Skip to content

Commit f5f4825

Browse files
goldsteinntstellar
authored andcommitted
[X86] Fix (shift X, (xor Y, N-1)) -> (shift X, (not Y)) by properly inserting not Y into DAG. [#61038]
Previously not inserting the `-1` in `not Y` (`xor Y, -1`) into the DAG. Not inserting `-1` as a DAG node comes up as a bug when doing `(xor (shl 1, A), B)` -> `(btc A, B)`. `btc` requires `B` (dst) to be a register. Differential Revision: https://reviews.llvm.org/D144984 (cherry picked from commit 4a23031)
1 parent da05802 commit f5f4825

File tree

2 files changed

+52
-1
lines changed

2 files changed

+52
-1
lines changed

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4020,7 +4020,10 @@ bool X86DAGToDAGISel::tryShiftAmountMod(SDNode *N) {
40204020

40214021
EVT OpVT = ShiftAmt.getValueType();
40224022

4023-
NewShiftAmt = CurDAG->getNOT(DL, Add0C == nullptr ? Add0 : Add1, OpVT);
4023+
SDValue AllOnes = CurDAG->getAllOnesConstant(DL, OpVT);
4024+
NewShiftAmt = CurDAG->getNode(ISD::XOR, DL, OpVT,
4025+
Add0C == nullptr ? Add0 : Add1, AllOnes);
4026+
insertDAGNode(*CurDAG, OrigShiftAmt, AllOnes);
40244027
insertDAGNode(*CurDAG, OrigShiftAmt, NewShiftAmt);
40254028
// If we are shifting by N-X where N == 0 mod Size, then just shift by
40264029
// -X to generate a NEG instead of a SUB of a constant.

llvm/test/CodeGen/X86/pr61038.ll

Lines changed: 48 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,48 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2+
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skylake-avx512 | FileCheck %s --check-prefix=CHECK-BMI2
3+
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefix=CHECK-BMI2
4+
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+bmi | FileCheck %s --check-prefix=CHECK-BMI
5+
6+
declare i32 @llvm.cttz.i32(i32, i1 immarg)
7+
define void @test_61038(ptr %tmp_buffer) {
8+
; CHECK-BMI2-LABEL: test_61038:
9+
; CHECK-BMI2: # %bb.0: # %entry
10+
; CHECK-BMI2-NEXT: tzcntl %eax, %eax
11+
; CHECK-BMI2-NEXT: movabsq $8589934591, %rcx # imm = 0x1FFFFFFFF
12+
; CHECK-BMI2-NEXT: movq $-1, %rdx
13+
; CHECK-BMI2-NEXT: btcq %rax, %rdx
14+
; CHECK-BMI2-NEXT: shrxq %rdx, %rcx, %rdx
15+
; CHECK-BMI2-NEXT: btcq %rax, %rcx
16+
; CHECK-BMI2-NEXT: xorl %eax, %eax
17+
; CHECK-BMI2-NEXT: cmpq $64, %rcx
18+
; CHECK-BMI2-NEXT: cmovael %eax, %edx
19+
; CHECK-BMI2-NEXT: movl %edx, (%rdi)
20+
; CHECK-BMI2-NEXT: retq
21+
;
22+
; CHECK-BMI-LABEL: test_61038:
23+
; CHECK-BMI: # %bb.0: # %entry
24+
; CHECK-BMI-NEXT: tzcntl %eax, %eax
25+
; CHECK-BMI-NEXT: movabsq $8589934591, %rdx # imm = 0x1FFFFFFFF
26+
; CHECK-BMI-NEXT: movq %rdx, %rsi
27+
; CHECK-BMI-NEXT: btcq %rax, %rsi
28+
; CHECK-BMI-NEXT: movq $-1, %rcx
29+
; CHECK-BMI-NEXT: btcq %rax, %rcx
30+
; CHECK-BMI-NEXT: # kill: def $cl killed $cl killed $rcx
31+
; CHECK-BMI-NEXT: shrq %cl, %rdx
32+
; CHECK-BMI-NEXT: xorl %eax, %eax
33+
; CHECK-BMI-NEXT: cmpq $64, %rsi
34+
; CHECK-BMI-NEXT: cmovael %eax, %edx
35+
; CHECK-BMI-NEXT: movl %edx, (%rdi)
36+
; CHECK-BMI-NEXT: retq
37+
entry:
38+
%0 = tail call i32 @llvm.cttz.i32(i32 poison, i1 false)
39+
%1 = zext i32 %0 to i64
40+
%2 = shl nuw nsw i64 1, %1
41+
%3 = xor i64 %2, 8589934591
42+
%4 = icmp ugt i64 %3, 63
43+
%x11.op.i = lshr i64 8589934591, %3
44+
%5 = trunc i64 %x11.op.i to i32
45+
%6 = select i1 %4, i32 0, i32 %5
46+
store i32 %6, ptr %tmp_buffer, align 4
47+
ret void
48+
}

0 commit comments

Comments
 (0)