[AMDGPU] Extend and reorganize memory legalizer tests

[scott.linder]

Summary

• Rename some tests to try to make a convention (where all components are optional) of:
  
  <addrspace>_<syncscope>_<memory-orders>_<operation>

• Split up at a level of granularity appropriate for the different RUN lines (i.e. split on addrspace so GFX6 can avoid FLAT) and that makes running a specific test reasonable in terms of wall time taken. This also means when run as part of the test suite the testing is not one serial bottleneck.

• Auto-generate check lines with update_llc_test_checks.py to make future maintenance more tractable.

[scott.linder]

Opened by mistake, intended to update D91545

llvm/test/CodeGen/AMDGPU/memory-legalizer-fence.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd- -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX6 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX7 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-WGP %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-CU %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx700 -amdgcn-skip-cache-invalidations -verify-machineinstrs < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s
+
+define amdgpu_kernel void @singlethread_acquire_fence() {
+; GFX6-LABEL: singlethread_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread") acquire
+  ret void
+}
+
+define amdgpu_kernel void @singlethread_release_fence() {
+; GFX6-LABEL: singlethread_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread") release
+  ret void
+}
+
+define amdgpu_kernel void @singlethread_acq_rel_fence() {
+; GFX6-LABEL: singlethread_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @singlethread_seq_cst_fence() {
+; GFX6-LABEL: singlethread_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @singlethread_one_as_acquire_fence() {
+; GFX6-LABEL: singlethread_one_as_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_one_as_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_one_as_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_one_as_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_one_as_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread-one-as") acquire
+  ret void
+}
+
+define amdgpu_kernel void @singlethread_one_as_release_fence() {
+; GFX6-LABEL: singlethread_one_as_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_one_as_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_one_as_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_one_as_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_one_as_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread-one-as") release
+  ret void
+}
+
+define amdgpu_kernel void @singlethread_one_as_acq_rel_fence() {
+; GFX6-LABEL: singlethread_one_as_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_one_as_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_one_as_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_one_as_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_one_as_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread-one-as") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @singlethread_one_as_seq_cst_fence() {
+; GFX6-LABEL: singlethread_one_as_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: singlethread_one_as_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: singlethread_one_as_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: singlethread_one_as_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: singlethread_one_as_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("singlethread-one-as") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_acquire_fence() {
+; GFX6-LABEL: wavefront_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront") acquire
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_release_fence() {
+; GFX6-LABEL: wavefront_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront") release
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_acq_rel_fence() {
+; GFX6-LABEL: wavefront_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_seq_cst_fence() {
+; GFX6-LABEL: wavefront_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_one_as_acquire_fence() {
+; GFX6-LABEL: wavefront_one_as_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_one_as_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_one_as_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_one_as_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_one_as_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront-one-as") acquire
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_one_as_release_fence() {
+; GFX6-LABEL: wavefront_one_as_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_one_as_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_one_as_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_one_as_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_one_as_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront-one-as") release
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_one_as_acq_rel_fence() {
+; GFX6-LABEL: wavefront_one_as_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_one_as_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_one_as_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_one_as_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_one_as_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront-one-as") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @wavefront_one_as_seq_cst_fence() {
+; GFX6-LABEL: wavefront_one_as_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: wavefront_one_as_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: wavefront_one_as_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: wavefront_one_as_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: wavefront_one_as_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("wavefront-one-as") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_acquire_fence() {
+; GFX6-LABEL: workgroup_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup") acquire
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_release_fence() {
+; GFX6-LABEL: workgroup_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup") release
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_acq_rel_fence() {
+; GFX6-LABEL: workgroup_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_seq_cst_fence() {
+; GFX6-LABEL: workgroup_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_one_as_acquire_fence() {
+; GFX6-LABEL: workgroup_one_as_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_one_as_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_one_as_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_one_as_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_one_as_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup-one-as") acquire
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_one_as_release_fence() {
+; GFX6-LABEL: workgroup_one_as_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_one_as_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_one_as_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_one_as_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_one_as_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup-one-as") release
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_one_as_acq_rel_fence() {
+; GFX6-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_one_as_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup-one-as") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @workgroup_one_as_seq_cst_fence() {
+; GFX6-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: workgroup_one_as_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("workgroup-one-as") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @agent_acquire_fence() {
+; GFX6-LABEL: agent_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent") acquire
+  ret void
+}
+
+define amdgpu_kernel void @agent_release_fence() {
+; GFX6-LABEL: agent_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent") release
+  ret void
+}
+
+define amdgpu_kernel void @agent_acq_rel_fence() {
+; GFX6-LABEL: agent_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @agent_seq_cst_fence() {
+; GFX6-LABEL: agent_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @agent_one_as_acquire_fence() {
+; GFX6-LABEL: agent_one_as_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_one_as_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_one_as_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_one_as_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_one_as_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent-one-as") acquire
+  ret void
+}
+
+define amdgpu_kernel void @agent_one_as_release_fence() {
+; GFX6-LABEL: agent_one_as_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_one_as_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_one_as_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_one_as_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_one_as_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent-one-as") release
+  ret void
+}
+
+define amdgpu_kernel void @agent_one_as_acq_rel_fence() {
+; GFX6-LABEL: agent_one_as_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_one_as_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_one_as_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_one_as_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_one_as_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent-one-as") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @agent_one_as_seq_cst_fence() {
+; GFX6-LABEL: agent_one_as_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: agent_one_as_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: agent_one_as_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: agent_one_as_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: agent_one_as_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("agent-one-as") seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @system_acquire_fence() {
+; GFX6-LABEL: system_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence acquire
+  ret void
+}
+
+define amdgpu_kernel void @system_release_fence() {
+; GFX6-LABEL: system_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence release
+  ret void
+}
+
+define amdgpu_kernel void @system_acq_rel_fence() {
+; GFX6-LABEL: system_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @system_seq_cst_fence() {
+; GFX6-LABEL: system_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence seq_cst
+  ret void
+}
+
+define amdgpu_kernel void @system_one_as_acquire_fence() {
+; GFX6-LABEL: system_one_as_acquire_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_one_as_acquire_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_one_as_acquire_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_one_as_acquire_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_one_as_acquire_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("one-as") acquire
+  ret void
+}
+
+define amdgpu_kernel void @system_one_as_release_fence() {
+; GFX6-LABEL: system_one_as_release_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_one_as_release_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_one_as_release_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_one_as_release_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_one_as_release_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("one-as") release
+  ret void
+}
+
+define amdgpu_kernel void @system_one_as_acq_rel_fence() {
+; GFX6-LABEL: system_one_as_acq_rel_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_one_as_acq_rel_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_one_as_acq_rel_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_one_as_acq_rel_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_one_as_acq_rel_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("one-as") acq_rel
+  ret void
+}
+
+define amdgpu_kernel void @system_one_as_seq_cst_fence() {
+; GFX6-LABEL: system_one_as_seq_cst_fence:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_wbinvl1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: system_one_as_seq_cst_fence:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    buffer_wbinvl1_vol
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: system_one_as_seq_cst_fence:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-WGP-NEXT:    buffer_gl0_inv
+; GFX10-WGP-NEXT:    buffer_gl1_inv
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: system_one_as_seq_cst_fence:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-CU-NEXT:    buffer_gl0_inv
+; GFX10-CU-NEXT:    buffer_gl1_inv
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: system_one_as_seq_cst_fence:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+entry:
+  fence syncscope("one-as") seq_cst
+  ret void
+}
+

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX7 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-WGP %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-CU %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx700 -amdgcn-skip-cache-invalidations -verify-machineinstrs < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s
+
+define amdgpu_kernel void @flat_nontemporal_load_0(
+; GFX7-LABEL: flat_nontemporal_load_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    flat_load_dword v0, v[0:1] glc slc
+; GFX7-NEXT:    v_mov_b32_e32 v2, s2
+; GFX7-NEXT:    v_mov_b32_e32 v3, s3
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[2:3], v0
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: flat_nontemporal_load_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1] slc
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: flat_nontemporal_load_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1] slc
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: flat_nontemporal_load_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
+; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1] glc slc
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32* %in, i32* %out) {
+entry:
+  %val = load i32, i32* %in, align 4, !nontemporal !0
+  store i32 %val, i32* %out
+  ret void
+}
+
+define amdgpu_kernel void @flat_nontemporal_load_1(
+; GFX7-LABEL: flat_nontemporal_load_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v3, s1
+; GFX7-NEXT:    v_add_i32_e32 v2, vcc, s0, v2
+; GFX7-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; GFX7-NEXT:    flat_load_dword v2, v[2:3] glc slc
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: flat_nontemporal_load_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_add_co_u32_e64 v0, s0, s0, v0
+; GFX10-WGP-NEXT:    v_add_co_ci_u32_e64 v1, s0, s1, 0, s0
+; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1] slc
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: flat_nontemporal_load_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_add_co_u32_e64 v0, s0, s0, v0
+; GFX10-CU-NEXT:    v_add_co_ci_u32_e64 v1, s0, s1, 0, s0
+; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1] slc
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: flat_nontemporal_load_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
+; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v2, vcc, s0, v2
+; SKIP-CACHE-INV-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[2:3] glc slc
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32* %in, i32* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val.gep = getelementptr inbounds i32, i32* %in, i32 %tid
+  %val = load i32, i32* %val.gep, align 4, !nontemporal !0
+  store i32 %val, i32* %out
+  ret void
+}
+
+define amdgpu_kernel void @flat_nontemporal_store_0(
+; GFX7-LABEL: flat_nontemporal_store_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    flat_load_dword v0, v[0:1]
+; GFX7-NEXT:    v_mov_b32_e32 v2, s2
+; GFX7-NEXT:    v_mov_b32_e32 v3, s3
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[2:3], v0 glc slc
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: flat_nontemporal_store_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-WGP-NEXT:    flat_load_dword v2, v[0:1]
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2 slc
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: flat_nontemporal_store_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-CU-NEXT:    flat_load_dword v2, v[0:1]
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2 slc
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: flat_nontemporal_store_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
+; SKIP-CACHE-INV-NEXT:    flat_load_dword v0, v[0:1]
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s3
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    flat_store_dword v[2:3], v0 glc slc
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32* %in, i32* %out) {
+entry:
+  %val = load i32, i32* %in, align 4
+  store i32 %val, i32* %out, !nontemporal !0
+  ret void
+}
+
+define amdgpu_kernel void @flat_nontemporal_store_1(
+; GFX7-LABEL: flat_nontemporal_store_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    v_mov_b32_e32 v2, s1
+; GFX7-NEXT:    flat_load_dword v2, v[1:2]
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
+; GFX7-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2 glc slc
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: flat_nontemporal_store_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v2, s1
+; GFX10-WGP-NEXT:    v_add_co_u32_e64 v0, s0, s2, v0
+; GFX10-WGP-NEXT:    flat_load_dword v2, v[1:2]
+; GFX10-WGP-NEXT:    v_add_co_ci_u32_e64 v1, s0, s3, 0, s0
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2 slc
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: flat_nontemporal_store_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
+; GFX10-CU-NEXT:    v_add_co_u32_e64 v0, s0, s2, v0
+; GFX10-CU-NEXT:    flat_load_dword v2, v[1:2]
+; GFX10-CU-NEXT:    v_add_co_ci_u32_e64 v1, s0, s3, 0, s0
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2 slc
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: flat_nontemporal_store_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
+; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[1:2]
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
+; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
+; SKIP-CACHE-INV-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2 glc slc
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32* %in, i32* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val = load i32, i32* %in, align 4
+  %out.gep = getelementptr inbounds i32, i32* %out, i32 %tid
+  store i32 %val, i32* %out.gep, !nontemporal !0
+  ret void
+}
+
+!0 = !{i32 1}
+declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd- -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX6 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX7 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-WGP %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-CU %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx700 -amdgcn-skip-cache-invalidations -verify-machineinstrs < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s
+
+define amdgpu_kernel void @global_nontemporal_load_0(
+; GFX6-LABEL: global_nontemporal_load_0:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX6-NEXT:    s_mov_b32 s4, s2
+; GFX6-NEXT:    s_mov_b32 s5, s3
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s0
+; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: global_nontemporal_load_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v2, s0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: global_nontemporal_load_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-WGP-NEXT:    global_store_dword v0, v1, s[2:3]
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: global_nontemporal_load_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: global_nontemporal_load_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(1)* %out) {
+entry:
+  %val = load i32, i32 addrspace(1)* %in, align 4, !nontemporal !0
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_kernel void @global_nontemporal_load_1(
+; GFX6-LABEL: global_nontemporal_load_1:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b32 s0, s6
+; GFX6-NEXT:    s_mov_b32 s1, s7
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    s_mov_b32 s7, s3
+; GFX6-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64 glc slc
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: global_nontemporal_load_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v3, s1
+; GFX7-NEXT:    v_add_i32_e32 v2, vcc, s0, v2
+; GFX7-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; GFX7-NEXT:    flat_load_dword v2, v[2:3] glc slc
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: global_nontemporal_load_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    global_load_dword v0, v0, s[0:1] slc
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    global_store_dword v1, v0, s[2:3]
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: global_nontemporal_load_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    global_load_dword v0, v0, s[0:1] slc
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    global_store_dword v1, v0, s[2:3]
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: global_nontemporal_load_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, 0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s0, s6
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s1, s7
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, 0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, s3
+; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64 glc slc
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(1)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val.gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %tid
+  %val = load i32, i32 addrspace(1)* %val.gep, align 4, !nontemporal !0
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_kernel void @global_nontemporal_store_0(
+; GFX6-LABEL: global_nontemporal_store_0:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX6-NEXT:    s_mov_b32 s4, s2
+; GFX6-NEXT:    s_mov_b32 s5, s3
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s0
+; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0 glc slc
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: global_nontemporal_store_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v2, s0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2 glc slc
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: global_nontemporal_store_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-WGP-NEXT:    global_store_dword v0, v1, s[2:3] slc
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: global_nontemporal_store_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3] slc
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: global_nontemporal_store_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[4:7], 0 glc slc
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(1)* %out) {
+entry:
+  %val = load i32, i32 addrspace(1)* %in, align 4
+  store i32 %val, i32 addrspace(1)* %out, !nontemporal !0
+  ret void
+}
+
+define amdgpu_kernel void @global_nontemporal_store_1(
+; GFX6-LABEL: global_nontemporal_store_1:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v2, s0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[4:7], 0 addr64 glc slc
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: global_nontemporal_store_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
+; GFX7-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v2, s0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2 glc slc
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: global_nontemporal_store_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-WGP-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-WGP-NEXT:    global_store_dword v0, v1, s[2:3] slc
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: global_nontemporal_store_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-CU-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3] slc
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: global_nontemporal_store_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, 0
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, 0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v2, v[0:1], s[4:7], 0 addr64 glc slc
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(1)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val = load i32, i32 addrspace(1)* %in, align 4
+  %out.gep = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 %tid
+  store i32 %val, i32 addrspace(1)* %out.gep, !nontemporal !0
+  ret void
+}
+
+!0 = !{i32 1}
+declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-nontemporal.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd- -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX6 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX7 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-WGP %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-CU %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx700 -amdgcn-skip-cache-invalidations -verify-machineinstrs < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s
+
+define amdgpu_kernel void @local_nontemporal_load_0(
+; GFX6-LABEL: local_nontemporal_load_0:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; GFX6-NEXT:    s_mov_b32 m0, -1
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
+; GFX6-NEXT:    ds_read_b32 v0, v0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: local_nontemporal_load_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
+; GFX7-NEXT:    s_mov_b32 m0, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    ds_read_b32 v2, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: local_nontemporal_load_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-WGP-NEXT:    ds_read_b32 v0, v0
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: local_nontemporal_load_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-CU-NEXT:    ds_read_b32 v0, v0
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: local_nontemporal_load_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
+; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(3)* %in, i32 addrspace(1)* %out) {
+entry:
+  %val = load i32, i32 addrspace(3)* %in, align 4, !nontemporal !0
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_kernel void @local_nontemporal_load_1(
+; GFX6-LABEL: local_nontemporal_load_1:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    s_mov_b32 m0, -1
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
+; GFX6-NEXT:    ds_read_b32 v0, v0
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: local_nontemporal_load_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX7-NEXT:    s_mov_b32 m0, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
+; GFX7-NEXT:    ds_read_b32 v2, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: local_nontemporal_load_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-WGP-NEXT:    ds_read_b32 v0, v0
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: local_nontemporal_load_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-CU-NEXT:    ds_read_b32 v0, v0
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: local_nontemporal_load_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
+; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(3)* %in, i32 addrspace(1)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val.gep = getelementptr inbounds i32, i32 addrspace(3)* %in, i32 %tid
+  %val = load i32, i32 addrspace(3)* %val.gep, align 4, !nontemporal !0
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_kernel void @local_nontemporal_store_0(
+; GFX6-LABEL: local_nontemporal_store_0:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0xb
+; GFX6-NEXT:    s_mov_b32 m0, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dword s1, s[2:3], 0x0
+; GFX6-NEXT:    v_mov_b32_e32 v0, s0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v1, s1
+; GFX6-NEXT:    ds_write_b32 v0, v1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: local_nontemporal_store_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
+; GFX7-NEXT:    s_mov_b32 m0, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    ds_write_b32 v0, v1
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: local_nontemporal_store_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-WGP-NEXT:    ds_write_b32 v0, v1
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: local_nontemporal_store_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-CU-NEXT:    ds_write_b32 v0, v1
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: local_nontemporal_store_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
+; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(3)* %out) {
+entry:
+  %val = load i32, i32 addrspace(1)* %in, align 4
+  store i32 %val, i32 addrspace(3)* %out, !nontemporal !0
+  ret void
+}
+
+define amdgpu_kernel void @local_nontemporal_store_1(
+; GFX6-LABEL: local_nontemporal_store_1:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0xb
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    s_mov_b32 m0, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dword s1, s[2:3], 0x0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v1, s1
+; GFX6-NEXT:    ds_write_b32 v0, v1
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: local_nontemporal_store_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX7-NEXT:    s_mov_b32 m0, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    ds_write_b32 v0, v1
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: local_nontemporal_store_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-WGP-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-WGP-NEXT:    ds_write_b32 v0, v1
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: local_nontemporal_store_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-CU-NEXT:    ds_write_b32 v0, v1
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: local_nontemporal_store_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
+; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
+; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(3)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val = load i32, i32 addrspace(1)* %in, align 4
+  %out.gep = getelementptr inbounds i32, i32 addrspace(3)* %out, i32 %tid
+  store i32 %val, i32 addrspace(3)* %out.gep, !nontemporal !0
+  ret void
+}
+
+!0 = !{i32 1}
+declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd- -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX6 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX7 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-WGP %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+cumode -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-CU %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx700 -amdgcn-skip-cache-invalidations -verify-machineinstrs < %s | FileCheck --check-prefixes=SKIP-CACHE-INV %s
+
+define amdgpu_kernel void @private_nontemporal_load_0(
+; GFX6-LABEL: private_nontemporal_load_0:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX6-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; GFX6-NEXT:    s_mov_b32 s10, -1
+; GFX6-NEXT:    s_mov_b32 s11, 0xe8f000
+; GFX6-NEXT:    s_add_u32 s8, s8, s3
+; GFX6-NEXT:    s_addc_u32 s9, s9, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
+; GFX6-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen glc slc
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: private_nontemporal_load_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX7-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
+; GFX7-NEXT:    s_add_u32 s8, s8, s7
+; GFX7-NEXT:    s_addc_u32 s9, s9, 0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    buffer_load_dword v2, v0, s[8:11], 0 offen glc slc
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: private_nontemporal_load_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-WGP-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-WGP-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-WGP-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-WGP-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen slc
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: private_nontemporal_load_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-CU-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-CU-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-CU-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-CU-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen slc
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: private_nontemporal_load_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[8:9]
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s8, s0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[8:11], s[8:9], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_add_u32 s8, s8, s3
+; SKIP-CACHE-INV-NEXT:    s_addc_u32 s9, s9, 0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
+; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen glc slc
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(5)* %in, i32 addrspace(1)* %out) {
+entry:
+  %val = load i32, i32 addrspace(5)* %in, align 4, !nontemporal !0
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_kernel void @private_nontemporal_load_1(
+; GFX6-LABEL: private_nontemporal_load_1:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
+; GFX6-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; GFX6-NEXT:    s_mov_b32 s10, -1
+; GFX6-NEXT:    s_mov_b32 s11, 0xe8f000
+; GFX6-NEXT:    s_add_u32 s8, s8, s3
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    s_addc_u32 s9, s9, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
+; GFX6-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen glc slc
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: private_nontemporal_load_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX7-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
+; GFX7-NEXT:    s_add_u32 s8, s8, s7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX7-NEXT:    s_addc_u32 s9, s9, 0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
+; GFX7-NEXT:    buffer_load_dword v2, v0, s[8:11], 0 offen glc slc
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: private_nontemporal_load_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-WGP-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-WGP-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-WGP-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-WGP-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen slc
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: private_nontemporal_load_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-CU-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-CU-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-CU-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-CU-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen slc
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: private_nontemporal_load_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[8:9]
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s8, s0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[8:11], s[8:9], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_add_u32 s8, s8, s3
+; SKIP-CACHE-INV-NEXT:    s_addc_u32 s9, s9, 0
+; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
+; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen glc slc
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(5)* %in, i32 addrspace(1)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val.gep = getelementptr inbounds i32, i32 addrspace(5)* %in, i32 %tid
+  %val = load i32, i32 addrspace(5)* %val.gep, align 4, !nontemporal !0
+  store i32 %val, i32 addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_kernel void @private_nontemporal_store_0(
+; GFX6-LABEL: private_nontemporal_store_0:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_mov_b32 s4, SCRATCH_RSRC_DWORD0
+; GFX6-NEXT:    s_mov_b32 s5, SCRATCH_RSRC_DWORD1
+; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    s_mov_b32 s7, 0xe8f000
+; GFX6-NEXT:    s_add_u32 s4, s4, s3
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0xb
+; GFX6-NEXT:    s_addc_u32 s5, s5, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dword s1, s[2:3], 0x0
+; GFX6-NEXT:    v_mov_b32_e32 v1, s0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s1
+; GFX6-NEXT:    buffer_store_dword v0, v1, s[4:7], 0 offen glc slc
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: private_nontemporal_store_0:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX7-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
+; GFX7-NEXT:    s_add_u32 s8, s8, s7
+; GFX7-NEXT:    s_addc_u32 s9, s9, 0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s2
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    buffer_store_dword v0, v1, s[8:11], 0 offen glc slc
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: private_nontemporal_store_0:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-WGP-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-WGP-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-WGP-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s2
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-WGP-NEXT:    buffer_store_dword v0, v1, s[8:11], 0 offen slc
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: private_nontemporal_store_0:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-CU-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-CU-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-CU-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-CU-NEXT:    buffer_store_dword v0, v1, s[8:11], 0 offen slc
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: private_nontemporal_store_0:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, v1, s[4:7], 0 offen glc slc
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(5)* %out) {
+entry:
+  %val = load i32, i32 addrspace(1)* %in, align 4
+  store i32 %val, i32 addrspace(5)* %out, !nontemporal !0
+  ret void
+}
+
+define amdgpu_kernel void @private_nontemporal_store_1(
+; GFX6-LABEL: private_nontemporal_store_1:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_mov_b32 s4, SCRATCH_RSRC_DWORD0
+; GFX6-NEXT:    s_mov_b32 s5, SCRATCH_RSRC_DWORD1
+; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    s_mov_b32 s7, 0xe8f000
+; GFX6-NEXT:    s_add_u32 s4, s4, s3
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0xb
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    s_addc_u32 s5, s5, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dword s1, s[2:3], 0x0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v1, s1
+; GFX6-NEXT:    buffer_store_dword v1, v0, s[4:7], 0 offen glc slc
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: private_nontemporal_store_1:
+; GFX7:       ; %bb.0: ; %entry
+; GFX7-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX7-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX7-NEXT:    s_load_dword s2, s[4:5], 0x2
+; GFX7-NEXT:    s_add_u32 s8, s8, s7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX7-NEXT:    s_addc_u32 s9, s9, 0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX7-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    buffer_store_dword v1, v0, s[8:11], 0 offen glc slc
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: private_nontemporal_store_1:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-WGP-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-WGP-NEXT:    s_clause 0x1
+; GFX10-WGP-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-WGP-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-WGP-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-WGP-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-WGP-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-WGP-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-WGP-NEXT:    buffer_store_dword v1, v0, s[8:11], 0 offen slc
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: private_nontemporal_store_1:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX10-CU-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX10-CU-NEXT:    s_clause 0x1
+; GFX10-CU-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-CU-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-CU-NEXT:    s_add_u32 s8, s8, s7
+; GFX10-CU-NEXT:    s_addc_u32 s9, s9, 0
+; GFX10-CU-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
+; GFX10-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
+; GFX10-CU-NEXT:    buffer_store_dword v1, v0, s[8:11], 0 offen slc
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: private_nontemporal_store_1:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
+; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v1, v0, s[4:7], 0 offen glc slc
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+    i32 addrspace(1)* %in, i32 addrspace(5)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %val = load i32, i32 addrspace(1)* %in, align 4
+  %out.gep = getelementptr inbounds i32, i32 addrspace(5)* %out, i32 %tid
+  store i32 %val, i32 addrspace(5)* %out.gep, !nontemporal !0
+  ret void
+}
+
+!0 = !{i32 1}
+declare i32 @llvm.amdgcn.workitem.id.x()