| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py |
| ; RUN: llc -mtriple=i686-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=X86 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=X64 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 < %s | FileCheck %s --check-prefixes=X64,SSE2 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.1 < %s | FileCheck %s --check-prefixes=X64,SSE41 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx < %s | FileCheck %s --check-prefixes=X64,AVX1 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 < %s | FileCheck %s --check-prefixes=X64,AVX2 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl < %s | FileCheck %s --check-prefixes=X64,AVX512VL |
| |
| define i1 @test_urem_odd(i13 %X) nounwind { |
| ; X86-LABEL: test_urem_odd: |
| ; X86: # %bb.0: |
| ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax |
| ; X86-NEXT: andl $8191, %eax # imm = 0x1FFF |
| ; X86-NEXT: imull $-13107, %eax, %eax # imm = 0xCCCD |
| ; X86-NEXT: movzwl %ax, %eax |
| ; X86-NEXT: cmpl $13108, %eax # imm = 0x3334 |
| ; X86-NEXT: setb %al |
| ; X86-NEXT: retl |
| ; |
| ; X64-LABEL: test_urem_odd: |
| ; X64: # %bb.0: |
| ; X64-NEXT: andl $8191, %edi # imm = 0x1FFF |
| ; X64-NEXT: imull $-13107, %edi, %eax # imm = 0xCCCD |
| ; X64-NEXT: movzwl %ax, %eax |
| ; X64-NEXT: cmpl $13108, %eax # imm = 0x3334 |
| ; X64-NEXT: setb %al |
| ; X64-NEXT: retq |
| %urem = urem i13 %X, 5 |
| %cmp = icmp eq i13 %urem, 0 |
| ret i1 %cmp |
| } |
| |
| define i1 @test_urem_even(i27 %X) nounwind { |
| ; X86-LABEL: test_urem_even: |
| ; X86: # %bb.0: |
| ; X86-NEXT: movl $134217727, %eax # imm = 0x7FFFFFF |
| ; X86-NEXT: andl {{[0-9]+}}(%esp), %eax |
| ; X86-NEXT: imull $-1227133513, %eax, %eax # imm = 0xB6DB6DB7 |
| ; X86-NEXT: rorl %eax |
| ; X86-NEXT: cmpl $306783379, %eax # imm = 0x12492493 |
| ; X86-NEXT: setb %al |
| ; X86-NEXT: retl |
| ; |
| ; X64-LABEL: test_urem_even: |
| ; X64: # %bb.0: |
| ; X64-NEXT: andl $134217727, %edi # imm = 0x7FFFFFF |
| ; X64-NEXT: imull $-1227133513, %edi, %eax # imm = 0xB6DB6DB7 |
| ; X64-NEXT: rorl %eax |
| ; X64-NEXT: cmpl $306783379, %eax # imm = 0x12492493 |
| ; X64-NEXT: setb %al |
| ; X64-NEXT: retq |
| %urem = urem i27 %X, 14 |
| %cmp = icmp eq i27 %urem, 0 |
| ret i1 %cmp |
| } |
| |
| define i1 @test_urem_odd_setne(i4 %X) nounwind { |
| ; X86-LABEL: test_urem_odd_setne: |
| ; X86: # %bb.0: |
| ; X86-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-NEXT: andb $15, %al |
| ; X86-NEXT: movzbl %al, %eax |
| ; X86-NEXT: imull $-51, %eax, %eax |
| ; X86-NEXT: cmpb $51, %al |
| ; X86-NEXT: seta %al |
| ; X86-NEXT: retl |
| ; |
| ; X64-LABEL: test_urem_odd_setne: |
| ; X64: # %bb.0: |
| ; X64-NEXT: andb $15, %dil |
| ; X64-NEXT: movzbl %dil, %eax |
| ; X64-NEXT: imull $-51, %eax, %eax |
| ; X64-NEXT: cmpb $51, %al |
| ; X64-NEXT: seta %al |
| ; X64-NEXT: retq |
| %urem = urem i4 %X, 5 |
| %cmp = icmp ne i4 %urem, 0 |
| ret i1 %cmp |
| } |
| |
| define i1 @test_urem_negative_odd(i9 %X) nounwind { |
| ; X86-LABEL: test_urem_negative_odd: |
| ; X86: # %bb.0: |
| ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax |
| ; X86-NEXT: andl $511, %eax # imm = 0x1FF |
| ; X86-NEXT: imull $-7885, %eax, %eax # imm = 0xE133 |
| ; X86-NEXT: movzwl %ax, %eax |
| ; X86-NEXT: cmpl $129, %eax |
| ; X86-NEXT: seta %al |
| ; X86-NEXT: retl |
| ; |
| ; X64-LABEL: test_urem_negative_odd: |
| ; X64: # %bb.0: |
| ; X64-NEXT: andl $511, %edi # imm = 0x1FF |
| ; X64-NEXT: imull $-7885, %edi, %eax # imm = 0xE133 |
| ; X64-NEXT: movzwl %ax, %eax |
| ; X64-NEXT: cmpl $129, %eax |
| ; X64-NEXT: seta %al |
| ; X64-NEXT: retq |
| %urem = urem i9 %X, -5 |
| %cmp = icmp ne i9 %urem, 0 |
| ret i1 %cmp |
| } |
| |
| define <3 x i1> @test_urem_vec(<3 x i11> %X) nounwind { |
| ; X86-LABEL: test_urem_vec: |
| ; X86: # %bb.0: |
| ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %edx |
| ; X86-NEXT: andl $2047, %edx # imm = 0x7FF |
| ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax |
| ; X86-NEXT: andl $2047, %eax # imm = 0x7FF |
| ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %ecx |
| ; X86-NEXT: andl $2047, %ecx # imm = 0x7FF |
| ; X86-NEXT: imull $-5325, %ecx, %ecx # imm = 0xEB33 |
| ; X86-NEXT: addl $10650, %ecx # imm = 0x299A |
| ; X86-NEXT: cmpw $32, %cx |
| ; X86-NEXT: seta %cl |
| ; X86-NEXT: imull $-21845, %eax, %eax # imm = 0xAAAB |
| ; X86-NEXT: rorw %ax |
| ; X86-NEXT: movzwl %ax, %eax |
| ; X86-NEXT: cmpl $10922, %eax # imm = 0x2AAA |
| ; X86-NEXT: seta %al |
| ; X86-NEXT: imull $28087, %edx, %edx # imm = 0x6DB7 |
| ; X86-NEXT: addl $-28087, %edx # imm = 0x9249 |
| ; X86-NEXT: movzwl %dx, %edx |
| ; X86-NEXT: cmpl $9362, %edx # imm = 0x2492 |
| ; X86-NEXT: seta %dl |
| ; X86-NEXT: retl |
| ; |
| ; SSE2-LABEL: test_urem_vec: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movl %esi, %eax |
| ; SSE2-NEXT: andl $2047, %eax # imm = 0x7FF |
| ; SSE2-NEXT: imull $9363, %eax, %ecx # imm = 0x2493 |
| ; SSE2-NEXT: shrl $16, %ecx |
| ; SSE2-NEXT: subl %ecx, %eax |
| ; SSE2-NEXT: movzwl %ax, %eax |
| ; SSE2-NEXT: shrl %eax |
| ; SSE2-NEXT: addl %ecx, %eax |
| ; SSE2-NEXT: shrl $2, %eax |
| ; SSE2-NEXT: leal (,%rax,8), %ecx |
| ; SSE2-NEXT: subl %ecx, %eax |
| ; SSE2-NEXT: addl %esi, %eax |
| ; SSE2-NEXT: andl $2047, %edi # imm = 0x7FF |
| ; SSE2-NEXT: imull $43691, %edi, %ecx # imm = 0xAAAB |
| ; SSE2-NEXT: shrl $17, %ecx |
| ; SSE2-NEXT: andl $-2, %ecx |
| ; SSE2-NEXT: leal (%rcx,%rcx,2), %ecx |
| ; SSE2-NEXT: subl %ecx, %edi |
| ; SSE2-NEXT: movd %edi, %xmm0 |
| ; SSE2-NEXT: pinsrw $2, %eax, %xmm0 |
| ; SSE2-NEXT: movl %edx, %eax |
| ; SSE2-NEXT: andl $2047, %eax # imm = 0x7FF |
| ; SSE2-NEXT: imull $161, %eax, %ecx |
| ; SSE2-NEXT: shrl $16, %ecx |
| ; SSE2-NEXT: subl %ecx, %eax |
| ; SSE2-NEXT: movzwl %ax, %eax |
| ; SSE2-NEXT: shrl %eax |
| ; SSE2-NEXT: addl %ecx, %eax |
| ; SSE2-NEXT: shrl $10, %eax |
| ; SSE2-NEXT: imull $2043, %eax, %eax # imm = 0x7FB |
| ; SSE2-NEXT: subl %eax, %edx |
| ; SSE2-NEXT: pinsrw $4, %edx, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: pcmpeqd {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE2-NEXT: pxor %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, -{{[0-9]+}}(%rsp) |
| ; SSE2-NEXT: movb -{{[0-9]+}}(%rsp), %al |
| ; SSE2-NEXT: movb -{{[0-9]+}}(%rsp), %dl |
| ; SSE2-NEXT: movb -{{[0-9]+}}(%rsp), %cl |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE41-LABEL: test_urem_vec: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movl %esi, %eax |
| ; SSE41-NEXT: andl $2047, %eax # imm = 0x7FF |
| ; SSE41-NEXT: imull $9363, %eax, %ecx # imm = 0x2493 |
| ; SSE41-NEXT: shrl $16, %ecx |
| ; SSE41-NEXT: subl %ecx, %eax |
| ; SSE41-NEXT: movzwl %ax, %eax |
| ; SSE41-NEXT: shrl %eax |
| ; SSE41-NEXT: addl %ecx, %eax |
| ; SSE41-NEXT: shrl $2, %eax |
| ; SSE41-NEXT: leal (,%rax,8), %ecx |
| ; SSE41-NEXT: subl %ecx, %eax |
| ; SSE41-NEXT: addl %esi, %eax |
| ; SSE41-NEXT: andl $2047, %edi # imm = 0x7FF |
| ; SSE41-NEXT: imull $43691, %edi, %ecx # imm = 0xAAAB |
| ; SSE41-NEXT: shrl $17, %ecx |
| ; SSE41-NEXT: andl $-2, %ecx |
| ; SSE41-NEXT: leal (%rcx,%rcx,2), %ecx |
| ; SSE41-NEXT: subl %ecx, %edi |
| ; SSE41-NEXT: movd %edi, %xmm0 |
| ; SSE41-NEXT: pinsrw $2, %eax, %xmm0 |
| ; SSE41-NEXT: movl %edx, %eax |
| ; SSE41-NEXT: andl $2047, %eax # imm = 0x7FF |
| ; SSE41-NEXT: imull $161, %eax, %ecx |
| ; SSE41-NEXT: shrl $16, %ecx |
| ; SSE41-NEXT: subl %ecx, %eax |
| ; SSE41-NEXT: movzwl %ax, %eax |
| ; SSE41-NEXT: shrl %eax |
| ; SSE41-NEXT: addl %ecx, %eax |
| ; SSE41-NEXT: shrl $10, %eax |
| ; SSE41-NEXT: imull $2043, %eax, %eax # imm = 0x7FB |
| ; SSE41-NEXT: subl %eax, %edx |
| ; SSE41-NEXT: pinsrw $4, %edx, %xmm0 |
| ; SSE41-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE41-NEXT: pcmpeqd {{.*}}(%rip), %xmm0 |
| ; SSE41-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE41-NEXT: pxor %xmm0, %xmm1 |
| ; SSE41-NEXT: movd %xmm1, %eax |
| ; SSE41-NEXT: pextrb $4, %xmm1, %edx |
| ; SSE41-NEXT: pextrb $8, %xmm1, %ecx |
| ; SSE41-NEXT: # kill: def $al killed $al killed $eax |
| ; SSE41-NEXT: # kill: def $dl killed $dl killed $edx |
| ; SSE41-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX1-LABEL: test_urem_vec: |
| ; AVX1: # %bb.0: |
| ; AVX1-NEXT: movl %esi, %eax |
| ; AVX1-NEXT: andl $2047, %eax # imm = 0x7FF |
| ; AVX1-NEXT: imull $9363, %eax, %ecx # imm = 0x2493 |
| ; AVX1-NEXT: shrl $16, %ecx |
| ; AVX1-NEXT: subl %ecx, %eax |
| ; AVX1-NEXT: movzwl %ax, %eax |
| ; AVX1-NEXT: shrl %eax |
| ; AVX1-NEXT: addl %ecx, %eax |
| ; AVX1-NEXT: shrl $2, %eax |
| ; AVX1-NEXT: leal (,%rax,8), %ecx |
| ; AVX1-NEXT: subl %ecx, %eax |
| ; AVX1-NEXT: addl %esi, %eax |
| ; AVX1-NEXT: andl $2047, %edi # imm = 0x7FF |
| ; AVX1-NEXT: imull $43691, %edi, %ecx # imm = 0xAAAB |
| ; AVX1-NEXT: shrl $17, %ecx |
| ; AVX1-NEXT: andl $-2, %ecx |
| ; AVX1-NEXT: leal (%rcx,%rcx,2), %ecx |
| ; AVX1-NEXT: subl %ecx, %edi |
| ; AVX1-NEXT: vmovd %edi, %xmm0 |
| ; AVX1-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 |
| ; AVX1-NEXT: movl %edx, %eax |
| ; AVX1-NEXT: andl $2047, %eax # imm = 0x7FF |
| ; AVX1-NEXT: imull $161, %eax, %ecx |
| ; AVX1-NEXT: shrl $16, %ecx |
| ; AVX1-NEXT: subl %ecx, %eax |
| ; AVX1-NEXT: movzwl %ax, %eax |
| ; AVX1-NEXT: shrl %eax |
| ; AVX1-NEXT: addl %ecx, %eax |
| ; AVX1-NEXT: shrl $10, %eax |
| ; AVX1-NEXT: imull $2043, %eax, %eax # imm = 0x7FB |
| ; AVX1-NEXT: subl %eax, %edx |
| ; AVX1-NEXT: vpinsrw $4, %edx, %xmm0, %xmm0 |
| ; AVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX1-NEXT: vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1 |
| ; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0 |
| ; AVX1-NEXT: vmovd %xmm0, %eax |
| ; AVX1-NEXT: vpextrb $4, %xmm0, %edx |
| ; AVX1-NEXT: vpextrb $8, %xmm0, %ecx |
| ; AVX1-NEXT: # kill: def $al killed $al killed $eax |
| ; AVX1-NEXT: # kill: def $dl killed $dl killed $edx |
| ; AVX1-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; AVX1-NEXT: retq |
| ; |
| ; AVX2-LABEL: test_urem_vec: |
| ; AVX2: # %bb.0: |
| ; AVX2-NEXT: andl $2047, %esi # imm = 0x7FF |
| ; AVX2-NEXT: imull $9363, %esi, %eax # imm = 0x2493 |
| ; AVX2-NEXT: shrl $16, %eax |
| ; AVX2-NEXT: movl %esi, %ecx |
| ; AVX2-NEXT: subl %eax, %ecx |
| ; AVX2-NEXT: movzwl %cx, %ecx |
| ; AVX2-NEXT: shrl %ecx |
| ; AVX2-NEXT: addl %eax, %ecx |
| ; AVX2-NEXT: shrl $2, %ecx |
| ; AVX2-NEXT: leal (,%rcx,8), %eax |
| ; AVX2-NEXT: subl %eax, %ecx |
| ; AVX2-NEXT: addl %esi, %ecx |
| ; AVX2-NEXT: andl $2047, %edi # imm = 0x7FF |
| ; AVX2-NEXT: imull $43691, %edi, %eax # imm = 0xAAAB |
| ; AVX2-NEXT: shrl $17, %eax |
| ; AVX2-NEXT: andl $-2, %eax |
| ; AVX2-NEXT: leal (%rax,%rax,2), %eax |
| ; AVX2-NEXT: subl %eax, %edi |
| ; AVX2-NEXT: vmovd %edi, %xmm0 |
| ; AVX2-NEXT: vpinsrw $2, %ecx, %xmm0, %xmm0 |
| ; AVX2-NEXT: andl $2047, %edx # imm = 0x7FF |
| ; AVX2-NEXT: imull $161, %edx, %eax |
| ; AVX2-NEXT: shrl $16, %eax |
| ; AVX2-NEXT: movl %edx, %ecx |
| ; AVX2-NEXT: subl %eax, %ecx |
| ; AVX2-NEXT: movzwl %cx, %ecx |
| ; AVX2-NEXT: shrl %ecx |
| ; AVX2-NEXT: addl %eax, %ecx |
| ; AVX2-NEXT: shrl $10, %ecx |
| ; AVX2-NEXT: imull $2043, %ecx, %eax # imm = 0x7FB |
| ; AVX2-NEXT: subl %eax, %edx |
| ; AVX2-NEXT: vpinsrw $4, %edx, %xmm0, %xmm0 |
| ; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2047,2047,2047,2047] |
| ; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0 |
| ; AVX2-NEXT: vpcmpeqd {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1 |
| ; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0 |
| ; AVX2-NEXT: vmovd %xmm0, %eax |
| ; AVX2-NEXT: vpextrb $4, %xmm0, %edx |
| ; AVX2-NEXT: vpextrb $8, %xmm0, %ecx |
| ; AVX2-NEXT: # kill: def $al killed $al killed $eax |
| ; AVX2-NEXT: # kill: def $dl killed $dl killed $edx |
| ; AVX2-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; AVX2-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: test_urem_vec: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: andl $2047, %esi # imm = 0x7FF |
| ; AVX512VL-NEXT: imull $9363, %esi, %eax # imm = 0x2493 |
| ; AVX512VL-NEXT: shrl $16, %eax |
| ; AVX512VL-NEXT: movl %esi, %ecx |
| ; AVX512VL-NEXT: subl %eax, %ecx |
| ; AVX512VL-NEXT: movzwl %cx, %ecx |
| ; AVX512VL-NEXT: shrl %ecx |
| ; AVX512VL-NEXT: addl %eax, %ecx |
| ; AVX512VL-NEXT: shrl $2, %ecx |
| ; AVX512VL-NEXT: leal (,%rcx,8), %eax |
| ; AVX512VL-NEXT: subl %eax, %ecx |
| ; AVX512VL-NEXT: addl %esi, %ecx |
| ; AVX512VL-NEXT: andl $2047, %edi # imm = 0x7FF |
| ; AVX512VL-NEXT: imull $43691, %edi, %eax # imm = 0xAAAB |
| ; AVX512VL-NEXT: shrl $17, %eax |
| ; AVX512VL-NEXT: andl $-2, %eax |
| ; AVX512VL-NEXT: leal (%rax,%rax,2), %eax |
| ; AVX512VL-NEXT: subl %eax, %edi |
| ; AVX512VL-NEXT: vmovd %edi, %xmm0 |
| ; AVX512VL-NEXT: vpinsrw $2, %ecx, %xmm0, %xmm0 |
| ; AVX512VL-NEXT: andl $2047, %edx # imm = 0x7FF |
| ; AVX512VL-NEXT: imull $161, %edx, %eax |
| ; AVX512VL-NEXT: shrl $16, %eax |
| ; AVX512VL-NEXT: movl %edx, %ecx |
| ; AVX512VL-NEXT: subl %eax, %ecx |
| ; AVX512VL-NEXT: movzwl %cx, %ecx |
| ; AVX512VL-NEXT: shrl %ecx |
| ; AVX512VL-NEXT: addl %eax, %ecx |
| ; AVX512VL-NEXT: shrl $10, %ecx |
| ; AVX512VL-NEXT: imull $2043, %ecx, %eax # imm = 0x7FB |
| ; AVX512VL-NEXT: subl %eax, %edx |
| ; AVX512VL-NEXT: vpinsrw $4, %edx, %xmm0, %xmm0 |
| ; AVX512VL-NEXT: vpandd {{.*}}(%rip){1to4}, %xmm0, %xmm0 |
| ; AVX512VL-NEXT: vpcmpneqd {{.*}}(%rip), %xmm0, %k0 |
| ; AVX512VL-NEXT: kshiftrw $1, %k0, %k1 |
| ; AVX512VL-NEXT: kmovw %k1, %edx |
| ; AVX512VL-NEXT: kshiftrw $2, %k0, %k1 |
| ; AVX512VL-NEXT: kmovw %k1, %ecx |
| ; AVX512VL-NEXT: kmovw %k0, %eax |
| ; AVX512VL-NEXT: # kill: def $al killed $al killed $eax |
| ; AVX512VL-NEXT: # kill: def $dl killed $dl killed $edx |
| ; AVX512VL-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; AVX512VL-NEXT: retq |
| %urem = urem <3 x i11> %X, <i11 6, i11 7, i11 -5> |
| %cmp = icmp ne <3 x i11> %urem, <i11 0, i11 1, i11 2> |
| ret <3 x i1> %cmp |
| } |
