[X86] Remove isel patterns that matched vXi16 X86VBroadcast with i8->i16 aextload input.

This was selecting VBROADCASTW which turned the 8-bit load into
a 16-bit load if it happened to be 2 byte aligned.

I have a plan to fix the regression with a follow up patch
which I'll post shortly.

Author: topperc

llvm/lib/Target/X86/X86InstrAVX512.td

@@ -1442,12 +1442,6 @@ let Predicates = [HasVLX, HasBWI] in {
   def : Pat<(v16i16 (X86VBroadcast
               (i16 (trunc (i32 (zextloadi16 addr:$src)))))),
             (VPBROADCASTWZ256rm addr:$src)>;
-
-  // FIXME this is to handle aligned extloads from i8.
-  def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
-            (VPBROADCASTWZ128rm addr:$src)>;
-  def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
-            (VPBROADCASTWZ256rm addr:$src)>;
 }
 let Predicates = [HasBWI] in {
   // loadi16 is tricky to fold, because !isTypeDesirableForOp, justifiably.
@@ -1460,10 +1454,6 @@ let Predicates = [HasBWI] in {
   def : Pat<(v32i16 (X86VBroadcast
               (i16 (trunc (i32 (zextloadi16 addr:$src)))))),
             (VPBROADCASTWZrm addr:$src)>;
-
-  // FIXME this is to handle aligned extloads from i8.
-  def : Pat<(v32i16 (X86VBroadcast (loadi16 addr:$src))),
-            (VPBROADCASTWZrm addr:$src)>;
 }
 
 //===----------------------------------------------------------------------===//

llvm/lib/Target/X86/X86InstrSSE.td

@@ -7533,12 +7533,6 @@ let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
   def : Pat<(v16i16 (X86VBroadcast
               (i16 (trunc (i32 (zextloadi16 addr:$src)))))),
             (VPBROADCASTWYrm addr:$src)>;
-
-  // FIXME this is to handle aligned extloads from i8.
-  def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
-            (VPBROADCASTWrm addr:$src)>;
-  def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
-            (VPBROADCASTWYrm addr:$src)>;
 }
 
 let Predicates = [HasAVX2, NoVLX] in {

llvm/test/CodeGen/X86/vector-sext.ll

@@ -2039,7 +2039,9 @@ define <8 x i16> @load_sext_8i1_to_8i16(<8 x i1> *%ptr) {
 ;
 ; AVX2-LABEL: load_sext_8i1_to_8i16:
 ; AVX2:       # %bb.0: # %entry
-; AVX2-NEXT:    vpbroadcastw (%rdi), %xmm0
+; AVX2-NEXT:    movzwl (%rdi), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm0
+; AVX2-NEXT:    vpbroadcastw %xmm0, %xmm0
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,2,4,8,16,32,64,128]
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0