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/X86/domain-reassignment.mir

755 lines
22 KiB
Plaintext
Raw Normal View History

[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
# RUN: llc -run-pass x86-domain-reassignment -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512dq -o - %s | FileCheck %s
--- |
; ModuleID = '../test/CodeGen/X86/gpr-to-mask.ll'
source_filename = "../test/CodeGen/X86/gpr-to-mask.ll"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-unknown"
define void @test_fcmp_storefloat(i1 %cond, float* %fptr, float %f1, float %f2, float %f3, float %f4, float %f5, float %f6) #0 {
entry:
br i1 %cond, label %if, label %else
if: ; preds = %entry
%cmp1 = fcmp oeq float %f3, %f4
br label %exit
else: ; preds = %entry
%cmp2 = fcmp oeq float %f5, %f6
br label %exit
exit: ; preds = %else, %if
%val = phi i1 [ %cmp1, %if ], [ %cmp2, %else ]
%selected = select i1 %val, float %f1, float %f2
store float %selected, float* %fptr
ret void
}
define void @test_8bitops() #0 {
ret void
}
define void @test_16bitops() #0 {
ret void
}
define void @test_32bitops() #0 {
ret void
}
define void @test_64bitops() #0 {
ret void
}
define void @test_16bitext() #0 {
ret void
}
define void @test_32bitext() #0 {
ret void
}
define void @test_64bitext() #0 {
ret void
}
...
---
name: test_fcmp_storefloat
# CHECK-LABEL: name: test_fcmp_storefloat
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr8, preferred-register: '' }
- { id: 1, class: gr8, preferred-register: '' }
- { id: 2, class: gr8, preferred-register: '' }
- { id: 3, class: gr32, preferred-register: '' }
- { id: 4, class: gr64, preferred-register: '' }
- { id: 5, class: vr128x, preferred-register: '' }
- { id: 6, class: fr32x, preferred-register: '' }
- { id: 7, class: fr32x, preferred-register: '' }
- { id: 8, class: fr32x, preferred-register: '' }
- { id: 9, class: fr32x, preferred-register: '' }
- { id: 10, class: fr32x, preferred-register: '' }
- { id: 11, class: gr8, preferred-register: '' }
- { id: 12, class: vk1, preferred-register: '' }
- { id: 13, class: gr32, preferred-register: '' }
- { id: 14, class: vk1, preferred-register: '' }
- { id: 15, class: gr32, preferred-register: '' }
- { id: 16, class: gr32, preferred-register: '' }
- { id: 17, class: gr32, preferred-register: '' }
- { id: 18, class: vk1wm, preferred-register: '' }
- { id: 19, class: vr128x, preferred-register: '' }
- { id: 20, class: fr128, preferred-register: '' }
- { id: 21, class: fr128, preferred-register: '' }
- { id: 22, class: fr32x, preferred-register: '' }
liveins:
- { reg: '%edi', virtual-reg: '%3' }
- { reg: '%rsi', virtual-reg: '%4' }
- { reg: '%xmm0', virtual-reg: '%5' }
- { reg: '%xmm1', virtual-reg: '%6' }
- { reg: '%xmm2', virtual-reg: '%7' }
- { reg: '%xmm3', virtual-reg: '%8' }
- { reg: '%xmm4', virtual-reg: '%9' }
- { reg: '%xmm5', virtual-reg: '%10' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0.entry:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
successors: %bb.1(0x40000000), %bb.2(0x40000000)
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
liveins: %edi, %rsi, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5
%10 = COPY %xmm5
%9 = COPY %xmm4
%8 = COPY %xmm3
%7 = COPY %xmm2
%6 = COPY %xmm1
%5 = COPY %xmm0
%4 = COPY %rsi
%3 = COPY %edi
%11 = COPY %3.sub_8bit
TEST8ri killed %11, 1, implicit-def %eflags
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
JE_1 %bb.2, implicit %eflags
JMP_1 %bb.1
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
bb.1.if:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
successors: %bb.3(0x80000000)
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%14 = VCMPSSZrr %7, %8, 0
; check that cross domain copies are replaced with same domain copies.
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %15:vk32 = COPY %14
; CHECK: %0:vk8 = COPY %15
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%15 = COPY %14
%0 = COPY %15.sub_8bit
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
JMP_1 %bb.3
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
bb.2.else:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
successors: %bb.3(0x80000000)
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%12 = VCMPSSZrr %9, %10, 0
; check that cross domain copies are replaced with same domain copies.
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %13:vk32 = COPY %12
; CHECK: %1:vk8 = COPY %13
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%13 = COPY %12
%1 = COPY %13.sub_8bit
bb.3.exit:
; check PHI, IMPLICIT_DEF, and INSERT_SUBREG replacers.
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; CHECK: %2:vk8 = PHI %1, %bb.2, %0, %bb.1
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %16:vk32 = COPY %2
; CHECK: %18:vk1wm = COPY %16
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
%2 = PHI %1, %bb.2, %0, %bb.1
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%17 = IMPLICIT_DEF
%16 = INSERT_SUBREG %17, %2, 1
%18 = COPY %16
%19 = COPY %6
%21 = IMPLICIT_DEF
%20 = VMOVSSZrrk %19, killed %18, killed %21, %5
%22 = COPY %20
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVSSZmr %4, 1, %noreg, 0, %noreg, killed %22 :: (store 4 into %ir.fptr)
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
RET 0
...
---
name: test_8bitops
# CHECK-LABEL: name: test_8bitops
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr64, preferred-register: '' }
- { id: 1, class: vr512, preferred-register: '' }
- { id: 2, class: vr512, preferred-register: '' }
- { id: 3, class: vr512, preferred-register: '' }
- { id: 4, class: vr512, preferred-register: '' }
- { id: 5, class: vk8, preferred-register: '' }
- { id: 6, class: gr32, preferred-register: '' }
- { id: 7, class: gr8, preferred-register: '' }
- { id: 8, class: gr32, preferred-register: '' }
- { id: 9, class: gr32, preferred-register: '' }
- { id: 10, class: vk8wm, preferred-register: '' }
- { id: 11, class: vr512, preferred-register: '' }
- { id: 12, class: gr8, preferred-register: '' }
- { id: 13, class: gr8, preferred-register: '' }
- { id: 14, class: gr8, preferred-register: '' }
- { id: 15, class: gr8, preferred-register: '' }
- { id: 16, class: gr8, preferred-register: '' }
- { id: 17, class: gr8, preferred-register: '' }
- { id: 18, class: gr8, preferred-register: '' }
liveins:
- { reg: '%rdi', virtual-reg: '%0' }
- { reg: '%zmm0', virtual-reg: '%1' }
- { reg: '%zmm1', virtual-reg: '%2' }
- { reg: '%zmm2', virtual-reg: '%3' }
- { reg: '%zmm3', virtual-reg: '%4' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0:
liveins: %rdi, %zmm0, %zmm1, %zmm2, %zmm3
%0 = COPY %rdi
%1 = COPY %zmm0
%2 = COPY %zmm1
%3 = COPY %zmm2
%4 = COPY %zmm3
%5 = VCMPPDZrri %3, %4, 0
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %6:vk32 = COPY %5
; CHECK: %7:vk8 = COPY %6
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%6 = COPY %5
%7 = COPY %6.sub_8bit
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %12:vk8 = KSHIFTRBri %7, 2
; CHECK: %13:vk8 = KSHIFTLBri %12, 1
; CHECK: %14:vk8 = KNOTBrr %13
; CHECK: %15:vk8 = KORBrr %14, %12
; CHECK: %16:vk8 = KANDBrr %15, %13
; CHECK: %17:vk8 = KXORBrr %16, %12
; CHECK: %18:vk8 = KADDBrr %17, %14
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%12 = SHR8ri %7, 2, implicit-def dead %eflags
%13 = SHL8ri %12, 1, implicit-def dead %eflags
%14 = NOT8r %13
%15 = OR8rr %14, %12, implicit-def dead %eflags
%16 = AND8rr %15, %13, implicit-def dead %eflags
%17 = XOR8rr %16, %12, implicit-def dead %eflags
%18 = ADD8rr %17, %14, implicit-def dead %eflags
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %9:vk32 = COPY %18
; CHECK: %10:vk8wm = COPY %9
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%8 = IMPLICIT_DEF
%9 = INSERT_SUBREG %8, %18, 1
%10 = COPY %9
%11 = VMOVAPDZrrk %2, killed %10, %1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVAPDZmr %0, 1, %noreg, 0, %noreg, killed %11
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK: KTESTBrr %18, %18, implicit-def %eflags
TEST8rr %18, %18, implicit-def %eflags
JE_1 %bb.1, implicit %eflags
JMP_1 %bb.2
bb.1:
bb.2:
RET 0
...
---
name: test_16bitops
# CHECK-LABEL: name: test_16bitops
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr64, preferred-register: '' }
- { id: 1, class: vr512, preferred-register: '' }
- { id: 2, class: vr512, preferred-register: '' }
- { id: 3, class: vr512, preferred-register: '' }
- { id: 4, class: vr512, preferred-register: '' }
- { id: 5, class: vk16, preferred-register: '' }
- { id: 6, class: gr32, preferred-register: '' }
- { id: 7, class: gr16, preferred-register: '' }
- { id: 8, class: gr32, preferred-register: '' }
- { id: 9, class: gr32, preferred-register: '' }
- { id: 10, class: vk16wm, preferred-register: '' }
- { id: 11, class: vr512, preferred-register: '' }
- { id: 12, class: gr16, preferred-register: '' }
- { id: 13, class: gr16, preferred-register: '' }
- { id: 14, class: gr16, preferred-register: '' }
- { id: 15, class: gr16, preferred-register: '' }
- { id: 16, class: gr16, preferred-register: '' }
- { id: 17, class: gr16, preferred-register: '' }
liveins:
- { reg: '%rdi', virtual-reg: '%0' }
- { reg: '%zmm0', virtual-reg: '%1' }
- { reg: '%zmm1', virtual-reg: '%2' }
- { reg: '%zmm2', virtual-reg: '%3' }
- { reg: '%zmm3', virtual-reg: '%4' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0:
liveins: %rdi, %zmm0, %zmm1, %zmm2, %zmm3
%0 = COPY %rdi
%1 = COPY %zmm0
%2 = COPY %zmm1
%3 = COPY %zmm2
%4 = COPY %zmm3
%5 = VCMPPSZrri %3, %4, 0
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %6:vk32 = COPY %5
; CHECK: %7:vk16 = COPY %6
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%6 = COPY %5
%7 = COPY %6.sub_16bit
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %12:vk16 = KSHIFTRWri %7, 2
; CHECK: %13:vk16 = KSHIFTLWri %12, 1
; CHECK: %14:vk16 = KNOTWrr %13
; CHECK: %15:vk16 = KORWrr %14, %12
; CHECK: %16:vk16 = KANDWrr %15, %13
; CHECK: %17:vk16 = KXORWrr %16, %12
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%12 = SHR16ri %7, 2, implicit-def dead %eflags
%13 = SHL16ri %12, 1, implicit-def dead %eflags
%14 = NOT16r %13
%15 = OR16rr %14, %12, implicit-def dead %eflags
%16 = AND16rr %15, %13, implicit-def dead %eflags
%17 = XOR16rr %16, %12, implicit-def dead %eflags
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %9:vk32 = COPY %17
; CHECK: %10:vk16wm = COPY %9
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%8 = IMPLICIT_DEF
%9 = INSERT_SUBREG %8, %17, 3
%10 = COPY %9
%11 = VMOVAPSZrrk %2, killed %10, %1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVAPSZmr %0, 1, %noreg, 0, %noreg, killed %11
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK: KTESTWrr %17, %17, implicit-def %eflags
TEST16rr %17, %17, implicit-def %eflags
JE_1 %bb.1, implicit %eflags
JMP_1 %bb.2
bb.1:
bb.2:
RET 0
...
---
name: test_32bitops
# CHECK-LABEL: name: test_32bitops
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr64, preferred-register: '' }
- { id: 1, class: vr512, preferred-register: '' }
- { id: 2, class: vr512, preferred-register: '' }
- { id: 3, class: vk32wm, preferred-register: '' }
- { id: 4, class: vr512, preferred-register: '' }
- { id: 5, class: gr32, preferred-register: '' }
- { id: 6, class: gr32, preferred-register: '' }
- { id: 7, class: gr32, preferred-register: '' }
- { id: 8, class: gr32, preferred-register: '' }
- { id: 9, class: gr32, preferred-register: '' }
- { id: 10, class: gr32, preferred-register: '' }
- { id: 11, class: gr32, preferred-register: '' }
- { id: 12, class: gr32, preferred-register: '' }
- { id: 13, class: gr32, preferred-register: '' }
liveins:
- { reg: '%rdi', virtual-reg: '%0' }
- { reg: '%zmm0', virtual-reg: '%1' }
- { reg: '%zmm1', virtual-reg: '%2' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0:
liveins: %rdi, %zmm0, %zmm1
%0 = COPY %rdi
%1 = COPY %zmm0
%2 = COPY %zmm1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
; CHECK: %5:vk32 = KMOVDkm %0, 1, %noreg, 0, %noreg
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %6:vk32 = KSHIFTRDri %5, 2
; CHECK: %7:vk32 = KSHIFTLDri %6, 1
; CHECK: %8:vk32 = KNOTDrr %7
; CHECK: %9:vk32 = KORDrr %8, %6
; CHECK: %10:vk32 = KANDDrr %9, %7
; CHECK: %11:vk32 = KXORDrr %10, %6
; CHECK: %12:vk32 = KANDNDrr %11, %9
; CHECK: %13:vk32 = KADDDrr %12, %11
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
%5 = MOV32rm %0, 1, %noreg, 0, %noreg
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%6 = SHR32ri %5, 2, implicit-def dead %eflags
%7 = SHL32ri %6, 1, implicit-def dead %eflags
%8 = NOT32r %7
%9 = OR32rr %8, %6, implicit-def dead %eflags
%10 = AND32rr %9, %7, implicit-def dead %eflags
%11 = XOR32rr %10, %6, implicit-def dead %eflags
%12 = ANDN32rr %11, %9, implicit-def dead %eflags
%13 = ADD32rr %12, %11, implicit-def dead %eflags
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %3:vk32wm = COPY %13
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%3 = COPY %13
%4 = VMOVDQU16Zrrk %2, killed %3, %1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVDQA32Zmr %0, 1, %noreg, 0, %noreg, killed %4
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK: KTESTDrr %13, %13, implicit-def %eflags
TEST32rr %13, %13, implicit-def %eflags
JE_1 %bb.1, implicit %eflags
JMP_1 %bb.2
bb.1:
bb.2:
RET 0
...
---
name: test_64bitops
# CHECK-LABEL: name: test_64bitops
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr64, preferred-register: '' }
- { id: 1, class: vr512, preferred-register: '' }
- { id: 2, class: vr512, preferred-register: '' }
- { id: 3, class: vk64wm, preferred-register: '' }
- { id: 4, class: vr512, preferred-register: '' }
- { id: 5, class: gr64, preferred-register: '' }
- { id: 6, class: gr64, preferred-register: '' }
- { id: 7, class: gr64, preferred-register: '' }
- { id: 8, class: gr64, preferred-register: '' }
- { id: 9, class: gr64, preferred-register: '' }
- { id: 10, class: gr64, preferred-register: '' }
- { id: 11, class: gr64, preferred-register: '' }
- { id: 12, class: gr64, preferred-register: '' }
- { id: 13, class: gr64, preferred-register: '' }
liveins:
- { reg: '%rdi', virtual-reg: '%0' }
- { reg: '%zmm0', virtual-reg: '%1' }
- { reg: '%zmm1', virtual-reg: '%2' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0:
liveins: %rdi, %zmm0, %zmm1
%0 = COPY %rdi
%1 = COPY %zmm0
%2 = COPY %zmm1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
; CHECK: %5:vk64 = KMOVQkm %0, 1, %noreg, 0, %noreg
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %6:vk64 = KSHIFTRQri %5, 2
; CHECK: %7:vk64 = KSHIFTLQri %6, 1
; CHECK: %8:vk64 = KNOTQrr %7
; CHECK: %9:vk64 = KORQrr %8, %6
; CHECK: %10:vk64 = KANDQrr %9, %7
; CHECK: %11:vk64 = KXORQrr %10, %6
; CHECK: %12:vk64 = KANDNQrr %11, %9
; CHECK: %13:vk64 = KADDQrr %12, %11
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
%5 = MOV64rm %0, 1, %noreg, 0, %noreg
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%6 = SHR64ri %5, 2, implicit-def dead %eflags
%7 = SHL64ri %6, 1, implicit-def dead %eflags
%8 = NOT64r %7
%9 = OR64rr %8, %6, implicit-def dead %eflags
%10 = AND64rr %9, %7, implicit-def dead %eflags
%11 = XOR64rr %10, %6, implicit-def dead %eflags
%12 = ANDN64rr %11, %9, implicit-def dead %eflags
%13 = ADD64rr %12, %11, implicit-def dead %eflags
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %3:vk64wm = COPY %13
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%3 = COPY %13
%4 = VMOVDQU8Zrrk %2, killed %3, %1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVDQA32Zmr %0, 1, %noreg, 0, %noreg, killed %4
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK: KTESTQrr %13, %13, implicit-def %eflags
TEST64rr %13, %13, implicit-def %eflags
JE_1 %bb.1, implicit %eflags
JMP_1 %bb.2
bb.1:
bb.2:
RET 0
...
---
name: test_16bitext
# CHECK-LABEL: name: test_16bitext
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr64, preferred-register: '' }
- { id: 1, class: vr512, preferred-register: '' }
- { id: 2, class: vr512, preferred-register: '' }
- { id: 3, class: vk16wm, preferred-register: '' }
- { id: 4, class: vr512, preferred-register: '' }
- { id: 5, class: gr16, preferred-register: '' }
- { id: 6, class: gr16, preferred-register: '' }
liveins:
- { reg: '%rdi', virtual-reg: '%0' }
- { reg: '%zmm0', virtual-reg: '%1' }
- { reg: '%zmm1', virtual-reg: '%2' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0:
liveins: %rdi, %zmm0, %zmm1
%0 = COPY %rdi
%1 = COPY %zmm0
%2 = COPY %zmm1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
; CHECK: %7:vk8 = KMOVBkm %0, 1, %noreg, 0, %noreg
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %5:vk16 = COPY %7
; CHECK: %6:vk16 = KNOTWrr %5
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
%5 = MOVZX16rm8 %0, 1, %noreg, 0, %noreg
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%6 = NOT16r %5
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %3:vk16wm = COPY %6
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%3 = COPY %6
%4 = VMOVAPSZrrk %2, killed %3, %1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVAPSZmr %0, 1, %noreg, 0, %noreg, killed %4
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
RET 0
...
---
name: test_32bitext
# CHECK-LABEL: name: test_32bitext
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr64, preferred-register: '' }
- { id: 1, class: vr512, preferred-register: '' }
- { id: 2, class: vr512, preferred-register: '' }
- { id: 3, class: vk64wm, preferred-register: '' }
- { id: 4, class: vr512, preferred-register: '' }
- { id: 5, class: gr32, preferred-register: '' }
- { id: 6, class: gr32, preferred-register: '' }
- { id: 7, class: gr32, preferred-register: '' }
liveins:
- { reg: '%rdi', virtual-reg: '%0' }
- { reg: '%zmm0', virtual-reg: '%1' }
- { reg: '%zmm1', virtual-reg: '%2' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0:
liveins: %rdi, %zmm0, %zmm1
%0 = COPY %rdi
%1 = COPY %zmm0
%2 = COPY %zmm1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
; CHECK: %8:vk8 = KMOVBkm %0, 1, %noreg, 0, %noreg
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %5:vk32 = COPY %8
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
; CHECK: %9:vk16 = KMOVWkm %0, 1, %noreg, 0, %noreg
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %6:vk32 = COPY %9
; CHECK: %7:vk32 = KADDDrr %5, %6
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
%5 = MOVZX32rm8 %0, 1, %noreg, 0, %noreg
%6 = MOVZX32rm16 %0, 1, %noreg, 0, %noreg
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%7 = ADD32rr %5, %6, implicit-def dead %eflags
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %3:vk64wm = COPY %7
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%3 = COPY %7
%4 = VMOVDQU16Zrrk %2, killed %3, %1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVDQA32Zmr %0, 1, %noreg, 0, %noreg, killed %4
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
RET 0
...
---
name: test_64bitext
# CHECK-LABEL: name: test_64bitext
alignment: 4
exposesReturnsTwice: false
legalized: false
regBankSelected: false
selected: false
tracksRegLiveness: true
registers:
- { id: 0, class: gr64, preferred-register: '' }
- { id: 1, class: vr512, preferred-register: '' }
- { id: 2, class: vr512, preferred-register: '' }
- { id: 3, class: vk64wm, preferred-register: '' }
- { id: 4, class: vr512, preferred-register: '' }
- { id: 5, class: gr64, preferred-register: '' }
- { id: 6, class: gr64, preferred-register: '' }
- { id: 7, class: gr64, preferred-register: '' }
liveins:
- { reg: '%rdi', virtual-reg: '%0' }
- { reg: '%zmm0', virtual-reg: '%1' }
- { reg: '%zmm1', virtual-reg: '%2' }
frameInfo:
isFrameAddressTaken: false
isReturnAddressTaken: false
hasStackMap: false
hasPatchPoint: false
stackSize: 0
offsetAdjustment: 0
maxAlignment: 0
adjustsStack: false
hasCalls: false
stackProtector: ''
maxCallFrameSize: 4294967295
hasOpaqueSPAdjustment: false
hasVAStart: false
hasMustTailInVarArgFunc: false
savePoint: ''
restorePoint: ''
fixedStack:
stack:
constants:
body: |
bb.0:
liveins: %rdi, %zmm0, %zmm1
%0 = COPY %rdi
%1 = COPY %zmm0
%2 = COPY %zmm1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
; CHECK: %8:vk8 = KMOVBkm %0, 1, %noreg, 0, %noreg
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %5:vk64 = COPY %8
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
; CHECK: %9:vk16 = KMOVWkm %0, 1, %noreg, 0, %noreg
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %6:vk64 = COPY %9
; CHECK: %7:vk64 = KADDQrr %5, %6
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
%5 = MOVZX64rm8 %0, 1, %noreg, 0, %noreg
%6 = MOVZX64rm16 %0, 1, %noreg, 0, %noreg
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%7 = ADD64rr %5, %6, implicit-def dead %eflags
MIR: Print the register class or bank in vreg defs This updates the MIRPrinter to include the regclass when printing virtual register defs, which is already valid syntax for the parser. That is, given 64 bit %0 and %1 in a "gpr" regbank, %1(s64) = COPY %0(s64) would now be written as %1:gpr(s64) = COPY %0(s64) While this change alone introduces a bit of redundancy with the registers block, it allows us to update the tests to be more concise and understandable and brings us closer to being able to remove the registers block completely. Note: We generally only print the class in defs, but there is one exception. If there are uses without any defs whatsoever, we'll print the class on all uses. I'm not completely convinced this comes up in meaningful machine IR, but for now the MIRParser and MachineVerifier both accept that kind of stuff, so we don't want to have a situation where we can print something we can't parse. llvm-svn: 316479
2017-10-25 02:04:54 +08:00
; CHECK: %3:vk64wm = COPY %7
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
%3 = COPY %7
%4 = VMOVDQU8Zrrk %2, killed %3, %1
[CodeGen] Always use `printReg` to print registers in both MIR and debug output As part of the unification of the debug format and the MIR format, always use `printReg` to print all kinds of registers. Updated the tests using '_' instead of '%noreg' until we decide which one we want to be the default one. Differential Revision: https://reviews.llvm.org/D40421 llvm-svn: 319445
2017-12-01 00:12:24 +08:00
VMOVDQA32Zmr %0, 1, %noreg, 0, %noreg, killed %4
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
RET 0
...