shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tensor.load.store.ll
Changpeng Fang eda3161c35
AMDGPU: Implement tensor load and store instructions for gfx1250 (#146636) 
Co-authored-by: Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com>
2025-07-03 13:49:34 -07:00

246 lines
12 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s
declare void @llvm.amdgcn.tensor.load.to.lds(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3, i32 %cpol)
declare void @llvm.amdgcn.tensor.load.to.lds.d2(<4 x i32> %D0, <8 x i32> %D1, i32 %cpol)
declare void @llvm.amdgcn.tensor.store.from.lds(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3, i32 %cpol)
declare void @llvm.amdgcn.tensor.store.from.lds.d2(<4 x i32> %D0, <8 x i32> %D1, i32 %cpol)
define amdgpu_ps void @tensor_load_to_lds(<4 x i32> inreg %D0, <8 x i32> inreg %D1, <4 x i32> inreg %D2, <4 x i32> inreg %D3) {
; GFX1250-LABEL: tensor_load_to_lds:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: tensor_load_to_lds s[0:3], s[4:11], s[12:15], s[16:19]
; GFX1250-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.load.to.lds(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3, i32 0)
ret void
}
define amdgpu_ps void @tensor_load_to_lds_vector(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3) {
; GFX1250-SDAG-LABEL: tensor_load_to_lds_vector:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s12, v12
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s13, v13
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s14, v14
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s15, v15
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s16, v16
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s17, v17
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s18, v18
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s19, v19
; GFX1250-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-SDAG-NEXT: tensor_load_to_lds s[8:11], s[0:7], s[12:15], s[16:19]
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: tensor_load_to_lds_vector:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s12, v12
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s13, v13
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s14, v14
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s15, v15
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s16, v16
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s17, v17
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s18, v18
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s19, v19
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: tensor_load_to_lds s[8:11], s[0:7], s[12:15], s[16:19]
; GFX1250-GISEL-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.load.to.lds(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3, i32 0)
ret void
}
define amdgpu_ps void @tensor_load_to_lds_d2(<4 x i32> inreg %D0, <8 x i32> inreg %D1) {
; GFX1250-LABEL: tensor_load_to_lds_d2:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: tensor_load_to_lds s[0:3], s[4:11] th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.load.to.lds.d2(<4 x i32> %D0, <8 x i32> %D1, i32 27)
ret void
}
define amdgpu_ps void @tensor_load_to_lds_d2_vector(<4 x i32> %D0, <8 x i32> %D1) {
; GFX1250-SDAG-LABEL: tensor_load_to_lds_d2_vector:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-SDAG-NEXT: tensor_load_to_lds s[8:11], s[0:7] th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: tensor_load_to_lds_d2_vector:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: tensor_load_to_lds s[8:11], s[0:7] th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-GISEL-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.load.to.lds.d2(<4 x i32> %D0, <8 x i32> %D1, i32 27)
ret void
}
define amdgpu_ps void @tensor_store_from_lds(<4 x i32> inreg %D0, <8 x i32> inreg %D1, <4 x i32> inreg %D2, <4 x i32> inreg %D3) {
; GFX1250-LABEL: tensor_store_from_lds:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: tensor_store_from_lds s[0:3], s[4:11], s[12:15], s[16:19] th:TH_STORE_NT_HT scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.store.from.lds(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3, i32 22)
ret void
}
define amdgpu_ps void @tensor_store_from_lds_vector(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3) {
; GFX1250-SDAG-LABEL: tensor_store_from_lds_vector:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s12, v12
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s13, v13
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s14, v14
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s15, v15
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s16, v16
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s17, v17
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s18, v18
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s19, v19
; GFX1250-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-SDAG-NEXT: tensor_store_from_lds s[8:11], s[0:7], s[12:15], s[16:19] th:TH_STORE_NT_HT scope:SCOPE_DEV
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: tensor_store_from_lds_vector:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s12, v12
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s13, v13
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s14, v14
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s15, v15
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s16, v16
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s17, v17
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s18, v18
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s19, v19
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: tensor_store_from_lds s[8:11], s[0:7], s[12:15], s[16:19] th:TH_STORE_NT_HT scope:SCOPE_DEV
; GFX1250-GISEL-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.store.from.lds(<4 x i32> %D0, <8 x i32> %D1, <4 x i32> %D2, <4 x i32> %D3, i32 22)
ret void
}
define amdgpu_ps void @tensor_store_from_lds_d2(<4 x i32> inreg %D0, <8 x i32> inreg %D1) {
; GFX1250-LABEL: tensor_store_from_lds_d2:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: tensor_store_from_lds s[0:3], s[4:11]
; GFX1250-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.store.from.lds.d2(<4 x i32> %D0, <8 x i32> %D1, i32 0)
ret void
}
define amdgpu_ps void @tensor_store_from_lds_d2_vector(<4 x i32> %D0, <8 x i32> %D1) {
; GFX1250-SDAG-LABEL: tensor_store_from_lds_d2_vector:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-SDAG-NEXT: tensor_store_from_lds s[8:11], s[0:7]
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: tensor_store_from_lds_d2_vector:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s8, v0
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s9, v1
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s10, v2
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s11, v3
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v4
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s1, v5
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v6
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s3, v7
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s4, v8
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s5, v9
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s6, v10
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s7, v11
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-GISEL-NEXT: tensor_store_from_lds s[8:11], s[0:7]
; GFX1250-GISEL-NEXT: s_endpgm
entry:
call void @llvm.amdgcn.tensor.store.from.lds.d2(<4 x i32> %D0, <8 x i32> %D1, i32 0)
ret void
}
Reference in New Issue View Git Blame Copy Permalink