45 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2;3; PR30298: Check if the target doesn't have SSE2, compiler will not crash4; or generate incorrect code because of vector mul width shrinking optimization.5;6; RUN: llc -mtriple=i386-pc-linux-gnu -mattr=+sse < %s | FileCheck %s7 8@c = external dso_local global ptr, align 89 10define void @mul_2xi8(ptr nocapture readonly %a, ptr nocapture readonly %b, i64 %index) nounwind {11; CHECK-LABEL: mul_2xi8:12; CHECK: # %bb.0: # %entry13; CHECK-NEXT: pushl %ebx14; CHECK-NEXT: pushl %edi15; CHECK-NEXT: pushl %esi16; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax17; CHECK-NEXT: movl {{[0-9]+}}(%esp), %ecx18; CHECK-NEXT: movl {{[0-9]+}}(%esp), %edx19; CHECK-NEXT: movl c, %esi20; CHECK-NEXT: movzbl 1(%edx,%ecx), %edi21; CHECK-NEXT: movzbl (%edx,%ecx), %edx22; CHECK-NEXT: movzbl 1(%eax,%ecx), %ebx23; CHECK-NEXT: imull %edi, %ebx24; CHECK-NEXT: movzbl (%eax,%ecx), %eax25; CHECK-NEXT: imull %edx, %eax26; CHECK-NEXT: movl %ebx, 4(%esi,%ecx,4)27; CHECK-NEXT: movl %eax, (%esi,%ecx,4)28; CHECK-NEXT: popl %esi29; CHECK-NEXT: popl %edi30; CHECK-NEXT: popl %ebx31; CHECK-NEXT: retl32entry:33 %pre = load ptr, ptr @c34 %tmp6 = getelementptr inbounds i8, ptr %a, i64 %index35 %wide.load = load <2 x i8>, ptr %tmp6, align 136 %tmp8 = zext <2 x i8> %wide.load to <2 x i32>37 %tmp10 = getelementptr inbounds i8, ptr %b, i64 %index38 %wide.load17 = load <2 x i8>, ptr %tmp10, align 139 %tmp12 = zext <2 x i8> %wide.load17 to <2 x i32>40 %tmp13 = mul nuw nsw <2 x i32> %tmp12, %tmp841 %tmp14 = getelementptr inbounds i32, ptr %pre, i64 %index42 store <2 x i32> %tmp13, ptr %tmp14, align 443 ret void44}45