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/atomic64.ll

236 lines
5.3 KiB
LLVM
Raw Normal View History

X86: Do not use llc -march in tests. `llc -march` is problematic because it only switches the target architecture, but leaves the operating system unchanged. This occasionally leads to indeterministic tests because the OS from LLVM_DEFAULT_TARGET_TRIPLE is used. However we can simply always use `llc -mtriple` instead. This changes all the tests to do this to avoid people using -march when they copy and paste parts of tests. See also the discussion in https://reviews.llvm.org/D35287 llvm-svn: 309774
2017-08-02 08:28:10 +08:00
; RUN: llc < %s -O0 -mtriple=x86_64-- -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X64
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
@sc64 = external global i64
define void @atomic_fetch_add64() 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_add64:
; X32-LABEL: atomic_fetch_add64:
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:
%t1 = atomicrmw add i64* @sc64, i64 1 acquire
; X64: lock
; X64: incq
%t2 = atomicrmw add i64* @sc64, i64 3 acquire
; X64: lock
; X64: addq $3
%t3 = atomicrmw add i64* @sc64, i64 5 acquire
; X64: lock
; X64: xaddq
%t4 = atomicrmw add i64* @sc64, i64 %t3 acquire
; X64: lock
; X64: addq
ret void
; X64: ret
}
define void @atomic_fetch_sub64() 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_sub64:
; X32-LABEL: atomic_fetch_sub64:
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 i64* @sc64, i64 1 acquire
; X64: lock
; X64: decq
%t2 = atomicrmw sub i64* @sc64, i64 3 acquire
; X64: lock
; X64: subq $3
%t3 = atomicrmw sub i64* @sc64, i64 5 acquire
; X64: lock
; X64: xaddq
%t4 = atomicrmw sub i64* @sc64, i64 %t3 acquire
; X64: lock
; X64: subq
ret void
; X64: ret
}
define void @atomic_fetch_and64() 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_and64:
; X32-LABEL: atomic_fetch_and64:
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 i64* @sc64, i64 3 acquire
; X64: lock
; X64: andq $3
%t2 = atomicrmw and i64* @sc64, i64 5 acquire
[X86] Don't use GR64 register 'and with immediate' instructions if the immediate is zero in the upper 33-bits or upper 57-bits. Use GR32 instructions instead. Previously the patterns didn't have high enough priority and we would only use the GR32 form if the only the upper 32 or 56 bits were zero. Fixes PR23100. llvm-svn: 234075
2015-04-04 10:08:20 +08:00
; X64: andl
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: cmpxchgq
%t3 = atomicrmw and i64* @sc64, i64 %t2 acquire
; X64: lock
; X64: andq
ret void
; X64: ret
}
define void @atomic_fetch_or64() 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_or64:
; X32-LABEL: atomic_fetch_or64:
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 i64* @sc64, i64 3 acquire
; X64: lock
; X64: orq $3
%t2 = atomicrmw or i64* @sc64, i64 5 acquire
; X64: orq
; X64: lock
; X64: cmpxchgq
%t3 = atomicrmw or i64* @sc64, i64 %t2 acquire
; X64: lock
; X64: orq
ret void
; X64: ret
}
define void @atomic_fetch_xor64() 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_xor64:
; X32-LABEL: atomic_fetch_xor64:
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 i64* @sc64, i64 3 acquire
; X64: lock
; X64: xorq $3
%t2 = atomicrmw xor i64* @sc64, i64 5 acquire
; X64: xorq
; X64: lock
; X64: cmpxchgq
%t3 = atomicrmw xor i64* @sc64, i64 %t2 acquire
; X64: lock
; X64: xorq
ret void
; X64: ret
}
define void @atomic_fetch_nand64(i64 %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_nand64:
; X32-LABEL: atomic_fetch_nand64:
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 i64* @sc64, i64 %x acquire
; X64: andq
; X64: notq
; X64: lock
; X64: cmpxchgq
; X32: andl
; X32: andl
; X32: notl
; X32: notl
; X32: lock
; X32: cmpxchg8b
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_max64(i64 %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_max64:
; X32-LABEL: atomic_fetch_max64:
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 i64* @sc64, i64 %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: subq
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: cmov
; X64: lock
; X64: cmpxchgq
; X32: cmpl
; X32: cmpl
; X32: cmov
; X32: cmov
; X32: cmov
; X32: lock
; X32: cmpxchg8b
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_min64(i64 %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_min64:
; X32-LABEL: atomic_fetch_min64:
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 i64* @sc64, i64 %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: subq
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: cmov
; X64: lock
; X64: cmpxchgq
; X32: cmpl
; X32: cmpl
; X32: cmov
; X32: cmov
; X32: cmov
; X32: lock
; X32: cmpxchg8b
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_umax64(i64 %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_umax64:
; X32-LABEL: atomic_fetch_umax64:
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 i64* @sc64, i64 %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: subq
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: cmov
; X64: lock
; X64: cmpxchgq
; X32: cmpl
; X32: cmpl
; X32: cmov
; X32: cmov
; X32: cmov
; X32: lock
; X32: cmpxchg8b
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_umin64(i64 %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_umin64:
; X32-LABEL: atomic_fetch_umin64:
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 i64* @sc64, i64 %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: subq
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: cmov
; X64: lock
; X64: cmpxchgq
; X32: cmpl
; X32: cmpl
; X32: cmov
; X32: cmov
; X32: cmov
; X32: lock
; X32: cmpxchg8b
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_cmpxchg64() 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_cmpxchg64:
; X32-LABEL: atomic_fetch_cmpxchg64:
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 i64* @sc64, i64 0, i64 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: cmpxchgq
; X32: lock
; X32: cmpxchg8b
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_store64(i64 %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_store64:
; X32-LABEL: atomic_fetch_store64:
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 i64 %x, i64* @sc64 release, align 8
; X64-NOT: lock
; X64: movq
; X32: lock
; X32: cmpxchg8b
ret void
; X64: ret
; X32: ret
}
define void @atomic_fetch_swap64(i64 %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_swap64:
; X32-LABEL: atomic_fetch_swap64:
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 i64* @sc64, i64 %x acquire
; X64-NOT: lock
; X64: xchgq
; X32: lock
; X32: xchg8b
ret void
; X64: ret
; X32: ret
}