6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-26 05:32:25 +00:00
Code Issues Projects Releases Wiki Activity
8838410a75
llvm-6502/test/CodeGen/X86/atomic_add.ll

203 lines
3.9 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) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176538 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-06 00:17:04 +00:00
; RUN: llc < %s -march=x86-64 -verify-machineinstrs | FileCheck %s
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
; rdar://7103704
define void @sub1(i32* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: sub1:
; CHECK: subl
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i32* %p, i32 %v monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @inc4(i64* nocapture %p) nounwind ssp {
entry:
; CHECK: inc4:
; CHECK: incq
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i64* %p, i64 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add8(i64* nocapture %p) nounwind ssp {
entry:
; CHECK: add8:
; CHECK: addq $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i64* %p, i64 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add4(i64* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: add4:
; CHECK: addq
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = sext i32 %v to i64 ; <i64> [#uses=1]
%1 = atomicrmw add i64* %p, i64 %0 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @inc3(i8* nocapture %p) nounwind ssp {
entry:
; CHECK: inc3:
; CHECK: incb
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i8* %p, i8 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add7(i8* nocapture %p) nounwind ssp {
entry:
; CHECK: add7:
; CHECK: addb $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i8* %p, i8 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add3(i8* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: add3:
; CHECK: addb
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = trunc i32 %v to i8 ; <i8> [#uses=1]
%1 = atomicrmw add i8* %p, i8 %0 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @inc2(i16* nocapture %p) nounwind ssp {
entry:
; CHECK: inc2:
; CHECK: incw
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i16* %p, i16 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add6(i16* nocapture %p) nounwind ssp {
entry:
; CHECK: add6:
; CHECK: addw $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i16* %p, i16 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add2(i16* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: add2:
; CHECK: addw
%0 = trunc i32 %v to i16 ; <i16> [#uses=1]
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%1 = atomicrmw add i16* %p, i16 %0 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @inc1(i32* nocapture %p) nounwind ssp {
entry:
; CHECK: inc1:
; CHECK: incl
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i32* %p, i32 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add5(i32* nocapture %p) nounwind ssp {
entry:
; CHECK: add5:
; CHECK: addl $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i32* %p, i32 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @add1(i32* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: add1:
; CHECK: addl
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw add i32* %p, i32 %v monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @dec4(i64* nocapture %p) nounwind ssp {
entry:
; CHECK: dec4:
; CHECK: decq
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i64* %p, i64 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @sub8(i64* nocapture %p) nounwind ssp {
entry:
; CHECK: sub8:
; CHECK: subq $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i64* %p, i64 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @sub4(i64* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: sub4:
; CHECK: subq
%0 = sext i32 %v to i64 ; <i64> [#uses=1]
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%1 = atomicrmw sub i64* %p, i64 %0 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @dec3(i8* nocapture %p) nounwind ssp {
entry:
; CHECK: dec3:
; CHECK: decb
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i8* %p, i8 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @sub7(i8* nocapture %p) nounwind ssp {
entry:
; CHECK: sub7:
; CHECK: subb $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i8* %p, i8 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @sub3(i8* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: sub3:
; CHECK: subb
%0 = trunc i32 %v to i8 ; <i8> [#uses=1]
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%1 = atomicrmw sub i8* %p, i8 %0 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @dec2(i16* nocapture %p) nounwind ssp {
entry:
; CHECK: dec2:
; CHECK: decw
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i16* %p, i16 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @sub6(i16* nocapture %p) nounwind ssp {
entry:
; CHECK: sub6:
; CHECK: subw $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i16* %p, i16 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @sub2(i16* nocapture %p, i32 %v) nounwind ssp {
entry:
; CHECK: sub2:
Unify the logic in SelectAtomicLoadAdd and SelectAtomicLoadArith - Merge the processing of LOAD_ADD with other atomic load-arith operations - Separate the logic getting target constant for atomic-load-op and add an optimization for atomic-load-add on i16 with negative value - Optimize a minor case for atomic-fetch-add i16 with negative operand. Test case is revised. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@164243 91177308-0d34-0410-b5e6-96231b3b80d8
2012-09-19 19:36:58 +00:00
; CHECK-NOT: negl
; CHECK: subw
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
%0 = trunc i32 %v to i16 ; <i16> [#uses=1]
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%1 = atomicrmw sub i16* %p, i16 %0 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @dec1(i32* nocapture %p) nounwind ssp {
entry:
; CHECK: dec1:
; CHECK: decl
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i32* %p, i32 1 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
define void @sub5(i32* nocapture %p) nounwind ssp {
entry:
; CHECK: sub5:
; CHECK: subl $2
Convert a bunch more tests over to the new atomic instructions. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140582 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 23:15:09 +00:00
%0 = atomicrmw sub i32* %p, i32 2 monotonic
ret void
Optimize some common usage patterns of atomic built-ins __sync_add_and_fetch() and __sync_sub_and_fetch. When the return value is not used (i.e. only care about the value in the memory), x86 does not have to use add to implement these. Instead, it can use add, sub, inc, dec instructions with the "lock" prefix. This is currently implemented using a bit of instruction selection trick. The issue is the target independent pattern produces one output and a chain and we want to map it into one that just output a chain. The current trick is to select it into a merge_values with the first definition being an implicit_def. The proper solution is to add new ISD opcodes for the no-output variant. DAG combiner can then transform the node before it gets to target node selection. Problem #2 is we are adding a whole bunch of x86 atomic instructions when in fact these instructions are identical to the non-lock versions. We need a way to add target specific information to target nodes and have this information carried over to machine instructions. Asm printer (or JIT) can use this information to add the "lock" prefix. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77582 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-30 08:33:02 +00:00
}
Reference in New Issue Copy Permalink