[llvm] [DAGCombiner] Fix exact power-of-two signed division for large integers (PR #177340)
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Sun Jan 25 06:40:46 PST 2026
================
@@ -0,0 +1,481 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefix=X64
+; RUN: llc < %s -mtriple=x86_64-- -O0 | FileCheck %s --check-prefix=X64-O0
+; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefix=X32
+; RUN: llc < %s -mtriple=i686-- -O0 | FileCheck %s --check-prefix=X32-O0
+
+define i129 @v_sdiv_i129_v_pow2k(i129 %lhs) {
+; X64-LABEL: v_sdiv_i129_v_pow2k:
+; X64: # %bb.0:
+; X64-NEXT: movq %rdx, %rcx
+; X64-NEXT: andl $1, %edx
+; X64-NEXT: negq %rdx
+; X64-NEXT: movl %edx, %eax
+; X64-NEXT: andl $1, %eax
+; X64-NEXT: shldq $32, %rdx, %rax
+; X64-NEXT: addq %rdi, %rax
+; X64-NEXT: adcq $0, %rsi
+; X64-NEXT: adcq $0, %rcx
+; X64-NEXT: shrdq $33, %rsi, %rax
+; X64-NEXT: andl $1, %ecx
+; X64-NEXT: movq %rcx, %rdx
+; X64-NEXT: negq %rdx
+; X64-NEXT: shldq $31, %rsi, %rdx
+; X64-NEXT: retq
+;
+; X64-O0-LABEL: v_sdiv_i129_v_pow2k:
+; X64-O0: # %bb.0:
+; X64-O0-NEXT: movl %edx, %eax
+; X64-O0-NEXT: andl $1, %eax
+; X64-O0-NEXT: movl %eax, %ecx
+; X64-O0-NEXT: negq %rcx
+; X64-O0-NEXT: movl %ecx, %r8d
+; X64-O0-NEXT: andl $1, %r8d
+; X64-O0-NEXT: # implicit-def: $rax
+; X64-O0-NEXT: movl %r8d, %eax
+; X64-O0-NEXT: shldq $32, %rcx, %rax
+; X64-O0-NEXT: addq %rax, %rdi
+; X64-O0-NEXT: adcq $0, %rsi
+; X64-O0-NEXT: adcq $0, %rdx
+; X64-O0-NEXT: movq %rsi, %rax
+; X64-O0-NEXT: shldq $31, %rdi, %rax
+; X64-O0-NEXT: movl %edx, %ecx
+; X64-O0-NEXT: andl $1, %ecx
+; X64-O0-NEXT: # kill: def $rcx killed $ecx
+; X64-O0-NEXT: movq %rcx, %rdx
+; X64-O0-NEXT: negq %rdx
+; X64-O0-NEXT: shldq $31, %rsi, %rdx
+; X64-O0-NEXT: retq
+;
+; X32-LABEL: v_sdiv_i129_v_pow2k:
+; X32: # %bb.0:
+; X32-NEXT: pushl %ebx
+; X32-NEXT: .cfi_def_cfa_offset 8
+; X32-NEXT: pushl %edi
+; X32-NEXT: .cfi_def_cfa_offset 12
+; X32-NEXT: pushl %esi
+; X32-NEXT: .cfi_def_cfa_offset 16
+; X32-NEXT: .cfi_offset %esi, -16
+; X32-NEXT: .cfi_offset %edi, -12
+; X32-NEXT: .cfi_offset %ebx, -8
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movl %ecx, %ebx
+; X32-NEXT: andl $1, %ebx
+; X32-NEXT: negl %ebx
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT: movl %ebx, %edi
+; X32-NEXT: andl $1, %edi
+; X32-NEXT: addl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT: adcl {{[0-9]+}}(%esp), %edi
+; X32-NEXT: adcl $0, %esi
+; X32-NEXT: adcl $0, %edx
+; X32-NEXT: adcl $0, %ecx
+; X32-NEXT: movl %ecx, %ebx
+; X32-NEXT: shldl $31, %edx, %ebx
+; X32-NEXT: shldl $31, %esi, %edx
+; X32-NEXT: shldl $31, %edi, %esi
+; X32-NEXT: andl $1, %ecx
+; X32-NEXT: movl %ecx, %edi
+; X32-NEXT: negl %edi
+; X32-NEXT: movl %esi, (%eax)
+; X32-NEXT: movl %edx, 4(%eax)
+; X32-NEXT: movl %ebx, 8(%eax)
+; X32-NEXT: movl %edi, 12(%eax)
+; X32-NEXT: movb %cl, 16(%eax)
+; X32-NEXT: popl %esi
+; X32-NEXT: .cfi_def_cfa_offset 12
+; X32-NEXT: popl %edi
+; X32-NEXT: .cfi_def_cfa_offset 8
+; X32-NEXT: popl %ebx
+; X32-NEXT: .cfi_def_cfa_offset 4
+; X32-NEXT: retl $4
+;
+; X32-O0-LABEL: v_sdiv_i129_v_pow2k:
+; X32-O0: # %bb.0:
+; X32-O0-NEXT: pushl %ebp
+; X32-O0-NEXT: .cfi_def_cfa_offset 8
+; X32-O0-NEXT: pushl %ebx
+; X32-O0-NEXT: .cfi_def_cfa_offset 12
+; X32-O0-NEXT: pushl %edi
+; X32-O0-NEXT: .cfi_def_cfa_offset 16
+; X32-O0-NEXT: pushl %esi
+; X32-O0-NEXT: .cfi_def_cfa_offset 20
+; X32-O0-NEXT: subl $8, %esp
+; X32-O0-NEXT: .cfi_def_cfa_offset 28
+; X32-O0-NEXT: .cfi_offset %esi, -20
+; X32-O0-NEXT: .cfi_offset %edi, -16
+; X32-O0-NEXT: .cfi_offset %ebx, -12
+; X32-O0-NEXT: .cfi_offset %ebp, -8
+; X32-O0-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-O0-NEXT: movl %eax, (%esp) # 4-byte Spill
+; X32-O0-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-O0-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X32-O0-NEXT: movl {{[0-9]+}}(%esp), %ebp
+; X32-O0-NEXT: movl {{[0-9]+}}(%esp), %ebx
+; X32-O0-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-O0-NEXT: movl {{[0-9]+}}(%esp), %esi
+; X32-O0-NEXT: movl %edx, %ecx
+; X32-O0-NEXT: andl $1, %ecx
+; X32-O0-NEXT: negl %ecx
+; X32-O0-NEXT: movl %ecx, %edi
+; X32-O0-NEXT: andl $1, %edi
+; X32-O0-NEXT: addl %ecx, %eax
+; X32-O0-NEXT: movl (%esp), %ecx # 4-byte Reload
+; X32-O0-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-O0-NEXT: adcl %edi, %esi
+; X32-O0-NEXT: adcl $0, %ebp
+; X32-O0-NEXT: adcl $0, %ebx
+; X32-O0-NEXT: adcl $0, %edx
+; X32-O0-NEXT: movl %edx, %edi
+; X32-O0-NEXT: shldl $31, %ebx, %edi
+; X32-O0-NEXT: shldl $31, %ebp, %ebx
+; X32-O0-NEXT: shldl $31, %esi, %ebp
+; X32-O0-NEXT: andl $1, %edx
+; X32-O0-NEXT: movl %edx, %esi
+; X32-O0-NEXT: negl %esi
+; X32-O0-NEXT: movl %ebp, (%ecx)
+; X32-O0-NEXT: movl %ebx, 4(%ecx)
+; X32-O0-NEXT: movl %edi, 8(%ecx)
+; X32-O0-NEXT: movl %esi, 12(%ecx)
+; X32-O0-NEXT: # kill: def $dl killed $dl killed $edx
+; X32-O0-NEXT: movb %dl, 16(%ecx)
+; X32-O0-NEXT: addl $8, %esp
+; X32-O0-NEXT: .cfi_def_cfa_offset 20
+; X32-O0-NEXT: popl %esi
+; X32-O0-NEXT: .cfi_def_cfa_offset 16
+; X32-O0-NEXT: popl %edi
+; X32-O0-NEXT: .cfi_def_cfa_offset 12
+; X32-O0-NEXT: popl %ebx
+; X32-O0-NEXT: .cfi_def_cfa_offset 8
+; X32-O0-NEXT: popl %ebp
+; X32-O0-NEXT: .cfi_def_cfa_offset 4
+; X32-O0-NEXT: retl $4
+ %div = sdiv i129 %lhs, 8589934592
+ ret i129 %div
+}
+
+define i129 @v_sdiv_exact_i129_v_pow2k(i129 %lhs) {
----------------
RKSimon wrote:
use nounwind
https://github.com/llvm/llvm-project/pull/177340
More information about the llvm-commits
mailing list