diff options
| author | Craig Topper <craig.topper@intel.com> | 2017-10-14 05:55:42 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@intel.com> | 2017-10-14 05:55:42 +0000 |
| commit | ee277e190c514228131a83b7bb80c8628295189c (patch) | |
| tree | 5fe13368f429b76a19bb146fde44787b17dbdab5 /llvm/lib | |
| parent | aec05a9303d70d1441f269e9988c764dff89c842 (diff) | |
| download | bcm5719-llvm-ee277e190c514228131a83b7bb80c8628295189c.tar.gz bcm5719-llvm-ee277e190c514228131a83b7bb80c8628295189c.zip | |
[X86] Add patterns for vzmovl+cvtpd2ps with a load.
llvm-svn: 315800
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrAVX512.td | 12 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrSSE.td | 24 |
2 files changed, 24 insertions, 12 deletions
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td index 9af9ad85929..f68516a34ce 100644 --- a/llvm/lib/Target/X86/X86InstrAVX512.td +++ b/llvm/lib/Target/X86/X86InstrAVX512.td @@ -6651,10 +6651,14 @@ def : Pat<(v8f64 (extloadv8f32 addr:$src)), (VCVTPS2PDZrm addr:$src)>; let Predicates = [HasVLX] in { - let AddedComplexity = 15 in - def : Pat<(X86vzmovl (v2f64 (bitconvert - (v4f32 (X86vfpround (v2f64 VR128X:$src)))))), - (VCVTPD2PSZ128rr VR128X:$src)>; + let AddedComplexity = 15 in { + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86vfpround (v2f64 VR128X:$src)))))), + (VCVTPD2PSZ128rr VR128X:$src)>; + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86vfpround (loadv2f64 addr:$src)))))), + (VCVTPD2PSZ128rm addr:$src)>; + } def : Pat<(v2f64 (extloadv2f32 addr:$src)), (VCVTPS2PDZ128rm addr:$src)>; def : Pat<(v4f64 (extloadv4f32 addr:$src)), diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td index 13a418a623b..556c64c4ea0 100644 --- a/llvm/lib/Target/X86/X86InstrSSE.td +++ b/llvm/lib/Target/X86/X86InstrSSE.td @@ -2054,18 +2054,26 @@ def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), let Predicates = [HasAVX, NoVLX] in { // Match fpround and fpextend for 128/256-bit conversions - let AddedComplexity = 15 in - def : Pat<(X86vzmovl (v2f64 (bitconvert - (v4f32 (X86vfpround (v2f64 VR128:$src)))))), - (VCVTPD2PSrr VR128:$src)>; + let AddedComplexity = 15 in { + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86vfpround (v2f64 VR128:$src)))))), + (VCVTPD2PSrr VR128:$src)>; + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86vfpround (loadv2f64 addr:$src)))))), + (VCVTPD2PSrm addr:$src)>; + } } let Predicates = [UseSSE2] in { // Match fpround and fpextend for 128 conversions - let AddedComplexity = 15 in - def : Pat<(X86vzmovl (v2f64 (bitconvert - (v4f32 (X86vfpround (v2f64 VR128:$src)))))), - (CVTPD2PSrr VR128:$src)>; + let AddedComplexity = 15 in { + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86vfpround (v2f64 VR128:$src)))))), + (CVTPD2PSrr VR128:$src)>; + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86vfpround (loadv2f64 addr:$src)))))), + (CVTPD2PSrm addr:$src)>; + } } //===----------------------------------------------------------------------===// |
