[X86] SimplifyDemandedBitsForTargetNode - add X86ISD::ANDNP handling

Add X86ISD::ANDNP handling to targetShrinkDemandedConstant as well, which allows us to replace a lot of truncated masks with (rematerializable) allones values

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -38669,14 +38669,14 @@ X86TargetLowering::targetShrinkDemandedConstant(SDValue Op,
       return false;
     };
     // For vectors - if we have a constant, then try to sign extend.
-    // TODO: Handle AND/ANDN cases.
+    // TODO: Handle AND cases.
     unsigned ActiveBits = DemandedBits.getActiveBits();
     if (EltSize > ActiveBits && EltSize > 1 && isTypeLegal(VT) &&
-        (Opcode == ISD::OR || Opcode == ISD::XOR) &&
+        (Opcode == ISD::OR || Opcode == ISD::XOR || Opcode == X86ISD::ANDNP) &&
         NeedsSignExtension(Op.getOperand(1), ActiveBits)) {
       EVT ExtSVT = EVT::getIntegerVT(*TLO.DAG.getContext(), ActiveBits);
       EVT ExtVT = EVT::getVectorVT(*TLO.DAG.getContext(), ExtSVT,
-                                    VT.getVectorNumElements());
+                                   VT.getVectorNumElements());
       SDValue NewC =
           TLO.DAG.getNode(ISD::SIGN_EXTEND_INREG, SDLoc(Op), VT,
                           Op.getOperand(1), TLO.DAG.getValueType(ExtVT));
@@ -43818,6 +43818,31 @@ bool X86TargetLowering::SimplifyDemandedBitsForTargetNode(
     }
     break;
   }
+  case X86ISD::ANDNP: {
+    KnownBits Known2;
+    SDValue Op0 = Op.getOperand(0);
+    SDValue Op1 = Op.getOperand(1);
+
+    if (SimplifyDemandedBits(Op1, OriginalDemandedBits, OriginalDemandedElts,
+                             Known, TLO, Depth + 1))
+      return true;
+    assert(!Known.hasConflict() && "Bits known to be one AND zero?");
+
+    if (SimplifyDemandedBits(Op0, ~Known.Zero & OriginalDemandedBits,
+                             OriginalDemandedElts, Known2, TLO, Depth + 1))
+      return true;
+    assert(!Known2.hasConflict() && "Bits known to be one AND zero?");
+
+    // If the RHS is a constant, see if we can simplify it.
+    if (ShrinkDemandedConstant(Op, ~Known2.One & OriginalDemandedBits,
+                               OriginalDemandedElts, TLO))
+      return true;
+
+    // ANDNP = (~Op0 & Op1);
+    Known.One &= Known2.Zero;
+    Known.Zero |= Known2.One;
+    break;
+  }
   case X86ISD::VSHLI: {
     SDValue Op0 = Op.getOperand(0);
 

llvm/test/CodeGen/X86/fpclamptosat_vec.ll

@@ -82,8 +82,9 @@ define <2 x i32> @utest_f64i32(<2 x double> %x) {
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
 ; CHECK-NEXT:    pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; CHECK-NEXT:    pand %xmm3, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
 ; CHECK-NEXT:    pand %xmm0, %xmm1
-; CHECK-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    pandn %xmm2, %xmm0
 ; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; CHECK-NEXT:    retq
@@ -270,31 +271,31 @@ define <4 x i32> @utest_f32i32(<4 x float> %x) {
 ; CHECK-NEXT:    orq %rax, %rdx
 ; CHECK-NEXT:    movq %rdx, %xmm0
 ; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [4294967295,4294967295]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
-; CHECK-NEXT:    movdqa %xmm0, %xmm4
-; CHECK-NEXT:    pxor %xmm3, %xmm4
-; CHECK-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm6 = [2147483647,2147483647,2147483647,2147483647]
-; CHECK-NEXT:    movdqa %xmm6, %xmm7
-; CHECK-NEXT:    pcmpgtd %xmm5, %xmm7
-; CHECK-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; CHECK-NEXT:    pcmpeqd %xmm3, %xmm4
-; CHECK-NEXT:    pand %xmm7, %xmm4
-; CHECK-NEXT:    pand %xmm4, %xmm0
-; CHECK-NEXT:    pandn %xmm2, %xmm4
-; CHECK-NEXT:    por %xmm0, %xmm4
+; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
+; CHECK-NEXT:    movdqa %xmm0, %xmm3
+; CHECK-NEXT:    pxor %xmm2, %xmm3
+; CHECK-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
+; CHECK-NEXT:    movdqa {{.*#+}} xmm5 = [2147483647,2147483647,2147483647,2147483647]
+; CHECK-NEXT:    movdqa %xmm5, %xmm6
+; CHECK-NEXT:    pcmpgtd %xmm4, %xmm6
+; CHECK-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; CHECK-NEXT:    pcmpeqd %xmm2, %xmm3
+; CHECK-NEXT:    pand %xmm6, %xmm3
+; CHECK-NEXT:    pcmpeqd %xmm4, %xmm4
+; CHECK-NEXT:    pand %xmm3, %xmm0
+; CHECK-NEXT:    pandn %xmm4, %xmm3
+; CHECK-NEXT:    por %xmm0, %xmm3
 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
-; CHECK-NEXT:    pxor %xmm3, %xmm0
-; CHECK-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[0,0,2,2]
-; CHECK-NEXT:    pcmpgtd %xmm5, %xmm6
+; CHECK-NEXT:    pxor %xmm2, %xmm0
+; CHECK-NEXT:    pshufd {{.*#+}} xmm6 = xmm0[0,0,2,2]
+; CHECK-NEXT:    pcmpgtd %xmm6, %xmm5
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; CHECK-NEXT:    pcmpeqd %xmm3, %xmm0
-; CHECK-NEXT:    pand %xmm6, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm2, %xmm0
+; CHECK-NEXT:    pand %xmm5, %xmm0
 ; CHECK-NEXT:    pand %xmm0, %xmm1
-; CHECK-NEXT:    pandn %xmm2, %xmm0
+; CHECK-NEXT:    pandn %xmm4, %xmm0
 ; CHECK-NEXT:    por %xmm1, %xmm0
-; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm4[0,2]
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
 ; CHECK-NEXT:    retq
 entry:
   %conv = fptoui <4 x float> %x to <4 x i64>
@@ -550,32 +551,32 @@ define <4 x i32> @utesth_f16i32(<4 x half> %x) {
 ; CHECK-NEXT:    movq %rdx, %xmm0
 ; CHECK-NEXT:    punpcklqdq (%rsp), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; CHECK-NEXT:    movdqa %xmm0, %xmm3
-; CHECK-NEXT:    pxor %xmm2, %xmm3
-; CHECK-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm5 = [2147483647,2147483647,2147483647,2147483647]
-; CHECK-NEXT:    movdqa %xmm5, %xmm6
-; CHECK-NEXT:    pcmpgtd %xmm4, %xmm6
-; CHECK-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; CHECK-NEXT:    pcmpeqd %xmm2, %xmm3
-; CHECK-NEXT:    pand %xmm6, %xmm3
-; CHECK-NEXT:    pand %xmm3, %xmm0
-; CHECK-NEXT:    pandn %xmm1, %xmm3
-; CHECK-NEXT:    por %xmm0, %xmm3
+; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    pxor %xmm1, %xmm2
+; CHECK-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
+; CHECK-NEXT:    movdqa {{.*#+}} xmm4 = [2147483647,2147483647,2147483647,2147483647]
+; CHECK-NEXT:    movdqa %xmm4, %xmm5
+; CHECK-NEXT:    pcmpgtd %xmm3, %xmm5
+; CHECK-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm2
+; CHECK-NEXT:    pand %xmm5, %xmm2
+; CHECK-NEXT:    pcmpeqd %xmm3, %xmm3
+; CHECK-NEXT:    pand %xmm2, %xmm0
+; CHECK-NEXT:    pandn %xmm3, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
 ; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
 ; CHECK-NEXT:    movdqa %xmm6, %xmm0
-; CHECK-NEXT:    pxor %xmm2, %xmm0
-; CHECK-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[0,0,2,2]
-; CHECK-NEXT:    pcmpgtd %xmm4, %xmm5
+; CHECK-NEXT:    pxor %xmm1, %xmm0
+; CHECK-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[0,0,2,2]
+; CHECK-NEXT:    pcmpgtd %xmm5, %xmm4
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; CHECK-NEXT:    pcmpeqd %xmm2, %xmm0
-; CHECK-NEXT:    pand %xmm5, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-NEXT:    pand %xmm4, %xmm0
 ; CHECK-NEXT:    pand %xmm0, %xmm6
-; CHECK-NEXT:    pandn %xmm1, %xmm0
+; CHECK-NEXT:    pandn %xmm3, %xmm0
 ; CHECK-NEXT:    por %xmm6, %xmm0
-; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
 ; CHECK-NEXT:    addq $72, %rsp
 ; CHECK-NEXT:    .cfi_def_cfa_offset 8
 ; CHECK-NEXT:    retq
@@ -733,8 +734,9 @@ define <2 x i16> @utest_f64i16(<2 x double> %x) {
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = <2147549183,2147549183,u,u>
 ; CHECK-NEXT:    pcmpgtd %xmm1, %xmm2
 ; CHECK-NEXT:    andpd %xmm2, %xmm0
-; CHECK-NEXT:    andnpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
-; CHECK-NEXT:    orpd %xmm0, %xmm2
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT:    pandn %xmm1, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
 ; CHECK-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,2,2,3,4,5,6,7]
 ; CHECK-NEXT:    retq
 entry:
@@ -802,7 +804,8 @@ define <4 x i16> @utest_f32i16(<4 x float> %x) {
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [2147549183,2147549183,2147549183,2147549183]
 ; CHECK-NEXT:    pcmpgtd %xmm1, %xmm2
 ; CHECK-NEXT:    pand %xmm2, %xmm0
-; CHECK-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT:    pandn %xmm1, %xmm2
 ; CHECK-NEXT:    por %xmm0, %xmm2
 ; CHECK-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,2,2,3,4,5,6,7]
 ; CHECK-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
@@ -1656,8 +1659,9 @@ define <2 x i32> @utest_f64i32_mm(<2 x double> %x) {
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
 ; CHECK-NEXT:    pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; CHECK-NEXT:    pand %xmm3, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
 ; CHECK-NEXT:    pand %xmm0, %xmm1
-; CHECK-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    pandn %xmm2, %xmm0
 ; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; CHECK-NEXT:    retq
@@ -1849,7 +1853,7 @@ define <4 x i32> @utest_f32i32_mm(<4 x float> %x) {
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
 ; CHECK-NEXT:    pcmpeqd %xmm2, %xmm3
 ; CHECK-NEXT:    pand %xmm6, %xmm3
-; CHECK-NEXT:    movdqa {{.*#+}} xmm4 = [4294967295,4294967295]
+; CHECK-NEXT:    pcmpeqd %xmm4, %xmm4
 ; CHECK-NEXT:    pand %xmm3, %xmm0
 ; CHECK-NEXT:    pandn %xmm4, %xmm3
 ; CHECK-NEXT:    por %xmm0, %xmm3
@@ -2124,7 +2128,7 @@ define <4 x i32> @utesth_f16i32_mm(<4 x half> %x) {
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; CHECK-NEXT:    pcmpeqd %xmm1, %xmm2
 ; CHECK-NEXT:    pand %xmm5, %xmm2
-; CHECK-NEXT:    movdqa {{.*#+}} xmm3 = [4294967295,4294967295]
+; CHECK-NEXT:    pcmpeqd %xmm3, %xmm3
 ; CHECK-NEXT:    pand %xmm2, %xmm0
 ; CHECK-NEXT:    pandn %xmm3, %xmm2
 ; CHECK-NEXT:    por %xmm0, %xmm2
@@ -2292,8 +2296,9 @@ define <2 x i16> @utest_f64i16_mm(<2 x double> %x) {
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = <2147549183,2147549183,u,u>
 ; CHECK-NEXT:    pcmpgtd %xmm1, %xmm2
 ; CHECK-NEXT:    andpd %xmm2, %xmm0
-; CHECK-NEXT:    andnpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
-; CHECK-NEXT:    orpd %xmm0, %xmm2
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT:    pandn %xmm1, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
 ; CHECK-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,2,2,3,4,5,6,7]
 ; CHECK-NEXT:    retq
 entry:
@@ -2356,7 +2361,8 @@ define <4 x i16> @utest_f32i16_mm(<4 x float> %x) {
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [2147549183,2147549183,2147549183,2147549183]
 ; CHECK-NEXT:    pcmpgtd %xmm1, %xmm2
 ; CHECK-NEXT:    pand %xmm2, %xmm0
-; CHECK-NEXT:    pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT:    pandn %xmm1, %xmm2
 ; CHECK-NEXT:    por %xmm0, %xmm2
 ; CHECK-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,2,2,3,4,5,6,7]
 ; CHECK-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]