blob: 158bab5f2f1d5e3914cf0e8a6a4e9499635b5405 [file] [log] [blame]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i386-pc-windows-msvc -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,X32,AVX512_32,AVX512_32_WIN,AVX512DQVL_32_WIN
; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,X32,AVX512_32,AVX512_32_LIN,AVX512DQVL_32_LIN
; RUN: llc < %s -mtriple=x86_64-pc-windows-msvc -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,X64,AVX512_64,AVX512_64_WIN,AVX512DQVL_64_WIN
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,X32,AVX512_64,AVX512_64_LIN,AVX512DQVL_64_LIN
; RUN: llc < %s -mtriple=i386-pc-windows-msvc -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,X32,AVX512_32,AVX512_32_WIN,AVX512DQ_32_WIN
; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,X32,AVX512_32,AVX512_32_LIN,AVX512DQ_32_LIN
; RUN: llc < %s -mtriple=x86_64-pc-windows-msvc -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,X64,AVX512_64,AVX512_64_WIN,AVX512DQ_64_WIN
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,X32,AVX512_64,AVX512_64_LIN,AVX512DQ_64_LIN
; RUN: llc < %s -mtriple=i386-pc-windows-msvc -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,X32,AVX512_32,AVX512_32_WIN,AVX512F_32_WIN
; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,X32,AVX512_32,AVX512_32_LIN,AVX512F_32_LIN
; RUN: llc < %s -mtriple=x86_64-pc-windows-msvc -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,X64,AVX512_64,AVX512_64_WIN,AVX512F_64_WIN
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,X32,AVX512_64,AVX512_64_LIN,AVX512F_64_LIN
; RUN: llc < %s -mtriple=i386-pc-windows-msvc -mattr=+sse3 | FileCheck %s --check-prefixes=CHECK,X32,SSE3_32,SSE3_32_WIN
; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mattr=+sse3 | FileCheck %s --check-prefixes=CHECK,X32,SSE3_32,SSE3_32_LIN
; RUN: llc < %s -mtriple=x86_64-pc-windows-msvc -mattr=+sse3 | FileCheck %s --check-prefixes=CHECK,X64,SSE3_64,SSE3_64_WIN
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse3 | FileCheck %s --check-prefixes=CHECK,X64,SSE3_64,SSE3_64_LIN
; RUN: llc < %s -mtriple=i386-pc-windows-msvc -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X32,SSE2_32,SSE2_32_WIN
; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X32,SSE2_32,SSE2_32_LIN
; RUN: llc < %s -mtriple=x86_64-pc-windows-msvc -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X64,SSE2_64,SSE2_64_WIN
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X64,SSE2_64,SSE2_64_LIN
; RUN: llc < %s -mtriple=i386-pc-windows-msvc -mattr=-sse | FileCheck %s --check-prefixes=CHECK,X32,X87,X87_WIN
; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mattr=-sse | FileCheck %s --check-prefixes=CHECK,X32,X87,X87_LIN
; Check that scalar FP conversions to signed and unsigned int64 are using
; reasonable sequences, across platforms and target switches.
;
; The signed case is straight forward, and the tests here basically
; ensure successful compilation (f80 with avx512 was broken at one point).
;
; For the unsigned case there are many possible sequences, so to avoid
; a fragile test we just check for the presence of a few key instructions.
; AVX512 on Intel64 can use vcvtts[ds]2usi directly for float and double.
; Otherwise the sequence will involve an FP subtract (fsub, subss or subsd),
; and a truncating conversion (cvtts[ds]2si, fisttp, or fnstcw+fist). When
; both a subtract and fnstcw are needed, they can occur in either order.
;
; The interesting subtargets are AVX512F (vcvtts[ds]2usi), SSE3 (fisttp),
; SSE2 (cvtts[ds]2si) and vanilla X87 (fnstcw+fist, 32-bit only).
define i64 @f_to_u64(float %a) nounwind {
; AVX512DQVL_32_WIN-LABEL: f_to_u64:
; AVX512DQVL_32_WIN: # %bb.0:
; AVX512DQVL_32_WIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQVL_32_WIN-NEXT: vcvttps2uqq %xmm0, %ymm0
; AVX512DQVL_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_WIN-NEXT: vzeroupper
; AVX512DQVL_32_WIN-NEXT: retl
;
; AVX512DQVL_32_LIN-LABEL: f_to_u64:
; AVX512DQVL_32_LIN: # %bb.0:
; AVX512DQVL_32_LIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQVL_32_LIN-NEXT: vcvttps2uqq %xmm0, %ymm0
; AVX512DQVL_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_LIN-NEXT: vzeroupper
; AVX512DQVL_32_LIN-NEXT: retl
;
; AVX512_64-LABEL: f_to_u64:
; AVX512_64: # %bb.0:
; AVX512_64-NEXT: vcvttss2usi %xmm0, %rax
; AVX512_64-NEXT: retq
;
; AVX512DQ_32_WIN-LABEL: f_to_u64:
; AVX512DQ_32_WIN: # %bb.0:
; AVX512DQ_32_WIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQ_32_WIN-NEXT: vcvttps2uqq %ymm0, %zmm0
; AVX512DQ_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_WIN-NEXT: vzeroupper
; AVX512DQ_32_WIN-NEXT: retl
;
; AVX512DQ_32_LIN-LABEL: f_to_u64:
; AVX512DQ_32_LIN: # %bb.0:
; AVX512DQ_32_LIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQ_32_LIN-NEXT: vcvttps2uqq %ymm0, %zmm0
; AVX512DQ_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_LIN-NEXT: vzeroupper
; AVX512DQ_32_LIN-NEXT: retl
;
; AVX512F_32_WIN-LABEL: f_to_u64:
; AVX512F_32_WIN: # %bb.0:
; AVX512F_32_WIN-NEXT: pushl %ebp
; AVX512F_32_WIN-NEXT: movl %esp, %ebp
; AVX512F_32_WIN-NEXT: andl $-8, %esp
; AVX512F_32_WIN-NEXT: subl $8, %esp
; AVX512F_32_WIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F_32_WIN-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX512F_32_WIN-NEXT: vcmpltss %xmm1, %xmm0, %k1
; AVX512F_32_WIN-NEXT: vsubss %xmm1, %xmm0, %xmm2
; AVX512F_32_WIN-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1}
; AVX512F_32_WIN-NEXT: vmovss %xmm2, (%esp)
; AVX512F_32_WIN-NEXT: flds (%esp)
; AVX512F_32_WIN-NEXT: fisttpll (%esp)
; AVX512F_32_WIN-NEXT: xorl %edx, %edx
; AVX512F_32_WIN-NEXT: vucomiss %xmm0, %xmm1
; AVX512F_32_WIN-NEXT: setbe %dl
; AVX512F_32_WIN-NEXT: shll $31, %edx
; AVX512F_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; AVX512F_32_WIN-NEXT: movl (%esp), %eax
; AVX512F_32_WIN-NEXT: movl %ebp, %esp
; AVX512F_32_WIN-NEXT: popl %ebp
; AVX512F_32_WIN-NEXT: retl
;
; AVX512F_32_LIN-LABEL: f_to_u64:
; AVX512F_32_LIN: # %bb.0:
; AVX512F_32_LIN-NEXT: subl $12, %esp
; AVX512F_32_LIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F_32_LIN-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX512F_32_LIN-NEXT: vcmpltss %xmm1, %xmm0, %k1
; AVX512F_32_LIN-NEXT: vsubss %xmm1, %xmm0, %xmm2
; AVX512F_32_LIN-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1}
; AVX512F_32_LIN-NEXT: vmovss %xmm2, (%esp)
; AVX512F_32_LIN-NEXT: flds (%esp)
; AVX512F_32_LIN-NEXT: fisttpll (%esp)
; AVX512F_32_LIN-NEXT: xorl %edx, %edx
; AVX512F_32_LIN-NEXT: vucomiss %xmm0, %xmm1
; AVX512F_32_LIN-NEXT: setbe %dl
; AVX512F_32_LIN-NEXT: shll $31, %edx
; AVX512F_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; AVX512F_32_LIN-NEXT: movl (%esp), %eax
; AVX512F_32_LIN-NEXT: addl $12, %esp
; AVX512F_32_LIN-NEXT: retl
;
; SSE3_32_WIN-LABEL: f_to_u64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl %ebp
; SSE3_32_WIN-NEXT: movl %esp, %ebp
; SSE3_32_WIN-NEXT: andl $-8, %esp
; SSE3_32_WIN-NEXT: subl $8, %esp
; SSE3_32_WIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE3_32_WIN-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE3_32_WIN-NEXT: movaps %xmm0, %xmm2
; SSE3_32_WIN-NEXT: cmpltss %xmm1, %xmm2
; SSE3_32_WIN-NEXT: movaps %xmm2, %xmm3
; SSE3_32_WIN-NEXT: andps %xmm0, %xmm2
; SSE3_32_WIN-NEXT: xorl %edx, %edx
; SSE3_32_WIN-NEXT: ucomiss %xmm0, %xmm1
; SSE3_32_WIN-NEXT: subss %xmm1, %xmm0
; SSE3_32_WIN-NEXT: andnps %xmm0, %xmm3
; SSE3_32_WIN-NEXT: orps %xmm3, %xmm2
; SSE3_32_WIN-NEXT: movss %xmm2, (%esp)
; SSE3_32_WIN-NEXT: flds (%esp)
; SSE3_32_WIN-NEXT: fisttpll (%esp)
; SSE3_32_WIN-NEXT: setbe %dl
; SSE3_32_WIN-NEXT: shll $31, %edx
; SSE3_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE3_32_WIN-NEXT: movl (%esp), %eax
; SSE3_32_WIN-NEXT: movl %ebp, %esp
; SSE3_32_WIN-NEXT: popl %ebp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: f_to_u64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE3_32_LIN-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE3_32_LIN-NEXT: movaps %xmm0, %xmm2
; SSE3_32_LIN-NEXT: cmpltss %xmm1, %xmm2
; SSE3_32_LIN-NEXT: movaps %xmm2, %xmm3
; SSE3_32_LIN-NEXT: andps %xmm0, %xmm2
; SSE3_32_LIN-NEXT: xorl %edx, %edx
; SSE3_32_LIN-NEXT: ucomiss %xmm0, %xmm1
; SSE3_32_LIN-NEXT: subss %xmm1, %xmm0
; SSE3_32_LIN-NEXT: andnps %xmm0, %xmm3
; SSE3_32_LIN-NEXT: orps %xmm3, %xmm2
; SSE3_32_LIN-NEXT: movss %xmm2, (%esp)
; SSE3_32_LIN-NEXT: flds (%esp)
; SSE3_32_LIN-NEXT: fisttpll (%esp)
; SSE3_32_LIN-NEXT: setbe %dl
; SSE3_32_LIN-NEXT: shll $31, %edx
; SSE3_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE3_32_LIN-NEXT: movl (%esp), %eax
; SSE3_32_LIN-NEXT: addl $12, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64-LABEL: f_to_u64:
; SSE3_64: # %bb.0:
; SSE3_64-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE3_64-NEXT: movaps %xmm0, %xmm2
; SSE3_64-NEXT: subss %xmm1, %xmm2
; SSE3_64-NEXT: cvttss2si %xmm2, %rax
; SSE3_64-NEXT: movabsq $-9223372036854775808, %rcx # imm = 0x8000000000000000
; SSE3_64-NEXT: xorq %rax, %rcx
; SSE3_64-NEXT: cvttss2si %xmm0, %rax
; SSE3_64-NEXT: ucomiss %xmm1, %xmm0
; SSE3_64-NEXT: cmovaeq %rcx, %rax
; SSE3_64-NEXT: retq
;
; SSE2_32_WIN-LABEL: f_to_u64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl %ebp
; SSE2_32_WIN-NEXT: movl %esp, %ebp
; SSE2_32_WIN-NEXT: andl $-8, %esp
; SSE2_32_WIN-NEXT: subl $16, %esp
; SSE2_32_WIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE2_32_WIN-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE2_32_WIN-NEXT: movaps %xmm0, %xmm2
; SSE2_32_WIN-NEXT: subss %xmm1, %xmm2
; SSE2_32_WIN-NEXT: movaps %xmm0, %xmm3
; SSE2_32_WIN-NEXT: cmpltss %xmm1, %xmm3
; SSE2_32_WIN-NEXT: movaps %xmm3, %xmm4
; SSE2_32_WIN-NEXT: andnps %xmm2, %xmm4
; SSE2_32_WIN-NEXT: andps %xmm0, %xmm3
; SSE2_32_WIN-NEXT: orps %xmm4, %xmm3
; SSE2_32_WIN-NEXT: movss %xmm3, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: flds {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: xorl %edx, %edx
; SSE2_32_WIN-NEXT: ucomiss %xmm0, %xmm1
; SSE2_32_WIN-NEXT: setbe %dl
; SSE2_32_WIN-NEXT: shll $31, %edx
; SSE2_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: movl %ebp, %esp
; SSE2_32_WIN-NEXT: popl %ebp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: f_to_u64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $20, %esp
; SSE2_32_LIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE2_32_LIN-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE2_32_LIN-NEXT: movaps %xmm0, %xmm2
; SSE2_32_LIN-NEXT: subss %xmm1, %xmm2
; SSE2_32_LIN-NEXT: movaps %xmm0, %xmm3
; SSE2_32_LIN-NEXT: cmpltss %xmm1, %xmm3
; SSE2_32_LIN-NEXT: movaps %xmm3, %xmm4
; SSE2_32_LIN-NEXT: andnps %xmm2, %xmm4
; SSE2_32_LIN-NEXT: andps %xmm0, %xmm3
; SSE2_32_LIN-NEXT: orps %xmm4, %xmm3
; SSE2_32_LIN-NEXT: movss %xmm3, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: flds {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: xorl %edx, %edx
; SSE2_32_LIN-NEXT: ucomiss %xmm0, %xmm1
; SSE2_32_LIN-NEXT: setbe %dl
; SSE2_32_LIN-NEXT: shll $31, %edx
; SSE2_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: addl $20, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64-LABEL: f_to_u64:
; SSE2_64: # %bb.0:
; SSE2_64-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; SSE2_64-NEXT: movaps %xmm0, %xmm2
; SSE2_64-NEXT: subss %xmm1, %xmm2
; SSE2_64-NEXT: cvttss2si %xmm2, %rax
; SSE2_64-NEXT: movabsq $-9223372036854775808, %rcx # imm = 0x8000000000000000
; SSE2_64-NEXT: xorq %rax, %rcx
; SSE2_64-NEXT: cvttss2si %xmm0, %rax
; SSE2_64-NEXT: ucomiss %xmm1, %xmm0
; SSE2_64-NEXT: cmovaeq %rcx, %rax
; SSE2_64-NEXT: retq
;
; X87_WIN-LABEL: f_to_u64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl %ebp
; X87_WIN-NEXT: movl %esp, %ebp
; X87_WIN-NEXT: andl $-8, %esp
; X87_WIN-NEXT: subl $16, %esp
; X87_WIN-NEXT: flds 8(%ebp)
; X87_WIN-NEXT: flds __real@5f000000
; X87_WIN-NEXT: fld %st(1)
; X87_WIN-NEXT: fsub %st(1), %st
; X87_WIN-NEXT: fxch %st(1)
; X87_WIN-NEXT: fucomp %st(2)
; X87_WIN-NEXT: fnstsw %ax
; X87_WIN-NEXT: # kill: def $ah killed $ah killed $ax
; X87_WIN-NEXT: sahf
; X87_WIN-NEXT: ja LBB0_2
; X87_WIN-NEXT: # %bb.1:
; X87_WIN-NEXT: fstp %st(1)
; X87_WIN-NEXT: fldz
; X87_WIN-NEXT: LBB0_2:
; X87_WIN-NEXT: fstp %st(0)
; X87_WIN-NEXT: setbe %al
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_WIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzbl %al, %edx
; X87_WIN-NEXT: shll $31, %edx
; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: movl %ebp, %esp
; X87_WIN-NEXT: popl %ebp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: f_to_u64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $20, %esp
; X87_LIN-NEXT: flds {{[0-9]+}}(%esp)
; X87_LIN-NEXT: flds {{\.LCPI.*}}
; X87_LIN-NEXT: fld %st(1)
; X87_LIN-NEXT: fsub %st(1), %st
; X87_LIN-NEXT: fxch %st(1)
; X87_LIN-NEXT: fucomp %st(2)
; X87_LIN-NEXT: fnstsw %ax
; X87_LIN-NEXT: # kill: def $ah killed $ah killed $ax
; X87_LIN-NEXT: sahf
; X87_LIN-NEXT: ja .LBB0_2
; X87_LIN-NEXT: # %bb.1:
; X87_LIN-NEXT: fstp %st(1)
; X87_LIN-NEXT: fldz
; X87_LIN-NEXT: .LBB0_2:
; X87_LIN-NEXT: fstp %st(0)
; X87_LIN-NEXT: setbe %al
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_LIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzbl %al, %edx
; X87_LIN-NEXT: shll $31, %edx
; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: addl $20, %esp
; X87_LIN-NEXT: retl
%r = fptoui float %a to i64
ret i64 %r
}
define i64 @f_to_s64(float %a) nounwind {
; AVX512DQVL_32_WIN-LABEL: f_to_s64:
; AVX512DQVL_32_WIN: # %bb.0:
; AVX512DQVL_32_WIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQVL_32_WIN-NEXT: vcvttps2qq %xmm0, %ymm0
; AVX512DQVL_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_WIN-NEXT: vzeroupper
; AVX512DQVL_32_WIN-NEXT: retl
;
; AVX512DQVL_32_LIN-LABEL: f_to_s64:
; AVX512DQVL_32_LIN: # %bb.0:
; AVX512DQVL_32_LIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQVL_32_LIN-NEXT: vcvttps2qq %xmm0, %ymm0
; AVX512DQVL_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_LIN-NEXT: vzeroupper
; AVX512DQVL_32_LIN-NEXT: retl
;
; AVX512_64-LABEL: f_to_s64:
; AVX512_64: # %bb.0:
; AVX512_64-NEXT: vcvttss2si %xmm0, %rax
; AVX512_64-NEXT: retq
;
; AVX512DQ_32_WIN-LABEL: f_to_s64:
; AVX512DQ_32_WIN: # %bb.0:
; AVX512DQ_32_WIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQ_32_WIN-NEXT: vcvttps2qq %ymm0, %zmm0
; AVX512DQ_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_WIN-NEXT: vzeroupper
; AVX512DQ_32_WIN-NEXT: retl
;
; AVX512DQ_32_LIN-LABEL: f_to_s64:
; AVX512DQ_32_LIN: # %bb.0:
; AVX512DQ_32_LIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512DQ_32_LIN-NEXT: vcvttps2qq %ymm0, %zmm0
; AVX512DQ_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_LIN-NEXT: vzeroupper
; AVX512DQ_32_LIN-NEXT: retl
;
; AVX512F_32_WIN-LABEL: f_to_s64:
; AVX512F_32_WIN: # %bb.0:
; AVX512F_32_WIN-NEXT: pushl %ebp
; AVX512F_32_WIN-NEXT: movl %esp, %ebp
; AVX512F_32_WIN-NEXT: andl $-8, %esp
; AVX512F_32_WIN-NEXT: subl $8, %esp
; AVX512F_32_WIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F_32_WIN-NEXT: vmovss %xmm0, (%esp)
; AVX512F_32_WIN-NEXT: flds (%esp)
; AVX512F_32_WIN-NEXT: fisttpll (%esp)
; AVX512F_32_WIN-NEXT: movl (%esp), %eax
; AVX512F_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; AVX512F_32_WIN-NEXT: movl %ebp, %esp
; AVX512F_32_WIN-NEXT: popl %ebp
; AVX512F_32_WIN-NEXT: retl
;
; AVX512F_32_LIN-LABEL: f_to_s64:
; AVX512F_32_LIN: # %bb.0:
; AVX512F_32_LIN-NEXT: subl $12, %esp
; AVX512F_32_LIN-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX512F_32_LIN-NEXT: vmovss %xmm0, (%esp)
; AVX512F_32_LIN-NEXT: flds (%esp)
; AVX512F_32_LIN-NEXT: fisttpll (%esp)
; AVX512F_32_LIN-NEXT: movl (%esp), %eax
; AVX512F_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; AVX512F_32_LIN-NEXT: addl $12, %esp
; AVX512F_32_LIN-NEXT: retl
;
; SSE3_32_WIN-LABEL: f_to_s64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl %ebp
; SSE3_32_WIN-NEXT: movl %esp, %ebp
; SSE3_32_WIN-NEXT: andl $-8, %esp
; SSE3_32_WIN-NEXT: subl $8, %esp
; SSE3_32_WIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE3_32_WIN-NEXT: movss %xmm0, (%esp)
; SSE3_32_WIN-NEXT: flds (%esp)
; SSE3_32_WIN-NEXT: fisttpll (%esp)
; SSE3_32_WIN-NEXT: movl (%esp), %eax
; SSE3_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE3_32_WIN-NEXT: movl %ebp, %esp
; SSE3_32_WIN-NEXT: popl %ebp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: f_to_s64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE3_32_LIN-NEXT: movss %xmm0, (%esp)
; SSE3_32_LIN-NEXT: flds (%esp)
; SSE3_32_LIN-NEXT: fisttpll (%esp)
; SSE3_32_LIN-NEXT: movl (%esp), %eax
; SSE3_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE3_32_LIN-NEXT: addl $12, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64-LABEL: f_to_s64:
; SSE3_64: # %bb.0:
; SSE3_64-NEXT: cvttss2si %xmm0, %rax
; SSE3_64-NEXT: retq
;
; SSE2_32_WIN-LABEL: f_to_s64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl %ebp
; SSE2_32_WIN-NEXT: movl %esp, %ebp
; SSE2_32_WIN-NEXT: andl $-8, %esp
; SSE2_32_WIN-NEXT: subl $16, %esp
; SSE2_32_WIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE2_32_WIN-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: flds {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE2_32_WIN-NEXT: movl %ebp, %esp
; SSE2_32_WIN-NEXT: popl %ebp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: f_to_s64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $20, %esp
; SSE2_32_LIN-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; SSE2_32_LIN-NEXT: movss %xmm0, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: flds {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE2_32_LIN-NEXT: addl $20, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64-LABEL: f_to_s64:
; SSE2_64: # %bb.0:
; SSE2_64-NEXT: cvttss2si %xmm0, %rax
; SSE2_64-NEXT: retq
;
; X87_WIN-LABEL: f_to_s64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl %ebp
; X87_WIN-NEXT: movl %esp, %ebp
; X87_WIN-NEXT: andl $-8, %esp
; X87_WIN-NEXT: subl $16, %esp
; X87_WIN-NEXT: flds 8(%ebp)
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; X87_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl %ebp, %esp
; X87_WIN-NEXT: popl %ebp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: f_to_s64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $20, %esp
; X87_LIN-NEXT: flds {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; X87_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: addl $20, %esp
; X87_LIN-NEXT: retl
%r = fptosi float %a to i64
ret i64 %r
}
define i64 @d_to_u64(double %a) nounwind {
; AVX512DQVL_32_WIN-LABEL: d_to_u64:
; AVX512DQVL_32_WIN: # %bb.0:
; AVX512DQVL_32_WIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQVL_32_WIN-NEXT: vcvttpd2uqq %ymm0, %ymm0
; AVX512DQVL_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_WIN-NEXT: vzeroupper
; AVX512DQVL_32_WIN-NEXT: retl
;
; AVX512DQVL_32_LIN-LABEL: d_to_u64:
; AVX512DQVL_32_LIN: # %bb.0:
; AVX512DQVL_32_LIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQVL_32_LIN-NEXT: vcvttpd2uqq %ymm0, %ymm0
; AVX512DQVL_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_LIN-NEXT: vzeroupper
; AVX512DQVL_32_LIN-NEXT: retl
;
; AVX512_64-LABEL: d_to_u64:
; AVX512_64: # %bb.0:
; AVX512_64-NEXT: vcvttsd2usi %xmm0, %rax
; AVX512_64-NEXT: retq
;
; AVX512DQ_32_WIN-LABEL: d_to_u64:
; AVX512DQ_32_WIN: # %bb.0:
; AVX512DQ_32_WIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ_32_WIN-NEXT: vcvttpd2uqq %zmm0, %zmm0
; AVX512DQ_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_WIN-NEXT: vzeroupper
; AVX512DQ_32_WIN-NEXT: retl
;
; AVX512DQ_32_LIN-LABEL: d_to_u64:
; AVX512DQ_32_LIN: # %bb.0:
; AVX512DQ_32_LIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ_32_LIN-NEXT: vcvttpd2uqq %zmm0, %zmm0
; AVX512DQ_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_LIN-NEXT: vzeroupper
; AVX512DQ_32_LIN-NEXT: retl
;
; AVX512F_32_WIN-LABEL: d_to_u64:
; AVX512F_32_WIN: # %bb.0:
; AVX512F_32_WIN-NEXT: pushl %ebp
; AVX512F_32_WIN-NEXT: movl %esp, %ebp
; AVX512F_32_WIN-NEXT: andl $-8, %esp
; AVX512F_32_WIN-NEXT: subl $8, %esp
; AVX512F_32_WIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512F_32_WIN-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX512F_32_WIN-NEXT: vcmpltsd %xmm1, %xmm0, %k1
; AVX512F_32_WIN-NEXT: vsubsd %xmm1, %xmm0, %xmm2
; AVX512F_32_WIN-NEXT: vmovsd %xmm0, %xmm2, %xmm2 {%k1}
; AVX512F_32_WIN-NEXT: vmovsd %xmm2, (%esp)
; AVX512F_32_WIN-NEXT: fldl (%esp)
; AVX512F_32_WIN-NEXT: fisttpll (%esp)
; AVX512F_32_WIN-NEXT: xorl %edx, %edx
; AVX512F_32_WIN-NEXT: vucomisd %xmm0, %xmm1
; AVX512F_32_WIN-NEXT: setbe %dl
; AVX512F_32_WIN-NEXT: shll $31, %edx
; AVX512F_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; AVX512F_32_WIN-NEXT: movl (%esp), %eax
; AVX512F_32_WIN-NEXT: movl %ebp, %esp
; AVX512F_32_WIN-NEXT: popl %ebp
; AVX512F_32_WIN-NEXT: retl
;
; AVX512F_32_LIN-LABEL: d_to_u64:
; AVX512F_32_LIN: # %bb.0:
; AVX512F_32_LIN-NEXT: subl $12, %esp
; AVX512F_32_LIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512F_32_LIN-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
; AVX512F_32_LIN-NEXT: vcmpltsd %xmm1, %xmm0, %k1
; AVX512F_32_LIN-NEXT: vsubsd %xmm1, %xmm0, %xmm2
; AVX512F_32_LIN-NEXT: vmovsd %xmm0, %xmm2, %xmm2 {%k1}
; AVX512F_32_LIN-NEXT: vmovsd %xmm2, (%esp)
; AVX512F_32_LIN-NEXT: fldl (%esp)
; AVX512F_32_LIN-NEXT: fisttpll (%esp)
; AVX512F_32_LIN-NEXT: xorl %edx, %edx
; AVX512F_32_LIN-NEXT: vucomisd %xmm0, %xmm1
; AVX512F_32_LIN-NEXT: setbe %dl
; AVX512F_32_LIN-NEXT: shll $31, %edx
; AVX512F_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; AVX512F_32_LIN-NEXT: movl (%esp), %eax
; AVX512F_32_LIN-NEXT: addl $12, %esp
; AVX512F_32_LIN-NEXT: retl
;
; SSE3_32_WIN-LABEL: d_to_u64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl %ebp
; SSE3_32_WIN-NEXT: movl %esp, %ebp
; SSE3_32_WIN-NEXT: andl $-8, %esp
; SSE3_32_WIN-NEXT: subl $8, %esp
; SSE3_32_WIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE3_32_WIN-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE3_32_WIN-NEXT: movapd %xmm0, %xmm2
; SSE3_32_WIN-NEXT: cmpltsd %xmm1, %xmm2
; SSE3_32_WIN-NEXT: movapd %xmm2, %xmm3
; SSE3_32_WIN-NEXT: andpd %xmm0, %xmm2
; SSE3_32_WIN-NEXT: xorl %edx, %edx
; SSE3_32_WIN-NEXT: ucomisd %xmm0, %xmm1
; SSE3_32_WIN-NEXT: subsd %xmm1, %xmm0
; SSE3_32_WIN-NEXT: andnpd %xmm0, %xmm3
; SSE3_32_WIN-NEXT: orpd %xmm3, %xmm2
; SSE3_32_WIN-NEXT: movlpd %xmm2, (%esp)
; SSE3_32_WIN-NEXT: fldl (%esp)
; SSE3_32_WIN-NEXT: fisttpll (%esp)
; SSE3_32_WIN-NEXT: setbe %dl
; SSE3_32_WIN-NEXT: shll $31, %edx
; SSE3_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE3_32_WIN-NEXT: movl (%esp), %eax
; SSE3_32_WIN-NEXT: movl %ebp, %esp
; SSE3_32_WIN-NEXT: popl %ebp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: d_to_u64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE3_32_LIN-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE3_32_LIN-NEXT: movapd %xmm0, %xmm2
; SSE3_32_LIN-NEXT: cmpltsd %xmm1, %xmm2
; SSE3_32_LIN-NEXT: movapd %xmm2, %xmm3
; SSE3_32_LIN-NEXT: andpd %xmm0, %xmm2
; SSE3_32_LIN-NEXT: xorl %edx, %edx
; SSE3_32_LIN-NEXT: ucomisd %xmm0, %xmm1
; SSE3_32_LIN-NEXT: subsd %xmm1, %xmm0
; SSE3_32_LIN-NEXT: andnpd %xmm0, %xmm3
; SSE3_32_LIN-NEXT: orpd %xmm3, %xmm2
; SSE3_32_LIN-NEXT: movlpd %xmm2, (%esp)
; SSE3_32_LIN-NEXT: fldl (%esp)
; SSE3_32_LIN-NEXT: fisttpll (%esp)
; SSE3_32_LIN-NEXT: setbe %dl
; SSE3_32_LIN-NEXT: shll $31, %edx
; SSE3_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE3_32_LIN-NEXT: movl (%esp), %eax
; SSE3_32_LIN-NEXT: addl $12, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64-LABEL: d_to_u64:
; SSE3_64: # %bb.0:
; SSE3_64-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE3_64-NEXT: movapd %xmm0, %xmm2
; SSE3_64-NEXT: subsd %xmm1, %xmm2
; SSE3_64-NEXT: cvttsd2si %xmm2, %rax
; SSE3_64-NEXT: movabsq $-9223372036854775808, %rcx # imm = 0x8000000000000000
; SSE3_64-NEXT: xorq %rax, %rcx
; SSE3_64-NEXT: cvttsd2si %xmm0, %rax
; SSE3_64-NEXT: ucomisd %xmm1, %xmm0
; SSE3_64-NEXT: cmovaeq %rcx, %rax
; SSE3_64-NEXT: retq
;
; SSE2_32_WIN-LABEL: d_to_u64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl %ebp
; SSE2_32_WIN-NEXT: movl %esp, %ebp
; SSE2_32_WIN-NEXT: andl $-8, %esp
; SSE2_32_WIN-NEXT: subl $16, %esp
; SSE2_32_WIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE2_32_WIN-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE2_32_WIN-NEXT: movapd %xmm0, %xmm2
; SSE2_32_WIN-NEXT: subsd %xmm1, %xmm2
; SSE2_32_WIN-NEXT: movapd %xmm0, %xmm3
; SSE2_32_WIN-NEXT: cmpltsd %xmm1, %xmm3
; SSE2_32_WIN-NEXT: movapd %xmm3, %xmm4
; SSE2_32_WIN-NEXT: andnpd %xmm2, %xmm4
; SSE2_32_WIN-NEXT: andpd %xmm0, %xmm3
; SSE2_32_WIN-NEXT: orpd %xmm4, %xmm3
; SSE2_32_WIN-NEXT: movlpd %xmm3, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: xorl %edx, %edx
; SSE2_32_WIN-NEXT: ucomisd %xmm0, %xmm1
; SSE2_32_WIN-NEXT: setbe %dl
; SSE2_32_WIN-NEXT: shll $31, %edx
; SSE2_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: movl %ebp, %esp
; SSE2_32_WIN-NEXT: popl %ebp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: d_to_u64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $20, %esp
; SSE2_32_LIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE2_32_LIN-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE2_32_LIN-NEXT: movapd %xmm0, %xmm2
; SSE2_32_LIN-NEXT: subsd %xmm1, %xmm2
; SSE2_32_LIN-NEXT: movapd %xmm0, %xmm3
; SSE2_32_LIN-NEXT: cmpltsd %xmm1, %xmm3
; SSE2_32_LIN-NEXT: movapd %xmm3, %xmm4
; SSE2_32_LIN-NEXT: andnpd %xmm2, %xmm4
; SSE2_32_LIN-NEXT: andpd %xmm0, %xmm3
; SSE2_32_LIN-NEXT: orpd %xmm4, %xmm3
; SSE2_32_LIN-NEXT: movlpd %xmm3, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: xorl %edx, %edx
; SSE2_32_LIN-NEXT: ucomisd %xmm0, %xmm1
; SSE2_32_LIN-NEXT: setbe %dl
; SSE2_32_LIN-NEXT: shll $31, %edx
; SSE2_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: addl $20, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64-LABEL: d_to_u64:
; SSE2_64: # %bb.0:
; SSE2_64-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
; SSE2_64-NEXT: movapd %xmm0, %xmm2
; SSE2_64-NEXT: subsd %xmm1, %xmm2
; SSE2_64-NEXT: cvttsd2si %xmm2, %rax
; SSE2_64-NEXT: movabsq $-9223372036854775808, %rcx # imm = 0x8000000000000000
; SSE2_64-NEXT: xorq %rax, %rcx
; SSE2_64-NEXT: cvttsd2si %xmm0, %rax
; SSE2_64-NEXT: ucomisd %xmm1, %xmm0
; SSE2_64-NEXT: cmovaeq %rcx, %rax
; SSE2_64-NEXT: retq
;
; X87_WIN-LABEL: d_to_u64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl %ebp
; X87_WIN-NEXT: movl %esp, %ebp
; X87_WIN-NEXT: andl $-8, %esp
; X87_WIN-NEXT: subl $16, %esp
; X87_WIN-NEXT: fldl 8(%ebp)
; X87_WIN-NEXT: flds __real@5f000000
; X87_WIN-NEXT: fld %st(1)
; X87_WIN-NEXT: fsub %st(1), %st
; X87_WIN-NEXT: fxch %st(1)
; X87_WIN-NEXT: fucomp %st(2)
; X87_WIN-NEXT: fnstsw %ax
; X87_WIN-NEXT: # kill: def $ah killed $ah killed $ax
; X87_WIN-NEXT: sahf
; X87_WIN-NEXT: ja LBB2_2
; X87_WIN-NEXT: # %bb.1:
; X87_WIN-NEXT: fstp %st(1)
; X87_WIN-NEXT: fldz
; X87_WIN-NEXT: LBB2_2:
; X87_WIN-NEXT: fstp %st(0)
; X87_WIN-NEXT: setbe %al
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_WIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzbl %al, %edx
; X87_WIN-NEXT: shll $31, %edx
; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: movl %ebp, %esp
; X87_WIN-NEXT: popl %ebp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: d_to_u64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $20, %esp
; X87_LIN-NEXT: fldl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: flds {{\.LCPI.*}}
; X87_LIN-NEXT: fld %st(1)
; X87_LIN-NEXT: fsub %st(1), %st
; X87_LIN-NEXT: fxch %st(1)
; X87_LIN-NEXT: fucomp %st(2)
; X87_LIN-NEXT: fnstsw %ax
; X87_LIN-NEXT: # kill: def $ah killed $ah killed $ax
; X87_LIN-NEXT: sahf
; X87_LIN-NEXT: ja .LBB2_2
; X87_LIN-NEXT: # %bb.1:
; X87_LIN-NEXT: fstp %st(1)
; X87_LIN-NEXT: fldz
; X87_LIN-NEXT: .LBB2_2:
; X87_LIN-NEXT: fstp %st(0)
; X87_LIN-NEXT: setbe %al
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_LIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzbl %al, %edx
; X87_LIN-NEXT: shll $31, %edx
; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: addl $20, %esp
; X87_LIN-NEXT: retl
%r = fptoui double %a to i64
ret i64 %r
}
define i64 @d_to_s64(double %a) nounwind {
; AVX512DQVL_32_WIN-LABEL: d_to_s64:
; AVX512DQVL_32_WIN: # %bb.0:
; AVX512DQVL_32_WIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQVL_32_WIN-NEXT: vcvttpd2qq %ymm0, %ymm0
; AVX512DQVL_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_WIN-NEXT: vzeroupper
; AVX512DQVL_32_WIN-NEXT: retl
;
; AVX512DQVL_32_LIN-LABEL: d_to_s64:
; AVX512DQVL_32_LIN: # %bb.0:
; AVX512DQVL_32_LIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQVL_32_LIN-NEXT: vcvttpd2qq %ymm0, %ymm0
; AVX512DQVL_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQVL_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQVL_32_LIN-NEXT: vzeroupper
; AVX512DQVL_32_LIN-NEXT: retl
;
; AVX512_64-LABEL: d_to_s64:
; AVX512_64: # %bb.0:
; AVX512_64-NEXT: vcvttsd2si %xmm0, %rax
; AVX512_64-NEXT: retq
;
; AVX512DQ_32_WIN-LABEL: d_to_s64:
; AVX512DQ_32_WIN: # %bb.0:
; AVX512DQ_32_WIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ_32_WIN-NEXT: vcvttpd2qq %zmm0, %zmm0
; AVX512DQ_32_WIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_WIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_WIN-NEXT: vzeroupper
; AVX512DQ_32_WIN-NEXT: retl
;
; AVX512DQ_32_LIN-LABEL: d_to_s64:
; AVX512DQ_32_LIN: # %bb.0:
; AVX512DQ_32_LIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512DQ_32_LIN-NEXT: vcvttpd2qq %zmm0, %zmm0
; AVX512DQ_32_LIN-NEXT: vmovd %xmm0, %eax
; AVX512DQ_32_LIN-NEXT: vpextrd $1, %xmm0, %edx
; AVX512DQ_32_LIN-NEXT: vzeroupper
; AVX512DQ_32_LIN-NEXT: retl
;
; AVX512F_32_WIN-LABEL: d_to_s64:
; AVX512F_32_WIN: # %bb.0:
; AVX512F_32_WIN-NEXT: pushl %ebp
; AVX512F_32_WIN-NEXT: movl %esp, %ebp
; AVX512F_32_WIN-NEXT: andl $-8, %esp
; AVX512F_32_WIN-NEXT: subl $8, %esp
; AVX512F_32_WIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512F_32_WIN-NEXT: vmovsd %xmm0, (%esp)
; AVX512F_32_WIN-NEXT: fldl (%esp)
; AVX512F_32_WIN-NEXT: fisttpll (%esp)
; AVX512F_32_WIN-NEXT: movl (%esp), %eax
; AVX512F_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; AVX512F_32_WIN-NEXT: movl %ebp, %esp
; AVX512F_32_WIN-NEXT: popl %ebp
; AVX512F_32_WIN-NEXT: retl
;
; AVX512F_32_LIN-LABEL: d_to_s64:
; AVX512F_32_LIN: # %bb.0:
; AVX512F_32_LIN-NEXT: subl $12, %esp
; AVX512F_32_LIN-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX512F_32_LIN-NEXT: vmovsd %xmm0, (%esp)
; AVX512F_32_LIN-NEXT: fldl (%esp)
; AVX512F_32_LIN-NEXT: fisttpll (%esp)
; AVX512F_32_LIN-NEXT: movl (%esp), %eax
; AVX512F_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; AVX512F_32_LIN-NEXT: addl $12, %esp
; AVX512F_32_LIN-NEXT: retl
;
; SSE3_32_WIN-LABEL: d_to_s64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl %ebp
; SSE3_32_WIN-NEXT: movl %esp, %ebp
; SSE3_32_WIN-NEXT: andl $-8, %esp
; SSE3_32_WIN-NEXT: subl $8, %esp
; SSE3_32_WIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE3_32_WIN-NEXT: movsd %xmm0, (%esp)
; SSE3_32_WIN-NEXT: fldl (%esp)
; SSE3_32_WIN-NEXT: fisttpll (%esp)
; SSE3_32_WIN-NEXT: movl (%esp), %eax
; SSE3_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE3_32_WIN-NEXT: movl %ebp, %esp
; SSE3_32_WIN-NEXT: popl %ebp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: d_to_s64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE3_32_LIN-NEXT: movsd %xmm0, (%esp)
; SSE3_32_LIN-NEXT: fldl (%esp)
; SSE3_32_LIN-NEXT: fisttpll (%esp)
; SSE3_32_LIN-NEXT: movl (%esp), %eax
; SSE3_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE3_32_LIN-NEXT: addl $12, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64-LABEL: d_to_s64:
; SSE3_64: # %bb.0:
; SSE3_64-NEXT: cvttsd2si %xmm0, %rax
; SSE3_64-NEXT: retq
;
; SSE2_32_WIN-LABEL: d_to_s64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl %ebp
; SSE2_32_WIN-NEXT: movl %esp, %ebp
; SSE2_32_WIN-NEXT: andl $-8, %esp
; SSE2_32_WIN-NEXT: subl $16, %esp
; SSE2_32_WIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE2_32_WIN-NEXT: movsd %xmm0, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE2_32_WIN-NEXT: movl %ebp, %esp
; SSE2_32_WIN-NEXT: popl %ebp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: d_to_s64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $20, %esp
; SSE2_32_LIN-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; SSE2_32_LIN-NEXT: movsd %xmm0, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE2_32_LIN-NEXT: addl $20, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64-LABEL: d_to_s64:
; SSE2_64: # %bb.0:
; SSE2_64-NEXT: cvttsd2si %xmm0, %rax
; SSE2_64-NEXT: retq
;
; X87_WIN-LABEL: d_to_s64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl %ebp
; X87_WIN-NEXT: movl %esp, %ebp
; X87_WIN-NEXT: andl $-8, %esp
; X87_WIN-NEXT: subl $16, %esp
; X87_WIN-NEXT: fldl 8(%ebp)
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; X87_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl %ebp, %esp
; X87_WIN-NEXT: popl %ebp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: d_to_s64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $20, %esp
; X87_LIN-NEXT: fldl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; X87_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: addl $20, %esp
; X87_LIN-NEXT: retl
%r = fptosi double %a to i64
ret i64 %r
}
define i64 @x_to_u64(x86_fp80 %a) nounwind {
; AVX512_32_WIN-LABEL: x_to_u64:
; AVX512_32_WIN: # %bb.0:
; AVX512_32_WIN-NEXT: pushl %ebp
; AVX512_32_WIN-NEXT: movl %esp, %ebp
; AVX512_32_WIN-NEXT: andl $-8, %esp
; AVX512_32_WIN-NEXT: subl $8, %esp
; AVX512_32_WIN-NEXT: fldt 8(%ebp)
; AVX512_32_WIN-NEXT: flds __real@5f000000
; AVX512_32_WIN-NEXT: fld %st(1)
; AVX512_32_WIN-NEXT: fsub %st(1), %st
; AVX512_32_WIN-NEXT: xorl %edx, %edx
; AVX512_32_WIN-NEXT: fxch %st(1)
; AVX512_32_WIN-NEXT: fucompi %st(2), %st
; AVX512_32_WIN-NEXT: fcmovnbe %st(1), %st
; AVX512_32_WIN-NEXT: fstp %st(1)
; AVX512_32_WIN-NEXT: fisttpll (%esp)
; AVX512_32_WIN-NEXT: setbe %dl
; AVX512_32_WIN-NEXT: shll $31, %edx
; AVX512_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; AVX512_32_WIN-NEXT: movl (%esp), %eax
; AVX512_32_WIN-NEXT: movl %ebp, %esp
; AVX512_32_WIN-NEXT: popl %ebp
; AVX512_32_WIN-NEXT: retl
;
; AVX512_32_LIN-LABEL: x_to_u64:
; AVX512_32_LIN: # %bb.0:
; AVX512_32_LIN-NEXT: subl $12, %esp
; AVX512_32_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; AVX512_32_LIN-NEXT: flds {{\.LCPI.*}}
; AVX512_32_LIN-NEXT: fld %st(1)
; AVX512_32_LIN-NEXT: fsub %st(1), %st
; AVX512_32_LIN-NEXT: xorl %edx, %edx
; AVX512_32_LIN-NEXT: fxch %st(1)
; AVX512_32_LIN-NEXT: fucompi %st(2), %st
; AVX512_32_LIN-NEXT: fcmovnbe %st(1), %st
; AVX512_32_LIN-NEXT: fstp %st(1)
; AVX512_32_LIN-NEXT: fisttpll (%esp)
; AVX512_32_LIN-NEXT: setbe %dl
; AVX512_32_LIN-NEXT: shll $31, %edx
; AVX512_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; AVX512_32_LIN-NEXT: movl (%esp), %eax
; AVX512_32_LIN-NEXT: addl $12, %esp
; AVX512_32_LIN-NEXT: retl
;
; AVX512_64_WIN-LABEL: x_to_u64:
; AVX512_64_WIN: # %bb.0:
; AVX512_64_WIN-NEXT: pushq %rax
; AVX512_64_WIN-NEXT: fldt (%rcx)
; AVX512_64_WIN-NEXT: flds __real@{{.*}}(%rip)
; AVX512_64_WIN-NEXT: fld %st(1)
; AVX512_64_WIN-NEXT: fsub %st(1), %st
; AVX512_64_WIN-NEXT: xorl %eax, %eax
; AVX512_64_WIN-NEXT: fxch %st(1)
; AVX512_64_WIN-NEXT: fucompi %st(2), %st
; AVX512_64_WIN-NEXT: fcmovnbe %st(1), %st
; AVX512_64_WIN-NEXT: fstp %st(1)
; AVX512_64_WIN-NEXT: fisttpll (%rsp)
; AVX512_64_WIN-NEXT: setbe %al
; AVX512_64_WIN-NEXT: shlq $63, %rax
; AVX512_64_WIN-NEXT: xorq (%rsp), %rax
; AVX512_64_WIN-NEXT: popq %rcx
; AVX512_64_WIN-NEXT: retq
;
; AVX512_64_LIN-LABEL: x_to_u64:
; AVX512_64_LIN: # %bb.0:
; AVX512_64_LIN-NEXT: fldt {{[0-9]+}}(%rsp)
; AVX512_64_LIN-NEXT: flds {{.*}}(%rip)
; AVX512_64_LIN-NEXT: fld %st(1)
; AVX512_64_LIN-NEXT: fsub %st(1), %st
; AVX512_64_LIN-NEXT: xorl %eax, %eax
; AVX512_64_LIN-NEXT: fxch %st(1)
; AVX512_64_LIN-NEXT: fucompi %st(2), %st
; AVX512_64_LIN-NEXT: fcmovnbe %st(1), %st
; AVX512_64_LIN-NEXT: fstp %st(1)
; AVX512_64_LIN-NEXT: fisttpll -{{[0-9]+}}(%rsp)
; AVX512_64_LIN-NEXT: setbe %al
; AVX512_64_LIN-NEXT: shlq $63, %rax
; AVX512_64_LIN-NEXT: xorq -{{[0-9]+}}(%rsp), %rax
; AVX512_64_LIN-NEXT: retq
;
; SSE3_32_WIN-LABEL: x_to_u64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl %ebp
; SSE3_32_WIN-NEXT: movl %esp, %ebp
; SSE3_32_WIN-NEXT: andl $-8, %esp
; SSE3_32_WIN-NEXT: subl $8, %esp
; SSE3_32_WIN-NEXT: fldt 8(%ebp)
; SSE3_32_WIN-NEXT: flds __real@5f000000
; SSE3_32_WIN-NEXT: fld %st(1)
; SSE3_32_WIN-NEXT: fsub %st(1), %st
; SSE3_32_WIN-NEXT: xorl %edx, %edx
; SSE3_32_WIN-NEXT: fxch %st(1)
; SSE3_32_WIN-NEXT: fucompi %st(2), %st
; SSE3_32_WIN-NEXT: fcmovnbe %st(1), %st
; SSE3_32_WIN-NEXT: fstp %st(1)
; SSE3_32_WIN-NEXT: fisttpll (%esp)
; SSE3_32_WIN-NEXT: setbe %dl
; SSE3_32_WIN-NEXT: shll $31, %edx
; SSE3_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE3_32_WIN-NEXT: movl (%esp), %eax
; SSE3_32_WIN-NEXT: movl %ebp, %esp
; SSE3_32_WIN-NEXT: popl %ebp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: x_to_u64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: flds {{\.LCPI.*}}
; SSE3_32_LIN-NEXT: fld %st(1)
; SSE3_32_LIN-NEXT: fsub %st(1), %st
; SSE3_32_LIN-NEXT: xorl %edx, %edx
; SSE3_32_LIN-NEXT: fxch %st(1)
; SSE3_32_LIN-NEXT: fucompi %st(2), %st
; SSE3_32_LIN-NEXT: fcmovnbe %st(1), %st
; SSE3_32_LIN-NEXT: fstp %st(1)
; SSE3_32_LIN-NEXT: fisttpll (%esp)
; SSE3_32_LIN-NEXT: setbe %dl
; SSE3_32_LIN-NEXT: shll $31, %edx
; SSE3_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE3_32_LIN-NEXT: movl (%esp), %eax
; SSE3_32_LIN-NEXT: addl $12, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64_WIN-LABEL: x_to_u64:
; SSE3_64_WIN: # %bb.0:
; SSE3_64_WIN-NEXT: pushq %rax
; SSE3_64_WIN-NEXT: fldt (%rcx)
; SSE3_64_WIN-NEXT: flds __real@{{.*}}(%rip)
; SSE3_64_WIN-NEXT: fld %st(1)
; SSE3_64_WIN-NEXT: fsub %st(1), %st
; SSE3_64_WIN-NEXT: xorl %eax, %eax
; SSE3_64_WIN-NEXT: fxch %st(1)
; SSE3_64_WIN-NEXT: fucompi %st(2), %st
; SSE3_64_WIN-NEXT: fcmovnbe %st(1), %st
; SSE3_64_WIN-NEXT: fstp %st(1)
; SSE3_64_WIN-NEXT: fisttpll (%rsp)
; SSE3_64_WIN-NEXT: setbe %al
; SSE3_64_WIN-NEXT: shlq $63, %rax
; SSE3_64_WIN-NEXT: xorq (%rsp), %rax
; SSE3_64_WIN-NEXT: popq %rcx
; SSE3_64_WIN-NEXT: retq
;
; SSE3_64_LIN-LABEL: x_to_u64:
; SSE3_64_LIN: # %bb.0:
; SSE3_64_LIN-NEXT: fldt {{[0-9]+}}(%rsp)
; SSE3_64_LIN-NEXT: flds {{.*}}(%rip)
; SSE3_64_LIN-NEXT: fld %st(1)
; SSE3_64_LIN-NEXT: fsub %st(1), %st
; SSE3_64_LIN-NEXT: xorl %eax, %eax
; SSE3_64_LIN-NEXT: fxch %st(1)
; SSE3_64_LIN-NEXT: fucompi %st(2), %st
; SSE3_64_LIN-NEXT: fcmovnbe %st(1), %st
; SSE3_64_LIN-NEXT: fstp %st(1)
; SSE3_64_LIN-NEXT: fisttpll -{{[0-9]+}}(%rsp)
; SSE3_64_LIN-NEXT: setbe %al
; SSE3_64_LIN-NEXT: shlq $63, %rax
; SSE3_64_LIN-NEXT: xorq -{{[0-9]+}}(%rsp), %rax
; SSE3_64_LIN-NEXT: retq
;
; SSE2_32_WIN-LABEL: x_to_u64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl %ebp
; SSE2_32_WIN-NEXT: movl %esp, %ebp
; SSE2_32_WIN-NEXT: andl $-8, %esp
; SSE2_32_WIN-NEXT: subl $16, %esp
; SSE2_32_WIN-NEXT: fldt 8(%ebp)
; SSE2_32_WIN-NEXT: flds __real@5f000000
; SSE2_32_WIN-NEXT: fld %st(1)
; SSE2_32_WIN-NEXT: fsub %st(1), %st
; SSE2_32_WIN-NEXT: xorl %edx, %edx
; SSE2_32_WIN-NEXT: fxch %st(1)
; SSE2_32_WIN-NEXT: fucompi %st(2), %st
; SSE2_32_WIN-NEXT: fcmovnbe %st(1), %st
; SSE2_32_WIN-NEXT: fstp %st(1)
; SSE2_32_WIN-NEXT: setbe %dl
; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: shll $31, %edx
; SSE2_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: movl %ebp, %esp
; SSE2_32_WIN-NEXT: popl %ebp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: x_to_u64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $20, %esp
; SSE2_32_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: flds {{\.LCPI.*}}
; SSE2_32_LIN-NEXT: fld %st(1)
; SSE2_32_LIN-NEXT: fsub %st(1), %st
; SSE2_32_LIN-NEXT: xorl %edx, %edx
; SSE2_32_LIN-NEXT: fxch %st(1)
; SSE2_32_LIN-NEXT: fucompi %st(2), %st
; SSE2_32_LIN-NEXT: fcmovnbe %st(1), %st
; SSE2_32_LIN-NEXT: fstp %st(1)
; SSE2_32_LIN-NEXT: setbe %dl
; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: shll $31, %edx
; SSE2_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: addl $20, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64_WIN-LABEL: x_to_u64:
; SSE2_64_WIN: # %bb.0:
; SSE2_64_WIN-NEXT: subq $16, %rsp
; SSE2_64_WIN-NEXT: fldt (%rcx)
; SSE2_64_WIN-NEXT: flds __real@{{.*}}(%rip)
; SSE2_64_WIN-NEXT: fld %st(1)
; SSE2_64_WIN-NEXT: fsub %st(1), %st
; SSE2_64_WIN-NEXT: xorl %eax, %eax
; SSE2_64_WIN-NEXT: fxch %st(1)
; SSE2_64_WIN-NEXT: fucompi %st(2), %st
; SSE2_64_WIN-NEXT: fcmovnbe %st(1), %st
; SSE2_64_WIN-NEXT: fstp %st(1)
; SSE2_64_WIN-NEXT: setbe %al
; SSE2_64_WIN-NEXT: fnstcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: movzwl {{[0-9]+}}(%rsp), %ecx
; SSE2_64_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE2_64_WIN-NEXT: movw %cx, {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fistpll {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: shlq $63, %rax
; SSE2_64_WIN-NEXT: xorq {{[0-9]+}}(%rsp), %rax
; SSE2_64_WIN-NEXT: addq $16, %rsp
; SSE2_64_WIN-NEXT: retq
;
; SSE2_64_LIN-LABEL: x_to_u64:
; SSE2_64_LIN: # %bb.0:
; SSE2_64_LIN-NEXT: fldt {{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: flds {{.*}}(%rip)
; SSE2_64_LIN-NEXT: fld %st(1)
; SSE2_64_LIN-NEXT: fsub %st(1), %st
; SSE2_64_LIN-NEXT: xorl %eax, %eax
; SSE2_64_LIN-NEXT: fxch %st(1)
; SSE2_64_LIN-NEXT: fucompi %st(2), %st
; SSE2_64_LIN-NEXT: fcmovnbe %st(1), %st
; SSE2_64_LIN-NEXT: fstp %st(1)
; SSE2_64_LIN-NEXT: setbe %al
; SSE2_64_LIN-NEXT: fnstcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: movzwl -{{[0-9]+}}(%rsp), %ecx
; SSE2_64_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE2_64_LIN-NEXT: movw %cx, -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fistpll -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: shlq $63, %rax
; SSE2_64_LIN-NEXT: xorq -{{[0-9]+}}(%rsp), %rax
; SSE2_64_LIN-NEXT: retq
;
; X87_WIN-LABEL: x_to_u64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl %ebp
; X87_WIN-NEXT: movl %esp, %ebp
; X87_WIN-NEXT: andl $-8, %esp
; X87_WIN-NEXT: subl $16, %esp
; X87_WIN-NEXT: fldt 8(%ebp)
; X87_WIN-NEXT: flds __real@5f000000
; X87_WIN-NEXT: fld %st(1)
; X87_WIN-NEXT: fsub %st(1), %st
; X87_WIN-NEXT: fxch %st(1)
; X87_WIN-NEXT: fucomp %st(2)
; X87_WIN-NEXT: fnstsw %ax
; X87_WIN-NEXT: # kill: def $ah killed $ah killed $ax
; X87_WIN-NEXT: sahf
; X87_WIN-NEXT: ja LBB4_2
; X87_WIN-NEXT: # %bb.1:
; X87_WIN-NEXT: fstp %st(1)
; X87_WIN-NEXT: fldz
; X87_WIN-NEXT: LBB4_2:
; X87_WIN-NEXT: fstp %st(0)
; X87_WIN-NEXT: setbe %al
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_WIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzbl %al, %edx
; X87_WIN-NEXT: shll $31, %edx
; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: movl %ebp, %esp
; X87_WIN-NEXT: popl %ebp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: x_to_u64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $20, %esp
; X87_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; X87_LIN-NEXT: flds {{\.LCPI.*}}
; X87_LIN-NEXT: fld %st(1)
; X87_LIN-NEXT: fsub %st(1), %st
; X87_LIN-NEXT: fxch %st(1)
; X87_LIN-NEXT: fucomp %st(2)
; X87_LIN-NEXT: fnstsw %ax
; X87_LIN-NEXT: # kill: def $ah killed $ah killed $ax
; X87_LIN-NEXT: sahf
; X87_LIN-NEXT: ja .LBB4_2
; X87_LIN-NEXT: # %bb.1:
; X87_LIN-NEXT: fstp %st(1)
; X87_LIN-NEXT: fldz
; X87_LIN-NEXT: .LBB4_2:
; X87_LIN-NEXT: fstp %st(0)
; X87_LIN-NEXT: setbe %al
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_LIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzbl %al, %edx
; X87_LIN-NEXT: shll $31, %edx
; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: addl $20, %esp
; X87_LIN-NEXT: retl
%r = fptoui x86_fp80 %a to i64
ret i64 %r
}
define i64 @x_to_s64(x86_fp80 %a) nounwind {
; AVX512_32_WIN-LABEL: x_to_s64:
; AVX512_32_WIN: # %bb.0:
; AVX512_32_WIN-NEXT: pushl %ebp
; AVX512_32_WIN-NEXT: movl %esp, %ebp
; AVX512_32_WIN-NEXT: andl $-8, %esp
; AVX512_32_WIN-NEXT: subl $8, %esp
; AVX512_32_WIN-NEXT: fldt 8(%ebp)
; AVX512_32_WIN-NEXT: fisttpll (%esp)
; AVX512_32_WIN-NEXT: movl (%esp), %eax
; AVX512_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; AVX512_32_WIN-NEXT: movl %ebp, %esp
; AVX512_32_WIN-NEXT: popl %ebp
; AVX512_32_WIN-NEXT: retl
;
; AVX512_32_LIN-LABEL: x_to_s64:
; AVX512_32_LIN: # %bb.0:
; AVX512_32_LIN-NEXT: subl $12, %esp
; AVX512_32_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; AVX512_32_LIN-NEXT: fisttpll (%esp)
; AVX512_32_LIN-NEXT: movl (%esp), %eax
; AVX512_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; AVX512_32_LIN-NEXT: addl $12, %esp
; AVX512_32_LIN-NEXT: retl
;
; AVX512_64_WIN-LABEL: x_to_s64:
; AVX512_64_WIN: # %bb.0:
; AVX512_64_WIN-NEXT: pushq %rax
; AVX512_64_WIN-NEXT: fldt (%rcx)
; AVX512_64_WIN-NEXT: fisttpll (%rsp)
; AVX512_64_WIN-NEXT: movq (%rsp), %rax
; AVX512_64_WIN-NEXT: popq %rcx
; AVX512_64_WIN-NEXT: retq
;
; AVX512_64_LIN-LABEL: x_to_s64:
; AVX512_64_LIN: # %bb.0:
; AVX512_64_LIN-NEXT: fldt {{[0-9]+}}(%rsp)
; AVX512_64_LIN-NEXT: fisttpll -{{[0-9]+}}(%rsp)
; AVX512_64_LIN-NEXT: movq -{{[0-9]+}}(%rsp), %rax
; AVX512_64_LIN-NEXT: retq
;
; SSE3_32_WIN-LABEL: x_to_s64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl %ebp
; SSE3_32_WIN-NEXT: movl %esp, %ebp
; SSE3_32_WIN-NEXT: andl $-8, %esp
; SSE3_32_WIN-NEXT: subl $8, %esp
; SSE3_32_WIN-NEXT: fldt 8(%ebp)
; SSE3_32_WIN-NEXT: fisttpll (%esp)
; SSE3_32_WIN-NEXT: movl (%esp), %eax
; SSE3_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE3_32_WIN-NEXT: movl %ebp, %esp
; SSE3_32_WIN-NEXT: popl %ebp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: x_to_s64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: fisttpll (%esp)
; SSE3_32_LIN-NEXT: movl (%esp), %eax
; SSE3_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE3_32_LIN-NEXT: addl $12, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64_WIN-LABEL: x_to_s64:
; SSE3_64_WIN: # %bb.0:
; SSE3_64_WIN-NEXT: pushq %rax
; SSE3_64_WIN-NEXT: fldt (%rcx)
; SSE3_64_WIN-NEXT: fisttpll (%rsp)
; SSE3_64_WIN-NEXT: movq (%rsp), %rax
; SSE3_64_WIN-NEXT: popq %rcx
; SSE3_64_WIN-NEXT: retq
;
; SSE3_64_LIN-LABEL: x_to_s64:
; SSE3_64_LIN: # %bb.0:
; SSE3_64_LIN-NEXT: fldt {{[0-9]+}}(%rsp)
; SSE3_64_LIN-NEXT: fisttpll -{{[0-9]+}}(%rsp)
; SSE3_64_LIN-NEXT: movq -{{[0-9]+}}(%rsp), %rax
; SSE3_64_LIN-NEXT: retq
;
; SSE2_32_WIN-LABEL: x_to_s64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl %ebp
; SSE2_32_WIN-NEXT: movl %esp, %ebp
; SSE2_32_WIN-NEXT: andl $-8, %esp
; SSE2_32_WIN-NEXT: subl $16, %esp
; SSE2_32_WIN-NEXT: fldt 8(%ebp)
; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE2_32_WIN-NEXT: movl %ebp, %esp
; SSE2_32_WIN-NEXT: popl %ebp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: x_to_s64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $20, %esp
; SSE2_32_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; SSE2_32_LIN-NEXT: addl $20, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64_WIN-LABEL: x_to_s64:
; SSE2_64_WIN: # %bb.0:
; SSE2_64_WIN-NEXT: subq $16, %rsp
; SSE2_64_WIN-NEXT: fldt (%rcx)
; SSE2_64_WIN-NEXT: fnstcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: movzwl {{[0-9]+}}(%rsp), %eax
; SSE2_64_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_64_WIN-NEXT: movw %ax, {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fistpll {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: movq {{[0-9]+}}(%rsp), %rax
; SSE2_64_WIN-NEXT: addq $16, %rsp
; SSE2_64_WIN-NEXT: retq
;
; SSE2_64_LIN-LABEL: x_to_s64:
; SSE2_64_LIN: # %bb.0:
; SSE2_64_LIN-NEXT: fldt {{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fnstcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: movzwl -{{[0-9]+}}(%rsp), %eax
; SSE2_64_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_64_LIN-NEXT: movw %ax, -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fistpll -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: movq -{{[0-9]+}}(%rsp), %rax
; SSE2_64_LIN-NEXT: retq
;
; X87_WIN-LABEL: x_to_s64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl %ebp
; X87_WIN-NEXT: movl %esp, %ebp
; X87_WIN-NEXT: andl $-8, %esp
; X87_WIN-NEXT: subl $16, %esp
; X87_WIN-NEXT: fldt 8(%ebp)
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; X87_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl %ebp, %esp
; X87_WIN-NEXT: popl %ebp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: x_to_s64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $20, %esp
; X87_LIN-NEXT: fldt {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; X87_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: addl $20, %esp
; X87_LIN-NEXT: retl
%r = fptosi x86_fp80 %a to i64
ret i64 %r
}
define i64 @t_to_u64(fp128 %a) nounwind {
; AVX512_32_WIN-LABEL: t_to_u64:
; AVX512_32_WIN: # %bb.0:
; AVX512_32_WIN-NEXT: subl $16, %esp
; AVX512_32_WIN-NEXT: vmovups {{[0-9]+}}(%esp), %xmm0
; AVX512_32_WIN-NEXT: vmovups %xmm0, (%esp)
; AVX512_32_WIN-NEXT: calll ___fixunstfdi
; AVX512_32_WIN-NEXT: addl $16, %esp
; AVX512_32_WIN-NEXT: retl
;
; AVX512_32_LIN-LABEL: t_to_u64:
; AVX512_32_LIN: # %bb.0:
; AVX512_32_LIN-NEXT: subl $28, %esp
; AVX512_32_LIN-NEXT: vmovaps {{[0-9]+}}(%esp), %xmm0
; AVX512_32_LIN-NEXT: vmovups %xmm0, (%esp)
; AVX512_32_LIN-NEXT: calll __fixunstfdi
; AVX512_32_LIN-NEXT: addl $28, %esp
; AVX512_32_LIN-NEXT: retl
;
; AVX512_64_WIN-LABEL: t_to_u64:
; AVX512_64_WIN: # %bb.0:
; AVX512_64_WIN-NEXT: subq $40, %rsp
; AVX512_64_WIN-NEXT: callq __fixunstfdi
; AVX512_64_WIN-NEXT: addq $40, %rsp
; AVX512_64_WIN-NEXT: retq
;
; AVX512_64_LIN-LABEL: t_to_u64:
; AVX512_64_LIN: # %bb.0:
; AVX512_64_LIN-NEXT: pushq %rax
; AVX512_64_LIN-NEXT: callq __fixunstfdi
; AVX512_64_LIN-NEXT: popq %rcx
; AVX512_64_LIN-NEXT: retq
;
; SSE3_32_WIN-LABEL: t_to_u64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: calll ___fixunstfdi
; SSE3_32_WIN-NEXT: addl $16, %esp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: t_to_u64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: calll __fixunstfdi
; SSE3_32_LIN-NEXT: addl $28, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64_WIN-LABEL: t_to_u64:
; SSE3_64_WIN: # %bb.0:
; SSE3_64_WIN-NEXT: subq $40, %rsp
; SSE3_64_WIN-NEXT: callq __fixunstfdi
; SSE3_64_WIN-NEXT: addq $40, %rsp
; SSE3_64_WIN-NEXT: retq
;
; SSE3_64_LIN-LABEL: t_to_u64:
; SSE3_64_LIN: # %bb.0:
; SSE3_64_LIN-NEXT: pushq %rax
; SSE3_64_LIN-NEXT: callq __fixunstfdi
; SSE3_64_LIN-NEXT: popq %rcx
; SSE3_64_LIN-NEXT: retq
;
; SSE2_32_WIN-LABEL: t_to_u64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: calll ___fixunstfdi
; SSE2_32_WIN-NEXT: addl $16, %esp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: t_to_u64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $12, %esp
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: calll __fixunstfdi
; SSE2_32_LIN-NEXT: addl $28, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64_WIN-LABEL: t_to_u64:
; SSE2_64_WIN: # %bb.0:
; SSE2_64_WIN-NEXT: subq $40, %rsp
; SSE2_64_WIN-NEXT: callq __fixunstfdi
; SSE2_64_WIN-NEXT: addq $40, %rsp
; SSE2_64_WIN-NEXT: retq
;
; SSE2_64_LIN-LABEL: t_to_u64:
; SSE2_64_LIN: # %bb.0:
; SSE2_64_LIN-NEXT: pushq %rax
; SSE2_64_LIN-NEXT: callq __fixunstfdi
; SSE2_64_LIN-NEXT: popq %rcx
; SSE2_64_LIN-NEXT: retq
;
; X87_WIN-LABEL: t_to_u64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: calll ___fixunstfdi
; X87_WIN-NEXT: addl $16, %esp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: t_to_u64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $12, %esp
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: calll __fixunstfdi
; X87_LIN-NEXT: addl $28, %esp
; X87_LIN-NEXT: retl
%r = fptoui fp128 %a to i64
ret i64 %r
}
define i64 @t_to_s64(fp128 %a) nounwind {
; AVX512_32_WIN-LABEL: t_to_s64:
; AVX512_32_WIN: # %bb.0:
; AVX512_32_WIN-NEXT: subl $16, %esp
; AVX512_32_WIN-NEXT: vmovups {{[0-9]+}}(%esp), %xmm0
; AVX512_32_WIN-NEXT: vmovups %xmm0, (%esp)
; AVX512_32_WIN-NEXT: calll ___fixtfdi
; AVX512_32_WIN-NEXT: addl $16, %esp
; AVX512_32_WIN-NEXT: retl
;
; AVX512_32_LIN-LABEL: t_to_s64:
; AVX512_32_LIN: # %bb.0:
; AVX512_32_LIN-NEXT: subl $28, %esp
; AVX512_32_LIN-NEXT: vmovaps {{[0-9]+}}(%esp), %xmm0
; AVX512_32_LIN-NEXT: vmovups %xmm0, (%esp)
; AVX512_32_LIN-NEXT: calll __fixtfdi
; AVX512_32_LIN-NEXT: addl $28, %esp
; AVX512_32_LIN-NEXT: retl
;
; AVX512_64_WIN-LABEL: t_to_s64:
; AVX512_64_WIN: # %bb.0:
; AVX512_64_WIN-NEXT: subq $40, %rsp
; AVX512_64_WIN-NEXT: callq __fixtfdi
; AVX512_64_WIN-NEXT: addq $40, %rsp
; AVX512_64_WIN-NEXT: retq
;
; AVX512_64_LIN-LABEL: t_to_s64:
; AVX512_64_LIN: # %bb.0:
; AVX512_64_LIN-NEXT: pushq %rax
; AVX512_64_LIN-NEXT: callq __fixtfdi
; AVX512_64_LIN-NEXT: popq %rcx
; AVX512_64_LIN-NEXT: retq
;
; SSE3_32_WIN-LABEL: t_to_s64:
; SSE3_32_WIN: # %bb.0:
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_WIN-NEXT: calll ___fixtfdi
; SSE3_32_WIN-NEXT: addl $16, %esp
; SSE3_32_WIN-NEXT: retl
;
; SSE3_32_LIN-LABEL: t_to_s64:
; SSE3_32_LIN: # %bb.0:
; SSE3_32_LIN-NEXT: subl $12, %esp
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE3_32_LIN-NEXT: calll __fixtfdi
; SSE3_32_LIN-NEXT: addl $28, %esp
; SSE3_32_LIN-NEXT: retl
;
; SSE3_64_WIN-LABEL: t_to_s64:
; SSE3_64_WIN: # %bb.0:
; SSE3_64_WIN-NEXT: subq $40, %rsp
; SSE3_64_WIN-NEXT: callq __fixtfdi
; SSE3_64_WIN-NEXT: addq $40, %rsp
; SSE3_64_WIN-NEXT: retq
;
; SSE3_64_LIN-LABEL: t_to_s64:
; SSE3_64_LIN: # %bb.0:
; SSE3_64_LIN-NEXT: pushq %rax
; SSE3_64_LIN-NEXT: callq __fixtfdi
; SSE3_64_LIN-NEXT: popq %rcx
; SSE3_64_LIN-NEXT: retq
;
; SSE2_32_WIN-LABEL: t_to_s64:
; SSE2_32_WIN: # %bb.0:
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: calll ___fixtfdi
; SSE2_32_WIN-NEXT: addl $16, %esp
; SSE2_32_WIN-NEXT: retl
;
; SSE2_32_LIN-LABEL: t_to_s64:
; SSE2_32_LIN: # %bb.0:
; SSE2_32_LIN-NEXT: subl $12, %esp
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: calll __fixtfdi
; SSE2_32_LIN-NEXT: addl $28, %esp
; SSE2_32_LIN-NEXT: retl
;
; SSE2_64_WIN-LABEL: t_to_s64:
; SSE2_64_WIN: # %bb.0:
; SSE2_64_WIN-NEXT: subq $40, %rsp
; SSE2_64_WIN-NEXT: callq __fixtfdi
; SSE2_64_WIN-NEXT: addq $40, %rsp
; SSE2_64_WIN-NEXT: retq
;
; SSE2_64_LIN-LABEL: t_to_s64:
; SSE2_64_LIN: # %bb.0:
; SSE2_64_LIN-NEXT: pushq %rax
; SSE2_64_LIN-NEXT: callq __fixtfdi
; SSE2_64_LIN-NEXT: popq %rcx
; SSE2_64_LIN-NEXT: retq
;
; X87_WIN-LABEL: t_to_s64:
; X87_WIN: # %bb.0:
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_WIN-NEXT: calll ___fixtfdi
; X87_WIN-NEXT: addl $16, %esp
; X87_WIN-NEXT: retl
;
; X87_LIN-LABEL: t_to_s64:
; X87_LIN: # %bb.0:
; X87_LIN-NEXT: subl $12, %esp
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: pushl {{[0-9]+}}(%esp)
; X87_LIN-NEXT: calll __fixtfdi
; X87_LIN-NEXT: addl $28, %esp
; X87_LIN-NEXT: retl
%r = fptosi fp128 %a to i64
ret i64 %r
}