maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/trunc-cmp-constant.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

175 lines
6.8 KiB
LLVM
Raw Normal View History

AMDGPU: Fix -enable-var-scope violations llvm-svn: 318004
2017-11-13 07:53:44 +08:00
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=SI -check-prefix=FUNC %s
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=SI -check-prefix=FUNC %s
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
[AMDGPU] Promote uniform (i1, i16] operations to i32 Differential Revision: https://reviews.llvm.org/D25302 llvm-svn: 283555
2016-10-07 22:22:58 +08:00
declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
[AMDGPU] Fix CHECK lines
2020-04-03 17:06:34 +08:00
; FUNC-LABEL: {{^}}sextload_i1_to_i32_trunc_cmp_eq_0:
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; SI: buffer_load_ubyte [[LOAD:v[0-9]+]]
; SI: v_and_b32_e32 [[TMP:v[0-9]+]], 1, [[LOAD]]
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; SI: v_cmp_eq_u32_e32 vcc, 0, [[TMP]]{{$}}
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; SI: v_cndmask_b32_e64
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sextload_i1_to_i32_trunc_cmp_eq_0(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = sext i1 %load to i32
%cmp = icmp eq i32 %ext, 0
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FIXME: The negate should be inverting the compare.
; FUNC-LABEL: {{^}}zextload_i1_to_i32_trunc_cmp_eq_0:
; SI: buffer_load_ubyte [[LOAD:v[0-9]+]]
; SI: v_and_b32_e32 [[TMP:v[0-9]+]], 1, [[LOAD]]
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; SI: v_cmp_eq_u32_e32 vcc, 1, [[TMP]]{{$}}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
; SI-NEXT: s_xor_b64 [[NEG:s\[[0-9]+:[0-9]+\]]], vcc, -1
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; SI-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, [[NEG]]
AMDGPU/SI: Enable lanemask tracking in misched Summary: This results in higher register usage, but should make it easier for the compiler to hide latency. This pass is a prerequisite for some more scheduler improvements, and I think the increase register usage with this patch is acceptable, because when combined with the scheduler improvements, the total register usage will decrease. shader-db stats: 2382 shaders in 478 tests Totals: SGPRS: 48672 -> 49088 (0.85 %) VGPRS: 34148 -> 34847 (2.05 %) Code Size: 1285816 -> 1289128 (0.26 %) bytes LDS: 28 -> 28 (0.00 %) blocks Scratch: 492544 -> 573440 (16.42 %) bytes per wave Max Waves: 6856 -> 6846 (-0.15 %) Wait states: 0 -> 0 (0.00 %) Depends on D18451 Reviewers: nhaehnle, arsenm Subscribers: arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D18452 llvm-svn: 264876
2016-03-31 00:35:09 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @zextload_i1_to_i32_trunc_cmp_eq_0(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = zext i1 %load to i32
%cmp = icmp eq i32 %ext, 0
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sextload_i1_to_i32_trunc_cmp_eq_1:
; SI: v_mov_b32_e32 [[RESULT:v[0-9]+]], 0{{$}}
R600/SI: Enable subreg liveness by default llvm-svn: 228228
2015-02-05 07:14:18 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sextload_i1_to_i32_trunc_cmp_eq_1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = sext i1 %load to i32
%cmp = icmp eq i32 %ext, 1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zextload_i1_to_i32_trunc_cmp_eq_1:
; SI: buffer_load_ubyte [[LOAD:v[0-9]+]]
; SI: v_and_b32_e32 [[RESULT:v[0-9]+]], 1, [[LOAD]]
AMDGPU/SI: Enable lanemask tracking in misched Summary: This results in higher register usage, but should make it easier for the compiler to hide latency. This pass is a prerequisite for some more scheduler improvements, and I think the increase register usage with this patch is acceptable, because when combined with the scheduler improvements, the total register usage will decrease. shader-db stats: 2382 shaders in 478 tests Totals: SGPRS: 48672 -> 49088 (0.85 %) VGPRS: 34148 -> 34847 (2.05 %) Code Size: 1285816 -> 1289128 (0.26 %) bytes LDS: 28 -> 28 (0.00 %) blocks Scratch: 492544 -> 573440 (16.42 %) bytes per wave Max Waves: 6856 -> 6846 (-0.15 %) Wait states: 0 -> 0 (0.00 %) Depends on D18451 Reviewers: nhaehnle, arsenm Subscribers: arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D18452 llvm-svn: 264876
2016-03-31 00:35:09 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @zextload_i1_to_i32_trunc_cmp_eq_1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = zext i1 %load to i32
%cmp = icmp eq i32 %ext, 1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sextload_i1_to_i32_trunc_cmp_eq_neg1:
; SI: buffer_load_ubyte [[LOAD:v[0-9]+]]
; SI: v_and_b32_e32 [[RESULT:v[0-9]+]], 1, [[LOAD]]
AMDGPU/SI: Enable lanemask tracking in misched Summary: This results in higher register usage, but should make it easier for the compiler to hide latency. This pass is a prerequisite for some more scheduler improvements, and I think the increase register usage with this patch is acceptable, because when combined with the scheduler improvements, the total register usage will decrease. shader-db stats: 2382 shaders in 478 tests Totals: SGPRS: 48672 -> 49088 (0.85 %) VGPRS: 34148 -> 34847 (2.05 %) Code Size: 1285816 -> 1289128 (0.26 %) bytes LDS: 28 -> 28 (0.00 %) blocks Scratch: 492544 -> 573440 (16.42 %) bytes per wave Max Waves: 6856 -> 6846 (-0.15 %) Wait states: 0 -> 0 (0.00 %) Depends on D18451 Reviewers: nhaehnle, arsenm Subscribers: arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D18452 llvm-svn: 264876
2016-03-31 00:35:09 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sextload_i1_to_i32_trunc_cmp_eq_neg1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = sext i1 %load to i32
%cmp = icmp eq i32 %ext, -1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zextload_i1_to_i32_trunc_cmp_eq_neg1:
; SI: v_mov_b32_e32 [[RESULT:v[0-9]+]], 0{{$}}
R600/SI: Enable subreg liveness by default llvm-svn: 228228
2015-02-05 07:14:18 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @zextload_i1_to_i32_trunc_cmp_eq_neg1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = zext i1 %load to i32
%cmp = icmp eq i32 %ext, -1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
[AMDGPU] Fix CHECK lines
2020-04-03 17:06:34 +08:00
; FUNC-LABEL: {{^}}sextload_i1_to_i32_trunc_cmp_ne_0:
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; SI: buffer_load_ubyte [[LOAD:v[0-9]+]]
[AMDGPU] Fix CHECK lines
2020-04-03 17:06:34 +08:00
; SI: v_and_b32_e32 [[RESULT:v[0-9]+]], 1, [[LOAD]]
AMDGPU/SI: Enable lanemask tracking in misched Summary: This results in higher register usage, but should make it easier for the compiler to hide latency. This pass is a prerequisite for some more scheduler improvements, and I think the increase register usage with this patch is acceptable, because when combined with the scheduler improvements, the total register usage will decrease. shader-db stats: 2382 shaders in 478 tests Totals: SGPRS: 48672 -> 49088 (0.85 %) VGPRS: 34148 -> 34847 (2.05 %) Code Size: 1285816 -> 1289128 (0.26 %) bytes LDS: 28 -> 28 (0.00 %) blocks Scratch: 492544 -> 573440 (16.42 %) bytes per wave Max Waves: 6856 -> 6846 (-0.15 %) Wait states: 0 -> 0 (0.00 %) Depends on D18451 Reviewers: nhaehnle, arsenm Subscribers: arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D18452 llvm-svn: 264876
2016-03-31 00:35:09 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sextload_i1_to_i32_trunc_cmp_ne_0(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = sext i1 %load to i32
%cmp = icmp ne i32 %ext, 0
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zextload_i1_to_i32_trunc_cmp_ne_0:
; SI: buffer_load_ubyte [[LOAD:v[0-9]+]]
AMDGPU: Fix -enable-var-scope violations llvm-svn: 318004
2017-11-13 07:53:44 +08:00
; SI: v_and_b32_e32 [[RESULT:v[0-9]+]], 1, [[LOAD]]
AMDGPU/SI: Enable lanemask tracking in misched Summary: This results in higher register usage, but should make it easier for the compiler to hide latency. This pass is a prerequisite for some more scheduler improvements, and I think the increase register usage with this patch is acceptable, because when combined with the scheduler improvements, the total register usage will decrease. shader-db stats: 2382 shaders in 478 tests Totals: SGPRS: 48672 -> 49088 (0.85 %) VGPRS: 34148 -> 34847 (2.05 %) Code Size: 1285816 -> 1289128 (0.26 %) bytes LDS: 28 -> 28 (0.00 %) blocks Scratch: 492544 -> 573440 (16.42 %) bytes per wave Max Waves: 6856 -> 6846 (-0.15 %) Wait states: 0 -> 0 (0.00 %) Depends on D18451 Reviewers: nhaehnle, arsenm Subscribers: arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D18452 llvm-svn: 264876
2016-03-31 00:35:09 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @zextload_i1_to_i32_trunc_cmp_ne_0(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = zext i1 %load to i32
%cmp = icmp ne i32 %ext, 0
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}sextload_i1_to_i32_trunc_cmp_ne_1:
; SI: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}}
R600/SI: Enable subreg liveness by default llvm-svn: 228228
2015-02-05 07:14:18 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sextload_i1_to_i32_trunc_cmp_ne_1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = sext i1 %load to i32
%cmp = icmp ne i32 %ext, 1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zextload_i1_to_i32_trunc_cmp_ne_1:
; SI: buffer_load_ubyte [[LOAD:v[0-9]+]]
; SI: v_and_b32_e32 [[TMP:v[0-9]+]], 1, [[LOAD]]
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; SI: v_cmp_eq_u32_e32 vcc, 1, [[TMP]]{{$}}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
; SI-NEXT: s_xor_b64 [[NEG:s\[[0-9]+:[0-9]+\]]], vcc, -1
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; SI-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, [[NEG]]
AMDGPU/SI: Enable lanemask tracking in misched Summary: This results in higher register usage, but should make it easier for the compiler to hide latency. This pass is a prerequisite for some more scheduler improvements, and I think the increase register usage with this patch is acceptable, because when combined with the scheduler improvements, the total register usage will decrease. shader-db stats: 2382 shaders in 478 tests Totals: SGPRS: 48672 -> 49088 (0.85 %) VGPRS: 34148 -> 34847 (2.05 %) Code Size: 1285816 -> 1289128 (0.26 %) bytes LDS: 28 -> 28 (0.00 %) blocks Scratch: 492544 -> 573440 (16.42 %) bytes per wave Max Waves: 6856 -> 6846 (-0.15 %) Wait states: 0 -> 0 (0.00 %) Depends on D18451 Reviewers: nhaehnle, arsenm Subscribers: arsenm, llvm-commits Differential Revision: http://reviews.llvm.org/D18452 llvm-svn: 264876
2016-03-31 00:35:09 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @zextload_i1_to_i32_trunc_cmp_ne_1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = zext i1 %load to i32
%cmp = icmp ne i32 %ext, 1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FIXME: This should be one compare.
; FUNC-LABEL: {{^}}sextload_i1_to_i32_trunc_cmp_ne_neg1:
; XSI: buffer_load_ubyte [[LOAD:v[0-9]+]]
; XSI: v_and_b32_e32 [[TMP:v[0-9]+]], 1, [[LOAD]]
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; XSI: v_cmp_eq_u32_e64 [[CMP0:s\[[0-9]+:[0-9]+\]]], [[TMP]], 0{{$}}
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; XSI-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, [[CMP0]]
; XSI-NEXT: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @sextload_i1_to_i32_trunc_cmp_ne_neg1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = sext i1 %load to i32
%cmp = icmp ne i32 %ext, -1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
; FUNC-LABEL: {{^}}zextload_i1_to_i32_trunc_cmp_ne_neg1:
; SI: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}}
R600/SI: Enable subreg liveness by default llvm-svn: 228228
2015-02-05 07:14:18 +08:00
; SI: buffer_store_byte [[RESULT]]
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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @zextload_i1_to_i32_trunc_cmp_ne_neg1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) nounwind {
[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]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%load = load i1, i1 addrspace(1)* %in
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%ext = zext i1 %load to i32
%cmp = icmp ne i32 %ext, -1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}
[AMDGPU] Promote uniform (i1, i16] operations to i32 Differential Revision: https://reviews.llvm.org/D25302 llvm-svn: 283555
2016-10-07 22:22:58 +08:00
; FIXME: Need to handle non-uniform case for function below (load without gep).
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; FUNC-LABEL: {{^}}masked_load_i1_to_i32_trunc_cmp_ne_neg1:
[AMDGPU] Promote uniform (i1, i16] operations to i32 Differential Revision: https://reviews.llvm.org/D25302 llvm-svn: 283555
2016-10-07 22:22:58 +08:00
; SI: {{buffer|flat}}_load_sbyte [[LOAD:v[0-9]+]]
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; SI: v_cmp_ne_u32_e32 vcc, -1, [[LOAD]]{{$}}
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
; SI-NEXT: v_cndmask_b32_e64
[AMDGPU] Promote uniform (i1, i16] operations to i32 Differential Revision: https://reviews.llvm.org/D25302 llvm-svn: 283555
2016-10-07 22:22:58 +08:00
; SI: {{buffer|flat}}_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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @masked_load_i1_to_i32_trunc_cmp_ne_neg1(i1 addrspace(1)* %out, i8 addrspace(1)* %in) nounwind {
[AMDGPU] Promote uniform (i1, i16] operations to i32 Differential Revision: https://reviews.llvm.org/D25302 llvm-svn: 283555
2016-10-07 22:22:58 +08:00
%tid.x = call i32 @llvm.amdgcn.workitem.id.x()
%in.ptr = getelementptr i8, i8 addrspace(1)* %in, i32 %tid.x
%load = load i8, i8 addrspace(1)* %in.ptr
Enable (sext x) == C --> x == (trunc C) combine Extend the existing code which handles this for zext. This makes this more useful for targets with ZeroOrNegativeOne BooleanContent and obsoletes a custom combine SI uses for i1 setcc (sext(i1), 0, setne) since the constant will now be shrunk to i1. llvm-svn: 224691
2014-12-22 00:48:42 +08:00
%masked = and i8 %load, 255
%ext = sext i8 %masked to i32
%cmp = icmp ne i32 %ext, -1
store i1 %cmp, i1 addrspace(1)* %out
ret void
}