[X86] Make sure load isn't volatile before shrinking it in MOVDDUP isel patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@365275 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 78ce2e339eccaf0b8d28a76808e06eaa5e516457..56aa2ecffe071486edfaa73412f8f47d2fe15ff8 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -10871,7 +10871,7 @@ def : Pat<(v2f64 (X86VBroadcast (loadf64 addr:$src))),
           (VMOVDDUPZ128rm addr:$src)>;
 def : Pat<(v2f64 (X86VBroadcast f64:$src)),
           (VMOVDDUPZ128rr (v2f64 (COPY_TO_REGCLASS FR64X:$src, VR128X)))>;
-def : Pat<(v2f64 (X86VBroadcast (loadv2f64 addr:$src))),
+def : Pat<(v2f64 (X86VBroadcast (v2f64 (nonvolatile_load addr:$src)))),
           (VMOVDDUPZ128rm addr:$src)>;
 def : Pat<(v2f64 (X86VBroadcast (v2f64 (X86vzload addr:$src)))),
           (VMOVDDUPZ128rm addr:$src)>;
@@ -10891,10 +10891,10 @@ def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (loadf64 addr:$src)
                    immAllZerosV),
           (VMOVDDUPZ128rmkz VK2WM:$mask, addr:$src)>;
 
-def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (loadv2f64 addr:$src))),
+def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (v2f64 (nonvolatile_load addr:$src)))),
                    (v2f64 VR128X:$src0)),
           (VMOVDDUPZ128rmk VR128X:$src0, VK2WM:$mask, addr:$src)>;
-def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (loadv2f64 addr:$src))),
+def : Pat<(vselect (v2i1 VK2WM:$mask), (v2f64 (X86VBroadcast (v2f64 (nonvolatile_load addr:$src)))),
                    immAllZerosV),
           (VMOVDDUPZ128rmkz VK2WM:$mask, addr:$src)>;
 }

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index ea14fb0600a7dd6421816e2c2dfe467637c906fa..f904f5a23d5f4b9d0a0dc06c890a922fa334438b 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -4364,7 +4364,7 @@ defm MOVDDUP : sse3_replicate_dfp<"movddup", SchedWriteFShuffle>;
 
 
 let Predicates = [HasAVX, NoVLX] in {
-  def : Pat<(X86Movddup (loadv2f64 addr:$src)),
+  def : Pat<(X86Movddup (v2f64 (nonvolatile_load addr:$src))),
             (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
   def : Pat<(X86Movddup (v2f64 (X86vzload addr:$src))),
             (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
@@ -4372,7 +4372,7 @@ let Predicates = [HasAVX, NoVLX] in {
 
 let Predicates = [UseSSE3] in {
   // No need for aligned memory as this only loads 64-bits.
-  def : Pat<(X86Movddup (loadv2f64 addr:$src)),
+  def : Pat<(X86Movddup (v2f64 (nonvolatile_load addr:$src))),
             (MOVDDUPrm addr:$src)>;
   def : Pat<(X86Movddup (v2f64 (X86vzload addr:$src))),
             (MOVDDUPrm addr:$src)>;