[x86] add/sub (X==0) --> sbb(neg X)

Our handling of select-of-constants is lumpy in IR (https://reviews.llvm.org/D24480),
lumpy in DAGCombiner, and lumpy in X86ISelLowering. That's why we only had the 'sbb'
codegen in 1 out of the 4 tests. This is a step towards smoothing that out.

First, show that all of these IR forms are equivalent:
http://rise4fun.com/Alive/mx

Second, show that the 'sbb' version is faster/smaller. IACA output for SandyBridge
(later Intel and AMD chips are similar based on Agner's tables):

This is the "obvious" x86 codegen (what gcc appears to produce currently):

| Num Of |              Ports pressure in cycles               |    |
|  Uops  |  0  - DV  |  1  |  2  -  D  |  3  -  D  |  4  |  5  |    |
---------------------------------------------------------------------
|   1*   |           |     |           |           |     |     |    | xor eax, eax
|   1    | 1.0       |     |           |           |     |     | CP | test edi, edi
|   1    |           |     |           |           |     | 1.0 | CP | setnz al
|   1    |           | 1.0 |           |           |     |     | CP | neg eax

This is the adc version:
|   1*   |           |     |           |           |     |     |    | xor eax, eax
|   1    | 1.0       |     |           |           |     |     | CP | cmp edi, 0x1
|   2    |           | 1.0 |           |           |     | 1.0 | CP | adc eax, 0xffffffff

And this is sbb:
|   1    | 1.0       |     |           |           |     |     |    | neg edi
|   2    |           | 1.0 |           |           |     | 1.0 | CP | sbb eax, eax

If IACA is trustworthy, then sbb became a single uop in Broadwell, so this will be
clearly better than the alternatives going forward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@306040 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 679fa81f7f8890d82a897c0695c8779527fb63ec..2b51137901fbed28db1389ed965c8328c1d0e33b 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -34897,13 +34897,29 @@ static SDValue combineAddOrSubToADCOrSBB(SDNode *N, SelectionDAG &DAG) {
       !Cmp.getOperand(0).getValueType().isInteger())
     return SDValue();
 
-  // (cmp Z, 1) sets the carry flag if Z is 0.
   SDValue Z = Cmp.getOperand(0);
+  SDVTList VTs = DAG.getVTList(N->getValueType(0), MVT::i32);
+
+  // If X is -1 or 0, then we have an opportunity to avoid constants required by
+  // the cmp transform below. 'neg' sets the carry flag when Z != 0, so create 0
+  // or -1 using 'sbb' with fake operands:
+  //  0 - (Z != 0) --> sbb %eax, %eax, (neg Z)
+  // -1 + (Z == 0) --> sbb %eax, %eax, (neg Z)
+  if (auto *ConstantX = dyn_cast<ConstantSDNode>(X)) {
+    if ((IsSub && CC == X86::COND_NE && ConstantX->isNullValue()) ||
+        (!IsSub && CC == X86::COND_E && ConstantX->isAllOnesValue())) {
+      SDValue Zero = DAG.getConstant(0, DL, VT);
+      SDValue Neg = DAG.getNode(X86ISD::SUB, DL, VTs, Zero, Z);
+      return DAG.getNode(X86ISD::SETCC_CARRY, DL, VT,
+                         DAG.getConstant(X86::COND_B, DL, MVT::i8),
+                         SDValue(Neg.getNode(), 1));
+    }
+  }
+
+  // (cmp Z, 1) sets the carry flag if Z is 0.
   SDValue NewCmp = DAG.getNode(X86ISD::CMP, DL, MVT::i32, Z,
                                DAG.getConstant(1, DL, Z.getValueType()));
 
-  SDVTList VTs = DAG.getVTList(N->getValueType(0), MVT::i32);
-
   // X - (Z != 0) --> sub X, (zext(setne Z, 0)) --> adc X, -1, (cmp Z, 1)
   // X + (Z != 0) --> add X, (zext(setne Z, 0)) --> sbb X, -1, (cmp Z, 1)
   if (CC == X86::COND_NE)

diff --git a/test/CodeGen/X86/sbb.ll b/test/CodeGen/X86/sbb.ll
index 6ff207ce35bb94f35a85da95c28893a302d7a039..062c5d26247daf2933fd34f984100baa3a12db11 100644 (file)
--- a/test/CodeGen/X86/sbb.ll
+++ b/test/CodeGen/X86/sbb.ll
@@ -8,9 +8,8 @@
 define i8 @i8_select_0_or_neg1(i8 %x) {
 ; CHECK-LABEL: i8_select_0_or_neg1:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    cmpb $1, %dil
-; CHECK-NEXT:    movb $-1, %al
-; CHECK-NEXT:    adcb $0, %al
+; CHECK-NEXT:    negb %dil
+; CHECK-NEXT:    sbbb %al, %al
 ; CHECK-NEXT:    retq
   %cmp = icmp eq i8 %x, 0
   %sel = select i1 %cmp, i8 0, i8 -1
@@ -22,9 +21,8 @@ define i8 @i8_select_0_or_neg1(i8 %x) {
 define i16 @i16_select_0_or_neg1_as_math(i16 %x) {
 ; CHECK-LABEL: i16_select_0_or_neg1_as_math:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    cmpw $1, %di
-; CHECK-NEXT:    movw $-1, %ax
-; CHECK-NEXT:    adcw $0, %ax
+; CHECK-NEXT:    negw %di
+; CHECK-NEXT:    sbbw %ax, %ax
 ; CHECK-NEXT:    retq
   %cmp = icmp eq i16 %x, 0
   %ext = zext i1 %cmp to i16
@@ -50,9 +48,8 @@ define i32 @i32_select_0_or_neg1_commuted(i32 %x) {
 define i64 @i64_select_0_or_neg1_commuted_as_math(i64 %x) {
 ; CHECK-LABEL: i64_select_0_or_neg1_commuted_as_math:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    cmpq $1, %rdi
-; CHECK-NEXT:    adcq $-1, %rax
+; CHECK-NEXT:    negq %rdi
+; CHECK-NEXT:    sbbq %rax, %rax
 ; CHECK-NEXT:    retq
   %cmp = icmp ne i64 %x, 0
   %ext = zext i1 %cmp to i64