This is an archive of the discontinued LLVM Phabricator instance.

Differential D155854

[AMDGPU] Add tests for SMEM addressing modes in CodeGenPrepare
ClosedPublic

Authored by foad on Jul 20 2023, 9:06 AM.

Details

Reviewers
arsenm
Group Reviewers
Restricted Project
Commits
rG787bef0bee82: [AMDGPU] Add tests for SMEM addressing modes in CodeGenPrepare

Diff Detail

Unit TestsFailed

TimeTest
80 msx64 windows > Clang.SemaCXX::static-assert-cxx26.cpp

Event Timeline

foad created this revision.Jul 20 2023, 9:06 AM
Herald added a project: Restricted Project. · View Herald TranscriptJul 20 2023, 9:06 AM
Herald added subscribers: StephenFan, kerbowa, tpr and 4 others. · View Herald Transcript
foad requested review of this revision.Jul 20 2023, 9:06 AM
Herald added a project: Restricted Project. · View Herald TranscriptJul 20 2023, 9:06 AM
Herald added subscribers: llvm-commits, wdng. · View Herald Transcript
foad updated this revision to Diff 542557.Jul 20 2023, 9:06 AM

Rebase

foad added a child revision: D155587: [AMDGPU][RFC] Update isLegalAddressingMode for GFX9 SMEM signed offsets.Jul 20 2023, 9:07 AM
Harbormaster completed remote builds in B246941: Diff 542557.Jul 20 2023, 12:54 PM
arsenm accepted this revision.Jul 20 2023, 3:40 PM
This revision is now accepted and ready to land.Jul 20 2023, 3:40 PM
This revision was landed with ongoing or failed builds.Jul 21 2023, 3:04 AM
Closed by commit rG787bef0bee82: [AMDGPU] Add tests for SMEM addressing modes in CodeGenPrepare (authored by foad). · Explain Why
This revision was automatically updated to reflect the committed changes.
foad added a commit: rG787bef0bee82: [AMDGPU] Add tests for SMEM addressing modes in CodeGenPrepare.

Revision Contents

PathSize
llvm/
test/
CodeGen/
AMDGPU/
199 lines

Diff 542557

llvm/test/CodeGen/AMDGPU/cgp-addressing-modes-smem.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc -march=amdgcn -mcpu=gfx600 < %s | FileCheck %s -check-prefixes=GFX6789,GFX678,GFX689,GFX67,GFX6
; RUN: llc -march=amdgcn -mcpu=gfx700 < %s | FileCheck %s -check-prefixes=GFX6789,GFX678,GFX67,GFX7
; RUN: llc -march=amdgcn -mcpu=gfx801 < %s | FileCheck %s -check-prefixes=GFX6789,GFX678,GFX689,GFX89
; RUN: llc -march=amdgcn -mcpu=gfx900 < %s | FileCheck %s -check-prefixes=GFX6789,GFX689,GFX89,GFX9
define amdgpu_cs void @test_sink_smem_offset_400(ptr addrspace(4) inreg %ptr, i32 inreg %val) {
; GFX67-LABEL: test_sink_smem_offset_400:
; GFX67: ; %bb.0: ; %entry
; GFX67-NEXT: .LBB0_1: ; %loop
; GFX67-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX67-NEXT: s_waitcnt lgkmcnt(0)
; GFX67-NEXT: s_load_dword s3, s[0:1], 0x64
; GFX67-NEXT: s_add_i32 s2, s2, -1
; GFX67-NEXT: s_cmp_lg_u32 s2, 0
; GFX67-NEXT: s_cbranch_scc1 .LBB0_1
; GFX67-NEXT: ; %bb.2: ; %end
; GFX67-NEXT: s_endpgm
;
; GFX89-LABEL: test_sink_smem_offset_400:
; GFX89: ; %bb.0: ; %entry
; GFX89-NEXT: .LBB0_1: ; %loop
; GFX89-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX89-NEXT: s_waitcnt lgkmcnt(0)
; GFX89-NEXT: s_load_dword s3, s[0:1], 0x190
; GFX89-NEXT: s_add_i32 s2, s2, -1
; GFX89-NEXT: s_cmp_lg_u32 s2, 0
; GFX89-NEXT: s_cbranch_scc1 .LBB0_1
; GFX89-NEXT: ; %bb.2: ; %end
; GFX89-NEXT: s_endpgm
entry:
%gep = getelementptr i8, ptr addrspace(4) %ptr, i64 400
br label %loop
loop:
%count = phi i32 [ %dec, %loop ], [ %val, %entry ]
%dec = sub i32 %count, 1
%load = load volatile i32, ptr addrspace(4) %gep
%cond = icmp eq i32 %dec, 0
br i1 %cond, label %end, label %loop
end:
ret void
}
define amdgpu_cs void @test_sink_smem_offset_4000(ptr addrspace(4) inreg %ptr, i32 inreg %val) {
; GFX6-LABEL: test_sink_smem_offset_4000:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_add_u32 s0, s0, 0xfa0
; GFX6-NEXT: s_addc_u32 s1, s1, 0
; GFX6-NEXT: .LBB1_1: ; %loop
; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_load_dword s3, s[0:1], 0x0
; GFX6-NEXT: s_add_i32 s2, s2, -1
; GFX6-NEXT: s_cmp_lg_u32 s2, 0
; GFX6-NEXT: s_cbranch_scc1 .LBB1_1
; GFX6-NEXT: ; %bb.2: ; %end
; GFX6-NEXT: s_endpgm
;
; GFX7-LABEL: test_sink_smem_offset_4000:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: .LBB1_1: ; %loop
; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
; GFX7-NEXT: s_load_dword s3, s[0:1], 0x3e8
; GFX7-NEXT: s_add_i32 s2, s2, -1
; GFX7-NEXT: s_cmp_lg_u32 s2, 0
; GFX7-NEXT: s_cbranch_scc1 .LBB1_1
; GFX7-NEXT: ; %bb.2: ; %end
; GFX7-NEXT: s_endpgm
;
; GFX89-LABEL: test_sink_smem_offset_4000:
; GFX89: ; %bb.0: ; %entry
; GFX89-NEXT: .LBB1_1: ; %loop
; GFX89-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX89-NEXT: s_waitcnt lgkmcnt(0)
; GFX89-NEXT: s_load_dword s3, s[0:1], 0xfa0
; GFX89-NEXT: s_add_i32 s2, s2, -1
; GFX89-NEXT: s_cmp_lg_u32 s2, 0
; GFX89-NEXT: s_cbranch_scc1 .LBB1_1
; GFX89-NEXT: ; %bb.2: ; %end
; GFX89-NEXT: s_endpgm
entry:
%gep = getelementptr i8, ptr addrspace(4) %ptr, i64 4000
br label %loop
loop:
%count = phi i32 [ %dec, %loop ], [ %val, %entry ]
%dec = sub i32 %count, 1
%load = load volatile i32, ptr addrspace(4) %gep
%cond = icmp eq i32 %dec, 0
br i1 %cond, label %end, label %loop
end:
ret void
}
define amdgpu_cs void @test_sink_smem_offset_4000000(ptr addrspace(4) inreg %ptr, i32 inreg %val) {
; GFX689-LABEL: test_sink_smem_offset_4000000:
; GFX689: ; %bb.0: ; %entry
; GFX689-NEXT: s_add_u32 s0, s0, 0x3d0900
; GFX689-NEXT: s_addc_u32 s1, s1, 0
; GFX689-NEXT: .LBB2_1: ; %loop
; GFX689-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX689-NEXT: s_waitcnt lgkmcnt(0)
; GFX689-NEXT: s_load_dword s3, s[0:1], 0x0
; GFX689-NEXT: s_add_i32 s2, s2, -1
; GFX689-NEXT: s_cmp_lg_u32 s2, 0
; GFX689-NEXT: s_cbranch_scc1 .LBB2_1
; GFX689-NEXT: ; %bb.2: ; %end
; GFX689-NEXT: s_endpgm
;
; GFX7-LABEL: test_sink_smem_offset_4000000:
; GFX7: ; %bb.0: ; %entry
; GFX7-NEXT: .LBB2_1: ; %loop
; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
; GFX7-NEXT: s_load_dword s3, s[0:1], 0xf4240
; GFX7-NEXT: s_add_i32 s2, s2, -1
; GFX7-NEXT: s_cmp_lg_u32 s2, 0
; GFX7-NEXT: s_cbranch_scc1 .LBB2_1
; GFX7-NEXT: ; %bb.2: ; %end
; GFX7-NEXT: s_endpgm
entry:
%gep = getelementptr i8, ptr addrspace(4) %ptr, i64 4000000
br label %loop
loop:
%count = phi i32 [ %dec, %loop ], [ %val, %entry ]
%dec = sub i32 %count, 1
%load = load volatile i32, ptr addrspace(4) %gep
%cond = icmp eq i32 %dec, 0
br i1 %cond, label %end, label %loop
end:
ret void
}
define amdgpu_cs void @test_sink_smem_offset_40000000000(ptr addrspace(4) inreg %ptr, i32 inreg %val) {
; GFX6789-LABEL: test_sink_smem_offset_40000000000:
; GFX6789: ; %bb.0: ; %entry
; GFX6789-NEXT: s_add_u32 s0, s0, 0x502f9000
; GFX6789-NEXT: s_addc_u32 s1, s1, 9
; GFX6789-NEXT: .LBB3_1: ; %loop
; GFX6789-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX6789-NEXT: s_waitcnt lgkmcnt(0)
; GFX6789-NEXT: s_load_dword s3, s[0:1], 0x0
; GFX6789-NEXT: s_add_i32 s2, s2, -1
; GFX6789-NEXT: s_cmp_lg_u32 s2, 0
; GFX6789-NEXT: s_cbranch_scc1 .LBB3_1
; GFX6789-NEXT: ; %bb.2: ; %end
; GFX6789-NEXT: s_endpgm
entry:
%gep = getelementptr i8, ptr addrspace(4) %ptr, i64 40000000000
br label %loop
loop:
%count = phi i32 [ %dec, %loop ], [ %val, %entry ]
%dec = sub i32 %count, 1
%load = load volatile i32, ptr addrspace(4) %gep
%cond = icmp eq i32 %dec, 0
br i1 %cond, label %end, label %loop
end:
ret void
}
define amdgpu_cs void @test_sink_smem_offset_neg400(ptr addrspace(4) inreg %ptr, i32 inreg %val) {
; GFX6789-LABEL: test_sink_smem_offset_neg400:
; GFX6789: ; %bb.0: ; %entry
; GFX6789-NEXT: s_add_u32 s0, s0, 0xfffffe70
; GFX6789-NEXT: s_addc_u32 s1, s1, -1
; GFX6789-NEXT: .LBB4_1: ; %loop
; GFX6789-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX6789-NEXT: s_waitcnt lgkmcnt(0)
; GFX6789-NEXT: s_load_dword s3, s[0:1], 0x0
; GFX6789-NEXT: s_add_i32 s2, s2, -1
; GFX6789-NEXT: s_cmp_lg_u32 s2, 0
; GFX6789-NEXT: s_cbranch_scc1 .LBB4_1
; GFX6789-NEXT: ; %bb.2: ; %end
; GFX6789-NEXT: s_endpgm
entry:
%gep = getelementptr i8, ptr addrspace(4) %ptr, i64 -400
br label %loop
loop:
%count = phi i32 [ %dec, %loop ], [ %val, %entry ]
%dec = sub i32 %count, 1
%load = load volatile i32, ptr addrspace(4) %gep
%cond = icmp eq i32 %dec, 0
br i1 %cond, label %end, label %loop
end:
ret void
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GFX678: {{.*}}
; GFX9: {{.*}}