[X86] Unsigned saturation subtraction canonicalization [the backend p…

…art]

Summary:
On behalf of julia.koval@intel.com

The patch transforms canonical version of unsigned saturation, which is sub(max(a,b),a) or sub(a,min(a,b)) to special psubus insturuction on targets, which support it(8bit and 16bit uints).
umax(a,b) - b -> subus(a,b)
a - umin(a,b) -> subus(a,b)

There is also extra case handled, when right part of sub is 32 bit and can be truncated, using UMIN(this transformation was discussed in https://reviews.llvm.org/D25987).

The example of special case code:

```
void foo(unsigned short *p, int max, int n) {

  int i;
  unsigned m;
  for (i = 0; i < n; i++) {
    m = *--p;
    *p = (unsigned short)(m >= max ? m-max : 0);
  }
}
```
Max in this example is truncated to max_short value, if it is greater than m, or just truncated to 16 bit, if it is not. It is vaid transformation, because if max > max_short, result of the expression will be zero.

Here is the table of types, I try to support, special case items are bold:

| Size | 128 | 256 | 512
| -----  | -----  | -----   | -----
| i8 | v16i8 | v32i8 | v64i8
| i16 | v8i16 | v16i16 | v32i16
| i32 | | **v8i32** | **v16i32**
| i64 | | | **v8i64**

Reviewers: zvi, spatel, DavidKreitzer, RKSimon

Reviewed By: zvi

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D37534

llvm-svn: 315237

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -35901,6 +35901,89 @@ static SDValue combineAdd(SDNode *N, SelectionDAG &DAG,
   return combineAddOrSubToADCOrSBB(N, DAG);
 }
 
+static SDValue combineSubToSubus(SDNode *N, SelectionDAG &DAG,
+                                 const X86Subtarget &Subtarget) {
+  SDValue Op0 = N->getOperand(0);
+  SDValue Op1 = N->getOperand(1);
+  EVT VT = N->getValueType(0);
+
+  // PSUBUS is supported, starting from SSE2, but special preprocessing
+  // for v8i32 requires umin, which appears in SSE41.
+  if (!(Subtarget.hasSSE2() && (VT == MVT::v16i8 || VT == MVT::v8i16)) &&
+      !(Subtarget.hasSSE41() && (VT == MVT::v8i32)) &&
+      !(Subtarget.hasAVX2() && (VT == MVT::v32i8 || VT == MVT::v16i16)) &&
+      !(Subtarget.hasAVX512() && Subtarget.hasBWI() &&
+        (VT == MVT::v64i8 || VT == MVT::v32i16 || VT == MVT::v16i32 ||
+         VT == MVT::v8i64)))
+    return SDValue();
+
+  SDValue SubusLHS, SubusRHS;
+  // Try to find umax(a,b) - b or a - umin(a,b) patterns
+  // they may be converted to subus(a,b).
+  // TODO: Need to add IR cannonicialization for this code.
+  if (Op0.getOpcode() == ISD::UMAX) {
+    SubusRHS = Op1;
+    SDValue MaxLHS = Op0.getOperand(0);
+    SDValue MaxRHS = Op0.getOperand(1);
+    if (DAG.isEqualTo(MaxLHS, Op1))
+      SubusLHS = MaxRHS;
+    else if (DAG.isEqualTo(MaxRHS, Op1))
+      SubusLHS = MaxLHS;
+    else
+      return SDValue();
+  } else if (Op1.getOpcode() == ISD::UMIN) {
+    SubusLHS = Op0;
+    SDValue MinLHS = Op1.getOperand(0);
+    SDValue MinRHS = Op1.getOperand(1);
+    if (DAG.isEqualTo(MinLHS, Op0))
+      SubusRHS = MinRHS;
+    else if (DAG.isEqualTo(MinRHS, Op0))
+      SubusRHS = MinLHS;
+    else
+      return SDValue();
+  } else
+    return SDValue();
+
+  // PSUBUS doesn't support v8i32/v8i64/v16i32, but it can be enabled with
+  // special preprocessing in some cases.
+  if (VT != MVT::v8i32 && VT != MVT::v16i32 && VT != MVT::v8i64)
+    return DAG.getNode(X86ISD::SUBUS, SDLoc(N), VT, SubusLHS, SubusRHS);
+
+  // Special preprocessing case can be only applied
+  // if the value was zero extended from 16 bit,
+  // so we require first 16 bits to be zeros for 32 bit
+  // values, or first 48 bits for 64 bit values.
+  KnownBits Known;
+  DAG.computeKnownBits(SubusLHS, Known);
+  unsigned NumZeros = Known.countMinLeadingZeros();
+  if ((VT == MVT::v8i64 && NumZeros < 48) || NumZeros < 16)
+    return SDValue();
+
+  EVT ExtType = SubusLHS.getValueType();
+  EVT ShrinkedType;
+  if (VT == MVT::v8i32 || VT == MVT::v8i64)
+    ShrinkedType = MVT::v8i16;
+  else
+    ShrinkedType = NumZeros >= 24 ? MVT::v16i8 : MVT::v16i16;
+
+  // If SubusLHS is zeroextended - truncate SubusRHS to it's
+  // size SubusRHS = umin(0xFFF.., SubusRHS).
+  SDValue SaturationConst =
+      DAG.getConstant(APInt::getLowBitsSet(ExtType.getScalarSizeInBits(),
+                                           ShrinkedType.getScalarSizeInBits()),
+                      SDLoc(SubusLHS), ExtType);
+  SDValue UMin = DAG.getNode(ISD::UMIN, SDLoc(SubusLHS), ExtType, SubusRHS,
+                             SaturationConst);
+  SDValue NewSubusLHS =
+      DAG.getZExtOrTrunc(SubusLHS, SDLoc(SubusLHS), ShrinkedType);
+  SDValue NewSubusRHS = DAG.getZExtOrTrunc(UMin, SDLoc(SubusRHS), ShrinkedType);
+  SDValue Psubus = DAG.getNode(X86ISD::SUBUS, SDLoc(N), ShrinkedType,
+                               NewSubusLHS, NewSubusRHS);
+  // Zero extend the result, it may be used somewhere as 32 bit,
+  // if not zext and following trunc will shrink.
+  return DAG.getZExtOrTrunc(Psubus, SDLoc(N), ExtType);
+}
+
 static SDValue combineSub(SDNode *N, SelectionDAG &DAG,
                           const X86Subtarget &Subtarget) {
   SDValue Op0 = N->getOperand(0);
@@ -35934,6 +36017,10 @@ static SDValue combineSub(SDNode *N, SelectionDAG &DAG,
   if (SDValue V = combineIncDecVector(N, DAG))
     return V;
 
+  // Try to create PSUBUS if SUB's argument is max/min
+  if (SDValue V = combineSubToSubus(N, DAG, Subtarget))
+    return V;
+
   return combineAddOrSubToADCOrSBB(N, DAG);
 }