93 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X863; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X644; RUN: llc < %s -mtriple=i686-windows -mattr=+sse2 | FileCheck %s --check-prefix=X86-WIN5; RUN: llc < %s -mtriple=x86_64-windows -mattr=+sse2 | FileCheck %s --check-prefix=X64-WIN6 7; Verify that we are using the efficient uitofp --> sitofp lowering illustrated8; by the compiler_rt implementation of __floatundisf.9; <rdar://problem/8493982>10 11define float @test(i64 %a) nounwind {12; X86-LABEL: test:13; X86: # %bb.0: # %entry14; X86-NEXT: pushl %ebp15; X86-NEXT: movl %esp, %ebp16; X86-NEXT: andl $-8, %esp17; X86-NEXT: subl $16, %esp18; X86-NEXT: movl 12(%ebp), %eax19; X86-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero20; X86-NEXT: movlps %xmm0, {{[0-9]+}}(%esp)21; X86-NEXT: shrl $31, %eax22; X86-NEXT: fildll {{[0-9]+}}(%esp)23; X86-NEXT: fadds {{\.?LCPI[0-9]+_[0-9]+}}(,%eax,4)24; X86-NEXT: fstps {{[0-9]+}}(%esp)25; X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero26; X86-NEXT: movss %xmm0, (%esp)27; X86-NEXT: flds (%esp)28; X86-NEXT: movl %ebp, %esp29; X86-NEXT: popl %ebp30; X86-NEXT: retl31;32; X64-LABEL: test:33; X64: # %bb.0: # %entry34; X64-NEXT: testq %rdi, %rdi35; X64-NEXT: js .LBB0_136; X64-NEXT: # %bb.2: # %entry37; X64-NEXT: cvtsi2ss %rdi, %xmm038; X64-NEXT: retq39; X64-NEXT: .LBB0_1:40; X64-NEXT: movq %rdi, %rax41; X64-NEXT: shrq %rax42; X64-NEXT: andl $1, %edi43; X64-NEXT: orq %rax, %rdi44; X64-NEXT: cvtsi2ss %rdi, %xmm045; X64-NEXT: addss %xmm0, %xmm046; X64-NEXT: retq47;48; X86-WIN-LABEL: test:49; X86-WIN: # %bb.0: # %entry50; X86-WIN-NEXT: pushl %ebp51; X86-WIN-NEXT: movl %esp, %ebp52; X86-WIN-NEXT: andl $-8, %esp53; X86-WIN-NEXT: subl $24, %esp54; X86-WIN-NEXT: movl 12(%ebp), %eax55; X86-WIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero56; X86-WIN-NEXT: movlps %xmm0, {{[0-9]+}}(%esp)57; X86-WIN-NEXT: shrl $31, %eax58; X86-WIN-NEXT: fildll {{[0-9]+}}(%esp)59; X86-WIN-NEXT: fnstcw {{[0-9]+}}(%esp)60; X86-WIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx61; X86-WIN-NEXT: orl $768, %ecx # imm = 0x30062; X86-WIN-NEXT: movw %cx, {{[0-9]+}}(%esp)63; X86-WIN-NEXT: fldcw {{[0-9]+}}(%esp)64; X86-WIN-NEXT: fadds __real@5f80000000000000(,%eax,4)65; X86-WIN-NEXT: fldcw {{[0-9]+}}(%esp)66; X86-WIN-NEXT: fstps {{[0-9]+}}(%esp)67; X86-WIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero68; X86-WIN-NEXT: movss %xmm0, {{[0-9]+}}(%esp)69; X86-WIN-NEXT: flds {{[0-9]+}}(%esp)70; X86-WIN-NEXT: movl %ebp, %esp71; X86-WIN-NEXT: popl %ebp72; X86-WIN-NEXT: retl73;74; X64-WIN-LABEL: test:75; X64-WIN: # %bb.0: # %entry76; X64-WIN-NEXT: testq %rcx, %rcx77; X64-WIN-NEXT: js .LBB0_178; X64-WIN-NEXT: # %bb.2: # %entry79; X64-WIN-NEXT: cvtsi2ss %rcx, %xmm080; X64-WIN-NEXT: retq81; X64-WIN-NEXT: .LBB0_1:82; X64-WIN-NEXT: movq %rcx, %rax83; X64-WIN-NEXT: shrq %rax84; X64-WIN-NEXT: andl $1, %ecx85; X64-WIN-NEXT: orq %rax, %rcx86; X64-WIN-NEXT: cvtsi2ss %rcx, %xmm087; X64-WIN-NEXT: addss %xmm0, %xmm088; X64-WIN-NEXT: retq89entry:90 %b = uitofp i64 %a to float91 ret float %b92}93