[DAGCombine] (A-C1)-C2 -> A-(C1+C2) constant-fold

Summary: https://rise4fun.com/Alive/Mb1M

Reviewers: RKSimon, craig.topper, spatel, t.p.northover

Reviewed By: t.p.northover

Subscribers: t.p.northover, javed.absar, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62689

llvm-svn: 362134

3 files changed, 15 insertions, 14 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index da205841a22..8bbd5cd2012 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -2885,6 +2885,16 @@ SDValue DAGCombiner::visitSUB(SDNode *N) {
     }
   }
 
+  // fold (A-C1)-C2 -> A-(C1+C2)
+  if (N0.getOpcode() == ISD::SUB &&
+      isConstantOrConstantVector(N1, /* NoOpaques */ true) &&
+      isConstantOrConstantVector(N0.getOperand(1), /* NoOpaques */ true)) {
+    SDValue NewC = DAG.FoldConstantArithmetic(
+        ISD::ADD, DL, VT, N0.getOperand(1).getNode(), N1.getNode());
+    assert(NewC && "Constant folding failed");
+    return DAG.getNode(ISD::SUB, DL, VT, N0.getOperand(0), NewC);
+  }
+
   // fold ((A+(B+or-C))-B) -> A+or-C
   if (N0.getOpcode() == ISD::ADD &&
       (N0.getOperand(1).getOpcode() == ISD::SUB ||
diff --git a/llvm/test/CodeGen/AArch64/addsub-constant-folding.ll b/llvm/test/CodeGen/AArch64/addsub-constant-folding.ll
index 8afa4ded1e0..de87aa4348a 100644
--- a/llvm/test/CodeGen/AArch64/addsub-constant-folding.ll
+++ b/llvm/test/CodeGen/AArch64/addsub-constant-folding.ll
@@ -200,9 +200,7 @@ define <4 x i32> @sub_const_add_const_nonsplat(<4 x i32> %arg) {
 define <4 x i32> @sub_const_sub_const(<4 x i32> %arg) {
 ; CHECK-LABEL: sub_const_sub_const:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    movi v1.4s, #8
-; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    movi v1.4s, #2
+; CHECK-NEXT:    movi v1.4s, #10
 ; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    ret
   %t0 = sub <4 x i32> %arg, <i32 8, i32 8, i32 8, i32 8>
@@ -218,12 +216,12 @@ define <4 x i32> @sub_const_sub_const_extrause(<4 x i32> %arg) {
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset w30, -16
 ; CHECK-NEXT:    movi v1.4s, #8
-; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
+; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    bl use
 ; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
 ; CHECK-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
-; CHECK-NEXT:    movi v0.4s, #2
+; CHECK-NEXT:    movi v0.4s, #10
 ; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
 ; CHECK-NEXT:    add sp, sp, #32 // =32
 ; CHECK-NEXT:    ret
@@ -238,10 +236,7 @@ define <4 x i32> @sub_const_sub_const_nonsplat(<4 x i32> %arg) {
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    adrp x8, .LCPI14_0
 ; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI14_0]
-; CHECK-NEXT:    adrp x8, .LCPI14_1
-; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI14_1]
 ; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    sub v0.4s, v0.4s, v2.4s
 ; CHECK-NEXT:    ret
   %t0 = sub <4 x i32> %arg, <i32 21, i32 undef, i32 8, i32 8>
   %t1 = sub <4 x i32> %t0, <i32 2, i32 3, i32 undef, i32 2>
diff --git a/llvm/test/CodeGen/X86/addsub-constant-folding.ll b/llvm/test/CodeGen/X86/addsub-constant-folding.ll
index 0053bc2b611..e24f35382fd 100644
--- a/llvm/test/CodeGen/X86/addsub-constant-folding.ll
+++ b/llvm/test/CodeGen/X86/addsub-constant-folding.ll
@@ -287,13 +287,11 @@ define <4 x i32> @sub_const_sub_const(<4 x i32> %arg) {
 ; X86-LABEL: sub_const_sub_const:
 ; X86:       # %bb.0:
 ; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
-; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: sub_const_sub_const:
 ; X64:       # %bb.0:
 ; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
-; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
 ; X64-NEXT:    retq
   %t0 = sub <4 x i32> %arg, <i32 8, i32 8, i32 8, i32 8>
   %t1 = sub <4 x i32> %t0, <i32 2, i32 2, i32 2, i32 2>
@@ -305,8 +303,8 @@ define <4 x i32> @sub_const_sub_const_extrause(<4 x i32> %arg) {
 ; X86:       # %bb.0:
 ; X86-NEXT:    subl $28, %esp
 ; X86-NEXT:    .cfi_def_cfa_offset 32
-; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    movdqu %xmm0, (%esp) # 16-byte Spill
+; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    calll use
 ; X86-NEXT:    movdqu (%esp), %xmm0 # 16-byte Reload
 ; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
@@ -318,8 +316,8 @@ define <4 x i32> @sub_const_sub_const_extrause(<4 x i32> %arg) {
 ; X64:       # %bb.0:
 ; X64-NEXT:    subq $24, %rsp
 ; X64-NEXT:    .cfi_def_cfa_offset 32
-; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
 ; X64-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
+; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
 ; X64-NEXT:    callq use
 ; X64-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
@@ -336,13 +334,11 @@ define <4 x i32> @sub_const_sub_const_nonsplat(<4 x i32> %arg) {
 ; X86-LABEL: sub_const_sub_const_nonsplat:
 ; X86:       # %bb.0:
 ; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
-; X86-NEXT:    psubd {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: sub_const_sub_const_nonsplat:
 ; X64:       # %bb.0:
 ; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
-; X64-NEXT:    psubd {{.*}}(%rip), %xmm0
 ; X64-NEXT:    retq
   %t0 = sub <4 x i32> %arg, <i32 21, i32 undef, i32 8, i32 8>
   %t1 = sub <4 x i32> %t0, <i32 2, i32 3, i32 undef, i32 2>