Changes in directory llvm/lib/Target/X86:
X86InstrSSE.td updated: 1.66 -> 1.67
---
Log message:
Added patterns for MOVHPSmr and MOVLPSmr.
---
Diffs of the changes: (+42 -4)
X86InstrSSE.td | 46 ++++++++++++++++++++++++++++++++++++++++++----
1 files changed, 42 insertions(+), 4 deletions(-)
Index: llvm/lib/Target/X86/X86InstrSSE.td
diff -u llvm/lib/Target/X86/X86InstrSSE.td:1.66
llvm/lib/Target/X86/X86InstrSSE.td:1.67
--- llvm/lib/Target/X86/X86InstrSSE.td:1.66 Thu Apr 6 18:53:29 2006
+++ llvm/lib/Target/X86/X86InstrSSE.td Fri Apr 7 16:20:58 2006
@@ -746,14 +746,23 @@
}
def MOVLPSmr : PSI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
- "movlps {$src, $dst|$dst, $src}", []>;
+ "movlps {$src, $dst|$dst, $src}",
+ [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
+ (i32 0))), addr:$dst)]>;
def MOVLPDmr : PDI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
"movlpd {$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (v2f64 VR128:$src),
(i32 0))), addr:$dst)]>;
+// v2f64 extract element 1 is always custom lowered to unpack high to low
+// and extract element 0 so the non-store version isn't too horrible.
def MOVHPSmr : PSI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
- "movhps {$src, $dst|$dst, $src}", []>;
+ "movhps {$src, $dst|$dst, $src}",
+ [(store (f64 (vector_extract
+ (v2f64 (vector_shuffle
+ (bc_v2f64 (v4f32 VR128:$src)),
(undef),
+ UNPCKH_shuffle_mask)), (i32 0))),
+ addr:$dst)]>;
def MOVHPDmr : PDI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
"movhpd {$src, $dst|$dst, $src}",
[(store (f64 (vector_extract
@@ -1703,31 +1712,60 @@
Requires<[HasSSE2]>;
def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>,
Requires<[HasSSE2]>;
+def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>,
+ Requires<[HasSSE2]>;
def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
+def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>,
+ Requires<[HasSSE2]>;
def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
+def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>,
+ Requires<[HasSSE2]>;
def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>,
Requires<[HasSSE2]>;
-
-def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>,
+def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>,
Requires<[HasSSE2]>;
def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>,
Requires<[HasSSE2]>;
+def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>,
+ Requires<[HasSSE2]>;
def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>,
Requires<[HasSSE2]>;
+def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>,
+ Requires<[HasSSE2]>;
+def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>,
+ Requires<[HasSSE2]>;
// Zeroing a VR128 then do a MOVS* to the lower bits.
def : Pat<(v2f64 (X86zexts2vec FR64:$src)),
_______________________________________________
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
