39 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i686-linux -mattr=+sse4.2 | FileCheck %s --check-prefix=X863; RUN: llc < %s -mtriple=x86_64-linux -mattr=+sse4.2 | FileCheck %s --check-prefix=X644 5; PR48916 7; Both loads should happen before either store.8 9define void @short2_int_swap(ptr nocapture %b, ptr nocapture %c) nounwind {10; X86-LABEL: short2_int_swap:11; X86: # %bb.0: # %entry12; X86-NEXT: pushl %esi13; X86-NEXT: movl {{[0-9]+}}(%esp), %eax14; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx15; X86-NEXT: movl (%ecx), %edx16; X86-NEXT: movl (%eax), %esi17; X86-NEXT: movl %esi, (%ecx)18; X86-NEXT: movl %edx, (%eax)19; X86-NEXT: popl %esi20; X86-NEXT: retl21;22; X64-LABEL: short2_int_swap:23; X64: # %bb.0: # %entry24; X64-NEXT: movl (%rdi), %eax25; X64-NEXT: movl (%rsi), %ecx26; X64-NEXT: movl %ecx, (%rdi)27; X64-NEXT: movl %eax, (%rsi)28; X64-NEXT: retq29entry:30 %0 = load <2 x i16>, ptr %b, align 2 ; <<2 x i16>> [#uses=1]31 %1 = load i32, ptr %c, align 4 ; <i32> [#uses=1]32 %tmp1 = bitcast i32 %1 to <2 x i16> ; <<2 x i16>> [#uses=1]33 store <2 x i16> %tmp1, ptr %b, align 234 %tmp5 = bitcast <2 x i16> %0 to <1 x i32> ; <<1 x i32>> [#uses=1]35 %tmp3 = extractelement <1 x i32> %tmp5, i32 0 ; <i32> [#uses=1]36 store i32 %tmp3, ptr %c, align 437 ret void38}39