RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-24 05:23:45 +02:00
Code Issues Projects Releases Wiki Activity
f028d10a98
llvm-mirror/test/CodeGen/AMDGPU/unaligned-load-store.ll

605 lines
17 KiB
LLVM
Raw Normal View History

AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; RUN: llc -march=amdgcn -verify-machineinstrs< %s | FileCheck -check-prefix=SI -check-prefix=ALIGNED %s
; RUN: llc -march=amdgcn -mcpu=bonaire -mattr=+unaligned-buffer-access -verify-machineinstrs< %s | FileCheck -check-prefix=SI -check-prefix=UNALIGNED %s
Enable FeatureFlatForGlobal on Volcanic Islands This switches to the workaround that HSA defaults to for the mesa path. This should be applied to the 4.0 branch. Patch by Vedran Miletić <vedran@miletic.net> llvm-svn: 292982
2017-01-24 23:02:15 +01:00
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -check-prefix=SI -check-prefix=ALIGNED %s
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; SI-LABEL: {{^}}local_unaligned_load_store_i16:
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_unaligned_load_store_i16(i16 addrspace(3)* %p, i16 addrspace(3)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load i16, i16 addrspace(3)* %p, align 1
R600/SI: Expand misaligned 16-bit memory accesses llvm-svn: 228190
2015-02-04 21:49:52 +01:00
store i16 %v, i16 addrspace(3)* %r, align 1
ret void
}
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; SI-LABEL: {{^}}global_unaligned_load_store_i16:
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; UNALIGNED: buffer_load_ushort
; UNALIGNED: buffer_store_short
; SI: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @global_unaligned_load_store_i16(i16 addrspace(1)* %p, i16 addrspace(1)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load i16, i16 addrspace(1)* %p, align 1
R600/SI: Expand misaligned 16-bit memory accesses llvm-svn: 228190
2015-02-04 21:49:52 +01:00
store i16 %v, i16 addrspace(1)* %r, align 1
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_unaligned_load_store_i32:
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI-NOT: v_or
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_unaligned_load_store_i32(i32 addrspace(3)* %p, i32 addrspace(3)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load i32, i32 addrspace(3)* %p, align 1
Fix CodeGen for unaligned loads with address spaces llvm-svn: 193721
2013-10-31 00:30:05 +01:00
store i32 %v, i32 addrspace(3)* %r, align 1
ret void
}
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; SI-LABEL: {{^}}global_unaligned_load_store_i32:
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; UNALIGNED: buffer_load_dword
; UNALIGNED: buffer_store_dword
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @global_unaligned_load_store_i32(i32 addrspace(1)* %p, i32 addrspace(1)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load i32, i32 addrspace(1)* %p, align 1
R600/SI: 64-bit and larger memory access must be at least 4-byte aligned This is true for SI only. CI+ supports unaligned memory accesses, but this requires driver support, so for now we disallow unaligned accesses for all GCN targets. llvm-svn: 227822
2015-02-02 19:02:28 +01:00
store i32 %v, i32 addrspace(1)* %r, align 1
ret void
}
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; SI-LABEL: {{^}}global_align2_load_store_i32:
; ALIGNED: buffer_load_ushort
; ALIGNED: buffer_load_ushort
; ALIGNED: buffer_store_short
; ALIGNED: buffer_store_short
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; UNALIGNED: buffer_load_dword
; UNALIGNED: buffer_store_dword
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @global_align2_load_store_i32(i32 addrspace(1)* %p, i32 addrspace(1)* %r) #0 {
AMDGPU: Custom lower v2i32 loads and stores This will allow us to split up 64-bit private accesses when necessary. llvm-svn: 268296
2016-05-02 22:13:51 +02:00
%v = load i32, i32 addrspace(1)* %p, align 2
store i32 %v, i32 addrspace(1)* %r, align 2
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_align2_load_store_i32:
; GCN: ds_read_u16
; GCN: ds_read_u16
; GCN: ds_write_b16
; GCN: ds_write_b16
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_align2_load_store_i32(i32 addrspace(3)* %p, i32 addrspace(3)* %r) #0 {
AMDGPU: Custom lower v2i32 loads and stores This will allow us to split up 64-bit private accesses when necessary. llvm-svn: 268296
2016-05-02 22:13:51 +02:00
%v = load i32, i32 addrspace(3)* %p, align 2
store i32 %v, i32 addrspace(3)* %r, align 2
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_unaligned_load_store_i64:
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_unaligned_load_store_i64(i64 addrspace(3)* %p, i64 addrspace(3)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load i64, i64 addrspace(3)* %p, align 1
R600/SI: 64-bit and larger memory access must be at least 4-byte aligned This is true for SI only. CI+ supports unaligned memory accesses, but this requires driver support, so for now we disallow unaligned accesses for all GCN targets. llvm-svn: 227822
2015-02-02 19:02:28 +01:00
store i64 %v, i64 addrspace(3)* %r, align 1
ret void
}
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
; SI-LABEL: {{^}}local_unaligned_load_store_v2i32:
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI-NOT: v_or_b32
; SI-NOT: v_lshl
; SI: ds_write_b8
; SI: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_unaligned_load_store_v2i32(<2 x i32> addrspace(3)* %p, <2 x i32> addrspace(3)* %r) #0 {
AMDGPU: Custom lower v2i32 loads and stores This will allow us to split up 64-bit private accesses when necessary. llvm-svn: 268296
2016-05-02 22:13:51 +02:00
%v = load <2 x i32>, <2 x i32> addrspace(3)* %p, align 1
store <2 x i32> %v, <2 x i32> addrspace(3)* %r, align 1
ret void
}
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
; SI-LABEL: {{^}}global_align2_load_store_i64:
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED: buffer_load_ushort
; ALIGNED: buffer_load_ushort
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED-NOT: v_or_
; ALIGNED-NOT: v_lshl
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED: buffer_load_ushort
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED-NOT: v_or_
; ALIGNED-NOT: v_lshl
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED: buffer_load_ushort
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED-NOT: v_or_
; ALIGNED-NOT: v_lshl
; ALIGNED: buffer_store_short
; ALIGNED: buffer_store_short
; ALIGNED: buffer_store_short
; ALIGNED: buffer_store_short
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; UNALIGNED: buffer_load_dwordx2
; UNALIGNED: buffer_store_dwordx2
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @global_align2_load_store_i64(i64 addrspace(1)* %p, i64 addrspace(1)* %r) #0 {
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
%v = load i64, i64 addrspace(1)* %p, align 2
store i64 %v, i64 addrspace(1)* %r, align 2
ret void
}
; SI-LABEL: {{^}}unaligned_load_store_i64_global:
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED-NOT: v_or_
; ALIGNED-NOT: v_lshl
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; UNALIGNED: buffer_load_dwordx2
; UNALIGNED: buffer_store_dwordx2
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @unaligned_load_store_i64_global(i64 addrspace(1)* %p, i64 addrspace(1)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load i64, i64 addrspace(1)* %p, align 1
R600/SI: 64-bit and larger memory access must be at least 4-byte aligned This is true for SI only. CI+ supports unaligned memory accesses, but this requires driver support, so for now we disallow unaligned accesses for all GCN targets. llvm-svn: 227822
2015-02-02 19:02:28 +01:00
store i64 %v, i64 addrspace(1)* %r, align 1
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_unaligned_load_store_v4i32:
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: ds_write_b8
; GCN: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_unaligned_load_store_v4i32(<4 x i32> addrspace(3)* %p, <4 x i32> addrspace(3)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load <4 x i32>, <4 x i32> addrspace(3)* %p, align 1
Fix CodeGen for unaligned loads with address spaces llvm-svn: 193721
2013-10-31 00:30:05 +01:00
store <4 x i32> %v, <4 x i32> addrspace(3)* %r, align 1
ret void
}
Add alignment value to allowsUnalignedMemoryAccess Rename to allowsMisalignedMemoryAccess. On R600, 8 and 16 byte accesses are mostly OK with 4-byte alignment, and don't need to be split into multiple accesses. Vector loads with an alignment of the element type are not uncommon in OpenCL code. llvm-svn: 214055
2014-07-27 19:46:40 +02:00
AMDGPU: Expand unaligned accesses early Due to visit order problems, in the case of an unaligned copy the legalized DAG fails to eliminate extra instructions introduced by the expansion of both unaligned parts. llvm-svn: 274397
2016-07-02 00:55:55 +02:00
; SI-LABEL: {{^}}global_unaligned_load_store_v4i32
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; ALIGNED: buffer_store_byte
; UNALIGNED: buffer_load_dwordx4
; UNALIGNED: buffer_store_dwordx4
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @global_unaligned_load_store_v4i32(<4 x i32> addrspace(1)* %p, <4 x i32> addrspace(1)* %r) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%v = load <4 x i32>, <4 x i32> addrspace(1)* %p, align 1
R600/SI: 64-bit and larger memory access must be at least 4-byte aligned This is true for SI only. CI+ supports unaligned memory accesses, but this requires driver support, so for now we disallow unaligned accesses for all GCN targets. llvm-svn: 227822
2015-02-02 19:02:28 +01:00
store <4 x i32> %v, <4 x i32> addrspace(1)* %r, align 1
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_load_i64_align_4:
; GCN: ds_read2_b32
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_load_i64_align_4(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%val = load i64, i64 addrspace(3)* %in, align 4
Add alignment value to allowsUnalignedMemoryAccess Rename to allowsMisalignedMemoryAccess. On R600, 8 and 16 byte accesses are mostly OK with 4-byte alignment, and don't need to be split into multiple accesses. Vector loads with an alignment of the element type are not uncommon in OpenCL code. llvm-svn: 214055
2014-07-27 19:46:40 +02:00
store i64 %val, i64 addrspace(1)* %out, align 8
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_load_i64_align_4_with_offset
; GCN: ds_read2_b32 v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]}} offset0:8 offset1:9
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_load_i64_align_4_with_offset(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
%ptr = getelementptr i64, i64 addrspace(3)* %in, i32 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%val = load i64, i64 addrspace(3)* %ptr, align 4
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
store i64 %val, i64 addrspace(1)* %out, align 8
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_load_i64_align_4_with_split_offset:
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
; The tests for the case where the lo offset is 8-bits, but the hi offset is 9-bits
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; GCN: ds_read2_b32 v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]}} offset1:1
; GCN: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_load_i64_align_4_with_split_offset(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
%ptr = bitcast i64 addrspace(3)* %in to i32 addrspace(3)*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
%ptr255 = getelementptr i32, i32 addrspace(3)* %ptr, i32 255
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
%ptri64 = bitcast i32 addrspace(3)* %ptr255 to i64 addrspace(3)*
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%val = load i64, i64 addrspace(3)* %ptri64, align 4
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
store i64 %val, i64 addrspace(1)* %out, align 8
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_load_i64_align_1:
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: ds_read_u8
; GCN: store_dwordx2
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_load_i64_align_1(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%val = load i64, i64 addrspace(3)* %in, align 1
R600/SI: Merge two test files llvm-svn: 227821
2015-02-02 19:02:23 +01:00
store i64 %val, i64 addrspace(1)* %out, align 8
ret void
}
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_store_i64_align_4:
; GCN: ds_write2_b32
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_store_i64_align_4(i64 addrspace(3)* %out, i64 %val) #0 {
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
store i64 %val, i64 addrspace(3)* %out, align 4
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_store_i64_align_4_with_offset
; GCN: ds_write2_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset0:8 offset1:9
; GCN: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_store_i64_align_4_with_offset(i64 addrspace(3)* %out) #0 {
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
%ptr = getelementptr i64, i64 addrspace(3)* %out, i32 4
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
store i64 0, i64 addrspace(3)* %ptr, align 4
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; FUNC-LABEL: {{^}}local_store_i64_align_4_with_split_offset:
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
; The tests for the case where the lo offset is 8-bits, but the hi offset is 9-bits
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
; GCN: ds_write2_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset1:1
; GCN: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_store_i64_align_4_with_split_offset(i64 addrspace(3)* %out) #0 {
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
%ptr = bitcast i64 addrspace(3)* %out to i32 addrspace(3)*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
%ptr255 = getelementptr i32, i32 addrspace(3)* %ptr, i32 255
R600/SI: Use READ2/WRITE2 instructions for 64-bit mem ops with 32-bit alignment llvm-svn: 216279
2014-08-22 20:49:35 +02:00
%ptri64 = bitcast i32 addrspace(3)* %ptr255 to i64 addrspace(3)*
store i64 0, i64 addrspace(3)* %out, align 4
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; SI-LABEL: {{^}}constant_unaligned_load_i32:
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; UNALIGNED: s_load_dword
; SI: buffer_store_dword
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_unaligned_load_i32(i32 addrspace(2)* %p, i32 addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load i32, i32 addrspace(2)* %p, align 1
store i32 %v, i32 addrspace(1)* %r, align 4
ret void
}
; SI-LABEL: {{^}}constant_align2_load_i32:
; ALIGNED: buffer_load_ushort
; ALIGNED: buffer_load_ushort
; UNALIGNED: s_load_dword
; UNALIGNED: buffer_store_dword
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_align2_load_i32(i32 addrspace(2)* %p, i32 addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load i32, i32 addrspace(2)* %p, align 2
store i32 %v, i32 addrspace(1)* %r, align 4
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
ret void
}
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; SI-LABEL: {{^}}constant_align2_load_i64:
; ALIGNED: buffer_load_ushort
; ALIGNED: buffer_load_ushort
; ALIGNED: buffer_load_ushort
; ALIGNED: buffer_load_ushort
; UNALIGNED: s_load_dwordx2
; UNALIGNED: buffer_store_dwordx2
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_align2_load_i64(i64 addrspace(2)* %p, i64 addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load i64, i64 addrspace(2)* %p, align 2
store i64 %v, i64 addrspace(1)* %r, align 4
ret void
}
; SI-LABEL: {{^}}constant_align4_load_i64:
; SI: s_load_dwordx2
; SI: buffer_store_dwordx2
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_align4_load_i64(i64 addrspace(2)* %p, i64 addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load i64, i64 addrspace(2)* %p, align 4
store i64 %v, i64 addrspace(1)* %r, align 4
ret void
}
; SI-LABEL: {{^}}constant_align4_load_v4i32:
; SI: s_load_dwordx4
; SI: buffer_store_dwordx4
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_align4_load_v4i32(<4 x i32> addrspace(2)* %p, <4 x i32> addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load <4 x i32>, <4 x i32> addrspace(2)* %p, align 4
store <4 x i32> %v, <4 x i32> addrspace(1)* %r, align 4
ret void
}
; SI-LABEL: {{^}}constant_unaligned_load_v2i32:
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; UNALIGNED: buffer_load_dwordx2
; SI: buffer_store_dwordx2
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_unaligned_load_v2i32(<2 x i32> addrspace(2)* %p, <2 x i32> addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load <2 x i32>, <2 x i32> addrspace(2)* %p, align 1
store <2 x i32> %v, <2 x i32> addrspace(1)* %r, align 4
ret void
}
; SI-LABEL: {{^}}constant_unaligned_load_v4i32:
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; ALIGNED: buffer_load_ubyte
; UNALIGNED: buffer_load_dwordx4
; SI: buffer_store_dwordx4
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_unaligned_load_v4i32(<4 x i32> addrspace(2)* %p, <4 x i32> addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load <4 x i32>, <4 x i32> addrspace(2)* %p, align 1
store <4 x i32> %v, <4 x i32> addrspace(1)* %r, align 4
ret void
}
; SI-LABEL: {{^}}constant_align4_load_i8:
; SI: buffer_load_ubyte
; SI: buffer_store_byte
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_align4_load_i8(i8 addrspace(2)* %p, i8 addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load i8, i8 addrspace(2)* %p, align 4
store i8 %v, i8 addrspace(1)* %r, align 4
ret void
}
; SI-LABEL: {{^}}constant_align2_load_i8:
; SI: buffer_load_ubyte
; SI: buffer_store_byte
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_align2_load_i8(i8 addrspace(2)* %p, i8 addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%v = load i8, i8 addrspace(2)* %p, align 2
store i8 %v, i8 addrspace(1)* %r, align 2
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
ret void
}
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
; SI-LABEL: {{^}}constant_align4_merge_load_2_i32:
; SI: s_load_dwordx2 s{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0x0{{$}}
; SI-DAG: v_mov_b32_e32 v[[VLO:[0-9]+]], s[[LO]]
; SI-DAG: v_mov_b32_e32 v[[VHI:[0-9]+]], s[[HI]]
; SI: buffer_store_dwordx2 v{{\[}}[[VLO]]:[[VHI]]{{\]}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @constant_align4_merge_load_2_i32(i32 addrspace(2)* %p, i32 addrspace(1)* %r) #0 {
AMDGPU: Add feature for unaligned access llvm-svn: 274398
2016-07-02 01:03:44 +02:00
%gep0 = getelementptr i32, i32 addrspace(2)* %p, i64 1
%v0 = load i32, i32 addrspace(2)* %p, align 4
%v1 = load i32, i32 addrspace(2)* %gep0, align 4
%gep1 = getelementptr i32, i32 addrspace(1)* %r, i64 1
store i32 %v0, i32 addrspace(1)* %r, align 4
store i32 %v1, i32 addrspace(1)* %gep1, align 4
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
ret void
}
AMDGPU: Fix introducing stack access on unaligned v16i8 llvm-svn: 280298
2016-08-31 23:52:27 +02:00
; SI-LABEL: {{^}}local_load_align1_v16i8:
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ds_read_u8
; SI: ScratchSize: 0{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_load_align1_v16i8(<16 x i8> addrspace(1)* %out, <16 x i8> addrspace(3)* %in) #0 {
AMDGPU: Fix introducing stack access on unaligned v16i8 llvm-svn: 280298
2016-08-31 23:52:27 +02:00
%ld = load <16 x i8>, <16 x i8> addrspace(3)* %in, align 1
store <16 x i8> %ld, <16 x i8> addrspace(1)* %out
ret void
}
; SI-LABEL: {{^}}local_store_align1_v16i8:
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ds_write_b8
; SI: ScratchSize: 0{{$}}
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @local_store_align1_v16i8(<16 x i8> addrspace(3)* %out) #0 {
AMDGPU: Fix introducing stack access on unaligned v16i8 llvm-svn: 280298
2016-08-31 23:52:27 +02:00
store <16 x i8> zeroinitializer, <16 x i8> addrspace(3)* %out, align 1
ret void
}
AMDGPU: Cleanup load tests There are a lot of different kinds of loads to test for, and these were scattered around inconsistently with some redundancy. Try to comprehensively test all loads in a consistent way. llvm-svn: 271571
2016-06-02 21:54:26 +02:00
attributes #0 = { nounwind }
Reference in New Issue Copy Permalink