diff options
| author | Alexey Volkov <avolkov.intel@gmail.com> | 2014-06-09 11:40:41 +0000 |
|---|---|---|
| committer | Alexey Volkov <avolkov.intel@gmail.com> | 2014-06-09 11:40:41 +0000 |
| commit | a2bc6951a028412e7af5e68aa369daa4611147b3 (patch) | |
| tree | e36b5b05d696ec346157589ad1e2ecb702e02123 /lib/Target/X86/X86InstrCompiler.td | |
| parent | a8d18fe94687b65f136b7222c2cc998c2148e39c (diff) | |
| download | llvm-a2bc6951a028412e7af5e68aa369daa4611147b3.tar.gz llvm-a2bc6951a028412e7af5e68aa369daa4611147b3.tar.bz2 llvm-a2bc6951a028412e7af5e68aa369daa4611147b3.tar.xz | |
[X86] Use ADD/SUB instead of INC/DEC for Silvermont
According to Intel Software Optimization Manual
on Silvermont INC or DEC instructions require
an additional uop to merge the flags.
As a result, a branch instruction depending
on an INC or a DEC instruction incurs a 1 cycle penalty.
Differential Revision: http://reviews.llvm.org/D3990
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210466 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/X86/X86InstrCompiler.td')
| -rw-r--r-- | lib/Target/X86/X86InstrCompiler.td | 38 |
1 files changed, 26 insertions, 12 deletions
diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td index 34d8fb9e18..9b3dce52f7 100644 --- a/lib/Target/X86/X86InstrCompiler.td +++ b/lib/Target/X86/X86InstrCompiler.td @@ -1696,20 +1696,34 @@ def : Pat<(mul (loadi64 addr:$src1), i64immSExt32:$src2), (IMUL64rmi32 addr:$src1, i64immSExt32:$src2)>; // Increment reg. -def : Pat<(add GR8 :$src, 1), (INC8r GR8 :$src)>; -def : Pat<(add GR16:$src, 1), (INC16r GR16:$src)>, Requires<[Not64BitMode]>; -def : Pat<(add GR16:$src, 1), (INC64_16r GR16:$src)>, Requires<[In64BitMode]>; -def : Pat<(add GR32:$src, 1), (INC32r GR32:$src)>, Requires<[Not64BitMode]>; -def : Pat<(add GR32:$src, 1), (INC64_32r GR32:$src)>, Requires<[In64BitMode]>; -def : Pat<(add GR64:$src, 1), (INC64r GR64:$src)>; +// Do not make INC if it is slow +def : Pat<(add GR8:$src, 1), + (INC8r GR8:$src)>, Requires<[NotSlowIncDec]>; +def : Pat<(add GR16:$src, 1), + (INC16r GR16:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>; +def : Pat<(add GR16:$src, 1), + (INC64_16r GR16:$src)>, Requires<[NotSlowIncDec, In64BitMode]>; +def : Pat<(add GR32:$src, 1), + (INC32r GR32:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>; +def : Pat<(add GR32:$src, 1), + (INC64_32r GR32:$src)>, Requires<[NotSlowIncDec, In64BitMode]>; +def : Pat<(add GR64:$src, 1), + (INC64r GR64:$src)>, Requires<[NotSlowIncDec]>; // Decrement reg. -def : Pat<(add GR8 :$src, -1), (DEC8r GR8 :$src)>; -def : Pat<(add GR16:$src, -1), (DEC16r GR16:$src)>, Requires<[Not64BitMode]>; -def : Pat<(add GR16:$src, -1), (DEC64_16r GR16:$src)>, Requires<[In64BitMode]>; -def : Pat<(add GR32:$src, -1), (DEC32r GR32:$src)>, Requires<[Not64BitMode]>; -def : Pat<(add GR32:$src, -1), (DEC64_32r GR32:$src)>, Requires<[In64BitMode]>; -def : Pat<(add GR64:$src, -1), (DEC64r GR64:$src)>; +// Do not make DEC if it is slow +def : Pat<(add GR8:$src, -1), + (DEC8r GR8:$src)>, Requires<[NotSlowIncDec]>; +def : Pat<(add GR16:$src, -1), + (DEC16r GR16:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>; +def : Pat<(add GR16:$src, -1), + (DEC64_16r GR16:$src)>, Requires<[NotSlowIncDec, In64BitMode]>; +def : Pat<(add GR32:$src, -1), + (DEC32r GR32:$src)>, Requires<[NotSlowIncDec, Not64BitMode]>; +def : Pat<(add GR32:$src, -1), + (DEC64_32r GR32:$src)>, Requires<[NotSlowIncDec, In64BitMode]>; +def : Pat<(add GR64:$src, -1), + (DEC64r GR64:$src)>, Requires<[NotSlowIncDec]>; // or reg/reg. def : Pat<(or GR8 :$src1, GR8 :$src2), (OR8rr GR8 :$src1, GR8 :$src2)>; |