The vbroadcast family of instructions has 'fallback patterns' in case where the

load source operand is used by multiple nodes. The v2i64 broadcast was emulated
by shuffling the two lower i32 elements to the upper two.
We had a bug in the immediate used for the broadcast.
Replacing 0 to 0x44.
0x44 means [01|00|01|00] which corresponds to the correct lane.

Patch by Michael Kuperstein.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@160430 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 10cc483..c2d169a 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -7781,11 +7781,12 @@ def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
   def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
             (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
               (VPSHUFDri
-                (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd), 0),
-                  sub_xmm),
+                (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd),
+                0x44),
+              sub_xmm),
               (VPSHUFDri
                 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd),
-              0), 1)>;
+                0x44), 1)>;
 
   def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
             (VPSHUFDri
@@ -7801,11 +7802,12 @@ def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
   def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
             (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
               (VPSHUFDri
-                (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd), 0),
-                  sub_xmm),
+                (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd),
+                0x44),
+              sub_xmm),
               (VPSHUFDri
                 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd),
-              0), 1)>;
+                0x44), 1)>;
   }
 }