[PATCH] D22225: [x86, SSE] optimize pcmp results better (PR28484)

[Elena Demikhovsky via llvm-commits]

delena added a subscriber: delena.

================
Comment at: lib/Target/X86/X86ISelLowering.cpp:28139
@@ +28138,3 @@
+  SDValue Op1 = peekThroughBitcasts(N->getOperand(1));
+  if (Op0.getOpcode() != X86ISD::PCMPEQ && Op0.getOpcode() != X86ISD::PCMPGT)
+    return SDValue();
----------------
We cam mark nodes like PCMP with AssertSext. And use this marker to simplify AND.

================
Comment at: lib/Target/X86/X86ISelLowering.cpp:28148
@@ +28147,3 @@
+  unsigned EltBitWidth = VT0.getScalarType().getSizeInBits();
+  if (VT0 != VT1 || EltBitWidth == 8)
+    return SDValue();
----------------
VT0 is always equal to VT1.

================
Comment at: lib/Target/X86/X86ISelLowering.cpp:28152
@@ +28151,3 @@
+  // TODO: Is this possible or worthwhile for AVX-512 variants?
+  if (VT0.getSizeInBits() != 128 && VT0.getSizeInBits() != 256)
+    return SDValue();
----------------
On AVX-512 (skylake-avx512) the result is in a mask reg, also for 256 and 128 vector inputs.


http://reviews.llvm.org/D22225