[llvm] [X86] Compress masked VPBLENDM using VBLENDV (PR #221838)

[Simon Pilgrim via llvm-commits]

================
@@ -314,6 +314,21 @@ static bool isCompressibleBlendVUse(unsigned BlendOpc, unsigned UseOpc) {
   }
 }
 
+static bool isCompressibleMaskedBlendUse(unsigned BlendOpc, unsigned UseOpc) {
+  switch (BlendOpc) {
+  case X86::VBLENDVPSrrr:
+    return UseOpc == X86::VPBLENDMDZ128rrk;
+  case X86::VBLENDVPSYrrr:
+    return UseOpc == X86::VPBLENDMDZ256rrk;
+  case X86::VBLENDVPDrrr:
+    return UseOpc == X86::VPBLENDMQZ128rrk;
+  case X86::VBLENDVPDYrrr:
+    return UseOpc == X86::VPBLENDMQZ256rrk;
----------------
RKSimon wrote:

Can we use the PS/PD variants as well? VBLENDMPDZ256rrk etc.

https://github.com/llvm/llvm-project/pull/221838