[llvm-commits] CVS: llvm/lib/Target/X86/X86ISelLowering.cpp X86ISelLowering.h X86InstrSSE.td

[Evan Cheng]

Changes in directory llvm/lib/Target/X86:

X86ISelLowering.cpp updated: 1.280 -> 1.281
X86ISelLowering.h updated: 1.76 -> 1.77
X86InstrSSE.td updated: 1.166 -> 1.167
---
Log message:

Fixed a significant bug where unpcklpd is incorrectly used to extract element 1 from a v2f64 value.

---
Diffs of the changes:  (+25 -6)

 X86ISelLowering.cpp |   11 +++++++++++
 X86ISelLowering.h   |    4 ++++
 X86InstrSSE.td      |   16 ++++++++++------
 3 files changed, 25 insertions(+), 6 deletions(-)


Index: llvm/lib/Target/X86/X86ISelLowering.cpp
diff -u llvm/lib/Target/X86/X86ISelLowering.cpp:1.280 llvm/lib/Target/X86/X86ISelLowering.cpp:1.281
--- llvm/lib/Target/X86/X86ISelLowering.cpp:1.280	Fri Oct 27 13:49:08 2006
+++ llvm/lib/Target/X86/X86ISelLowering.cpp	Fri Oct 27 16:08:32 2006
@@ -2868,6 +2868,17 @@
   return ::isSplatMask(N);
 }
 
+/// isSplatLoMask - Return true if the specified VECTOR_SHUFFLE operand
+/// specifies a splat of zero element.
+bool X86::isSplatLoMask(SDNode *N) {
+  assert(N->getOpcode() == ISD::BUILD_VECTOR);
+
+  for (unsigned i = 0, e = N->getNumOperands(); i < e; ++i) 
+    if (!isUndefOrEqual(N->getOperand(i), 0))
+      return false;
+  return true;
+}
+
 /// getShuffleSHUFImmediate - Return the appropriate immediate to shuffle
 /// the specified isShuffleMask VECTOR_SHUFFLE mask with PSHUF* and SHUFP*
 /// instructions.


Index: llvm/lib/Target/X86/X86ISelLowering.h
diff -u llvm/lib/Target/X86/X86ISelLowering.h:1.76 llvm/lib/Target/X86/X86ISelLowering.h:1.77
--- llvm/lib/Target/X86/X86ISelLowering.h:1.76	Fri Oct 20 12:42:20 2006
+++ llvm/lib/Target/X86/X86ISelLowering.h	Fri Oct 27 16:08:32 2006
@@ -225,6 +225,10 @@
    /// specifies a splat of a single element.
    bool isSplatMask(SDNode *N);
 
+   /// isSplatLoMask - Return true if the specified VECTOR_SHUFFLE operand
+   /// specifies a splat of zero element.
+   bool isSplatLoMask(SDNode *N);
+
    /// getShuffleSHUFImmediate - Return the appropriate immediate to shuffle
    /// the specified isShuffleMask VECTOR_SHUFFLE mask with PSHUF* and SHUFP*
    /// instructions.


Index: llvm/lib/Target/X86/X86InstrSSE.td
diff -u llvm/lib/Target/X86/X86InstrSSE.td:1.166 llvm/lib/Target/X86/X86InstrSSE.td:1.167
--- llvm/lib/Target/X86/X86InstrSSE.td:1.166	Wed Oct 25 16:35:05 2006
+++ llvm/lib/Target/X86/X86InstrSSE.td	Fri Oct 27 16:08:32 2006
@@ -104,8 +104,8 @@
   return X86::isSplatMask(N);
 }], SHUFFLE_get_shuf_imm>;
 
-def SSE_splat_v2_mask : PatLeaf<(build_vector), [{
-  return X86::isSplatMask(N);
+def SSE_splat_lo_mask : PatLeaf<(build_vector), [{
+  return X86::isSplatLoMask(N);
 }]>;
 
 def MOVHLPS_shuffle_mask : PatLeaf<(build_vector), [{
@@ -812,13 +812,13 @@
                       "movddup {$src, $dst|$dst, $src}",
                   [(set VR128:$dst, (v2f64 (vector_shuffle
                                             VR128:$src, (undef),
-                                            SSE_splat_v2_mask)))]>;
+                                            SSE_splat_lo_mask)))]>;
 def MOVDDUPrm : S3DI<0x12, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
                   "movddup {$src, $dst|$dst, $src}",
                   [(set VR128:$dst, (v2f64 (vector_shuffle
                                          (scalar_to_vector (loadf64 addr:$src)),
                                              (undef),
-                                            SSE_splat_v2_mask)))]>;
+                                            SSE_splat_lo_mask)))]>;
 
 // SSE2 instructions without OpSize prefix
 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
@@ -1908,10 +1908,14 @@
 
 // Splat v2f64 / v2i64
 let AddedComplexity = 10 in {
-def : Pat<(vector_shuffle (v2f64 VR128:$src), (undef), SSE_splat_v2_mask:$sm),
+def : Pat<(vector_shuffle (v2f64 VR128:$src), (undef), SSE_splat_lo_mask:$sm),
           (UNPCKLPDrr VR128:$src, VR128:$src)>,   Requires<[HasSSE2]>;
-def : Pat<(vector_shuffle (v2i64 VR128:$src), (undef), SSE_splat_v2_mask:$sm),
+def : Pat<(vector_shuffle (v2f64 VR128:$src), (undef), UNPCKH_shuffle_mask:$sm),
+          (UNPCKHPDrr VR128:$src, VR128:$src)>,   Requires<[HasSSE2]>;
+def : Pat<(vector_shuffle (v2i64 VR128:$src), (undef), SSE_splat_lo_mask:$sm),
           (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
+def : Pat<(vector_shuffle (v2i64 VR128:$src), (undef), UNPCKH_shuffle_mask:$sm),
+          (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
 }
 
 // Splat v4f32