Skip to content

Commit

Permalink
[SDAG] allow vector types for select->logic folds
Browse files Browse the repository at this point in the history 
This prepares codegen for a change that will remove the identical
folds from IR because they are not poison-safe. See
D93065 / D97360
for details.

We already generically support scalar types, and there are various
target-specific transforms that overlap the vector folds. For example,
x86 recognizes the and patterns, but not or. We can end up with 1
extra instruction there, but I think that is still preferred over the
blendv alternative that loads a constant vector.

If this is not optimal, then it should be fixed with a later transform
(this change is not expected to result in any regressions because
InstCombine currently does the same thing).

Removing custom code and supporting undefs in constant-pattern-matching
can be follow-up changes.

Differential Revision: https://reviews.llvm.org/D97730
  • Loading branch information
@rotateright
rotateright committed Mar 2, 2021
1 parent 4096ae0 commit 7fce332
Show file tree
Hide file tree
Showing 3 changed files with 20 additions and 22 deletions.
16 changes: 10 additions & 6 deletions llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View file
Expand Up @@ -9302,31 +9302,32 @@ SDValue DAGCombiner::foldSelectOfConstants(SDNode *N) {
}

static SDValue foldBoolSelectToLogic(SDNode *N, SelectionDAG &DAG) {
assert(N->getOpcode() == ISD::SELECT && "Expected a select");
assert((N->getOpcode() == ISD::SELECT || N->getOpcode() == ISD::VSELECT) &&
"Expected a (v)select");
SDValue Cond = N->getOperand(0);
SDValue T = N->getOperand(1), F = N->getOperand(2);
EVT VT = N->getValueType(0);
if (VT != Cond.getValueType() || VT != MVT::i1)
if (VT != Cond.getValueType() || VT.getScalarSizeInBits() != 1)
return SDValue();

// select Cond, Cond, F --> or Cond, F
// select Cond, 1, F --> or Cond, F
if (Cond == T || isOneConstant(T))
if (Cond == T || isOneOrOneSplat(T))
return DAG.getNode(ISD::OR, SDLoc(N), VT, Cond, F);

// select Cond, T, Cond --> and Cond, T
// select Cond, T, 0 --> and Cond, T
if (Cond == F || isNullConstant(F))
if (Cond == F || isNullOrNullSplat(F))
return DAG.getNode(ISD::AND, SDLoc(N), VT, Cond, T);

// select Cond, T, 1 --> or (not Cond), T
if (isOneConstant(F)) {
if (isOneOrOneSplat(F)) {
SDValue NotCond = DAG.getNOT(SDLoc(N), Cond, VT);
return DAG.getNode(ISD::OR, SDLoc(N), VT, NotCond, T);
}

// select Cond, 0, F --> and (not Cond), F
if (isNullConstant(T)) {
if (isNullOrNullSplat(T)) {
SDValue NotCond = DAG.getNOT(SDLoc(N), Cond, VT);
return DAG.getNode(ISD::AND, SDLoc(N), VT, NotCond, F);
}
Expand Down Expand Up @@ -9788,6 +9789,9 @@ SDValue DAGCombiner::visitVSELECT(SDNode *N) {
if (SDValue V = DAG.simplifySelect(N0, N1, N2))
return V;

if (SDValue V = foldBoolSelectToLogic(N, DAG))
return V;

// vselect (not Cond), N1, N2 -> vselect Cond, N2, N1
if (SDValue F = extractBooleanFlip(N0, DAG, TLI, false))
return DAG.getSelect(DL, VT, F, N2, N1);
Expand Down
15 changes: 4 additions & 11 deletions llvm/test/CodeGen/AArch64/select-with-and-or.ll
View file
Expand Up @@ -66,7 +66,7 @@ define <4 x i1> @and_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32> %w)
; CHECK: // %bb.0:
; CHECK-NEXT: cmeq v0.4s, v0.4s, v1.4s
; CHECK-NEXT: cmgt v1.4s, v2.4s, v3.4s
; CHECK-NEXT: and v0.16b, v1.16b, v0.16b
; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: ret
%a = icmp eq <4 x i32> %x, %y
Expand All @@ -80,10 +80,8 @@ define <4 x i1> @or_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32> %w)
; CHECK: // %bb.0:
; CHECK-NEXT: cmeq v0.4s, v0.4s, v1.4s
; CHECK-NEXT: cmgt v1.4s, v2.4s, v3.4s
; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: xtn v1.4h, v1.4s
; CHECK-NEXT: movi v2.4h, #1
; CHECK-NEXT: bsl v0.8b, v2.8b, v1.8b
; CHECK-NEXT: ret
%a = icmp eq <4 x i32> %x, %y
%b = icmp sgt <4 x i32> %z, %w
Expand All @@ -96,9 +94,8 @@ define <4 x i1> @and_not_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32>
; CHECK: // %bb.0:
; CHECK-NEXT: cmeq v0.4s, v0.4s, v1.4s
; CHECK-NEXT: cmgt v1.4s, v2.4s, v3.4s
; CHECK-NEXT: bic v0.16b, v1.16b, v0.16b
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: xtn v1.4h, v1.4s
; CHECK-NEXT: bic v0.8b, v1.8b, v0.8b
; CHECK-NEXT: ret
%a = icmp eq <4 x i32> %x, %y
%b = icmp sgt <4 x i32> %z, %w
Expand All @@ -111,12 +108,8 @@ define <4 x i1> @or_not_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32>
; CHECK: // %bb.0:
; CHECK-NEXT: cmeq v0.4s, v0.4s, v1.4s
; CHECK-NEXT: cmgt v1.4s, v2.4s, v3.4s
; CHECK-NEXT: movi v2.4h, #1
; CHECK-NEXT: xtn v3.4h, v0.4s
; CHECK-NEXT: and v0.16b, v1.16b, v0.16b
; CHECK-NEXT: orn v0.16b, v1.16b, v0.16b
; CHECK-NEXT: xtn v0.4h, v0.4s
; CHECK-NEXT: bic v1.8b, v2.8b, v3.8b
; CHECK-NEXT: orr v0.8b, v0.8b, v1.8b
; CHECK-NEXT: ret
%a = icmp eq <4 x i32> %x, %y
%b = icmp sgt <4 x i32> %z, %w
Expand Down
11 changes: 6 additions & 5 deletions llvm/test/CodeGen/X86/select-with-and-or.ll
View file
Expand Up @@ -233,7 +233,7 @@ define <4 x i1> @or_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32> %w)
; CHECK: # %bb.0:
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vpcmpgtd %xmm3, %xmm2, %xmm1
; CHECK-NEXT: vblendvps %xmm0, {{.*}}(%rip), %xmm1, %xmm0
; CHECK-NEXT: vpor %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
%a = icmp eq <4 x i32> %x, %y
%b = icmp sgt <4 x i32> %z, %w
Expand All @@ -244,9 +244,9 @@ define <4 x i1> @or_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32> %w)
define <4 x i1> @and_not_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32> %w) {
; CHECK-LABEL: and_not_vec:
; CHECK: # %bb.0:
; CHECK-NEXT: vpcmpgtd %xmm3, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vpandn %xmm2, %xmm0, %xmm0
; CHECK-NEXT: vpcmpgtd %xmm3, %xmm2, %xmm1
; CHECK-NEXT: vpandn %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
%a = icmp eq <4 x i32> %x, %y
%b = icmp sgt <4 x i32> %z, %w
Expand All @@ -258,9 +258,10 @@ define <4 x i1> @or_not_vec(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32>
; CHECK-LABEL: or_not_vec:
; CHECK: # %bb.0:
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpxor %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vpcmpgtd %xmm3, %xmm2, %xmm1
; CHECK-NEXT: vmovaps {{.*#+}} xmm2 = [1,1,1,1]
; CHECK-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0
; CHECK-NEXT: vpor %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
%a = icmp eq <4 x i32> %x, %y
%b = icmp sgt <4 x i32> %z, %w
Expand Down

0 comments on commit 7fce332

Please sign in to comment.