Skip to content

[AMDGPU] MC support for async load and store on gfx1250 #151030

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Jul 28, 2025
Merged

[AMDGPU] MC support for async load and store on gfx1250 #151030

merged 1 commit into from
Jul 28, 2025

Conversation

@changpeng
Copy link
Contributor

@changpeng changpeng commented Jul 28, 2025

No description provided.

@llvmbot llvmbot added backend:AMDGPU llvm:mc Machine (object) code labels Jul 28, 2025
@llvmbot
Copy link
Member

llvmbot commented Jul 28, 2025 •
edited
Loading

@llvm/pr-subscribers-mc

@llvm/pr-subscribers-backend-amdgpu

Author: Changpeng Fang (changpeng)

Changes

Patch is 39.99 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/151030.diff

5 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/FLATInstructions.td (+64-9)
  • (modified) llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp (+8-1)
  • (modified) llvm/test/MC/AMDGPU/gfx1250_asm_vflat.s (+244)
  • (modified) llvm/test/MC/AMDGPU/gfx1250_asm_vflat_err.s (+48)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_vflat.txt (+156)
diff --git a/llvm/lib/Target/AMDGPU/FLATInstructions.td b/llvm/lib/Target/AMDGPU/FLATInstructions.td
index 7207c251994ad..0f172e0ddee56 100644
--- a/llvm/lib/Target/AMDGPU/FLATInstructions.td
+++ b/llvm/lib/Target/AMDGPU/FLATInstructions.td
@@ -369,31 +369,68 @@ multiclass FLAT_Global_Store_Pseudo_t16<string opName> {
   }
 }
 
-class FLAT_Global_Load_LDS_Pseudo <string opName, bit EnableSaddr = 0> : FLAT_Pseudo<
+// Async loads, introduced in gfx1250, will store directly
+// to a DS address in vdst (they will not use M0 for DS addess).
+class FLAT_Global_Load_LDS_Pseudo <string opName, bit EnableSaddr = 0, bit IsAsync = 0> : FLAT_Pseudo<
   opName,
   (outs ),
   !con(
-      !if(EnableSaddr, (ins SReg_64:$saddr, VGPR_32:$vaddr), (ins VReg_64:$vaddr)),
-      (ins flat_offset:$offset, CPol_0:$cpol)),
-  " $vaddr"#!if(EnableSaddr, ", $saddr", ", off")#"$offset$cpol"> {
-  let LGKM_CNT = 1;
+       !if(IsAsync, (ins VGPR_32:$vdst), (ins)),
+       !if(EnableSaddr, (ins SReg_64:$saddr, VGPR_32:$vaddr), (ins VReg_64:$vaddr)),
+       (ins flat_offset:$offset, CPol_0:$cpol)),
+  !if(IsAsync, " $vdst,", "")#" $vaddr"#!if(EnableSaddr, ", $saddr", ", off")#"$offset$cpol"> {
+  let LGKM_CNT = !not(IsAsync);
+  let VM_CNT = !not(IsAsync);
+  let ASYNC_CNT = IsAsync;
   let is_flat_global = 1;
   let lds = 1;
   let has_data = 0;
+  let has_vdst = IsAsync; // vdst for ds address with IsAsync
+  let mayLoad = 1;
+  let mayStore = 1;
+  let has_saddr = 1;
+  let enabled_saddr = EnableSaddr;
+  let VALU = 1;
+  let PseudoInstr = opName#!if(EnableSaddr, "_SADDR", "");
+  let Uses = !if(IsAsync, [EXEC, ASYNCcnt], [M0, EXEC]);
+  let Defs = !if(IsAsync, [ASYNCcnt], []);
+  let SchedRW = [WriteVMEM, WriteLDS];
+}
+
+multiclass FLAT_Global_Load_LDS_Pseudo<string opName, bit IsAsync = 0> {
+  def ""     : FLAT_Global_Load_LDS_Pseudo<opName, 0, IsAsync>,
+    GlobalSaddrTable<0, opName>;
+  def _SADDR : FLAT_Global_Load_LDS_Pseudo<opName, 1, IsAsync>,
+    GlobalSaddrTable<1, opName>;
+}
+
+class FLAT_Global_STORE_LDS_Pseudo <string opName, bit EnableSaddr = 0> : FLAT_Pseudo<
+  opName,
+  (outs ),
+  !con(
+      !if(EnableSaddr, (ins SReg_64:$saddr, VGPR_32:$vaddr), (ins VReg_64:$vaddr)), (ins VGPR_32:$vdata),
+      (ins flat_offset:$offset, CPol_0:$cpol)),
+  " $vaddr, $vdata"#!if(EnableSaddr, ", $saddr", ", off")#"$offset$cpol"> {
+  let VM_CNT = 0;
+  let ASYNC_CNT = 1;
+  let is_flat_global = 1;
+  let lds = 1;
+  let has_data = 1; // vdata for ds address
   let has_vdst = 0;
   let mayLoad = 1;
   let mayStore = 1;
   let has_saddr = 1;
   let enabled_saddr = EnableSaddr;
   let VALU = 1;
-  let Uses = [M0, EXEC];
+  let Uses = [EXEC, ASYNCcnt];
+  let Defs = [ASYNCcnt];
   let SchedRW = [WriteVMEM, WriteLDS];
 }
 
-multiclass FLAT_Global_Load_LDS_Pseudo<string opName> {
-  def ""     : FLAT_Global_Load_LDS_Pseudo<opName>,
+multiclass FLAT_Global_STORE_LDS_Pseudo<string opName> {
+  def ""     : FLAT_Global_STORE_LDS_Pseudo<opName>,
     GlobalSaddrTable<0, opName>;
-  def _SADDR : FLAT_Global_Load_LDS_Pseudo<opName, 1>,
+  def _SADDR : FLAT_Global_STORE_LDS_Pseudo<opName, 1>,
     GlobalSaddrTable<1, opName>;
 }
 
@@ -1156,6 +1193,15 @@ let SubtargetPredicate = isGFX12Plus in {
 
 let SubtargetPredicate = isGFX1250Plus in {
 
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B8       :  FLAT_Global_Load_LDS_Pseudo<"global_load_async_to_lds_b8",    1>;
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B32      :  FLAT_Global_Load_LDS_Pseudo<"global_load_async_to_lds_b32",   1>;
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B64      :  FLAT_Global_Load_LDS_Pseudo<"global_load_async_to_lds_b64",   1>;
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B128     :  FLAT_Global_Load_LDS_Pseudo<"global_load_async_to_lds_b128",  1>;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B8    :  FLAT_Global_STORE_LDS_Pseudo<"global_store_async_from_lds_b8">;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B32   :  FLAT_Global_STORE_LDS_Pseudo<"global_store_async_from_lds_b32">;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B64   :  FLAT_Global_STORE_LDS_Pseudo<"global_store_async_from_lds_b64">;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B128  :  FLAT_Global_STORE_LDS_Pseudo<"global_store_async_from_lds_b128">;
+
 def TENSOR_SAVE : FLAT_Global_Tensor_Pseudo<"tensor_save", 1>;
 def TENSOR_STOP : FLAT_Global_Tensor_Pseudo<"tensor_stop">;
 } // End SubtargetPredicate = isGFX1250Plus
@@ -3374,6 +3420,15 @@ defm GLOBAL_LOAD_MONITOR_B32          : VFLAT_Real_AllAddr_gfx1250<0x070>;
 defm GLOBAL_LOAD_MONITOR_B64          : VFLAT_Real_AllAddr_gfx1250<0x071>;
 defm GLOBAL_LOAD_MONITOR_B128         : VFLAT_Real_AllAddr_gfx1250<0x072>;
 
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B8      : VFLAT_Real_AllAddr_gfx1250<0x5f>;
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B32     : VFLAT_Real_AllAddr_gfx1250<0x60>;
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B64     : VFLAT_Real_AllAddr_gfx1250<0x61>;
+defm GLOBAL_LOAD_ASYNC_TO_LDS_B128    : VFLAT_Real_AllAddr_gfx1250<0x62>;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B8   : VFLAT_Real_AllAddr_gfx1250<0x63>;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B32  : VFLAT_Real_AllAddr_gfx1250<0x64>;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B64  : VFLAT_Real_AllAddr_gfx1250<0x65>;
+defm GLOBAL_STORE_ASYNC_FROM_LDS_B128 : VFLAT_Real_AllAddr_gfx1250<0x66>;
+
 defm GLOBAL_LOAD_TR_B128_w32          : VFLAT_Real_AllAddr_gfx1250<0x057, "global_load_tr16_b128">;
 defm GLOBAL_LOAD_TR_B64_w32           : VFLAT_Real_AllAddr_gfx1250<0x058, "global_load_tr8_b64">;
 
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
index b5b3cc97569ed..83e63ac4e9777 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
@@ -732,7 +732,14 @@ bool isGenericAtomic(unsigned Opc) {
 }
 
 bool isAsyncStore(unsigned Opc) {
-  return false; // placeholder before async store implementation.
+  return Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B8_gfx1250 ||
+         Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B32_gfx1250 ||
+         Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B64_gfx1250 ||
+         Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B128_gfx1250 ||
+         Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B8_SADDR_gfx1250 ||
+         Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B32_SADDR_gfx1250 ||
+         Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B64_SADDR_gfx1250 ||
+         Opc == GLOBAL_STORE_ASYNC_FROM_LDS_B128_SADDR_gfx1250;
 }
 
 bool isTensorStore(unsigned Opc) {
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_vflat.s b/llvm/test/MC/AMDGPU/gfx1250_asm_vflat.s
index b9eb2d2dc8c70..c5288a76e5721 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_vflat.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_vflat.s
@@ -249,6 +249,250 @@ flat_load_monitor_b32 v1, v[2:3] offset:64
 // GFX1250: flat_load_monitor_b32 v1, v[2:3] offset:64 ; encoding: [0x7c,0x00,0x1c,0xec,0x01,0x00,0x00,0x00,0x02,0x40,0x00,0x00]
 // GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
 
+flat_load_monitor_b32 v1, v[2:3] offset:-64 th:TH_LOAD_BYPASS scope:SCOPE_SYS
+// GFX1250: flat_load_monitor_b32 v1, v[2:3] offset:-64 th:TH_LOAD_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0x00,0x1c,0xec,0x01,0x00,0x3c,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b64 v[0:1], v[2:3]
+// GFX1250: flat_load_monitor_b64 v[0:1], v[2:3] ; encoding: [0x7c,0x40,0x1c,0xec,0x00,0x00,0x00,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b64 v[0:1], v[2:3] offset:64
+// GFX1250: flat_load_monitor_b64 v[0:1], v[2:3] offset:64 ; encoding: [0x7c,0x40,0x1c,0xec,0x00,0x00,0x00,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b64 v[0:1], v[2:3] offset:-64 th:TH_LOAD_BYPASS scope:SCOPE_SYS
+// GFX1250: flat_load_monitor_b64 v[0:1], v[2:3] offset:-64 th:TH_LOAD_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0x40,0x1c,0xec,0x00,0x00,0x3c,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b128 v[0:3], v[4:5]
+// GFX1250: flat_load_monitor_b128 v[0:3], v[4:5] ; encoding: [0x7c,0x80,0x1c,0xec,0x00,0x00,0x00,0x00,0x04,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b128 v[0:3], v[4:5] offset:64
+// GFX1250: flat_load_monitor_b128 v[0:3], v[4:5] offset:64 ; encoding: [0x7c,0x80,0x1c,0xec,0x00,0x00,0x00,0x00,0x04,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b128 v[0:3], v[4:5] offset:-64 th:TH_LOAD_BYPASS scope:SCOPE_SYS
+// GFX1250: flat_load_monitor_b128 v[0:3], v[4:5] offset:-64 th:TH_LOAD_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0x80,0x1c,0xec,0x00,0x00,0x3c,0x00,0x04,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b32 v1, v2, s[4:5] offset:64 scale_offset
+// GFX1250: flat_load_monitor_b32 v1, v2, s[4:5] offset:64 scale_offset ; encoding: [0x04,0x00,0x1c,0xec,0x01,0x00,0x01,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+flat_load_monitor_b64 v[2:3], v2, s[4:5] offset:64 scale_offset
+// GFX1250: flat_load_monitor_b64 v[2:3], v2, s[4:5] offset:64 scale_offset ; encoding: [0x04,0x40,0x1c,0xec,0x02,0x00,0x01,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b8 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS
+// GFX1250: global_store_async_from_lds_b8 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0xc0,0x18,0xee,0x00,0x00,0xbc,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b8 v[2:3], v1, off offset:64
+// GFX1250: global_store_async_from_lds_b8 v[2:3], v1, off offset:64 ; encoding: [0x7c,0xc0,0x18,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b8 v[2:3], v1, off offset:-64
+// GFX1250: global_store_async_from_lds_b8 v[2:3], v1, off offset:-64 ; encoding: [0x7c,0xc0,0x18,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b8 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV
+// GFX1250: global_store_async_from_lds_b8 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV ; encoding: [0x02,0xc0,0x18,0xee,0x00,0x00,0xe8,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b8 v2, v1, s[2:3] offset:64
+// GFX1250: global_store_async_from_lds_b8 v2, v1, s[2:3] offset:64 ; encoding: [0x02,0xc0,0x18,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b8 v2, v1, s[2:3] offset:-64
+// GFX1250: global_store_async_from_lds_b8 v2, v1, s[2:3] offset:-64 ; encoding: [0x02,0xc0,0x18,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b32 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS
+// GFX1250: global_store_async_from_lds_b32 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0x00,0x19,0xee,0x00,0x00,0xbc,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b32 v[2:3], v1, off offset:64
+// GFX1250: global_store_async_from_lds_b32 v[2:3], v1, off offset:64 ; encoding: [0x7c,0x00,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b32 v[2:3], v1, off offset:-64
+// GFX1250: global_store_async_from_lds_b32 v[2:3], v1, off offset:-64 ; encoding: [0x7c,0x00,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b32 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV
+// GFX1250: global_store_async_from_lds_b32 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV ; encoding: [0x02,0x00,0x19,0xee,0x00,0x00,0xe8,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b32 v2, v1, s[2:3] offset:64
+// GFX1250: global_store_async_from_lds_b32 v2, v1, s[2:3] offset:64 ; encoding: [0x02,0x00,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b32 v2, v1, s[2:3] offset:-64
+// GFX1250: global_store_async_from_lds_b32 v2, v1, s[2:3] offset:-64 ; encoding: [0x02,0x00,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b64 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS
+// GFX1250: global_store_async_from_lds_b64 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0x40,0x19,0xee,0x00,0x00,0xbc,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b64 v[2:3], v1, off offset:64
+// GFX1250: global_store_async_from_lds_b64 v[2:3], v1, off offset:64 ; encoding: [0x7c,0x40,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b64 v[2:3], v1, off offset:-64
+// GFX1250: global_store_async_from_lds_b64 v[2:3], v1, off offset:-64 ; encoding: [0x7c,0x40,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b64 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV
+// GFX1250: global_store_async_from_lds_b64 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV ; encoding: [0x02,0x40,0x19,0xee,0x00,0x00,0xe8,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b64 v2, v1, s[2:3] offset:64
+// GFX1250: global_store_async_from_lds_b64 v2, v1, s[2:3] offset:64 ; encoding: [0x02,0x40,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b64 v2, v1, s[2:3] offset:-64
+// GFX1250: global_store_async_from_lds_b64 v2, v1, s[2:3] offset:-64 ; encoding: [0x02,0x40,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b128 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS
+// GFX1250: global_store_async_from_lds_b128 v[2:3], v1, off th:TH_STORE_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0x80,0x19,0xee,0x00,0x00,0xbc,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b128 v[2:3], v1, off offset:64
+// GFX1250: global_store_async_from_lds_b128 v[2:3], v1, off offset:64 ; encoding: [0x7c,0x80,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b128 v[2:3], v1, off offset:-64
+// GFX1250: global_store_async_from_lds_b128 v[2:3], v1, off offset:-64 ; encoding: [0x7c,0x80,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b128 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV
+// GFX1250: global_store_async_from_lds_b128 v2, v1, s[2:3] th:TH_STORE_NT_HT scope:SCOPE_DEV ; encoding: [0x02,0x80,0x19,0xee,0x00,0x00,0xe8,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b128 v2, v1, s[2:3] offset:64
+// GFX1250: global_store_async_from_lds_b128 v2, v1, s[2:3] offset:64 ; encoding: [0x02,0x80,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b128 v2, v1, s[2:3] offset:-64
+// GFX1250: global_store_async_from_lds_b128 v2, v1, s[2:3] offset:-64 ; encoding: [0x02,0x80,0x19,0xee,0x00,0x00,0x80,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b32 v2, v1, s[4:5] scale_offset th:TH_STORE_BYPASS scope:SCOPE_SYS
+// GFX1250: global_store_async_from_lds_b32 v2, v1, s[4:5] scale_offset th:TH_STORE_BYPASS scope:SCOPE_SYS ; encoding: [0x04,0x00,0x19,0xee,0x00,0x00,0xbd,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_store_async_from_lds_b64 v2, v1, s[4:5] scale_offset th:TH_STORE_BYPASS scope:SCOPE_SYS
+// GFX1250: global_store_async_from_lds_b64 v2, v1, s[4:5] scale_offset th:TH_STORE_BYPASS scope:SCOPE_SYS ; encoding: [0x04,0x40,0x19,0xee,0x00,0x00,0xbd,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b8 v1, v[2:3], off th:TH_LOAD_BYPASS scope:SCOPE_SYS
+// GFX1250: global_load_async_to_lds_b8 v1, v[2:3], off th:TH_LOAD_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0xc0,0x17,0xee,0x01,0x00,0x3c,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b8 v1, v[2:3], off offset:64
+// GFX1250: global_load_async_to_lds_b8 v1, v[2:3], off offset:64 ; encoding: [0x7c,0xc0,0x17,0xee,0x01,0x00,0x00,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b8 v1, v[2:3], off offset:-64
+// GFX1250: global_load_async_to_lds_b8 v1, v[2:3], off offset:-64 ; encoding: [0x7c,0xc0,0x17,0xee,0x01,0x00,0x00,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b8 v1, v2, s[2:3] th:TH_LOAD_NT_HT scope:SCOPE_DEV
+// GFX1250: global_load_async_to_lds_b8 v1, v2, s[2:3] th:TH_LOAD_NT_HT scope:SCOPE_DEV ; encoding: [0x02,0xc0,0x17,0xee,0x01,0x00,0x68,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b8 v1, v2, s[2:3] offset:64
+// GFX1250: global_load_async_to_lds_b8 v1, v2, s[2:3] offset:64 ; encoding: [0x02,0xc0,0x17,0xee,0x01,0x00,0x00,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b8 v1, v2, s[2:3] offset:-64
+// GFX1250: global_load_async_to_lds_b8 v1, v2, s[2:3] offset:-64 ; encoding: [0x02,0xc0,0x17,0xee,0x01,0x00,0x00,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b32 v1, v[2:3], off th:TH_LOAD_BYPASS scope:SCOPE_SYS
+// GFX1250: global_load_async_to_lds_b32 v1, v[2:3], off th:TH_LOAD_BYPASS scope:SCOPE_SYS ; encoding: [0x7c,0x00,0x18,0xee,0x01,0x00,0x3c,0x00,0x02,0x00,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b32 v1, v[2:3], off offset:64
+// GFX1250: global_load_async_to_lds_b32 v1, v[2:3], off offset:64 ; encoding: [0x7c,0x00,0x18,0xee,0x01,0x00,0x00,0x00,0x02,0x40,0x00,0x00]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b32 v1, v[2:3], off offset:-64
+// GFX1250: global_load_async_to_lds_b32 v1, v[2:3], off offset:-64 ; encoding: [0x7c,0x00,0x18,0xee,0x01,0x00,0x00,0x00,0x02,0xc0,0xff,0xff]
+// GFX12-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+global_load_async_to_lds_b32 v1, v2, s[2:3] th:TH_LOAD_NT_HT scope:SCOPE_DEV
+// GFX1250: global_load_async_to_lds_b32 v1, v2, s[2:3] th:TH_LOAD_NT_HT scope:SCOPE_DEV ; encoding: [0x02,0...
[truncated]

@changpeng changpeng requested review from rampitec and shiltian July 28, 2025 20:01
@changpeng changpeng merged commit 67e2faa into llvm:main Jul 28, 2025
12 checks passed
@changpeng changpeng deleted the async-load-store branch July 28, 2025 20:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@shiltian shiltian shiltian approved these changes

@rampitec rampitec Awaiting requested review from rampitec

Assignees

No one assigned

Labels

backend:AMDGPU llvm:mc Machine (object) code

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@changpeng @llvmbot @shiltian