caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
121cd7c6f0270540ce976b98061ba765edc1675d
clang-p2996/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll
Brox Chen 8bc0f879a0 [AMDGPU][True16][CodeGen] D16 LDS load/store pseudo instructions in true16 (#131427) 
Implement new pseudos with the suffix _t16 which have VGPR_16 as the
store src or load dst. This affects LDS 8 and 16-bit loads and stores.
Lower the pseudos to the existing real Hi/Lo instructions in MC inst
layer with VGPR_32 src or dst

---------

Co-authored-by: Abhinav <abhinav.garg@amd.com>
2025-03-17 10:28:45 -04:00

471 lines
18 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,CI %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s
define void @atomic_store_monotonic_i8(ptr addrspace(3) %ptr, i8 %val) {
; CI-LABEL: atomic_store_monotonic_i8:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b8 v0, v1
; CI-NEXT: ds_write_b8 v0, v2
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_i8:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b8 v0, v1
; GFX9-NEXT: ds_write_b8 v0, v2
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_i8:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b8 v0, v1
; GFX11-TRUE16-NEXT: ds_store_b8_d16_hi v0, v1
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_i8:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b8 v0, v1
; GFX11-FAKE16-NEXT: ds_store_b8 v0, v2
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%val1 = add i8 %val, 2
store atomic i8 %val, ptr addrspace(3) %ptr monotonic, align 1
store atomic i8 %val1, ptr addrspace(3) %ptr monotonic, align 1
ret void
}
define void @atomic_store_monotonic_offset_i8(ptr addrspace(3) %ptr, i8 %val) {
; CI-LABEL: atomic_store_monotonic_offset_i8:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b8 v0, v1 offset:8
; CI-NEXT: ds_write_b8 v0, v2 offset:16
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_offset_i8:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b8 v0, v1 offset:8
; GFX9-NEXT: ds_write_b8 v0, v2 offset:16
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_offset_i8:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b8 v0, v1 offset:8
; GFX11-TRUE16-NEXT: ds_store_b8_d16_hi v0, v1 offset:16
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_offset_i8:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b8 v0, v1 offset:8
; GFX11-FAKE16-NEXT: ds_store_b8 v0, v2 offset:16
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%val1 = add i8 %val, 2
%gep_1 = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 8
%gep_2 = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 16
store atomic i8 %val, ptr addrspace(3) %gep_1 monotonic, align 1
store atomic i8 %val1, ptr addrspace(3) %gep_2 monotonic, align 1
ret void
}
define void @atomic_store_monotonic_i16(ptr addrspace(3) %ptr, i16 %val) {
; CI-LABEL: atomic_store_monotonic_i16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b16 v0, v1
; CI-NEXT: ds_write_b16 v0, v2
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b16 v0, v1
; GFX9-NEXT: ds_write_b16 v0, v2
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_i16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b16 v0, v1
; GFX11-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_i16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v1
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%val1 = add i16 %val, 2
store atomic i16 %val, ptr addrspace(3) %ptr monotonic, align 2
store atomic i16 %val1, ptr addrspace(3) %ptr monotonic, align 2
ret void
}
define void @atomic_store_monotonic_offset_i16(ptr addrspace(3) %ptr, i16 %val) {
; CI-LABEL: atomic_store_monotonic_offset_i16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b16 v0, v1 offset:32
; CI-NEXT: ds_write_b16 v0, v2 offset:32
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_offset_i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b16 v0, v1 offset:32
; GFX9-NEXT: ds_write_b16 v0, v2 offset:32
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_offset_i16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32
; GFX11-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_offset_i16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%val1 = add i16 %val, 2
%gep = getelementptr inbounds i16, ptr addrspace(3) %ptr, i16 16
store atomic i16 %val, ptr addrspace(3) %gep monotonic, align 2
store atomic i16 %val1, ptr addrspace(3) %gep monotonic, align 2
ret void
}
define void @atomic_store_monotonic_i32(ptr addrspace(3) %ptr, i32 %val) {
; CI-LABEL: atomic_store_monotonic_i32:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: ds_write_b32 v0, v1
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_i32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: ds_write_b32 v0, v1
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: atomic_store_monotonic_i32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: ds_store_b32 v0, v1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
store atomic i32 %val, ptr addrspace(3) %ptr monotonic, align 4
ret void
}
define void @atomic_store_monotonic_offset_i32(ptr addrspace(3) %ptr, i32 %val) {
; CI-LABEL: atomic_store_monotonic_offset_i32:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: ds_write_b32 v0, v1 offset:64
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_offset_i32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: ds_write_b32 v0, v1 offset:64
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: atomic_store_monotonic_offset_i32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: ds_store_b32 v0, v1 offset:64
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr inbounds i32, ptr addrspace(3) %ptr, i32 16
store atomic i32 %val, ptr addrspace(3) %gep monotonic, align 4
ret void
}
define void @atomic_store_monotonic_i64(ptr addrspace(3) %ptr, i64 %val) {
; CI-LABEL: atomic_store_monotonic_i64:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: ds_write_b64 v0, v[1:2]
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_i64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: ds_write_b64 v0, v[1:2]
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: atomic_store_monotonic_i64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: ds_store_b64 v0, v[1:2]
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
store atomic i64 %val, ptr addrspace(3) %ptr monotonic, align 8
ret void
}
define void @atomic_store_monotonic_offset_i64(ptr addrspace(3) %ptr, i64 %val) {
; CI-LABEL: atomic_store_monotonic_offset_i64:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: ds_write_b64 v0, v[1:2] offset:128
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_offset_i64:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: ds_write_b64 v0, v[1:2] offset:128
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: atomic_store_monotonic_offset_i64:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: ds_store_b64 v0, v[1:2] offset:128
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr inbounds i64, ptr addrspace(3) %ptr, i64 16
store atomic i64 %val, ptr addrspace(3) %gep monotonic, align 8
ret void
}
define void @atomic_store_monotonic_f16(ptr addrspace(3) %ptr, i16 %arg.val) {
; CI-LABEL: atomic_store_monotonic_f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b16 v0, v1
; CI-NEXT: ds_write_b16 v0, v2
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b16 v0, v1
; GFX9-NEXT: ds_write_b16 v0, v2
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_f16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b16 v0, v1
; GFX11-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_f16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v1
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%arg.val1 = add i16 %arg.val, 2
%val = bitcast i16 %arg.val to half
%val1 = bitcast i16 %arg.val1 to half
store atomic half %val, ptr addrspace(3) %ptr monotonic, align 2
store atomic half %val1, ptr addrspace(3) %ptr monotonic, align 2
ret void
}
define void @atomic_store_monotonic_offset_f16(ptr addrspace(3) %ptr, i16 %arg.val) {
; CI-LABEL: atomic_store_monotonic_offset_f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b16 v0, v1 offset:32
; CI-NEXT: ds_write_b16 v0, v2 offset:32
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_offset_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b16 v0, v1 offset:32
; GFX9-NEXT: ds_write_b16 v0, v2 offset:32
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_offset_f16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32
; GFX11-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_offset_f16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%arg.val1 = add i16 %arg.val, 2
%val1 = bitcast i16 %arg.val1 to half
%val = bitcast i16 %arg.val to half
%gep = getelementptr inbounds half, ptr addrspace(3) %ptr, i32 16
store atomic half %val, ptr addrspace(3) %gep monotonic, align 2
store atomic half %val1, ptr addrspace(3) %gep monotonic, align 2
ret void
}
define void @atomic_store_monotonic_bf16(ptr addrspace(3) %ptr, i16 %arg.val) {
; CI-LABEL: atomic_store_monotonic_bf16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b16 v0, v1
; CI-NEXT: ds_write_b16 v0, v2
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_bf16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b16 v0, v1
; GFX9-NEXT: ds_write_b16 v0, v2
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_bf16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b16 v0, v1
; GFX11-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_bf16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v1
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%arg.val1 = add i16 %arg.val, 2
%val1 = bitcast i16 %arg.val1 to bfloat
%val = bitcast i16 %arg.val to bfloat
store atomic bfloat %val, ptr addrspace(3) %ptr monotonic, align 2
store atomic bfloat %val1, ptr addrspace(3) %ptr monotonic, align 2
ret void
}
define void @atomic_store_monotonic_offset_bf16(ptr addrspace(3) %ptr, i16 %arg.val) {
; CI-LABEL: atomic_store_monotonic_offset_bf16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 m0, -1
; CI-NEXT: v_add_i32_e32 v2, vcc, 2, v1
; CI-NEXT: ds_write_b16 v0, v1 offset:32
; CI-NEXT: ds_write_b16 v0, v2 offset:32
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: atomic_store_monotonic_offset_bf16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u16_e32 v2, 2, v1
; GFX9-NEXT: ds_write_b16 v0, v1 offset:32
; GFX9-NEXT: ds_write_b16 v0, v2 offset:32
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: atomic_store_monotonic_offset_bf16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2
; GFX11-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32
; GFX11-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: atomic_store_monotonic_offset_bf16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32
; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%arg.val1 = add i16 %arg.val, 2
%val1 = bitcast i16 %arg.val1 to bfloat
%val = bitcast i16 %arg.val to bfloat
%gep = getelementptr inbounds bfloat, ptr addrspace(3) %ptr, i32 16
store atomic bfloat %val, ptr addrspace(3) %gep monotonic, align 2
store atomic bfloat %val1, ptr addrspace(3) %gep monotonic, align 2
ret void
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GCN: {{.*}}
Reference in New Issue View Git Blame Copy Permalink