Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / e22fdb9cac18f1ddc4900804e6e626027b5579af / . / llvm / test / CodeGen / AMDGPU / large-alloca-compute.ll
blob: 4f6dbf9dc2bf7db0e340fc61a129b0d0cba67054 [file] [log] [blame]
Tom Stellardf3af8412016-06-10 19:26:38 +0000[diff] [blame]1; RUN: llc -march=amdgcn -mcpu=bonaire -show-mc-encoding < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=ALL %s
2; RUN: llc -march=amdgcn -mcpu=carrizo --show-mc-encoding < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=ALL %s
Changpeng Fangb41574a2015-12-22 20:55:23 +0000[diff] [blame]3; RUN: llc -march=amdgcn -mcpu=bonaire -mtriple=amdgcn-unknown-amdhsa < %s -mattr=-flat-for-global | FileCheck -check-prefix=GCNHSA -check-prefix=CIHSA -check-prefix=ALL %s
Nicolai Haehnle60355042016-01-05 20:42:49 +0000[diff] [blame]4; RUN: llc -march=amdgcn -mcpu=carrizo -mtriple=amdgcn-unknown-amdhsa -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCNHSA -check-prefix=VIHSA -check-prefix=ALL %s
Matt Arsenault0e3d3892015-11-30 21:15:53 +0000[diff] [blame]5
6; FIXME: align on alloca seems to be ignored for private_segment_alignment
7
8; ALL-LABEL: {{^}}large_alloca_compute_shader:
9
Matt Arsenaulte8ed8e52016-05-11 00:28:54 +0000[diff] [blame]10; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
Tom Stellard1c89eb72016-06-20 16:59:44 +0000[diff] [blame]11; GCN-DAG: ; fixup A - offset: 4, value: SCRATCH_RSRC_DWORD0
Matt Arsenaulte8ed8e52016-05-11 00:28:54 +0000[diff] [blame]12; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
Tom Stellard1c89eb72016-06-20 16:59:44 +0000[diff] [blame]13; GCN-DAG: ; fixup A - offset: 4, value: SCRATCH_RSRC_DWORD1
Matt Arsenaulte8ed8e52016-05-11 00:28:54 +0000[diff] [blame]14; GCN-DAG: s_mov_b32 s{{[0-9]+}}, -1
Marek Olsake93f6d62016-06-13 16:05:57 +0000[diff] [blame]15; CI-DAG: s_mov_b32 s{{[0-9]+}}, 0xe8f000
16; VI-DAG: s_mov_b32 s{{[0-9]+}}, 0xe80000
Matt Arsenault0e3d3892015-11-30 21:15:53 +0000[diff] [blame]17
18
19; GCNHSA: .amd_kernel_code_t
Matt Arsenault26f8f3d2015-11-30 21:16:03 +0000[diff] [blame]20
Sam Koltona2e5c882016-09-09 10:08:02 +0000[diff] [blame]21; GCNHSA: enable_sgpr_private_segment_wave_byte_offset = 1
22; GCNHSA: user_sgpr_count = 8
23; GCNHSA: enable_sgpr_workgroup_id_x = 1
24; GCNHSA: enable_sgpr_workgroup_id_y = 0
25; GCNHSA: enable_sgpr_workgroup_id_z = 0
26; GCNHSA: enable_sgpr_workgroup_info = 0
27; GCNHSA: enable_vgpr_workitem_id = 0
Matt Arsenault26f8f3d2015-11-30 21:16:03 +0000[diff] [blame]28
29; GCNHSA: enable_sgpr_private_segment_buffer = 1
30; GCNHSA: enable_sgpr_dispatch_ptr = 0
31; GCNHSA: enable_sgpr_queue_ptr = 0
32; GCNHSA: enable_sgpr_kernarg_segment_ptr = 1
33; GCNHSA: enable_sgpr_dispatch_id = 0
Matt Arsenault296b8492016-02-12 06:31:30 +0000[diff] [blame]34; GCNHSA: enable_sgpr_flat_scratch_init = 1
Matt Arsenault26f8f3d2015-11-30 21:16:03 +0000[diff] [blame]35; GCNHSA: enable_sgpr_private_segment_size = 0
36; GCNHSA: enable_sgpr_grid_workgroup_count_x = 0
37; GCNHSA: enable_sgpr_grid_workgroup_count_y = 0
38; GCNHSA: enable_sgpr_grid_workgroup_count_z = 0
Tom Stellarda4953072015-12-15 22:55:30 +0000[diff] [blame]39; GCNHSA: workitem_private_segment_byte_size = 32772
Matt Arsenault0e3d3892015-11-30 21:15:53 +0000[diff] [blame]40; GCNHSA: private_segment_alignment = 4
41; GCNHSA: .end_amd_kernel_code_t
42
Matt Arsenault0e3d3892015-11-30 21:15:53 +0000[diff] [blame]43
Matt Arsenault296b8492016-02-12 06:31:30 +0000[diff] [blame]44; GCNHSA: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], s9 offen
45; GCNHSA: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], s9 offen
Matt Arsenault0e3d3892015-11-30 21:15:53 +0000[diff] [blame]46
47; Scratch size = alloca size + emergency stack slot
48; ALL: ; ScratchSize: 32772
49define void @large_alloca_compute_shader(i32 %x, i32 %y) #0 {
50 %large = alloca [8192 x i32], align 4
51 %gep = getelementptr [8192 x i32], [8192 x i32]* %large, i32 0, i32 8191
52 store volatile i32 %x, i32* %gep
53 %gep1 = getelementptr [8192 x i32], [8192 x i32]* %large, i32 0, i32 %y
54 %val = load volatile i32, i32* %gep1
55 store volatile i32 %val, i32 addrspace(1)* undef
56 ret void
57}
58
59attributes #0 = { nounwind }