[llvm] [X86] Fold AND(Y, XOR(X, SUB(0, X))) to ANDN(Y, BLSMSK(X)) (PR #128348)

[Simon Pilgrim via llvm-commits]

================
@@ -0,0 +1,346 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=i686-- -mattr=-bmi,+sse2 | FileCheck %s --check-prefixes=X86,X86-NOBMI
+; RUN: llc < %s -mtriple=i686-- -mattr=+bmi,+sse2 | FileCheck %s --check-prefixes=X86,X86-BMI
+; RUN: llc < %s -mtriple=x86_64-- -mattr=-bmi | FileCheck %s --check-prefixes=X64,X64-NOBMI
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+bmi | FileCheck %s --check-prefixes=X64,X64-BMI
+
+declare void @use(i32)
+
+define i32 @fold_and_xor_neg_v1_32(i32 %x, i32 %y) {
+; X86-NOBMI-LABEL: fold_and_xor_neg_v1_32:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl %ecx, %eax
+; X86-NOBMI-NEXT:    negl %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %eax
+; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI-LABEL: fold_and_xor_neg_v1_32:
+; X86-BMI:       # %bb.0:
+; X86-BMI-NEXT:    blsmskl {{[0-9]+}}(%esp), %eax
+; X86-BMI-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI-NEXT:    retl
+;
+; X64-NOBMI-LABEL: fold_and_xor_neg_v1_32:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    negl %eax
+; X64-NOBMI-NEXT:    xorl %edi, %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI-LABEL: fold_and_xor_neg_v1_32:
+; X64-BMI:       # %bb.0:
+; X64-BMI-NEXT:    blsmskl %edi, %eax
+; X64-BMI-NEXT:    andnl %esi, %eax, %eax
+; X64-BMI-NEXT:    retq
+  %neg = sub i32 0, %x
+  %xor = xor i32 %x, %neg
+  %and = and i32 %xor, %y
+  ret i32 %and
+}
+
+define i32 @fold_and_xor_neg_v2_32(i32 %x, i32 %y) {
+; X86-NOBMI-LABEL: fold_and_xor_neg_v2_32:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl %ecx, %eax
+; X86-NOBMI-NEXT:    negl %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %eax
+; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI-LABEL: fold_and_xor_neg_v2_32:
+; X86-BMI:       # %bb.0:
+; X86-BMI-NEXT:    blsmskl {{[0-9]+}}(%esp), %eax
+; X86-BMI-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI-NEXT:    retl
+;
+; X64-NOBMI-LABEL: fold_and_xor_neg_v2_32:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    negl %eax
+; X64-NOBMI-NEXT:    xorl %edi, %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI-LABEL: fold_and_xor_neg_v2_32:
+; X64-BMI:       # %bb.0:
+; X64-BMI-NEXT:    blsmskl %edi, %eax
+; X64-BMI-NEXT:    andnl %esi, %eax, %eax
+; X64-BMI-NEXT:    retq
+  %neg = sub i32 0, %x
+  %xor = xor i32 %x, %neg
+  %and = and i32 %y, %xor
+  ret i32 %and
+}
+
+define i32 @fold_and_xor_neg_v3_32(i32 %x, i32 %y) {
+; X86-NOBMI-LABEL: fold_and_xor_neg_v3_32:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl %ecx, %eax
+; X86-NOBMI-NEXT:    negl %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %eax
+; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI-LABEL: fold_and_xor_neg_v3_32:
+; X86-BMI:       # %bb.0:
+; X86-BMI-NEXT:    blsmskl {{[0-9]+}}(%esp), %eax
+; X86-BMI-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI-NEXT:    retl
+;
+; X64-NOBMI-LABEL: fold_and_xor_neg_v3_32:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    negl %eax
+; X64-NOBMI-NEXT:    xorl %edi, %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI-LABEL: fold_and_xor_neg_v3_32:
+; X64-BMI:       # %bb.0:
+; X64-BMI-NEXT:    blsmskl %edi, %eax
+; X64-BMI-NEXT:    andnl %esi, %eax, %eax
+; X64-BMI-NEXT:    retq
+  %neg = sub i32 0, %x
+  %xor = xor i32 %neg, %x
+  %and = and i32 %xor, %y
+  ret i32 %and
+}
+
+define i32 @fold_and_xor_neg_v4_32(i32 %x, i32 %y) {
+; X86-NOBMI-LABEL: fold_and_xor_neg_v4_32:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl %ecx, %eax
+; X86-NOBMI-NEXT:    negl %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %eax
+; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI-LABEL: fold_and_xor_neg_v4_32:
+; X86-BMI:       # %bb.0:
+; X86-BMI-NEXT:    blsmskl {{[0-9]+}}(%esp), %eax
+; X86-BMI-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI-NEXT:    retl
+;
+; X64-NOBMI-LABEL: fold_and_xor_neg_v4_32:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    negl %eax
+; X64-NOBMI-NEXT:    xorl %edi, %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI-LABEL: fold_and_xor_neg_v4_32:
+; X64-BMI:       # %bb.0:
+; X64-BMI-NEXT:    blsmskl %edi, %eax
+; X64-BMI-NEXT:    andnl %esi, %eax, %eax
+; X64-BMI-NEXT:    retq
+  %neg = sub i32 0, %x
+  %xor = xor i32 %neg, %x
+  %and = and i32 %y, %xor
+  ret i32 %and
+}
+
+define i64 @fold_and_xor_neg_v1_64(i64 %x, i64 %y) {
----------------
RKSimon wrote:

`define i64 @fold_and_xor_neg_v1_64(i64 %x, i64 %y) nounwind {` to silence cfi noise - same with others

https://github.com/llvm/llvm-project/pull/128348