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/atomic8.ll

274 lines
6.0 KiB
LLVM
Raw Normal View History

Fix PR15355 - Clear 'mayStore' flag when loading from the atomic variable before the spin loop - Clear kill flag from one use to multiple use in registers forming the address to that atomic variable - don't use a physical register as live-in register in BB (neither entry nor landing pad.) by copying it into virtual register (patch by Cameron Zwarich) llvm-svn: 176538
2013-03-06 08:17:04 +08:00
; RUN: llc < %s -O0 -march=x86-64 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X64
; RUN: llc < %s -O0 -march=x86 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X32
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
@sc8 = external global i8
define void @atomic_fetch_add8() nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_add8:
; X32-LABEL: atomic_fetch_add8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
entry:
; 32-bit
%t1 = atomicrmw add i8* @sc8, i8 1 acquire
; X64: lock
; X64: incb
; X32: lock
; X32: incb
%t2 = atomicrmw add i8* @sc8, i8 3 acquire
; X64: lock
; X64: addb $3
; X32: lock
; X32: addb $3
%t3 = atomicrmw add i8* @sc8, i8 5 acquire
; X64: lock
; X64: xaddb
; X32: lock
; X32: xaddb
%t4 = atomicrmw add i8* @sc8, i8 %t3 acquire
; X64: lock
; X64: addb
; X32: lock
; X32: addb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_sub8() nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_sub8:
; X32-LABEL: atomic_fetch_sub8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw sub i8* @sc8, i8 1 acquire
; X64: lock
; X64: decb
; X32: lock
; X32: decb
%t2 = atomicrmw sub i8* @sc8, i8 3 acquire
; X64: lock
; X64: subb $3
; X32: lock
; X32: subb $3
%t3 = atomicrmw sub i8* @sc8, i8 5 acquire
; X64: lock
; X64: xaddb
; X32: lock
; X32: xaddb
%t4 = atomicrmw sub i8* @sc8, i8 %t3 acquire
; X64: lock
; X64: subb
; X32: lock
; X32: subb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_and8() nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_and8:
; X32-LABEL: atomic_fetch_and8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw and i8* @sc8, i8 3 acquire
; X64: lock
; X64: andb $3
; X32: lock
; X32: andb $3
%t2 = atomicrmw and i8* @sc8, i8 5 acquire
; X64: andb
; X64: lock
; X64: cmpxchgb
; X32: andb
; X32: lock
; X32: cmpxchgb
%t3 = atomicrmw and i8* @sc8, i8 %t2 acquire
; X64: lock
; X64: andb
; X32: lock
; X32: andb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_or8() nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_or8:
; X32-LABEL: atomic_fetch_or8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw or i8* @sc8, i8 3 acquire
; X64: lock
; X64: orb $3
; X32: lock
; X32: orb $3
%t2 = atomicrmw or i8* @sc8, i8 5 acquire
; X64: orb
; X64: lock
; X64: cmpxchgb
; X32: orb
; X32: lock
; X32: cmpxchgb
%t3 = atomicrmw or i8* @sc8, i8 %t2 acquire
; X64: lock
; X64: orb
; X32: lock
; X32: orb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_xor8() nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_xor8:
; X32-LABEL: atomic_fetch_xor8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw xor i8* @sc8, i8 3 acquire
; X64: lock
; X64: xorb $3
; X32: lock
; X32: xorb $3
%t2 = atomicrmw xor i8* @sc8, i8 5 acquire
; X64: xorb
; X64: lock
; X64: cmpxchgb
; X32: xorb
; X32: lock
; X32: cmpxchgb
%t3 = atomicrmw xor i8* @sc8, i8 %t2 acquire
; X64: lock
; X64: xorb
; X32: lock
; X32: xorb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_nand8(i8 %x) nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_nand8:
; X32-LABEL: atomic_fetch_nand8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw nand i8* @sc8, i8 %x acquire
; X64: andb
; X64: notb
; X64: lock
; X64: cmpxchgb
; X32: andb
; X32: notb
; X32: lock
; X32: cmpxchgb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_max8(i8 %x) nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_max8:
; X32-LABEL: atomic_fetch_max8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw max i8* @sc8, i8 %x acquire
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64: movsbl
; X64: movsbl
; X64: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X64: lock
; X64: cmpxchgb
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X32: movsbl
; X32: movsbl
; X32: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X32: lock
; X32: cmpxchgb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_min8(i8 %x) nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_min8:
; X32-LABEL: atomic_fetch_min8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw min i8* @sc8, i8 %x acquire
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64: movsbl
; X64: movsbl
; X64: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X64: lock
; X64: cmpxchgb
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X32: movsbl
; X32: movsbl
; X32: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X32: lock
; X32: cmpxchgb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_umax8(i8 %x) nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_umax8:
; X32-LABEL: atomic_fetch_umax8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw umax i8* @sc8, i8 %x acquire
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64: movzbl
; X64: movzbl
; X64: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X64: lock
; X64: cmpxchgb
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X32: movzbl
; X32: movzbl
; X32: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X32: lock
; X32: cmpxchgb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_umin8(i8 %x) nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_umin8:
; X32-LABEL: atomic_fetch_umin8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw umin i8* @sc8, i8 %x acquire
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64: movzbl
; X64: movzbl
; X64: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X64: lock
; X64: cmpxchgb
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X32: movzbl
; X32: movzbl
; X32: subl
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X32: lock
; X32: cmpxchgb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_cmpxchg8() nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_cmpxchg8:
; X32-LABEL: atomic_fetch_cmpxchg8:
IR: add a second ordering operand to cmpxhg for failure The syntax for "cmpxchg" should now look something like: cmpxchg i32* %addr, i32 42, i32 3 acquire monotonic where the second ordering argument gives the required semantics in the case that no exchange takes place. It should be no stronger than the first ordering constraint and cannot be either "release" or "acq_rel" (since no store will have taken place). rdar://problem/15996804 llvm-svn: 203559
2014-03-11 18:48:52 +08:00
%t1 = cmpxchg i8* @sc8, i8 0, i8 1 acquire acquire
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
; X64: lock
; X64: cmpxchgb
; X32: lock
; X32: cmpxchgb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_store8(i8 %x) nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_store8:
; X32-LABEL: atomic_fetch_store8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
store atomic i8 %x, i8* @sc8 release, align 4
; X64-NOT: lock
; X64: movb
; X32-NOT: lock
; X32: movb
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_swap8(i8 %x) nounwind {
X86: expand atomics in IR instead of as MachineInstrs. The logic for expanding atomics that aren't natively supported in terms of cmpxchg loops is much simpler to express at the IR level. It also allows the normal optimisations and CodeGen improvements to help out with atomics, instead of using a limited set of possible instructions.. rdar://problem/13496295 llvm-svn: 212119
2014-07-02 02:53:31 +08:00
; X64-LABEL: atomic_fetch_swap8:
; X32-LABEL: atomic_fetch_swap8:
Re-work X86 code generation of atomic ops with spin-loop - Rewrite/merge pseudo-atomic instruction emitters to address the following issue: * Reduce one unnecessary load in spin-loop previously the spin-loop looks like thisMBB: newMBB: ld t1 = [bitinstr.addr] op t2 = t1, [bitinstr.val] not t3 = t2 (if Invert) mov EAX = t1 lcs dest = [bitinstr.addr], t3 [EAX is implicit] bz newMBB fallthrough -->nextMBB the 'ld' at the beginning of newMBB should be lift out of the loop as lcs (or CMPXCHG on x86) will load the current memory value into EAX. This loop is refined as: thisMBB: EAX = LOAD [MI.addr] mainMBB: t1 = OP [MI.val], EAX LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined] JNE mainMBB sinkMBB: * Remove immopc as, so far, all pseudo-atomic instructions has all-register form only, there is no immedidate operand. * Remove unnecessary attributes/modifiers in pseudo-atomic instruction td * Fix issues in PR13458 - Add comprehensive tests on atomic ops on various data types. NOTE: Some of them are turned off due to missing functionality. - Revise tests due to the new spin-loop generated. llvm-svn: 164281
2012-09-20 11:06:15 +08:00
%t1 = atomicrmw xchg i8* @sc8, i8 %x acquire
; X64-NOT: lock
; X64: xchgb
; X32-NOT: lock
; X32: xchgb
ret void
; X64: ret
; X32: ret
}