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/s_movk_i32.ll

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

192 lines
8.9 KiB
LLVM
Raw Normal View History

[AMDGPU] Switch scalarize global loads ON by default Differential revision: https://reviews.llvm.org/D34407 llvm-svn: 307097
2017-07-05 01:32:00 +08:00
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
; SI-LABEL: {{^}}s_movk_i32_k0:
; SI-DAG: s_mov_b32 [[LO_S_IMM:s[0-9]+]], 0xffff{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
AMDGPU: Reduce number of copies emitted Instead of always inserting a copy in case the super register is itself a subregister, only extract to the super reg class if this is actually the case. This shouldn't really change codegen, but makes looking at the output of SIFixSGPRCopies easier to read. llvm-svn: 248467
2015-09-24 15:16:37 +08:00
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 1, v[[HI_VREG]]
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k0(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 4295032831 ; ((1 << 16) - 1) | (1 << 32)
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 4295032831)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k1:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x7fff{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
AMDGPU: Reduce number of copies emitted Instead of always inserting a copy in case the super register is itself a subregister, only extract to the super reg class if this is actually the case. This shouldn't really change codegen, but makes looking at the output of SIFixSGPRCopies easier to read. llvm-svn: 248467
2015-09-24 15:16:37 +08:00
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 1, v[[HI_VREG]]
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k1(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 4295000063 ; ((1 << 15) - 1) | (1 << 32)
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 4295000063)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k2:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x7fff{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
AMDGPU: Reduce number of copies emitted Instead of always inserting a copy in case the super register is itself a subregister, only extract to the super reg class if this is actually the case. This shouldn't really change codegen, but makes looking at the output of SIFixSGPRCopies easier to read. llvm-svn: 248467
2015-09-24 15:16:37 +08:00
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 64, v[[HI_VREG]]
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k2(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 274877939711 ; ((1 << 15) - 1) | (64 << 32)
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 274877939711)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k3:
; SI-DAG: s_mov_b32 [[LO_S_IMM:s[0-9]+]], 0x8000{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
AMDGPU: Reduce number of copies emitted Instead of always inserting a copy in case the super register is itself a subregister, only extract to the super reg class if this is actually the case. This shouldn't really change codegen, but makes looking at the output of SIFixSGPRCopies easier to read. llvm-svn: 248467
2015-09-24 15:16:37 +08:00
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 1, v[[HI_VREG]]
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k3(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 4295000064 ; (1 << 15) | (1 << 32)
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 4295000064)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k4:
; SI-DAG: s_mov_b32 [[LO_S_IMM:s[0-9]+]], 0x20000{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
AMDGPU: Reduce number of copies emitted Instead of always inserting a copy in case the super register is itself a subregister, only extract to the super reg class if this is actually the case. This shouldn't really change codegen, but makes looking at the output of SIFixSGPRCopies easier to read. llvm-svn: 248467
2015-09-24 15:16:37 +08:00
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 1, v[[HI_VREG]]
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k4(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 4295098368 ; (1 << 17) | (1 << 32)
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 4295098368)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k5:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0xffef{{$}}
; SI-DAG: s_mov_b32 [[HI_S_IMM:s[0-9]+]], 0xff00ffff{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[HI_S_IMM]], v[[HI_VREG]]
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k5(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 18374967954648334319 ; -17 & 0xff00ffffffffffff
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 18374967954648334319)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k6:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x41{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
AMDGPU: Reduce number of copies emitted Instead of always inserting a copy in case the super register is itself a subregister, only extract to the super reg class if this is actually the case. This shouldn't really change codegen, but makes looking at the output of SIFixSGPRCopies easier to read. llvm-svn: 248467
2015-09-24 15:16:37 +08:00
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 63, v[[HI_VREG]]
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k6(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 270582939713 ; 65 | (63 << 32)
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 270582939713)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k7:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x2000{{$}}
; SI-DAG: s_movk_i32 [[HI_S_IMM:s[0-9]+]], 0x4000{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[HI_S_IMM]], v[[HI_VREG]]
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k7(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 70368744185856; ((1 << 13)) | ((1 << 14) << 32)
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 70368744185856)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k8:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x8000{{$}}
; SI-DAG: s_mov_b32 [[HI_S_IMM:s[0-9]+]], 0x11111111{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[HI_S_IMM]], v[[HI_VREG]]
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k8(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 1229782942255906816 ; 0x11111111ffff8000
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 1229782942255906816)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k9:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x8001{{$}}
; SI-DAG: s_mov_b32 [[HI_S_IMM:s[0-9]+]], 0x11111111{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[HI_S_IMM]], v[[HI_VREG]]
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k9(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 1229782942255906817 ; 0x11111111ffff8001
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 1229782942255906817)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k10:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x8888{{$}}
; SI-DAG: s_mov_b32 [[HI_S_IMM:s[0-9]+]], 0x11111111{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[HI_S_IMM]], v[[HI_VREG]]
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k10(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 1229782942255909000 ; 0x11111111ffff8888
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 1229782942255909000)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k11:
; SI-DAG: s_movk_i32 [[LO_S_IMM:s[0-9]+]], 0x8fff{{$}}
; SI-DAG: s_mov_b32 [[HI_S_IMM:s[0-9]+]], 0x11111111{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[HI_S_IMM]], v[[HI_VREG]]
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k11(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 1229782942255910911 ; 0x11111111ffff8fff
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 1229782942255910911)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}
; SI-LABEL: {{^}}s_movk_i32_k12:
; SI-DAG: s_mov_b32 [[LO_S_IMM:s[0-9]+]], 0xffff7001{{$}}
; SI-DAG: s_mov_b32 [[HI_S_IMM:s[0-9]+]], 0x11111111{{$}}
; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]{{\]}},
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[LO_S_IMM]], v[[LO_VREG]]
; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, [[HI_S_IMM]], v[[HI_VREG]]
; 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
2017-03-22 05:39:51 +08:00
define amdgpu_kernel void @s_movk_i32_k12(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64 addrspace(1)* %b) {
[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
%loada = load i64, i64 addrspace(1)* %a, align 4
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
%or = or i64 %loada, 1229782942255902721 ; 0x11111111ffff7001
store i64 %or, i64 addrspace(1)* %out
AMDGPU: Fix immediate folding logic when shrinking instructions If the literal is being folded into src0, it doesn't matter if it's an SGPR because it's being replaced with the literal. Also fixes initially selecting 32-bit versions of some instructions which also confused commuting. llvm-svn: 281117
2016-09-10 07:32:53 +08:00
call void asm sideeffect "; use $0", "s"(i64 1229782942255902721)
R600/SI: Use s_movk_i32 llvm-svn: 221922
2014-11-14 04:44:23 +08:00
ret void
}