11# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
22# RUN: llc -x mir -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1102 -run-pass=machine-scheduler %s -o - | FileCheck %s
33
4- --- |
5- declare void @llvm.amdgcn.s.waitcnt(i32 immarg)
6-
7- declare <2 x i32> @llvm.amdgcn.raw.buffer.load.v2i32(<4 x i32>, i32, i32, i32 immarg)
8-
9- define amdgpu_kernel void @foo(ptr %p) {
10- entry :
11- %foo.kernarg.segment = call nonnull align 16 dereferenceable(264) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
12- %p.kernarg.offset1 = bitcast ptr addrspace(4) %foo.kernarg.segment to ptr addrspace(4)
13- %p.load = load ptr, ptr addrspace(4) %p.kernarg.offset1, align 16
14- %call = tail call <2 x i32> @llvm.amdgcn.raw.buffer.load.v2i32(<4 x i32> zeroinitializer, i32 0, i32 0, i32 0)
15- %cast = bitcast <2 x i32> %call to <8 x i8>
16- %shuffle = shufflevector <8 x i8> zeroinitializer, <8 x i8> %cast, <2 x i32> <i32 3, i32 11>
17- %zext = zext <2 x i8> %shuffle to <2 x i16>
18- %shl = shl <2 x i16> %zext, splat (i16 8)
19- store <2 x i16> %shl, ptr %p.load, align 4
20- tail call void @llvm.amdgcn.s.waitcnt(i32 0)
21- ret void
22- }
23-
24- declare noundef align 4 ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
25- ...
264---
275name : foo
286tracksRegLiveness : true
297liveins :
30- - { reg: '$sgpr4_sgpr5', virtual-reg: '%3 ' }
8+ - { reg: '$sgpr4_sgpr5', virtual-reg: '%0 ' }
319body : |
3210 bb.0.entry:
3311 liveins: $sgpr4_sgpr5
@@ -42,29 +20,29 @@ body: |
4220 ; CHECK-NEXT: [[COPY1:%[0-9]+]].sub2:sgpr_128 = COPY [[S_MOV_B32_]]
4321 ; CHECK-NEXT: [[COPY1:%[0-9]+]].sub3:sgpr_128 = COPY [[S_MOV_B32_]]
4422 ; CHECK-NEXT: [[BUFFER_LOAD_DWORDX2_OFFSET:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_OFFSET [[COPY1]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 8)
45- ; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 0, 0 :: (dereferenceable invariant load (s64) from %ir.p.kernarg.offset1 , align 16, addrspace 4)
23+ ; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 0, 0 :: (dereferenceable invariant load (s64), align 16, addrspace 4)
4624 ; CHECK-NEXT: [[V_LSHRREV_B64_e64_:%[0-9]+]]:vreg_64 = V_LSHRREV_B64_e64 24, [[BUFFER_LOAD_DWORDX2_OFFSET]], implicit $exec
4725 ; CHECK-NEXT: undef [[COPY2:%[0-9]+]].lo16:vgpr_32 = COPY [[V_LSHRREV_B64_e64_]].lo16
4826 ; CHECK-NEXT: [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 16, [[COPY2]], implicit $exec
4927 ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[S_LOAD_DWORDX2_IMM]]
5028 ; CHECK-NEXT: [[V_PK_LSHLREV_B16_:%[0-9]+]]:vgpr_32 = V_PK_LSHLREV_B16 0, 8, 8, [[V_LSHLREV_B32_e64_]], 0, 0, 0, 0, 0, implicit $exec
51- ; CHECK-NEXT: FLAT_STORE_DWORD [[COPY3]], [[V_PK_LSHLREV_B16_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %ir.p.load )
29+ ; CHECK-NEXT: FLAT_STORE_DWORD [[COPY3]], [[V_PK_LSHLREV_B16_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
5230 ; CHECK-NEXT: S_WAITCNT 0
5331 ; CHECK-NEXT: S_ENDPGM 0
54- %3 :sgpr_64(p4) = COPY killed $sgpr4_sgpr5
55- %13 :sreg_64_xexec = S_LOAD_DWORDX2_IMM killed %3 (p4), 0, 0 :: (dereferenceable invariant load (s64) from %ir.p.kernarg.offset1 , align 16, addrspace 4)
56- %14 :sreg_32 = S_MOV_B32 0
57- undef %15 .sub0:sgpr_128 = COPY %14
58- %15 .sub1:sgpr_128 = COPY %14
59- %15 .sub2:sgpr_128 = COPY %14
60- %15 .sub3:sgpr_128 = COPY killed %14
61- %16 :vreg_64 = BUFFER_LOAD_DWORDX2_OFFSET killed %15 , 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 8)
62- %26 :vreg_64 = V_LSHRREV_B64_e64 24, killed %16 , implicit $exec
63- undef %28 .lo16:vgpr_32 = COPY killed %26 .lo16
64- %30 :vgpr_32 = V_LSHLREV_B32_e64 16, killed %28 , implicit $exec
65- %24 :vgpr_32 = V_PK_LSHLREV_B16 0, 8, 8, killed %30 , 0, 0, 0, 0, 0, implicit $exec
66- %25 :vreg_64 = COPY killed %13
67- FLAT_STORE_DWORD killed %25 , killed %24 , 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %ir.p.load )
32+ %0 :sgpr_64(p4) = COPY killed $sgpr4_sgpr5
33+ %1 :sreg_64_xexec = S_LOAD_DWORDX2_IMM killed %0 (p4), 0, 0 :: (dereferenceable invariant load (s64), align 16, addrspace 4)
34+ %2 :sreg_32 = S_MOV_B32 0
35+ undef %3 .sub0:sgpr_128 = COPY %2
36+ %3 .sub1:sgpr_128 = COPY %2
37+ %3 .sub2:sgpr_128 = COPY %2
38+ %3 .sub3:sgpr_128 = COPY killed %2
39+ %4 :vreg_64 = BUFFER_LOAD_DWORDX2_OFFSET killed %3 , 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 8)
40+ %5 :vreg_64 = V_LSHRREV_B64_e64 24, killed %4 , implicit $exec
41+ undef %6 .lo16:vgpr_32 = COPY killed %5 .lo16
42+ %7 :vgpr_32 = V_LSHLREV_B32_e64 16, killed %6 , implicit $exec
43+ %8 :vgpr_32 = V_PK_LSHLREV_B16 0, 8, 8, killed %7 , 0, 0, 0, 0, 0, implicit $exec
44+ %9 :vreg_64 = COPY killed %1
45+ FLAT_STORE_DWORD killed %9 , killed %8 , 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
6846 S_WAITCNT 0
6947 S_ENDPGM 0
7048 ...
0 commit comments