blob: 764285fea64d2d0dd884043ecf02d4e7f1cca1d2 [file]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+f16c < %s | FileCheck %s --check-prefixes=ALL,F16C
; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx < %s | FileCheck %s --check-prefixes=ALL,AVX
; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx512fp16 < %s | FileCheck %s --check-prefixes=FP16
define zeroext i16 @test1_fast(double %d) #0 {
; ALL-LABEL: test1_fast:
; ALL: # %bb.0: # %entry
; ALL-NEXT: pushq %rax
; ALL-NEXT: callq __truncdfhf2@PLT
; ALL-NEXT: vpextrw $0, %xmm0, %eax
; ALL-NEXT: # kill: def $ax killed $ax killed $eax
; ALL-NEXT: popq %rcx
; ALL-NEXT: retq
;
; FP16-LABEL: test1_fast:
; FP16: # %bb.0: # %entry
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm0
; FP16-NEXT: vmovw %xmm0, %eax
; FP16-NEXT: # kill: def $ax killed $ax killed $eax
; FP16-NEXT: retq
entry:
%0 = tail call i16 @llvm.convert.to.fp16.f64(double %d)
ret i16 %0
}
define zeroext i16 @test2_fast(x86_fp80 %d) #0 {
; ALL-LABEL: test2_fast:
; ALL: # %bb.0: # %entry
; ALL-NEXT: subq $24, %rsp
; ALL-NEXT: fldt {{[0-9]+}}(%rsp)
; ALL-NEXT: fstpt (%rsp)
; ALL-NEXT: callq __truncxfhf2@PLT
; ALL-NEXT: vpextrw $0, %xmm0, %eax
; ALL-NEXT: # kill: def $ax killed $ax killed $eax
; ALL-NEXT: addq $24, %rsp
; ALL-NEXT: retq
;
; FP16-LABEL: test2_fast:
; FP16: # %bb.0: # %entry
; FP16-NEXT: subq $24, %rsp
; FP16-NEXT: fldt {{[0-9]+}}(%rsp)
; FP16-NEXT: fstpt (%rsp)
; FP16-NEXT: callq __truncxfhf2@PLT
; FP16-NEXT: vmovw %xmm0, %eax
; FP16-NEXT: # kill: def $ax killed $ax killed $eax
; FP16-NEXT: addq $24, %rsp
; FP16-NEXT: retq
entry:
%0 = tail call i16 @llvm.convert.to.fp16.f80(x86_fp80 %d)
ret i16 %0
}
define zeroext i16 @test1(double %d) #1 {
; ALL-LABEL: test1:
; ALL: # %bb.0: # %entry
; ALL-NEXT: pushq %rax
; ALL-NEXT: callq __truncdfhf2@PLT
; ALL-NEXT: vpextrw $0, %xmm0, %eax
; ALL-NEXT: # kill: def $ax killed $ax killed $eax
; ALL-NEXT: popq %rcx
; ALL-NEXT: retq
;
; FP16-LABEL: test1:
; FP16: # %bb.0: # %entry
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm0
; FP16-NEXT: vmovw %xmm0, %eax
; FP16-NEXT: # kill: def $ax killed $ax killed $eax
; FP16-NEXT: retq
entry:
%0 = tail call i16 @llvm.convert.to.fp16.f64(double %d)
ret i16 %0
}
define zeroext i16 @test2(x86_fp80 %d) #1 {
; ALL-LABEL: test2:
; ALL: # %bb.0: # %entry
; ALL-NEXT: subq $24, %rsp
; ALL-NEXT: fldt {{[0-9]+}}(%rsp)
; ALL-NEXT: fstpt (%rsp)
; ALL-NEXT: callq __truncxfhf2@PLT
; ALL-NEXT: vpextrw $0, %xmm0, %eax
; ALL-NEXT: # kill: def $ax killed $ax killed $eax
; ALL-NEXT: addq $24, %rsp
; ALL-NEXT: retq
;
; FP16-LABEL: test2:
; FP16: # %bb.0: # %entry
; FP16-NEXT: subq $24, %rsp
; FP16-NEXT: fldt {{[0-9]+}}(%rsp)
; FP16-NEXT: fstpt (%rsp)
; FP16-NEXT: callq __truncxfhf2@PLT
; FP16-NEXT: vmovw %xmm0, %eax
; FP16-NEXT: # kill: def $ax killed $ax killed $eax
; FP16-NEXT: addq $24, %rsp
; FP16-NEXT: retq
entry:
%0 = tail call i16 @llvm.convert.to.fp16.f80(x86_fp80 %d)
ret i16 %0
}
define half @fptrunc_afn(double %d) #0 {
; F16C-LABEL: fptrunc_afn:
; F16C: # %bb.0:
; F16C-NEXT: vcvtsd2ss %xmm0, %xmm0, %xmm0
; F16C-NEXT: vcvtps2ph $4, %xmm0, %xmm0
; F16C-NEXT: retq
;
; AVX-LABEL: fptrunc_afn:
; AVX: # %bb.0:
; AVX-NEXT: pushq %rax
; AVX-NEXT: callq __truncdfhf2@PLT
; AVX-NEXT: popq %rax
; AVX-NEXT: retq
;
; FP16-LABEL: fptrunc_afn:
; FP16: # %bb.0:
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm0
; FP16-NEXT: retq
%r = fptrunc afn double %d to half
ret half %r
}
define half @fptrunc_afn_f80(x86_fp80 %d) #0 {
; F16C-LABEL: fptrunc_afn_f80:
; F16C: # %bb.0:
; F16C-NEXT: fldt {{[0-9]+}}(%rsp)
; F16C-NEXT: fstps -{{[0-9]+}}(%rsp)
; F16C-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; F16C-NEXT: vcvtps2ph $4, %xmm0, %xmm0
; F16C-NEXT: retq
;
; AVX-LABEL: fptrunc_afn_f80:
; AVX: # %bb.0:
; AVX-NEXT: subq $24, %rsp
; AVX-NEXT: fldt {{[0-9]+}}(%rsp)
; AVX-NEXT: fstpt (%rsp)
; AVX-NEXT: callq __truncxfhf2@PLT
; AVX-NEXT: addq $24, %rsp
; AVX-NEXT: retq
;
; FP16-LABEL: fptrunc_afn_f80:
; FP16: # %bb.0:
; FP16-NEXT: subq $24, %rsp
; FP16-NEXT: fldt {{[0-9]+}}(%rsp)
; FP16-NEXT: fstpt (%rsp)
; FP16-NEXT: callq __truncxfhf2@PLT
; FP16-NEXT: addq $24, %rsp
; FP16-NEXT: retq
%r = fptrunc afn x86_fp80 %d to half
ret half %r
}
define half @fptrunc_afn_twostep(double %d) #0 {
; F16C-LABEL: fptrunc_afn_twostep:
; F16C: # %bb.0:
; F16C-NEXT: vcvtsd2ss %xmm0, %xmm0, %xmm0
; F16C-NEXT: vcvtps2ph $4, %xmm0, %xmm0
; F16C-NEXT: retq
;
; AVX-LABEL: fptrunc_afn_twostep:
; AVX: # %bb.0:
; AVX-NEXT: pushq %rax
; AVX-NEXT: callq __truncdfhf2@PLT
; AVX-NEXT: popq %rax
; AVX-NEXT: retq
;
; FP16-LABEL: fptrunc_afn_twostep:
; FP16: # %bb.0:
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm0
; FP16-NEXT: retq
%f = fptrunc contract afn double %d to float
%r = fptrunc contract afn float %f to half
ret half %r
}
define <4 x half> @fptrunc_afn_v4f64(<4 x double> %d) #0 {
; F16C-LABEL: fptrunc_afn_v4f64:
; F16C: # %bb.0:
; F16C-NEXT: vextractf128 $1, %ymm0, %xmm1
; F16C-NEXT: vcvtsd2ss %xmm1, %xmm1, %xmm2
; F16C-NEXT: vcvtps2ph $4, %xmm2, %xmm2
; F16C-NEXT: vshufpd {{.*#+}} xmm1 = xmm1[1,0]
; F16C-NEXT: vcvtsd2ss %xmm1, %xmm1, %xmm1
; F16C-NEXT: vcvtps2ph $4, %xmm1, %xmm1
; F16C-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; F16C-NEXT: vcvtsd2ss %xmm0, %xmm0, %xmm2
; F16C-NEXT: vcvtps2ph $4, %xmm2, %xmm2
; F16C-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; F16C-NEXT: vcvtsd2ss %xmm0, %xmm0, %xmm0
; F16C-NEXT: vcvtps2ph $4, %xmm0, %xmm0
; F16C-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; F16C-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
; F16C-NEXT: vzeroupper
; F16C-NEXT: retq
;
; AVX-LABEL: fptrunc_afn_v4f64:
; AVX: # %bb.0:
; AVX-NEXT: subq $88, %rsp
; AVX-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX-NEXT: vzeroupper
; AVX-NEXT: callq __truncdfhf2@PLT
; AVX-NEXT: vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX-NEXT: vzeroupper
; AVX-NEXT: callq __truncdfhf2@PLT
; AVX-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVX-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX-NEXT: vzeroupper
; AVX-NEXT: callq __truncdfhf2@PLT
; AVX-NEXT: vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
; AVX-NEXT: callq __truncdfhf2@PLT
; AVX-NEXT: vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
; AVX-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
; AVX-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; AVX-NEXT: vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
; AVX-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3]
; AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],zero,zero
; AVX-NEXT: addq $88, %rsp
; AVX-NEXT: retq
;
; FP16-LABEL: fptrunc_afn_v4f64:
; FP16: # %bb.0:
; FP16-NEXT: vextractf128 $1, %ymm0, %xmm1
; FP16-NEXT: vcvtsd2sh %xmm1, %xmm1, %xmm2
; FP16-NEXT: vshufpd {{.*#+}} xmm1 = xmm1[1,0]
; FP16-NEXT: vcvtsd2sh %xmm1, %xmm1, %xmm1
; FP16-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm2
; FP16-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0]
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm0
; FP16-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
; FP16-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm31, %xmm1
; FP16-NEXT: vpbroadcastw %xmm1, %xmm1
; FP16-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
; FP16-NEXT: vzeroupper
; FP16-NEXT: retq
%r = fptrunc afn <4 x double> %d to <4 x half>
ret <4 x half> %r
}
define half @fptrunc_contract(double %d) #0 {
; ALL-LABEL: fptrunc_contract:
; ALL: # %bb.0:
; ALL-NEXT: pushq %rax
; ALL-NEXT: callq __truncdfhf2@PLT
; ALL-NEXT: popq %rax
; ALL-NEXT: retq
;
; FP16-LABEL: fptrunc_contract:
; FP16: # %bb.0:
; FP16-NEXT: vcvtsd2sh %xmm0, %xmm0, %xmm0
; FP16-NEXT: retq
%r = fptrunc contract double %d to half
ret half %r
}
declare i16 @llvm.convert.to.fp16.f64(double)
declare i16 @llvm.convert.to.fp16.f80(x86_fp80)
attributes #0 = { nounwind readnone "use-soft-float"="false" }
attributes #1 = { nounwind readnone "use-soft-float"="false" }