Fix the MASM asmprinter's lies. It does not want to emit code to .text/.data

[oota-llvm.git] / lib / Target / X86 / X86InstrSSE.td

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index d86a6cf83059522a53a0fbb6fb35fce2d061bea8..3c1db6714621fa785b504562692aadd2c9918e82 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -526,7 +526,7 @@ def CVTSS2SDrr: I<0x5A, MRMSrcReg, (ops FR64:$dst, FR32:$src),
                 Requires<[HasSSE2]>;
 def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops FR64:$dst, f32mem:$src),
                  "cvtss2sd {$src, $dst|$dst, $src}",
-                 [(set FR64:$dst, (fextend (loadf32 addr:$src)))]>, XS,
+                 [(set FR64:$dst, (extload addr:$src, f32))]>, XS,
                 Requires<[HasSSE2]>;
 
 // Match intrinsics which expect XMM operand(s).
@@ -1988,7 +1988,7 @@ def PUNPCKHDQrm : PDI<0x6A, MRMSrcMem,
                                 UNPCKH_shuffle_mask)))]>;
 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg, 
                        (ops VR128:$dst, VR128:$src1, VR128:$src2),
-                       "punpckhdq {$src2, $dst|$dst, $src2}",
+                       "punpckhqdq {$src2, $dst|$dst, $src2}",
                       [(set VR128:$dst,
                         (v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
                                 UNPCKH_shuffle_mask)))]>;
@@ -2481,6 +2481,14 @@ def : Pat<(v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
                   MOVL_shuffle_mask)),
           (v2i64 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
 
+// vector_shuffle v1, v2 <4, 5, 2, 3> using MOVLPDrr (movsd)
+def : Pat<(v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
+                  MOVLP_shuffle_mask)),
+          (v4f32 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
+def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
+                  MOVLP_shuffle_mask)),
+          (v4i32 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
+
 // Set lowest element and zero upper elements.
 def : Pat<(bc_v2i64 (vector_shuffle immAllZerosV,
                      (v2f64 (scalar_to_vector (loadf64 addr:$src))),