RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-09 21:32:49 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Following icc's lead: use movdqa to load / store 128-bit integer vectors

Browse Source 
llvm-svn: 26980
This commit is contained in:
Evan Cheng 2006-03-23 07:44:07 +00:00
parent 3629ca4268
commit a6dc6e535d
1 changed files with 29 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
lib/Target/X86/X86InstrSSE.td
View File

@ -40,6 +40,10 @@ def X86loadpf64 : PatFrag<(ops node:$ptr), (f64 (X86loadp node:$ptr))>;
def loadv4f32 : PatFrag<(ops node:$ptr), (v4f32 (load node:$ptr))>;
def loadv2f64 : PatFrag<(ops node:$ptr), (v2f64 (load node:$ptr))>;
def loadv16i8 : PatFrag<(ops node:$ptr), (v16i8 (load node:$ptr))>;
def loadv8i16 : PatFrag<(ops node:$ptr), (v8i16 (load node:$ptr))>;
def loadv4i32 : PatFrag<(ops node:$ptr), (v4i32 (load node:$ptr))>;
def loadv2i64 : PatFrag<(ops node:$ptr), (v2i64 (load node:$ptr))>;
// SHUFFLE_get_shuf_imm xform function: convert vector_shuffle mask to PSHUF*,
// SHUFP* etc. imm.
@ -750,7 +754,7 @@ def UNPCKLPDrm : PDI<0x14, MRMSrcMem,
// Move Instructions
def MOVD128rr : PDI<0x6E, MRMSrcReg, (ops VR128:$dst, R32:$src),
"movd {$src, $dst|$dst, $src}",
"movd {$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v4i32 (scalar_to_vector R32:$src)))]>;
def MOVD128rm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
@ -758,6 +762,15 @@ def MOVD128rm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
def MOVD128mr : PDI<0x7E, MRMDestMem, (ops i32mem:$dst, VR128:$src),
"movd {$src, $dst|$dst, $src}", []>;
def MOVDQArr : PDI<0x6F, MRMSrcReg, (ops VR128:$dst, VR128:$src),
"movdqa {$src, $dst|$dst, $src}", []>;
def MOVDQArm : PDI<0x6F, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
"movdqa {$src, $dst|$dst, $src}",
[(set VR128:$dst, (loadv4i32 addr:$src))]>;
def MOVDQAmr : PDI<0x7F, MRMDestMem, (ops i128mem:$dst, VR128:$src),
"movdqa {$src, $dst|$dst, $src}",
[(store (v4i32 VR128:$src), addr:$dst)]>;
// SSE2 instructions with XS prefix
def MOVQ128rr : I<0x7E, MRMSrcReg, (ops VR128:$dst, VR64:$src),
"movq {$src, $dst|$dst, $src}",
@ -823,24 +836,24 @@ def : Pat<(v4i32 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
def : Pat<(v2i64 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
// Load 128-bit integer vector values.
def : Pat<(v16i8 (load addr:$src)), (MOVAPSrm addr:$src)>,
Requires<[HasSSE1]>;
def : Pat<(v8i16 (load addr:$src)), (MOVAPSrm addr:$src)>,
Requires<[HasSSE1]>;
def : Pat<(v4i32 (load addr:$src)), (MOVAPSrm addr:$src)>,
Requires<[HasSSE1]>;
def : Pat<(v2i64 (load addr:$src)), (MOVAPDrm addr:$src)>,
def : Pat<(v16i8 (load addr:$src)), (MOVDQArm addr:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v8i16 (load addr:$src)), (MOVDQArm addr:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v4i32 (load addr:$src)), (MOVDQArm addr:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v2i64 (load addr:$src)), (MOVDQArm addr:$src)>,
Requires<[HasSSE2]>;
// Store 128-bit integer vector values.
def : Pat<(store (v16i8 VR128:$src), addr:$dst), (MOVAPSmr addr:$dst, VR128:$src)>,
Requires<[HasSSE1]>;
def : Pat<(store (v8i16 VR128:$src), addr:$dst), (MOVAPSmr addr:$dst, VR128:$src)>,
Requires<[HasSSE1]>;
def : Pat<(store (v4i32 VR128:$src), addr:$dst), (MOVAPSmr addr:$dst, VR128:$src)>,
Requires<[HasSSE1]>;
def : Pat<(store (v2i64 VR128:$src), addr:$dst), (MOVAPDmr addr:$dst, VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(store (v16i8 VR128:$src), addr:$dst),
(MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE1]>;
def : Pat<(store (v8i16 VR128:$src), addr:$dst),
(MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE1]>;
def : Pat<(store (v4i32 VR128:$src), addr:$dst),
(MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE1]>;
def : Pat<(store (v2i64 VR128:$src), addr:$dst),
(MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
// Scalar to v8i16 / v16i8. The source may be a R32, but only the lower 8 or
// 16-bits matter.