[llvm] ed51e26 - [X86] combineAddOrSubToADCOrSBB - commute + neg subtraction patterns

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2022-03-21T13:55:35Z
New Revision: ed51e26ab4f077c4ca5cb14fe94b257855b0293f

URL: https://github.com/llvm/llvm-project/commit/ed51e26ab4f077c4ca5cb14fe94b257855b0293f
DIFF: https://github.com/llvm/llvm-project/commit/ed51e26ab4f077c4ca5cb14fe94b257855b0293f.diff

LOG: [X86] combineAddOrSubToADCOrSBB - commute + neg subtraction patterns

Handle SUB(AND(SRL(Y,Z),1),X) -> NEG(SBB(X,0,BT(Y,Z))) folds

I'll address the X86 lost folded-load regressions in a follow-up patch

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp
    llvm/test/CodeGen/X86/add-sub-bool.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 8c6e9e4c2fb63..8b0d6bd958957 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -52477,9 +52477,13 @@ static SDValue combineAddOrSubToADCOrSBB(SDNode *N, SelectionDAG &DAG) {
   if (SDValue ADCOrSBB = combineAddOrSubToADCOrSBB(IsSub, DL, VT, X, Y, DAG))
     return ADCOrSBB;
 
-  // If this is an add, commute and try again.
-  if (!IsSub)
-    return combineAddOrSubToADCOrSBB(IsSub, DL, VT, Y, X, DAG);
+  // Commute and try again (negate the result for subtracts).
+  if (SDValue ADCOrSBB = combineAddOrSubToADCOrSBB(IsSub, DL, VT, Y, X, DAG)) {
+    if (IsSub)
+      ADCOrSBB =
+          DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, DL, VT), ADCOrSBB);
+    return ADCOrSBB;
+  }
 
   return SDValue();
 }

diff  --git a/llvm/test/CodeGen/X86/add-sub-bool.ll b/llvm/test/CodeGen/X86/add-sub-bool.ll
index a97f293e58968..b5480a1bd383d 100644
--- a/llvm/test/CodeGen/X86/add-sub-bool.ll
+++ b/llvm/test/CodeGen/X86/add-sub-bool.ll
@@ -215,28 +215,19 @@ define i32 @test_i32_sub_sub_idx(i32 %x, i32 %y, i32 %z) nounwind {
 ; X86-LABEL: test_i32_sub_sub_idx:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    shrl $16, %eax
-; X86-NEXT:    andl $1, %eax
-; X86-NEXT:    subl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    btl $16, {{[0-9]+}}(%esp)
+; X86-NEXT:    sbbl $0, %ecx
+; X86-NEXT:    subl %ecx, %eax
 ; X86-NEXT:    retl
 ;
-; NOTBM-LABEL: test_i32_sub_sub_idx:
-; NOTBM:       # %bb.0:
-; NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; NOTBM-NEXT:    # kill: def $edi killed $edi def $rdi
-; NOTBM-NEXT:    shrl $16, %edx
-; NOTBM-NEXT:    andl $1, %edx
-; NOTBM-NEXT:    subl %esi, %edx
-; NOTBM-NEXT:    leal (%rdx,%rdi), %eax
-; NOTBM-NEXT:    retq
-;
-; TBM-LABEL: test_i32_sub_sub_idx:
-; TBM:       # %bb.0:
-; TBM-NEXT:    bextrl $272, %edx, %eax # imm = 0x110
-; TBM-NEXT:    subl %esi, %eax
-; TBM-NEXT:    addl %edi, %eax
-; TBM-NEXT:    retq
+; X64-LABEL: test_i32_sub_sub_idx:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    btl $16, %edx
+; X64-NEXT:    sbbl $0, %esi
+; X64-NEXT:    subl %esi, %eax
+; X64-NEXT:    retq
   %shift = lshr i32 %z, 16
   %mask = and i32 %shift, 1
   %sub0 = sub i32 %y, %mask
@@ -436,23 +427,23 @@ define i32 @test_i32_sub_add_var(i32 %x, i32 %y, i32 %z, i32 %w) nounwind {
 define i32 @test_i32_sub_sub_var(i32 %x, i32 %y, i32 %z, i32 %w) nounwind {
 ; X86-LABEL: test_i32_sub_sub_var:
 ; X86:       # %bb.0:
-; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    shrl %cl, %eax
-; X86-NEXT:    andl $1, %eax
-; X86-NEXT:    subl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    btl %edx, %esi
+; X86-NEXT:    sbbl $0, %ecx
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    popl %esi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_i32_sub_sub_var:
 ; X64:       # %bb.0:
-; X64-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-NEXT:    # kill: def $edi killed $edi def $rdi
-; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shrl %cl, %edx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    subl %esi, %edx
-; X64-NEXT:    leal (%rdx,%rdi), %eax
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    btl %ecx, %edx
+; X64-NEXT:    sbbl $0, %esi
+; X64-NEXT:    subl %esi, %eax
 ; X64-NEXT:    retq
   %shift = lshr i32 %z, %w
   %mask = and i32 %shift, 1