[llvm] Revert "[DAG] Enhance SDPatternMatch to match integer minimum and maximum patterns in addition to the existing ISD nodes." (PR #112200)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 14 06:41:18 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-aarch64
Author: None (c8ef)
<details>
<summary>Changes</summary>
Reverts llvm/llvm-project#<!-- -->111774
---
Patch is 110.63 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/112200.diff
11 Files Affected:
- (modified) llvm/include/llvm/CodeGen/SDPatternMatch.h (-99)
- (modified) llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (+8-8)
- (modified) llvm/test/CodeGen/AArch64/abds.ll (+24-19)
- (modified) llvm/test/CodeGen/AArch64/abdu.ll (+24-20)
- (modified) llvm/test/CodeGen/AArch64/midpoint-int.ll (+92-74)
- (modified) llvm/test/CodeGen/RISCV/abds.ll (+136-123)
- (modified) llvm/test/CodeGen/RISCV/abdu.ll (+126-129)
- (modified) llvm/test/CodeGen/X86/abds.ll (+93-71)
- (modified) llvm/test/CodeGen/X86/abdu.ll (+104-68)
- (modified) llvm/test/CodeGen/X86/midpoint-int.ll (+491-447)
- (modified) llvm/unittests/CodeGen/SelectionDAGPatternMatchTest.cpp (-29)
``````````diff
diff --git a/llvm/include/llvm/CodeGen/SDPatternMatch.h b/llvm/include/llvm/CodeGen/SDPatternMatch.h
index 0af4f73b869c3c..04135ee7e1c022 100644
--- a/llvm/include/llvm/CodeGen/SDPatternMatch.h
+++ b/llvm/include/llvm/CodeGen/SDPatternMatch.h
@@ -542,81 +542,6 @@ struct BinaryOpc_match {
}
};
-template <typename LHS_P, typename RHS_P, typename Pred_t,
- bool Commutable = false, bool ExcludeChain = false>
-struct MaxMin_match {
- using PredType = Pred_t;
- LHS_P LHS;
- RHS_P RHS;
-
- MaxMin_match(const LHS_P &L, const RHS_P &R) : LHS(L), RHS(R) {}
-
- template <typename MatchContext>
- bool match(const MatchContext &Ctx, SDValue N) {
- if (sd_context_match(N, Ctx, m_Opc(ISD::SELECT)) ||
- sd_context_match(N, Ctx, m_Opc(ISD::VSELECT))) {
- EffectiveOperands<ExcludeChain> EO_SELECT(N, Ctx);
- assert(EO_SELECT.Size == 3);
- SDValue Cond = N->getOperand(EO_SELECT.FirstIndex);
- SDValue TrueValue = N->getOperand(EO_SELECT.FirstIndex + 1);
- SDValue FalseValue = N->getOperand(EO_SELECT.FirstIndex + 2);
-
- if (sd_context_match(Cond, Ctx, m_Opc(ISD::SETCC))) {
- EffectiveOperands<ExcludeChain> EO_SETCC(Cond, Ctx);
- assert(EO_SETCC.Size == 3);
- SDValue L = Cond->getOperand(EO_SETCC.FirstIndex);
- SDValue R = Cond->getOperand(EO_SETCC.FirstIndex + 1);
- auto *CondNode =
- cast<CondCodeSDNode>(Cond->getOperand(EO_SETCC.FirstIndex + 2));
-
- if ((TrueValue != L || FalseValue != R) &&
- (TrueValue != R || FalseValue != L)) {
- return false;
- }
-
- ISD::CondCode Cond =
- TrueValue == L ? CondNode->get()
- : getSetCCInverse(CondNode->get(), L.getValueType());
- if (!Pred_t::match(Cond)) {
- return false;
- }
- return (LHS.match(Ctx, L) && RHS.match(Ctx, R)) ||
- (Commutable && LHS.match(Ctx, R) && RHS.match(Ctx, L));
- }
- }
-
- return false;
- }
-};
-
-// Helper class for identifying signed max predicates.
-struct smax_pred_ty {
- static bool match(ISD::CondCode Cond) {
- return Cond == ISD::CondCode::SETGT || Cond == ISD::CondCode::SETGE;
- }
-};
-
-// Helper class for identifying unsigned max predicates.
-struct umax_pred_ty {
- static bool match(ISD::CondCode Cond) {
- return Cond == ISD::CondCode::SETUGT || Cond == ISD::CondCode::SETUGE;
- }
-};
-
-// Helper class for identifying signed min predicates.
-struct smin_pred_ty {
- static bool match(ISD::CondCode Cond) {
- return Cond == ISD::CondCode::SETLT || Cond == ISD::CondCode::SETLE;
- }
-};
-
-// Helper class for identifying unsigned min predicates.
-struct umin_pred_ty {
- static bool match(ISD::CondCode Cond) {
- return Cond == ISD::CondCode::SETULT || Cond == ISD::CondCode::SETULE;
- }
-};
-
template <typename LHS, typename RHS>
inline BinaryOpc_match<LHS, RHS> m_BinOp(unsigned Opc, const LHS &L,
const RHS &R) {
@@ -688,45 +613,21 @@ inline BinaryOpc_match<LHS, RHS, true> m_SMin(const LHS &L, const RHS &R) {
return BinaryOpc_match<LHS, RHS, true>(ISD::SMIN, L, R);
}
-template <typename LHS, typename RHS>
-inline auto m_SMinLike(const LHS &L, const RHS &R) {
- return m_AnyOf(BinaryOpc_match<LHS, RHS, true>(ISD::SMIN, L, R),
- MaxMin_match<LHS, RHS, smin_pred_ty, true>(L, R));
-}
-
template <typename LHS, typename RHS>
inline BinaryOpc_match<LHS, RHS, true> m_SMax(const LHS &L, const RHS &R) {
return BinaryOpc_match<LHS, RHS, true>(ISD::SMAX, L, R);
}
-template <typename LHS, typename RHS>
-inline auto m_SMaxLike(const LHS &L, const RHS &R) {
- return m_AnyOf(BinaryOpc_match<LHS, RHS, true>(ISD::SMAX, L, R),
- MaxMin_match<LHS, RHS, smax_pred_ty, true>(L, R));
-}
-
template <typename LHS, typename RHS>
inline BinaryOpc_match<LHS, RHS, true> m_UMin(const LHS &L, const RHS &R) {
return BinaryOpc_match<LHS, RHS, true>(ISD::UMIN, L, R);
}
-template <typename LHS, typename RHS>
-inline auto m_UMinLike(const LHS &L, const RHS &R) {
- return m_AnyOf(BinaryOpc_match<LHS, RHS, true>(ISD::UMIN, L, R),
- MaxMin_match<LHS, RHS, umin_pred_ty, true>(L, R));
-}
-
template <typename LHS, typename RHS>
inline BinaryOpc_match<LHS, RHS, true> m_UMax(const LHS &L, const RHS &R) {
return BinaryOpc_match<LHS, RHS, true>(ISD::UMAX, L, R);
}
-template <typename LHS, typename RHS>
-inline auto m_UMaxLike(const LHS &L, const RHS &R) {
- return m_AnyOf(BinaryOpc_match<LHS, RHS, true>(ISD::UMAX, L, R),
- MaxMin_match<LHS, RHS, umax_pred_ty, true>(L, R));
-}
-
template <typename LHS, typename RHS>
inline BinaryOpc_match<LHS, RHS> m_UDiv(const LHS &L, const RHS &R) {
return BinaryOpc_match<LHS, RHS>(ISD::UDIV, L, R);
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 924558a298fd85..810ca458bc8787 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -4190,26 +4190,26 @@ SDValue DAGCombiner::visitSUB(SDNode *N) {
// smax(a,b) - smin(a,b) --> abds(a,b)
if ((!LegalOperations || hasOperation(ISD::ABDS, VT)) &&
- sd_match(N0, m_SMaxLike(m_Value(A), m_Value(B))) &&
- sd_match(N1, m_SMinLike(m_Specific(A), m_Specific(B))))
+ sd_match(N0, m_SMax(m_Value(A), m_Value(B))) &&
+ sd_match(N1, m_SMin(m_Specific(A), m_Specific(B))))
return DAG.getNode(ISD::ABDS, DL, VT, A, B);
// smin(a,b) - smax(a,b) --> neg(abds(a,b))
if (hasOperation(ISD::ABDS, VT) &&
- sd_match(N0, m_SMinLike(m_Value(A), m_Value(B))) &&
- sd_match(N1, m_SMaxLike(m_Specific(A), m_Specific(B))))
+ sd_match(N0, m_SMin(m_Value(A), m_Value(B))) &&
+ sd_match(N1, m_SMax(m_Specific(A), m_Specific(B))))
return DAG.getNegative(DAG.getNode(ISD::ABDS, DL, VT, A, B), DL, VT);
// umax(a,b) - umin(a,b) --> abdu(a,b)
if ((!LegalOperations || hasOperation(ISD::ABDU, VT)) &&
- sd_match(N0, m_UMaxLike(m_Value(A), m_Value(B))) &&
- sd_match(N1, m_UMinLike(m_Specific(A), m_Specific(B))))
+ sd_match(N0, m_UMax(m_Value(A), m_Value(B))) &&
+ sd_match(N1, m_UMin(m_Specific(A), m_Specific(B))))
return DAG.getNode(ISD::ABDU, DL, VT, A, B);
// umin(a,b) - umax(a,b) --> neg(abdu(a,b))
if (hasOperation(ISD::ABDU, VT) &&
- sd_match(N0, m_UMinLike(m_Value(A), m_Value(B))) &&
- sd_match(N1, m_UMaxLike(m_Specific(A), m_Specific(B))))
+ sd_match(N0, m_UMin(m_Value(A), m_Value(B))) &&
+ sd_match(N1, m_UMax(m_Specific(A), m_Specific(B))))
return DAG.getNegative(DAG.getNode(ISD::ABDU, DL, VT, A, B), DL, VT);
return SDValue();
diff --git a/llvm/test/CodeGen/AArch64/abds.ll b/llvm/test/CodeGen/AArch64/abds.ll
index 62db30f17747cf..e5cc04f9be1a1f 100644
--- a/llvm/test/CodeGen/AArch64/abds.ll
+++ b/llvm/test/CodeGen/AArch64/abds.ll
@@ -547,9 +547,10 @@ define i8 @abd_select_i8(i8 %a, i8 %b) nounwind {
; CHECK-LABEL: abd_select_i8:
; CHECK: // %bb.0:
; CHECK-NEXT: sxtb w8, w0
-; CHECK-NEXT: sub w8, w8, w1, sxtb
-; CHECK-NEXT: cmp w8, #0
-; CHECK-NEXT: cneg w0, w8, mi
+; CHECK-NEXT: cmp w8, w1, sxtb
+; CHECK-NEXT: csel w8, w0, w1, lt
+; CHECK-NEXT: csel w9, w1, w0, lt
+; CHECK-NEXT: sub w0, w9, w8
; CHECK-NEXT: ret
%cmp = icmp slt i8 %a, %b
%ab = select i1 %cmp, i8 %a, i8 %b
@@ -562,9 +563,10 @@ define i16 @abd_select_i16(i16 %a, i16 %b) nounwind {
; CHECK-LABEL: abd_select_i16:
; CHECK: // %bb.0:
; CHECK-NEXT: sxth w8, w0
-; CHECK-NEXT: sub w8, w8, w1, sxth
-; CHECK-NEXT: cmp w8, #0
-; CHECK-NEXT: cneg w0, w8, mi
+; CHECK-NEXT: cmp w8, w1, sxth
+; CHECK-NEXT: csel w8, w0, w1, le
+; CHECK-NEXT: csel w9, w1, w0, le
+; CHECK-NEXT: sub w0, w9, w8
; CHECK-NEXT: ret
%cmp = icmp sle i16 %a, %b
%ab = select i1 %cmp, i16 %a, i16 %b
@@ -576,9 +578,10 @@ define i16 @abd_select_i16(i16 %a, i16 %b) nounwind {
define i32 @abd_select_i32(i32 %a, i32 %b) nounwind {
; CHECK-LABEL: abd_select_i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub w8, w1, w0
-; CHECK-NEXT: subs w9, w0, w1
-; CHECK-NEXT: csel w0, w9, w8, gt
+; CHECK-NEXT: cmp w0, w1
+; CHECK-NEXT: csel w8, w0, w1, gt
+; CHECK-NEXT: csel w9, w1, w0, gt
+; CHECK-NEXT: sub w0, w8, w9
; CHECK-NEXT: ret
%cmp = icmp sgt i32 %a, %b
%ab = select i1 %cmp, i32 %a, i32 %b
@@ -590,9 +593,10 @@ define i32 @abd_select_i32(i32 %a, i32 %b) nounwind {
define i64 @abd_select_i64(i64 %a, i64 %b) nounwind {
; CHECK-LABEL: abd_select_i64:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub x8, x1, x0
-; CHECK-NEXT: subs x9, x0, x1
-; CHECK-NEXT: csel x0, x9, x8, gt
+; CHECK-NEXT: cmp x0, x1
+; CHECK-NEXT: csel x8, x0, x1, ge
+; CHECK-NEXT: csel x9, x1, x0, ge
+; CHECK-NEXT: sub x0, x8, x9
; CHECK-NEXT: ret
%cmp = icmp sge i64 %a, %b
%ab = select i1 %cmp, i64 %a, i64 %b
@@ -604,13 +608,14 @@ define i64 @abd_select_i64(i64 %a, i64 %b) nounwind {
define i128 @abd_select_i128(i128 %a, i128 %b) nounwind {
; CHECK-LABEL: abd_select_i128:
; CHECK: // %bb.0:
-; CHECK-NEXT: subs x8, x0, x2
-; CHECK-NEXT: sbc x9, x1, x3
-; CHECK-NEXT: subs x10, x2, x0
-; CHECK-NEXT: sbc x11, x3, x1
-; CHECK-NEXT: sbcs xzr, x3, x1
-; CHECK-NEXT: csel x0, x8, x10, lt
-; CHECK-NEXT: csel x1, x9, x11, lt
+; CHECK-NEXT: cmp x0, x2
+; CHECK-NEXT: sbcs xzr, x1, x3
+; CHECK-NEXT: csel x8, x0, x2, lt
+; CHECK-NEXT: csel x9, x2, x0, lt
+; CHECK-NEXT: csel x10, x1, x3, lt
+; CHECK-NEXT: csel x11, x3, x1, lt
+; CHECK-NEXT: subs x0, x9, x8
+; CHECK-NEXT: sbc x1, x11, x10
; CHECK-NEXT: ret
%cmp = icmp slt i128 %a, %b
%ab = select i1 %cmp, i128 %a, i128 %b
diff --git a/llvm/test/CodeGen/AArch64/abdu.ll b/llvm/test/CodeGen/AArch64/abdu.ll
index 4585de96c848f2..0a44ae16884582 100644
--- a/llvm/test/CodeGen/AArch64/abdu.ll
+++ b/llvm/test/CodeGen/AArch64/abdu.ll
@@ -408,9 +408,10 @@ define i8 @abd_select_i8(i8 %a, i8 %b) nounwind {
; CHECK-LABEL: abd_select_i8:
; CHECK: // %bb.0:
; CHECK-NEXT: and w8, w0, #0xff
-; CHECK-NEXT: sub w8, w8, w1, uxtb
-; CHECK-NEXT: cmp w8, #0
-; CHECK-NEXT: cneg w0, w8, mi
+; CHECK-NEXT: cmp w8, w1, uxtb
+; CHECK-NEXT: csel w8, w0, w1, lo
+; CHECK-NEXT: csel w9, w1, w0, lo
+; CHECK-NEXT: sub w0, w9, w8
; CHECK-NEXT: ret
%cmp = icmp ult i8 %a, %b
%ab = select i1 %cmp, i8 %a, i8 %b
@@ -423,9 +424,10 @@ define i16 @abd_select_i16(i16 %a, i16 %b) nounwind {
; CHECK-LABEL: abd_select_i16:
; CHECK: // %bb.0:
; CHECK-NEXT: and w8, w0, #0xffff
-; CHECK-NEXT: sub w8, w8, w1, uxth
-; CHECK-NEXT: cmp w8, #0
-; CHECK-NEXT: cneg w0, w8, mi
+; CHECK-NEXT: cmp w8, w1, uxth
+; CHECK-NEXT: csel w8, w0, w1, ls
+; CHECK-NEXT: csel w9, w1, w0, ls
+; CHECK-NEXT: sub w0, w9, w8
; CHECK-NEXT: ret
%cmp = icmp ule i16 %a, %b
%ab = select i1 %cmp, i16 %a, i16 %b
@@ -437,9 +439,10 @@ define i16 @abd_select_i16(i16 %a, i16 %b) nounwind {
define i32 @abd_select_i32(i32 %a, i32 %b) nounwind {
; CHECK-LABEL: abd_select_i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub w8, w1, w0
-; CHECK-NEXT: subs w9, w0, w1
-; CHECK-NEXT: csel w0, w9, w8, hi
+; CHECK-NEXT: cmp w0, w1
+; CHECK-NEXT: csel w8, w0, w1, hi
+; CHECK-NEXT: csel w9, w1, w0, hi
+; CHECK-NEXT: sub w0, w8, w9
; CHECK-NEXT: ret
%cmp = icmp ugt i32 %a, %b
%ab = select i1 %cmp, i32 %a, i32 %b
@@ -451,9 +454,10 @@ define i32 @abd_select_i32(i32 %a, i32 %b) nounwind {
define i64 @abd_select_i64(i64 %a, i64 %b) nounwind {
; CHECK-LABEL: abd_select_i64:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub x8, x1, x0
-; CHECK-NEXT: subs x9, x0, x1
-; CHECK-NEXT: csel x0, x9, x8, hi
+; CHECK-NEXT: cmp x0, x1
+; CHECK-NEXT: csel x8, x0, x1, hs
+; CHECK-NEXT: csel x9, x1, x0, hs
+; CHECK-NEXT: sub x0, x8, x9
; CHECK-NEXT: ret
%cmp = icmp uge i64 %a, %b
%ab = select i1 %cmp, i64 %a, i64 %b
@@ -465,14 +469,14 @@ define i64 @abd_select_i64(i64 %a, i64 %b) nounwind {
define i128 @abd_select_i128(i128 %a, i128 %b) nounwind {
; CHECK-LABEL: abd_select_i128:
; CHECK: // %bb.0:
-; CHECK-NEXT: subs x8, x0, x2
-; CHECK-NEXT: sbcs x9, x1, x3
-; CHECK-NEXT: cset w10, lo
-; CHECK-NEXT: sbfx x10, x10, #0, #1
-; CHECK-NEXT: eor x8, x8, x10
-; CHECK-NEXT: eor x9, x9, x10
-; CHECK-NEXT: subs x0, x8, x10
-; CHECK-NEXT: sbc x1, x9, x10
+; CHECK-NEXT: cmp x0, x2
+; CHECK-NEXT: sbcs xzr, x1, x3
+; CHECK-NEXT: csel x8, x0, x2, lo
+; CHECK-NEXT: csel x9, x2, x0, lo
+; CHECK-NEXT: csel x10, x1, x3, lo
+; CHECK-NEXT: csel x11, x3, x1, lo
+; CHECK-NEXT: subs x0, x9, x8
+; CHECK-NEXT: sbc x1, x11, x10
; CHECK-NEXT: ret
%cmp = icmp ult i128 %a, %b
%ab = select i1 %cmp, i128 %a, i128 %b
diff --git a/llvm/test/CodeGen/AArch64/midpoint-int.ll b/llvm/test/CodeGen/AArch64/midpoint-int.ll
index bbdce7c6e933b3..1043fa5c4565ee 100644
--- a/llvm/test/CodeGen/AArch64/midpoint-int.ll
+++ b/llvm/test/CodeGen/AArch64/midpoint-int.ll
@@ -13,11 +13,12 @@
define i32 @scalar_i32_signed_reg_reg(i32 %a1, i32 %a2) nounwind {
; CHECK-LABEL: scalar_i32_signed_reg_reg:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub w9, w1, w0
-; CHECK-NEXT: subs w10, w0, w1
+; CHECK-NEXT: cmp w0, w1
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
-; CHECK-NEXT: csel w9, w10, w9, gt
+; CHECK-NEXT: csel w9, w1, w0, gt
+; CHECK-NEXT: csel w10, w0, w1, gt
; CHECK-NEXT: cneg w8, w8, le
+; CHECK-NEXT: sub w9, w10, w9
; CHECK-NEXT: lsr w9, w9, #1
; CHECK-NEXT: madd w0, w9, w8, w0
; CHECK-NEXT: ret
@@ -35,11 +36,12 @@ define i32 @scalar_i32_signed_reg_reg(i32 %a1, i32 %a2) nounwind {
define i32 @scalar_i32_unsigned_reg_reg(i32 %a1, i32 %a2) nounwind {
; CHECK-LABEL: scalar_i32_unsigned_reg_reg:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub w9, w1, w0
-; CHECK-NEXT: subs w10, w0, w1
+; CHECK-NEXT: cmp w0, w1
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
-; CHECK-NEXT: csel w9, w10, w9, hi
+; CHECK-NEXT: csel w9, w1, w0, hi
+; CHECK-NEXT: csel w10, w0, w1, hi
; CHECK-NEXT: cneg w8, w8, ls
+; CHECK-NEXT: sub w9, w10, w9
; CHECK-NEXT: lsr w9, w9, #1
; CHECK-NEXT: madd w0, w9, w8, w0
; CHECK-NEXT: ret
@@ -62,10 +64,10 @@ define i32 @scalar_i32_signed_mem_reg(ptr %a1_addr, i32 %a2) nounwind {
; CHECK-NEXT: ldr w9, [x0]
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cmp w9, w1
-; CHECK-NEXT: sub w10, w1, w9
+; CHECK-NEXT: csel w10, w1, w9, gt
+; CHECK-NEXT: csel w11, w9, w1, gt
; CHECK-NEXT: cneg w8, w8, le
-; CHECK-NEXT: subs w11, w9, w1
-; CHECK-NEXT: csel w10, w11, w10, gt
+; CHECK-NEXT: sub w10, w11, w10
; CHECK-NEXT: lsr w10, w10, #1
; CHECK-NEXT: madd w0, w10, w8, w9
; CHECK-NEXT: ret
@@ -87,10 +89,10 @@ define i32 @scalar_i32_signed_reg_mem(i32 %a1, ptr %a2_addr) nounwind {
; CHECK-NEXT: ldr w9, [x1]
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cmp w0, w9
-; CHECK-NEXT: sub w10, w9, w0
+; CHECK-NEXT: csel w10, w9, w0, gt
+; CHECK-NEXT: csel w9, w0, w9, gt
; CHECK-NEXT: cneg w8, w8, le
-; CHECK-NEXT: subs w9, w0, w9
-; CHECK-NEXT: csel w9, w9, w10, gt
+; CHECK-NEXT: sub w9, w9, w10
; CHECK-NEXT: lsr w9, w9, #1
; CHECK-NEXT: madd w0, w9, w8, w0
; CHECK-NEXT: ret
@@ -113,10 +115,10 @@ define i32 @scalar_i32_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
; CHECK-NEXT: ldr w10, [x1]
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
; CHECK-NEXT: cmp w9, w10
-; CHECK-NEXT: sub w11, w10, w9
+; CHECK-NEXT: csel w11, w10, w9, gt
+; CHECK-NEXT: csel w10, w9, w10, gt
; CHECK-NEXT: cneg w8, w8, le
-; CHECK-NEXT: subs w10, w9, w10
-; CHECK-NEXT: csel w10, w10, w11, gt
+; CHECK-NEXT: sub w10, w10, w11
; CHECK-NEXT: lsr w10, w10, #1
; CHECK-NEXT: madd w0, w10, w8, w9
; CHECK-NEXT: ret
@@ -142,11 +144,12 @@ define i32 @scalar_i32_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
define i64 @scalar_i64_signed_reg_reg(i64 %a1, i64 %a2) nounwind {
; CHECK-LABEL: scalar_i64_signed_reg_reg:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub x9, x1, x0
-; CHECK-NEXT: subs x10, x0, x1
+; CHECK-NEXT: cmp x0, x1
; CHECK-NEXT: mov x8, #-1 // =0xffffffffffffffff
-; CHECK-NEXT: csel x9, x10, x9, gt
+; CHECK-NEXT: csel x9, x1, x0, gt
+; CHECK-NEXT: csel x10, x0, x1, gt
; CHECK-NEXT: cneg x8, x8, le
+; CHECK-NEXT: sub x9, x10, x9
; CHECK-NEXT: lsr x9, x9, #1
; CHECK-NEXT: madd x0, x9, x8, x0
; CHECK-NEXT: ret
@@ -164,11 +167,12 @@ define i64 @scalar_i64_signed_reg_reg(i64 %a1, i64 %a2) nounwind {
define i64 @scalar_i64_unsigned_reg_reg(i64 %a1, i64 %a2) nounwind {
; CHECK-LABEL: scalar_i64_unsigned_reg_reg:
; CHECK: // %bb.0:
-; CHECK-NEXT: sub x9, x1, x0
-; CHECK-NEXT: subs x10, x0, x1
+; CHECK-NEXT: cmp x0, x1
; CHECK-NEXT: mov x8, #-1 // =0xffffffffffffffff
-; CHECK-NEXT: csel x9, x10, x9, hi
+; CHECK-NEXT: csel x9, x1, x0, hi
+; CHECK-NEXT: csel x10, x0, x1, hi
; CHECK-NEXT: cneg x8, x8, ls
+; CHECK-NEXT: sub x9, x10, x9
; CHECK-NEXT: lsr x9, x9, #1
; CHECK-NEXT: madd x0, x9, x8, x0
; CHECK-NEXT: ret
@@ -191,10 +195,10 @@ define i64 @scalar_i64_signed_mem_reg(ptr %a1_addr, i64 %a2) nounwind {
; CHECK-NEXT: ldr x9, [x0]
; CHECK-NEXT: mov x8, #-1 // =0xffffffffffffffff
; CHECK-NEXT: cmp x9, x1
-; CHECK-NEXT: sub x10, x1, x9
+; CHECK-NEXT: csel x10, x1, x9, gt
+; CHECK-NEXT: csel x11, x9, x1, gt
; CHECK-NEXT: cneg x8, x8, le
-; CHECK-NEXT: subs x11, x9, x1
-; CHECK-NEXT: csel x10, x11, x10, gt
+; CHECK-NEXT: sub x10, x11, x10
; CHECK-NEXT: lsr x10, x10, #1
; CHECK-NEXT: madd x0, x10, x8, x9
; CHECK-NEXT: ret
@@ -216,10 +220,10 @@ define i64 @scalar_i64_signed_reg_mem(i64 %a1, ptr %a2_addr) nounwind {
; CHECK-NEXT: ldr x9, [x1]
; CHECK-NEXT: mov x8, #-1 // =0xffffffffffffffff
; CHECK-NEXT: cmp x0, x9
-; CHECK-NEXT: sub x10, x9, x0
+; CHECK-NEXT: csel x10, x9, x0, gt
+; CHECK-NEXT: csel x9, x0, x9, gt
; CHECK-NEXT: cneg x8, x8, le
-; CHECK-NEXT: subs x9, x0, x9
-; CHECK-NEXT: csel x9, x9, x10, gt
+; CHECK-NEXT: sub x9, x9, x10
; CHECK-NEXT: lsr x9, x9, #1
; CHECK-NEXT: madd x0, x9, x8, x0
; CHECK-NEXT: ret
@@ -242,10 +246,10 @@ define i64 @scalar_i64_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
; CHECK-NEXT: ldr x10, [x1]
; CHECK-NEXT: mov x8, #-1 // =0xffffffffffffffff
; CHECK-NEXT: cmp x9, x10
-; CHECK-NEXT: sub x11, x10, x9
+; CHECK-NEXT: csel x11, x10, x9, gt
+; CHECK-NEXT: csel x10, x9, x10, gt
; CHECK-NEXT: cneg x8, x8, le
-; CHECK-NEXT: subs x10, x9, x10
-; CHECK-NEXT: csel x10, x10, x11, gt
+; CHECK-NEXT: sub x10, x10, x11
; CHECK-NEXT: lsr x10, x10, #1
; CHECK-NEXT: madd x0, x10, x8, x9
; CHECK-NEXT: ret
@@ -271,13 +275,14 @@ define i64 @scalar_i64_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
define i16 @scalar_i16_signed_reg_reg(i16 %a1, i16 %a2) nounwind {
; CHECK-LABEL: scalar_i16_signed_reg_reg:
; CHECK: // %bb.0:
-; CHECK-NEXT: sxth w9, w1
-; CHECK-NEXT: sxth w10, w0
+; CHECK-NEXT: sxth w9, w0
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
-; CHECK-NEXT: subs w9, w10, w9
-; CHECK-NEXT: cneg w9, w9, mi
+; CHECK-NEXT: cmp w9, w1, sxth
+; CHECK-NEXT: csel w9, w1, w0, gt
+; CHECK-NEXT: csel w10, w0, w1, gt
; CHECK-NEXT: cneg w8, w8, le
-; CHECK-NEXT: lsr w9, w9, #1
+; CHECK-NEXT: sub w9, w10, w9
+; CHECK-NEXT: ubfx w9, w9, #1, #15
; CHECK-NEXT: madd w0, w9, w8, w0
; CHECK-NEXT: ret
%t3 = icmp sgt i16 %a1, %a2 ; signed
@@ -294,13 +299,14 @@ define i16 @scalar_i16_signed_reg_reg(i16 %a1, i16 %a2) nounwind {
define i16 @scalar_i16_unsigned_reg_reg(i16 %a1, i16 %a2) nounwind {
; CHECK-LABEL: scalar_i16_unsigned_reg_reg:
; CHECK: // %bb.0:
-; CHECK-NEXT: and w9, w1, #0xffff
-; CHECK-NEXT: and w10, w0, #0xffff
+; CHECK-NEXT: and w9, w0, #0xffff
; CHECK-NEXT: mov w8, #-1 // =0xffffffff
-; CHECK-NEXT: subs w9, w10, w9
-; CHECK-NEXT: cneg w9, w9, mi
+; CHECK-NEXT: cmp w9, w1, uxth
+; CHECK-NEXT: csel w9, w1, w0, hi
+; CHECK-NEXT: csel w10, w0, w1, hi
; CHECK-NEXT: cneg w8, w8, ls
-; CHECK-NEXT: lsr w9, w...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/112200
More information about the llvm-commits
mailing list