[llvm] [AMDGPU] Fix setting nontemporal in memory legalizer (PR #83815)

Mirko BrkuĊĦanin via llvm-commits llvm-commits at lists.llvm.org
Mon Mar 4 02:23:14 PST 2024


https://github.com/mbrkusanin updated https://github.com/llvm/llvm-project/pull/83815

>From 67a49733d61a2467ff4bce3c7462441cf31674ae Mon Sep 17 00:00:00 2001
From: Mirko Brkusanin <Mirko.Brkusanin at amd.com>
Date: Fri, 1 Mar 2024 18:25:40 +0100
Subject: [PATCH 1/2] Add nontemporal+volatile tests

---
 .../memory-legalizer-flat-nontemporal.ll      | 165 ++++++++++++++
 .../memory-legalizer-global-nontemporal.ll    | 158 ++++++++++++++
 .../memory-legalizer-local-nontemporal.ll     | 179 +++++++++++++++
 .../memory-legalizer-private-nontemporal.ll   | 203 ++++++++++++++++++
 4 files changed, 705 insertions(+)

diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll
index 9f0556023aa24a..35a9ab72b04362 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll
@@ -684,5 +684,170 @@ entry:
   ret void
 }
 
+define amdgpu_kernel void @flat_nontemporal_volatile_load(
+; GFX7-LABEL: flat_nontemporal_volatile_load:
+; 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 v2, v[0:1] glc
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: flat_nontemporal_volatile_load:
+; GFX10-WGP:       ; %bb.0: ; %entry
+; GFX10-WGP-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; 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] glc dlc
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-WGP-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    flat_store_dword v[0:1], v2
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: flat_nontemporal_volatile_load:
+; GFX10-CU:       ; %bb.0: ; %entry
+; GFX10-CU-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; 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] glc dlc
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: flat_nontemporal_volatile_load:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
+; 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 v2, v[0:1] glc
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; 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 lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+;
+; GFX90A-NOTTGSPLIT-LABEL: flat_nontemporal_volatile_load:
+; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, s1
+; GFX90A-NOTTGSPLIT-NEXT:    flat_load_dword v2, v[0:1] glc
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, s3
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    flat_store_dword v[0:1], v2
+; GFX90A-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX90A-TGSPLIT-LABEL: flat_nontemporal_volatile_load:
+; GFX90A-TGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-TGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v0, s0
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v1, s1
+; GFX90A-TGSPLIT-NEXT:    flat_load_dword v2, v[0:1] glc
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v1, s3
+; GFX90A-TGSPLIT-NEXT:    flat_store_dword v[0:1], v2
+; GFX90A-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-NOTTGSPLIT-LABEL: flat_nontemporal_volatile_load:
+; GFX940-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX940-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, s1
+; GFX940-NOTTGSPLIT-NEXT:    flat_load_dword v2, v[0:1] sc0 sc1
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, s3
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    flat_store_dword v[0:1], v2 sc0 sc1
+; GFX940-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-TGSPLIT-LABEL: flat_nontemporal_volatile_load:
+; GFX940-TGSPLIT:       ; %bb.0: ; %entry
+; GFX940-TGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
+; GFX940-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v0, s0
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v1, s1
+; GFX940-TGSPLIT-NEXT:    flat_load_dword v2, v[0:1] sc0 sc1
+; GFX940-TGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v1, s3
+; GFX940-TGSPLIT-NEXT:    flat_store_dword v[0:1], v2 sc0 sc1
+; GFX940-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX11-WGP-LABEL: flat_nontemporal_volatile_load:
+; GFX11-WGP:       ; %bb.0: ; %entry
+; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-WGP-NEXT:    flat_load_b32 v2, v[0:1] glc dlc
+; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-WGP-NEXT:    flat_store_b32 v[0:1], v2
+; GFX11-WGP-NEXT:    s_endpgm
+;
+; GFX11-CU-LABEL: flat_nontemporal_volatile_load:
+; GFX11-CU:       ; %bb.0: ; %entry
+; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-CU-NEXT:    flat_load_b32 v2, v[0:1] glc dlc
+; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-CU-NEXT:    flat_store_b32 v[0:1], v2
+; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: flat_nontemporal_volatile_load:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX12-WGP-NEXT:    s_wait_kmcnt 0x0
+; GFX12-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX12-WGP-NEXT:    flat_load_b32 v2, v[0:1] scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    flat_store_b32 v[0:1], v2
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: flat_nontemporal_volatile_load:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX12-CU-NEXT:    s_wait_kmcnt 0x0
+; GFX12-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX12-CU-NEXT:    flat_load_b32 v2, v[0:1] scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    flat_store_b32 v[0:1], v2
+; GFX12-CU-NEXT:    s_endpgm
+    ptr %in, ptr %out) {
+entry:
+  %val = load volatile i32, ptr %in, align 4, !nontemporal !0
+  store i32 %val, ptr %out
+  ret void
+}
+
 !0 = !{i32 1}
 declare i32 @llvm.amdgcn.workitem.id.x()
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll
index b2751326fa784b..672f803a12235d 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll
@@ -674,5 +674,163 @@ entry:
   ret void
 }
 
+define amdgpu_kernel void @global_nontemporal_volatile_load(
+; GFX6-LABEL: global_nontemporal_volatile_load:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX6-NEXT:    s_mov_b32 s7, 0x100f000
+; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b32 s4, s0
+; GFX6-NEXT:    s_mov_b32 s5, s1
+; GFX6-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    s_mov_b32 s4, s2
+; GFX6-NEXT:    s_mov_b32 s5, s3
+; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: global_nontemporal_volatile_load:
+; 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 v2, v[0:1] glc
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: global_nontemporal_volatile_load:
+; 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:    s_waitcnt lgkmcnt(0)
+; GFX10-WGP-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
+; GFX10-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-WGP-NEXT:    global_store_dword v0, v1, s[2:3]
+; GFX10-WGP-NEXT:    s_endpgm
+;
+; GFX10-CU-LABEL: global_nontemporal_volatile_load:
+; 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:    s_waitcnt lgkmcnt(0)
+; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
+; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
+; GFX10-CU-NEXT:    s_endpgm
+;
+; SKIP-CACHE-INV-LABEL: global_nontemporal_volatile_load:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
+; 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_mov_b32 s4, s0
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
+; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+;
+; GFX90A-NOTTGSPLIT-LABEL: global_nontemporal_volatile_load:
+; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, 0
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    global_load_dword v1, v0, s[0:1] glc
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    global_store_dword v0, v1, s[2:3]
+; GFX90A-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX90A-TGSPLIT-LABEL: global_nontemporal_volatile_load:
+; GFX90A-TGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-TGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v0, 0
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    global_load_dword v1, v0, s[0:1] glc
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    global_store_dword v0, v1, s[2:3]
+; GFX90A-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-NOTTGSPLIT-LABEL: global_nontemporal_volatile_load:
+; GFX940-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX940-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, 0
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    global_load_dword v1, v0, s[0:1] sc0 sc1
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    global_store_dword v0, v1, s[2:3] sc0 sc1
+; GFX940-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-TGSPLIT-LABEL: global_nontemporal_volatile_load:
+; GFX940-TGSPLIT:       ; %bb.0: ; %entry
+; GFX940-TGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v0, 0
+; GFX940-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-TGSPLIT-NEXT:    global_load_dword v1, v0, s[0:1] sc0 sc1
+; GFX940-TGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-TGSPLIT-NEXT:    global_store_dword v0, v1, s[2:3] sc0 sc1
+; GFX940-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX11-WGP-LABEL: global_nontemporal_volatile_load:
+; GFX11-WGP:       ; %bb.0: ; %entry
+; GFX11-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX11-WGP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-WGP-NEXT:    global_load_b32 v1, v0, s[0:1] glc dlc
+; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-WGP-NEXT:    global_store_b32 v0, v1, s[2:3]
+; GFX11-WGP-NEXT:    s_nop 0
+; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-WGP-NEXT:    s_endpgm
+;
+; GFX11-CU-LABEL: global_nontemporal_volatile_load:
+; GFX11-CU:       ; %bb.0: ; %entry
+; GFX11-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX11-CU-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-CU-NEXT:    global_load_b32 v1, v0, s[0:1] glc dlc
+; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-CU-NEXT:    global_store_b32 v0, v1, s[2:3]
+; GFX11-CU-NEXT:    s_nop 0
+; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: global_nontemporal_volatile_load:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX12-WGP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX12-WGP-NEXT:    s_wait_kmcnt 0x0
+; GFX12-WGP-NEXT:    global_load_b32 v1, v0, s[0:1] scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    global_store_b32 v0, v1, s[2:3]
+; GFX12-WGP-NEXT:    s_nop 0
+; GFX12-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: global_nontemporal_volatile_load:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX12-CU-NEXT:    v_mov_b32_e32 v0, 0
+; GFX12-CU-NEXT:    s_wait_kmcnt 0x0
+; GFX12-CU-NEXT:    global_load_b32 v1, v0, s[0:1] scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    global_store_b32 v0, v1, s[2:3]
+; GFX12-CU-NEXT:    s_nop 0
+; GFX12-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-CU-NEXT:    s_endpgm
+    ptr addrspace(1) %in, ptr addrspace(1) %out) {
+entry:
+  %val = load volatile i32, ptr addrspace(1) %in, align 4, !nontemporal !0
+  store i32 %val, ptr addrspace(1) %out
+  ret void
+}
+
 !0 = !{i32 1}
 declare i32 @llvm.amdgcn.workitem.id.x()
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-local-nontemporal.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-local-nontemporal.ll
index 0eb2f7b1d694ec..16ede2665400ec 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-local-nontemporal.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-local-nontemporal.ll
@@ -717,5 +717,184 @@ entry:
   ret void
 }
 
+define amdgpu_kernel void @local_nontemporal_volatile_load(
+; GFX6-LABEL: local_nontemporal_volatile_load:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
+; GFX6-NEXT:    s_mov_b32 m0, -1
+; GFX6-NEXT:    s_mov_b32 s3, 0x100f000
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s2
+; 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_volatile_load:
+; 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_volatile_load:
+; 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:    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_volatile_load:
+; 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:    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_volatile_load:
+; SKIP-CACHE-INV:       ; %bb.0: ; %entry
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
+; 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_mov_b32_e32 v0, s2
+; 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
+;
+; GFX90A-NOTTGSPLIT-LABEL: local_nontemporal_volatile_load:
+; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-NOTTGSPLIT-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX90A-NOTTGSPLIT-NEXT:    ds_read_b32 v0, v0
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX90A-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX90A-TGSPLIT-LABEL: local_nontemporal_volatile_load:
+; GFX90A-TGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-TGSPLIT-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX90A-TGSPLIT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX90A-TGSPLIT-NEXT:    ds_read_b32 v0, v0
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX90A-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-NOTTGSPLIT-LABEL: local_nontemporal_volatile_load:
+; GFX940-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX940-NOTTGSPLIT-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX940-NOTTGSPLIT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s4
+; GFX940-NOTTGSPLIT-NEXT:    ds_read_b32 v0, v0
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    global_store_dword v1, v0, s[2:3] sc0 sc1
+; GFX940-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-TGSPLIT-LABEL: local_nontemporal_volatile_load:
+; GFX940-TGSPLIT:       ; %bb.0: ; %entry
+; GFX940-TGSPLIT-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX940-TGSPLIT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX940-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v0, s4
+; GFX940-TGSPLIT-NEXT:    ds_read_b32 v0, v0
+; GFX940-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-TGSPLIT-NEXT:    global_store_dword v1, v0, s[2:3] sc0 sc1
+; GFX940-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX11-WGP-LABEL: local_nontemporal_volatile_load:
+; GFX11-WGP:       ; %bb.0: ; %entry
+; GFX11-WGP-NEXT:    s_clause 0x1
+; GFX11-WGP-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX11-WGP-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-WGP-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX11-WGP-NEXT:    ds_load_b32 v0, v0
+; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-WGP-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX11-WGP-NEXT:    s_nop 0
+; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-WGP-NEXT:    s_endpgm
+;
+; GFX11-CU-LABEL: local_nontemporal_volatile_load:
+; GFX11-CU:       ; %bb.0: ; %entry
+; GFX11-CU-NEXT:    s_clause 0x1
+; GFX11-CU-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX11-CU-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-CU-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX11-CU-NEXT:    ds_load_b32 v0, v0
+; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-CU-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX11-CU-NEXT:    s_nop 0
+; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: local_nontemporal_volatile_load:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_clause 0x1
+; GFX12-WGP-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX12-WGP-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX12-WGP-NEXT:    s_wait_kmcnt 0x0
+; GFX12-WGP-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-WGP-NEXT:    ds_load_b32 v0, v0
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX12-WGP-NEXT:    s_nop 0
+; GFX12-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: local_nontemporal_volatile_load:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_clause 0x1
+; GFX12-CU-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX12-CU-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX12-CU-NEXT:    s_wait_kmcnt 0x0
+; GFX12-CU-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-CU-NEXT:    ds_load_b32 v0, v0
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX12-CU-NEXT:    s_nop 0
+; GFX12-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-CU-NEXT:    s_endpgm
+    ptr addrspace(3) %in, ptr addrspace(1) %out) {
+entry:
+  %val = load volatile i32, ptr addrspace(3) %in, align 4, !nontemporal !0
+  store i32 %val, ptr addrspace(1) %out
+  ret void
+}
+
 !0 = !{i32 1}
 declare i32 @llvm.amdgcn.workitem.id.x()
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll
index 069443791b3e1b..e0eff90c4f40ec 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll
@@ -813,5 +813,208 @@ entry:
   ret void
 }
 
+define amdgpu_kernel void @private_nontemporal_volatile_load(
+; GFX6-LABEL: private_nontemporal_volatile_load:
+; GFX6:       ; %bb.0: ; %entry
+; GFX6-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX6-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX6-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2
+; GFX6-NEXT:    s_add_u32 s8, s8, s7
+; GFX6-NEXT:    s_addc_u32 s9, s9, 0
+; GFX6-NEXT:    s_mov_b32 s3, 0x100f000
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s2
+; GFX6-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen glc
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT:    s_endpgm
+;
+; GFX7-LABEL: private_nontemporal_volatile_load:
+; 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
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+;
+; GFX10-WGP-LABEL: private_nontemporal_volatile_load:
+; 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:    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 glc dlc
+; 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_volatile_load:
+; 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:    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 glc dlc
+; 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_volatile_load:
+; 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_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
+; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
+; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
+; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen glc
+; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
+; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
+; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SKIP-CACHE-INV-NEXT:    s_endpgm
+;
+; GFX90A-NOTTGSPLIT-LABEL: private_nontemporal_volatile_load:
+; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-NOTTGSPLIT-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX90A-NOTTGSPLIT-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX90A-NOTTGSPLIT-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX90A-NOTTGSPLIT-NEXT:    s_add_u32 s8, s8, s7
+; GFX90A-NOTTGSPLIT-NEXT:    s_addc_u32 s9, s9, 0
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX90A-NOTTGSPLIT-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen glc
+; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-NOTTGSPLIT-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX90A-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX90A-TGSPLIT-LABEL: private_nontemporal_volatile_load:
+; GFX90A-TGSPLIT:       ; %bb.0: ; %entry
+; GFX90A-TGSPLIT-NEXT:    s_mov_b64 s[10:11], s[2:3]
+; GFX90A-TGSPLIT-NEXT:    s_mov_b64 s[8:9], s[0:1]
+; GFX90A-TGSPLIT-NEXT:    s_load_dword s2, s[4:5], 0x0
+; GFX90A-TGSPLIT-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX90A-TGSPLIT-NEXT:    s_add_u32 s8, s8, s7
+; GFX90A-TGSPLIT-NEXT:    s_addc_u32 s9, s9, 0
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    v_mov_b32_e32 v0, s2
+; GFX90A-TGSPLIT-NEXT:    buffer_load_dword v0, v0, s[8:11], 0 offen glc
+; GFX90A-TGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX90A-TGSPLIT-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX90A-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-NOTTGSPLIT-LABEL: private_nontemporal_volatile_load:
+; GFX940-NOTTGSPLIT:       ; %bb.0: ; %entry
+; GFX940-NOTTGSPLIT-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX940-NOTTGSPLIT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX940-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    scratch_load_dword v0, off, s4 sc0 sc1
+; GFX940-NOTTGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NOTTGSPLIT-NEXT:    global_store_dword v1, v0, s[2:3] sc0 sc1
+; GFX940-NOTTGSPLIT-NEXT:    s_endpgm
+;
+; GFX940-TGSPLIT-LABEL: private_nontemporal_volatile_load:
+; GFX940-TGSPLIT:       ; %bb.0: ; %entry
+; GFX940-TGSPLIT-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX940-TGSPLIT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX940-TGSPLIT-NEXT:    v_mov_b32_e32 v1, 0
+; GFX940-TGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX940-TGSPLIT-NEXT:    scratch_load_dword v0, off, s4 sc0 sc1
+; GFX940-TGSPLIT-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-TGSPLIT-NEXT:    global_store_dword v1, v0, s[2:3] sc0 sc1
+; GFX940-TGSPLIT-NEXT:    s_endpgm
+;
+; GFX11-WGP-LABEL: private_nontemporal_volatile_load:
+; GFX11-WGP:       ; %bb.0: ; %entry
+; GFX11-WGP-NEXT:    s_clause 0x1
+; GFX11-WGP-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX11-WGP-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX11-WGP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-WGP-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-WGP-NEXT:    scratch_load_b32 v0, off, s2 glc dlc
+; GFX11-WGP-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-WGP-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX11-WGP-NEXT:    s_nop 0
+; GFX11-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-WGP-NEXT:    s_endpgm
+;
+; GFX11-CU-LABEL: private_nontemporal_volatile_load:
+; GFX11-CU:       ; %bb.0: ; %entry
+; GFX11-CU-NEXT:    s_clause 0x1
+; GFX11-CU-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX11-CU-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX11-CU-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-CU-NEXT:    scratch_load_b32 v0, off, s2 glc dlc
+; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-CU-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX11-CU-NEXT:    s_nop 0
+; GFX11-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: private_nontemporal_volatile_load:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_clause 0x1
+; GFX12-WGP-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX12-WGP-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX12-WGP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX12-WGP-NEXT:    s_wait_kmcnt 0x0
+; GFX12-WGP-NEXT:    scratch_load_b32 v0, off, s2 scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX12-WGP-NEXT:    s_nop 0
+; GFX12-WGP-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: private_nontemporal_volatile_load:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_clause 0x1
+; GFX12-CU-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX12-CU-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
+; GFX12-CU-NEXT:    v_mov_b32_e32 v1, 0
+; GFX12-CU-NEXT:    s_wait_kmcnt 0x0
+; GFX12-CU-NEXT:    scratch_load_b32 v0, off, s2 scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX12-CU-NEXT:    s_nop 0
+; GFX12-CU-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-CU-NEXT:    s_endpgm
+    ptr addrspace(5) %in, ptr addrspace(1) %out) {
+entry:
+  %val = load volatile i32, ptr addrspace(5) %in, align 4, !nontemporal !0
+  store i32 %val, ptr addrspace(1) %out
+  ret void
+}
+
 !0 = !{i32 1}
 declare i32 @llvm.amdgcn.workitem.id.x()

>From febe0d929b8b26d4402a71cf401ce36bd606a808 Mon Sep 17 00:00:00 2001
From: Mirko Brkusanin <Mirko.Brkusanin at amd.com>
Date: Fri, 1 Mar 2024 18:29:18 +0100
Subject: [PATCH 2/2] Propely set nontemporal when volatile is also present

Iterator MI can advance in insertWait() but we need original
instruction to set temporal hint. Just move it before handling
volatile.
---
 llvm/lib/Target/AMDGPU/SIMemoryLegalizer.cpp           | 10 +++++-----
 .../AMDGPU/memory-legalizer-flat-nontemporal.ll        |  4 ++--
 .../AMDGPU/memory-legalizer-global-nontemporal.ll      |  4 ++--
 .../AMDGPU/memory-legalizer-private-nontemporal.ll     |  4 ++--
 4 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIMemoryLegalizer.cpp b/llvm/lib/Target/AMDGPU/SIMemoryLegalizer.cpp
index 4069a368f68719..73c23f0f987c3d 100644
--- a/llvm/lib/Target/AMDGPU/SIMemoryLegalizer.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMemoryLegalizer.cpp
@@ -2392,6 +2392,11 @@ bool SIGfx12CacheControl::enableVolatileAndOrNonTemporal(
 
   bool Changed = false;
 
+  if (IsNonTemporal) {
+    // Set non-temporal hint for all cache levels.
+    Changed |= setTH(MI, AMDGPU::CPol::TH_NT);
+  }
+
   if (IsVolatile) {
     Changed |= setScope(MI, AMDGPU::CPol::SCOPE_SYS);
 
@@ -2407,11 +2412,6 @@ bool SIGfx12CacheControl::enableVolatileAndOrNonTemporal(
                           Position::AFTER);
   }
 
-  if (IsNonTemporal) {
-    // Set non-temporal hint for all cache levels.
-    Changed |= setTH(MI, AMDGPU::CPol::TH_NT);
-  }
-
   return Changed;
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll
index 35a9ab72b04362..f6fac11d634c5b 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll
@@ -824,7 +824,7 @@ define amdgpu_kernel void @flat_nontemporal_volatile_load(
 ; GFX12-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX12-WGP-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-WGP-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GFX12-WGP-NEXT:    flat_load_b32 v2, v[0:1] scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    flat_load_b32 v2, v[0:1] th:TH_LOAD_NT scope:SCOPE_SYS
 ; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-WGP-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
@@ -836,7 +836,7 @@ define amdgpu_kernel void @flat_nontemporal_volatile_load(
 ; GFX12-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX12-CU-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-CU-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GFX12-CU-NEXT:    flat_load_b32 v2, v[0:1] scope:SCOPE_SYS
+; GFX12-CU-NEXT:    flat_load_b32 v2, v[0:1] th:TH_LOAD_NT scope:SCOPE_SYS
 ; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-CU-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
 ; GFX12-CU-NEXT:    s_wait_dscnt 0x0
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll
index 672f803a12235d..c95ff2c691a3cb 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll
@@ -807,7 +807,7 @@ define amdgpu_kernel void @global_nontemporal_volatile_load(
 ; GFX12-WGP-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX12-WGP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX12-WGP-NEXT:    s_wait_kmcnt 0x0
-; GFX12-WGP-NEXT:    global_load_b32 v1, v0, s[0:1] scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    global_load_b32 v1, v0, s[0:1] th:TH_LOAD_NT scope:SCOPE_SYS
 ; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-WGP-NEXT:    global_store_b32 v0, v1, s[2:3]
 ; GFX12-WGP-NEXT:    s_nop 0
@@ -819,7 +819,7 @@ define amdgpu_kernel void @global_nontemporal_volatile_load(
 ; GFX12-CU-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; GFX12-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX12-CU-NEXT:    s_wait_kmcnt 0x0
-; GFX12-CU-NEXT:    global_load_b32 v1, v0, s[0:1] scope:SCOPE_SYS
+; GFX12-CU-NEXT:    global_load_b32 v1, v0, s[0:1] th:TH_LOAD_NT scope:SCOPE_SYS
 ; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-CU-NEXT:    global_store_b32 v0, v1, s[2:3]
 ; GFX12-CU-NEXT:    s_nop 0
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll
index e0eff90c4f40ec..30296a9c3d8963 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll
@@ -989,7 +989,7 @@ define amdgpu_kernel void @private_nontemporal_volatile_load(
 ; GFX12-WGP-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
 ; GFX12-WGP-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX12-WGP-NEXT:    s_wait_kmcnt 0x0
-; GFX12-WGP-NEXT:    scratch_load_b32 v0, off, s2 scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    scratch_load_b32 v0, off, s2 th:TH_LOAD_NT scope:SCOPE_SYS
 ; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-WGP-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12-WGP-NEXT:    s_nop 0
@@ -1003,7 +1003,7 @@ define amdgpu_kernel void @private_nontemporal_volatile_load(
 ; GFX12-CU-NEXT:    s_load_b64 s[0:1], s[0:1], 0x8
 ; GFX12-CU-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX12-CU-NEXT:    s_wait_kmcnt 0x0
-; GFX12-CU-NEXT:    scratch_load_b32 v0, off, s2 scope:SCOPE_SYS
+; GFX12-CU-NEXT:    scratch_load_b32 v0, off, s2 th:TH_LOAD_NT scope:SCOPE_SYS
 ; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
 ; GFX12-CU-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12-CU-NEXT:    s_nop 0



More information about the llvm-commits mailing list