; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+sse2 -O3 | FileCheck %s --check-prefix=SSE-32
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+sse2 -O3 | FileCheck %s --check-prefix=SSE-64
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx -O3 | FileCheck %s --check-prefixes=AVX,AVX-32
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx -O3 | FileCheck %s --check-prefixes=AVX,AVX-64
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx512f -O3 | FileCheck %s --check-prefixes=AVX512F,AVX512F-32
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f -O3 | FileCheck %s --check-prefixes=AVX512F,AVX512F-64
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx512vl -O3 | FileCheck %s --check-prefixes=AVX512VL,AVX512VL-32
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512vl -O3 | FileCheck %s --check-prefixes=AVX512VL,AVX512VL-64
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=avx512dq -O3 | FileCheck %s --check-prefixes=AVX512DQ,AVX512DQ-32
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=avx512dq -O3 | FileCheck %s --check-prefixes=AVX512DQ,AVX512DQ-64
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=avx512dq,avx512vl -O3 | FileCheck %s --check-prefixes=AVX512VLDQ,AVX512VLDQ-32
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=avx512dq,avx512vl -O3 | FileCheck %s --check-prefixes=AVX512VLDQ,AVX512VLDQ-64
declare <2 x i64> @llvm.experimental.constrained.fptosi.v2i64.v2f64(<2 x double>, metadata)
declare <2 x i64> @llvm.experimental.constrained.fptoui.v2i64.v2f64(<2 x double>, metadata)
declare <2 x i64> @llvm.experimental.constrained.fptosi.v2i64.v2f32(<2 x float>, metadata)
declare <2 x i64> @llvm.experimental.constrained.fptoui.v2i64.v2f32(<2 x float>, metadata)
declare <2 x i32> @llvm.experimental.constrained.fptosi.v2i32.v2f64(<2 x double>, metadata)
declare <2 x i32> @llvm.experimental.constrained.fptoui.v2i32.v2f64(<2 x double>, metadata)
declare <2 x i32> @llvm.experimental.constrained.fptosi.v2i32.v2f32(<2 x float>, metadata)
declare <2 x i32> @llvm.experimental.constrained.fptoui.v2i32.v2f32(<2 x float>, metadata)
declare <2 x i16> @llvm.experimental.constrained.fptosi.v2i16.v2f64(<2 x double>, metadata)
declare <2 x i16> @llvm.experimental.constrained.fptoui.v2i16.v2f64(<2 x double>, metadata)
declare <2 x i16> @llvm.experimental.constrained.fptosi.v2i16.v2f32(<2 x float>, metadata)
declare <2 x i16> @llvm.experimental.constrained.fptoui.v2i16.v2f32(<2 x float>, metadata)
declare <2 x i8> @llvm.experimental.constrained.fptosi.v2i8.v2f64(<2 x double>, metadata)
declare <2 x i8> @llvm.experimental.constrained.fptoui.v2i8.v2f64(<2 x double>, metadata)
declare <2 x i8> @llvm.experimental.constrained.fptosi.v2i8.v2f32(<2 x float>, metadata)
declare <2 x i8> @llvm.experimental.constrained.fptoui.v2i8.v2f32(<2 x float>, metadata)
declare <2 x i1> @llvm.experimental.constrained.fptosi.v2i1.v2f64(<2 x double>, metadata)
declare <2 x i1> @llvm.experimental.constrained.fptoui.v2i1.v2f64(<2 x double>, metadata)
declare <2 x i1> @llvm.experimental.constrained.fptosi.v2i1.v2f32(<2 x float>, metadata)
declare <2 x i1> @llvm.experimental.constrained.fptoui.v2i1.v2f32(<2 x float>, metadata)
declare <4 x i32> @llvm.experimental.constrained.fptosi.v4i32.v4f32(<4 x float>, metadata)
declare <4 x i32> @llvm.experimental.constrained.fptoui.v4i32.v4f32(<4 x float>, metadata)
declare <4 x i16> @llvm.experimental.constrained.fptosi.v4i16.v4f32(<4 x float>, metadata)
declare <4 x i16> @llvm.experimental.constrained.fptoui.v4i16.v4f32(<4 x float>, metadata)
declare <4 x i8> @llvm.experimental.constrained.fptosi.v4i8.v4f32(<4 x float>, metadata)
declare <4 x i8> @llvm.experimental.constrained.fptoui.v4i8.v4f32(<4 x float>, metadata)
declare <4 x i1> @llvm.experimental.constrained.fptosi.v4i1.v4f32(<4 x float>, metadata)
declare <4 x i1> @llvm.experimental.constrained.fptoui.v4i1.v4f32(<4 x float>, metadata)
define <2 x i64> @strict_vector_fptosi_v2f64_to_v2i64(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movhps %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: movlps %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE-32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE-32-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm1
; SSE-64-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovlps %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: vmovhps %xmm0, (%esp)
; AVX-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: fldl (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm1
; AVX-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-32-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl %esp, %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512F-32-NEXT: andl $-8, %esp
; AVX512F-32-NEXT: subl $16, %esp
; AVX512F-32-NEXT: vmovlps %xmm0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: vmovhps %xmm0, (%esp)
; AVX512F-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fldl (%esp)
; AVX512F-32-NEXT: fisttpll (%esp)
; AVX512F-32-NEXT: wait
; AVX512F-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: movl %ebp, %esp
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512F-32-NEXT: retl
;
; AVX512F-64-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX512F-64: # %bb.0:
; AVX512F-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm1
; AVX512F-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512F-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm0
; AVX512F-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512F-64-NEXT: retq
;
; AVX512VL-32-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX512VL-32: # %bb.0:
; AVX512VL-32-NEXT: pushl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_offset 8
; AVX512VL-32-NEXT: .cfi_offset %ebp, -8
; AVX512VL-32-NEXT: movl %esp, %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512VL-32-NEXT: andl $-8, %esp
; AVX512VL-32-NEXT: subl $16, %esp
; AVX512VL-32-NEXT: vmovlps %xmm0, {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: vmovhps %xmm0, (%esp)
; AVX512VL-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fldl (%esp)
; AVX512VL-32-NEXT: fisttpll (%esp)
; AVX512VL-32-NEXT: wait
; AVX512VL-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: movl %ebp, %esp
; AVX512VL-32-NEXT: popl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512VL-32-NEXT: retl
;
; AVX512VL-64-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX512VL-64: # %bb.0:
; AVX512VL-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm1
; AVX512VL-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512VL-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm0
; AVX512VL-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512VL-64-NEXT: retq
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovaps %xmm0, %xmm0
; AVX512DQ-NEXT: vcvttpd2qq %zmm0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f64_to_v2i64:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2qq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i64> @llvm.experimental.constrained.fptosi.v2i64.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i64> %ret
}
define <2 x i64> @strict_vector_fptoui_v2f64_to_v2i64(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE-32-NEXT: comisd %xmm1, %xmm0
; SSE-32-NEXT: movapd %xmm1, %xmm2
; SSE-32-NEXT: jae .LBB1_2
; SSE-32-NEXT: # %bb.1:
; SSE-32-NEXT: xorpd %xmm2, %xmm2
; SSE-32-NEXT: .LBB1_2:
; SSE-32-NEXT: movapd %xmm0, %xmm3
; SSE-32-NEXT: subsd %xmm2, %xmm3
; SSE-32-NEXT: movsd %xmm3, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; SSE-32-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE-32-NEXT: movw %cx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-32-NEXT: comisd %xmm1, %xmm0
; SSE-32-NEXT: jae .LBB1_4
; SSE-32-NEXT: # %bb.3:
; SSE-32-NEXT: xorpd %xmm1, %xmm1
; SSE-32-NEXT: .LBB1_4:
; SSE-32-NEXT: subsd %xmm1, %xmm0
; SSE-32-NEXT: movsd %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %cl
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %edx
; SSE-32-NEXT: orl $3072, %edx # imm = 0xC00
; SSE-32-NEXT: movw %dx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-32-NEXT: movzbl %cl, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-32-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movsd {{.*#+}} xmm3 = mem[0],zero
; SSE-64-NEXT: comisd %xmm3, %xmm0
; SSE-64-NEXT: xorpd %xmm2, %xmm2
; SSE-64-NEXT: xorpd %xmm1, %xmm1
; SSE-64-NEXT: jb .LBB1_2
; SSE-64-NEXT: # %bb.1:
; SSE-64-NEXT: movapd %xmm3, %xmm1
; SSE-64-NEXT: .LBB1_2:
; SSE-64-NEXT: movapd %xmm0, %xmm4
; SSE-64-NEXT: subsd %xmm1, %xmm4
; SSE-64-NEXT: cvttsd2si %xmm4, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm1
; SSE-64-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-64-NEXT: comisd %xmm3, %xmm0
; SSE-64-NEXT: jb .LBB1_4
; SSE-64-NEXT: # %bb.3:
; SSE-64-NEXT: movapd %xmm3, %xmm2
; SSE-64-NEXT: .LBB1_4:
; SSE-64-NEXT: subsd %xmm2, %xmm0
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
; AVX-32-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX-32-NEXT: vcomisd %xmm1, %xmm2
; AVX-32-NEXT: vmovapd %xmm1, %xmm3
; AVX-32-NEXT: jae .LBB1_2
; AVX-32-NEXT: # %bb.1:
; AVX-32-NEXT: vxorpd %xmm3, %xmm3, %xmm3
; AVX-32-NEXT: .LBB1_2:
; AVX-32-NEXT: vsubsd %xmm3, %xmm2, %xmm2
; AVX-32-NEXT: vmovsd %xmm2, (%esp)
; AVX-32-NEXT: fldl (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %al
; AVX-32-NEXT: movzbl %al, %eax
; AVX-32-NEXT: shll $31, %eax
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX-32-NEXT: vcomisd %xmm1, %xmm0
; AVX-32-NEXT: jae .LBB1_4
; AVX-32-NEXT: # %bb.3:
; AVX-32-NEXT: vxorpd %xmm1, %xmm1, %xmm1
; AVX-32-NEXT: .LBB1_4:
; AVX-32-NEXT: vsubsd %xmm1, %xmm0, %xmm0
; AVX-32-NEXT: vmovsd %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %cl
; AVX-32-NEXT: movzbl %cl, %ecx
; AVX-32-NEXT: shll $31, %ecx
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX-64-NEXT: vcomisd %xmm1, %xmm0
; AVX-64-NEXT: vxorpd %xmm2, %xmm2, %xmm2
; AVX-64-NEXT: vxorpd %xmm3, %xmm3, %xmm3
; AVX-64-NEXT: jb .LBB1_2
; AVX-64-NEXT: # %bb.1:
; AVX-64-NEXT: vmovapd %xmm1, %xmm3
; AVX-64-NEXT: .LBB1_2:
; AVX-64-NEXT: vsubsd %xmm3, %xmm0, %xmm3
; AVX-64-NEXT: vcvttsd2si %xmm3, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm3
; AVX-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX-64-NEXT: vcomisd %xmm1, %xmm0
; AVX-64-NEXT: jb .LBB1_4
; AVX-64-NEXT: # %bb.3:
; AVX-64-NEXT: vmovapd %xmm1, %xmm2
; AVX-64-NEXT: .LBB1_4:
; AVX-64-NEXT: vsubsd %xmm2, %xmm0, %xmm0
; AVX-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-32-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl %esp, %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512F-32-NEXT: andl $-8, %esp
; AVX512F-32-NEXT: subl $16, %esp
; AVX512F-32-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX512F-32-NEXT: vmovsd {{.*#+}} xmm2 = mem[0],zero
; AVX512F-32-NEXT: xorl %eax, %eax
; AVX512F-32-NEXT: vcomisd %xmm2, %xmm1
; AVX512F-32-NEXT: setae %al
; AVX512F-32-NEXT: kmovw %eax, %k1
; AVX512F-32-NEXT: vmovsd %xmm2, %xmm2, %xmm3 {%k1} {z}
; AVX512F-32-NEXT: vsubsd %xmm3, %xmm1, %xmm1
; AVX512F-32-NEXT: vmovsd %xmm1, (%esp)
; AVX512F-32-NEXT: xorl %ecx, %ecx
; AVX512F-32-NEXT: vcomisd %xmm2, %xmm0
; AVX512F-32-NEXT: setae %cl
; AVX512F-32-NEXT: kmovw %ecx, %k1
; AVX512F-32-NEXT: vmovsd %xmm2, %xmm2, %xmm1 {%k1} {z}
; AVX512F-32-NEXT: vsubsd %xmm1, %xmm0, %xmm0
; AVX512F-32-NEXT: vmovsd %xmm0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fldl (%esp)
; AVX512F-32-NEXT: fisttpll (%esp)
; AVX512F-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: wait
; AVX512F-32-NEXT: shll $31, %eax
; AVX512F-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX512F-32-NEXT: shll $31, %ecx
; AVX512F-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX512F-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX512F-32-NEXT: movl %ebp, %esp
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512F-32-NEXT: retl
;
; AVX512F-64-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX512F-64: # %bb.0:
; AVX512F-64-NEXT: vcvttsd2usi %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm1
; AVX512F-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512F-64-NEXT: vcvttsd2usi %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm0
; AVX512F-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512F-64-NEXT: retq
;
; AVX512VL-32-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX512VL-32: # %bb.0:
; AVX512VL-32-NEXT: pushl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_offset 8
; AVX512VL-32-NEXT: .cfi_offset %ebp, -8
; AVX512VL-32-NEXT: movl %esp, %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512VL-32-NEXT: andl $-8, %esp
; AVX512VL-32-NEXT: subl $16, %esp
; AVX512VL-32-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX512VL-32-NEXT: vmovsd {{.*#+}} xmm2 = mem[0],zero
; AVX512VL-32-NEXT: xorl %eax, %eax
; AVX512VL-32-NEXT: vcomisd %xmm2, %xmm1
; AVX512VL-32-NEXT: setae %al
; AVX512VL-32-NEXT: kmovw %eax, %k1
; AVX512VL-32-NEXT: vmovsd %xmm2, %xmm2, %xmm3 {%k1} {z}
; AVX512VL-32-NEXT: vsubsd %xmm3, %xmm1, %xmm1
; AVX512VL-32-NEXT: vmovsd %xmm1, (%esp)
; AVX512VL-32-NEXT: xorl %ecx, %ecx
; AVX512VL-32-NEXT: vcomisd %xmm2, %xmm0
; AVX512VL-32-NEXT: setae %cl
; AVX512VL-32-NEXT: kmovw %ecx, %k1
; AVX512VL-32-NEXT: vmovsd %xmm2, %xmm2, %xmm1 {%k1} {z}
; AVX512VL-32-NEXT: vsubsd %xmm1, %xmm0, %xmm0
; AVX512VL-32-NEXT: vmovsd %xmm0, {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fldl (%esp)
; AVX512VL-32-NEXT: fisttpll (%esp)
; AVX512VL-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: wait
; AVX512VL-32-NEXT: shll $31, %eax
; AVX512VL-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX512VL-32-NEXT: shll $31, %ecx
; AVX512VL-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX512VL-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX512VL-32-NEXT: movl %ebp, %esp
; AVX512VL-32-NEXT: popl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512VL-32-NEXT: retl
;
; AVX512VL-64-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX512VL-64: # %bb.0:
; AVX512VL-64-NEXT: vcvttsd2usi %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm1
; AVX512VL-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512VL-64-NEXT: vcvttsd2usi %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm0
; AVX512VL-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512VL-64-NEXT: retq
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovaps %xmm0, %xmm0
; AVX512DQ-NEXT: vcvttpd2uqq %zmm0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f64_to_v2i64:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2uqq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i64> @llvm.experimental.constrained.fptoui.v2i64.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i64> %ret
}
define <2 x i64> @strict_vector_fptosi_v2f32_to_v2i64(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE-32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE-32-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm1
; SSE-64-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovss %xmm0, (%esp)
; AVX-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: flds (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vcvttss2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm1
; AVX-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX-64-NEXT: vcvttss2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl %esp, %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512F-32-NEXT: andl $-8, %esp
; AVX512F-32-NEXT: subl $16, %esp
; AVX512F-32-NEXT: vmovd %xmm0, (%esp)
; AVX512F-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: flds (%esp)
; AVX512F-32-NEXT: fisttpll (%esp)
; AVX512F-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: wait
; AVX512F-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: movl %ebp, %esp
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512F-32-NEXT: retl
;
; AVX512F-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX512F-64: # %bb.0:
; AVX512F-64-NEXT: vcvttss2si %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm1
; AVX512F-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512F-64-NEXT: vcvttss2si %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm0
; AVX512F-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512F-64-NEXT: retq
;
; AVX512VL-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX512VL-32: # %bb.0:
; AVX512VL-32-NEXT: pushl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_offset 8
; AVX512VL-32-NEXT: .cfi_offset %ebp, -8
; AVX512VL-32-NEXT: movl %esp, %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512VL-32-NEXT: andl $-8, %esp
; AVX512VL-32-NEXT: subl $16, %esp
; AVX512VL-32-NEXT: vmovd %xmm0, (%esp)
; AVX512VL-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: flds (%esp)
; AVX512VL-32-NEXT: fisttpll (%esp)
; AVX512VL-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: wait
; AVX512VL-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: movl %ebp, %esp
; AVX512VL-32-NEXT: popl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512VL-32-NEXT: retl
;
; AVX512VL-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX512VL-64: # %bb.0:
; AVX512VL-64-NEXT: vcvttss2si %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm1
; AVX512VL-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512VL-64-NEXT: vcvttss2si %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm0
; AVX512VL-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512VL-64-NEXT: retq
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2qq %ymm0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f32_to_v2i64:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2qq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i64> @llvm.experimental.constrained.fptosi.v2i64.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i64> %ret
}
define <2 x i64> @strict_vector_fptosi_v2f32_to_v2i64_load128(ptr %x) strictfp {
; SSE-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movl 8(%ebp), %eax
; SSE-32-NEXT: movaps (%eax), %xmm0
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE-32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE-32-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movaps (%rdi), %xmm1
; SSE-64-NEXT: cvttss2si %xmm1, %rax
; SSE-64-NEXT: movq %rax, %xmm0
; SSE-64-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1,1,1]
; SSE-64-NEXT: cvttss2si %xmm1, %rax
; SSE-64-NEXT: movq %rax, %xmm1
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: movl 8(%ebp), %eax
; AVX-32-NEXT: vmovaps (%eax), %xmm0
; AVX-32-NEXT: vmovss %xmm0, (%esp)
; AVX-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: flds (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vcvttss2si 4(%rdi), %rax
; AVX-64-NEXT: vmovq %rax, %xmm0
; AVX-64-NEXT: vcvttss2si (%rdi), %rax
; AVX-64-NEXT: vmovq %rax, %xmm1
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl %esp, %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512F-32-NEXT: andl $-8, %esp
; AVX512F-32-NEXT: subl $16, %esp
; AVX512F-32-NEXT: movl 8(%ebp), %eax
; AVX512F-32-NEXT: vmovdqa (%eax), %xmm0
; AVX512F-32-NEXT: vmovd %xmm0, (%esp)
; AVX512F-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: flds (%esp)
; AVX512F-32-NEXT: fisttpll (%esp)
; AVX512F-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: wait
; AVX512F-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: movl %ebp, %esp
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512F-32-NEXT: retl
;
; AVX512F-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512F-64: # %bb.0:
; AVX512F-64-NEXT: vcvttss2si 4(%rdi), %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm0
; AVX512F-64-NEXT: vcvttss2si (%rdi), %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm1
; AVX512F-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512F-64-NEXT: retq
;
; AVX512VL-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512VL-32: # %bb.0:
; AVX512VL-32-NEXT: pushl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_offset 8
; AVX512VL-32-NEXT: .cfi_offset %ebp, -8
; AVX512VL-32-NEXT: movl %esp, %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512VL-32-NEXT: andl $-8, %esp
; AVX512VL-32-NEXT: subl $16, %esp
; AVX512VL-32-NEXT: movl 8(%ebp), %eax
; AVX512VL-32-NEXT: vmovdqa (%eax), %xmm0
; AVX512VL-32-NEXT: vmovd %xmm0, (%esp)
; AVX512VL-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: flds (%esp)
; AVX512VL-32-NEXT: fisttpll (%esp)
; AVX512VL-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: wait
; AVX512VL-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: movl %ebp, %esp
; AVX512VL-32-NEXT: popl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512VL-32-NEXT: retl
;
; AVX512VL-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512VL-64: # %bb.0:
; AVX512VL-64-NEXT: vcvttss2si 4(%rdi), %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm0
; AVX512VL-64-NEXT: vcvttss2si (%rdi), %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm1
; AVX512VL-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512VL-64-NEXT: retq
;
; AVX512DQ-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512DQ-32: # %bb.0:
; AVX512DQ-32-NEXT: movl {{[0-9]+}}(%esp), %eax
; AVX512DQ-32-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ-32-NEXT: vcvttps2qq %ymm0, %zmm0
; AVX512DQ-32-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-32-NEXT: vzeroupper
; AVX512DQ-32-NEXT: retl
;
; AVX512DQ-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512DQ-64: # %bb.0:
; AVX512DQ-64-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ-64-NEXT: vcvttps2qq %ymm0, %zmm0
; AVX512DQ-64-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-64-NEXT: vzeroupper
; AVX512DQ-64-NEXT: retq
;
; AVX512VLDQ-32-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512VLDQ-32: # %bb.0:
; AVX512VLDQ-32-NEXT: movl {{[0-9]+}}(%esp), %eax
; AVX512VLDQ-32-NEXT: vcvttps2qq (%eax), %xmm0
; AVX512VLDQ-32-NEXT: retl
;
; AVX512VLDQ-64-LABEL: strict_vector_fptosi_v2f32_to_v2i64_load128:
; AVX512VLDQ-64: # %bb.0:
; AVX512VLDQ-64-NEXT: vcvttps2qq (%rdi), %xmm0
; AVX512VLDQ-64-NEXT: retq
%a = load <4 x float>, ptr %x
%b = shufflevector <4 x float> %a, <4 x float> undef, <2 x i32> <i32 0, i32 1>
%c = call <2 x i64> @llvm.experimental.constrained.fptosi.v2i64.v2f32(<2 x float> %b, metadata !"fpexcept.strict") #0
ret <2 x i64> %c
}
define <2 x i64> @strict_vector_fptoui_v2f32_to_v2i64(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE-32-NEXT: comiss %xmm1, %xmm0
; SSE-32-NEXT: movaps %xmm1, %xmm2
; SSE-32-NEXT: jae .LBB4_2
; SSE-32-NEXT: # %bb.1:
; SSE-32-NEXT: xorps %xmm2, %xmm2
; SSE-32-NEXT: .LBB4_2:
; SSE-32-NEXT: movaps %xmm0, %xmm3
; SSE-32-NEXT: subss %xmm2, %xmm3
; SSE-32-NEXT: movss %xmm3, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; SSE-32-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE-32-NEXT: movw %cx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-32-NEXT: comiss %xmm1, %xmm0
; SSE-32-NEXT: jae .LBB4_4
; SSE-32-NEXT: # %bb.3:
; SSE-32-NEXT: xorps %xmm1, %xmm1
; SSE-32-NEXT: .LBB4_4:
; SSE-32-NEXT: subss %xmm1, %xmm0
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %cl
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %edx
; SSE-32-NEXT: orl $3072, %edx # imm = 0xC00
; SSE-32-NEXT: movw %dx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-32-NEXT: movzbl %cl, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-32-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-64-NEXT: comiss %xmm3, %xmm0
; SSE-64-NEXT: xorps %xmm2, %xmm2
; SSE-64-NEXT: xorps %xmm1, %xmm1
; SSE-64-NEXT: jb .LBB4_2
; SSE-64-NEXT: # %bb.1:
; SSE-64-NEXT: movaps %xmm3, %xmm1
; SSE-64-NEXT: .LBB4_2:
; SSE-64-NEXT: movaps %xmm0, %xmm4
; SSE-64-NEXT: subss %xmm1, %xmm4
; SSE-64-NEXT: cvttss2si %xmm4, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm1
; SSE-64-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-64-NEXT: comiss %xmm3, %xmm0
; SSE-64-NEXT: jb .LBB4_4
; SSE-64-NEXT: # %bb.3:
; SSE-64-NEXT: movaps %xmm3, %xmm2
; SSE-64-NEXT: .LBB4_4:
; SSE-64-NEXT: subss %xmm2, %xmm0
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
; AVX-32-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX-32-NEXT: vcomiss %xmm1, %xmm2
; AVX-32-NEXT: vmovaps %xmm1, %xmm3
; AVX-32-NEXT: jae .LBB4_2
; AVX-32-NEXT: # %bb.1:
; AVX-32-NEXT: vxorps %xmm3, %xmm3, %xmm3
; AVX-32-NEXT: .LBB4_2:
; AVX-32-NEXT: vsubss %xmm3, %xmm2, %xmm2
; AVX-32-NEXT: vmovss %xmm2, {{[0-9]+}}(%esp)
; AVX-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %al
; AVX-32-NEXT: movzbl %al, %eax
; AVX-32-NEXT: shll $31, %eax
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX-32-NEXT: vcomiss %xmm1, %xmm0
; AVX-32-NEXT: jae .LBB4_4
; AVX-32-NEXT: # %bb.3:
; AVX-32-NEXT: vxorps %xmm1, %xmm1, %xmm1
; AVX-32-NEXT: .LBB4_4:
; AVX-32-NEXT: vsubss %xmm1, %xmm0, %xmm0
; AVX-32-NEXT: vmovss %xmm0, (%esp)
; AVX-32-NEXT: flds (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %cl
; AVX-32-NEXT: movzbl %cl, %ecx
; AVX-32-NEXT: shll $31, %ecx
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX-64-NEXT: vcomiss %xmm1, %xmm0
; AVX-64-NEXT: vxorps %xmm2, %xmm2, %xmm2
; AVX-64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; AVX-64-NEXT: jb .LBB4_2
; AVX-64-NEXT: # %bb.1:
; AVX-64-NEXT: vmovaps %xmm1, %xmm3
; AVX-64-NEXT: .LBB4_2:
; AVX-64-NEXT: vsubss %xmm3, %xmm0, %xmm3
; AVX-64-NEXT: vcvttss2si %xmm3, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm3
; AVX-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX-64-NEXT: vcomiss %xmm1, %xmm0
; AVX-64-NEXT: jb .LBB4_4
; AVX-64-NEXT: # %bb.3:
; AVX-64-NEXT: vmovaps %xmm1, %xmm2
; AVX-64-NEXT: .LBB4_4:
; AVX-64-NEXT: vsubss %xmm2, %xmm0, %xmm0
; AVX-64-NEXT: vcvttss2si %xmm0, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl %esp, %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512F-32-NEXT: andl $-8, %esp
; AVX512F-32-NEXT: subl $16, %esp
; AVX512F-32-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
; AVX512F-32-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: xorl %eax, %eax
; AVX512F-32-NEXT: vcomiss %xmm2, %xmm1
; AVX512F-32-NEXT: setae %al
; AVX512F-32-NEXT: kmovw %eax, %k1
; AVX512F-32-NEXT: vmovss %xmm2, %xmm2, %xmm3 {%k1} {z}
; AVX512F-32-NEXT: vsubss %xmm3, %xmm1, %xmm1
; AVX512F-32-NEXT: vmovss %xmm1, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: xorl %ecx, %ecx
; AVX512F-32-NEXT: vcomiss %xmm2, %xmm0
; AVX512F-32-NEXT: setae %cl
; AVX512F-32-NEXT: kmovw %ecx, %k1
; AVX512F-32-NEXT: vmovss %xmm2, %xmm2, %xmm1 {%k1} {z}
; AVX512F-32-NEXT: vsubss %xmm1, %xmm0, %xmm0
; AVX512F-32-NEXT: vmovss %xmm0, (%esp)
; AVX512F-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: flds (%esp)
; AVX512F-32-NEXT: fisttpll (%esp)
; AVX512F-32-NEXT: wait
; AVX512F-32-NEXT: shll $31, %eax
; AVX512F-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX512F-32-NEXT: shll $31, %ecx
; AVX512F-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX512F-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX512F-32-NEXT: movl %ebp, %esp
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512F-32-NEXT: retl
;
; AVX512F-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX512F-64: # %bb.0:
; AVX512F-64-NEXT: vcvttss2usi %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm1
; AVX512F-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512F-64-NEXT: vcvttss2usi %xmm0, %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm0
; AVX512F-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512F-64-NEXT: retq
;
; AVX512VL-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX512VL-32: # %bb.0:
; AVX512VL-32-NEXT: pushl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_offset 8
; AVX512VL-32-NEXT: .cfi_offset %ebp, -8
; AVX512VL-32-NEXT: movl %esp, %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512VL-32-NEXT: andl $-8, %esp
; AVX512VL-32-NEXT: subl $16, %esp
; AVX512VL-32-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
; AVX512VL-32-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: xorl %eax, %eax
; AVX512VL-32-NEXT: vcomiss %xmm2, %xmm1
; AVX512VL-32-NEXT: setae %al
; AVX512VL-32-NEXT: kmovw %eax, %k1
; AVX512VL-32-NEXT: vmovss %xmm2, %xmm2, %xmm3 {%k1} {z}
; AVX512VL-32-NEXT: vsubss %xmm3, %xmm1, %xmm1
; AVX512VL-32-NEXT: vmovss %xmm1, {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: xorl %ecx, %ecx
; AVX512VL-32-NEXT: vcomiss %xmm2, %xmm0
; AVX512VL-32-NEXT: setae %cl
; AVX512VL-32-NEXT: kmovw %ecx, %k1
; AVX512VL-32-NEXT: vmovss %xmm2, %xmm2, %xmm1 {%k1} {z}
; AVX512VL-32-NEXT: vsubss %xmm1, %xmm0, %xmm0
; AVX512VL-32-NEXT: vmovss %xmm0, (%esp)
; AVX512VL-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: flds (%esp)
; AVX512VL-32-NEXT: fisttpll (%esp)
; AVX512VL-32-NEXT: wait
; AVX512VL-32-NEXT: shll $31, %eax
; AVX512VL-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX512VL-32-NEXT: shll $31, %ecx
; AVX512VL-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX512VL-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX512VL-32-NEXT: movl %ebp, %esp
; AVX512VL-32-NEXT: popl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512VL-32-NEXT: retl
;
; AVX512VL-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX512VL-64: # %bb.0:
; AVX512VL-64-NEXT: vcvttss2usi %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm1
; AVX512VL-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512VL-64-NEXT: vcvttss2usi %xmm0, %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm0
; AVX512VL-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512VL-64-NEXT: retq
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2uqq %ymm0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f32_to_v2i64:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2uqq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i64> @llvm.experimental.constrained.fptoui.v2i64.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i64> %ret
}
define <2 x i64> @strict_vector_fptoui_v2f32_to_v2i64_load128(ptr %x) strictfp {
; SSE-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movl 8(%ebp), %eax
; SSE-32-NEXT: movaps (%eax), %xmm0
; SSE-32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE-32-NEXT: comiss %xmm1, %xmm0
; SSE-32-NEXT: movaps %xmm1, %xmm2
; SSE-32-NEXT: jae .LBB5_2
; SSE-32-NEXT: # %bb.1:
; SSE-32-NEXT: xorps %xmm2, %xmm2
; SSE-32-NEXT: .LBB5_2:
; SSE-32-NEXT: movaps %xmm0, %xmm3
; SSE-32-NEXT: subss %xmm2, %xmm3
; SSE-32-NEXT: movss %xmm3, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; SSE-32-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE-32-NEXT: movw %cx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-32-NEXT: comiss %xmm1, %xmm0
; SSE-32-NEXT: jae .LBB5_4
; SSE-32-NEXT: # %bb.3:
; SSE-32-NEXT: xorps %xmm1, %xmm1
; SSE-32-NEXT: .LBB5_4:
; SSE-32-NEXT: subss %xmm1, %xmm0
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %cl
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %edx
; SSE-32-NEXT: orl $3072, %edx # imm = 0xC00
; SSE-32-NEXT: movw %dx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-32-NEXT: movzbl %cl, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-32-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movaps (%rdi), %xmm1
; SSE-64-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-64-NEXT: comiss %xmm3, %xmm1
; SSE-64-NEXT: xorps %xmm2, %xmm2
; SSE-64-NEXT: xorps %xmm0, %xmm0
; SSE-64-NEXT: jb .LBB5_2
; SSE-64-NEXT: # %bb.1:
; SSE-64-NEXT: movaps %xmm3, %xmm0
; SSE-64-NEXT: .LBB5_2:
; SSE-64-NEXT: movaps %xmm1, %xmm4
; SSE-64-NEXT: subss %xmm0, %xmm4
; SSE-64-NEXT: cvttss2si %xmm4, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm0
; SSE-64-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1,1,1]
; SSE-64-NEXT: comiss %xmm3, %xmm1
; SSE-64-NEXT: jb .LBB5_4
; SSE-64-NEXT: # %bb.3:
; SSE-64-NEXT: movaps %xmm3, %xmm2
; SSE-64-NEXT: .LBB5_4:
; SSE-64-NEXT: subss %xmm2, %xmm1
; SSE-64-NEXT: cvttss2si %xmm1, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm1
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: movl 8(%ebp), %eax
; AVX-32-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX-32-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX-32-NEXT: vcomiss %xmm1, %xmm2
; AVX-32-NEXT: vmovaps %xmm1, %xmm3
; AVX-32-NEXT: jae .LBB5_2
; AVX-32-NEXT: # %bb.1:
; AVX-32-NEXT: vxorps %xmm3, %xmm3, %xmm3
; AVX-32-NEXT: .LBB5_2:
; AVX-32-NEXT: vsubss %xmm3, %xmm2, %xmm2
; AVX-32-NEXT: vmovss %xmm2, {{[0-9]+}}(%esp)
; AVX-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %al
; AVX-32-NEXT: movzbl %al, %eax
; AVX-32-NEXT: shll $31, %eax
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX-32-NEXT: vcomiss %xmm1, %xmm0
; AVX-32-NEXT: jae .LBB5_4
; AVX-32-NEXT: # %bb.3:
; AVX-32-NEXT: vxorps %xmm1, %xmm1, %xmm1
; AVX-32-NEXT: .LBB5_4:
; AVX-32-NEXT: vsubss %xmm1, %xmm0, %xmm0
; AVX-32-NEXT: vmovss %xmm0, (%esp)
; AVX-32-NEXT: flds (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %cl
; AVX-32-NEXT: movzbl %cl, %ecx
; AVX-32-NEXT: shll $31, %ecx
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-64-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; AVX-64-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX-64-NEXT: vcomiss %xmm1, %xmm3
; AVX-64-NEXT: vxorps %xmm2, %xmm2, %xmm2
; AVX-64-NEXT: vxorps %xmm4, %xmm4, %xmm4
; AVX-64-NEXT: jb .LBB5_2
; AVX-64-NEXT: # %bb.1:
; AVX-64-NEXT: vmovaps %xmm1, %xmm4
; AVX-64-NEXT: .LBB5_2:
; AVX-64-NEXT: vsubss %xmm4, %xmm3, %xmm3
; AVX-64-NEXT: vcvttss2si %xmm3, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm3
; AVX-64-NEXT: vcomiss %xmm1, %xmm0
; AVX-64-NEXT: jb .LBB5_4
; AVX-64-NEXT: # %bb.3:
; AVX-64-NEXT: vmovaps %xmm1, %xmm2
; AVX-64-NEXT: .LBB5_4:
; AVX-64-NEXT: vsubss %xmm2, %xmm0, %xmm0
; AVX-64-NEXT: vcvttss2si %xmm0, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0]
; AVX-64-NEXT: retq
;
; AVX512F-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl %esp, %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512F-32-NEXT: andl $-8, %esp
; AVX512F-32-NEXT: subl $16, %esp
; AVX512F-32-NEXT: movl 8(%ebp), %eax
; AVX512F-32-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: xorl %eax, %eax
; AVX512F-32-NEXT: vcomiss %xmm2, %xmm1
; AVX512F-32-NEXT: setae %al
; AVX512F-32-NEXT: kmovw %eax, %k1
; AVX512F-32-NEXT: vmovss %xmm2, %xmm2, %xmm3 {%k1} {z}
; AVX512F-32-NEXT: vsubss %xmm3, %xmm1, %xmm1
; AVX512F-32-NEXT: vmovss %xmm1, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: xorl %ecx, %ecx
; AVX512F-32-NEXT: vcomiss %xmm2, %xmm0
; AVX512F-32-NEXT: setae %cl
; AVX512F-32-NEXT: kmovw %ecx, %k1
; AVX512F-32-NEXT: vmovss %xmm2, %xmm2, %xmm1 {%k1} {z}
; AVX512F-32-NEXT: vsubss %xmm1, %xmm0, %xmm0
; AVX512F-32-NEXT: vmovss %xmm0, (%esp)
; AVX512F-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: flds (%esp)
; AVX512F-32-NEXT: fisttpll (%esp)
; AVX512F-32-NEXT: wait
; AVX512F-32-NEXT: shll $31, %eax
; AVX512F-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX512F-32-NEXT: shll $31, %ecx
; AVX512F-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX512F-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512F-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX512F-32-NEXT: movl %ebp, %esp
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512F-32-NEXT: retl
;
; AVX512F-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512F-64: # %bb.0:
; AVX512F-64-NEXT: vcvttss2usi 4(%rdi), %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm0
; AVX512F-64-NEXT: vcvttss2usi (%rdi), %rax
; AVX512F-64-NEXT: vmovq %rax, %xmm1
; AVX512F-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512F-64-NEXT: retq
;
; AVX512VL-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512VL-32: # %bb.0:
; AVX512VL-32-NEXT: pushl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_offset 8
; AVX512VL-32-NEXT: .cfi_offset %ebp, -8
; AVX512VL-32-NEXT: movl %esp, %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa_register %ebp
; AVX512VL-32-NEXT: andl $-8, %esp
; AVX512VL-32-NEXT: subl $16, %esp
; AVX512VL-32-NEXT: movl 8(%ebp), %eax
; AVX512VL-32-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: xorl %eax, %eax
; AVX512VL-32-NEXT: vcomiss %xmm2, %xmm1
; AVX512VL-32-NEXT: setae %al
; AVX512VL-32-NEXT: kmovw %eax, %k1
; AVX512VL-32-NEXT: vmovss %xmm2, %xmm2, %xmm3 {%k1} {z}
; AVX512VL-32-NEXT: vsubss %xmm3, %xmm1, %xmm1
; AVX512VL-32-NEXT: vmovss %xmm1, {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: xorl %ecx, %ecx
; AVX512VL-32-NEXT: vcomiss %xmm2, %xmm0
; AVX512VL-32-NEXT: setae %cl
; AVX512VL-32-NEXT: kmovw %ecx, %k1
; AVX512VL-32-NEXT: vmovss %xmm2, %xmm2, %xmm1 {%k1} {z}
; AVX512VL-32-NEXT: vsubss %xmm1, %xmm0, %xmm0
; AVX512VL-32-NEXT: vmovss %xmm0, (%esp)
; AVX512VL-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX512VL-32-NEXT: flds (%esp)
; AVX512VL-32-NEXT: fisttpll (%esp)
; AVX512VL-32-NEXT: wait
; AVX512VL-32-NEXT: shll $31, %eax
; AVX512VL-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX512VL-32-NEXT: shll $31, %ecx
; AVX512VL-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX512VL-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512VL-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX512VL-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX512VL-32-NEXT: movl %ebp, %esp
; AVX512VL-32-NEXT: popl %ebp
; AVX512VL-32-NEXT: .cfi_def_cfa %esp, 4
; AVX512VL-32-NEXT: retl
;
; AVX512VL-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512VL-64: # %bb.0:
; AVX512VL-64-NEXT: vcvttss2usi 4(%rdi), %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm0
; AVX512VL-64-NEXT: vcvttss2usi (%rdi), %rax
; AVX512VL-64-NEXT: vmovq %rax, %xmm1
; AVX512VL-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512VL-64-NEXT: retq
;
; AVX512DQ-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512DQ-32: # %bb.0:
; AVX512DQ-32-NEXT: movl {{[0-9]+}}(%esp), %eax
; AVX512DQ-32-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ-32-NEXT: vcvttps2uqq %ymm0, %zmm0
; AVX512DQ-32-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-32-NEXT: vzeroupper
; AVX512DQ-32-NEXT: retl
;
; AVX512DQ-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512DQ-64: # %bb.0:
; AVX512DQ-64-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ-64-NEXT: vcvttps2uqq %ymm0, %zmm0
; AVX512DQ-64-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-64-NEXT: vzeroupper
; AVX512DQ-64-NEXT: retq
;
; AVX512VLDQ-32-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512VLDQ-32: # %bb.0:
; AVX512VLDQ-32-NEXT: movl {{[0-9]+}}(%esp), %eax
; AVX512VLDQ-32-NEXT: vcvttps2uqq (%eax), %xmm0
; AVX512VLDQ-32-NEXT: retl
;
; AVX512VLDQ-64-LABEL: strict_vector_fptoui_v2f32_to_v2i64_load128:
; AVX512VLDQ-64: # %bb.0:
; AVX512VLDQ-64-NEXT: vcvttps2uqq (%rdi), %xmm0
; AVX512VLDQ-64-NEXT: retq
%a = load <4 x float>, ptr %x
%b = shufflevector <4 x float> %a, <4 x float> undef, <2 x i32> <i32 0, i32 1>
%c = call <2 x i64> @llvm.experimental.constrained.fptoui.v2i64.v2f32(<2 x float> %b, metadata !"fpexcept.strict") #0
ret <2 x i64> %c
}
define <2 x i32> @strict_vector_fptosi_v2f64_to_v2i32(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f64_to_v2i32:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f64_to_v2i32:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v2f64_to_v2i32:
; AVX: # %bb.0:
; AVX-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v2f64_to_v2i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f64_to_v2i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f64_to_v2i32:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f64_to_v2i32:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i32> @llvm.experimental.constrained.fptosi.v2i32.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i32> %ret
}
define <2 x i32> @strict_vector_fptoui_v2f64_to_v2i32(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movsd {{.*#+}} xmm3 = mem[0],zero
; SSE-32-NEXT: comisd %xmm3, %xmm0
; SSE-32-NEXT: xorpd %xmm2, %xmm2
; SSE-32-NEXT: xorpd %xmm1, %xmm1
; SSE-32-NEXT: jb .LBB7_2
; SSE-32-NEXT: # %bb.1:
; SSE-32-NEXT: movapd %xmm3, %xmm1
; SSE-32-NEXT: .LBB7_2:
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: movapd %xmm0, %xmm4
; SSE-32-NEXT: subsd %xmm1, %xmm4
; SSE-32-NEXT: cvttsd2si %xmm4, %ecx
; SSE-32-NEXT: xorl %eax, %ecx
; SSE-32-NEXT: movd %ecx, %xmm1
; SSE-32-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-32-NEXT: comisd %xmm3, %xmm0
; SSE-32-NEXT: jb .LBB7_4
; SSE-32-NEXT: # %bb.3:
; SSE-32-NEXT: movapd %xmm3, %xmm2
; SSE-32-NEXT: .LBB7_4:
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: subsd %xmm2, %xmm0
; SSE-32-NEXT: cvttsd2si %xmm0, %ecx
; SSE-32-NEXT: xorl %eax, %ecx
; SSE-32-NEXT: movd %ecx, %xmm0
; SSE-32-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE-32-NEXT: movdqa %xmm1, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: movd %eax, %xmm1
; SSE-64-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: movd %eax, %xmm0
; SSE-64-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovlps %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: vmovhps %xmm0, (%esp)
; AVX-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: fldl (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, (%esp), %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX-64-NEXT: vcvttsd2si %xmm1, %rax
; AVX-64-NEXT: vcvttsd2si %xmm0, %rcx
; AVX-64-NEXT: vmovd %ecx, %xmm0
; AVX-64-NEXT: vpinsrd $1, %eax, %xmm0, %xmm0
; AVX-64-NEXT: retq
;
; AVX512F-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovaps %xmm0, %xmm0
; AVX512F-NEXT: vcvttpd2udq %zmm0, %ymm0
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2udq %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovaps %xmm0, %xmm0
; AVX512DQ-NEXT: vcvttpd2udq %zmm0, %ymm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f64_to_v2i32:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2udq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i32> @llvm.experimental.constrained.fptoui.v2i32.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i32> %ret
}
define <2 x i32> @strict_vector_fptosi_v2f32_to_v2i32(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f32_to_v2i32:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f32_to_v2i32:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v2f32_to_v2i32:
; AVX: # %bb.0:
; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v2f32_to_v2i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f32_to_v2i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f32_to_v2i32:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f32_to_v2i32:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i32> @llvm.experimental.constrained.fptosi.v2i32.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i32> %ret
}
define <2 x i32> @strict_vector_fptoui_v2f32_to_v2i32(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-32-NEXT: comiss %xmm3, %xmm0
; SSE-32-NEXT: xorps %xmm2, %xmm2
; SSE-32-NEXT: xorps %xmm1, %xmm1
; SSE-32-NEXT: jb .LBB9_2
; SSE-32-NEXT: # %bb.1:
; SSE-32-NEXT: movaps %xmm3, %xmm1
; SSE-32-NEXT: .LBB9_2:
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: movaps %xmm0, %xmm4
; SSE-32-NEXT: subss %xmm1, %xmm4
; SSE-32-NEXT: cvttss2si %xmm4, %ecx
; SSE-32-NEXT: xorl %eax, %ecx
; SSE-32-NEXT: movd %ecx, %xmm1
; SSE-32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-32-NEXT: comiss %xmm3, %xmm0
; SSE-32-NEXT: jb .LBB9_4
; SSE-32-NEXT: # %bb.3:
; SSE-32-NEXT: movaps %xmm3, %xmm2
; SSE-32-NEXT: .LBB9_4:
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: subss %xmm2, %xmm0
; SSE-32-NEXT: cvttss2si %xmm0, %ecx
; SSE-32-NEXT: xorl %eax, %ecx
; SSE-32-NEXT: movd %ecx, %xmm0
; SSE-32-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE-32-NEXT: movdqa %xmm1, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: movd %eax, %xmm1
; SSE-64-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: movd %eax, %xmm0
; SSE-64-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovss %xmm0, (%esp)
; AVX-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: flds (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
; AVX-64-NEXT: vcvttss2si %xmm1, %rax
; AVX-64-NEXT: vcvttss2si %xmm0, %rcx
; AVX-64-NEXT: vmovd %ecx, %xmm0
; AVX-64-NEXT: vpinsrd $1, %eax, %xmm0, %xmm0
; AVX-64-NEXT: retq
;
; AVX512F-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512F-NEXT: vcvttps2udq %zmm0, %zmm0
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VL-NEXT: vcvttps2udq %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2udq %zmm0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f32_to_v2i32:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VLDQ-NEXT: vcvttps2udq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i32> @llvm.experimental.constrained.fptoui.v2i32.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i32> %ret
}
define <2 x i16> @strict_vector_fptosi_v2f64_to_v2i16(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f64_to_v2i16:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-32-NEXT: packssdw %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f64_to_v2i16:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-64-NEXT: packssdw %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v2f64_to_v2i16:
; AVX: # %bb.0:
; AVX-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v2f64_to_v2i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f64_to_v2i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f64_to_v2i16:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f64_to_v2i16:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i16> @llvm.experimental.constrained.fptosi.v2i16.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i16> %ret
}
define <2 x i16> @strict_vector_fptoui_v2f64_to_v2i16(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f64_to_v2i16:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-32-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f64_to_v2i16:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-64-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptoui_v2f64_to_v2i16:
; AVX: # %bb.0:
; AVX-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptoui_v2f64_to_v2i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f64_to_v2i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f64_to_v2i16:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f64_to_v2i16:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i16> @llvm.experimental.constrained.fptoui.v2i16.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i16> %ret
}
define <2 x i16> @strict_vector_fptosi_v2f32_to_v2i16(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f32_to_v2i16:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: packssdw %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f32_to_v2i16:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: packssdw %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v2f32_to_v2i16:
; AVX: # %bb.0:
; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v2f32_to_v2i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f32_to_v2i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f32_to_v2i16:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f32_to_v2i16:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i16> @llvm.experimental.constrained.fptosi.v2i16.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i16> %ret
}
define <2 x i16> @strict_vector_fptoui_v2f32_to_v2i16(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f32_to_v2i16:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f32_to_v2i16:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptoui_v2f32_to_v2i16:
; AVX: # %bb.0:
; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptoui_v2f32_to_v2i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f32_to_v2i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f32_to_v2i16:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f32_to_v2i16:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i16> @llvm.experimental.constrained.fptoui.v2i16.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i16> %ret
}
define <2 x i8> @strict_vector_fptosi_v2f64_to_v2i8(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f64_to_v2i8:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-32-NEXT: packssdw %xmm0, %xmm0
; SSE-32-NEXT: packsswb %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f64_to_v2i8:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-64-NEXT: packssdw %xmm0, %xmm0
; SSE-64-NEXT: packsswb %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v2f64_to_v2i8:
; AVX: # %bb.0:
; AVX-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v2f64_to_v2i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f64_to_v2i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f64_to_v2i8:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f64_to_v2i8:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i8> @llvm.experimental.constrained.fptosi.v2i8.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i8> %ret
}
define <2 x i8> @strict_vector_fptoui_v2f64_to_v2i8(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f64_to_v2i8:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-32-NEXT: packuswb %xmm0, %xmm0
; SSE-32-NEXT: packuswb %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f64_to_v2i8:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttpd2dq %xmm0, %xmm0
; SSE-64-NEXT: packuswb %xmm0, %xmm0
; SSE-64-NEXT: packuswb %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptoui_v2f64_to_v2i8:
; AVX: # %bb.0:
; AVX-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptoui_v2f64_to_v2i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f64_to_v2i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f64_to_v2i8:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f64_to_v2i8:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i8> @llvm.experimental.constrained.fptoui.v2i8.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i8> %ret
}
define <2 x i8> @strict_vector_fptosi_v2f32_to_v2i8(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f32_to_v2i8:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: packssdw %xmm0, %xmm0
; SSE-32-NEXT: packsswb %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f32_to_v2i8:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: packssdw %xmm0, %xmm0
; SSE-64-NEXT: packsswb %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v2f32_to_v2i8:
; AVX: # %bb.0:
; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v2f32_to_v2i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f32_to_v2i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f32_to_v2i8:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f32_to_v2i8:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i8> @llvm.experimental.constrained.fptosi.v2i8.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i8> %ret
}
define <2 x i8> @strict_vector_fptoui_v2f32_to_v2i8(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f32_to_v2i8:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: packuswb %xmm0, %xmm0
; SSE-32-NEXT: packuswb %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f32_to_v2i8:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: packuswb %xmm0, %xmm0
; SSE-64-NEXT: packuswb %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptoui_v2f32_to_v2i8:
; AVX: # %bb.0:
; AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptoui_v2f32_to_v2i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f32_to_v2i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f32_to_v2i8:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f32_to_v2i8:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i8> @llvm.experimental.constrained.fptoui.v2i8.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i8> %ret
}
define <2 x i1> @strict_vector_fptosi_v2f64_to_v2i1(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movhps %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: movlps %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE-32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE-32-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm1
; SSE-64-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovlps %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: vmovhps %xmm0, (%esp)
; AVX-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: fldl (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm1
; AVX-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512F-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k1
; AVX512F-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512VL-NEXT: vptestmd %xmm0, %xmm0, %k1
; AVX512VL-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: vmovdqa64 %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512DQ-NEXT: vpmovd2m %zmm0, %k0
; AVX512DQ-NEXT: vpmovm2q %k0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f64_to_v2i1:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovd2m %xmm0, %k0
; AVX512VLDQ-NEXT: vpmovm2q %k0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i1> @llvm.experimental.constrained.fptosi.v2i1.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i1> %ret
}
define <2 x i1> @strict_vector_fptoui_v2f64_to_v2i1(<2 x double> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE-32-NEXT: comisd %xmm1, %xmm0
; SSE-32-NEXT: movapd %xmm1, %xmm2
; SSE-32-NEXT: jae .LBB19_2
; SSE-32-NEXT: # %bb.1:
; SSE-32-NEXT: xorpd %xmm2, %xmm2
; SSE-32-NEXT: .LBB19_2:
; SSE-32-NEXT: movapd %xmm0, %xmm3
; SSE-32-NEXT: subsd %xmm2, %xmm3
; SSE-32-NEXT: movsd %xmm3, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; SSE-32-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE-32-NEXT: movw %cx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-32-NEXT: comisd %xmm1, %xmm0
; SSE-32-NEXT: jae .LBB19_4
; SSE-32-NEXT: # %bb.3:
; SSE-32-NEXT: xorpd %xmm1, %xmm1
; SSE-32-NEXT: .LBB19_4:
; SSE-32-NEXT: subsd %xmm1, %xmm0
; SSE-32-NEXT: movsd %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %cl
; SSE-32-NEXT: fldl {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %edx
; SSE-32-NEXT: orl $3072, %edx # imm = 0xC00
; SSE-32-NEXT: movw %dx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-32-NEXT: movzbl %cl, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-32-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movsd {{.*#+}} xmm3 = mem[0],zero
; SSE-64-NEXT: comisd %xmm3, %xmm0
; SSE-64-NEXT: xorpd %xmm2, %xmm2
; SSE-64-NEXT: xorpd %xmm1, %xmm1
; SSE-64-NEXT: jb .LBB19_2
; SSE-64-NEXT: # %bb.1:
; SSE-64-NEXT: movapd %xmm3, %xmm1
; SSE-64-NEXT: .LBB19_2:
; SSE-64-NEXT: movapd %xmm0, %xmm4
; SSE-64-NEXT: subsd %xmm1, %xmm4
; SSE-64-NEXT: cvttsd2si %xmm4, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm1
; SSE-64-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
; SSE-64-NEXT: comisd %xmm3, %xmm0
; SSE-64-NEXT: jb .LBB19_4
; SSE-64-NEXT: # %bb.3:
; SSE-64-NEXT: movapd %xmm3, %xmm2
; SSE-64-NEXT: .LBB19_4:
; SSE-64-NEXT: subsd %xmm2, %xmm0
; SSE-64-NEXT: cvttsd2si %xmm0, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
; AVX-32-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX-32-NEXT: vcomisd %xmm1, %xmm2
; AVX-32-NEXT: vmovapd %xmm1, %xmm3
; AVX-32-NEXT: jae .LBB19_2
; AVX-32-NEXT: # %bb.1:
; AVX-32-NEXT: vxorpd %xmm3, %xmm3, %xmm3
; AVX-32-NEXT: .LBB19_2:
; AVX-32-NEXT: vsubsd %xmm3, %xmm2, %xmm2
; AVX-32-NEXT: vmovsd %xmm2, (%esp)
; AVX-32-NEXT: fldl (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %al
; AVX-32-NEXT: movzbl %al, %eax
; AVX-32-NEXT: shll $31, %eax
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX-32-NEXT: vcomisd %xmm1, %xmm0
; AVX-32-NEXT: jae .LBB19_4
; AVX-32-NEXT: # %bb.3:
; AVX-32-NEXT: vxorpd %xmm1, %xmm1, %xmm1
; AVX-32-NEXT: .LBB19_4:
; AVX-32-NEXT: vsubsd %xmm1, %xmm0, %xmm0
; AVX-32-NEXT: vmovsd %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: fldl {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %cl
; AVX-32-NEXT: movzbl %cl, %ecx
; AVX-32-NEXT: shll $31, %ecx
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, (%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX-64-NEXT: vcomisd %xmm1, %xmm0
; AVX-64-NEXT: vxorpd %xmm2, %xmm2, %xmm2
; AVX-64-NEXT: vxorpd %xmm3, %xmm3, %xmm3
; AVX-64-NEXT: jb .LBB19_2
; AVX-64-NEXT: # %bb.1:
; AVX-64-NEXT: vmovapd %xmm1, %xmm3
; AVX-64-NEXT: .LBB19_2:
; AVX-64-NEXT: vsubsd %xmm3, %xmm0, %xmm3
; AVX-64-NEXT: vcvttsd2si %xmm3, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm3
; AVX-64-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX-64-NEXT: vcomisd %xmm1, %xmm0
; AVX-64-NEXT: jb .LBB19_4
; AVX-64-NEXT: # %bb.3:
; AVX-64-NEXT: vmovapd %xmm1, %xmm2
; AVX-64-NEXT: .LBB19_4:
; AVX-64-NEXT: vsubsd %xmm2, %xmm0, %xmm0
; AVX-64-NEXT: vcvttsd2si %xmm0, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovaps %xmm0, %xmm0
; AVX512F-NEXT: vcvttpd2udq %zmm0, %ymm0
; AVX512F-NEXT: vpslld $31, %ymm0, %ymm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k1
; AVX512F-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttpd2udq %xmm0, %xmm0
; AVX512VL-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512VL-NEXT: vptestmd %xmm0, %xmm0, %k1
; AVX512VL-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: vmovdqa64 %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovaps %xmm0, %xmm0
; AVX512DQ-NEXT: vcvttpd2udq %zmm0, %ymm0
; AVX512DQ-NEXT: vpslld $31, %ymm0, %ymm0
; AVX512DQ-NEXT: vpmovd2m %zmm0, %k0
; AVX512DQ-NEXT: vpmovm2q %k0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f64_to_v2i1:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttpd2udq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovd2m %xmm0, %k0
; AVX512VLDQ-NEXT: vpmovm2q %k0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i1> @llvm.experimental.constrained.fptoui.v2i1.v2f64(<2 x double> %a,
metadata !"fpexcept.strict") #0
ret <2 x i1> %ret
}
define <2 x i1> @strict_vector_fptosi_v2f32_to_v2i1(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %eax
; SSE-32-NEXT: orl $3072, %eax # imm = 0xC00
; SSE-32-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE-32-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE-32-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm1
; SSE-64-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: movq %rax, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovss %xmm0, (%esp)
; AVX-32-NEXT: vextractps $1, %xmm0, {{[0-9]+}}(%esp)
; AVX-32-NEXT: flds (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vcvttss2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm1
; AVX-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX-64-NEXT: vcvttss2si %xmm0, %rax
; AVX-64-NEXT: vmovq %rax, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttss2si %xmm0, %eax
; AVX512F-NEXT: andl $1, %eax
; AVX512F-NEXT: kmovw %eax, %k0
; AVX512F-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512F-NEXT: vcvttss2si %xmm0, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kshiftlw $1, %k1, %k1
; AVX512F-NEXT: korw %k1, %k0, %k1
; AVX512F-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttss2si %xmm0, %eax
; AVX512VL-NEXT: andl $1, %eax
; AVX512VL-NEXT: kmovw %eax, %k0
; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512VL-NEXT: vcvttss2si %xmm0, %eax
; AVX512VL-NEXT: kmovw %eax, %k1
; AVX512VL-NEXT: kshiftlw $1, %k1, %k1
; AVX512VL-NEXT: korw %k1, %k0, %k1
; AVX512VL-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: vmovdqa64 %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
; AVX512DQ-NEXT: vcvttss2si %xmm1, %eax
; AVX512DQ-NEXT: kmovw %eax, %k0
; AVX512DQ-NEXT: kshiftlb $1, %k0, %k0
; AVX512DQ-NEXT: vcvttss2si %xmm0, %eax
; AVX512DQ-NEXT: kmovw %eax, %k1
; AVX512DQ-NEXT: kshiftlb $7, %k1, %k1
; AVX512DQ-NEXT: kshiftrb $7, %k1, %k1
; AVX512DQ-NEXT: korw %k0, %k1, %k0
; AVX512DQ-NEXT: vpmovm2q %k0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v2f32_to_v2i1:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
; AVX512VLDQ-NEXT: vcvttss2si %xmm1, %eax
; AVX512VLDQ-NEXT: kmovw %eax, %k0
; AVX512VLDQ-NEXT: kshiftlb $1, %k0, %k0
; AVX512VLDQ-NEXT: vcvttss2si %xmm0, %eax
; AVX512VLDQ-NEXT: kmovw %eax, %k1
; AVX512VLDQ-NEXT: kshiftlb $7, %k1, %k1
; AVX512VLDQ-NEXT: kshiftrb $7, %k1, %k1
; AVX512VLDQ-NEXT: korw %k0, %k1, %k0
; AVX512VLDQ-NEXT: vpmovm2q %k0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i1> @llvm.experimental.constrained.fptosi.v2i1.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i1> %ret
}
define <2 x i1> @strict_vector_fptoui_v2f32_to_v2i1(<2 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; SSE-32: # %bb.0:
; SSE-32-NEXT: pushl %ebp
; SSE-32-NEXT: .cfi_def_cfa_offset 8
; SSE-32-NEXT: .cfi_offset %ebp, -8
; SSE-32-NEXT: movl %esp, %ebp
; SSE-32-NEXT: .cfi_def_cfa_register %ebp
; SSE-32-NEXT: andl $-8, %esp
; SSE-32-NEXT: subl $24, %esp
; SSE-32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE-32-NEXT: comiss %xmm1, %xmm0
; SSE-32-NEXT: movaps %xmm1, %xmm2
; SSE-32-NEXT: jae .LBB21_2
; SSE-32-NEXT: # %bb.1:
; SSE-32-NEXT: xorps %xmm2, %xmm2
; SSE-32-NEXT: .LBB21_2:
; SSE-32-NEXT: movaps %xmm0, %xmm3
; SSE-32-NEXT: subss %xmm2, %xmm3
; SSE-32-NEXT: movss %xmm3, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %al
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; SSE-32-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE-32-NEXT: movw %cx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-32-NEXT: comiss %xmm1, %xmm0
; SSE-32-NEXT: jae .LBB21_4
; SSE-32-NEXT: # %bb.3:
; SSE-32-NEXT: xorps %xmm1, %xmm1
; SSE-32-NEXT: .LBB21_4:
; SSE-32-NEXT: subss %xmm1, %xmm0
; SSE-32-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE-32-NEXT: setae %cl
; SSE-32-NEXT: flds {{[0-9]+}}(%esp)
; SSE-32-NEXT: wait
; SSE-32-NEXT: fnstcw (%esp)
; SSE-32-NEXT: movzwl (%esp), %edx
; SSE-32-NEXT: orl $3072, %edx # imm = 0xC00
; SSE-32-NEXT: movw %dx, {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE-32-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE-32-NEXT: fldcw (%esp)
; SSE-32-NEXT: movzbl %al, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-32-NEXT: movzbl %cl, %eax
; SSE-32-NEXT: shll $31, %eax
; SSE-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; SSE-32-NEXT: movd %eax, %xmm1
; SSE-32-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-32-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-32-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
; SSE-32-NEXT: movl %ebp, %esp
; SSE-32-NEXT: popl %ebp
; SSE-32-NEXT: .cfi_def_cfa %esp, 4
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-64-NEXT: comiss %xmm3, %xmm0
; SSE-64-NEXT: xorps %xmm2, %xmm2
; SSE-64-NEXT: xorps %xmm1, %xmm1
; SSE-64-NEXT: jb .LBB21_2
; SSE-64-NEXT: # %bb.1:
; SSE-64-NEXT: movaps %xmm3, %xmm1
; SSE-64-NEXT: .LBB21_2:
; SSE-64-NEXT: movaps %xmm0, %xmm4
; SSE-64-NEXT: subss %xmm1, %xmm4
; SSE-64-NEXT: cvttss2si %xmm4, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm1
; SSE-64-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; SSE-64-NEXT: comiss %xmm3, %xmm0
; SSE-64-NEXT: jb .LBB21_4
; SSE-64-NEXT: # %bb.3:
; SSE-64-NEXT: movaps %xmm3, %xmm2
; SSE-64-NEXT: .LBB21_4:
; SSE-64-NEXT: subss %xmm2, %xmm0
; SSE-64-NEXT: cvttss2si %xmm0, %rax
; SSE-64-NEXT: setae %cl
; SSE-64-NEXT: movzbl %cl, %ecx
; SSE-64-NEXT: shlq $63, %rcx
; SSE-64-NEXT: xorq %rax, %rcx
; SSE-64-NEXT: movq %rcx, %xmm0
; SSE-64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-64-NEXT: movdqa %xmm1, %xmm0
; SSE-64-NEXT: retq
;
; AVX-32-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; AVX-32: # %bb.0:
; AVX-32-NEXT: pushl %ebp
; AVX-32-NEXT: .cfi_def_cfa_offset 8
; AVX-32-NEXT: .cfi_offset %ebp, -8
; AVX-32-NEXT: movl %esp, %ebp
; AVX-32-NEXT: .cfi_def_cfa_register %ebp
; AVX-32-NEXT: andl $-8, %esp
; AVX-32-NEXT: subl $16, %esp
; AVX-32-NEXT: vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
; AVX-32-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX-32-NEXT: vcomiss %xmm1, %xmm2
; AVX-32-NEXT: vmovaps %xmm1, %xmm3
; AVX-32-NEXT: jae .LBB21_2
; AVX-32-NEXT: # %bb.1:
; AVX-32-NEXT: vxorps %xmm3, %xmm3, %xmm3
; AVX-32-NEXT: .LBB21_2:
; AVX-32-NEXT: vsubss %xmm3, %xmm2, %xmm2
; AVX-32-NEXT: vmovss %xmm2, {{[0-9]+}}(%esp)
; AVX-32-NEXT: flds {{[0-9]+}}(%esp)
; AVX-32-NEXT: fisttpll {{[0-9]+}}(%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %al
; AVX-32-NEXT: movzbl %al, %eax
; AVX-32-NEXT: shll $31, %eax
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %eax
; AVX-32-NEXT: vcomiss %xmm1, %xmm0
; AVX-32-NEXT: jae .LBB21_4
; AVX-32-NEXT: # %bb.3:
; AVX-32-NEXT: vxorps %xmm1, %xmm1, %xmm1
; AVX-32-NEXT: .LBB21_4:
; AVX-32-NEXT: vsubss %xmm1, %xmm0, %xmm0
; AVX-32-NEXT: vmovss %xmm0, (%esp)
; AVX-32-NEXT: flds (%esp)
; AVX-32-NEXT: fisttpll (%esp)
; AVX-32-NEXT: wait
; AVX-32-NEXT: setae %cl
; AVX-32-NEXT: movzbl %cl, %ecx
; AVX-32-NEXT: shll $31, %ecx
; AVX-32-NEXT: xorl {{[0-9]+}}(%esp), %ecx
; AVX-32-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-32-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $2, {{[0-9]+}}(%esp), %xmm0, %xmm0
; AVX-32-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0
; AVX-32-NEXT: movl %ebp, %esp
; AVX-32-NEXT: popl %ebp
; AVX-32-NEXT: .cfi_def_cfa %esp, 4
; AVX-32-NEXT: retl
;
; AVX-64-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; AVX-64: # %bb.0:
; AVX-64-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX-64-NEXT: vcomiss %xmm1, %xmm0
; AVX-64-NEXT: vxorps %xmm2, %xmm2, %xmm2
; AVX-64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; AVX-64-NEXT: jb .LBB21_2
; AVX-64-NEXT: # %bb.1:
; AVX-64-NEXT: vmovaps %xmm1, %xmm3
; AVX-64-NEXT: .LBB21_2:
; AVX-64-NEXT: vsubss %xmm3, %xmm0, %xmm3
; AVX-64-NEXT: vcvttss2si %xmm3, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm3
; AVX-64-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX-64-NEXT: vcomiss %xmm1, %xmm0
; AVX-64-NEXT: jb .LBB21_4
; AVX-64-NEXT: # %bb.3:
; AVX-64-NEXT: vmovaps %xmm1, %xmm2
; AVX-64-NEXT: .LBB21_4:
; AVX-64-NEXT: vsubss %xmm2, %xmm0, %xmm0
; AVX-64-NEXT: vcvttss2si %xmm0, %rax
; AVX-64-NEXT: setae %cl
; AVX-64-NEXT: movzbl %cl, %ecx
; AVX-64-NEXT: shlq $63, %rcx
; AVX-64-NEXT: xorq %rax, %rcx
; AVX-64-NEXT: vmovq %rcx, %xmm0
; AVX-64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm3[0],xmm0[0]
; AVX-64-NEXT: retq
;
; AVX512F-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttss2si %xmm0, %eax
; AVX512F-NEXT: andl $1, %eax
; AVX512F-NEXT: kmovw %eax, %k0
; AVX512F-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512F-NEXT: vcvttss2si %xmm0, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kshiftlw $1, %k1, %k1
; AVX512F-NEXT: korw %k1, %k0, %k1
; AVX512F-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttss2si %xmm0, %eax
; AVX512VL-NEXT: andl $1, %eax
; AVX512VL-NEXT: kmovw %eax, %k0
; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX512VL-NEXT: vcvttss2si %xmm0, %eax
; AVX512VL-NEXT: kmovw %eax, %k1
; AVX512VL-NEXT: kshiftlw $1, %k1, %k1
; AVX512VL-NEXT: korw %k1, %k0, %k1
; AVX512VL-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: vmovdqa64 %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
; AVX512DQ-NEXT: vcvttss2si %xmm1, %eax
; AVX512DQ-NEXT: kmovw %eax, %k0
; AVX512DQ-NEXT: kshiftlb $1, %k0, %k0
; AVX512DQ-NEXT: vcvttss2si %xmm0, %eax
; AVX512DQ-NEXT: kmovw %eax, %k1
; AVX512DQ-NEXT: kshiftlb $7, %k1, %k1
; AVX512DQ-NEXT: kshiftrb $7, %k1, %k1
; AVX512DQ-NEXT: korw %k0, %k1, %k0
; AVX512DQ-NEXT: vpmovm2q %k0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v2f32_to_v2i1:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
; AVX512VLDQ-NEXT: vcvttss2si %xmm1, %eax
; AVX512VLDQ-NEXT: kmovw %eax, %k0
; AVX512VLDQ-NEXT: kshiftlb $1, %k0, %k0
; AVX512VLDQ-NEXT: vcvttss2si %xmm0, %eax
; AVX512VLDQ-NEXT: kmovw %eax, %k1
; AVX512VLDQ-NEXT: kshiftlb $7, %k1, %k1
; AVX512VLDQ-NEXT: kshiftrb $7, %k1, %k1
; AVX512VLDQ-NEXT: korw %k0, %k1, %k0
; AVX512VLDQ-NEXT: vpmovm2q %k0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <2 x i1> @llvm.experimental.constrained.fptoui.v2i1.v2f32(<2 x float> %a,
metadata !"fpexcept.strict") #0
ret <2 x i1> %ret
}
define <4 x i32> @strict_vector_fptosi_v4f32_to_v4i32(<4 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v4f32_to_v4i32:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v4f32_to_v4i32:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v4f32_to_v4i32:
; AVX: # %bb.0:
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v4f32_to_v4i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v4f32_to_v4i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v4f32_to_v4i32:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v4f32_to_v4i32:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <4 x i32> @llvm.experimental.constrained.fptosi.v4i32.v4f32(<4 x float> %a,
metadata !"fpexcept.strict") #0
ret <4 x i32> %ret
}
define <4 x i32> @strict_vector_fptoui_v4f32_to_v4i32(<4 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v4f32_to_v4i32:
; SSE-32: # %bb.0:
; SSE-32-NEXT: movaps {{.*#+}} xmm1 = [2.14748365E+9,2.14748365E+9,2.14748365E+9,2.14748365E+9]
; SSE-32-NEXT: movaps %xmm0, %xmm2
; SSE-32-NEXT: cmpltps %xmm1, %xmm2
; SSE-32-NEXT: movaps %xmm2, %xmm3
; SSE-32-NEXT: andnps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm3
; SSE-32-NEXT: andnps %xmm1, %xmm2
; SSE-32-NEXT: subps %xmm2, %xmm0
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: xorps %xmm3, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v4f32_to_v4i32:
; SSE-64: # %bb.0:
; SSE-64-NEXT: movaps {{.*#+}} xmm1 = [2.14748365E+9,2.14748365E+9,2.14748365E+9,2.14748365E+9]
; SSE-64-NEXT: movaps %xmm0, %xmm2
; SSE-64-NEXT: cmpltps %xmm1, %xmm2
; SSE-64-NEXT: movaps %xmm2, %xmm3
; SSE-64-NEXT: andnps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
; SSE-64-NEXT: andnps %xmm1, %xmm2
; SSE-64-NEXT: subps %xmm2, %xmm0
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: xorps %xmm3, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptoui_v4f32_to_v4i32:
; AVX: # %bb.0:
; AVX-NEXT: vmovaps {{.*#+}} xmm1 = [2.14748365E+9,2.14748365E+9,2.14748365E+9,2.14748365E+9]
; AVX-NEXT: vcmpltps %xmm1, %xmm0, %xmm2
; AVX-NEXT: vxorps %xmm3, %xmm3, %xmm3
; AVX-NEXT: vmovaps {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
; AVX-NEXT: vblendvps %xmm2, %xmm3, %xmm4, %xmm4
; AVX-NEXT: vblendvps %xmm2, %xmm3, %xmm1, %xmm1
; AVX-NEXT: vsubps %xmm1, %xmm0, %xmm0
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: vxorps %xmm4, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptoui_v4f32_to_v4i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovaps %xmm0, %xmm0
; AVX512F-NEXT: vcvttps2udq %zmm0, %zmm0
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v4f32_to_v4i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttps2udq %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v4f32_to_v4i32:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovaps %xmm0, %xmm0
; AVX512DQ-NEXT: vcvttps2udq %zmm0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v4f32_to_v4i32:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2udq %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <4 x i32> @llvm.experimental.constrained.fptoui.v4i32.v4f32(<4 x float> %a,
metadata !"fpexcept.strict") #0
ret <4 x i32> %ret
}
define <4 x i8> @strict_vector_fptosi_v4f32_to_v4i8(<4 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v4f32_to_v4i8:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: packssdw %xmm0, %xmm0
; SSE-32-NEXT: packsswb %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v4f32_to_v4i8:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: packssdw %xmm0, %xmm0
; SSE-64-NEXT: packsswb %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v4f32_to_v4i8:
; AVX: # %bb.0:
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v4f32_to_v4i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v4f32_to_v4i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v4f32_to_v4i8:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v4f32_to_v4i8:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <4 x i8> @llvm.experimental.constrained.fptosi.v4i8.v4f32(<4 x float> %a,
metadata !"fpexcept.strict") #0
ret <4 x i8> %ret
}
define <4 x i8> @strict_vector_fptoui_v4f32_to_v4i8(<4 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v4f32_to_v4i8:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: packuswb %xmm0, %xmm0
; SSE-32-NEXT: packuswb %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v4f32_to_v4i8:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: packuswb %xmm0, %xmm0
; SSE-64-NEXT: packuswb %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptoui_v4f32_to_v4i8:
; AVX: # %bb.0:
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptoui_v4f32_to_v4i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v4f32_to_v4i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v4f32_to_v4i8:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v4f32_to_v4i8:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovdb %xmm0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <4 x i8> @llvm.experimental.constrained.fptoui.v4i8.v4f32(<4 x float> %a,
metadata !"fpexcept.strict") #0
ret <4 x i8> %ret
}
define <4 x i1> @strict_vector_fptosi_v4f32_to_v4i1(<4 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
; AVX: # %bb.0:
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k1
; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vptestmd %xmm0, %xmm0, %k1
; AVX512VL-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: vmovdqa32 %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpmovd2m %zmm0, %k0
; AVX512DQ-NEXT: vpmovm2d %k0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptosi_v4f32_to_v4i1:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovd2m %xmm0, %k0
; AVX512VLDQ-NEXT: vpmovm2d %k0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <4 x i1> @llvm.experimental.constrained.fptosi.v4i1.v4f32(<4 x float> %a,
metadata !"fpexcept.strict") #0
ret <4 x i1> %ret
}
define <4 x i1> @strict_vector_fptoui_v4f32_to_v4i1(<4 x float> %a) #0 {
; SSE-32-LABEL: strict_vector_fptoui_v4f32_to_v4i1:
; SSE-32: # %bb.0:
; SSE-32-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-32-NEXT: retl
;
; SSE-64-LABEL: strict_vector_fptoui_v4f32_to_v4i1:
; SSE-64: # %bb.0:
; SSE-64-NEXT: cvttps2dq %xmm0, %xmm0
; SSE-64-NEXT: retq
;
; AVX-LABEL: strict_vector_fptoui_v4f32_to_v4i1:
; AVX: # %bb.0:
; AVX-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX-NEXT: ret{{[l|q]}}
;
; AVX512F-LABEL: strict_vector_fptoui_v4f32_to_v4i1:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512F-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k1
; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: ret{{[l|q]}}
;
; AVX512VL-LABEL: strict_vector_fptoui_v4f32_to_v4i1:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VL-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512VL-NEXT: vptestmd %xmm0, %xmm0, %k1
; AVX512VL-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: vmovdqa32 %xmm0, %xmm0 {%k1} {z}
; AVX512VL-NEXT: ret{{[l|q]}}
;
; AVX512DQ-LABEL: strict_vector_fptoui_v4f32_to_v4i1:
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512DQ-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512DQ-NEXT: vpmovd2m %zmm0, %k0
; AVX512DQ-NEXT: vpmovm2d %k0, %zmm0
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: ret{{[l|q]}}
;
; AVX512VLDQ-LABEL: strict_vector_fptoui_v4f32_to_v4i1:
; AVX512VLDQ: # %bb.0:
; AVX512VLDQ-NEXT: vcvttps2dq %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpslld $31, %xmm0, %xmm0
; AVX512VLDQ-NEXT: vpmovd2m %xmm0, %k0
; AVX512VLDQ-NEXT: vpmovm2d %k0, %xmm0
; AVX512VLDQ-NEXT: ret{{[l|q]}}
%ret = call <4 x i1> @llvm.experimental.constrained.fptoui.v4i1.v4f32(<4 x float> %a,
metadata !"fpexcept.strict") #0
ret <4 x i1> %ret
}
attributes #0 = { strictfp }