v6 +; GFX908-NEXT: s_mov_b64 s[12:13], exec +; GFX908-NEXT: v_mov_b32_e32 v6, v7 ; GFX908-NEXT: .LBB2_4: ; Parent Loop BB2_3 Depth=1 ; GFX908-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX908-NEXT: v_readfirstlane_b32 s8, v0 @@ -621,27 +607,26 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX908-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB2_4 ; GFX908-NEXT: ; %bb.5: ; in Loop: Header=BB2_3 Depth=1 ; GFX908-NEXT: s_mov_b64 exec, s[12:13] ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v6 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v5, v7 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v6, v4 +; GFX908-NEXT: v_mov_b32_e32 v7, v5 ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB2_3 ; GFX908-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX908-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v0, v4 +; GFX908-NEXT: v_mov_b32_e32 v0, v5 ; GFX908-NEXT: s_setpc_b64 s[30:31] ; ; GFX8-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_u32_e32 v7, vcc, 0x400, v4 ; GFX8-NEXT: s_mov_b64 s[6:7], exec ; GFX8-NEXT: .LBB2_1: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: v_readfirstlane_b32 s8, v0 @@ -653,8 +638,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX8-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_nop 0 -; GFX8-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX8-NEXT: ; implicit-def: $vgpr4 +; GFX8-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB2_1 ; GFX8-NEXT: ; %bb.2: @@ -665,11 +649,11 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX8-NEXT: ; =>This Loop Header: Depth=1 ; GFX8-NEXT: ; Child Loop BB2_4 Depth 2 ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: v_mul_f32_e32 v4, 1.0, v6 -; GFX8-NEXT: v_max_f32_e32 v5, v4, v8 -; GFX8-NEXT: v_mov_b32_e32 v4, v5 -; GFX8-NEXT: s_mov_b64 s[12:13], exec +; GFX8-NEXT: v_mul_f32_e32 v5, 1.0, v7 +; GFX8-NEXT: v_max_f32_e32 v6, v5, v8 ; GFX8-NEXT: v_mov_b32_e32 v5, v6 +; GFX8-NEXT: s_mov_b64 s[12:13], exec +; GFX8-NEXT: v_mov_b32_e32 v6, v7 ; GFX8-NEXT: .LBB2_4: ; Parent Loop BB2_3 Depth=1 ; GFX8-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX8-NEXT: v_readfirstlane_b32 s8, v0 @@ -681,21 +665,21 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgp ; GFX8-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB2_4 ; GFX8-NEXT: ; %bb.5: ; in Loop: Header=BB2_3 Depth=1 ; GFX8-NEXT: s_mov_b64 exec, s[12:13] ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v6 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v5, v7 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v7, v5 ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB2_3 ; GFX8-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX8-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v0, v4 +; GFX8-NEXT: v_mov_b32_e32 v0, v5 ; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__waterfall__amdgpu_no_fine_grained_memory: @@ -777,10 +761,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX942-NEXT: v_mov_b32_e32 v1, v0 ; GFX942-NEXT: v_mov_b32_e32 v0, s16 ; GFX942-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s6, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[4:5], 0 ; GFX942-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX942-NEXT: v_mov_b32_e32 v3, s6 +; GFX942-NEXT: v_mov_b32_e32 v3, s16 ; GFX942-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -789,7 +772,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX942-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX942-NEXT: v_mov_b64_e32 v[0:1], v[4:5] ; GFX942-NEXT: buffer_wbl2 sc1 -; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -804,11 +787,10 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s16 -; GFX11-NEXT: s_add_i32 s4, s16, 0x400 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) -; GFX11-NEXT: v_dual_mov_b32 v3, s4 :: v_dual_max_f32 v2, v1, v1 -; GFX11-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024 ; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_dual_mov_b32 v3, s16 :: v_dual_max_f32 v2, v1, v1 +; GFX11-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024 ; GFX11-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt vmcnt(0) @@ -819,7 +801,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX11-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5 -; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen offset:1024 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv @@ -837,11 +819,10 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mov_b32_e32 v1, v0 ; GFX10-NEXT: v_mov_b32_e32 v0, s20 -; GFX10-NEXT: s_add_i32 s4, s20, 0x400 -; GFX10-NEXT: v_mov_b32_e32 v3, s4 +; GFX10-NEXT: v_mov_b32_e32 v3, s20 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: v_max_f32_e32 v2, v1, v1 ; GFX10-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt vmcnt(0) @@ -851,7 +832,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX10-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX10-NEXT: v_mov_b32_e32 v0, v4 ; GFX10-NEXT: v_mov_b32_e32 v1, v5 -; GFX10-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv @@ -869,10 +850,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX90A-NEXT: v_mov_b32_e32 v1, v0 ; GFX90A-NEXT: v_mov_b32_e32 v0, s20 ; GFX90A-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX90A-NEXT: v_mov_b32_e32 v3, s6 +; GFX90A-NEXT: v_mov_b32_e32 v3, s20 ; GFX90A-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -880,7 +860,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX90A-NEXT: v_max_f32_e32 v0, v5, v5 ; GFX90A-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[4:5] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -897,10 +877,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX908-NEXT: v_mov_b32_e32 v1, v0 ; GFX908-NEXT: v_mov_b32_e32 v0, s20 ; GFX908-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s6, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 ; GFX908-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX908-NEXT: v_mov_b32_e32 v3, s6 +; GFX908-NEXT: v_mov_b32_e32 v3, s20 ; GFX908-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -909,7 +888,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX908-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX908-NEXT: v_mov_b32_e32 v0, v4 ; GFX908-NEXT: v_mov_b32_e32 v1, v5 -; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -926,10 +905,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX8-NEXT: v_mov_b32_e32 v1, v0 ; GFX8-NEXT: v_mov_b32_e32 v0, s20 ; GFX8-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s6, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 ; GFX8-NEXT: v_mul_f32_e32 v2, 1.0, v1 -; GFX8-NEXT: v_mov_b32_e32 v3, s6 +; GFX8-NEXT: v_mov_b32_e32 v3, s20 ; GFX8-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -938,7 +916,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX8-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX8-NEXT: v_mov_b32_e32 v0, v4 ; GFX8-NEXT: v_mov_b32_e32 v1, v5 -; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -955,10 +933,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX7-NEXT: v_mov_b32_e32 v1, v0 ; GFX7-NEXT: v_mov_b32_e32 v0, s20 ; GFX7-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX7-NEXT: s_add_i32 s6, s20, 0x400 ; GFX7-NEXT: s_mov_b64 s[4:5], 0 ; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v1 -; GFX7-NEXT: v_mov_b32_e32 v3, s6 +; GFX7-NEXT: v_mov_b32_e32 v3, s20 ; GFX7-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt vmcnt(0) @@ -967,7 +944,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_remote ; GFX7-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX7-NEXT: v_mov_b32_e32 v0, v4 ; GFX7-NEXT: v_mov_b32_e32 v1, v5 -; GFX7-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -1035,10 +1012,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX942-NEXT: v_mov_b32_e32 v1, v0 ; GFX942-NEXT: v_mov_b32_e32 v0, s16 ; GFX942-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s6, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[4:5], 0 ; GFX942-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX942-NEXT: v_mov_b32_e32 v3, s6 +; GFX942-NEXT: v_mov_b32_e32 v3, s16 ; GFX942-NEXT: .LBB4_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -1047,7 +1023,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX942-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX942-NEXT: v_mov_b64_e32 v[0:1], v[4:5] ; GFX942-NEXT: buffer_wbl2 sc1 -; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -1086,10 +1062,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX90A-NEXT: v_mov_b32_e32 v1, v0 ; GFX90A-NEXT: v_mov_b32_e32 v0, s20 ; GFX90A-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX90A-NEXT: v_mov_b32_e32 v3, s6 +; GFX90A-NEXT: v_mov_b32_e32 v3, s20 ; GFX90A-NEXT: .LBB4_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -1097,7 +1072,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX90A-NEXT: v_max_f32_e32 v0, v5, v5 ; GFX90A-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[4:5] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -1114,10 +1089,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX908-NEXT: v_mov_b32_e32 v1, v0 ; GFX908-NEXT: v_mov_b32_e32 v0, s20 ; GFX908-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s6, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 ; GFX908-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX908-NEXT: v_mov_b32_e32 v3, s6 +; GFX908-NEXT: v_mov_b32_e32 v3, s20 ; GFX908-NEXT: .LBB4_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -1126,7 +1100,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX908-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX908-NEXT: v_mov_b32_e32 v0, v4 ; GFX908-NEXT: v_mov_b32_e32 v1, v5 -; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -1143,10 +1117,9 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX8-NEXT: v_mov_b32_e32 v1, v0 ; GFX8-NEXT: v_mov_b32_e32 v0, s20 ; GFX8-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s6, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 ; GFX8-NEXT: v_mul_f32_e32 v2, 1.0, v1 -; GFX8-NEXT: v_mov_b32_e32 v3, s6 +; GFX8-NEXT: v_mov_b32_e32 v3, s20 ; GFX8-NEXT: .LBB4_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -1155,7 +1128,7 @@ define float @buffer_fat_ptr_agent_atomic_fmax_ret_f32__offset__amdgpu_no_fine_g ; GFX8-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX8-NEXT: v_mov_b32_e32 v0, v4 ; GFX8-NEXT: v_mov_b32_e32 v1, v5 -; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -1203,12 +1176,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 ; GFX12-NEXT: v_mov_b32_e32 v0, s16 -; GFX12-NEXT: s_add_co_i32 s4, s16, 0x800 -; GFX12-NEXT: s_wait_alu 0xfffe -; GFX12-NEXT: v_mov_b32_e32 v6, s4 +; GFX12-NEXT: v_mov_b32_e32 v6, s16 +; GFX12-NEXT: s_mov_b32 s4, 0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) ; GFX12-NEXT: v_max_num_f64_e32 v[4:5], v[2:3], v[2:3] ; GFX12-NEXT: buffer_load_b64 v[0:1], v0, s[0:3], null offen offset:2048 -; GFX12-NEXT: s_mov_b32 s4, 0 ; GFX12-NEXT: .LBB5_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_loadcnt 0x0 @@ -1220,7 +1192,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-NEXT: v_dual_mov_b32 v0, v7 :: v_dual_mov_b32 v1, v8 ; GFX12-NEXT: v_dual_mov_b32 v2, v9 :: v_dual_mov_b32 v3, v10 -; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], null offen offset:2048 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV ; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[9:10] @@ -1248,12 +1220,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 ; GFX11-NEXT: v_mov_b32_e32 v0, s16 -; GFX11-NEXT: s_add_i32 s4, s16, 0x800 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NEXT: v_mov_b32_e32 v6, s4 +; GFX11-NEXT: v_mov_b32_e32 v6, s16 +; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) ; GFX11-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] ; GFX11-NEXT: buffer_load_b64 v[0:1], v0, s[0:3], 0 offen offset:2048 -; GFX11-NEXT: s_mov_b32 s4, 0 ; GFX11-NEXT: .LBB5_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt vmcnt(0) @@ -1265,7 +1236,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_dual_mov_b32 v0, v7 :: v_dual_mov_b32 v1, v8 ; GFX11-NEXT: v_dual_mov_b32 v2, v9 :: v_dual_mov_b32 v3, v10 -; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], 0 offen offset:2048 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv @@ -1306,9 +1277,8 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_mov_b32_e32 v3, v1 ; GFX908-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 ; GFX908-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX908-NEXT: s_add_i32 s6, s20, 0x800 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 -; GFX908-NEXT: v_mov_b32_e32 v6, s6 +; GFX908-NEXT: v_mov_b32_e32 v6, s20 ; GFX908-NEXT: .LBB5_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -1320,7 +1290,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_mov_b32_e32 v1, v8 ; GFX908-NEXT: v_mov_b32_e32 v2, v9 ; GFX908-NEXT: v_mov_b32_e32 v3, v10 -; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10] @@ -1339,9 +1309,8 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_mov_b32_e32 v3, v1 ; GFX8-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 ; GFX8-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX8-NEXT: s_add_i32 s6, s20, 0x800 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 -; GFX8-NEXT: v_mov_b32_e32 v6, s6 +; GFX8-NEXT: v_mov_b32_e32 v6, s20 ; GFX8-NEXT: .LBB5_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -1353,7 +1322,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_mov_b32_e32 v1, v8 ; GFX8-NEXT: v_mov_b32_e32 v2, v9 ; GFX8-NEXT: v_mov_b32_e32 v3, v10 -; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10] @@ -1397,11 +1366,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: v_mov_b32_e32 v2, s16 ; GFX12-NEXT: v_max_num_f64_e32 v[4:5], v[0:1], v[0:1] -; GFX12-NEXT: s_add_co_i32 s4, s16, 0x800 -; GFX12-NEXT: s_wait_alu 0xfffe -; GFX12-NEXT: v_mov_b32_e32 v6, s4 -; GFX12-NEXT: buffer_load_b64 v[2:3], v2, s[0:3], null offen offset:2048 +; GFX12-NEXT: v_mov_b32_e32 v6, s16 ; GFX12-NEXT: s_mov_b32 s4, 0 +; GFX12-NEXT: buffer_load_b64 v[2:3], v2, s[0:3], null offen offset:2048 ; GFX12-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_loadcnt 0x0 @@ -1411,7 +1378,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX12-NEXT: v_max_num_f64_e32 v[0:1], v[0:1], v[4:5] ; GFX12-NEXT: v_dual_mov_b32 v10, v3 :: v_dual_mov_b32 v9, v2 ; GFX12-NEXT: v_dual_mov_b32 v8, v1 :: v_dual_mov_b32 v7, v0 -; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[7:10], v6, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[7:10], v6, s[0:3], null offen offset:2048 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV ; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[2:3] @@ -1440,11 +1407,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_mov_b32_e32 v2, s16 ; GFX11-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1] -; GFX11-NEXT: s_add_i32 s4, s16, 0x800 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: v_mov_b32_e32 v6, s4 -; GFX11-NEXT: buffer_load_b64 v[2:3], v2, s[0:3], 0 offen offset:2048 +; GFX11-NEXT: v_mov_b32_e32 v6, s16 ; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: buffer_load_b64 v[2:3], v2, s[0:3], 0 offen offset:2048 ; GFX11-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt vmcnt(0) @@ -1454,7 +1419,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[4:5] ; GFX11-NEXT: v_dual_mov_b32 v10, v3 :: v_dual_mov_b32 v9, v2 ; GFX11-NEXT: v_dual_mov_b32 v8, v1 :: v_dual_mov_b32 v7, v0 -; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[7:10], v6, s[0:3], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[7:10], v6, s[0:3], 0 offen offset:2048 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv @@ -1494,9 +1459,8 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_mov_b32_e32 v2, s20 ; GFX908-NEXT: buffer_load_dwordx2 v[2:3], v2, s[16:19], 0 offen offset:2048 ; GFX908-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1] -; GFX908-NEXT: s_add_i32 s6, s20, 0x800 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 -; GFX908-NEXT: v_mov_b32_e32 v6, s6 +; GFX908-NEXT: v_mov_b32_e32 v6, s20 ; GFX908-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -1506,7 +1470,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_mov_b32_e32 v9, v2 ; GFX908-NEXT: v_mov_b32_e32 v8, v1 ; GFX908-NEXT: v_mov_b32_e32 v7, v0 -; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[7:10], v6, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[7:10], v6, s[16:19], 0 offen offset:2048 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[7:8], v[2:3] @@ -1525,9 +1489,8 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_mov_b32_e32 v2, s20 ; GFX8-NEXT: buffer_load_dwordx2 v[2:3], v2, s[16:19], 0 offen offset:2048 ; GFX8-NEXT: v_max_f64 v[4:5], v[0:1], v[0:1] -; GFX8-NEXT: s_add_i32 s6, s20, 0x800 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 -; GFX8-NEXT: v_mov_b32_e32 v6, s6 +; GFX8-NEXT: v_mov_b32_e32 v6, s20 ; GFX8-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -1537,7 +1500,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_f64__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_mov_b32_e32 v9, v2 ; GFX8-NEXT: v_mov_b32_e32 v8, v1 ; GFX8-NEXT: v_mov_b32_e32 v7, v0 -; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[7:10], v6, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[7:10], v6, s[16:19], 0 offen offset:2048 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[7:8], v[2:3] @@ -1583,10 +1546,9 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: v_dual_mov_b32 v8, v3 :: v_dual_mov_b32 v7, v2 ; GFX12-NEXT: v_dual_mov_b32 v10, v1 :: v_dual_mov_b32 v9, v0 -; GFX12-NEXT: v_add_nc_u32_e32 v15, 0x800, v4 ; GFX12-NEXT: s_mov_b32 s1, exec_lo ; GFX12-NEXT: .LBB7_1: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-NEXT: v_readfirstlane_b32 s4, v9 ; GFX12-NEXT: v_readfirstlane_b32 s5, v10 ; GFX12-NEXT: v_readfirstlane_b32 s6, v7 @@ -1600,12 +1562,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX12-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: buffer_load_b64 v[13:14], v4, s[4:7], null offen offset:2048 -; GFX12-NEXT: ; implicit-def: $vgpr4 ; GFX12-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_cbranch_execnz .LBB7_1 ; GFX12-NEXT: ; %bb.2: ; GFX12-NEXT: s_mov_b32 exec_lo, s1 -; GFX12-NEXT: v_max_num_f64_e32 v[4:5], v[5:6], v[5:6] +; GFX12-NEXT: v_max_num_f64_e32 v[5:6], v[5:6], v[5:6] ; GFX12-NEXT: s_mov_b32 s1, 0 ; GFX12-NEXT: .LBB7_3: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Loop Header: Depth=1 @@ -1615,7 +1576,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX12-NEXT: s_mov_b32 s2, exec_lo ; GFX12-NEXT: s_wait_storecnt 0x0 ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_max_num_f64_e32 v[11:12], v[0:1], v[4:5] +; GFX12-NEXT: v_max_num_f64_e32 v[11:12], v[0:1], v[5:6] ; GFX12-NEXT: v_dual_mov_b32 v0, v11 :: v_dual_mov_b32 v1, v12 ; GFX12-NEXT: v_dual_mov_b32 v2, v13 :: v_dual_mov_b32 v3, v14 ; GFX12-NEXT: .LBB7_4: ; Parent Loop BB7_3 Depth=1 @@ -1632,7 +1593,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-NEXT: s_wait_loadcnt 0x0 -; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v15, s[4:7], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v4, s[4:7], null offen offset:2048 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_cbranch_execnz .LBB7_4 ; GFX12-NEXT: ; %bb.5: ; in Loop: Header=BB7_3 Depth=1 @@ -1686,27 +1647,26 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v8, v3 :: v_dual_mov_b32 v7, v2 ; GFX11-NEXT: v_dual_mov_b32 v10, v1 :: v_dual_mov_b32 v9, v0 -; GFX11-NEXT: v_add_nc_u32_e32 v15, 0x800, v4 ; GFX11-NEXT: s_mov_b32 s1, 0 ; GFX11-NEXT: s_mov_b32 s2, exec_lo ; GFX11-NEXT: .LBB7_1: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3) +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-NEXT: v_readfirstlane_b32 s4, v9 ; GFX11-NEXT: v_readfirstlane_b32 s5, v10 ; GFX11-NEXT: v_readfirstlane_b32 s6, v7 ; GFX11-NEXT: v_readfirstlane_b32 s7, v8 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[9:10] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) ; GFX11-NEXT: v_cmp_eq_u64_e64 s0, s[6:7], v[7:8] ; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_saveexec_b32 s0, s0 ; GFX11-NEXT: buffer_load_b64 v[13:14], v4, s[4:7], 0 offen offset:2048 -; GFX11-NEXT: ; implicit-def: $vgpr4 ; GFX11-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_cbranch_execnz .LBB7_1 ; GFX11-NEXT: ; %bb.2: ; GFX11-NEXT: s_mov_b32 exec_lo, s2 -; GFX11-NEXT: v_max_f64 v[4:5], v[5:6], v[5:6] +; GFX11-NEXT: v_max_f64 v[5:6], v[5:6], v[5:6] ; GFX11-NEXT: .p2align 6 ; GFX11-NEXT: .LBB7_3: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Loop Header: Depth=1 @@ -1716,7 +1676,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX11-NEXT: s_mov_b32 s2, exec_lo ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_max_f64 v[11:12], v[0:1], v[4:5] +; GFX11-NEXT: v_max_f64 v[11:12], v[0:1], v[5:6] ; GFX11-NEXT: v_dual_mov_b32 v0, v11 :: v_dual_mov_b32 v1, v12 ; GFX11-NEXT: v_dual_mov_b32 v2, v13 :: v_dual_mov_b32 v3, v14 ; GFX11-NEXT: .LBB7_4: ; Parent Loop BB7_3 Depth=1 @@ -1732,7 +1692,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_saveexec_b32 s0, s0 ; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v15, s[4:7], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v4, s[4:7], 0 offen offset:2048 glc ; GFX11-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_cbranch_execnz .LBB7_4 ; GFX11-NEXT: ; %bb.5: ; in Loop: Header=BB7_3 Depth=1 @@ -1816,7 +1776,6 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX908-NEXT: v_mov_b32_e32 v7, v2 ; GFX908-NEXT: v_mov_b32_e32 v10, v1 ; GFX908-NEXT: v_mov_b32_e32 v9, v0 -; GFX908-NEXT: v_add_u32_e32 v15, 0x800, v4 ; GFX908-NEXT: s_mov_b64 s[6:7], exec ; GFX908-NEXT: .LBB7_1: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: v_readfirstlane_b32 s8, v9 @@ -1829,12 +1788,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_nop 0 ; GFX908-NEXT: buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048 -; GFX908-NEXT: ; implicit-def: $vgpr4 ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB7_1 ; GFX908-NEXT: ; %bb.2: ; GFX908-NEXT: s_mov_b64 exec, s[6:7] -; GFX908-NEXT: v_max_f64 v[4:5], v[5:6], v[5:6] +; GFX908-NEXT: v_max_f64 v[5:6], v[5:6], v[5:6] ; GFX908-NEXT: s_mov_b64 s[6:7], 0 ; GFX908-NEXT: .LBB7_3: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Loop Header: Depth=1 @@ -1842,7 +1800,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: v_max_f64 v[0:1], v[13:14], v[13:14] ; GFX908-NEXT: s_mov_b64 s[12:13], exec -; GFX908-NEXT: v_max_f64 v[11:12], v[0:1], v[4:5] +; GFX908-NEXT: v_max_f64 v[11:12], v[0:1], v[5:6] ; GFX908-NEXT: v_mov_b32_e32 v0, v11 ; GFX908-NEXT: v_mov_b32_e32 v1, v12 ; GFX908-NEXT: v_mov_b32_e32 v2, v13 @@ -1858,7 +1816,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX908-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v4, s[8:11], 0 offen offset:2048 glc ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB7_4 ; GFX908-NEXT: ; %bb.5: ; in Loop: Header=BB7_3 Depth=1 @@ -1882,7 +1840,6 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX8-NEXT: v_mov_b32_e32 v7, v2 ; GFX8-NEXT: v_mov_b32_e32 v10, v1 ; GFX8-NEXT: v_mov_b32_e32 v9, v0 -; GFX8-NEXT: v_add_u32_e32 v15, vcc, 0x800, v4 ; GFX8-NEXT: s_mov_b64 s[6:7], exec ; GFX8-NEXT: .LBB7_1: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: v_readfirstlane_b32 s8, v9 @@ -1895,12 +1852,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_nop 0 ; GFX8-NEXT: buffer_load_dwordx2 v[13:14], v4, s[8:11], 0 offen offset:2048 -; GFX8-NEXT: ; implicit-def: $vgpr4 ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB7_1 ; GFX8-NEXT: ; %bb.2: ; GFX8-NEXT: s_mov_b64 exec, s[6:7] -; GFX8-NEXT: v_max_f64 v[4:5], v[5:6], v[5:6] +; GFX8-NEXT: v_max_f64 v[5:6], v[5:6], v[5:6] ; GFX8-NEXT: s_mov_b64 s[6:7], 0 ; GFX8-NEXT: .LBB7_3: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Loop Header: Depth=1 @@ -1908,7 +1864,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: v_max_f64 v[0:1], v[13:14], v[13:14] ; GFX8-NEXT: s_mov_b64 s[12:13], exec -; GFX8-NEXT: v_max_f64 v[11:12], v[0:1], v[4:5] +; GFX8-NEXT: v_max_f64 v[11:12], v[0:1], v[5:6] ; GFX8-NEXT: v_mov_b32_e32 v0, v11 ; GFX8-NEXT: v_mov_b32_e32 v1, v12 ; GFX8-NEXT: v_mov_b32_e32 v2, v13 @@ -1924,7 +1880,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__waterfall__amdg ; GFX8-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v15, s[8:11], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v4, s[8:11], 0 offen offset:2048 glc ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB7_4 ; GFX8-NEXT: ; %bb.5: ; in Loop: Header=BB7_3 Depth=1 @@ -2010,12 +1966,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 ; GFX12-NEXT: v_mov_b32_e32 v0, s16 -; GFX12-NEXT: s_add_co_i32 s4, s16, 0x800 -; GFX12-NEXT: s_wait_alu 0xfffe -; GFX12-NEXT: v_mov_b32_e32 v6, s4 +; GFX12-NEXT: v_mov_b32_e32 v6, s16 +; GFX12-NEXT: s_mov_b32 s4, 0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) ; GFX12-NEXT: v_max_num_f64_e32 v[4:5], v[2:3], v[2:3] ; GFX12-NEXT: buffer_load_b64 v[0:1], v0, s[0:3], null offen offset:2048 -; GFX12-NEXT: s_mov_b32 s4, 0 ; GFX12-NEXT: .LBB8_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_loadcnt 0x0 @@ -2027,7 +1982,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-NEXT: v_dual_mov_b32 v0, v7 :: v_dual_mov_b32 v1, v8 ; GFX12-NEXT: v_dual_mov_b32 v2, v9 :: v_dual_mov_b32 v3, v10 -; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], null offen offset:2048 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV ; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[9:10] @@ -2055,12 +2010,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 ; GFX11-NEXT: v_mov_b32_e32 v0, s16 -; GFX11-NEXT: s_add_i32 s4, s16, 0x800 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NEXT: v_mov_b32_e32 v6, s4 +; GFX11-NEXT: v_mov_b32_e32 v6, s16 +; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) ; GFX11-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] ; GFX11-NEXT: buffer_load_b64 v[0:1], v0, s[0:3], 0 offen offset:2048 -; GFX11-NEXT: s_mov_b32 s4, 0 ; GFX11-NEXT: .LBB8_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt vmcnt(0) @@ -2072,7 +2026,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_dual_mov_b32 v0, v7 :: v_dual_mov_b32 v1, v8 ; GFX11-NEXT: v_dual_mov_b32 v2, v9 :: v_dual_mov_b32 v3, v10 -; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], 0 offen offset:2048 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv @@ -2091,11 +2045,10 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX10-NEXT: v_mov_b32_e32 v2, v0 ; GFX10-NEXT: v_mov_b32_e32 v0, s20 ; GFX10-NEXT: v_mov_b32_e32 v3, v1 -; GFX10-NEXT: s_add_i32 s4, s20, 0x800 -; GFX10-NEXT: v_mov_b32_e32 v6, s4 +; GFX10-NEXT: v_mov_b32_e32 v6, s20 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 ; GFX10-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: .LBB8_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt vmcnt(0) @@ -2108,7 +2061,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX10-NEXT: v_mov_b32_e32 v1, v8 ; GFX10-NEXT: v_mov_b32_e32 v2, v9 ; GFX10-NEXT: v_mov_b32_e32 v3, v10 -; GFX10-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv @@ -2127,10 +2080,9 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX90A-NEXT: v_mov_b32_e32 v0, s20 ; GFX90A-NEXT: v_mov_b32_e32 v3, v1 ; GFX90A-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x800 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX90A-NEXT: v_mov_b32_e32 v6, s6 +; GFX90A-NEXT: v_mov_b32_e32 v6, s20 ; GFX90A-NEXT: .LBB8_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -2139,7 +2091,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX90A-NEXT: v_max_f64 v[8:9], v[0:1], v[4:5] ; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[8:9], v[8:9] op_sel:[0,1] ; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[10:11], v[10:11] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[10:11] @@ -2158,9 +2110,8 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX908-NEXT: v_mov_b32_e32 v3, v1 ; GFX908-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 ; GFX908-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX908-NEXT: s_add_i32 s6, s20, 0x800 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 -; GFX908-NEXT: v_mov_b32_e32 v6, s6 +; GFX908-NEXT: v_mov_b32_e32 v6, s20 ; GFX908-NEXT: .LBB8_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -2172,7 +2123,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX908-NEXT: v_mov_b32_e32 v1, v8 ; GFX908-NEXT: v_mov_b32_e32 v2, v9 ; GFX908-NEXT: v_mov_b32_e32 v3, v10 -; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10] @@ -2191,9 +2142,8 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX8-NEXT: v_mov_b32_e32 v3, v1 ; GFX8-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 ; GFX8-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX8-NEXT: s_add_i32 s6, s20, 0x800 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 -; GFX8-NEXT: v_mov_b32_e32 v6, s6 +; GFX8-NEXT: v_mov_b32_e32 v6, s20 ; GFX8-NEXT: .LBB8_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -2205,7 +2155,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX8-NEXT: v_mov_b32_e32 v1, v8 ; GFX8-NEXT: v_mov_b32_e32 v2, v9 ; GFX8-NEXT: v_mov_b32_e32 v3, v10 -; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10] @@ -2223,10 +2173,9 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX7-NEXT: v_mov_b32_e32 v0, s20 ; GFX7-NEXT: v_mov_b32_e32 v3, v1 ; GFX7-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 -; GFX7-NEXT: s_add_i32 s6, s20, 0x800 ; GFX7-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] ; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mov_b32_e32 v6, s6 +; GFX7-NEXT: v_mov_b32_e32 v6, s20 ; GFX7-NEXT: .LBB8_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt vmcnt(0) @@ -2238,7 +2187,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_remot ; GFX7-NEXT: v_mov_b32_e32 v1, v8 ; GFX7-NEXT: v_mov_b32_e32 v2, v9 ; GFX7-NEXT: v_mov_b32_e32 v3, v10 -; GFX7-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10] @@ -2298,12 +2247,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 ; GFX12-NEXT: v_mov_b32_e32 v0, s16 -; GFX12-NEXT: s_add_co_i32 s4, s16, 0x800 -; GFX12-NEXT: s_wait_alu 0xfffe -; GFX12-NEXT: v_mov_b32_e32 v6, s4 +; GFX12-NEXT: v_mov_b32_e32 v6, s16 +; GFX12-NEXT: s_mov_b32 s4, 0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) ; GFX12-NEXT: v_max_num_f64_e32 v[4:5], v[2:3], v[2:3] ; GFX12-NEXT: buffer_load_b64 v[0:1], v0, s[0:3], null offen offset:2048 -; GFX12-NEXT: s_mov_b32 s4, 0 ; GFX12-NEXT: .LBB9_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_loadcnt 0x0 @@ -2315,7 +2263,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-NEXT: v_dual_mov_b32 v0, v7 :: v_dual_mov_b32 v1, v8 ; GFX12-NEXT: v_dual_mov_b32 v2, v9 :: v_dual_mov_b32 v3, v10 -; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], null offen offset:2048 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV ; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[9:10] @@ -2343,12 +2291,11 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 ; GFX11-NEXT: v_mov_b32_e32 v0, s16 -; GFX11-NEXT: s_add_i32 s4, s16, 0x800 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NEXT: v_mov_b32_e32 v6, s4 +; GFX11-NEXT: v_mov_b32_e32 v6, s16 +; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) ; GFX11-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] ; GFX11-NEXT: buffer_load_b64 v[0:1], v0, s[0:3], 0 offen offset:2048 -; GFX11-NEXT: s_mov_b32 s4, 0 ; GFX11-NEXT: .LBB9_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt vmcnt(0) @@ -2360,7 +2307,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_dual_mov_b32 v0, v7 :: v_dual_mov_b32 v1, v8 ; GFX11-NEXT: v_dual_mov_b32 v2, v9 :: v_dual_mov_b32 v3, v10 -; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b64 v[0:3], v6, s[0:3], 0 offen offset:2048 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv @@ -2401,9 +2348,8 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_mov_b32_e32 v3, v1 ; GFX908-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 ; GFX908-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX908-NEXT: s_add_i32 s6, s20, 0x800 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 -; GFX908-NEXT: v_mov_b32_e32 v6, s6 +; GFX908-NEXT: v_mov_b32_e32 v6, s20 ; GFX908-NEXT: .LBB9_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -2415,7 +2361,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_mov_b32_e32 v1, v8 ; GFX908-NEXT: v_mov_b32_e32 v2, v9 ; GFX908-NEXT: v_mov_b32_e32 v3, v10 -; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10] @@ -2434,9 +2380,8 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_mov_b32_e32 v3, v1 ; GFX8-NEXT: buffer_load_dwordx2 v[0:1], v0, s[16:19], 0 offen offset:2048 ; GFX8-NEXT: v_max_f64 v[4:5], v[2:3], v[2:3] -; GFX8-NEXT: s_add_i32 s6, s20, 0x800 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 -; GFX8-NEXT: v_mov_b32_e32 v6, s6 +; GFX8-NEXT: v_mov_b32_e32 v6, s20 ; GFX8-NEXT: .LBB9_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -2448,7 +2393,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_mov_b32_e32 v1, v8 ; GFX8-NEXT: v_mov_b32_e32 v2, v9 ; GFX8-NEXT: v_mov_b32_e32 v3, v10 -; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap_x2 v[0:3], v6, s[16:19], 0 offen offset:2048 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[9:10] @@ -6146,13 +6091,11 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s16 -; GFX12-NEXT: s_add_co_i32 s4, s16, 0x400 -; GFX12-NEXT: s_wait_alu 0xfffe -; GFX12-NEXT: v_mov_b32_e32 v3, s4 +; GFX12-NEXT: v_mov_b32_e32 v3, s16 +; GFX12-NEXT: s_mov_b32 s4, 0 ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) ; GFX12-NEXT: v_pk_max_num_f16 v2, v1, v1 ; GFX12-NEXT: buffer_load_b32 v0, v0, s[0:3], null offen offset:1024 -; GFX12-NEXT: s_mov_b32 s4, 0 ; GFX12-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_loadcnt 0x0 @@ -6163,7 +6106,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX12-NEXT: v_pk_max_num_f16 v4, v0, v2 ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-NEXT: v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5 -; GFX12-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV ; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v5 @@ -6182,10 +6125,9 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX942-NEXT: v_mov_b32_e32 v1, v0 ; GFX942-NEXT: v_mov_b32_e32 v0, s16 ; GFX942-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s6, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[4:5], 0 ; GFX942-NEXT: v_pk_max_f16 v2, v1, v1 -; GFX942-NEXT: v_mov_b32_e32 v3, s6 +; GFX942-NEXT: v_mov_b32_e32 v3, s16 ; GFX942-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -6195,7 +6137,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX942-NEXT: v_pk_max_f16 v4, v0, v2 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_mov_b64_e32 v[0:1], v[4:5] -; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -6210,12 +6152,11 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s16 -; GFX11-NEXT: s_add_i32 s4, s16, 0x400 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NEXT: v_mov_b32_e32 v3, s4 +; GFX11-NEXT: v_mov_b32_e32 v3, s16 +; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) ; GFX11-NEXT: v_pk_max_f16 v2, v1, v1 ; GFX11-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024 -; GFX11-NEXT: s_mov_b32 s4, 0 ; GFX11-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt vmcnt(0) @@ -6226,7 +6167,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX11-NEXT: v_pk_max_f16 v4, v0, v2 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_dual_mov_b32 v0, v4 :: v_dual_mov_b32 v1, v5 -; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v3, s[0:3], 0 offen offset:1024 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv @@ -6244,11 +6185,10 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mov_b32_e32 v1, v0 ; GFX10-NEXT: v_mov_b32_e32 v0, s20 -; GFX10-NEXT: s_add_i32 s4, s20, 0x400 -; GFX10-NEXT: v_mov_b32_e32 v3, s4 +; GFX10-NEXT: v_mov_b32_e32 v3, s20 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: v_pk_max_f16 v2, v1, v1 ; GFX10-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt vmcnt(0) @@ -6258,7 +6198,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX10-NEXT: v_pk_max_f16 v4, v0, v2 ; GFX10-NEXT: v_mov_b32_e32 v0, v4 ; GFX10-NEXT: v_mov_b32_e32 v1, v5 -; GFX10-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv @@ -6276,10 +6216,9 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX90A-NEXT: v_mov_b32_e32 v1, v0 ; GFX90A-NEXT: v_mov_b32_e32 v0, s20 ; GFX90A-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_pk_max_f16 v2, v1, v1 -; GFX90A-NEXT: v_mov_b32_e32 v3, s6 +; GFX90A-NEXT: v_mov_b32_e32 v3, s20 ; GFX90A-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -6287,7 +6226,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX90A-NEXT: v_pk_max_f16 v0, v5, v5 ; GFX90A-NEXT: v_pk_max_f16 v4, v0, v2 ; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[4:5] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -6304,10 +6243,9 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX908-NEXT: v_mov_b32_e32 v1, v0 ; GFX908-NEXT: v_mov_b32_e32 v0, s20 ; GFX908-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s6, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 ; GFX908-NEXT: v_pk_max_f16 v2, v1, v1 -; GFX908-NEXT: v_mov_b32_e32 v3, s6 +; GFX908-NEXT: v_mov_b32_e32 v3, s20 ; GFX908-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -6316,7 +6254,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX908-NEXT: v_pk_max_f16 v4, v0, v2 ; GFX908-NEXT: v_mov_b32_e32 v0, v4 ; GFX908-NEXT: v_mov_b32_e32 v1, v5 -; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -6333,11 +6271,10 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX8-NEXT: v_mov_b32_e32 v1, v0 ; GFX8-NEXT: v_mov_b32_e32 v0, s20 ; GFX8-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s6, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 ; GFX8-NEXT: v_max_f16_sdwa v2, v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 ; GFX8-NEXT: v_max_f16_e32 v3, v1, v1 -; GFX8-NEXT: v_mov_b32_e32 v4, s6 +; GFX8-NEXT: v_mov_b32_e32 v4, s20 ; GFX8-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -6349,7 +6286,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX8-NEXT: v_or_b32_e32 v5, v1, v0 ; GFX8-NEXT: v_mov_b32_e32 v0, v5 ; GFX8-NEXT: v_mov_b32_e32 v1, v6 -; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v0, v6 @@ -6367,7 +6304,6 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX7-NEXT: buffer_load_dword v3, v2, s[16:19], 0 offen offset:1024 ; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v0 -; GFX7-NEXT: s_add_i32 s6, s20, 0x400 ; GFX7-NEXT: s_mov_b64 s[4:5], 0 ; GFX7-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX7-NEXT: s_waitcnt vmcnt(0) @@ -6375,7 +6311,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v3 ; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v1 ; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v4 -; GFX7-NEXT: v_mov_b32_e32 v4, s6 +; GFX7-NEXT: v_mov_b32_e32 v4, s20 ; GFX7-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1 @@ -6392,7 +6328,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__amdgpu_no ; GFX7-NEXT: v_or_b32_e32 v5, v7, v0 ; GFX7-NEXT: v_mov_b32_e32 v8, v6 ; GFX7-NEXT: v_mov_b32_e32 v7, v5 -; GFX7-NEXT: buffer_atomic_cmpswap v[7:8], v4, s[16:19], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap v[7:8], v4, s[16:19], 0 offen offset:1024 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: v_lshrrev_b32_e32 v1, 16, v7 @@ -6467,10 +6403,8 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: v_mov_b32_e32 v1, s16 -; GFX12-NEXT: s_add_co_i32 s4, s16, 0x400 ; GFX12-NEXT: v_pk_max_num_f16 v2, v0, v0 -; GFX12-NEXT: s_wait_alu 0xfffe -; GFX12-NEXT: v_mov_b32_e32 v3, s4 +; GFX12-NEXT: v_mov_b32_e32 v3, s16 ; GFX12-NEXT: s_mov_b32 s4, 0 ; GFX12-NEXT: buffer_load_b32 v1, v1, s[0:3], null offen offset:1024 ; GFX12-NEXT: .LBB17_1: ; %atomicrmw.start @@ -6481,7 +6415,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-NEXT: v_pk_max_num_f16 v0, v0, v2 ; GFX12-NEXT: v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0 -; GFX12-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV ; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v1 @@ -6500,10 +6434,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX942-NEXT: v_mov_b32_e32 v1, s16 ; GFX942-NEXT: buffer_load_dword v1, v1, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s6, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[4:5], 0 ; GFX942-NEXT: v_pk_max_f16 v2, v0, v0 -; GFX942-NEXT: v_mov_b32_e32 v3, s6 +; GFX942-NEXT: v_mov_b32_e32 v3, s16 ; GFX942-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -6512,7 +6445,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX942-NEXT: v_pk_max_f16 v0, v0, v2 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_mov_b64_e32 v[4:5], v[0:1] -; GFX942-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v1 @@ -6528,9 +6461,8 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_mov_b32_e32 v1, s16 -; GFX11-NEXT: s_add_i32 s4, s16, 0x400 ; GFX11-NEXT: v_pk_max_f16 v2, v0, v0 -; GFX11-NEXT: v_mov_b32_e32 v3, s4 +; GFX11-NEXT: v_mov_b32_e32 v3, s16 ; GFX11-NEXT: s_mov_b32 s4, 0 ; GFX11-NEXT: buffer_load_b32 v1, v1, s[0:3], 0 offen offset:1024 ; GFX11-NEXT: .LBB17_1: ; %atomicrmw.start @@ -6541,7 +6473,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-NEXT: v_pk_max_f16 v0, v0, v2 ; GFX11-NEXT: v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0 -; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], 0 offen offset:1024 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv @@ -6559,9 +6491,8 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mov_b32_e32 v1, s20 -; GFX10-NEXT: s_add_i32 s4, s20, 0x400 ; GFX10-NEXT: v_pk_max_f16 v2, v0, v0 -; GFX10-NEXT: v_mov_b32_e32 v3, s4 +; GFX10-NEXT: v_mov_b32_e32 v3, s20 ; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 ; GFX10-NEXT: .LBB17_1: ; %atomicrmw.start @@ -6572,7 +6503,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX10-NEXT: v_pk_max_f16 v0, v0, v2 ; GFX10-NEXT: v_mov_b32_e32 v5, v1 ; GFX10-NEXT: v_mov_b32_e32 v4, v0 -; GFX10-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen offset:1024 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv @@ -6590,17 +6521,16 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_mov_b32_e32 v1, s20 ; GFX90A-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_pk_max_f16 v2, v0, v0 -; GFX90A-NEXT: v_mov_b32_e32 v3, s6 +; GFX90A-NEXT: v_mov_b32_e32 v3, s20 ; GFX90A-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_pk_max_f16 v0, v1, v1 ; GFX90A-NEXT: v_pk_max_f16 v0, v0, v2 ; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[0:1], v[0:1] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v1 @@ -6617,10 +6547,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX908-NEXT: v_mov_b32_e32 v1, s20 ; GFX908-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s6, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 ; GFX908-NEXT: v_pk_max_f16 v2, v0, v0 -; GFX908-NEXT: v_mov_b32_e32 v3, s6 +; GFX908-NEXT: v_mov_b32_e32 v3, s20 ; GFX908-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -6628,7 +6557,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX908-NEXT: v_pk_max_f16 v0, v0, v2 ; GFX908-NEXT: v_mov_b32_e32 v5, v1 ; GFX908-NEXT: v_mov_b32_e32 v4, v0 -; GFX908-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[4:5], v3, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v1 @@ -6645,11 +6574,10 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-NEXT: v_mov_b32_e32 v1, s20 ; GFX8-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s6, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 ; GFX8-NEXT: v_max_f16_sdwa v2, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 ; GFX8-NEXT: v_max_f16_e32 v3, v0, v0 -; GFX8-NEXT: v_mov_b32_e32 v4, s6 +; GFX8-NEXT: v_mov_b32_e32 v4, s20 ; GFX8-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -6660,7 +6588,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX8-NEXT: v_or_b32_e32 v0, v5, v0 ; GFX8-NEXT: v_mov_b32_e32 v6, v1 ; GFX8-NEXT: v_mov_b32_e32 v5, v0 -; GFX8-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v5, v1 @@ -6679,7 +6607,6 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX7-NEXT: buffer_load_dword v2, v2, s[16:19], 0 offen offset:1024 ; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX7-NEXT: v_cvt_f16_f32_e32 v5, v0 -; GFX7-NEXT: s_add_i32 s6, s20, 0x400 ; GFX7-NEXT: s_mov_b64 s[4:5], 0 ; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v1 ; GFX7-NEXT: s_waitcnt vmcnt(0) @@ -6687,7 +6614,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v2 ; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v1 ; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v5 -; GFX7-NEXT: v_mov_b32_e32 v2, s6 +; GFX7-NEXT: v_mov_b32_e32 v2, s20 ; GFX7-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 @@ -6704,7 +6631,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2f16__offset__amdgpu_no_fin ; GFX7-NEXT: v_or_b32_e32 v4, v6, v3 ; GFX7-NEXT: v_mov_b32_e32 v7, v5 ; GFX7-NEXT: v_mov_b32_e32 v6, v4 -; GFX7-NEXT: buffer_atomic_cmpswap v[6:7], v2, s[16:19], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap v[6:7], v2, s[16:19], 0 offen offset:1024 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v6 @@ -6778,7 +6705,6 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX12-NEXT: s_mov_b32 s1, exec_lo ; GFX12-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: v_readfirstlane_b32 s4, v0 @@ -6793,8 +6719,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-NEXT: s_wait_loadcnt 0x0 -; GFX12-NEXT: buffer_load_b32 v6, v4, s[4:7], null offen offset:1024 -; GFX12-NEXT: ; implicit-def: $vgpr4 +; GFX12-NEXT: buffer_load_b32 v7, v4, s[4:7], null offen offset:1024 ; GFX12-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_cbranch_execnz .LBB18_1 ; GFX12-NEXT: ; %bb.2: @@ -6805,13 +6730,13 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX12-NEXT: ; =>This Loop Header: Depth=1 ; GFX12-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX12-NEXT: s_wait_loadcnt 0x0 -; GFX12-NEXT: v_pk_max_num_f16 v4, v6, v6 +; GFX12-NEXT: v_pk_max_num_f16 v5, v7, v7 ; GFX12-NEXT: s_mov_b32 s2, exec_lo ; GFX12-NEXT: s_wait_storecnt 0x0 ; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_pk_max_num_f16 v5, v4, v8 -; GFX12-NEXT: v_mov_b32_e32 v4, v5 +; GFX12-NEXT: v_pk_max_num_f16 v6, v5, v8 ; GFX12-NEXT: v_mov_b32_e32 v5, v6 +; GFX12-NEXT: v_mov_b32_e32 v6, v7 ; GFX12-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX12-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX12-NEXT: v_readfirstlane_b32 s4, v0 @@ -6826,14 +6751,14 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-NEXT: s_wait_loadcnt 0x0 -; GFX12-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], null offen th:TH_ATOMIC_RETURN +; GFX12-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[4:7], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_cbranch_execnz .LBB18_4 ; GFX12-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX12-NEXT: s_mov_b32 exec_lo, s2 ; GFX12-NEXT: s_wait_loadcnt 0x0 -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX12-NEXT: v_mov_b32_e32 v6, v4 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX12-NEXT: v_mov_b32_e32 v7, v5 ; GFX12-NEXT: global_inv scope:SCOPE_DEV ; GFX12-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -6841,14 +6766,13 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX12-NEXT: s_cbranch_execnz .LBB18_3 ; GFX12-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX12-NEXT: v_mov_b32_e32 v0, v4 +; GFX12-NEXT: v_mov_b32_e32 v0, v5 ; GFX12-NEXT: s_wait_loadcnt 0x0 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX942-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX942: ; %bb.0: ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX942-NEXT: v_add_u32_e32 v8, 0x400, v4 ; GFX942-NEXT: s_mov_b64 s[2:3], exec ; GFX942-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: v_readfirstlane_b32 s4, v0 @@ -6860,23 +6784,22 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX942-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], v[2:3] ; GFX942-NEXT: s_and_b64 s[0:1], vcc, s[0:1] ; GFX942-NEXT: s_and_saveexec_b64 s[0:1], s[0:1] -; GFX942-NEXT: buffer_load_dword v7, v4, s[4:7], 0 offen offset:1024 -; GFX942-NEXT: ; implicit-def: $vgpr4 +; GFX942-NEXT: buffer_load_dword v9, v4, s[4:7], 0 offen offset:1024 ; GFX942-NEXT: s_xor_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB18_1 ; GFX942-NEXT: ; %bb.2: ; GFX942-NEXT: s_mov_b64 exec, s[2:3] ; GFX942-NEXT: s_mov_b64 s[2:3], 0 -; GFX942-NEXT: v_pk_max_f16 v9, v5, v5 +; GFX942-NEXT: v_pk_max_f16 v5, v5, v5 ; GFX942-NEXT: .LBB18_3: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Loop Header: Depth=1 ; GFX942-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: v_pk_max_f16 v4, v7, v7 +; GFX942-NEXT: v_pk_max_f16 v6, v9, v9 ; GFX942-NEXT: s_mov_b64 s[8:9], exec -; GFX942-NEXT: v_pk_max_f16 v6, v4, v9 +; GFX942-NEXT: v_pk_max_f16 v8, v6, v5 ; GFX942-NEXT: buffer_wbl2 sc1 -; GFX942-NEXT: v_mov_b64_e32 v[4:5], v[6:7] +; GFX942-NEXT: v_mov_b64_e32 v[6:7], v[8:9] ; GFX942-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX942-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX942-NEXT: v_readfirstlane_b32 s4, v0 @@ -6889,27 +6812,26 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX942-NEXT: s_and_b64 s[0:1], vcc, s[0:1] ; GFX942-NEXT: s_and_saveexec_b64 s[0:1], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: buffer_atomic_cmpswap v[4:5], v8, s[4:7], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[4:7], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_xor_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB18_4 ; GFX942-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX942-NEXT: s_mov_b64 exec, s[8:9] ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v7 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v6, v9 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v7, v4 +; GFX942-NEXT: v_mov_b32_e32 v9, v6 ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB18_3 ; GFX942-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX942-NEXT: s_or_b64 exec, exec, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v0, v4 +; GFX942-NEXT: v_mov_b32_e32 v0, v6 ; GFX942-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX11-NEXT: s_mov_b32 s1, 0 ; GFX11-NEXT: s_mov_b32 s2, exec_lo ; GFX11-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 @@ -6923,8 +6845,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_saveexec_b32 s0, s0 -; GFX11-NEXT: buffer_load_b32 v6, v4, s[4:7], 0 offen offset:1024 -; GFX11-NEXT: ; implicit-def: $vgpr4 +; GFX11-NEXT: buffer_load_b32 v7, v4, s[4:7], 0 offen offset:1024 ; GFX11-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_cbranch_execnz .LBB18_1 ; GFX11-NEXT: ; %bb.2: @@ -6935,13 +6856,13 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX11-NEXT: ; =>This Loop Header: Depth=1 ; GFX11-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: v_pk_max_f16 v4, v6, v6 +; GFX11-NEXT: v_pk_max_f16 v5, v7, v7 ; GFX11-NEXT: s_mov_b32 s2, exec_lo ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_pk_max_f16 v5, v4, v8 -; GFX11-NEXT: v_mov_b32_e32 v4, v5 +; GFX11-NEXT: v_pk_max_f16 v6, v5, v8 ; GFX11-NEXT: v_mov_b32_e32 v5, v6 +; GFX11-NEXT: v_mov_b32_e32 v6, v7 ; GFX11-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX11-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX11-NEXT: v_readfirstlane_b32 s4, v0 @@ -6955,14 +6876,14 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_saveexec_b32 s0, s0 ; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], 0 offen glc +; GFX11-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[4:7], 0 offen offset:1024 glc ; GFX11-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_cbranch_execnz .LBB18_4 ; GFX11-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX11-NEXT: s_mov_b32 exec_lo, s2 ; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX11-NEXT: v_mov_b32_e32 v6, v4 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX11-NEXT: v_mov_b32_e32 v7, v5 ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv ; GFX11-NEXT: s_or_b32 s1, vcc_lo, s1 @@ -6971,13 +6892,12 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX11-NEXT: s_cbranch_execnz .LBB18_3 ; GFX11-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX11-NEXT: v_mov_b32_e32 v0, v4 +; GFX11-NEXT: v_mov_b32_e32 v0, v5 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX10-NEXT: s_mov_b32 s5, 0 ; GFX10-NEXT: s_mov_b32 s6, exec_lo ; GFX10-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 @@ -6989,8 +6909,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX10-NEXT: v_cmp_eq_u64_e64 s4, s[10:11], v[2:3] ; GFX10-NEXT: s_and_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_and_saveexec_b32 s4, s4 -; GFX10-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX10-NEXT: ; implicit-def: $vgpr4 +; GFX10-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_xor_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB18_1 @@ -7001,12 +6920,12 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX10-NEXT: ; =>This Loop Header: Depth=1 ; GFX10-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: v_pk_max_f16 v4, v6, v6 +; GFX10-NEXT: v_pk_max_f16 v5, v7, v7 ; GFX10-NEXT: s_mov_b32 s6, exec_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: v_pk_max_f16 v5, v4, v8 -; GFX10-NEXT: v_mov_b32_e32 v4, v5 +; GFX10-NEXT: v_pk_max_f16 v6, v5, v8 ; GFX10-NEXT: v_mov_b32_e32 v5, v6 +; GFX10-NEXT: v_mov_b32_e32 v6, v7 ; GFX10-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX10-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX10-NEXT: v_readfirstlane_b32 s8, v0 @@ -7018,15 +6937,15 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX10-NEXT: s_and_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_and_saveexec_b32 s4, s4 ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_xor_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB18_4 ; GFX10-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX10-NEXT: s_mov_b32 exec_lo, s6 ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX10-NEXT: v_mov_b32_e32 v6, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX10-NEXT: v_mov_b32_e32 v7, v5 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 @@ -7035,13 +6954,12 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX10-NEXT: s_cbranch_execnz .LBB18_3 ; GFX10-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s5 -; GFX10-NEXT: v_mov_b32_e32 v0, v4 +; GFX10-NEXT: v_mov_b32_e32 v0, v5 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_add_u32_e32 v8, 0x400, v4 ; GFX90A-NEXT: s_mov_b64 s[6:7], exec ; GFX90A-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: v_readfirstlane_b32 s8, v0 @@ -7053,22 +6971,21 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX90A-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX90A-NEXT: s_nop 0 -; GFX90A-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 -; GFX90A-NEXT: ; implicit-def: $vgpr4 +; GFX90A-NEXT: buffer_load_dword v9, v4, s[8:11], 0 offen offset:1024 ; GFX90A-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB18_1 ; GFX90A-NEXT: ; %bb.2: ; GFX90A-NEXT: s_mov_b64 exec, s[6:7] ; GFX90A-NEXT: s_mov_b64 s[6:7], 0 -; GFX90A-NEXT: v_pk_max_f16 v9, v5, v5 +; GFX90A-NEXT: v_pk_max_f16 v5, v5, v5 ; GFX90A-NEXT: .LBB18_3: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Loop Header: Depth=1 ; GFX90A-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_pk_max_f16 v4, v7, v7 -; GFX90A-NEXT: v_pk_max_f16 v6, v4, v9 +; GFX90A-NEXT: v_pk_max_f16 v6, v9, v9 +; GFX90A-NEXT: v_pk_max_f16 v8, v6, v5 ; GFX90A-NEXT: s_mov_b64 s[12:13], exec -; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[6:7], v[6:7] op_sel:[0,1] +; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[8:9], v[8:9] op_sel:[0,1] ; GFX90A-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX90A-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX90A-NEXT: v_readfirstlane_b32 s8, v0 @@ -7080,27 +6997,26 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX90A-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: buffer_atomic_cmpswap v[4:5], v8, s[8:11], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[8:11], 0 offen offset:1024 glc ; GFX90A-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB18_4 ; GFX90A-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX90A-NEXT: s_mov_b64 exec, s[12:13] ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v7 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v6, v9 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v7, v4 +; GFX90A-NEXT: v_mov_b32_e32 v9, v6 ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB18_3 ; GFX90A-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX90A-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v0, v4 +; GFX90A-NEXT: v_mov_b32_e32 v0, v6 ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX908: ; %bb.0: ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX908-NEXT: v_add_u32_e32 v7, 0x400, v4 ; GFX908-NEXT: s_mov_b64 s[6:7], exec ; GFX908-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: v_readfirstlane_b32 s8, v0 @@ -7112,8 +7028,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX908-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_nop 0 -; GFX908-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX908-NEXT: ; implicit-def: $vgpr4 +; GFX908-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB18_1 ; GFX908-NEXT: ; %bb.2: @@ -7124,11 +7039,11 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX908-NEXT: ; =>This Loop Header: Depth=1 ; GFX908-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: v_pk_max_f16 v4, v6, v6 -; GFX908-NEXT: v_pk_max_f16 v5, v4, v8 -; GFX908-NEXT: v_mov_b32_e32 v4, v5 -; GFX908-NEXT: s_mov_b64 s[12:13], exec +; GFX908-NEXT: v_pk_max_f16 v5, v7, v7 +; GFX908-NEXT: v_pk_max_f16 v6, v5, v8 ; GFX908-NEXT: v_mov_b32_e32 v5, v6 +; GFX908-NEXT: s_mov_b64 s[12:13], exec +; GFX908-NEXT: v_mov_b32_e32 v6, v7 ; GFX908-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX908-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX908-NEXT: v_readfirstlane_b32 s8, v0 @@ -7140,27 +7055,26 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX908-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB18_4 ; GFX908-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX908-NEXT: s_mov_b64 exec, s[12:13] ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v6 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v5, v7 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v6, v4 +; GFX908-NEXT: v_mov_b32_e32 v7, v5 ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB18_3 ; GFX908-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX908-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v0, v4 +; GFX908-NEXT: v_mov_b32_e32 v0, v5 ; GFX908-NEXT: s_setpc_b64 s[30:31] ; ; GFX8-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_u32_e32 v7, vcc, 0x400, v4 ; GFX8-NEXT: s_mov_b64 s[6:7], exec ; GFX8-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: v_readfirstlane_b32 s8, v0 @@ -7172,8 +7086,7 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX8-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_nop 0 -; GFX8-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX8-NEXT: ; implicit-def: $vgpr4 +; GFX8-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB18_1 ; GFX8-NEXT: ; %bb.2: @@ -7185,14 +7098,14 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX8-NEXT: ; =>This Loop Header: Depth=1 ; GFX8-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: v_max_f16_sdwa v4, v6, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_max_f16_e32 v5, v6, v6 -; GFX8-NEXT: v_max_f16_sdwa v4, v4, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_max_f16_e32 v5, v5, v9 -; GFX8-NEXT: v_or_b32_e32 v5, v5, v4 -; GFX8-NEXT: v_mov_b32_e32 v4, v5 -; GFX8-NEXT: s_mov_b64 s[12:13], exec +; GFX8-NEXT: v_max_f16_sdwa v5, v7, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_max_f16_e32 v6, v7, v7 +; GFX8-NEXT: v_max_f16_sdwa v5, v5, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_max_f16_e32 v6, v6, v9 +; GFX8-NEXT: v_or_b32_e32 v6, v6, v5 ; GFX8-NEXT: v_mov_b32_e32 v5, v6 +; GFX8-NEXT: s_mov_b64 s[12:13], exec +; GFX8-NEXT: v_mov_b32_e32 v6, v7 ; GFX8-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX8-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX8-NEXT: v_readfirstlane_b32 s8, v0 @@ -7204,27 +7117,26 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX8-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB18_4 ; GFX8-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX8-NEXT: s_mov_b64 exec, s[12:13] ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v6 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v5, v7 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v7, v5 ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB18_3 ; GFX8-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX8-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v0, v4 +; GFX8-NEXT: v_mov_b32_e32 v0, v5 ; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v9, vcc, 0x400, v4 ; GFX7-NEXT: s_mov_b64 s[6:7], exec ; GFX7-NEXT: .LBB18_1: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: v_readfirstlane_b32 s8, v0 @@ -7235,39 +7147,38 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX7-NEXT: v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3] ; GFX7-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX7-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] -; GFX7-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 -; GFX7-NEXT: ; implicit-def: $vgpr4 +; GFX7-NEXT: buffer_load_dword v8, v4, s[8:11], 0 offen offset:1024 ; GFX7-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB18_1 ; GFX7-NEXT: ; %bb.2: ; GFX7-NEXT: s_mov_b64 exec, s[6:7] ; GFX7-NEXT: v_cvt_f16_f32_e32 v6, v6 -; GFX7-NEXT: v_cvt_f16_f32_e32 v8, v5 +; GFX7-NEXT: v_cvt_f16_f32_e32 v9, v5 ; GFX7-NEXT: s_waitcnt vmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v5, 16, v7 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v7 +; GFX7-NEXT: v_lshrrev_b32_e32 v5, 16, v8 +; GFX7-NEXT: v_cvt_f32_f16_e32 v7, v8 ; GFX7-NEXT: v_cvt_f32_f16_e32 v5, v5 ; GFX7-NEXT: v_cvt_f32_f16_e32 v10, v6 -; GFX7-NEXT: v_cvt_f32_f16_e32 v11, v8 +; GFX7-NEXT: v_cvt_f32_f16_e32 v11, v9 ; GFX7-NEXT: s_mov_b64 s[6:7], 0 ; GFX7-NEXT: .LBB18_3: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Loop Header: Depth=1 ; GFX7-NEXT: ; Child Loop BB18_4 Depth 2 ; GFX7-NEXT: v_cvt_f16_f32_e32 v5, v5 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 +; GFX7-NEXT: v_cvt_f16_f32_e32 v6, v7 ; GFX7-NEXT: s_mov_b64 s[12:13], exec -; GFX7-NEXT: v_cvt_f32_f16_e32 v6, v5 -; GFX7-NEXT: v_cvt_f32_f16_e32 v7, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v7, v5 +; GFX7-NEXT: v_cvt_f32_f16_e32 v8, v6 ; GFX7-NEXT: v_lshlrev_b32_e32 v5, 16, v5 -; GFX7-NEXT: v_max_f32_e32 v6, v6, v10 -; GFX7-NEXT: v_max_f32_e32 v7, v7, v11 -; GFX7-NEXT: v_cvt_f16_f32_e32 v8, v6 +; GFX7-NEXT: v_or_b32_e32 v6, v6, v5 +; GFX7-NEXT: v_max_f32_e32 v7, v7, v10 +; GFX7-NEXT: v_max_f32_e32 v8, v8, v11 ; GFX7-NEXT: v_cvt_f16_f32_e32 v7, v7 -; GFX7-NEXT: v_or_b32_e32 v6, v4, v5 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v8 -; GFX7-NEXT: v_or_b32_e32 v5, v7, v4 -; GFX7-NEXT: v_mov_b32_e32 v8, v6 -; GFX7-NEXT: v_mov_b32_e32 v7, v5 +; GFX7-NEXT: v_cvt_f16_f32_e32 v8, v8 +; GFX7-NEXT: v_lshlrev_b32_e32 v5, 16, v7 +; GFX7-NEXT: v_or_b32_e32 v5, v8, v5 +; GFX7-NEXT: v_mov_b32_e32 v9, v6 +; GFX7-NEXT: v_mov_b32_e32 v8, v5 ; GFX7-NEXT: .LBB18_4: ; Parent Loop BB18_3 Depth=1 ; GFX7-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX7-NEXT: v_readfirstlane_b32 s8, v0 @@ -7279,23 +7190,23 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fmax_ret_v2f16__offset__waterfall ; GFX7-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX7-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX7-NEXT: s_waitcnt vmcnt(0) -; GFX7-NEXT: buffer_atomic_cmpswap v[7:8], v9, s[8:11], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap v[8:9], v4, s[8:11], 0 offen offset:1024 glc ; GFX7-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB18_4 ; GFX7-NEXT: ; %bb.5: ; in Loop: Header=BB18_3 Depth=1 ; GFX7-NEXT: s_mov_b64 exec, s[12:13] ; GFX7-NEXT: s_waitcnt vmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v5, 16, v7 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v7 +; GFX7-NEXT: v_lshrrev_b32_e32 v5, 16, v8 +; GFX7-NEXT: v_cvt_f32_f16_e32 v7, v8 ; GFX7-NEXT: v_cvt_f32_f16_e32 v5, v5 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v7, v6 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v8, v6 ; GFX7-NEXT: s_or_b64 s[6:7], vcc, s[6:7] ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX7-NEXT: s_cbranch_execnz .LBB18_3 ; GFX7-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX7-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX7-NEXT: v_mov_b32_e32 v0, v4 +; GFX7-NEXT: v_mov_b32_e32 v0, v7 ; GFX7-NEXT: v_mov_b32_e32 v1, v5 ; GFX7-NEXT: s_setpc_b64 s[30:31] ; @@ -7396,13 +7307,11 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 ; GFX12-TRUE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s16 -; GFX12-TRUE16-NEXT: s_add_co_i32 s4, s16, 0x400 -; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe +; GFX12-TRUE16-NEXT: s_mov_b32 s4, 0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_lshlrev_b32 v3, 16, v1 +; GFX12-TRUE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_lshlrev_b32 v3, 16, v1 ; GFX12-TRUE16-NEXT: buffer_load_b32 v0, v0, s[0:3], null offen offset:1024 ; GFX12-TRUE16-NEXT: v_and_b32_e32 v2, 0xffff0000, v1 -; GFX12-TRUE16-NEXT: s_mov_b32 s4, 0 ; GFX12-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 @@ -7431,7 +7340,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v0.h ; GFX12-TRUE16-NEXT: v_dual_mov_b32 v0, v5 :: v_dual_mov_b32 v1, v6 -; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_DEV ; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 @@ -7452,11 +7361,9 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0 ; GFX12-FAKE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s16 -; GFX12-FAKE16-NEXT: s_add_co_i32 s4, s16, 0x400 ; GFX12-FAKE16-NEXT: s_mov_b32 s5, 0 -; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_and_b32 v3, 0xffff0000, v1 +; GFX12-FAKE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_and_b32 v3, 0xffff0000, v1 ; GFX12-FAKE16-NEXT: buffer_load_b32 v0, v0, s[0:3], null offen offset:1024 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 ; GFX12-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start @@ -7487,7 +7394,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX12-FAKE16-NEXT: v_perm_b32 v5, v1, v0, 0x7060302 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-FAKE16-NEXT: v_dual_mov_b32 v0, v5 :: v_dual_mov_b32 v1, v6 -; GFX12-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_DEV ; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 @@ -7506,13 +7413,12 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX942-NEXT: v_mov_b32_e32 v1, v0 ; GFX942-NEXT: v_mov_b32_e32 v0, s16 ; GFX942-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s4, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[6:7], 0 ; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1 ; GFX942-NEXT: s_movk_i32 s8, 0x7fff ; GFX942-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 ; GFX942-NEXT: s_mov_b32 s9, 0x7060302 -; GFX942-NEXT: v_mov_b32_e32 v4, s4 +; GFX942-NEXT: v_mov_b32_e32 v4, s16 ; GFX942-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -7534,7 +7440,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX942-NEXT: v_cndmask_b32_e64 v0, v5, v6, s[4:5] ; GFX942-NEXT: v_perm_b32 v6, v1, v0, s9 ; GFX942-NEXT: v_mov_b64_e32 v[0:1], v[6:7] -; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v0, v7 @@ -7549,12 +7455,11 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-TRUE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s16 -; GFX11-TRUE16-NEXT: s_add_i32 s4, s16, 0x400 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) -; GFX11-TRUE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_lshlrev_b32 v3, 16, v1 +; GFX11-TRUE16-NEXT: s_mov_b32 s4, 0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_lshlrev_b32 v3, 16, v1 ; GFX11-TRUE16-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024 ; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0xffff0000, v1 -; GFX11-TRUE16-NEXT: s_mov_b32 s4, 0 ; GFX11-TRUE16-NEXT: s_set_inst_prefetch_distance 0x1 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start @@ -7583,7 +7488,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v0.h ; GFX11-TRUE16-NEXT: v_dual_mov_b32 v0, v5 :: v_dual_mov_b32 v1, v6 -; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], 0 offen glc +; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], 0 offen offset:1024 glc ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl1_inv ; GFX11-TRUE16-NEXT: buffer_gl0_inv @@ -7601,10 +7506,9 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-FAKE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s16 -; GFX11-FAKE16-NEXT: s_add_i32 s4, s16, 0x400 ; GFX11-FAKE16-NEXT: s_mov_b32 s5, 0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_and_b32 v3, 0xffff0000, v1 +; GFX11-FAKE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_and_b32 v3, 0xffff0000, v1 ; GFX11-FAKE16-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen offset:1024 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 ; GFX11-FAKE16-NEXT: s_set_inst_prefetch_distance 0x1 @@ -7635,7 +7539,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX11-FAKE16-NEXT: v_perm_b32 v5, v1, v0, 0x7060302 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-FAKE16-NEXT: v_dual_mov_b32 v0, v5 :: v_dual_mov_b32 v1, v6 -; GFX11-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], 0 offen glc +; GFX11-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[0:1], v4, s[0:3], 0 offen offset:1024 glc ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl1_inv ; GFX11-FAKE16-NEXT: buffer_gl0_inv @@ -7654,9 +7558,8 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mov_b32_e32 v1, v0 ; GFX10-NEXT: v_mov_b32_e32 v0, s20 -; GFX10-NEXT: s_add_i32 s4, s20, 0x400 +; GFX10-NEXT: v_mov_b32_e32 v4, s20 ; GFX10-NEXT: s_mov_b32 s5, 0 -; GFX10-NEXT: v_mov_b32_e32 v4, s4 ; GFX10-NEXT: v_lshlrev_b32_e32 v2, 16, v1 ; GFX10-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 ; GFX10-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 @@ -7682,7 +7585,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX10-NEXT: v_perm_b32 v5, v1, v0, 0x7060302 ; GFX10-NEXT: v_mov_b32_e32 v0, v5 ; GFX10-NEXT: v_mov_b32_e32 v1, v6 -; GFX10-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen offset:1024 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv @@ -7700,13 +7603,12 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX90A-NEXT: v_mov_b32_e32 v1, v0 ; GFX90A-NEXT: v_mov_b32_e32 v0, s20 ; GFX90A-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s4, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[6:7], 0 ; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1 ; GFX90A-NEXT: s_movk_i32 s8, 0x7fff ; GFX90A-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 ; GFX90A-NEXT: s_mov_b32 s9, 0x7060302 -; GFX90A-NEXT: v_mov_b32_e32 v4, s4 +; GFX90A-NEXT: v_mov_b32_e32 v4, s20 ; GFX90A-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -7727,7 +7629,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX90A-NEXT: v_cndmask_b32_e32 v1, v8, v9, vcc ; GFX90A-NEXT: v_perm_b32 v6, v1, v0, s9 ; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[6:7], v[6:7] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v0, v7 @@ -7744,13 +7646,12 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX908-NEXT: v_mov_b32_e32 v1, v0 ; GFX908-NEXT: v_mov_b32_e32 v0, s20 ; GFX908-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s4, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[6:7], 0 ; GFX908-NEXT: v_lshlrev_b32_e32 v2, 16, v1 ; GFX908-NEXT: s_movk_i32 s8, 0x7fff ; GFX908-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 ; GFX908-NEXT: s_mov_b32 s9, 0x7060302 -; GFX908-NEXT: v_mov_b32_e32 v4, s4 +; GFX908-NEXT: v_mov_b32_e32 v4, s20 ; GFX908-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -7772,7 +7673,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX908-NEXT: v_perm_b32 v5, v1, v0, s9 ; GFX908-NEXT: v_mov_b32_e32 v0, v5 ; GFX908-NEXT: v_mov_b32_e32 v1, v6 -; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v0, v6 @@ -7789,11 +7690,10 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX8-NEXT: v_mov_b32_e32 v1, v0 ; GFX8-NEXT: v_mov_b32_e32 v0, s20 ; GFX8-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s4, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[6:7], 0 ; GFX8-NEXT: v_lshlrev_b32_e32 v2, 16, v1 ; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_mov_b32_e32 v4, s4 +; GFX8-NEXT: v_mov_b32_e32 v4, s20 ; GFX8-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -7818,7 +7718,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX8-NEXT: v_alignbit_b32 v5, v1, v0, 16 ; GFX8-NEXT: v_mov_b32_e32 v0, v5 ; GFX8-NEXT: v_mov_b32_e32 v1, v6 -; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v4, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v0, v6 @@ -7834,7 +7734,6 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_mov_b32_e32 v2, s20 ; GFX7-NEXT: buffer_load_dword v4, v2, s[16:19], 0 offen offset:1024 -; GFX7-NEXT: s_add_i32 s6, s20, 0x400 ; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v1 ; GFX7-NEXT: v_mul_f32_e32 v0, 1.0, v0 ; GFX7-NEXT: s_mov_b64 s[4:5], 0 @@ -7843,7 +7742,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: v_and_b32_e32 v1, 0xffff0000, v4 ; GFX7-NEXT: v_lshlrev_b32_e32 v0, 16, v4 -; GFX7-NEXT: v_mov_b32_e32 v4, s6 +; GFX7-NEXT: v_mov_b32_e32 v4, s20 ; GFX7-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v1 @@ -7858,7 +7757,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__amdgpu ; GFX7-NEXT: v_alignbit_b32 v0, v0, v6, 16 ; GFX7-NEXT: v_mov_b32_e32 v6, v1 ; GFX7-NEXT: v_mov_b32_e32 v5, v0 -; GFX7-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen offset:1024 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v5, v1 @@ -7928,11 +7827,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 ; GFX12-TRUE16-NEXT: v_dual_mov_b32 v1, s16 :: v_dual_and_b32 v2, 0xffff0000, v0 -; GFX12-TRUE16-NEXT: s_add_co_i32 s4, s16, 0x400 -; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe -; GFX12-TRUE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_lshlrev_b32 v3, 16, v0 -; GFX12-TRUE16-NEXT: buffer_load_b32 v1, v1, s[0:3], null offen offset:1024 +; GFX12-TRUE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_lshlrev_b32 v3, 16, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s4, 0 +; GFX12-TRUE16-NEXT: buffer_load_b32 v1, v1, s[0:3], null offen offset:1024 ; GFX12-TRUE16-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 @@ -7958,7 +7855,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v0.l, v6.h ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-TRUE16-NEXT: v_dual_mov_b32 v6, v1 :: v_dual_mov_b32 v5, v0 -; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_DEV ; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v1 @@ -7980,11 +7877,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0 ; GFX12-FAKE16-NEXT: v_dual_mov_b32 v1, s16 :: v_dual_lshlrev_b32 v2, 16, v0 -; GFX12-FAKE16-NEXT: s_add_co_i32 s4, s16, 0x400 -; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe -; GFX12-FAKE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_and_b32 v3, 0xffff0000, v0 -; GFX12-FAKE16-NEXT: buffer_load_b32 v1, v1, s[0:3], null offen offset:1024 +; GFX12-FAKE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_and_b32 v3, 0xffff0000, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s5, 0 +; GFX12-FAKE16-NEXT: buffer_load_b32 v1, v1, s[0:3], null offen offset:1024 ; GFX12-FAKE16-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 @@ -8010,7 +7905,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-FAKE16-NEXT: v_perm_b32 v0, v5, v0, 0x7060302 ; GFX12-FAKE16-NEXT: v_dual_mov_b32 v6, v1 :: v_dual_mov_b32 v5, v0 -; GFX12-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_DEV ; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v1 @@ -8029,13 +7924,12 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX942-NEXT: v_mov_b32_e32 v1, s16 ; GFX942-NEXT: buffer_load_dword v1, v1, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s4, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[6:7], 0 ; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v0 ; GFX942-NEXT: s_movk_i32 s8, 0x7fff ; GFX942-NEXT: v_and_b32_e32 v3, 0xffff0000, v0 ; GFX942-NEXT: s_mov_b32 s9, 0x7060302 -; GFX942-NEXT: v_mov_b32_e32 v4, s4 +; GFX942-NEXT: v_mov_b32_e32 v4, s16 ; GFX942-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -8056,7 +7950,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX942-NEXT: v_cndmask_b32_e64 v0, v6, v7, s[4:5] ; GFX942-NEXT: v_perm_b32 v0, v5, v0, s9 ; GFX942-NEXT: v_mov_b64_e32 v[6:7], v[0:1] -; GFX942-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v6, v1 @@ -8072,11 +7966,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-TRUE16-NEXT: v_dual_mov_b32 v1, s16 :: v_dual_and_b32 v2, 0xffff0000, v0 -; GFX11-TRUE16-NEXT: s_add_i32 s4, s16, 0x400 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-TRUE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_lshlrev_b32 v3, 16, v0 -; GFX11-TRUE16-NEXT: buffer_load_b32 v1, v1, s[0:3], 0 offen offset:1024 +; GFX11-TRUE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_lshlrev_b32 v3, 16, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s4, 0 +; GFX11-TRUE16-NEXT: buffer_load_b32 v1, v1, s[0:3], 0 offen offset:1024 ; GFX11-TRUE16-NEXT: s_set_inst_prefetch_distance 0x1 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB20_1: ; %atomicrmw.start @@ -8102,7 +7994,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, v6.h ; GFX11-TRUE16-NEXT: v_dual_mov_b32 v6, v1 :: v_dual_mov_b32 v5, v0 -; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], 0 offen glc +; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], 0 offen offset:1024 glc ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl1_inv ; GFX11-TRUE16-NEXT: buffer_gl0_inv @@ -8121,11 +8013,9 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-FAKE16-NEXT: v_dual_mov_b32 v1, s16 :: v_dual_lshlrev_b32 v2, 16, v0 -; GFX11-FAKE16-NEXT: s_add_i32 s4, s16, 0x400 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-FAKE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_and_b32 v3, 0xffff0000, v0 -; GFX11-FAKE16-NEXT: buffer_load_b32 v1, v1, s[0:3], 0 offen offset:1024 +; GFX11-FAKE16-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_and_b32 v3, 0xffff0000, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s5, 0 +; GFX11-FAKE16-NEXT: buffer_load_b32 v1, v1, s[0:3], 0 offen offset:1024 ; GFX11-FAKE16-NEXT: s_set_inst_prefetch_distance 0x1 ; GFX11-FAKE16-NEXT: .p2align 6 ; GFX11-FAKE16-NEXT: .LBB20_1: ; %atomicrmw.start @@ -8151,7 +8041,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-FAKE16-NEXT: v_perm_b32 v0, v5, v0, 0x7060302 ; GFX11-FAKE16-NEXT: v_dual_mov_b32 v6, v1 :: v_dual_mov_b32 v5, v0 -; GFX11-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], 0 offen glc +; GFX11-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[0:3], 0 offen offset:1024 glc ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl1_inv ; GFX11-FAKE16-NEXT: buffer_gl0_inv @@ -8170,12 +8060,11 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_mov_b32_e32 v1, s20 -; GFX10-NEXT: s_add_i32 s4, s20, 0x400 ; GFX10-NEXT: v_lshlrev_b32_e32 v2, 16, v0 ; GFX10-NEXT: v_and_b32_e32 v3, 0xffff0000, v0 -; GFX10-NEXT: v_mov_b32_e32 v4, s4 -; GFX10-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 +; GFX10-NEXT: v_mov_b32_e32 v4, s20 ; GFX10-NEXT: s_mov_b32 s5, 0 +; GFX10-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 ; GFX10-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt vmcnt(0) @@ -8197,7 +8086,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX10-NEXT: v_perm_b32 v0, v5, v0, 0x7060302 ; GFX10-NEXT: v_mov_b32_e32 v6, v1 ; GFX10-NEXT: v_mov_b32_e32 v5, v0 -; GFX10-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen offset:1024 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv @@ -8215,13 +8104,12 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_mov_b32_e32 v1, s20 ; GFX90A-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s4, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[6:7], 0 ; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v0 ; GFX90A-NEXT: s_movk_i32 s8, 0x7fff ; GFX90A-NEXT: v_and_b32_e32 v3, 0xffff0000, v0 ; GFX90A-NEXT: s_mov_b32 s9, 0x7060302 -; GFX90A-NEXT: v_mov_b32_e32 v4, s4 +; GFX90A-NEXT: v_mov_b32_e32 v4, s20 ; GFX90A-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -8241,7 +8129,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc ; GFX90A-NEXT: v_perm_b32 v0, v5, v0, s9 ; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[0:1], v[0:1] op_sel:[0,1] -; GFX90A-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v6, v1 @@ -8258,13 +8146,12 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX908-NEXT: v_mov_b32_e32 v1, s20 ; GFX908-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s4, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[6:7], 0 ; GFX908-NEXT: v_lshlrev_b32_e32 v2, 16, v0 ; GFX908-NEXT: s_movk_i32 s8, 0x7fff ; GFX908-NEXT: v_and_b32_e32 v3, 0xffff0000, v0 ; GFX908-NEXT: s_mov_b32 s9, 0x7060302 -; GFX908-NEXT: v_mov_b32_e32 v4, s4 +; GFX908-NEXT: v_mov_b32_e32 v4, s20 ; GFX908-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -8285,7 +8172,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX908-NEXT: v_perm_b32 v0, v5, v0, s9 ; GFX908-NEXT: v_mov_b32_e32 v6, v1 ; GFX908-NEXT: v_mov_b32_e32 v5, v0 -; GFX908-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v5, v1 @@ -8302,11 +8189,10 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-NEXT: v_mov_b32_e32 v1, s20 ; GFX8-NEXT: buffer_load_dword v1, v1, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s4, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[6:7], 0 ; GFX8-NEXT: v_lshlrev_b32_e32 v2, 16, v0 ; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v0 -; GFX8-NEXT: v_mov_b32_e32 v4, s4 +; GFX8-NEXT: v_mov_b32_e32 v4, s20 ; GFX8-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -8330,7 +8216,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX8-NEXT: v_alignbit_b32 v0, v5, v0, 16 ; GFX8-NEXT: v_mov_b32_e32 v6, v1 ; GFX8-NEXT: v_mov_b32_e32 v5, v0 -; GFX8-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v5, v1 @@ -8347,7 +8233,6 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_mov_b32_e32 v2, s20 ; GFX7-NEXT: buffer_load_dword v2, v2, s[16:19], 0 offen offset:1024 -; GFX7-NEXT: s_add_i32 s6, s20, 0x400 ; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v1 ; GFX7-NEXT: v_mul_f32_e32 v3, 1.0, v0 ; GFX7-NEXT: s_mov_b64 s[4:5], 0 @@ -8356,7 +8241,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: v_and_b32_e32 v4, 0xffff0000, v2 ; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v2 -; GFX7-NEXT: v_mov_b32_e32 v2, s6 +; GFX7-NEXT: v_mov_b32_e32 v2, s20 ; GFX7-NEXT: .LBB20_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v4 @@ -8371,7 +8256,7 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_v2bf16__offset__amdgpu_no_fi ; GFX7-NEXT: v_alignbit_b32 v3, v3, v6, 16 ; GFX7-NEXT: v_mov_b32_e32 v6, v4 ; GFX7-NEXT: v_mov_b32_e32 v5, v3 -; GFX7-NEXT: buffer_atomic_cmpswap v[5:6], v2, s[16:19], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap v[5:6], v2, s[16:19], 0 offen offset:1024 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v5, v4 @@ -8440,7 +8325,6 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 -; GFX12-TRUE16-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX12-TRUE16-NEXT: s_mov_b32 s1, exec_lo ; GFX12-TRUE16-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -8455,8 +8339,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: buffer_load_b32 v6, v4, s[4:7], null offen offset:1024 -; GFX12-TRUE16-NEXT: ; implicit-def: $vgpr4 +; GFX12-TRUE16-NEXT: buffer_load_b32 v7, v4, s[4:7], null offen offset:1024 ; GFX12-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-TRUE16-NEXT: s_cbranch_execnz .LBB21_1 ; GFX12-TRUE16-NEXT: ; %bb.2: @@ -8468,30 +8351,30 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-TRUE16-NEXT: ; =>This Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v6 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v6, 0xffff0000, v7 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v7 ; GFX12-TRUE16-NEXT: s_mov_b32 s2, exec_lo ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_max_num_f32 v5, v5, v8 :: v_dual_max_num_f32 v4, v4, v9 -; GFX12-TRUE16-NEXT: v_bfe_u32 v11, v5, 16, 1 +; GFX12-TRUE16-NEXT: v_dual_max_num_f32 v6, v6, v8 :: v_dual_max_num_f32 v5, v5, v9 +; GFX12-TRUE16-NEXT: v_bfe_u32 v11, v6, 16, 1 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_bfe_u32 v10, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v12, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v13, 0x400000, v5 -; GFX12-TRUE16-NEXT: v_add3_u32 v11, v11, v5, 0x7fff -; GFX12-TRUE16-NEXT: v_add3_u32 v10, v10, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_bfe_u32 v10, v5, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v12, 0x400000, v5 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v13, 0x400000, v6 +; GFX12-TRUE16-NEXT: v_add3_u32 v11, v11, v6, 0x7fff +; GFX12-TRUE16-NEXT: v_add3_u32 v10, v10, v5, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v10, v12, vcc_lo -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v10, v12, vcc_lo +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v6, v6 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v11, v13, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v6, v11, v13, vcc_lo +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v6.l, v5.h ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v5 ; GFX12-TRUE16-NEXT: v_mov_b32_e32 v5, v6 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v6, v7 ; GFX12-TRUE16-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX12-TRUE16-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -8506,14 +8389,14 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], null offen th:TH_ATOMIC_RETURN +; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[4:7], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-TRUE16-NEXT: s_cbranch_execnz .LBB21_4 ; GFX12-TRUE16-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX12-TRUE16-NEXT: s_mov_b32 exec_lo, s2 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v6, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v7, v5 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_DEV ; GFX12-TRUE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -8521,7 +8404,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-TRUE16-NEXT: s_cbranch_execnz .LBB21_3 ; GFX12-TRUE16-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX12-TRUE16-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, v4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v0, v5 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] ; @@ -8532,7 +8415,6 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0 ; GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0 -; GFX12-FAKE16-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX12-FAKE16-NEXT: s_mov_b32 s1, exec_lo ; GFX12-FAKE16-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -8547,8 +8429,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-FAKE16-NEXT: buffer_load_b32 v6, v4, s[4:7], null offen offset:1024 -; GFX12-FAKE16-NEXT: ; implicit-def: $vgpr4 +; GFX12-FAKE16-NEXT: buffer_load_b32 v7, v4, s[4:7], null offen offset:1024 ; GFX12-FAKE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-FAKE16-NEXT: s_cbranch_execnz .LBB21_1 ; GFX12-FAKE16-NEXT: ; %bb.2: @@ -8560,30 +8441,30 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-FAKE16-NEXT: ; =>This Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v6 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v6, 0xffff0000, v7 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v5, 16, v7 ; GFX12-FAKE16-NEXT: s_mov_b32 s2, exec_lo ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_max_num_f32 v5, v5, v9 :: v_dual_max_num_f32 v4, v4, v8 -; GFX12-FAKE16-NEXT: v_bfe_u32 v11, v5, 16, 1 +; GFX12-FAKE16-NEXT: v_dual_max_num_f32 v6, v6, v9 :: v_dual_max_num_f32 v5, v5, v8 +; GFX12-FAKE16-NEXT: v_bfe_u32 v11, v6, 16, 1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_bfe_u32 v10, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v12, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v13, 0x400000, v5 -; GFX12-FAKE16-NEXT: v_add3_u32 v11, v11, v5, 0x7fff -; GFX12-FAKE16-NEXT: v_add3_u32 v10, v10, v4, 0x7fff +; GFX12-FAKE16-NEXT: v_bfe_u32 v10, v5, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v12, 0x400000, v5 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v13, 0x400000, v6 +; GFX12-FAKE16-NEXT: v_add3_u32 v11, v11, v6, 0x7fff +; GFX12-FAKE16-NEXT: v_add3_u32 v10, v10, v5, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v10, v12, vcc_lo -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v10, v12, vcc_lo +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v6, v6 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v11, v13, vcc_lo -; GFX12-FAKE16-NEXT: v_perm_b32 v5, v5, v4, 0x7060302 +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v6, v11, v13, vcc_lo +; GFX12-FAKE16-NEXT: v_perm_b32 v6, v6, v5, 0x7060302 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v5 ; GFX12-FAKE16-NEXT: v_mov_b32_e32 v5, v6 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v6, v7 ; GFX12-FAKE16-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX12-FAKE16-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX12-FAKE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -8598,14 +8479,14 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], null offen th:TH_ATOMIC_RETURN +; GFX12-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[4:7], null offen offset:1024 th:TH_ATOMIC_RETURN ; GFX12-FAKE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-FAKE16-NEXT: s_cbranch_execnz .LBB21_4 ; GFX12-FAKE16-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX12-FAKE16-NEXT: s_mov_b32 exec_lo, s2 ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v6, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v7, v5 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_DEV ; GFX12-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -8613,14 +8494,13 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX12-FAKE16-NEXT: s_cbranch_execnz .LBB21_3 ; GFX12-FAKE16-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX12-FAKE16-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, v5 ; GFX12-FAKE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-FAKE16-NEXT: s_setpc_b64 s[30:31] ; ; GFX942-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX942: ; %bb.0: ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX942-NEXT: v_add_u32_e32 v8, 0x400, v4 ; GFX942-NEXT: s_mov_b64 s[2:3], exec ; GFX942-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: v_readfirstlane_b32 s4, v0 @@ -8632,40 +8512,39 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX942-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], v[2:3] ; GFX942-NEXT: s_and_b64 s[0:1], vcc, s[0:1] ; GFX942-NEXT: s_and_saveexec_b64 s[0:1], s[0:1] -; GFX942-NEXT: buffer_load_dword v7, v4, s[4:7], 0 offen offset:1024 -; GFX942-NEXT: ; implicit-def: $vgpr4 +; GFX942-NEXT: buffer_load_dword v9, v4, s[4:7], 0 offen offset:1024 ; GFX942-NEXT: s_xor_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB21_1 ; GFX942-NEXT: ; %bb.2: ; GFX942-NEXT: s_mov_b64 exec, s[2:3] ; GFX942-NEXT: s_mov_b64 s[2:3], 0 -; GFX942-NEXT: v_lshlrev_b32_e32 v9, 16, v5 +; GFX942-NEXT: v_lshlrev_b32_e32 v10, 16, v5 ; GFX942-NEXT: s_movk_i32 s10, 0x7fff -; GFX942-NEXT: v_and_b32_e32 v10, 0xffff0000, v5 +; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v5 ; GFX942-NEXT: s_mov_b32 s11, 0x7060302 ; GFX942-NEXT: .LBB21_3: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Loop Header: Depth=1 ; GFX942-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v4, 16, v7 -; GFX942-NEXT: v_max_f32_e32 v4, v4, v9 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s10 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_lshlrev_b32_e32 v6, 16, v9 +; GFX942-NEXT: v_max_f32_e32 v6, v6, v10 +; GFX942-NEXT: v_bfe_u32 v7, v6, 16, 1 +; GFX942-NEXT: v_add3_u32 v7, v7, v6, s10 +; GFX942-NEXT: v_or_b32_e32 v8, 0x400000, v6 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v6, v6 ; GFX942-NEXT: s_mov_b64 s[8:9], exec ; GFX942-NEXT: buffer_wbl2 sc1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v7 -; GFX942-NEXT: v_max_f32_e32 v5, v5, v10 -; GFX942-NEXT: v_bfe_u32 v6, v5, 16, 1 -; GFX942-NEXT: v_add3_u32 v6, v6, v5, s10 -; GFX942-NEXT: v_or_b32_e32 v11, 0x400000, v5 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 +; GFX942-NEXT: v_cndmask_b32_e32 v6, v7, v8, vcc +; GFX942-NEXT: v_and_b32_e32 v7, 0xffff0000, v9 +; GFX942-NEXT: v_max_f32_e32 v7, v7, v5 +; GFX942-NEXT: v_bfe_u32 v8, v7, 16, 1 +; GFX942-NEXT: v_add3_u32 v8, v8, v7, s10 +; GFX942-NEXT: v_or_b32_e32 v11, 0x400000, v7 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v7, v7 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v5, v6, v11, vcc -; GFX942-NEXT: v_perm_b32 v6, v5, v4, s11 -; GFX942-NEXT: v_mov_b64_e32 v[4:5], v[6:7] +; GFX942-NEXT: v_cndmask_b32_e32 v7, v8, v11, vcc +; GFX942-NEXT: v_perm_b32 v8, v7, v6, s11 +; GFX942-NEXT: v_mov_b64_e32 v[6:7], v[8:9] ; GFX942-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX942-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX942-NEXT: v_readfirstlane_b32 s4, v0 @@ -8678,27 +8557,26 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX942-NEXT: s_and_b64 s[0:1], vcc, s[0:1] ; GFX942-NEXT: s_and_saveexec_b64 s[0:1], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: buffer_atomic_cmpswap v[4:5], v8, s[4:7], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[4:7], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_xor_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB21_4 ; GFX942-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX942-NEXT: s_mov_b64 exec, s[8:9] ; GFX942-NEXT: s_waitcnt vmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v7 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v6, v9 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v7, v4 +; GFX942-NEXT: v_mov_b32_e32 v9, v6 ; GFX942-NEXT: buffer_inv sc1 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB21_3 ; GFX942-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX942-NEXT: s_or_b64 exec, exec, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v0, v4 +; GFX942-NEXT: v_mov_b32_e32 v0, v6 ; GFX942-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-TRUE16-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX11-TRUE16-NEXT: s_mov_b32 s1, 0 ; GFX11-TRUE16-NEXT: s_mov_b32 s2, exec_lo ; GFX11-TRUE16-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 @@ -8712,8 +8590,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-TRUE16-NEXT: s_and_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 -; GFX11-TRUE16-NEXT: buffer_load_b32 v6, v4, s[4:7], 0 offen offset:1024 -; GFX11-TRUE16-NEXT: ; implicit-def: $vgpr4 +; GFX11-TRUE16-NEXT: buffer_load_b32 v7, v4, s[4:7], 0 offen offset:1024 ; GFX11-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-TRUE16-NEXT: s_cbranch_execnz .LBB21_1 ; GFX11-TRUE16-NEXT: ; %bb.2: @@ -8726,28 +8603,28 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-TRUE16-NEXT: ; =>This Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v6 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v6, 0xffff0000, v7 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v7 ; GFX11-TRUE16-NEXT: s_mov_b32 s2, exec_lo ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_max_f32 v5, v5, v8 :: v_dual_max_f32 v4, v4, v9 -; GFX11-TRUE16-NEXT: v_bfe_u32 v11, v5, 16, 1 +; GFX11-TRUE16-NEXT: v_dual_max_f32 v6, v6, v8 :: v_dual_max_f32 v5, v5, v9 +; GFX11-TRUE16-NEXT: v_bfe_u32 v11, v6, 16, 1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_bfe_u32 v10, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v12, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v13, 0x400000, v5 -; GFX11-TRUE16-NEXT: v_add3_u32 v11, v11, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v10, v10, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v10, v12, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v10, v5, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v12, 0x400000, v5 ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v11, v13, vcc_lo +; GFX11-TRUE16-NEXT: v_or_b32_e32 v13, 0x400000, v6 +; GFX11-TRUE16-NEXT: v_add3_u32 v11, v11, v6, 0x7fff +; GFX11-TRUE16-NEXT: v_add3_u32 v10, v10, v5, 0x7fff +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v10, v12, vcc_lo +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v6, v6 +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v6, v11, v13, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v5 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v5.h ; GFX11-TRUE16-NEXT: v_mov_b32_e32 v5, v6 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v6, v7 ; GFX11-TRUE16-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX11-TRUE16-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX11-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -8761,14 +8638,14 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], 0 offen glc +; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[4:7], 0 offen offset:1024 glc ; GFX11-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-TRUE16-NEXT: s_cbranch_execnz .LBB21_4 ; GFX11-TRUE16-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX11-TRUE16-NEXT: s_mov_b32 exec_lo, s2 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v6, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v7, v5 ; GFX11-TRUE16-NEXT: buffer_gl1_inv ; GFX11-TRUE16-NEXT: buffer_gl0_inv ; GFX11-TRUE16-NEXT: s_or_b32 s1, vcc_lo, s1 @@ -8778,13 +8655,12 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-TRUE16-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX11-TRUE16-NEXT: s_set_inst_prefetch_distance 0x2 ; GFX11-TRUE16-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v0, v4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v0, v5 ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-FAKE16-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX11-FAKE16: ; %bb.0: ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX11-FAKE16-NEXT: s_mov_b32 s1, 0 ; GFX11-FAKE16-NEXT: s_mov_b32 s2, exec_lo ; GFX11-FAKE16-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 @@ -8798,8 +8674,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-FAKE16-NEXT: s_and_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_saveexec_b32 s0, s0 -; GFX11-FAKE16-NEXT: buffer_load_b32 v6, v4, s[4:7], 0 offen offset:1024 -; GFX11-FAKE16-NEXT: ; implicit-def: $vgpr4 +; GFX11-FAKE16-NEXT: buffer_load_b32 v7, v4, s[4:7], 0 offen offset:1024 ; GFX11-FAKE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-FAKE16-NEXT: s_cbranch_execnz .LBB21_1 ; GFX11-FAKE16-NEXT: ; %bb.2: @@ -8812,28 +8687,28 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-FAKE16-NEXT: ; =>This Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v6 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v6, 0xffff0000, v7 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v5, 16, v7 ; GFX11-FAKE16-NEXT: s_mov_b32 s2, exec_lo ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_max_f32 v5, v5, v9 :: v_dual_max_f32 v4, v4, v8 -; GFX11-FAKE16-NEXT: v_bfe_u32 v11, v5, 16, 1 +; GFX11-FAKE16-NEXT: v_dual_max_f32 v6, v6, v9 :: v_dual_max_f32 v5, v5, v8 +; GFX11-FAKE16-NEXT: v_bfe_u32 v11, v6, 16, 1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_bfe_u32 v10, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v12, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v13, 0x400000, v5 -; GFX11-FAKE16-NEXT: v_add3_u32 v11, v11, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v10, v10, v4, 0x7fff -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v10, v12, vcc_lo +; GFX11-FAKE16-NEXT: v_bfe_u32 v10, v5, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v12, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v11, v13, vcc_lo +; GFX11-FAKE16-NEXT: v_or_b32_e32 v13, 0x400000, v6 +; GFX11-FAKE16-NEXT: v_add3_u32 v11, v11, v6, 0x7fff +; GFX11-FAKE16-NEXT: v_add3_u32 v10, v10, v5, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v10, v12, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v6, v6 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v6, v11, v13, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v5, v5, v4, 0x7060302 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v5 +; GFX11-FAKE16-NEXT: v_perm_b32 v6, v6, v5, 0x7060302 ; GFX11-FAKE16-NEXT: v_mov_b32_e32 v5, v6 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v6, v7 ; GFX11-FAKE16-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX11-FAKE16-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX11-FAKE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -8847,14 +8722,14 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v7, s[4:7], 0 offen glc +; GFX11-FAKE16-NEXT: buffer_atomic_cmpswap_b32 v[5:6], v4, s[4:7], 0 offen offset:1024 glc ; GFX11-FAKE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-FAKE16-NEXT: s_cbranch_execnz .LBB21_4 ; GFX11-FAKE16-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX11-FAKE16-NEXT: s_mov_b32 exec_lo, s2 ; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v6, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v7, v5 ; GFX11-FAKE16-NEXT: buffer_gl1_inv ; GFX11-FAKE16-NEXT: buffer_gl0_inv ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 @@ -8864,13 +8739,12 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX11-FAKE16-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX11-FAKE16-NEXT: s_set_inst_prefetch_distance 0x2 ; GFX11-FAKE16-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, v5 ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_nc_u32_e32 v7, 0x400, v4 ; GFX10-NEXT: s_mov_b32 s5, 0 ; GFX10-NEXT: s_mov_b32 s6, exec_lo ; GFX10-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 @@ -8882,8 +8756,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX10-NEXT: v_cmp_eq_u64_e64 s4, s[10:11], v[2:3] ; GFX10-NEXT: s_and_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_and_saveexec_b32 s4, s4 -; GFX10-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX10-NEXT: ; implicit-def: $vgpr4 +; GFX10-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_xor_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB21_1 @@ -8895,25 +8768,25 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX10-NEXT: ; =>This Loop Header: Depth=1 ; GFX10-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v6 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 +; GFX10-NEXT: v_lshlrev_b32_e32 v5, 16, v7 +; GFX10-NEXT: v_and_b32_e32 v6, 0xffff0000, v7 ; GFX10-NEXT: s_mov_b32 s6, exec_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: v_max_f32_e32 v4, v4, v8 -; GFX10-NEXT: v_max_f32_e32 v5, v5, v9 -; GFX10-NEXT: v_bfe_u32 v10, v4, 16, 1 -; GFX10-NEXT: v_bfe_u32 v11, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v12, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_or_b32_e32 v13, 0x400000, v5 -; GFX10-NEXT: v_add3_u32 v10, v10, v4, 0x7fff -; GFX10-NEXT: v_add3_u32 v11, v11, v5, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v10, v12, vcc_lo +; GFX10-NEXT: v_max_f32_e32 v5, v5, v8 +; GFX10-NEXT: v_max_f32_e32 v6, v6, v9 +; GFX10-NEXT: v_bfe_u32 v10, v5, 16, 1 +; GFX10-NEXT: v_bfe_u32 v11, v6, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v12, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_cndmask_b32_e32 v5, v11, v13, vcc_lo -; GFX10-NEXT: v_perm_b32 v5, v5, v4, 0x7060302 -; GFX10-NEXT: v_mov_b32_e32 v4, v5 +; GFX10-NEXT: v_or_b32_e32 v13, 0x400000, v6 +; GFX10-NEXT: v_add3_u32 v10, v10, v5, 0x7fff +; GFX10-NEXT: v_add3_u32 v11, v11, v6, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v5, v10, v12, vcc_lo +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v6, v6 +; GFX10-NEXT: v_cndmask_b32_e32 v6, v11, v13, vcc_lo +; GFX10-NEXT: v_perm_b32 v6, v6, v5, 0x7060302 ; GFX10-NEXT: v_mov_b32_e32 v5, v6 +; GFX10-NEXT: v_mov_b32_e32 v6, v7 ; GFX10-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX10-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX10-NEXT: v_readfirstlane_b32 s8, v0 @@ -8925,15 +8798,15 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX10-NEXT: s_and_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_and_saveexec_b32 s4, s4 ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX10-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_xor_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB21_4 ; GFX10-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX10-NEXT: s_mov_b32 exec_lo, s6 ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX10-NEXT: v_mov_b32_e32 v6, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v5, v7 +; GFX10-NEXT: v_mov_b32_e32 v7, v5 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 @@ -8942,13 +8815,12 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX10-NEXT: s_cbranch_execnz .LBB21_3 ; GFX10-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s5 -; GFX10-NEXT: v_mov_b32_e32 v0, v4 +; GFX10-NEXT: v_mov_b32_e32 v0, v5 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_add_u32_e32 v8, 0x400, v4 ; GFX90A-NEXT: s_mov_b64 s[6:7], exec ; GFX90A-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: v_readfirstlane_b32 s8, v0 @@ -8960,38 +8832,37 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX90A-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX90A-NEXT: s_nop 0 -; GFX90A-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 -; GFX90A-NEXT: ; implicit-def: $vgpr4 +; GFX90A-NEXT: buffer_load_dword v9, v4, s[8:11], 0 offen offset:1024 ; GFX90A-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB21_1 ; GFX90A-NEXT: ; %bb.2: ; GFX90A-NEXT: s_mov_b64 exec, s[6:7] ; GFX90A-NEXT: s_mov_b64 s[6:7], 0 -; GFX90A-NEXT: v_lshlrev_b32_e32 v9, 16, v5 +; GFX90A-NEXT: v_lshlrev_b32_e32 v10, 16, v5 ; GFX90A-NEXT: s_movk_i32 s14, 0x7fff -; GFX90A-NEXT: v_and_b32_e32 v10, 0xffff0000, v5 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v5 ; GFX90A-NEXT: s_mov_b32 s15, 0x7060302 ; GFX90A-NEXT: .LBB21_3: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Loop Header: Depth=1 ; GFX90A-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v7 -; GFX90A-NEXT: v_max_f32_e32 v4, v4, v9 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s14 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v7 -; GFX90A-NEXT: v_max_f32_e32 v5, v5, v10 -; GFX90A-NEXT: v_bfe_u32 v6, v5, 16, 1 -; GFX90A-NEXT: v_add3_u32 v6, v6, v5, s14 -; GFX90A-NEXT: v_or_b32_e32 v11, 0x400000, v5 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cndmask_b32_e32 v5, v6, v11, vcc -; GFX90A-NEXT: v_perm_b32 v6, v5, v4, s15 +; GFX90A-NEXT: v_lshlrev_b32_e32 v6, 16, v9 +; GFX90A-NEXT: v_max_f32_e32 v6, v6, v10 +; GFX90A-NEXT: v_bfe_u32 v7, v6, 16, 1 +; GFX90A-NEXT: v_add3_u32 v7, v7, v6, s14 +; GFX90A-NEXT: v_or_b32_e32 v8, 0x400000, v6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v6, v6 +; GFX90A-NEXT: v_cndmask_b32_e32 v6, v7, v8, vcc +; GFX90A-NEXT: v_and_b32_e32 v7, 0xffff0000, v9 +; GFX90A-NEXT: v_max_f32_e32 v7, v7, v5 +; GFX90A-NEXT: v_bfe_u32 v8, v7, 16, 1 +; GFX90A-NEXT: v_add3_u32 v8, v8, v7, s14 +; GFX90A-NEXT: v_or_b32_e32 v11, 0x400000, v7 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v7, v7 +; GFX90A-NEXT: v_cndmask_b32_e32 v7, v8, v11, vcc +; GFX90A-NEXT: v_perm_b32 v8, v7, v6, s15 ; GFX90A-NEXT: s_mov_b64 s[12:13], exec -; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[6:7], v[6:7] op_sel:[0,1] +; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[8:9], v[8:9] op_sel:[0,1] ; GFX90A-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX90A-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX90A-NEXT: v_readfirstlane_b32 s8, v0 @@ -9003,27 +8874,26 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX90A-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX90A-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: buffer_atomic_cmpswap v[4:5], v8, s[8:11], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[6:7], v4, s[8:11], 0 offen offset:1024 glc ; GFX90A-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB21_4 ; GFX90A-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX90A-NEXT: s_mov_b64 exec, s[12:13] ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v7 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v6, v9 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v7, v4 +; GFX90A-NEXT: v_mov_b32_e32 v9, v6 ; GFX90A-NEXT: buffer_wbinvl1 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB21_3 ; GFX90A-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX90A-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v0, v4 +; GFX90A-NEXT: v_mov_b32_e32 v0, v6 ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX908: ; %bb.0: ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX908-NEXT: v_add_u32_e32 v7, 0x400, v4 ; GFX908-NEXT: s_mov_b64 s[6:7], exec ; GFX908-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: v_readfirstlane_b32 s8, v0 @@ -9035,8 +8905,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX908-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_nop 0 -; GFX908-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX908-NEXT: ; implicit-def: $vgpr4 +; GFX908-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB21_1 ; GFX908-NEXT: ; %bb.2: @@ -9050,24 +8919,24 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX908-NEXT: ; =>This Loop Header: Depth=1 ; GFX908-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v6 -; GFX908-NEXT: v_max_f32_e32 v4, v4, v8 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s14 -; GFX908-NEXT: v_or_b32_e32 v10, 0x400000, v4 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v10, vcc -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 -; GFX908-NEXT: v_max_f32_e32 v5, v5, v9 -; GFX908-NEXT: v_bfe_u32 v10, v5, 16, 1 -; GFX908-NEXT: v_add3_u32 v10, v10, v5, s14 -; GFX908-NEXT: v_or_b32_e32 v11, 0x400000, v5 +; GFX908-NEXT: v_lshlrev_b32_e32 v5, 16, v7 +; GFX908-NEXT: v_max_f32_e32 v5, v5, v8 +; GFX908-NEXT: v_bfe_u32 v6, v5, 16, 1 +; GFX908-NEXT: v_add3_u32 v6, v6, v5, s14 +; GFX908-NEXT: v_or_b32_e32 v10, 0x400000, v5 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cndmask_b32_e32 v5, v10, v11, vcc -; GFX908-NEXT: v_perm_b32 v5, v5, v4, s15 -; GFX908-NEXT: v_mov_b32_e32 v4, v5 -; GFX908-NEXT: s_mov_b64 s[12:13], exec +; GFX908-NEXT: v_cndmask_b32_e32 v5, v6, v10, vcc +; GFX908-NEXT: v_and_b32_e32 v6, 0xffff0000, v7 +; GFX908-NEXT: v_max_f32_e32 v6, v6, v9 +; GFX908-NEXT: v_bfe_u32 v10, v6, 16, 1 +; GFX908-NEXT: v_add3_u32 v10, v10, v6, s14 +; GFX908-NEXT: v_or_b32_e32 v11, 0x400000, v6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v6, v6 +; GFX908-NEXT: v_cndmask_b32_e32 v6, v10, v11, vcc +; GFX908-NEXT: v_perm_b32 v6, v6, v5, s15 ; GFX908-NEXT: v_mov_b32_e32 v5, v6 +; GFX908-NEXT: s_mov_b64 s[12:13], exec +; GFX908-NEXT: v_mov_b32_e32 v6, v7 ; GFX908-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX908-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX908-NEXT: v_readfirstlane_b32 s8, v0 @@ -9079,27 +8948,26 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX908-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX908-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX908-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB21_4 ; GFX908-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX908-NEXT: s_mov_b64 exec, s[12:13] ; GFX908-NEXT: s_waitcnt vmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v6 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v5, v7 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v6, v4 +; GFX908-NEXT: v_mov_b32_e32 v7, v5 ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB21_3 ; GFX908-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX908-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v0, v4 +; GFX908-NEXT: v_mov_b32_e32 v0, v5 ; GFX908-NEXT: s_setpc_b64 s[30:31] ; ; GFX8-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_u32_e32 v7, vcc, 0x400, v4 ; GFX8-NEXT: s_mov_b64 s[6:7], exec ; GFX8-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: v_readfirstlane_b32 s8, v0 @@ -9111,8 +8979,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX8-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_nop 0 -; GFX8-NEXT: buffer_load_dword v6, v4, s[8:11], 0 offen offset:1024 -; GFX8-NEXT: ; implicit-def: $vgpr4 +; GFX8-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB21_1 ; GFX8-NEXT: ; %bb.2: @@ -9124,27 +8991,27 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX8-NEXT: ; =>This Loop Header: Depth=1 ; GFX8-NEXT: ; Child Loop BB21_4 Depth 2 ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v6 -; GFX8-NEXT: v_max_f32_e32 v4, v4, v8 -; GFX8-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v5, vcc, v5, v4 -; GFX8-NEXT: v_add_u32_e32 v5, vcc, 0x7fff, v5 -; GFX8-NEXT: v_or_b32_e32 v10, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v5, v10, vcc -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 -; GFX8-NEXT: v_max_f32_e32 v5, v5, v9 -; GFX8-NEXT: v_bfe_u32 v10, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v10, vcc, v10, v5 -; GFX8-NEXT: v_add_u32_e32 v10, vcc, 0x7fff, v10 -; GFX8-NEXT: v_or_b32_e32 v11, 0x400000, v5 +; GFX8-NEXT: v_lshlrev_b32_e32 v5, 16, v7 +; GFX8-NEXT: v_max_f32_e32 v5, v5, v8 +; GFX8-NEXT: v_bfe_u32 v6, v5, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v5 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 +; GFX8-NEXT: v_or_b32_e32 v10, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_cndmask_b32_e32 v5, v10, v11, vcc -; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v5, v5, v4, 16 -; GFX8-NEXT: v_mov_b32_e32 v4, v5 -; GFX8-NEXT: s_mov_b64 s[12:13], exec +; GFX8-NEXT: v_cndmask_b32_e32 v5, v6, v10, vcc +; GFX8-NEXT: v_and_b32_e32 v6, 0xffff0000, v7 +; GFX8-NEXT: v_max_f32_e32 v6, v6, v9 +; GFX8-NEXT: v_bfe_u32 v10, v6, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v10, vcc, v10, v6 +; GFX8-NEXT: v_add_u32_e32 v10, vcc, 0x7fff, v10 +; GFX8-NEXT: v_or_b32_e32 v11, 0x400000, v6 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v6, v6 +; GFX8-NEXT: v_cndmask_b32_e32 v6, v10, v11, vcc +; GFX8-NEXT: v_lshrrev_b32_e32 v6, 16, v6 +; GFX8-NEXT: v_alignbit_b32 v6, v6, v5, 16 ; GFX8-NEXT: v_mov_b32_e32 v5, v6 +; GFX8-NEXT: s_mov_b64 s[12:13], exec +; GFX8-NEXT: v_mov_b32_e32 v6, v7 ; GFX8-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX8-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX8-NEXT: v_readfirstlane_b32 s8, v0 @@ -9156,27 +9023,26 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX8-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX8-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: buffer_atomic_cmpswap v[4:5], v7, s[8:11], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[5:6], v4, s[8:11], 0 offen offset:1024 glc ; GFX8-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB21_4 ; GFX8-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX8-NEXT: s_mov_b64 exec, s[12:13] ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v6 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v5, v7 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v7, v5 ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB21_3 ; GFX8-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX8-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v0, v4 +; GFX8-NEXT: v_mov_b32_e32 v0, v5 ; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX7-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v8, vcc, 0x400, v4 ; GFX7-NEXT: s_mov_b64 s[6:7], exec ; GFX7-NEXT: .LBB21_1: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: v_readfirstlane_b32 s8, v0 @@ -9187,8 +9053,7 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX7-NEXT: v_cmp_eq_u64_e64 s[4:5], s[10:11], v[2:3] ; GFX7-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX7-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] -; GFX7-NEXT: buffer_load_dword v7, v4, s[8:11], 0 offen offset:1024 -; GFX7-NEXT: ; implicit-def: $vgpr4 +; GFX7-NEXT: buffer_load_dword v8, v4, s[8:11], 0 offen offset:1024 ; GFX7-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB21_1 ; GFX7-NEXT: ; %bb.2: @@ -9196,27 +9061,27 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX7-NEXT: v_mul_f32_e32 v6, 1.0, v6 ; GFX7-NEXT: v_mul_f32_e32 v5, 1.0, v5 ; GFX7-NEXT: s_waitcnt vmcnt(0) -; GFX7-NEXT: v_and_b32_e32 v4, 0xffff0000, v7 -; GFX7-NEXT: v_lshlrev_b32_e32 v7, 16, v7 +; GFX7-NEXT: v_and_b32_e32 v7, 0xffff0000, v8 +; GFX7-NEXT: v_lshlrev_b32_e32 v8, 16, v8 ; GFX7-NEXT: s_mov_b64 s[6:7], 0 -; GFX7-NEXT: v_and_b32_e32 v9, 0xffff0000, v6 -; GFX7-NEXT: v_and_b32_e32 v10, 0xffff0000, v5 +; GFX7-NEXT: v_and_b32_e32 v10, 0xffff0000, v6 +; GFX7-NEXT: v_and_b32_e32 v11, 0xffff0000, v5 ; GFX7-NEXT: .LBB21_3: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Loop Header: Depth=1 ; GFX7-NEXT: ; Child Loop BB21_4 Depth 2 -; GFX7-NEXT: v_mul_f32_e32 v5, 1.0, v4 -; GFX7-NEXT: v_and_b32_e32 v4, 0xffff0000, v5 ; GFX7-NEXT: v_mul_f32_e32 v6, 1.0, v7 -; GFX7-NEXT: v_max_f32_e32 v4, v4, v9 -; GFX7-NEXT: v_and_b32_e32 v7, 0xffff0000, v6 +; GFX7-NEXT: v_and_b32_e32 v5, 0xffff0000, v6 +; GFX7-NEXT: v_mul_f32_e32 v7, 1.0, v8 +; GFX7-NEXT: v_max_f32_e32 v5, v5, v10 +; GFX7-NEXT: v_and_b32_e32 v8, 0xffff0000, v7 +; GFX7-NEXT: v_lshrrev_b32_e32 v6, 16, v6 ; GFX7-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_max_f32_e32 v7, v7, v10 -; GFX7-NEXT: v_alignbit_b32 v5, v5, v6, 16 -; GFX7-NEXT: v_alignbit_b32 v4, v4, v7, 16 -; GFX7-NEXT: v_mov_b32_e32 v7, v5 +; GFX7-NEXT: v_max_f32_e32 v8, v8, v11 +; GFX7-NEXT: v_alignbit_b32 v6, v6, v7, 16 +; GFX7-NEXT: v_alignbit_b32 v5, v5, v8, 16 +; GFX7-NEXT: v_mov_b32_e32 v9, v6 ; GFX7-NEXT: s_mov_b64 s[12:13], exec -; GFX7-NEXT: v_mov_b32_e32 v6, v4 +; GFX7-NEXT: v_mov_b32_e32 v8, v5 ; GFX7-NEXT: .LBB21_4: ; Parent Loop BB21_3 Depth=1 ; GFX7-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX7-NEXT: v_readfirstlane_b32 s8, v0 @@ -9228,23 +9093,23 @@ define <2 x bfloat> @buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterf ; GFX7-NEXT: s_and_b64 s[4:5], vcc, s[4:5] ; GFX7-NEXT: s_and_saveexec_b64 s[4:5], s[4:5] ; GFX7-NEXT: s_waitcnt vmcnt(0) -; GFX7-NEXT: buffer_atomic_cmpswap v[6:7], v8, s[8:11], 0 offen glc +; GFX7-NEXT: buffer_atomic_cmpswap v[8:9], v4, s[8:11], 0 offen offset:1024 glc ; GFX7-NEXT: s_xor_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB21_4 ; GFX7-NEXT: ; %bb.5: ; in Loop: Header=BB21_3 Depth=1 ; GFX7-NEXT: s_mov_b64 exec, s[12:13] ; GFX7-NEXT: s_waitcnt vmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v6, v5 -; GFX7-NEXT: v_and_b32_e32 v4, 0xffff0000, v6 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v8, v6 +; GFX7-NEXT: v_and_b32_e32 v7, 0xffff0000, v8 ; GFX7-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX7-NEXT: v_lshlrev_b32_e32 v7, 16, v6 +; GFX7-NEXT: v_lshlrev_b32_e32 v8, 16, v8 ; GFX7-NEXT: buffer_wbinvl1 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX7-NEXT: s_cbranch_execnz .LBB21_3 ; GFX7-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX7-NEXT: s_or_b64 exec, exec, s[6:7] -; GFX7-NEXT: v_mov_b32_e32 v0, v7 -; GFX7-NEXT: v_mov_b32_e32 v1, v4 +; GFX7-NEXT: v_mov_b32_e32 v0, v8 +; GFX7-NEXT: v_mov_b32_e32 v1, v7 ; GFX7-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_v2bf16__offset__waterfall__amdgpu_no_fine_grained_memory: @@ -9353,10 +9218,9 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX942-NEXT: v_mov_b32_e32 v1, v0 ; GFX942-NEXT: v_mov_b32_e32 v0, s16 ; GFX942-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen offset:1024 -; GFX942-NEXT: s_add_i32 s6, s16, 0x400 ; GFX942-NEXT: s_mov_b64 s[4:5], 0 ; GFX942-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX942-NEXT: v_mov_b32_e32 v3, s6 +; GFX942-NEXT: v_mov_b32_e32 v3, s16 ; GFX942-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -9365,7 +9229,7 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX942-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX942-NEXT: v_mov_b64_e32 v[0:1], v[4:5] ; GFX942-NEXT: buffer_wbl2 sc0 sc1 -; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen sc0 +; GFX942-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[0:3], 0 offen offset:1024 sc0 ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: buffer_inv sc0 sc1 ; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -9404,10 +9268,9 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX90A-NEXT: v_mov_b32_e32 v1, v0 ; GFX90A-NEXT: v_mov_b32_e32 v0, s20 ; GFX90A-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX90A-NEXT: s_add_i32 s6, s20, 0x400 ; GFX90A-NEXT: s_mov_b64 s[4:5], 0 ; GFX90A-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX90A-NEXT: v_mov_b32_e32 v3, s6 +; GFX90A-NEXT: v_mov_b32_e32 v3, s20 ; GFX90A-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -9416,7 +9279,7 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX90A-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[4:5] op_sel:[0,1] ; GFX90A-NEXT: buffer_wbl2 -; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX90A-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: buffer_invl2 ; GFX90A-NEXT: buffer_wbinvl1 @@ -9434,10 +9297,9 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_mov_b32_e32 v1, v0 ; GFX908-NEXT: v_mov_b32_e32 v0, s20 ; GFX908-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX908-NEXT: s_add_i32 s6, s20, 0x400 ; GFX908-NEXT: s_mov_b64 s[4:5], 0 ; GFX908-NEXT: v_max_f32_e32 v2, v1, v1 -; GFX908-NEXT: v_mov_b32_e32 v3, s6 +; GFX908-NEXT: v_mov_b32_e32 v3, s20 ; GFX908-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt vmcnt(0) @@ -9446,7 +9308,7 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX908-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX908-NEXT: v_mov_b32_e32 v0, v4 ; GFX908-NEXT: v_mov_b32_e32 v1, v5 -; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX908-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: buffer_wbinvl1 ; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 @@ -9463,10 +9325,9 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_mov_b32_e32 v1, v0 ; GFX8-NEXT: v_mov_b32_e32 v0, s20 ; GFX8-NEXT: buffer_load_dword v0, v0, s[16:19], 0 offen offset:1024 -; GFX8-NEXT: s_add_i32 s6, s20, 0x400 ; GFX8-NEXT: s_mov_b64 s[4:5], 0 ; GFX8-NEXT: v_mul_f32_e32 v2, 1.0, v1 -; GFX8-NEXT: v_mov_b32_e32 v3, s6 +; GFX8-NEXT: v_mov_b32_e32 v3, s20 ; GFX8-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt vmcnt(0) @@ -9475,7 +9336,7 @@ define float @buffer_fat_ptr_system_atomic_fmax_ret_f32__offset__amdgpu_no_fine_ ; GFX8-NEXT: v_max_f32_e32 v4, v0, v2 ; GFX8-NEXT: v_mov_b32_e32 v0, v4 ; GFX8-NEXT: v_mov_b32_e32 v1, v5 -; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen glc +; GFX8-NEXT: buffer_atomic_cmpswap v[0:1], v3, s[16:19], 0 offen offset:1024 glc ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: buffer_wbinvl1 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v0, v5 |