[PATCH] D120116: [SDAG] enable binop identity constant folds for sub

[LuoYuanke via Phabricator via llvm-commits]

LuoYuanke created this revision.
Herald added subscribers: ecnelises, pengfei, hiraditya.
LuoYuanke requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

This patch extract the sub folding from D119654 <https://reviews.llvm.org/D119654> and leave only add
folding int that patch.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D120116

Files:
  llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
  llvm/test/CodeGen/X86/vector-bo-select.ll


Index: llvm/test/CodeGen/X86/vector-bo-select.ll
===================================================================
--- llvm/test/CodeGen/X86/vector-bo-select.ll
+++ llvm/test/CodeGen/X86/vector-bo-select.ll
@@ -915,8 +915,8 @@
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm2, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpsubd %xmm2, %xmm1, %xmm1 {%k1}
+; AVX512VL-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VL-NEXT:    retq
   %s = select <4 x i1> %b, <4 x i32> %y, <4 x i32> zeroinitializer
   %r = sub <4 x i32> %x, %s
@@ -979,9 +979,9 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; AVX512-NEXT:    vmovdqa32 %zmm2, %zmm0 {%k1} {z}
-; AVX512-NEXT:    vpsubd %zmm0, %zmm1, %zmm0
+; AVX512-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; AVX512-NEXT:    vpsubd %zmm2, %zmm1, %zmm0
+; AVX512-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512-NEXT:    retq
   %s = select <16 x i1> %b, <16 x i32> zeroinitializer, <16 x i32> %y
   %r = sub <16 x i32> %x, %s
@@ -1044,8 +1044,7 @@
 ; AVX512VL-LABEL: sub_v8i32_cast_cond:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    kmovw %edi, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm1, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsubd %ymm1, %ymm0, %ymm0 {%k1}
 ; AVX512VL-NEXT:    retq
   %b = bitcast i8 %pb to <8 x i1>
   %s = select <8 x i1> %b, <8 x i32> %y, <8 x i32> zeroinitializer
@@ -1110,8 +1109,7 @@
 ; AVX512-LABEL: sub_v8i64_cast_cond:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    kmovw %edi, %k1
-; AVX512-NEXT:    vmovdqa64 %zmm1, %zmm1 {%k1} {z}
-; AVX512-NEXT:    vpsubq %zmm1, %zmm0, %zmm0
+; AVX512-NEXT:    vpsubq %zmm1, %zmm0, %zmm0 {%k1}
 ; AVX512-NEXT:    retq
   %b = bitcast i8 %pb to <8 x i1>
   %s = select <8 x i1> %b, <8 x i64> %y, <8 x i64> zeroinitializer
Index: llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -2144,6 +2144,12 @@
         return C->isExactlyValue(1.0);
       }
     }
+    if (ConstantSDNode *C = isConstOrConstSplat(V)) {
+      switch (Opcode) {
+      case ISD::SUB: // X - 0 --> X
+        return C->isZero();
+      }
+    }
     return false;
   };
 
@@ -3326,9 +3332,15 @@
   EVT VT = N0.getValueType();
   SDLoc DL(N);
 
+  auto PeekThroughFreeze = [](SDValue N) {
+    if (N->getOpcode() == ISD::FREEZE && N.hasOneUse())
+      return N->getOperand(0);
+    return N;
+  };
+
   // fold (sub x, x) -> 0
   // FIXME: Refactor this and xor and other similar operations together.
-  if (N0 == N1)
+  if (PeekThroughFreeze(N0) == PeekThroughFreeze(N1))
     return tryFoldToZero(DL, TLI, VT, DAG, LegalOperations);
 
   // fold (sub c1, c2) -> c3


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D120116.409901.patch
Type: text/x-patch
Size: 3008 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220218/989ee887/attachment.bin>