llvm/test/CodeGen/X86/scalar_widen_div.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.2 |  FileCheck %s

; Verify when widening a divide/remainder operation, we only generate a
; divide/rem per element since divide/remainder can trap.

define void @div_extract(ptr %dst, ptr %lhs, ptr %rhs) nounwind {
; CHECK-LABEL: div_extract:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movzwl (%rsi), %eax
; CHECK-NEXT:    movd %eax, %xmm0
; CHECK-NEXT:    movzwl (%rdx), %eax
; CHECK-NEXT:    movd %eax, %xmm1
; CHECK-NEXT:    pmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
; CHECK-NEXT:    cvtdq2ps %xmm1, %xmm1
; CHECK-NEXT:    pmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
; CHECK-NEXT:    cvtdq2ps %xmm0, %xmm0
; CHECK-NEXT:    divps %xmm1, %xmm0
; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
; CHECK-NEXT:    pextrb $0, %xmm0, (%rdi)
; CHECK-NEXT:    pextrb $4, %xmm0, 1(%rdi)
; CHECK-NEXT:    retq
  %a = load <2 x i8>, ptr %lhs
  %b = load <2 x i8>, ptr %rhs
  %quot = udiv <2 x i8> %a, %b
  %elt0 = extractelement <2 x i8> %quot, i64 0
  store i8 %elt0, ptr %dst, align 1
  %next = getelementptr inbounds nuw i8, ptr %dst, i64 1
  %elt1 = extractelement <2 x i8> %quot, i64 1
  store i8 %elt1, ptr %next, align 1
  ret void
}

define void @vectorDiv (ptr addrspace(1) %nsource, ptr addrspace(1) %dsource, ptr addrspace(1) %qdest) nounwind {
; CHECK-LABEL: vectorDiv:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp)
; CHECK-NEXT:    movslq -{{[0-9]+}}(%rsp), %rax
; CHECK-NEXT:    cvtdq2pd (%rsi,%rax,8), %xmm0
; CHECK-NEXT:    cvtdq2pd (%rdi,%rax,8), %xmm1
; CHECK-NEXT:    divpd %xmm0, %xmm1
; CHECK-NEXT:    cvttpd2dq %xmm1, %xmm0
; CHECK-NEXT:    movlpd %xmm0, (%rdx,%rax,8)
; CHECK-NEXT:    retq
entry:
  %nsource.addr = alloca ptr addrspace(1), align 4
  %dsource.addr = alloca ptr addrspace(1), align 4
  %qdest.addr = alloca ptr addrspace(1), align 4
  %index = alloca i32, align 4
  store ptr addrspace(1) %nsource, ptr %nsource.addr
  store ptr addrspace(1) %dsource, ptr %dsource.addr
  store ptr addrspace(1) %qdest, ptr %qdest.addr
  %tmp = load ptr addrspace(1), ptr %qdest.addr
  %tmp1 = load i32, ptr %index
  %arrayidx = getelementptr <2 x i32>, ptr addrspace(1) %tmp, i32 %tmp1
  %tmp2 = load ptr addrspace(1), ptr %nsource.addr
  %tmp3 = load i32, ptr %index
  %arrayidx4 = getelementptr <2 x i32>, ptr addrspace(1) %tmp2, i32 %tmp3
  %tmp5 = load <2 x i32>, ptr addrspace(1) %arrayidx4
  %tmp6 = load ptr addrspace(1), ptr %dsource.addr
  %tmp7 = load i32, ptr %index
  %arrayidx8 = getelementptr <2 x i32>, ptr addrspace(1) %tmp6, i32 %tmp7
  %tmp9 = load <2 x i32>, ptr addrspace(1) %arrayidx8
  %tmp10 = sdiv <2 x i32> %tmp5, %tmp9
  store <2 x i32> %tmp10, ptr addrspace(1) %arrayidx
  ret void
}

define <3 x i8> @test_char_div(<3 x i8> %num, <3 x i8> %div) {
; CHECK-LABEL: test_char_div:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movd %edi, %xmm0
; CHECK-NEXT:    pinsrb $1, %esi, %xmm0
; CHECK-NEXT:    pinsrb $2, %edx, %xmm0
; CHECK-NEXT:    movd %ecx, %xmm1
; CHECK-NEXT:    pinsrb $1, %r8d, %xmm1
; CHECK-NEXT:    pinsrb $2, %r9d, %xmm1
; CHECK-NEXT:    pmovsxbd %xmm1, %xmm1
; CHECK-NEXT:    cvtdq2ps %xmm1, %xmm1
; CHECK-NEXT:    pmovsxbd %xmm0, %xmm0
; CHECK-NEXT:    cvtdq2ps %xmm0, %xmm0
; CHECK-NEXT:    divps %xmm1, %xmm0
; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
; CHECK-NEXT:    packssdw %xmm0, %xmm0
; CHECK-NEXT:    packsswb %xmm0, %xmm0
; CHECK-NEXT:    movd %xmm0, %eax
; CHECK-NEXT:    pextrb $1, %xmm0, %edx
; CHECK-NEXT:    pextrb $2, %xmm0, %ecx
; CHECK-NEXT:    # kill: def $al killed $al killed $eax
; CHECK-NEXT:    # kill: def $dl killed $dl killed $edx
; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
; CHECK-NEXT:    retq
  %div.r = sdiv <3 x i8> %num, %div
  ret <3 x i8>  %div.r
}

define <3 x i8> @test_uchar_div(<3 x i8> %num, <3 x i8> %div) {
; CHECK-LABEL: test_uchar_div:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movd %edi, %xmm0
; CHECK-NEXT:    pinsrb $1, %esi, %xmm0
; CHECK-NEXT:    pinsrb $2, %edx, %xmm0
; CHECK-NEXT:    movd %ecx, %xmm1
; CHECK-NEXT:    pinsrb $1, %r8d, %xmm1
; CHECK-NEXT:    pinsrb $2, %r9d, %xmm1
; CHECK-NEXT:    pmovzxbd {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
; CHECK-NEXT:    cvtdq2ps %xmm1, %xmm1
; CHECK-NEXT:    pmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
; CHECK-NEXT:    cvtdq2ps %xmm0, %xmm0
; CHECK-NEXT:    divps %xmm1, %xmm0
; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
; CHECK-NEXT:    packusdw %xmm0, %xmm0
; CHECK-NEXT:    packuswb %xmm0, %xmm0
; CHECK-NEXT:    movd %xmm0, %eax
; CHECK-NEXT:    pextrb $1, %xmm0, %edx
; CHECK-NEXT:    pextrb $2, %xmm0, %ecx
; CHECK-NEXT:    # kill: def $al killed $al killed $eax
; CHECK-NEXT:    # kill: def $dl killed $dl killed $edx
; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
; CHECK-NEXT:    retq
  %div.r = udiv <3 x i8> %num, %div
  ret <3 x i8>  %div.r
}

define <5 x i16> @test_short_div(<5 x i16> %num, <5 x i16> %div) {
; CHECK-LABEL: test_short_div:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pmovsxwd %xmm1, %xmm2
; CHECK-NEXT:    cvtdq2ps %xmm2, %xmm2
; CHECK-NEXT:    pmovsxwd %xmm0, %xmm3
; CHECK-NEXT:    cvtdq2ps %xmm3, %xmm3
; CHECK-NEXT:    divps %xmm2, %xmm3
; CHECK-NEXT:    cvttps2dq %xmm3, %xmm2
; CHECK-NEXT:    packssdw %xmm2, %xmm2
; CHECK-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
; CHECK-NEXT:    pmovsxwd %xmm1, %xmm1
; CHECK-NEXT:    cvtdq2ps %xmm1, %xmm1
; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; CHECK-NEXT:    pmovsxwd %xmm0, %xmm0
; CHECK-NEXT:    cvtdq2ps %xmm0, %xmm0
; CHECK-NEXT:    divps %xmm1, %xmm0
; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
; CHECK-NEXT:    movdqa %xmm2, %xmm0
; CHECK-NEXT:    retq
  %div.r = sdiv <5 x i16> %num, %div
  ret <5 x i16>  %div.r
}

define <4 x i16> @test_ushort_div(<4 x i16> %num, <4 x i16> %div) {
; CHECK-LABEL: test_ushort_div:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; CHECK-NEXT:    cvtdq2ps %xmm1, %xmm1
; CHECK-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
; CHECK-NEXT:    cvtdq2ps %xmm0, %xmm0
; CHECK-NEXT:    divps %xmm1, %xmm0
; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
; CHECK-NEXT:    packusdw %xmm0, %xmm0
; CHECK-NEXT:    retq
  %div.r = udiv <4 x i16> %num, %div
  ret <4 x i16>  %div.r
}

define <3 x i32> @test_uint_div(<3 x i32> %num, <3 x i32> %div) {
; CHECK-LABEL: test_uint_div:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pextrd $2, %xmm0, %eax
; CHECK-NEXT:    pextrd $2, %xmm1, %ecx
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divl %ecx
; CHECK-NEXT:    movl %eax, %ecx
; CHECK-NEXT:    pextrd $1, %xmm0, %eax
; CHECK-NEXT:    pextrd $1, %xmm1, %esi
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divl %esi
; CHECK-NEXT:    movl %eax, %esi
; CHECK-NEXT:    movd %xmm0, %eax
; CHECK-NEXT:    movd %xmm1, %edi
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divl %edi
; CHECK-NEXT:    movd %eax, %xmm0
; CHECK-NEXT:    pinsrd $1, %esi, %xmm0
; CHECK-NEXT:    pinsrd $2, %ecx, %xmm0
; CHECK-NEXT:    retq
  %div.r = udiv <3 x i32> %num, %div
  ret <3 x i32>  %div.r
}

define <3 x i64> @test_long_div(<3 x i64> %num, <3 x i64> %div) {
; CHECK-LABEL: test_long_div:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %rdx, %r10
; CHECK-NEXT:    movq %rdi, %rax
; CHECK-NEXT:    cqto
; CHECK-NEXT:    idivq %rcx
; CHECK-NEXT:    movq %rax, %rcx
; CHECK-NEXT:    movq %rsi, %rax
; CHECK-NEXT:    cqto
; CHECK-NEXT:    idivq %r8
; CHECK-NEXT:    movq %rax, %rsi
; CHECK-NEXT:    movq %r10, %rax
; CHECK-NEXT:    cqto
; CHECK-NEXT:    idivq %r9
; CHECK-NEXT:    movq %rax, %rdi
; CHECK-NEXT:    movq %rcx, %rax
; CHECK-NEXT:    movq %rsi, %rdx
; CHECK-NEXT:    movq %rdi, %rcx
; CHECK-NEXT:    retq
  %div.r = sdiv <3 x i64> %num, %div
  ret <3 x i64>  %div.r
}

define <3 x i64> @test_ulong_div(<3 x i64> %num, <3 x i64> %div) {
; CHECK-LABEL: test_ulong_div:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %rdx, %r10
; CHECK-NEXT:    movq %rdi, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq %rcx
; CHECK-NEXT:    movq %rax, %rcx
; CHECK-NEXT:    movq %rsi, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq %r8
; CHECK-NEXT:    movq %rax, %rsi
; CHECK-NEXT:    movq %r10, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq %r9
; CHECK-NEXT:    movq %rax, %rdi
; CHECK-NEXT:    movq %rcx, %rax
; CHECK-NEXT:    movq %rsi, %rdx
; CHECK-NEXT:    movq %rdi, %rcx
; CHECK-NEXT:    retq
  %div.r = udiv <3 x i64> %num, %div
  ret <3 x i64>  %div.r
}

define <4 x i8> @test_char_rem(<4 x i8> %num, <4 x i8> %rem) {
; CHECK-LABEL: test_char_rem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pmovsxbd %xmm1, %xmm2
; CHECK-NEXT:    cvtdq2ps %xmm2, %xmm2
; CHECK-NEXT:    pmovsxbd %xmm0, %xmm3
; CHECK-NEXT:    cvtdq2ps %xmm3, %xmm3
; CHECK-NEXT:    divps %xmm2, %xmm3
; CHECK-NEXT:    cvttps2dq %xmm3, %xmm2
; CHECK-NEXT:    packssdw %xmm2, %xmm2
; CHECK-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; CHECK-NEXT:    pmullw %xmm2, %xmm1
; CHECK-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,2,4,6,u,u,u,u,u,u,u,u,u,u,u,u]
; CHECK-NEXT:    psubb %xmm1, %xmm0
; CHECK-NEXT:    retq
  %rem.r = srem <4 x i8> %num, %rem
  ret <4 x i8>  %rem.r
}

define <5 x i16> @test_short_rem(<5 x i16> %num, <5 x i16> %rem) {
; CHECK-LABEL: test_short_rem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pmovsxwd %xmm1, %xmm2
; CHECK-NEXT:    cvtdq2ps %xmm2, %xmm2
; CHECK-NEXT:    pmovsxwd %xmm0, %xmm3
; CHECK-NEXT:    cvtdq2ps %xmm3, %xmm3
; CHECK-NEXT:    divps %xmm2, %xmm3
; CHECK-NEXT:    cvttps2dq %xmm3, %xmm2
; CHECK-NEXT:    packssdw %xmm2, %xmm2
; CHECK-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[2,3,2,3]
; CHECK-NEXT:    pmovsxwd %xmm3, %xmm3
; CHECK-NEXT:    cvtdq2ps %xmm3, %xmm3
; CHECK-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
; CHECK-NEXT:    pmovsxwd %xmm4, %xmm4
; CHECK-NEXT:    cvtdq2ps %xmm4, %xmm4
; CHECK-NEXT:    divps %xmm3, %xmm4
; CHECK-NEXT:    cvttps2dq %xmm4, %xmm3
; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; CHECK-NEXT:    pmullw %xmm1, %xmm2
; CHECK-NEXT:    psubw %xmm2, %xmm0
; CHECK-NEXT:    retq
  %rem.r = srem <5 x i16> %num, %rem
  ret <5 x i16>  %rem.r
}

define <4 x i32> @test_uint_rem(<4 x i32> %num, <4 x i32> %rem) {
; CHECK-LABEL: test_uint_rem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    cvtdq2pd %xmm1, %xmm2
; CHECK-NEXT:    cvtdq2pd %xmm0, %xmm3
; CHECK-NEXT:    divpd %xmm2, %xmm3
; CHECK-NEXT:    cvttpd2dq %xmm3, %xmm2
; CHECK-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[2,3,2,3]
; CHECK-NEXT:    cvtdq2pd %xmm3, %xmm3
; CHECK-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
; CHECK-NEXT:    cvtdq2pd %xmm4, %xmm4
; CHECK-NEXT:    divpd %xmm3, %xmm4
; CHECK-NEXT:    cvttpd2dq %xmm4, %xmm3
; CHECK-NEXT:    unpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; CHECK-NEXT:    pmulld %xmm1, %xmm2
; CHECK-NEXT:    psubd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %rem.r = srem <4 x i32> %num, %rem
  ret <4 x i32>  %rem.r
}


define <5 x i64> @test_ulong_rem(<5 x i64> %num, <5 x i64> %rem) {
; CHECK-LABEL: test_ulong_rem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %rdx, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movq %rdx, %xmm0
; CHECK-NEXT:    movq %rsi, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movq %rdx, %xmm1
; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; CHECK-NEXT:    movq %r8, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movq %rdx, %xmm0
; CHECK-NEXT:    movq %rcx, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movq %rdx, %xmm2
; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
; CHECK-NEXT:    movq %r9, %rax
; CHECK-NEXT:    xorl %edx, %edx
; CHECK-NEXT:    divq {{[0-9]+}}(%rsp)
; CHECK-NEXT:    movq %rdx, 32(%rdi)
; CHECK-NEXT:    movdqa %xmm2, 16(%rdi)
; CHECK-NEXT:    movdqa %xmm1, (%rdi)
; CHECK-NEXT:    movq %rdi, %rax
; CHECK-NEXT:    retq
  %rem.r = urem <5 x i64> %num, %rem
  ret <5 x i64>  %rem.r
}

define void @test_int_div(ptr %dest, ptr %old, i32 %n) {
; CHECK-LABEL: test_int_div:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    testl %edx, %edx
; CHECK-NEXT:    jle .LBB13_3
; CHECK-NEXT:  # %bb.1: # %bb.nph
; CHECK-NEXT:    xorl %eax, %eax
; CHECK-NEXT:    .p2align 4
; CHECK-NEXT:  .LBB13_2: # %for.body
; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    movdqa (%rdi,%rax), %xmm0
; CHECK-NEXT:    movdqa (%rsi,%rax), %xmm1
; CHECK-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
; CHECK-NEXT:    cvtdq2pd %xmm2, %xmm2
; CHECK-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]
; CHECK-NEXT:    cvtdq2pd %xmm3, %xmm3
; CHECK-NEXT:    divpd %xmm2, %xmm3
; CHECK-NEXT:    cvttpd2dq %xmm3, %xmm2
; CHECK-NEXT:    cvtdq2pd %xmm1, %xmm1
; CHECK-NEXT:    cvtdq2pd %xmm0, %xmm0
; CHECK-NEXT:    divpd %xmm1, %xmm0
; CHECK-NEXT:    cvttpd2dq %xmm0, %xmm0
; CHECK-NEXT:    movlpd %xmm0, (%rdi,%rax)
; CHECK-NEXT:    movss %xmm2, 8(%rdi,%rax)
; CHECK-NEXT:    addq $16, %rax
; CHECK-NEXT:    decl %edx
; CHECK-NEXT:    jne .LBB13_2
; CHECK-NEXT:  .LBB13_3: # %for.end
; CHECK-NEXT:    retq
entry:
  %cmp13 = icmp sgt i32 %n, 0
  br i1 %cmp13, label %bb.nph, label %for.end

bb.nph:
  br label %for.body

for.body:
  %i.014 = phi i32 [ 0, %bb.nph ], [ %inc, %for.body ]
  %arrayidx11 = getelementptr <3 x i32>, ptr %dest, i32 %i.014
  %tmp4 = load <3 x i32>, ptr %arrayidx11 ; <<3 x i32>> [#uses=1]
  %arrayidx7 = getelementptr inbounds <3 x i32>, ptr %old, i32 %i.014
  %tmp8 = load <3 x i32>, ptr %arrayidx7 ; <<3 x i32>> [#uses=1]
  %div = sdiv <3 x i32> %tmp4, %tmp8
  store <3 x i32> %div, ptr %arrayidx11
  %inc = add nsw i32 %i.014, 1
  %exitcond = icmp eq i32 %inc, %n
  br i1 %exitcond, label %for.end, label %for.body

for.end:                                          ; preds = %for.body, %entry
  ret void
}