| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=i686-unknown | FileCheck %s --check-prefix=X86-I32 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=i686-unknown -mattr=-x87 | FileCheck %s --check-prefix=X86-NOX87 | |
| ; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=i686-unknown | FileCheck %s --check-prefix=X86-I64 | |
| ; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=i686-unknown -mattr=-x87 | FileCheck %s --check-prefix=X86-I64-NOX87 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=i686-unknown -mattr=sse2 | FileCheck %s --check-prefix=X86-SSE2 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=i686-unknown -mattr=avx | FileCheck %s --check-prefixes=X86-AVX,X86-AVX1 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=i686-unknown -mattr=avx512f | FileCheck %s --check-prefixes=X86-AVX,AVX512-i32 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=i686-unknown -mattr=avx512dq,avx512vl | FileCheck %s --check-prefixes=X86-AVX,AVX512-i32 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=x86_64-unknown -mattr=avx | FileCheck %s --check-prefixes=X64-AVX-i32,X64-AVX1-i32 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=x86_64-unknown -mattr=avx512f | FileCheck %s --check-prefixes=X64-AVX-i32,AVX512-i32 | |
| ; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=x86_64-unknown -mattr=avx512dq,avx512vl | FileCheck %s --check-prefixes=X64-AVX-i32,AVX512-i32 | |
| ; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=x86_64-unknown -mattr=avx | FileCheck %s --check-prefixes=X64-AVX-i64,X64-AVX1-i64 | |
| ; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=x86_64-unknown -mattr=avx512f | FileCheck %s --check-prefixes=X64-AVX-i64,AVX512-i64 | |
| ; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=x86_64-unknown -mattr=avx512dq,avx512vl | FileCheck %s --check-prefixes=X64-AVX-i64,AVX512DQ-i64 | |
| define <1 x iXLen> @lrint_v1f32(<1 x float> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v1f32: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: pushl %eax | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl (%esp) | |
| ; X86-I32-NEXT: movl (%esp), %eax | |
| ; X86-I32-NEXT: popl %ecx | |
| ; X86-I32-NEXT: retl | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v1f32: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: retl | |
| ; | |
| ; X86-I64-LABEL: lrint_v1f32: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $8, %esp | |
| ; X86-I64-NEXT: flds 8(%ebp) | |
| ; X86-I64-NEXT: fistpll (%esp) | |
| ; X86-I64-NEXT: movl (%esp), %eax | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl %ebp, %esp | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v1f32: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: retl | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v1f32: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: cvtss2si {{[0-9]+}}(%esp), %eax | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX-LABEL: lrint_v1f32: | |
| ; X86-AVX: # %bb.0: | |
| ; X86-AVX-NEXT: vcvtss2si {{[0-9]+}}(%esp), %eax | |
| ; X86-AVX-NEXT: retl | |
| ; | |
| ; X64-AVX-i32-LABEL: lrint_v1f32: | |
| ; X64-AVX-i32: # %bb.0: | |
| ; X64-AVX-i32-NEXT: vcvtss2si %xmm0, %eax | |
| ; X64-AVX-i32-NEXT: retq | |
| ; | |
| ; X64-AVX-i64-LABEL: lrint_v1f32: | |
| ; X64-AVX-i64: # %bb.0: | |
| ; X64-AVX-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX-i64-NEXT: retq | |
| %a = call <1 x iXLen> @llvm.lrint.v1iXLen.v1f32(<1 x float> %x) | |
| ret <1 x iXLen> %a | |
| } | |
| declare <1 x iXLen> @llvm.lrint.v1iXLen.v1f32(<1 x float>) | |
| define <2 x iXLen> @lrint_v2f32(<2 x float> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v2f32: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: subl $8, %esp | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl (%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: movl (%esp), %eax | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I32-NEXT: addl $8, %esp | |
| ; X86-I32-NEXT: retl | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v2f32: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: pushl %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %esi | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %edx | |
| ; X86-NOX87-NEXT: movl %esi, %eax | |
| ; X86-NOX87-NEXT: popl %esi | |
| ; X86-NOX87-NEXT: popl %edi | |
| ; X86-NOX87-NEXT: retl | |
| ; | |
| ; X86-I64-LABEL: lrint_v2f32: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: pushl %edi | |
| ; X86-I64-NEXT: pushl %esi | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $16, %esp | |
| ; X86-I64-NEXT: movl 8(%ebp), %eax | |
| ; X86-I64-NEXT: flds 16(%ebp) | |
| ; X86-I64-NEXT: flds 12(%ebp) | |
| ; X86-I64-NEXT: fistpll (%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: movl (%esp), %ecx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NEXT: movl %edi, 12(%eax) | |
| ; X86-I64-NEXT: movl %esi, 8(%eax) | |
| ; X86-I64-NEXT: movl %edx, 4(%eax) | |
| ; X86-I64-NEXT: movl %ecx, (%eax) | |
| ; X86-I64-NEXT: leal -8(%ebp), %esp | |
| ; X86-I64-NEXT: popl %esi | |
| ; X86-I64-NEXT: popl %edi | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl $4 | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v2f32: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: pushl %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, %edi | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %edx, 12(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebx, 4(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %edi, (%esi) | |
| ; X86-I64-NOX87-NEXT: movl %esi, %eax | |
| ; X86-I64-NOX87-NEXT: popl %esi | |
| ; X86-I64-NOX87-NEXT: popl %edi | |
| ; X86-I64-NOX87-NEXT: popl %ebx | |
| ; X86-I64-NOX87-NEXT: popl %ebp | |
| ; X86-I64-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v2f32: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: cvtps2dq %xmm0, %xmm0 | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX-LABEL: lrint_v2f32: | |
| ; X86-AVX: # %bb.0: | |
| ; X86-AVX-NEXT: vcvtps2dq %xmm0, %xmm0 | |
| ; X86-AVX-NEXT: retl | |
| ; | |
| ; X64-AVX-i32-LABEL: lrint_v2f32: | |
| ; X64-AVX-i32: # %bb.0: | |
| ; X64-AVX-i32-NEXT: vcvtps2dq %xmm0, %xmm0 | |
| ; X64-AVX-i32-NEXT: retq | |
| ; | |
| ; X64-AVX1-i64-LABEL: lrint_v2f32: | |
| ; X64-AVX1-i64: # %bb.0: | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm0 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] | |
| ; X64-AVX1-i64-NEXT: retq | |
| ; | |
| ; AVX512-i64-LABEL: lrint_v2f32: | |
| ; AVX512-i64: # %bb.0: | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm1 | |
| ; AVX512-i64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm0 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] | |
| ; AVX512-i64-NEXT: retq | |
| ; | |
| ; AVX512DQ-i64-LABEL: lrint_v2f32: | |
| ; AVX512DQ-i64: # %bb.0: | |
| ; AVX512DQ-i64-NEXT: vcvtps2qq %xmm0, %xmm0 | |
| ; AVX512DQ-i64-NEXT: retq | |
| %a = call <2 x iXLen> @llvm.lrint.v2iXLen.v2f32(<2 x float> %x) | |
| ret <2 x iXLen> %a | |
| } | |
| declare <2 x iXLen> @llvm.lrint.v2iXLen.v2f32(<2 x float>) | |
| define <4 x iXLen> @lrint_v4f32(<4 x float> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v4f32: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: pushl %edi | |
| ; X86-I32-NEXT: pushl %esi | |
| ; X86-I32-NEXT: subl $16, %esp | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %eax | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl (%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: movl (%esp), %ecx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I32-NEXT: movl %edi, 12(%eax) | |
| ; X86-I32-NEXT: movl %esi, 8(%eax) | |
| ; X86-I32-NEXT: movl %edx, 4(%eax) | |
| ; X86-I32-NEXT: movl %ecx, (%eax) | |
| ; X86-I32-NEXT: addl $16, %esp | |
| ; X86-I32-NEXT: popl %esi | |
| ; X86-I32-NEXT: popl %edi | |
| ; X86-I32-NEXT: retl $4 | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v4f32: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: pushl %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %edi | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebx | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebp | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, 12(%esi) | |
| ; X86-NOX87-NEXT: movl %ebp, 8(%esi) | |
| ; X86-NOX87-NEXT: movl %ebx, 4(%esi) | |
| ; X86-NOX87-NEXT: movl %edi, (%esi) | |
| ; X86-NOX87-NEXT: movl %esi, %eax | |
| ; X86-NOX87-NEXT: popl %esi | |
| ; X86-NOX87-NEXT: popl %edi | |
| ; X86-NOX87-NEXT: popl %ebx | |
| ; X86-NOX87-NEXT: popl %ebp | |
| ; X86-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-I64-LABEL: lrint_v4f32: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: pushl %ebx | |
| ; X86-I64-NEXT: pushl %edi | |
| ; X86-I64-NEXT: pushl %esi | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $48, %esp | |
| ; X86-I64-NEXT: movl 8(%ebp), %eax | |
| ; X86-I64-NEXT: flds 24(%ebp) | |
| ; X86-I64-NEXT: flds 20(%ebp) | |
| ; X86-I64-NEXT: flds 16(%ebp) | |
| ; X86-I64-NEXT: flds 12(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NEXT: movl %esi, 28(%eax) | |
| ; X86-I64-NEXT: movl %ecx, 24(%eax) | |
| ; X86-I64-NEXT: movl %edx, 20(%eax) | |
| ; X86-I64-NEXT: movl %ebx, 16(%eax) | |
| ; X86-I64-NEXT: movl %edi, 12(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 8(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 4(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, (%eax) | |
| ; X86-I64-NEXT: leal -12(%ebp), %esp | |
| ; X86-I64-NEXT: popl %esi | |
| ; X86-I64-NEXT: popl %edi | |
| ; X86-I64-NEXT: popl %ebx | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl $4 | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v4f32: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: pushl %esi | |
| ; X86-I64-NOX87-NEXT: subl $12, %esp | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esp) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, %edi | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, %ebx | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %edx, 28(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %eax, 24(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebp, 20(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebx, 16(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %edi, 12(%esi) | |
| ; X86-I64-NOX87-NEXT: movl (%esp), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 4(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esi) | |
| ; X86-I64-NOX87-NEXT: movl %esi, %eax | |
| ; X86-I64-NOX87-NEXT: addl $12, %esp | |
| ; X86-I64-NOX87-NEXT: popl %esi | |
| ; X86-I64-NOX87-NEXT: popl %edi | |
| ; X86-I64-NOX87-NEXT: popl %ebx | |
| ; X86-I64-NOX87-NEXT: popl %ebp | |
| ; X86-I64-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v4f32: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: cvtps2dq %xmm0, %xmm0 | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX-LABEL: lrint_v4f32: | |
| ; X86-AVX: # %bb.0: | |
| ; X86-AVX-NEXT: vcvtps2dq %xmm0, %xmm0 | |
| ; X86-AVX-NEXT: retl | |
| ; | |
| ; X64-AVX-i32-LABEL: lrint_v4f32: | |
| ; X64-AVX-i32: # %bb.0: | |
| ; X64-AVX-i32-NEXT: vcvtps2dq %xmm0, %xmm0 | |
| ; X64-AVX-i32-NEXT: retq | |
| ; | |
| ; X64-AVX1-i64-LABEL: lrint_v4f32: | |
| ; X64-AVX1-i64: # %bb.0: | |
| ; X64-AVX1-i64-NEXT: vshufps {{.*#+}} xmm1 = xmm0[3,3,3,3] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm1, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm2 = xmm0[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm2, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm0 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] | |
| ; X64-AVX1-i64-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 | |
| ; X64-AVX1-i64-NEXT: retq | |
| ; | |
| ; AVX512-i64-LABEL: lrint_v4f32: | |
| ; AVX512-i64: # %bb.0: | |
| ; AVX512-i64-NEXT: vshufps {{.*#+}} xmm1 = xmm0[3,3,3,3] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm1, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm1 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm2 = xmm0[1,0] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm2, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm0 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] | |
| ; AVX512-i64-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 | |
| ; AVX512-i64-NEXT: retq | |
| ; | |
| ; AVX512DQ-i64-LABEL: lrint_v4f32: | |
| ; AVX512DQ-i64: # %bb.0: | |
| ; AVX512DQ-i64-NEXT: vcvtps2qq %xmm0, %ymm0 | |
| ; AVX512DQ-i64-NEXT: retq | |
| %a = call <4 x iXLen> @llvm.lrint.v4iXLen.v4f32(<4 x float> %x) | |
| ret <4 x iXLen> %a | |
| } | |
| declare <4 x iXLen> @llvm.lrint.v4iXLen.v4f32(<4 x float>) | |
| define <8 x iXLen> @lrint_v8f32(<8 x float> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v8f32: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: pushl %ebp | |
| ; X86-I32-NEXT: pushl %ebx | |
| ; X86-I32-NEXT: pushl %edi | |
| ; X86-I32-NEXT: pushl %esi | |
| ; X86-I32-NEXT: subl $40, %esp | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: flds {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %eax | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I32-NEXT: movl %ecx, (%esp) # 4-byte Spill | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I32-NEXT: movl %edx, 28(%eax) | |
| ; X86-I32-NEXT: movl %ecx, 24(%eax) | |
| ; X86-I32-NEXT: movl %ebp, 20(%eax) | |
| ; X86-I32-NEXT: movl %ebx, 16(%eax) | |
| ; X86-I32-NEXT: movl %edi, 12(%eax) | |
| ; X86-I32-NEXT: movl %esi, 8(%eax) | |
| ; X86-I32-NEXT: movl (%esp), %ecx # 4-byte Reload | |
| ; X86-I32-NEXT: movl %ecx, 4(%eax) | |
| ; X86-I32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I32-NEXT: movl %ecx, (%eax) | |
| ; X86-I32-NEXT: addl $40, %esp | |
| ; X86-I32-NEXT: popl %esi | |
| ; X86-I32-NEXT: popl %edi | |
| ; X86-I32-NEXT: popl %ebx | |
| ; X86-I32-NEXT: popl %ebp | |
| ; X86-I32-NEXT: retl $4 | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v8f32: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: pushl %esi | |
| ; X86-NOX87-NEXT: subl $16, %esp | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, (%esp) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebx | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebp | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %edi | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrintf | |
| ; X86-NOX87-NEXT: addl $4, %esp | |
| ; X86-NOX87-NEXT: movl %eax, 28(%esi) | |
| ; X86-NOX87-NEXT: movl %edi, 24(%esi) | |
| ; X86-NOX87-NEXT: movl %ebp, 20(%esi) | |
| ; X86-NOX87-NEXT: movl %ebx, 16(%esi) | |
| ; X86-NOX87-NEXT: movl (%esp), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, 12(%esi) | |
| ; X86-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, 4(%esi) | |
| ; X86-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, (%esi) | |
| ; X86-NOX87-NEXT: movl %esi, %eax | |
| ; X86-NOX87-NEXT: addl $16, %esp | |
| ; X86-NOX87-NEXT: popl %esi | |
| ; X86-NOX87-NEXT: popl %edi | |
| ; X86-NOX87-NEXT: popl %ebx | |
| ; X86-NOX87-NEXT: popl %ebp | |
| ; X86-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-I64-LABEL: lrint_v8f32: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: pushl %ebx | |
| ; X86-I64-NEXT: pushl %edi | |
| ; X86-I64-NEXT: pushl %esi | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $112, %esp | |
| ; X86-I64-NEXT: flds 12(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: flds 16(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: flds 20(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: flds 24(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: flds 28(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: flds 32(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: flds 36(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: flds 40(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: movl 8(%ebp), %eax | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NEXT: movl %ebx, 60(%eax) | |
| ; X86-I64-NEXT: movl %ecx, 56(%eax) | |
| ; X86-I64-NEXT: movl %edx, 52(%eax) | |
| ; X86-I64-NEXT: movl %esi, 48(%eax) | |
| ; X86-I64-NEXT: movl %edi, 44(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 40(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 36(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 32(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 28(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 24(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 20(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 16(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 12(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 8(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 4(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, (%eax) | |
| ; X86-I64-NEXT: leal -12(%ebp), %esp | |
| ; X86-I64-NEXT: popl %esi | |
| ; X86-I64-NEXT: popl %edi | |
| ; X86-I64-NEXT: popl %ebx | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl $4 | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v8f32: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: pushl %esi | |
| ; X86-I64-NOX87-NEXT: subl $44, %esp | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esp) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, %edi | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebx | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrintf | |
| ; X86-I64-NOX87-NEXT: addl $4, %esp | |
| ; X86-I64-NOX87-NEXT: movl %edx, 60(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %eax, 56(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebx, 52(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %edi, 48(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebp, 44(%esi) | |
| ; X86-I64-NOX87-NEXT: movl (%esp), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 40(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 36(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 32(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 28(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 24(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 20(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 16(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 12(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 4(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esi) | |
| ; X86-I64-NOX87-NEXT: movl %esi, %eax | |
| ; X86-I64-NOX87-NEXT: addl $44, %esp | |
| ; X86-I64-NOX87-NEXT: popl %esi | |
| ; X86-I64-NOX87-NEXT: popl %edi | |
| ; X86-I64-NOX87-NEXT: popl %ebx | |
| ; X86-I64-NOX87-NEXT: popl %ebp | |
| ; X86-I64-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v8f32: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: cvtps2dq %xmm0, %xmm0 | |
| ; X86-SSE2-NEXT: cvtps2dq %xmm1, %xmm1 | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX-LABEL: lrint_v8f32: | |
| ; X86-AVX: # %bb.0: | |
| ; X86-AVX-NEXT: vcvtps2dq %ymm0, %ymm0 | |
| ; X86-AVX-NEXT: retl | |
| ; | |
| ; X64-AVX-i32-LABEL: lrint_v8f32: | |
| ; X64-AVX-i32: # %bb.0: | |
| ; X64-AVX-i32-NEXT: vcvtps2dq %ymm0, %ymm0 | |
| ; X64-AVX-i32-NEXT: retq | |
| ; | |
| ; X64-AVX1-i64-LABEL: lrint_v8f32: | |
| ; X64-AVX1-i64: # %bb.0: | |
| ; X64-AVX1-i64-NEXT: vshufps {{.*#+}} xmm1 = xmm0[3,3,3,3] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm1, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm2 = xmm0[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm2, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vmovshdup {{.*#+}} xmm3 = xmm0[1,1,3,3] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm3, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm3 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0] | |
| ; X64-AVX1-i64-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm2 | |
| ; X64-AVX1-i64-NEXT: vextractf128 $1, %ymm0, %xmm0 | |
| ; X64-AVX1-i64-NEXT: vshufps {{.*#+}} xmm1 = xmm0[3,3,3,3] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm1, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm3 = xmm0[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm3, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm3 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm3[0],xmm1[0] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm3 | |
| ; X64-AVX1-i64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] | |
| ; X64-AVX1-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm0 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0] | |
| ; X64-AVX1-i64-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm1 | |
| ; X64-AVX1-i64-NEXT: vmovaps %ymm2, %ymm0 | |
| ; X64-AVX1-i64-NEXT: retq | |
| ; | |
| ; AVX512-i64-LABEL: lrint_v8f32: | |
| ; AVX512-i64: # %bb.0: | |
| ; AVX512-i64-NEXT: vextractf128 $1, %ymm0, %xmm1 | |
| ; AVX512-i64-NEXT: vshufps {{.*#+}} xmm2 = xmm1[3,3,3,3] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm2, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm3 = xmm1[1,0] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm3, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm3 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm1, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm3 | |
| ; AVX512-i64-NEXT: vmovshdup {{.*#+}} xmm1 = xmm1[1,1,3,3] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm1, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm1 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm3[0],xmm1[0] | |
| ; AVX512-i64-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1 | |
| ; AVX512-i64-NEXT: vshufps {{.*#+}} xmm2 = xmm0[3,3,3,3] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm2, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm3 = xmm0[1,0] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm3, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm3 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm3 | |
| ; AVX512-i64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] | |
| ; AVX512-i64-NEXT: vcvtss2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm0 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0] | |
| ; AVX512-i64-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0 | |
| ; AVX512-i64-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 | |
| ; AVX512-i64-NEXT: retq | |
| ; | |
| ; AVX512DQ-i64-LABEL: lrint_v8f32: | |
| ; AVX512DQ-i64: # %bb.0: | |
| ; AVX512DQ-i64-NEXT: vcvtps2qq %ymm0, %zmm0 | |
| ; AVX512DQ-i64-NEXT: retq | |
| %a = call <8 x iXLen> @llvm.lrint.v8iXLen.v8f32(<8 x float> %x) | |
| ret <8 x iXLen> %a | |
| } | |
| declare <8 x iXLen> @llvm.lrint.v8iXLen.v8f32(<8 x float>) | |
| define <16 x iXLen> @lrint_v16iXLen_v16f32(<16 x float> %x) nounwind { | |
| %a = call <16 x iXLen> @llvm.lrint.v16iXLen.v16f32(<16 x float> %x) | |
| ret <16 x iXLen> %a | |
| } | |
| declare <16 x iXLen> @llvm.lrint.v16iXLen.v16f32(<16 x float>) | |
| define <1 x iXLen> @lrint_v1f64(<1 x double> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v1f64: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: pushl %eax | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl (%esp) | |
| ; X86-I32-NEXT: movl (%esp), %eax | |
| ; X86-I32-NEXT: popl %ecx | |
| ; X86-I32-NEXT: retl | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v1f64: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: retl | |
| ; | |
| ; X86-I64-LABEL: lrint_v1f64: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $8, %esp | |
| ; X86-I64-NEXT: fldl 8(%ebp) | |
| ; X86-I64-NEXT: fistpll (%esp) | |
| ; X86-I64-NEXT: movl (%esp), %eax | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl %ebp, %esp | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v1f64: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: retl | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v1f64: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: cvtsd2si {{[0-9]+}}(%esp), %eax | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX-LABEL: lrint_v1f64: | |
| ; X86-AVX: # %bb.0: | |
| ; X86-AVX-NEXT: vcvtsd2si {{[0-9]+}}(%esp), %eax | |
| ; X86-AVX-NEXT: retl | |
| ; | |
| ; X64-AVX-i32-LABEL: lrint_v1f64: | |
| ; X64-AVX-i32: # %bb.0: | |
| ; X64-AVX-i32-NEXT: vcvtsd2si %xmm0, %eax | |
| ; X64-AVX-i32-NEXT: retq | |
| ; | |
| ; X64-AVX-i64-LABEL: lrint_v1f64: | |
| ; X64-AVX-i64: # %bb.0: | |
| ; X64-AVX-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; X64-AVX-i64-NEXT: retq | |
| %a = call <1 x iXLen> @llvm.lrint.v1iXLen.v1f64(<1 x double> %x) | |
| ret <1 x iXLen> %a | |
| } | |
| declare <1 x iXLen> @llvm.lrint.v1iXLen.v1f64(<1 x double>) | |
| define <2 x iXLen> @lrint_v2f64(<2 x double> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v2f64: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: subl $8, %esp | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl (%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: movl (%esp), %eax | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I32-NEXT: addl $8, %esp | |
| ; X86-I32-NEXT: retl | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v2f64: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: pushl %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %esi | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %edx | |
| ; X86-NOX87-NEXT: movl %esi, %eax | |
| ; X86-NOX87-NEXT: popl %esi | |
| ; X86-NOX87-NEXT: popl %edi | |
| ; X86-NOX87-NEXT: popl %ebx | |
| ; X86-NOX87-NEXT: retl | |
| ; | |
| ; X86-I64-LABEL: lrint_v2f64: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: pushl %edi | |
| ; X86-I64-NEXT: pushl %esi | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $16, %esp | |
| ; X86-I64-NEXT: movl 8(%ebp), %eax | |
| ; X86-I64-NEXT: fldl 20(%ebp) | |
| ; X86-I64-NEXT: fldl 12(%ebp) | |
| ; X86-I64-NEXT: fistpll (%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: movl (%esp), %ecx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NEXT: movl %edi, 12(%eax) | |
| ; X86-I64-NEXT: movl %esi, 8(%eax) | |
| ; X86-I64-NEXT: movl %edx, 4(%eax) | |
| ; X86-I64-NEXT: movl %ecx, (%eax) | |
| ; X86-I64-NEXT: leal -8(%ebp), %esp | |
| ; X86-I64-NEXT: popl %esi | |
| ; X86-I64-NEXT: popl %edi | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl $4 | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v2f64: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: pushl %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, %edi | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %edx, 12(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebx, 4(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %edi, (%esi) | |
| ; X86-I64-NOX87-NEXT: movl %esi, %eax | |
| ; X86-I64-NOX87-NEXT: popl %esi | |
| ; X86-I64-NOX87-NEXT: popl %edi | |
| ; X86-I64-NOX87-NEXT: popl %ebx | |
| ; X86-I64-NOX87-NEXT: popl %ebp | |
| ; X86-I64-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v2f64: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: cvtpd2dq %xmm0, %xmm0 | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX-LABEL: lrint_v2f64: | |
| ; X86-AVX: # %bb.0: | |
| ; X86-AVX-NEXT: vcvtpd2dq %xmm0, %xmm0 | |
| ; X86-AVX-NEXT: retl | |
| ; | |
| ; X64-AVX-i32-LABEL: lrint_v2f64: | |
| ; X64-AVX-i32: # %bb.0: | |
| ; X64-AVX-i32-NEXT: vcvtpd2dq %xmm0, %xmm0 | |
| ; X64-AVX-i32-NEXT: retq | |
| ; | |
| ; X64-AVX1-i64-LABEL: lrint_v2f64: | |
| ; X64-AVX1-i64: # %bb.0: | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm0 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] | |
| ; X64-AVX1-i64-NEXT: retq | |
| ; | |
| ; AVX512-i64-LABEL: lrint_v2f64: | |
| ; AVX512-i64: # %bb.0: | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm1 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm0 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] | |
| ; AVX512-i64-NEXT: retq | |
| ; | |
| ; AVX512DQ-i64-LABEL: lrint_v2f64: | |
| ; AVX512DQ-i64: # %bb.0: | |
| ; AVX512DQ-i64-NEXT: vcvtpd2qq %xmm0, %xmm0 | |
| ; AVX512DQ-i64-NEXT: retq | |
| %a = call <2 x iXLen> @llvm.lrint.v2iXLen.v2f64(<2 x double> %x) | |
| ret <2 x iXLen> %a | |
| } | |
| declare <2 x iXLen> @llvm.lrint.v2iXLen.v2f64(<2 x double>) | |
| define <4 x iXLen> @lrint_v4f64(<4 x double> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v4f64: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: pushl %edi | |
| ; X86-I32-NEXT: pushl %esi | |
| ; X86-I32-NEXT: subl $16, %esp | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %eax | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl (%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: movl (%esp), %ecx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I32-NEXT: movl %edi, 12(%eax) | |
| ; X86-I32-NEXT: movl %esi, 8(%eax) | |
| ; X86-I32-NEXT: movl %edx, 4(%eax) | |
| ; X86-I32-NEXT: movl %ecx, (%eax) | |
| ; X86-I32-NEXT: addl $16, %esp | |
| ; X86-I32-NEXT: popl %esi | |
| ; X86-I32-NEXT: popl %edi | |
| ; X86-I32-NEXT: retl $4 | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v4f64: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: pushl %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %edi | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebx | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebp | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, 12(%esi) | |
| ; X86-NOX87-NEXT: movl %ebp, 8(%esi) | |
| ; X86-NOX87-NEXT: movl %ebx, 4(%esi) | |
| ; X86-NOX87-NEXT: movl %edi, (%esi) | |
| ; X86-NOX87-NEXT: movl %esi, %eax | |
| ; X86-NOX87-NEXT: popl %esi | |
| ; X86-NOX87-NEXT: popl %edi | |
| ; X86-NOX87-NEXT: popl %ebx | |
| ; X86-NOX87-NEXT: popl %ebp | |
| ; X86-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-I64-LABEL: lrint_v4f64: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: pushl %ebx | |
| ; X86-I64-NEXT: pushl %edi | |
| ; X86-I64-NEXT: pushl %esi | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $48, %esp | |
| ; X86-I64-NEXT: movl 8(%ebp), %eax | |
| ; X86-I64-NEXT: fldl 36(%ebp) | |
| ; X86-I64-NEXT: fldl 28(%ebp) | |
| ; X86-I64-NEXT: fldl 20(%ebp) | |
| ; X86-I64-NEXT: fldl 12(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NEXT: movl %esi, 28(%eax) | |
| ; X86-I64-NEXT: movl %ecx, 24(%eax) | |
| ; X86-I64-NEXT: movl %edx, 20(%eax) | |
| ; X86-I64-NEXT: movl %ebx, 16(%eax) | |
| ; X86-I64-NEXT: movl %edi, 12(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 8(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 4(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, (%eax) | |
| ; X86-I64-NEXT: leal -12(%ebp), %esp | |
| ; X86-I64-NEXT: popl %esi | |
| ; X86-I64-NEXT: popl %edi | |
| ; X86-I64-NEXT: popl %ebx | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl $4 | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v4f64: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: pushl %esi | |
| ; X86-I64-NOX87-NEXT: subl $12, %esp | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esp) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, %edi | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, %ebx | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %edx, 28(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %eax, 24(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebp, 20(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebx, 16(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %edi, 12(%esi) | |
| ; X86-I64-NOX87-NEXT: movl (%esp), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 4(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esi) | |
| ; X86-I64-NOX87-NEXT: movl %esi, %eax | |
| ; X86-I64-NOX87-NEXT: addl $12, %esp | |
| ; X86-I64-NOX87-NEXT: popl %esi | |
| ; X86-I64-NOX87-NEXT: popl %edi | |
| ; X86-I64-NOX87-NEXT: popl %ebx | |
| ; X86-I64-NOX87-NEXT: popl %ebp | |
| ; X86-I64-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v4f64: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: cvtpd2dq %xmm1, %xmm1 | |
| ; X86-SSE2-NEXT: cvtpd2dq %xmm0, %xmm0 | |
| ; X86-SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX-LABEL: lrint_v4f64: | |
| ; X86-AVX: # %bb.0: | |
| ; X86-AVX-NEXT: vcvtpd2dq %ymm0, %xmm0 | |
| ; X86-AVX-NEXT: vzeroupper | |
| ; X86-AVX-NEXT: retl | |
| ; | |
| ; X64-AVX-i32-LABEL: lrint_v4f64: | |
| ; X64-AVX-i32: # %bb.0: | |
| ; X64-AVX-i32-NEXT: vcvtpd2dq %ymm0, %xmm0 | |
| ; X64-AVX-i32-NEXT: vzeroupper | |
| ; X64-AVX-i32-NEXT: retq | |
| ; | |
| ; X64-AVX1-i64-LABEL: lrint_v4f64: | |
| ; X64-AVX1-i64: # %bb.0: | |
| ; X64-AVX1-i64-NEXT: vextractf128 $1, %ymm0, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm1 = xmm1[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm0 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] | |
| ; X64-AVX1-i64-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 | |
| ; X64-AVX1-i64-NEXT: retq | |
| ; | |
| ; AVX512-i64-LABEL: lrint_v4f64: | |
| ; AVX512-i64: # %bb.0: | |
| ; AVX512-i64-NEXT: vextractf128 $1, %ymm0, %xmm1 | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm1 = xmm1[1,0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm1 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm0 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] | |
| ; AVX512-i64-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 | |
| ; AVX512-i64-NEXT: retq | |
| ; | |
| ; AVX512DQ-i64-LABEL: lrint_v4f64: | |
| ; AVX512DQ-i64: # %bb.0: | |
| ; AVX512DQ-i64-NEXT: vcvtpd2qq %ymm0, %ymm0 | |
| ; AVX512DQ-i64-NEXT: retq | |
| %a = call <4 x iXLen> @llvm.lrint.v4iXLen.v4f64(<4 x double> %x) | |
| ret <4 x iXLen> %a | |
| } | |
| declare <4 x iXLen> @llvm.lrint.v4iXLen.v4f64(<4 x double>) | |
| define <8 x iXLen> @lrint_v8f64(<8 x double> %x) nounwind { | |
| ; X86-I32-LABEL: lrint_v8f64: | |
| ; X86-I32: # %bb.0: | |
| ; X86-I32-NEXT: pushl %ebp | |
| ; X86-I32-NEXT: pushl %ebx | |
| ; X86-I32-NEXT: pushl %edi | |
| ; X86-I32-NEXT: pushl %esi | |
| ; X86-I32-NEXT: subl $40, %esp | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fldl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: fistpl {{[0-9]+}}(%esp) | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %eax | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I32-NEXT: movl %ecx, (%esp) # 4-byte Spill | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I32-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I32-NEXT: movl %edx, 28(%eax) | |
| ; X86-I32-NEXT: movl %ecx, 24(%eax) | |
| ; X86-I32-NEXT: movl %ebp, 20(%eax) | |
| ; X86-I32-NEXT: movl %ebx, 16(%eax) | |
| ; X86-I32-NEXT: movl %edi, 12(%eax) | |
| ; X86-I32-NEXT: movl %esi, 8(%eax) | |
| ; X86-I32-NEXT: movl (%esp), %ecx # 4-byte Reload | |
| ; X86-I32-NEXT: movl %ecx, 4(%eax) | |
| ; X86-I32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I32-NEXT: movl %ecx, (%eax) | |
| ; X86-I32-NEXT: addl $40, %esp | |
| ; X86-I32-NEXT: popl %esi | |
| ; X86-I32-NEXT: popl %edi | |
| ; X86-I32-NEXT: popl %ebx | |
| ; X86-I32-NEXT: popl %ebp | |
| ; X86-I32-NEXT: retl $4 | |
| ; | |
| ; X86-NOX87-LABEL: lrint_v8f64: | |
| ; X86-NOX87: # %bb.0: | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: pushl %esi | |
| ; X86-NOX87-NEXT: subl $16, %esp | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl %ebp | |
| ; X86-NOX87-NEXT: pushl %ebx | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl %edi | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, (%esp) # 4-byte Spill | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebx | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %ebp | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, %edi | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-NOX87-NEXT: calll lrint | |
| ; X86-NOX87-NEXT: addl $8, %esp | |
| ; X86-NOX87-NEXT: movl %eax, 28(%esi) | |
| ; X86-NOX87-NEXT: movl %edi, 24(%esi) | |
| ; X86-NOX87-NEXT: movl %ebp, 20(%esi) | |
| ; X86-NOX87-NEXT: movl %ebx, 16(%esi) | |
| ; X86-NOX87-NEXT: movl (%esp), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, 12(%esi) | |
| ; X86-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, 4(%esi) | |
| ; X86-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-NOX87-NEXT: movl %eax, (%esi) | |
| ; X86-NOX87-NEXT: movl %esi, %eax | |
| ; X86-NOX87-NEXT: addl $16, %esp | |
| ; X86-NOX87-NEXT: popl %esi | |
| ; X86-NOX87-NEXT: popl %edi | |
| ; X86-NOX87-NEXT: popl %ebx | |
| ; X86-NOX87-NEXT: popl %ebp | |
| ; X86-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-I64-LABEL: lrint_v8f64: | |
| ; X86-I64: # %bb.0: | |
| ; X86-I64-NEXT: pushl %ebp | |
| ; X86-I64-NEXT: movl %esp, %ebp | |
| ; X86-I64-NEXT: pushl %ebx | |
| ; X86-I64-NEXT: pushl %edi | |
| ; X86-I64-NEXT: pushl %esi | |
| ; X86-I64-NEXT: andl $-8, %esp | |
| ; X86-I64-NEXT: subl $112, %esp | |
| ; X86-I64-NEXT: fldl 12(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fldl 20(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fldl 28(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fldl 36(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fldl 44(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fldl 52(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fldl 60(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: fldl 68(%ebp) | |
| ; X86-I64-NEXT: fistpll {{[0-9]+}}(%esp) | |
| ; X86-I64-NEXT: movl 8(%ebp), %eax | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %edx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ecx | |
| ; X86-I64-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NEXT: movl %ebx, 60(%eax) | |
| ; X86-I64-NEXT: movl %ecx, 56(%eax) | |
| ; X86-I64-NEXT: movl %edx, 52(%eax) | |
| ; X86-I64-NEXT: movl %esi, 48(%eax) | |
| ; X86-I64-NEXT: movl %edi, 44(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 40(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 36(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 32(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 28(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 24(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 20(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 16(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 12(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 8(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, 4(%eax) | |
| ; X86-I64-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload | |
| ; X86-I64-NEXT: movl %ecx, (%eax) | |
| ; X86-I64-NEXT: leal -12(%ebp), %esp | |
| ; X86-I64-NEXT: popl %esi | |
| ; X86-I64-NEXT: popl %edi | |
| ; X86-I64-NEXT: popl %ebx | |
| ; X86-I64-NEXT: popl %ebp | |
| ; X86-I64-NEXT: retl $4 | |
| ; | |
| ; X86-I64-NOX87-LABEL: lrint_v8f64: | |
| ; X86-I64-NOX87: # %bb.0: | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: pushl %esi | |
| ; X86-I64-NOX87-NEXT: subl $44, %esp | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %esi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %edi | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebx | |
| ; X86-I64-NOX87-NEXT: movl {{[0-9]+}}(%esp), %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl %ebp | |
| ; X86-I64-NOX87-NEXT: pushl %ebx | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl %edi | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esp) # 4-byte Spill | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebp | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %eax, %edi | |
| ; X86-I64-NOX87-NEXT: movl %edx, %ebx | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: pushl {{[0-9]+}}(%esp) | |
| ; X86-I64-NOX87-NEXT: calll lrint | |
| ; X86-I64-NOX87-NEXT: addl $8, %esp | |
| ; X86-I64-NOX87-NEXT: movl %edx, 60(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %eax, 56(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebx, 52(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %edi, 48(%esi) | |
| ; X86-I64-NOX87-NEXT: movl %ebp, 44(%esi) | |
| ; X86-I64-NOX87-NEXT: movl (%esp), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 40(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 36(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 32(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 28(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 24(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 20(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 16(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 12(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 8(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, 4(%esi) | |
| ; X86-I64-NOX87-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload | |
| ; X86-I64-NOX87-NEXT: movl %eax, (%esi) | |
| ; X86-I64-NOX87-NEXT: movl %esi, %eax | |
| ; X86-I64-NOX87-NEXT: addl $44, %esp | |
| ; X86-I64-NOX87-NEXT: popl %esi | |
| ; X86-I64-NOX87-NEXT: popl %edi | |
| ; X86-I64-NOX87-NEXT: popl %ebx | |
| ; X86-I64-NOX87-NEXT: popl %ebp | |
| ; X86-I64-NOX87-NEXT: retl $4 | |
| ; | |
| ; X86-SSE2-LABEL: lrint_v8f64: | |
| ; X86-SSE2: # %bb.0: | |
| ; X86-SSE2-NEXT: pushl %ebp | |
| ; X86-SSE2-NEXT: movl %esp, %ebp | |
| ; X86-SSE2-NEXT: andl $-16, %esp | |
| ; X86-SSE2-NEXT: subl $16, %esp | |
| ; X86-SSE2-NEXT: cvtpd2dq %xmm1, %xmm1 | |
| ; X86-SSE2-NEXT: cvtpd2dq %xmm0, %xmm0 | |
| ; X86-SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] | |
| ; X86-SSE2-NEXT: cvtpd2dq %xmm2, %xmm1 | |
| ; X86-SSE2-NEXT: cvtpd2dq 8(%ebp), %xmm2 | |
| ; X86-SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm2[0] | |
| ; X86-SSE2-NEXT: movl %ebp, %esp | |
| ; X86-SSE2-NEXT: popl %ebp | |
| ; X86-SSE2-NEXT: retl | |
| ; | |
| ; X86-AVX1-LABEL: lrint_v8f64: | |
| ; X86-AVX1: # %bb.0: | |
| ; X86-AVX1-NEXT: vcvtpd2dq %ymm0, %xmm0 | |
| ; X86-AVX1-NEXT: vcvtpd2dq %ymm1, %xmm1 | |
| ; X86-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 | |
| ; X86-AVX1-NEXT: retl | |
| ; | |
| ; AVX512-i32-LABEL: lrint_v8f64: | |
| ; AVX512-i32: # %bb.0: | |
| ; AVX512-i32-NEXT: vcvtpd2dq %zmm0, %ymm0 | |
| ; AVX512-i32-NEXT: ret{{[l|q]}} | |
| ; | |
| ; X64-AVX1-i32-LABEL: lrint_v8f64: | |
| ; X64-AVX1-i32: # %bb.0: | |
| ; X64-AVX1-i32-NEXT: vcvtpd2dq %ymm0, %xmm0 | |
| ; X64-AVX1-i32-NEXT: vcvtpd2dq %ymm1, %xmm1 | |
| ; X64-AVX1-i32-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 | |
| ; X64-AVX1-i32-NEXT: retq | |
| ; | |
| ; X64-AVX1-i64-LABEL: lrint_v8f64: | |
| ; X64-AVX1-i64: # %bb.0: | |
| ; X64-AVX1-i64-NEXT: vextractf128 $1, %ymm0, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm3 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm2 = xmm2[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm3 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm0 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0] | |
| ; X64-AVX1-i64-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 | |
| ; X64-AVX1-i64-NEXT: vextractf128 $1, %ymm1, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm3 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm2 = xmm2[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm2 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm3 | |
| ; X64-AVX1-i64-NEXT: vshufpd {{.*#+}} xmm1 = xmm1[1,0] | |
| ; X64-AVX1-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; X64-AVX1-i64-NEXT: vmovq %rax, %xmm1 | |
| ; X64-AVX1-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm3[0],xmm1[0] | |
| ; X64-AVX1-i64-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1 | |
| ; X64-AVX1-i64-NEXT: retq | |
| ; | |
| ; AVX512-i64-LABEL: lrint_v8f64: | |
| ; AVX512-i64: # %bb.0: | |
| ; AVX512-i64-NEXT: vextractf32x4 $3, %zmm0, %xmm1 | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm1 = xmm1[1,0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm1, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm1 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] | |
| ; AVX512-i64-NEXT: vextractf32x4 $2, %zmm0, %xmm2 | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm3 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm2 = xmm2[1,0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0] | |
| ; AVX512-i64-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1 | |
| ; AVX512-i64-NEXT: vextractf128 $1, %ymm0, %xmm2 | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm3 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm2 = xmm2[1,0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm2, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm2 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm3 | |
| ; AVX512-i64-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0] | |
| ; AVX512-i64-NEXT: vcvtsd2si %xmm0, %rax | |
| ; AVX512-i64-NEXT: vmovq %rax, %xmm0 | |
| ; AVX512-i64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0] | |
| ; AVX512-i64-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0 | |
| ; AVX512-i64-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 | |
| ; AVX512-i64-NEXT: retq | |
| ; | |
| ; AVX512DQ-i64-LABEL: lrint_v8f64: | |
| ; AVX512DQ-i64: # %bb.0: | |
| ; AVX512DQ-i64-NEXT: vcvtpd2qq %zmm0, %zmm0 | |
| ; AVX512DQ-i64-NEXT: retq | |
| %a = call <8 x iXLen> @llvm.lrint.v8iXLen.v8f64(<8 x double> %x) | |
| ret <8 x iXLen> %a | |
| } | |
| declare <8 x iXLen> @llvm.lrint.v8iXLen.v8f64(<8 x double>) | |
| declare <1 x iXLen> @llvm.lrint.v1iXLen.v1fp128(<1 x fp128>) | |
| declare <2 x iXLen> @llvm.lrint.v2iXLen.v2fp128(<2 x fp128>) | |
| declare <4 x iXLen> @llvm.lrint.v4iXLen.v4fp128(<4 x fp128>) | |
| declare <8 x iXLen> @llvm.lrint.v8iXLen.v8fp128(<8 x fp128>) |