[X86] Fix some patterns that select VLX instructions, but were incorrectly also check...

The EVEX->VEX pass probably obscures this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@315365 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index aeb08676bf988ceb94a259d758a11b37c556f237..499440ea48fefca87fc8dffd05c8b1b511939700 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -1339,12 +1339,14 @@ let Predicates = [HasAVX512] in {
             (VPBROADCASTQZm addr:$src)>;
 }
 
-let Predicates = [HasVLX, HasBWI] in {
+let Predicates = [HasVLX] in {
   // 32-bit targets will fail to load a i64 directly but can use ZEXT_LOAD.
   def : Pat<(v2i64 (X86VBroadcast (v2i64 (X86vzload addr:$src)))),
             (VPBROADCASTQZ128m addr:$src)>;
   def : Pat<(v4i64 (X86VBroadcast (v4i64 (X86vzload addr:$src)))),
             (VPBROADCASTQZ256m addr:$src)>;
+}
+let Predicates = [HasVLX, HasBWI] in {
   // loadi16 is tricky to fold, because !isTypeDesirableForOp, justifiably.
   // This means we'll encounter truncated i32 loads; match that here.
   def : Pat<(v8i16 (X86VBroadcast (i16 (trunc (i32 (load addr:$src)))))),

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index bbaa6f4df07409632627328958bc8ac46ea31c6e..c2f086428518ddf62dd228f7c4f88c8703e83b68 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -7855,12 +7855,14 @@ defm VPBROADCASTD  : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
 defm VPBROADCASTQ  : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
                                     v2i64, v4i64, NoVLX>;
 
-let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
+let Predicates = [HasAVX2, NoVLX] in {
   // 32-bit targets will fail to load a i64 directly but can use ZEXT_LOAD.
   def : Pat<(v2i64 (X86VBroadcast (v2i64 (X86vzload addr:$src)))),
             (VPBROADCASTQrm addr:$src)>;
   def : Pat<(v4i64 (X86VBroadcast (v4i64 (X86vzload addr:$src)))),
             (VPBROADCASTQYrm addr:$src)>;
+}
+let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
   // loadi16 is tricky to fold, because !isTypeDesirableForOp, justifiably.
   // This means we'll encounter truncated i32 loads; match that here.
   def : Pat<(v8i16 (X86VBroadcast (i16 (trunc (i32 (load addr:$src)))))),