aboutsummaryrefslogtreecommitdiff
path: root/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll')
-rw-r--r--llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll222
1 files changed, 64 insertions, 158 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll b/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll
index 1e4b633..fc36ed9 100644
--- a/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll
+++ b/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll
@@ -45,27 +45,18 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p
; GFX9-GISEL: ; %bb.0: ; %entry
; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10
-; GFX9-GISEL-NEXT: s_mov_b32 s11, 0
-; GFX9-GISEL-NEXT: s_mov_b32 s4, s11
-; GFX9-GISEL-NEXT: s_mov_b32 s6, s11
; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s1
+; GFX9-GISEL-NEXT: s_mov_b32 s4, s1
; GFX9-GISEL-NEXT: s_mov_b32 s5, s2
-; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5]
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s3
-; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7]
+; GFX9-GISEL-NEXT: s_mov_b32 s6, s3
; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX9-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen glc slc
; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20
; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30
-; GFX9-GISEL-NEXT: s_mov_b32 s4, s11
-; GFX9-GISEL-NEXT: s_mov_b32 s6, s11
; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s1
+; GFX9-GISEL-NEXT: s_mov_b32 s4, s1
; GFX9-GISEL-NEXT: s_mov_b32 s5, s2
-; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5]
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s3
-; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7]
+; GFX9-GISEL-NEXT: s_mov_b32 s6, s3
; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s0
; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX9-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen glc slc
@@ -105,27 +96,18 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p
; GFX942-GISEL: ; %bb.0: ; %entry
; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-GISEL-NEXT: s_load_dword s11, s[4:5], 0x10
-; GFX942-GISEL-NEXT: s_mov_b32 s7, 0
-; GFX942-GISEL-NEXT: s_mov_b32 s8, s7
-; GFX942-GISEL-NEXT: s_mov_b32 s10, s7
; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-NEXT: s_mov_b32 s6, s1
+; GFX942-GISEL-NEXT: s_mov_b32 s8, s1
; GFX942-GISEL-NEXT: s_mov_b32 s9, s2
-; GFX942-GISEL-NEXT: s_or_b64 s[8:9], s[6:7], s[8:9]
-; GFX942-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX942-GISEL-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11]
+; GFX942-GISEL-NEXT: s_mov_b32 s10, s3
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX942-GISEL-NEXT: buffer_load_dword v0, v0, s[8:11], 0 offen nt
; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x20
-; GFX942-GISEL-NEXT: s_load_dword s9, s[4:5], 0x30
-; GFX942-GISEL-NEXT: s_mov_b32 s4, s7
-; GFX942-GISEL-NEXT: s_mov_b32 s8, s7
+; GFX942-GISEL-NEXT: s_load_dword s7, s[4:5], 0x30
; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-NEXT: s_mov_b32 s6, s1
+; GFX942-GISEL-NEXT: s_mov_b32 s4, s1
; GFX942-GISEL-NEXT: s_mov_b32 s5, s2
-; GFX942-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5]
; GFX942-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX942-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9]
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s0
; GFX942-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX942-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen nt
@@ -168,29 +150,22 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p
; GFX10-GISEL: ; %bb.0: ; %entry
; GFX10-GISEL-NEXT: s_clause 0x1
; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0
-; GFX10-GISEL-NEXT: s_load_dword s5, s[8:9], 0x10
-; GFX10-GISEL-NEXT: s_mov_b32 s7, 0
-; GFX10-GISEL-NEXT: s_mov_b32 s10, s7
-; GFX10-GISEL-NEXT: s_mov_b32 s4, s7
+; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10
; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-GISEL-NEXT: s_mov_b32 s6, s1
-; GFX10-GISEL-NEXT: s_mov_b32 s11, s2
; GFX10-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX10-GISEL-NEXT: s_or_b64 s[0:1], s[6:7], s[10:11]
+; GFX10-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX10-GISEL-NEXT: s_mov_b32 s5, s2
; GFX10-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX10-GISEL-NEXT: s_or_b64 s[2:3], s[6:7], s[4:5]
-; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen slc
+; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen slc
; GFX10-GISEL-NEXT: s_clause 0x1
-; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3
; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20
-; GFX10-GISEL-NEXT: s_load_dword s11, s[8:9], 0x30
+; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3
+; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30
; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-GISEL-NEXT: s_mov_b32 s6, s1
-; GFX10-GISEL-NEXT: s_mov_b32 s5, s2
; GFX10-GISEL-NEXT: v_mov_b32_e32 v1, s0
-; GFX10-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5]
+; GFX10-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX10-GISEL-NEXT: s_mov_b32 s5, s2
; GFX10-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX10-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[10:11]
; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX10-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen glc slc
; GFX10-GISEL-NEXT: s_endpgm
@@ -234,32 +209,21 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p
; GFX11-GISEL: ; %bb.0: ; %entry
; GFX11-GISEL-NEXT: s_clause 0x1
; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0
-; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10
-; GFX11-GISEL-NEXT: s_mov_b32 s9, 0
-; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-GISEL-NEXT: s_mov_b32 s10, s9
-; GFX11-GISEL-NEXT: s_mov_b32 s6, s9
+; GFX11-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10
; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX11-GISEL-NEXT: s_mov_b32 s11, s2
; GFX11-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX11-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11]
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7]
-; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen slc dlc
+; GFX11-GISEL-NEXT: s_mov_b32 s8, s1
+; GFX11-GISEL-NEXT: s_mov_b32 s9, s2
+; GFX11-GISEL-NEXT: s_mov_b32 s10, s3
+; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], 0 offen slc dlc
; GFX11-GISEL-NEXT: s_clause 0x1
; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20
; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30
-; GFX11-GISEL-NEXT: s_mov_b32 s4, s9
; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX11-GISEL-NEXT: s_mov_b32 s5, s2
; GFX11-GISEL-NEXT: v_mov_b32_e32 v1, s0
-; GFX11-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5]
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7]
+; GFX11-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX11-GISEL-NEXT: s_mov_b32 s5, s2
+; GFX11-GISEL-NEXT: s_mov_b32 s6, s3
; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX11-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], 0 offen glc slc dlc
; GFX11-GISEL-NEXT: s_endpgm
@@ -303,32 +267,21 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p
; GFX12-GISEL: ; %bb.0: ; %entry
; GFX12-GISEL-NEXT: s_clause 0x1
; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0
-; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10
-; GFX12-GISEL-NEXT: s_mov_b32 s9, 0
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-GISEL-NEXT: s_mov_b32 s10, s9
-; GFX12-GISEL-NEXT: s_mov_b32 s6, s9
+; GFX12-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX12-GISEL-NEXT: s_mov_b32 s11, s2
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11]
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7]
-; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], null offen th:TH_LOAD_NT
+; GFX12-GISEL-NEXT: s_mov_b32 s8, s1
+; GFX12-GISEL-NEXT: s_mov_b32 s9, s2
+; GFX12-GISEL-NEXT: s_mov_b32 s10, s3
+; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], null offen th:TH_LOAD_NT
; GFX12-GISEL-NEXT: s_clause 0x1
; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20
; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30
-; GFX12-GISEL-NEXT: s_mov_b32 s4, s9
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX12-GISEL-NEXT: s_mov_b32 s5, s2
; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, s0
-; GFX12-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5]
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7]
+; GFX12-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX12-GISEL-NEXT: s_mov_b32 s5, s2
+; GFX12-GISEL-NEXT: s_mov_b32 s6, s3
; GFX12-GISEL-NEXT: s_wait_loadcnt 0x0
; GFX12-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], null offen th:TH_STORE_NT
; GFX12-GISEL-NEXT: s_endpgm
@@ -374,28 +327,19 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp
; GFX9-GISEL: ; %bb.0: ; %entry
; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10
-; GFX9-GISEL-NEXT: s_mov_b32 s11, 0
-; GFX9-GISEL-NEXT: s_mov_b32 s4, s11
-; GFX9-GISEL-NEXT: s_mov_b32 s6, s11
; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s1
+; GFX9-GISEL-NEXT: s_mov_b32 s4, s1
; GFX9-GISEL-NEXT: s_mov_b32 s5, s2
-; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5]
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s3
-; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7]
+; GFX9-GISEL-NEXT: s_mov_b32 s6, s3
; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX9-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen glc
; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20
; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30
-; GFX9-GISEL-NEXT: s_mov_b32 s4, s11
-; GFX9-GISEL-NEXT: s_mov_b32 s6, s11
; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s1
+; GFX9-GISEL-NEXT: s_mov_b32 s4, s1
; GFX9-GISEL-NEXT: s_mov_b32 s5, s2
-; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5]
-; GFX9-GISEL-NEXT: s_mov_b32 s10, s3
-; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7]
+; GFX9-GISEL-NEXT: s_mov_b32 s6, s3
; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s0
; GFX9-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen
; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0)
@@ -436,28 +380,19 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp
; GFX942-GISEL: ; %bb.0: ; %entry
; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-GISEL-NEXT: s_load_dword s11, s[4:5], 0x10
-; GFX942-GISEL-NEXT: s_mov_b32 s7, 0
-; GFX942-GISEL-NEXT: s_mov_b32 s8, s7
-; GFX942-GISEL-NEXT: s_mov_b32 s10, s7
; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-NEXT: s_mov_b32 s6, s1
+; GFX942-GISEL-NEXT: s_mov_b32 s8, s1
; GFX942-GISEL-NEXT: s_mov_b32 s9, s2
-; GFX942-GISEL-NEXT: s_or_b64 s[8:9], s[6:7], s[8:9]
-; GFX942-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX942-GISEL-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11]
+; GFX942-GISEL-NEXT: s_mov_b32 s10, s3
; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX942-GISEL-NEXT: buffer_load_dword v0, v0, s[8:11], 0 offen sc0 sc1
; GFX942-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x20
-; GFX942-GISEL-NEXT: s_load_dword s9, s[4:5], 0x30
-; GFX942-GISEL-NEXT: s_mov_b32 s4, s7
-; GFX942-GISEL-NEXT: s_mov_b32 s8, s7
+; GFX942-GISEL-NEXT: s_load_dword s7, s[4:5], 0x30
; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX942-GISEL-NEXT: s_mov_b32 s6, s1
+; GFX942-GISEL-NEXT: s_mov_b32 s4, s1
; GFX942-GISEL-NEXT: s_mov_b32 s5, s2
-; GFX942-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5]
; GFX942-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX942-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9]
; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s0
; GFX942-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen sc0 sc1
; GFX942-GISEL-NEXT: s_waitcnt vmcnt(0)
@@ -501,30 +436,23 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp
; GFX10-GISEL: ; %bb.0: ; %entry
; GFX10-GISEL-NEXT: s_clause 0x1
; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0
-; GFX10-GISEL-NEXT: s_load_dword s5, s[8:9], 0x10
-; GFX10-GISEL-NEXT: s_mov_b32 s7, 0
-; GFX10-GISEL-NEXT: s_mov_b32 s10, s7
-; GFX10-GISEL-NEXT: s_mov_b32 s4, s7
+; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10
; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-GISEL-NEXT: s_mov_b32 s6, s1
-; GFX10-GISEL-NEXT: s_mov_b32 s11, s2
; GFX10-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX10-GISEL-NEXT: s_or_b64 s[0:1], s[6:7], s[10:11]
+; GFX10-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX10-GISEL-NEXT: s_mov_b32 s5, s2
; GFX10-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX10-GISEL-NEXT: s_or_b64 s[2:3], s[6:7], s[4:5]
-; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen glc dlc
+; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen glc dlc
; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX10-GISEL-NEXT: s_clause 0x1
-; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3
; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20
-; GFX10-GISEL-NEXT: s_load_dword s11, s[8:9], 0x30
+; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3
+; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30
; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX10-GISEL-NEXT: s_mov_b32 s6, s1
-; GFX10-GISEL-NEXT: s_mov_b32 s5, s2
; GFX10-GISEL-NEXT: v_mov_b32_e32 v1, s0
-; GFX10-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5]
+; GFX10-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX10-GISEL-NEXT: s_mov_b32 s5, s2
; GFX10-GISEL-NEXT: s_mov_b32 s6, s3
-; GFX10-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[10:11]
; GFX10-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen
; GFX10-GISEL-NEXT: s_waitcnt_vscnt null, 0x0
; GFX10-GISEL-NEXT: s_endpgm
@@ -569,33 +497,22 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp
; GFX11-GISEL: ; %bb.0: ; %entry
; GFX11-GISEL-NEXT: s_clause 0x1
; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0
-; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10
-; GFX11-GISEL-NEXT: s_mov_b32 s9, 0
-; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-GISEL-NEXT: s_mov_b32 s10, s9
-; GFX11-GISEL-NEXT: s_mov_b32 s6, s9
+; GFX11-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10
; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX11-GISEL-NEXT: s_mov_b32 s11, s2
; GFX11-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX11-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11]
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7]
-; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen glc dlc
+; GFX11-GISEL-NEXT: s_mov_b32 s8, s1
+; GFX11-GISEL-NEXT: s_mov_b32 s9, s2
+; GFX11-GISEL-NEXT: s_mov_b32 s10, s3
+; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], 0 offen glc dlc
; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0)
; GFX11-GISEL-NEXT: s_clause 0x1
; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20
; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30
-; GFX11-GISEL-NEXT: s_mov_b32 s4, s9
; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX11-GISEL-NEXT: s_mov_b32 s5, s2
; GFX11-GISEL-NEXT: v_mov_b32_e32 v1, s0
-; GFX11-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5]
-; GFX11-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7]
+; GFX11-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX11-GISEL-NEXT: s_mov_b32 s5, s2
+; GFX11-GISEL-NEXT: s_mov_b32 s6, s3
; GFX11-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], 0 offen dlc
; GFX11-GISEL-NEXT: s_waitcnt_vscnt null, 0x0
; GFX11-GISEL-NEXT: s_endpgm
@@ -640,33 +557,22 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp
; GFX12-GISEL: ; %bb.0: ; %entry
; GFX12-GISEL-NEXT: s_clause 0x1
; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0
-; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10
-; GFX12-GISEL-NEXT: s_mov_b32 s9, 0
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-GISEL-NEXT: s_mov_b32 s10, s9
-; GFX12-GISEL-NEXT: s_mov_b32 s6, s9
+; GFX12-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX12-GISEL-NEXT: s_mov_b32 s11, s2
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11]
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7]
-; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], null offen th:TH_LOAD_NT scope:SCOPE_SYS
+; GFX12-GISEL-NEXT: s_mov_b32 s8, s1
+; GFX12-GISEL-NEXT: s_mov_b32 s9, s2
+; GFX12-GISEL-NEXT: s_mov_b32 s10, s3
+; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], null offen th:TH_LOAD_NT scope:SCOPE_SYS
; GFX12-GISEL-NEXT: s_wait_loadcnt 0x0
; GFX12-GISEL-NEXT: s_clause 0x1
; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20
; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30
-; GFX12-GISEL-NEXT: s_mov_b32 s4, s9
; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s1
-; GFX12-GISEL-NEXT: s_mov_b32 s5, s2
; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, s0
-; GFX12-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5]
-; GFX12-GISEL-NEXT: s_mov_b32 s8, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
-; GFX12-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7]
+; GFX12-GISEL-NEXT: s_mov_b32 s4, s1
+; GFX12-GISEL-NEXT: s_mov_b32 s5, s2
+; GFX12-GISEL-NEXT: s_mov_b32 s6, s3
; GFX12-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], null offen th:TH_STORE_NT scope:SCOPE_SYS
; GFX12-GISEL-NEXT: s_wait_storecnt 0x0
; GFX12-GISEL-NEXT: s_endpgm