; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq | FileCheck %s --check-prefixes=AVX,AVX512,AVX512DQ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bf16,+avx512vl | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BF16 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx10.2 | FileCheck %s --check-prefixes=AVX10_2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=X86 declare bfloat @llvm.maximumnum.bf16(bfloat, bfloat) declare float @llvm.maximumnum.f32(float, float) declare double @llvm.maximumnum.f64(double, double) declare fp128 @llvm.maximumnum.f128(fp128, fp128) declare bfloat @llvm.minimumnum.bf16(bfloat, bfloat) declare float @llvm.minimumnum.f32(float, float) declare double @llvm.minimumnum.f64(double, double) declare fp128 @llvm.minimumnum.f128(fp128, fp128) declare <2 x double> @llvm.minimumnum.v2f64(<2 x double>, <2 x double>) declare <4 x float> @llvm.maximumnum.v4f32(<4 x float>, <4 x float>) declare <4 x half> @llvm.maximumnum.v4f16(<4 x half>, <4 x half>) declare <4 x bfloat> @llvm.maximumnum.v4bf16(<4 x bfloat>, <4 x bfloat>) ; ; fmaximumnum ; define float @test_fmaximumnum(float %x, float %y) nounwind { ; SSE2-LABEL: test_fmaximumnum: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: maxss %xmm1, %xmm3 ; SSE2-NEXT: cmpunordss %xmm1, %xmm1 ; SSE2-NEXT: movaps %xmm1, %xmm2 ; SSE2-NEXT: andps %xmm0, %xmm2 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: andps %xmm3, %xmm0 ; SSE2-NEXT: andnps %xmm0, %xmm1 ; SSE2-NEXT: orps %xmm1, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum: ; AVX1: # %bb.0: ; AVX1-NEXT: vmaxss %xmm1, %xmm0, %xmm2 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm3 ; AVX1-NEXT: vandps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vorps %xmm2, %xmm0, %xmm2 ; AVX512F-NEXT: vmaxss %xmm1, %xmm0, %xmm3 ; AVX512F-NEXT: vandps %xmm3, %xmm2, %xmm2 ; AVX512F-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512F-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512F-NEXT: vmovaps %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vorps %xmm2, %xmm0, %xmm2 ; AVX512DQ-NEXT: vmaxss %xmm1, %xmm0, %xmm3 ; AVX512DQ-NEXT: vandps %xmm3, %xmm2, %xmm2 ; AVX512DQ-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512DQ-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512DQ-NEXT: vmovaps %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vmaxss %xmm1, %xmm0, %xmm2 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm2 = xmm2 & (xmm0 | m32bcst) ; AVX512BF16-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512BF16-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512BF16-NEXT: vmovaps %xmm2, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxss $21, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero ; X86-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm3 ; X86-NEXT: vandps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: flds (%esp) ; X86-NEXT: popl %eax ; X86-NEXT: retl %1 = tail call float @llvm.maximumnum.f32(float %x, float %y) ret float %1 } define <4 x float> @test_fmaximumnum_scalarize(<4 x float> %x, <4 x float> %y) { ; SSE2-LABEL: test_fmaximumnum_scalarize: ; SSE2: # %bb.0: ; SSE2-NEXT: maxps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_scalarize: ; AVX: # %bb.0: ; AVX-NEXT: vmaxps %xmm1, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_scalarize: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxps $21, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_scalarize: ; X86: # %bb.0: ; X86-NEXT: vmaxps %xmm1, %xmm0, %xmm0 ; X86-NEXT: retl %r = call nnan nsz <4 x float> @llvm.maximumnum.v4f32(<4 x float> %x, <4 x float> %y) ret <4 x float> %r } define float @test_fmaximumnum_nan0(float %x, float %y) { ; SSE2-LABEL: test_fmaximumnum_nan0: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_nan0: ; AVX: # %bb.0: ; AVX-NEXT: vmovaps %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_nan0: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovaps %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_nan0: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = tail call float @llvm.maximumnum.f32(float 0x7fff000000000000, float %y) ret float %1 } define float @test_fmaximumnum_nan1(float %x, float %y) { ; SSE2-LABEL: test_fmaximumnum_nan1: ; SSE2: # %bb.0: ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_nan1: ; AVX: # %bb.0: ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_nan1: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_nan1: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = tail call float @llvm.maximumnum.f32(float %x, float 0x7fff000000000000) ret float %1 } define float @test_fmaximumnum_nnan(float %x, float %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_nnan: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: addss %xmm1, %xmm2 ; SSE2-NEXT: subss %xmm1, %xmm0 ; SSE2-NEXT: movaps %xmm2, %xmm1 ; SSE2-NEXT: maxss %xmm0, %xmm1 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2 ; SSE2-NEXT: andps %xmm1, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_nnan: ; AVX1: # %bb.0: ; AVX1-NEXT: vaddss %xmm1, %xmm0, %xmm2 ; AVX1-NEXT: vsubss %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: vmaxss %xmm0, %xmm2, %xmm0 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 ; AVX1-NEXT: vandps %xmm0, %xmm1, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_nnan: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vaddss %xmm1, %xmm0, %xmm2 ; AVX512F-NEXT: vsubss %xmm1, %xmm0, %xmm0 ; AVX512F-NEXT: vmaxss %xmm0, %xmm2, %xmm0 ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vorps %xmm1, %xmm2, %xmm1 ; AVX512F-NEXT: vandps %xmm0, %xmm1, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_nnan: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vaddss %xmm1, %xmm0, %xmm2 ; AVX512DQ-NEXT: vsubss %xmm1, %xmm0, %xmm0 ; AVX512DQ-NEXT: vfpclassss $3, %xmm0, %k0 # k0 = isQuietNaN(xmm0) | isPositiveZero(xmm0) ; AVX512DQ-NEXT: kmovw %k0, %k1 ; AVX512DQ-NEXT: vmovaps %xmm2, %xmm1 ; AVX512DQ-NEXT: vmovss %xmm0, %xmm1, %xmm1 {%k1} ; AVX512DQ-NEXT: vmovss %xmm2, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: vmaxss %xmm1, %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_nnan: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vaddss %xmm1, %xmm0, %xmm2 ; AVX512BF16-NEXT: vsubss %xmm1, %xmm0, %xmm0 ; AVX512BF16-NEXT: vmaxss %xmm0, %xmm2, %xmm0 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm0 = xmm0 & (xmm2 | m32bcst) ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_nnan: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vaddss %xmm1, %xmm0, %xmm2 ; AVX10_2-NEXT: vsubss %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: vminmaxss $21, %xmm0, %xmm2 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_nnan: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero ; X86-NEXT: vaddss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vsubss %xmm0, %xmm1, %xmm0 ; X86-NEXT: vmaxss %xmm0, %xmm2, %xmm0 ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm2, %xmm1 ; X86-NEXT: vandps %xmm0, %xmm1, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: flds (%esp) ; X86-NEXT: popl %eax ; X86-NEXT: retl %1 = fadd nnan float %x, %y %2 = fsub nnan float %x, %y %3 = tail call float @llvm.maximumnum.f32(float %1, float %2) ret float %3 } define double @test_fmaximumnum_zero0(double %x, double %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_zero0: ; SSE2: # %bb.0: ; SSE2-NEXT: movapd %xmm1, %xmm0 ; SSE2-NEXT: xorpd %xmm1, %xmm1 ; SSE2-NEXT: maxsd %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_zero0: ; AVX: # %bb.0: ; AVX-NEXT: vxorpd %xmm0, %xmm0, %xmm0 ; AVX-NEXT: vmaxsd %xmm0, %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_zero0: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm0, %xmm0, %xmm0 ; AVX10_2-NEXT: vminmaxsd $21, %xmm0, %xmm1 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_zero0: ; X86: # %bb.0: ; X86-NEXT: pushl %ebp ; X86-NEXT: movl %esp, %ebp ; X86-NEXT: andl $-8, %esp ; X86-NEXT: subl $8, %esp ; X86-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero ; X86-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vmaxsd %xmm1, %xmm0, %xmm0 ; X86-NEXT: vmovsd %xmm0, (%esp) ; X86-NEXT: fldl (%esp) ; X86-NEXT: movl %ebp, %esp ; X86-NEXT: popl %ebp ; X86-NEXT: retl %1 = tail call double @llvm.maximumnum.f64(double 0.0, double %y) ret double %1 } define double @test_fmaximumnum_zero1(double %x, double %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_zero1: ; SSE2: # %bb.0: ; SSE2-NEXT: xorpd %xmm1, %xmm1 ; SSE2-NEXT: maxsd %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_zero1: ; AVX: # %bb.0: ; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX-NEXT: vmaxsd %xmm1, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_zero1: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vminmaxsd $21, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_zero1: ; X86: # %bb.0: ; X86-NEXT: pushl %ebp ; X86-NEXT: movl %esp, %ebp ; X86-NEXT: andl $-8, %esp ; X86-NEXT: subl $8, %esp ; X86-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero ; X86-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vmaxsd %xmm1, %xmm0, %xmm0 ; X86-NEXT: vmovsd %xmm0, (%esp) ; X86-NEXT: fldl (%esp) ; X86-NEXT: movl %ebp, %esp ; X86-NEXT: popl %ebp ; X86-NEXT: retl %1 = tail call double @llvm.maximumnum.f64(double %x, double 0.0) ret double %1 } define double @test_fmaximumnum_zero2(double %x, double %y) { ; SSE2-LABEL: test_fmaximumnum_zero2: ; SSE2: # %bb.0: ; SSE2-NEXT: xorps %xmm0, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_zero2: ; AVX: # %bb.0: ; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_zero2: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorps %xmm0, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_zero2: ; X86: # %bb.0: ; X86-NEXT: fldz ; X86-NEXT: retl %1 = tail call double @llvm.maximumnum.f64(double 0.0, double -0.0) ret double %1 } define float @test_fmaximumnum_nsz(float %x, float %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_nsz: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm2 ; SSE2-NEXT: cmpunordss %xmm1, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm3 ; SSE2-NEXT: andps %xmm0, %xmm3 ; SSE2-NEXT: maxss %xmm1, %xmm0 ; SSE2-NEXT: andnps %xmm0, %xmm2 ; SSE2-NEXT: orps %xmm3, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_nsz: ; AVX1: # %bb.0: ; AVX1-NEXT: vmaxss %xmm1, %xmm0, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512-LABEL: test_fmaximumnum_nsz: ; AVX512: # %bb.0: ; AVX512-NEXT: vmaxss %xmm1, %xmm0, %xmm2 ; AVX512-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512-NEXT: vmovaps %xmm2, %xmm0 ; AVX512-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_nsz: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxss $21, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_nsz: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero ; X86-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: flds (%esp) ; X86-NEXT: popl %eax ; X86-NEXT: retl %1 = tail call nsz float @llvm.maximumnum.f32(float %x, float %y) ret float %1 } define float @test_fmaximumnum_combine_cmps(float %x, float %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_combine_cmps: ; SSE2: # %bb.0: ; SSE2-NEXT: divss %xmm0, %xmm1 ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: maxss %xmm1, %xmm2 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: andps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_combine_cmps: ; AVX1: # %bb.0: ; AVX1-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX1-NEXT: vmaxss %xmm1, %xmm0, %xmm1 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX1-NEXT: vandps %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_combine_cmps: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vorps %xmm2, %xmm0, %xmm2 ; AVX512F-NEXT: vmaxss %xmm1, %xmm0, %xmm0 ; AVX512F-NEXT: vandps %xmm0, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_combine_cmps: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX512DQ-NEXT: vfpclassss $3, %xmm0, %k0 # k0 = isQuietNaN(xmm0) | isPositiveZero(xmm0) ; AVX512DQ-NEXT: kmovw %k0, %k1 ; AVX512DQ-NEXT: vmovaps %xmm1, %xmm2 ; AVX512DQ-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512DQ-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: vmaxss %xmm2, %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_combine_cmps: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX512BF16-NEXT: vmaxss %xmm1, %xmm0, %xmm1 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm0 = xmm1 & (xmm0 | m32bcst) ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_combine_cmps: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX10_2-NEXT: vminmaxss $21, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_combine_cmps: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero ; X86-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; X86-NEXT: vmaxss %xmm1, %xmm0, %xmm1 ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: vandps %xmm1, %xmm0, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: flds (%esp) ; X86-NEXT: popl %eax ; X86-NEXT: retl %1 = fdiv nnan float %y, %x %2 = tail call float @llvm.maximumnum.f32(float %x, float %1) ret float %2 } ; ; fminimumnum ; define float @test_fminimumnum(float %x, float %y) nounwind { ; SSE2-LABEL: test_fminimumnum: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: minss %xmm1, %xmm3 ; SSE2-NEXT: cmpunordss %xmm1, %xmm1 ; SSE2-NEXT: movaps %xmm1, %xmm2 ; SSE2-NEXT: andps %xmm0, %xmm2 ; SSE2-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: orps %xmm3, %xmm0 ; SSE2-NEXT: andnps %xmm0, %xmm1 ; SSE2-NEXT: orps %xmm1, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum: ; AVX1: # %bb.0: ; AVX1-NEXT: vminss %xmm1, %xmm0, %xmm2 ; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm3 ; AVX1-NEXT: vorps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm2 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512F-NEXT: vandps %xmm2, %xmm0, %xmm2 ; AVX512F-NEXT: vminss %xmm1, %xmm0, %xmm3 ; AVX512F-NEXT: vorps %xmm3, %xmm2, %xmm2 ; AVX512F-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512F-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512F-NEXT: vmovaps %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm2 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512DQ-NEXT: vandps %xmm2, %xmm0, %xmm2 ; AVX512DQ-NEXT: vminss %xmm1, %xmm0, %xmm3 ; AVX512DQ-NEXT: vorps %xmm3, %xmm2, %xmm2 ; AVX512DQ-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512DQ-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512DQ-NEXT: vmovaps %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vminss %xmm1, %xmm0, %xmm2 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm2 = xmm2 | (xmm0 & m32bcst) ; AVX512BF16-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512BF16-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512BF16-NEXT: vmovaps %xmm2, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxss $20, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero ; X86-NEXT: vminss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm3 ; X86-NEXT: vorps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: flds (%esp) ; X86-NEXT: popl %eax ; X86-NEXT: retl %1 = tail call float @llvm.minimumnum.f32(float %x, float %y) ret float %1 } define <2 x double> @test_fminimumnum_scalarize(<2 x double> %x, <2 x double> %y) { ; SSE2-LABEL: test_fminimumnum_scalarize: ; SSE2: # %bb.0: ; SSE2-NEXT: minpd %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_scalarize: ; AVX: # %bb.0: ; AVX-NEXT: vminpd %xmm1, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_scalarize: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxpd $20, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_scalarize: ; X86: # %bb.0: ; X86-NEXT: vminpd %xmm1, %xmm0, %xmm0 ; X86-NEXT: retl %r = call nnan nsz <2 x double> @llvm.minimumnum.v2f64(<2 x double> %x, <2 x double> %y) ret <2 x double> %r } define float @test_fminimumnum_nan0(float %x, float %y) { ; SSE2-LABEL: test_fminimumnum_nan0: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_nan0: ; AVX: # %bb.0: ; AVX-NEXT: vmovaps %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_nan0: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovaps %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_nan0: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = tail call float @llvm.minimumnum.f32(float 0x7fff000000000000, float %y) ret float %1 } define float @test_fminimumnum_nan1(float %x, float %y) { ; SSE2-LABEL: test_fminimumnum_nan1: ; SSE2: # %bb.0: ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_nan1: ; AVX: # %bb.0: ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_nan1: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_nan1: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = tail call float @llvm.minimumnum.f32(float %x, float 0x7fff000000000000) ret float %1 } define double @test_fminimumnum_nnan(double %x, double %y) nounwind { ; SSE2-LABEL: test_fminimumnum_nnan: ; SSE2: # %bb.0: ; SSE2-NEXT: movapd %xmm0, %xmm2 ; SSE2-NEXT: minsd %xmm1, %xmm2 ; SSE2-NEXT: andpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: orpd %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_nnan: ; AVX1: # %bb.0: ; AVX1-NEXT: vminsd %xmm1, %xmm0, %xmm1 ; AVX1-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX1-NEXT: vorpd %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_nnan: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vminsd %xmm1, %xmm0, %xmm1 ; AVX512F-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX512F-NEXT: vorpd %xmm1, %xmm0, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_nnan: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vfpclasssd $5, %xmm0, %k0 # k0 = isQuietNaN(xmm0) | isNegativeZero(xmm0) ; AVX512DQ-NEXT: kmovw %k0, %k1 ; AVX512DQ-NEXT: vmovapd %xmm1, %xmm2 ; AVX512DQ-NEXT: vmovsd %xmm0, %xmm2, %xmm2 {%k1} ; AVX512DQ-NEXT: vmovsd %xmm1, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: vminsd %xmm2, %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_nnan: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vminsd %xmm1, %xmm0, %xmm1 ; AVX512BF16-NEXT: vpternlogq {{.*#+}} xmm0 = (xmm0 & m64bcst) | xmm1 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_nnan: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxsd $20, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_nnan: ; X86: # %bb.0: ; X86-NEXT: pushl %ebp ; X86-NEXT: movl %esp, %ebp ; X86-NEXT: andl $-8, %esp ; X86-NEXT: subl $8, %esp ; X86-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero ; X86-NEXT: vminsd 16(%ebp), %xmm0, %xmm1 ; X86-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: vorpd %xmm1, %xmm0, %xmm0 ; X86-NEXT: vmovlpd %xmm0, (%esp) ; X86-NEXT: fldl (%esp) ; X86-NEXT: movl %ebp, %esp ; X86-NEXT: popl %ebp ; X86-NEXT: retl %1 = tail call nnan double @llvm.minimumnum.f64(double %x, double %y) ret double %1 } define double @test_fminimumnum_zero0(double %x, double %y) nounwind { ; SSE2-LABEL: test_fminimumnum_zero0: ; SSE2: # %bb.0: ; SSE2-NEXT: movapd %xmm1, %xmm0 ; SSE2-NEXT: minsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_zero0: ; AVX: # %bb.0: ; AVX-NEXT: vminsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_zero0: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxsd $20, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_zero0: ; X86: # %bb.0: ; X86-NEXT: pushl %ebp ; X86-NEXT: movl %esp, %ebp ; X86-NEXT: andl $-8, %esp ; X86-NEXT: subl $8, %esp ; X86-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero ; X86-NEXT: vminsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: vmovsd %xmm0, (%esp) ; X86-NEXT: fldl (%esp) ; X86-NEXT: movl %ebp, %esp ; X86-NEXT: popl %ebp ; X86-NEXT: retl %1 = tail call double @llvm.minimumnum.f64(double -0.0, double %y) ret double %1 } define double @test_fminimumnum_zero1(double %x, double %y) nounwind { ; SSE2-LABEL: test_fminimumnum_zero1: ; SSE2: # %bb.0: ; SSE2-NEXT: minsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_zero1: ; AVX: # %bb.0: ; AVX-NEXT: vminsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_zero1: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxsd $20, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_zero1: ; X86: # %bb.0: ; X86-NEXT: pushl %ebp ; X86-NEXT: movl %esp, %ebp ; X86-NEXT: andl $-8, %esp ; X86-NEXT: subl $8, %esp ; X86-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero ; X86-NEXT: vminsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: vmovsd %xmm0, (%esp) ; X86-NEXT: fldl (%esp) ; X86-NEXT: movl %ebp, %esp ; X86-NEXT: popl %ebp ; X86-NEXT: retl %1 = tail call double @llvm.minimumnum.f64(double %x, double -0.0) ret double %1 } define double @test_fminimumnum_zero2(double %x, double %y) { ; SSE2-LABEL: test_fminimumnum_zero2: ; SSE2: # %bb.0: ; SSE2-NEXT: movsd {{.*#+}} xmm0 = [-0.0E+0,0.0E+0] ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_zero2: ; AVX: # %bb.0: ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = [-0.0E+0,0.0E+0] ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_zero2: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovsd {{.*#+}} xmm0 = [-0.0E+0,0.0E+0] ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_zero2: ; X86: # %bb.0: ; X86-NEXT: fldz ; X86-NEXT: fchs ; X86-NEXT: retl %1 = tail call double @llvm.minimumnum.f64(double -0.0, double 0.0) ret double %1 } define float @test_fminimumnum_nsz(float %x, float %y) nounwind { ; SSE2-LABEL: test_fminimumnum_nsz: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm2 ; SSE2-NEXT: cmpunordss %xmm1, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm3 ; SSE2-NEXT: andps %xmm0, %xmm3 ; SSE2-NEXT: minss %xmm1, %xmm0 ; SSE2-NEXT: andnps %xmm0, %xmm2 ; SSE2-NEXT: orps %xmm3, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_nsz: ; AVX1: # %bb.0: ; AVX1-NEXT: vminss %xmm1, %xmm0, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512-LABEL: test_fminimumnum_nsz: ; AVX512: # %bb.0: ; AVX512-NEXT: vminss %xmm1, %xmm0, %xmm2 ; AVX512-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512-NEXT: vmovaps %xmm2, %xmm0 ; AVX512-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_nsz: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxss $20, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_nsz: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero ; X86-NEXT: vminss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: flds (%esp) ; X86-NEXT: popl %eax ; X86-NEXT: retl %1 = tail call nsz float @llvm.minimumnum.f32(float %x, float %y) ret float %1 } define float @test_fminimumnum_combine_cmps(float %x, float %y) nounwind { ; SSE2-LABEL: test_fminimumnum_combine_cmps: ; SSE2: # %bb.0: ; SSE2-NEXT: divss %xmm0, %xmm1 ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: minss %xmm1, %xmm2 ; SSE2-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: orps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_combine_cmps: ; AVX1: # %bb.0: ; AVX1-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX1-NEXT: vminss %xmm1, %xmm0, %xmm1 ; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX1-NEXT: vorps %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_combine_cmps: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm2 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512F-NEXT: vandps %xmm2, %xmm0, %xmm2 ; AVX512F-NEXT: vminss %xmm1, %xmm0, %xmm0 ; AVX512F-NEXT: vorps %xmm0, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_combine_cmps: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX512DQ-NEXT: vfpclassss $5, %xmm0, %k0 # k0 = isQuietNaN(xmm0) | isNegativeZero(xmm0) ; AVX512DQ-NEXT: kmovw %k0, %k1 ; AVX512DQ-NEXT: vmovaps %xmm1, %xmm2 ; AVX512DQ-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512DQ-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: vminss %xmm2, %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_combine_cmps: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX512BF16-NEXT: vminss %xmm1, %xmm0, %xmm1 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm0 = (xmm0 & m32bcst) | xmm1 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_combine_cmps: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; AVX10_2-NEXT: vminmaxss $20, %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_combine_cmps: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero ; X86-NEXT: vdivss %xmm0, %xmm1, %xmm1 ; X86-NEXT: vminss %xmm1, %xmm0, %xmm1 ; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: vorps %xmm1, %xmm0, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: flds (%esp) ; X86-NEXT: popl %eax ; X86-NEXT: retl %1 = fdiv nnan float %y, %x %2 = tail call float @llvm.minimumnum.f32(float %x, float %1) ret float %2 } define <2 x double> @test_fminimumnum_vector(<2 x double> %x, <2 x double> %y) { ; SSE2-LABEL: test_fminimumnum_vector: ; SSE2: # %bb.0: ; SSE2-NEXT: movapd %xmm0, %xmm3 ; SSE2-NEXT: minpd %xmm1, %xmm3 ; SSE2-NEXT: cmpunordpd %xmm1, %xmm1 ; SSE2-NEXT: movapd %xmm0, %xmm2 ; SSE2-NEXT: andpd %xmm1, %xmm2 ; SSE2-NEXT: andpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: orpd %xmm3, %xmm0 ; SSE2-NEXT: andnpd %xmm0, %xmm1 ; SSE2-NEXT: orpd %xmm1, %xmm2 ; SSE2-NEXT: movapd %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_vector: ; AVX1: # %bb.0: ; AVX1-NEXT: vminpd %xmm1, %xmm0, %xmm2 ; AVX1-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm3 ; AVX1-NEXT: vorpd %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordpd %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vblendvpd %xmm1, %xmm0, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_vector: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vminpd %xmm1, %xmm0, %xmm2 ; AVX512F-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm3 ; AVX512F-NEXT: vorpd %xmm2, %xmm3, %xmm2 ; AVX512F-NEXT: vcmpunordpd %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vblendvpd %xmm1, %xmm0, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_vector: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vminpd %xmm1, %xmm0, %xmm2 ; AVX512DQ-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm3 ; AVX512DQ-NEXT: vorpd %xmm2, %xmm3, %xmm2 ; AVX512DQ-NEXT: vcmpunordpd %xmm1, %xmm1, %xmm1 ; AVX512DQ-NEXT: vblendvpd %xmm1, %xmm0, %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_vector: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vminpd %xmm1, %xmm0, %xmm2 ; AVX512BF16-NEXT: vpternlogq {{.*#+}} xmm2 = xmm2 | (xmm0 & m64bcst) ; AVX512BF16-NEXT: vcmpunordpd %xmm1, %xmm1, %k1 ; AVX512BF16-NEXT: vmovapd %xmm0, %xmm2 {%k1} ; AVX512BF16-NEXT: vmovapd %xmm2, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_vector: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxpd $20, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_vector: ; X86: # %bb.0: ; X86-NEXT: vminpd %xmm1, %xmm0, %xmm2 ; X86-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm3 ; X86-NEXT: vorpd %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vblendvpd %xmm1, %xmm0, %xmm2, %xmm0 ; X86-NEXT: retl %r = call <2 x double> @llvm.minimumnum.v2f64(<2 x double> %x, <2 x double> %y) ret <2 x double> %r } define <4 x float> @test_fmaximumnum_vector(<4 x float> %x, <4 x float> %y) { ; SSE2-LABEL: test_fmaximumnum_vector: ; SSE2: # %bb.0: ; SSE2-NEXT: maxps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_vector: ; AVX: # %bb.0: ; AVX-NEXT: vmaxps %xmm1, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_vector: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxps $21, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_vector: ; X86: # %bb.0: ; X86-NEXT: vmaxps %xmm1, %xmm0, %xmm0 ; X86-NEXT: retl %r = call nnan nsz <4 x float> @llvm.maximumnum.v4f32(<4 x float> %x, <4 x float> %y) ret <4 x float> %r } define <2 x double> @test_fminimumnum_vector_zero(<2 x double> %x) { ; SSE2-LABEL: test_fminimumnum_vector_zero: ; SSE2: # %bb.0: ; SSE2-NEXT: xorpd %xmm1, %xmm1 ; SSE2-NEXT: minpd %xmm0, %xmm1 ; SSE2-NEXT: cmpunordpd %xmm0, %xmm0 ; SSE2-NEXT: andnpd %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_vector_zero: ; AVX1: # %bb.0: ; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; AVX1-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_vector_zero: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; AVX512F-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512F-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_vector_zero: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512DQ-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; AVX512DQ-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512DQ-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_vector_zero: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vcmpunordpd %xmm0, %xmm0, %k0 ; AVX512BF16-NEXT: knotw %k0, %k1 ; AVX512BF16-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512BF16-NEXT: vminpd %xmm0, %xmm1, %xmm0 {%k1} {z} ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_vector_zero: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vminmaxpd $20, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_vector_zero: ; X86: # %bb.0: ; X86-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; X86-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; X86-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; X86-NEXT: retl %r = call <2 x double> @llvm.minimumnum.v2f64(<2 x double> %x, <2 x double> ) ret <2 x double> %r } define <4 x float> @test_fmaximumnum_vector_signed_zero(<4 x float> %x) { ; SSE2-LABEL: test_fmaximumnum_vector_signed_zero: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; SSE2-NEXT: movaps %xmm1, %xmm2 ; SSE2-NEXT: maxps %xmm0, %xmm2 ; SSE2-NEXT: cmpunordps %xmm0, %xmm0 ; SSE2-NEXT: andps %xmm0, %xmm1 ; SSE2-NEXT: andnps %xmm2, %xmm0 ; SSE2-NEXT: orps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_vector_signed_zero: ; AVX1: # %bb.0: ; AVX1-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX1-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_vector_signed_zero: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512F-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; AVX512F-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; AVX512F-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_vector_signed_zero: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512DQ-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; AVX512DQ-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; AVX512DQ-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_vector_signed_zero: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vbroadcastss {{.*#+}} xmm2 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512BF16-NEXT: vmaxps %xmm0, %xmm2, %xmm1 ; AVX512BF16-NEXT: vcmpunordps %xmm0, %xmm0, %k1 ; AVX512BF16-NEXT: vmovaps %xmm2, %xmm1 {%k1} ; AVX512BF16-NEXT: vmovaps %xmm1, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_vector_signed_zero: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxps $21, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_vector_signed_zero: ; X86: # %bb.0: ; X86-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; X86-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; X86-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: retl %r = call <4 x float> @llvm.maximumnum.v4f32(<4 x float> %x, <4 x float> ) ret <4 x float> %r } define <2 x double> @test_fminimumnum_vector_partially_zero(<2 x double> %x) { ; SSE2-LABEL: test_fminimumnum_vector_partially_zero: ; SSE2: # %bb.0: ; SSE2-NEXT: xorpd %xmm1, %xmm1 ; SSE2-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; SSE2-NEXT: movapd %xmm1, %xmm2 ; SSE2-NEXT: minpd %xmm0, %xmm1 ; SSE2-NEXT: cmpunordpd %xmm0, %xmm0 ; SSE2-NEXT: andpd %xmm0, %xmm2 ; SSE2-NEXT: andnpd %xmm1, %xmm0 ; SSE2-NEXT: orpd %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_vector_partially_zero: ; AVX1: # %bb.0: ; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX1-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_vector_partially_zero: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX512F-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; AVX512F-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512F-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_vector_partially_zero: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512DQ-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX512DQ-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; AVX512DQ-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512DQ-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_vector_partially_zero: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512BF16-NEXT: vmovhpd {{.*#+}} xmm2 = xmm1[0],mem[0] ; AVX512BF16-NEXT: vminpd %xmm0, %xmm2, %xmm1 ; AVX512BF16-NEXT: vcmpunordpd %xmm0, %xmm0, %k1 ; AVX512BF16-NEXT: vmovapd %xmm2, %xmm1 {%k1} ; AVX512BF16-NEXT: vmovapd %xmm1, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_vector_partially_zero: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX10_2-NEXT: vminmaxpd $20, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_vector_partially_zero: ; X86: # %bb.0: ; X86-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; X86-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; X86-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: retl %r = call <2 x double> @llvm.minimumnum.v2f64(<2 x double> %x, <2 x double> ) ret <2 x double> %r } define <2 x double> @test_fminimumnum_vector_different_zeros(<2 x double> %x) { ; SSE2-LABEL: test_fminimumnum_vector_different_zeros: ; SSE2: # %bb.0: ; SSE2-NEXT: xorpd %xmm1, %xmm1 ; SSE2-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; SSE2-NEXT: movapd %xmm0, %xmm2 ; SSE2-NEXT: minpd %xmm1, %xmm2 ; SSE2-NEXT: andpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: orpd %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_vector_different_zeros: ; AVX1: # %bb.0: ; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX1-NEXT: vminpd %xmm1, %xmm0, %xmm1 ; AVX1-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX1-NEXT: vorpd %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_vector_different_zeros: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX512F-NEXT: vminpd %xmm1, %xmm0, %xmm1 ; AVX512F-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX512F-NEXT: vorpd %xmm1, %xmm0, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_vector_different_zeros: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512DQ-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX512DQ-NEXT: vminpd %xmm1, %xmm0, %xmm1 ; AVX512DQ-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX512DQ-NEXT: vorpd %xmm1, %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_vector_different_zeros: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512BF16-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX512BF16-NEXT: vminpd %xmm1, %xmm0, %xmm1 ; AVX512BF16-NEXT: vpternlogq {{.*#+}} xmm0 = (xmm0 & m64bcst) | xmm1 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_vector_different_zeros: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX10_2-NEXT: vminmaxpd $20, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_vector_different_zeros: ; X86: # %bb.0: ; X86-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; X86-NEXT: vminpd %xmm1, %xmm0, %xmm1 ; X86-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: vorpd %xmm1, %xmm0, %xmm0 ; X86-NEXT: retl %r = call <2 x double> @llvm.minimumnum.v2f64(<2 x double> %x, <2 x double> ) ret <2 x double> %r } define <4 x float> @test_fmaximumnum_vector_non_zero(<4 x float> %x) { ; SSE2-LABEL: test_fmaximumnum_vector_non_zero: ; SSE2: # %bb.0: ; SSE2-NEXT: maxps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_vector_non_zero: ; AVX: # %bb.0: ; AVX-NEXT: vmaxps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_vector_non_zero: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxps $21, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_vector_non_zero: ; X86: # %bb.0: ; X86-NEXT: vmaxps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: retl %r = call <4 x float> @llvm.maximumnum.v4f32(<4 x float> %x, <4 x float> ) ret <4 x float> %r } define <2 x double> @test_fminimumnum_vector_nan(<2 x double> %x) { ; SSE2-LABEL: test_fminimumnum_vector_nan: ; SSE2: # %bb.0: ; SSE2-NEXT: xorpd %xmm1, %xmm1 ; SSE2-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; SSE2-NEXT: movapd %xmm1, %xmm2 ; SSE2-NEXT: minpd %xmm0, %xmm1 ; SSE2-NEXT: cmpunordpd %xmm0, %xmm0 ; SSE2-NEXT: andpd %xmm0, %xmm2 ; SSE2-NEXT: andnpd %xmm1, %xmm0 ; SSE2-NEXT: orpd %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_vector_nan: ; AVX1: # %bb.0: ; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX1-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_vector_nan: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX512F-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; AVX512F-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512F-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_vector_nan: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512DQ-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX512DQ-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; AVX512DQ-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512DQ-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_vector_nan: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512BF16-NEXT: vmovhpd {{.*#+}} xmm2 = xmm1[0],mem[0] ; AVX512BF16-NEXT: vminpd %xmm0, %xmm2, %xmm1 ; AVX512BF16-NEXT: vcmpunordpd %xmm0, %xmm0, %k1 ; AVX512BF16-NEXT: vmovapd %xmm2, %xmm1 {%k1} ; AVX512BF16-NEXT: vmovapd %xmm1, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_vector_nan: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; AVX10_2-NEXT: vminmaxpd $20, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_vector_nan: ; X86: # %bb.0: ; X86-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] ; X86-NEXT: vminpd %xmm0, %xmm1, %xmm2 ; X86-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvpd %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: retl %r = call <2 x double> @llvm.minimumnum.v2f64(<2 x double> %x, <2 x double> ) ret <2 x double> %r } define <2 x double> @test_fminimumnum_vector_zero_first(<2 x double> %x) { ; SSE2-LABEL: test_fminimumnum_vector_zero_first: ; SSE2: # %bb.0: ; SSE2-NEXT: xorpd %xmm1, %xmm1 ; SSE2-NEXT: minpd %xmm0, %xmm1 ; SSE2-NEXT: cmpunordpd %xmm0, %xmm0 ; SSE2-NEXT: andnpd %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_vector_zero_first: ; AVX1: # %bb.0: ; AVX1-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; AVX1-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_vector_zero_first: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; AVX512F-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512F-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_vector_zero_first: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512DQ-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; AVX512DQ-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; AVX512DQ-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_vector_zero_first: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vcmpunordpd %xmm0, %xmm0, %k0 ; AVX512BF16-NEXT: knotw %k0, %k1 ; AVX512BF16-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX512BF16-NEXT: vminpd %xmm0, %xmm1, %xmm0 {%k1} {z} ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_vector_zero_first: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vminmaxpd $20, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_vector_zero_first: ; X86: # %bb.0: ; X86-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; X86-NEXT: vminpd %xmm0, %xmm1, %xmm1 ; X86-NEXT: vcmpunordpd %xmm0, %xmm0, %xmm0 ; X86-NEXT: vandnpd %xmm1, %xmm0, %xmm0 ; X86-NEXT: retl %r = call <2 x double> @llvm.minimumnum.v2f64(<2 x double> , <2 x double> %x) ret <2 x double> %r } define <2 x double> @test_fminimumnum_vector_signed_zero(<2 x double> %x) { ; SSE2-LABEL: test_fminimumnum_vector_signed_zero: ; SSE2: # %bb.0: ; SSE2-NEXT: minpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_vector_signed_zero: ; AVX1: # %bb.0: ; AVX1-NEXT: vminpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_vector_signed_zero: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vminpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_vector_signed_zero: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vminpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_vector_signed_zero: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vminpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_vector_signed_zero: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxpd $20, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_vector_signed_zero: ; X86: # %bb.0: ; X86-NEXT: vminpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0 ; X86-NEXT: retl %r = call <2 x double> @llvm.minimumnum.v2f64(<2 x double> %x, <2 x double> ) ret <2 x double> %r } define <4 x float> @test_fmaximumnum_vector_signed_zero_first(<4 x float> %x) { ; SSE2-LABEL: test_fmaximumnum_vector_signed_zero_first: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; SSE2-NEXT: movaps %xmm1, %xmm2 ; SSE2-NEXT: maxps %xmm0, %xmm2 ; SSE2-NEXT: cmpunordps %xmm0, %xmm0 ; SSE2-NEXT: andps %xmm0, %xmm1 ; SSE2-NEXT: andnps %xmm2, %xmm0 ; SSE2-NEXT: orps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_vector_signed_zero_first: ; AVX1: # %bb.0: ; AVX1-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX1-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_vector_signed_zero_first: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512F-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; AVX512F-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; AVX512F-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_vector_signed_zero_first: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512DQ-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; AVX512DQ-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; AVX512DQ-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_vector_signed_zero_first: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vbroadcastss {{.*#+}} xmm2 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512BF16-NEXT: vmaxps %xmm0, %xmm2, %xmm1 ; AVX512BF16-NEXT: vcmpunordps %xmm0, %xmm0, %k1 ; AVX512BF16-NEXT: vmovaps %xmm2, %xmm1 {%k1} ; AVX512BF16-NEXT: vmovaps %xmm1, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_vector_signed_zero_first: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxps $21, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_vector_signed_zero_first: ; X86: # %bb.0: ; X86-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; X86-NEXT: vmaxps %xmm0, %xmm1, %xmm2 ; X86-NEXT: vcmpunordps %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: retl %r = call <4 x float> @llvm.maximumnum.v4f32(<4 x float> , <4 x float> %x) ret <4 x float> %r } define <4 x float> @test_fmaximumnum_vector_zero(<4 x float> %x) { ; SSE2-LABEL: test_fmaximumnum_vector_zero: ; SSE2: # %bb.0: ; SSE2-NEXT: xorps %xmm1, %xmm1 ; SSE2-NEXT: maxps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_vector_zero: ; AVX: # %bb.0: ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 ; AVX-NEXT: vmaxps %xmm1, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_vector_zero: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vxorpd %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vminmaxps $21, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_vector_zero: ; X86: # %bb.0: ; X86-NEXT: vxorps %xmm1, %xmm1, %xmm1 ; X86-NEXT: vmaxps %xmm1, %xmm0, %xmm0 ; X86-NEXT: retl %r = call <4 x float> @llvm.maximumnum.v4f32(<4 x float> %x, <4 x float> ) ret <4 x float> %r } ; PR77805: Check that signed zeroes are handled correctly in this case (FIXME) define <4 x float> @test_fmaximumnum_v4f32_splat(<4 x float> %x, float %y) { ; SSE2-LABEL: test_fmaximumnum_v4f32_splat: ; SSE2: # %bb.0: ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0,0,0] ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: maxps %xmm1, %xmm3 ; SSE2-NEXT: cmpunordps %xmm1, %xmm1 ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: andps %xmm1, %xmm2 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 ; SSE2-NEXT: andps %xmm3, %xmm0 ; SSE2-NEXT: andnps %xmm0, %xmm1 ; SSE2-NEXT: orps %xmm1, %xmm2 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_v4f32_splat: ; AVX1: # %bb.0: ; AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[0,0,0,0] ; AVX1-NEXT: vmaxps %xmm1, %xmm0, %xmm2 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm3 ; AVX1-NEXT: vandps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordps %xmm1, %xmm1, %xmm1 ; AVX1-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_v4f32_splat: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vbroadcastss %xmm1, %xmm1 ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vorps %xmm2, %xmm0, %xmm2 ; AVX512F-NEXT: vmaxps %xmm1, %xmm0, %xmm3 ; AVX512F-NEXT: vandps %xmm3, %xmm2, %xmm2 ; AVX512F-NEXT: vcmpunordps %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_v4f32_splat: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vbroadcastss %xmm1, %xmm1 ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vorps %xmm2, %xmm0, %xmm2 ; AVX512DQ-NEXT: vmaxps %xmm1, %xmm0, %xmm3 ; AVX512DQ-NEXT: vandps %xmm3, %xmm2, %xmm2 ; AVX512DQ-NEXT: vcmpunordps %xmm1, %xmm1, %xmm1 ; AVX512DQ-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_v4f32_splat: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vbroadcastss %xmm1, %xmm2 ; AVX512BF16-NEXT: vmaxps %xmm2, %xmm0, %xmm1 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm1 = xmm1 & (xmm0 | m32bcst) ; AVX512BF16-NEXT: vcmpunordps %xmm2, %xmm2, %k1 ; AVX512BF16-NEXT: vmovaps %xmm0, %xmm1 {%k1} ; AVX512BF16-NEXT: vmovaps %xmm1, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_v4f32_splat: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vbroadcastss %xmm1, %xmm1 ; AVX10_2-NEXT: vminmaxps $21, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_v4f32_splat: ; X86: # %bb.0: ; X86-NEXT: vbroadcastss {{[0-9]+}}(%esp), %xmm1 ; X86-NEXT: vmaxps %xmm1, %xmm0, %xmm2 ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm3 ; X86-NEXT: vandps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordps %xmm1, %xmm1, %xmm1 ; X86-NEXT: vblendvps %xmm1, %xmm0, %xmm2, %xmm0 ; X86-NEXT: retl %splatinsert = insertelement <4 x float> poison, float %y, i64 0 %vec = shufflevector <4 x float> %splatinsert, <4 x float> poison, <4 x i32> zeroinitializer %r = call <4 x float> @llvm.maximumnum.v4f32(<4 x float> %x, <4 x float> %vec) readnone ret <4 x float> %r } define <4 x half> @test_fmaximumnum_v4f16(<4 x half> %x, <4 x half> %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_v4f16: ; SSE2: # %bb.0: ; SSE2-NEXT: subq $168, %rsp ; SSE2-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movdqa %xmm0, (%rsp) # 16-byte Spill ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: psrlq $48, %xmm2 ; SSE2-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movdqa %xmm0, %xmm2 ; SSE2-NEXT: psrlq $48, %xmm2 ; SSE2-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[1,1] ; SSE2-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movdqa %xmm0, %xmm2 ; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm0[1,1] ; SSE2-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: psrld $16, %xmm1 ; SSE2-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: psrld $16, %xmm0 ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN] ; SSE2-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movdqa %xmm0, %xmm2 ; SSE2-NEXT: por %xmm1, %xmm2 ; SSE2-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload ; SSE2-NEXT: movaps %xmm3, %xmm1 ; SSE2-NEXT: maxss %xmm0, %xmm1 ; SSE2-NEXT: andps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: andnps %xmm1, %xmm2 ; SSE2-NEXT: andps %xmm3, %xmm0 ; SSE2-NEXT: orps %xmm2, %xmm0 ; SSE2-NEXT: callq __truncsfhf2@PLT ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload ; SSE2-NEXT: movaps %xmm3, %xmm1 ; SSE2-NEXT: maxss %xmm0, %xmm1 ; SSE2-NEXT: andps (%rsp), %xmm1 # 16-byte Folded Reload ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: andnps %xmm1, %xmm2 ; SSE2-NEXT: andps %xmm3, %xmm0 ; SSE2-NEXT: orps %xmm2, %xmm0 ; SSE2-NEXT: callq __truncsfhf2@PLT ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movaps (%rsp), %xmm3 # 16-byte Reload ; SSE2-NEXT: movaps %xmm3, %xmm1 ; SSE2-NEXT: maxss %xmm0, %xmm1 ; SSE2-NEXT: andps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: andnps %xmm1, %xmm2 ; SSE2-NEXT: andps %xmm3, %xmm0 ; SSE2-NEXT: orps %xmm2, %xmm0 ; SSE2-NEXT: callq __truncsfhf2@PLT ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: orps %xmm1, %xmm0 ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: callq __extendhfsf2@PLT ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload ; SSE2-NEXT: movaps %xmm3, %xmm1 ; SSE2-NEXT: maxss %xmm0, %xmm1 ; SSE2-NEXT: andps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm2 ; SSE2-NEXT: andnps %xmm1, %xmm2 ; SSE2-NEXT: andps %xmm3, %xmm0 ; SSE2-NEXT: orps %xmm2, %xmm0 ; SSE2-NEXT: callq __truncsfhf2@PLT ; SSE2-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; SSE2-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] ; SSE2-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload ; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3] ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] ; SSE2-NEXT: addq $168, %rsp ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_v4f16: ; AVX1: # %bb.0: ; AVX1-NEXT: subq $152, %rsp ; AVX1-NEXT: vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovdqa %xmm0, %xmm2 ; AVX1-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill ; AVX1-NEXT: vpsrlq $48, %xmm1, %xmm0 ; AVX1-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vpsrlq $48, %xmm2, %xmm0 ; AVX1-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovshdup {{.*#+}} xmm0 = xmm1[1,1,3,3] ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovshdup {{.*#+}} xmm0 = xmm2[1,1,3,3] ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vpsrld $16, %xmm1, %xmm0 ; AVX1-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vpsrld $16, %xmm2, %xmm0 ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1 ; AVX1-NEXT: vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload ; AVX1-NEXT: vmaxss %xmm0, %xmm2, %xmm1 ; AVX1-NEXT: vandps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm2, %xmm1, %xmm0 ; AVX1-NEXT: callq __truncsfhf2@PLT ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1 ; AVX1-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload ; AVX1-NEXT: vmaxss %xmm0, %xmm2, %xmm1 ; AVX1-NEXT: vandps (%rsp), %xmm1, %xmm1 # 16-byte Folded Reload ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm2, %xmm1, %xmm0 ; AVX1-NEXT: callq __truncsfhf2@PLT ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1 ; AVX1-NEXT: vmovaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovaps (%rsp), %xmm2 # 16-byte Reload ; AVX1-NEXT: vmaxss %xmm0, %xmm2, %xmm1 ; AVX1-NEXT: vandps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm2, %xmm1, %xmm0 ; AVX1-NEXT: callq __truncsfhf2@PLT ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1 ; AVX1-NEXT: vmovaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX1-NEXT: callq __extendhfsf2@PLT ; AVX1-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload ; AVX1-NEXT: vmaxss %xmm0, %xmm2, %xmm1 ; AVX1-NEXT: vandps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm2, %xmm1, %xmm0 ; AVX1-NEXT: callq __truncsfhf2@PLT ; AVX1-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX1-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] ; AVX1-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX1-NEXT: vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload ; AVX1-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3] ; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],zero,zero ; AVX1-NEXT: addq $152, %rsp ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_v4f16: ; AVX512F: # %bb.0: ; AVX512F-NEXT: vpsrldq {{.*#+}} xmm2 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm3 ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vorps %xmm2, %xmm3, %xmm4 ; AVX512F-NEXT: vpsrldq {{.*#+}} xmm5 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512F-NEXT: vmaxss %xmm5, %xmm3, %xmm6 ; AVX512F-NEXT: vandps %xmm6, %xmm4, %xmm4 ; AVX512F-NEXT: vcmpunordss %xmm5, %xmm5, %k1 ; AVX512F-NEXT: vmovss %xmm3, %xmm4, %xmm4 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm4, %xmm3 ; AVX512F-NEXT: vshufps {{.*#+}} xmm4 = xmm0[3,3,3,3] ; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512F-NEXT: vorps %xmm2, %xmm4, %xmm5 ; AVX512F-NEXT: vpshufd {{.*#+}} xmm6 = xmm1[3,3,3,3] ; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512F-NEXT: vmaxss %xmm6, %xmm4, %xmm7 ; AVX512F-NEXT: vandps %xmm7, %xmm5, %xmm5 ; AVX512F-NEXT: vcmpunordss %xmm6, %xmm6, %k1 ; AVX512F-NEXT: vmovss %xmm4, %xmm5, %xmm5 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm5, %xmm4 ; AVX512F-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3] ; AVX512F-NEXT: vpsrldq {{.*#+}} xmm4 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512F-NEXT: vorps %xmm2, %xmm4, %xmm5 ; AVX512F-NEXT: vpsrldq {{.*#+}} xmm6 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512F-NEXT: vmaxss %xmm6, %xmm4, %xmm7 ; AVX512F-NEXT: vandps %xmm7, %xmm5, %xmm5 ; AVX512F-NEXT: vcmpunordss %xmm6, %xmm6, %k1 ; AVX512F-NEXT: vmovss %xmm4, %xmm5, %xmm5 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm5, %xmm4 ; AVX512F-NEXT: vshufpd {{.*#+}} xmm5 = xmm0[1,0] ; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512F-NEXT: vorps %xmm2, %xmm5, %xmm6 ; AVX512F-NEXT: vshufpd {{.*#+}} xmm7 = xmm1[1,0] ; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7 ; AVX512F-NEXT: vmaxss %xmm7, %xmm5, %xmm8 ; AVX512F-NEXT: vandps %xmm6, %xmm8, %xmm6 ; AVX512F-NEXT: vcmpunordss %xmm7, %xmm7, %k1 ; AVX512F-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm6, %xmm5 ; AVX512F-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3] ; AVX512F-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1] ; AVX512F-NEXT: vpsrlq $48, %xmm0, %xmm4 ; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512F-NEXT: vorps %xmm2, %xmm4, %xmm5 ; AVX512F-NEXT: vpsrlq $48, %xmm1, %xmm6 ; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512F-NEXT: vmaxss %xmm6, %xmm4, %xmm7 ; AVX512F-NEXT: vandps %xmm7, %xmm5, %xmm5 ; AVX512F-NEXT: vcmpunordss %xmm6, %xmm6, %k1 ; AVX512F-NEXT: vmovss %xmm4, %xmm5, %xmm5 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm5, %xmm4 ; AVX512F-NEXT: vmovshdup {{.*#+}} xmm5 = xmm0[1,1,3,3] ; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512F-NEXT: vorps %xmm2, %xmm5, %xmm6 ; AVX512F-NEXT: vmovshdup {{.*#+}} xmm7 = xmm1[1,1,3,3] ; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7 ; AVX512F-NEXT: vmaxss %xmm7, %xmm5, %xmm8 ; AVX512F-NEXT: vandps %xmm6, %xmm8, %xmm6 ; AVX512F-NEXT: vcmpunordss %xmm7, %xmm7, %k1 ; AVX512F-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm6, %xmm5 ; AVX512F-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3] ; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm5 ; AVX512F-NEXT: vorps %xmm2, %xmm5, %xmm6 ; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm7 ; AVX512F-NEXT: vmaxss %xmm7, %xmm5, %xmm8 ; AVX512F-NEXT: vandps %xmm6, %xmm8, %xmm6 ; AVX512F-NEXT: vcmpunordss %xmm7, %xmm7, %k1 ; AVX512F-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm6, %xmm5 ; AVX512F-NEXT: vpsrld $16, %xmm0, %xmm0 ; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0 ; AVX512F-NEXT: vorps %xmm2, %xmm0, %xmm2 ; AVX512F-NEXT: vpsrld $16, %xmm1, %xmm1 ; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1 ; AVX512F-NEXT: vmaxss %xmm1, %xmm0, %xmm6 ; AVX512F-NEXT: vandps %xmm6, %xmm2, %xmm2 ; AVX512F-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512F-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm0 ; AVX512F-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3] ; AVX512F-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1] ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0] ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_v4f16: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: vpsrldq {{.*#+}} xmm2 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512DQ-NEXT: vcvtph2ps %xmm2, %xmm3 ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vorps %xmm2, %xmm3, %xmm4 ; AVX512DQ-NEXT: vpsrldq {{.*#+}} xmm5 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512DQ-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512DQ-NEXT: vmaxss %xmm5, %xmm3, %xmm6 ; AVX512DQ-NEXT: vandps %xmm6, %xmm4, %xmm4 ; AVX512DQ-NEXT: vcmpunordss %xmm5, %xmm5, %k1 ; AVX512DQ-NEXT: vmovss %xmm3, %xmm4, %xmm4 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm4, %xmm3 ; AVX512DQ-NEXT: vshufps {{.*#+}} xmm4 = xmm0[3,3,3,3] ; AVX512DQ-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512DQ-NEXT: vorps %xmm2, %xmm4, %xmm5 ; AVX512DQ-NEXT: vpshufd {{.*#+}} xmm6 = xmm1[3,3,3,3] ; AVX512DQ-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512DQ-NEXT: vmaxss %xmm6, %xmm4, %xmm7 ; AVX512DQ-NEXT: vandps %xmm7, %xmm5, %xmm5 ; AVX512DQ-NEXT: vcmpunordss %xmm6, %xmm6, %k1 ; AVX512DQ-NEXT: vmovss %xmm4, %xmm5, %xmm5 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm5, %xmm4 ; AVX512DQ-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3] ; AVX512DQ-NEXT: vpsrldq {{.*#+}} xmm4 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512DQ-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512DQ-NEXT: vorps %xmm2, %xmm4, %xmm5 ; AVX512DQ-NEXT: vpsrldq {{.*#+}} xmm6 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512DQ-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512DQ-NEXT: vmaxss %xmm6, %xmm4, %xmm7 ; AVX512DQ-NEXT: vandps %xmm7, %xmm5, %xmm5 ; AVX512DQ-NEXT: vcmpunordss %xmm6, %xmm6, %k1 ; AVX512DQ-NEXT: vmovss %xmm4, %xmm5, %xmm5 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm5, %xmm4 ; AVX512DQ-NEXT: vshufpd {{.*#+}} xmm5 = xmm0[1,0] ; AVX512DQ-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512DQ-NEXT: vorps %xmm2, %xmm5, %xmm6 ; AVX512DQ-NEXT: vshufpd {{.*#+}} xmm7 = xmm1[1,0] ; AVX512DQ-NEXT: vcvtph2ps %xmm7, %xmm7 ; AVX512DQ-NEXT: vmaxss %xmm7, %xmm5, %xmm8 ; AVX512DQ-NEXT: vandps %xmm6, %xmm8, %xmm6 ; AVX512DQ-NEXT: vcmpunordss %xmm7, %xmm7, %k1 ; AVX512DQ-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm6, %xmm5 ; AVX512DQ-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3] ; AVX512DQ-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1] ; AVX512DQ-NEXT: vpsrlq $48, %xmm0, %xmm4 ; AVX512DQ-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512DQ-NEXT: vorps %xmm2, %xmm4, %xmm5 ; AVX512DQ-NEXT: vpsrlq $48, %xmm1, %xmm6 ; AVX512DQ-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512DQ-NEXT: vmaxss %xmm6, %xmm4, %xmm7 ; AVX512DQ-NEXT: vandps %xmm7, %xmm5, %xmm5 ; AVX512DQ-NEXT: vcmpunordss %xmm6, %xmm6, %k1 ; AVX512DQ-NEXT: vmovss %xmm4, %xmm5, %xmm5 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm5, %xmm4 ; AVX512DQ-NEXT: vmovshdup {{.*#+}} xmm5 = xmm0[1,1,3,3] ; AVX512DQ-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512DQ-NEXT: vorps %xmm2, %xmm5, %xmm6 ; AVX512DQ-NEXT: vmovshdup {{.*#+}} xmm7 = xmm1[1,1,3,3] ; AVX512DQ-NEXT: vcvtph2ps %xmm7, %xmm7 ; AVX512DQ-NEXT: vmaxss %xmm7, %xmm5, %xmm8 ; AVX512DQ-NEXT: vandps %xmm6, %xmm8, %xmm6 ; AVX512DQ-NEXT: vcmpunordss %xmm7, %xmm7, %k1 ; AVX512DQ-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm6, %xmm5 ; AVX512DQ-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3] ; AVX512DQ-NEXT: vcvtph2ps %xmm0, %xmm5 ; AVX512DQ-NEXT: vorps %xmm2, %xmm5, %xmm6 ; AVX512DQ-NEXT: vcvtph2ps %xmm1, %xmm7 ; AVX512DQ-NEXT: vmaxss %xmm7, %xmm5, %xmm8 ; AVX512DQ-NEXT: vandps %xmm6, %xmm8, %xmm6 ; AVX512DQ-NEXT: vcmpunordss %xmm7, %xmm7, %k1 ; AVX512DQ-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm6, %xmm5 ; AVX512DQ-NEXT: vpsrld $16, %xmm0, %xmm0 ; AVX512DQ-NEXT: vcvtph2ps %xmm0, %xmm0 ; AVX512DQ-NEXT: vorps %xmm2, %xmm0, %xmm2 ; AVX512DQ-NEXT: vpsrld $16, %xmm1, %xmm1 ; AVX512DQ-NEXT: vcvtph2ps %xmm1, %xmm1 ; AVX512DQ-NEXT: vmaxss %xmm1, %xmm0, %xmm6 ; AVX512DQ-NEXT: vandps %xmm6, %xmm2, %xmm2 ; AVX512DQ-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512DQ-NEXT: vmovss %xmm0, %xmm2, %xmm2 {%k1} ; AVX512DQ-NEXT: vcvtps2ph $4, %xmm2, %xmm0 ; AVX512DQ-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3] ; AVX512DQ-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1] ; AVX512DQ-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0] ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_v4f16: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vpsrldq {{.*#+}} xmm2 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512BF16-NEXT: vcvtph2ps %xmm2, %xmm3 ; AVX512BF16-NEXT: vpsrldq {{.*#+}} xmm2 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512BF16-NEXT: vcvtph2ps %xmm2, %xmm4 ; AVX512BF16-NEXT: vmaxss %xmm3, %xmm4, %xmm5 ; AVX512BF16-NEXT: vpbroadcastd {{.*#+}} xmm2 = [NaN,NaN,NaN,NaN] ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm5 = xmm5 & (xmm4 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm3, %xmm3, %k1 ; AVX512BF16-NEXT: vmovss %xmm4, %xmm5, %xmm5 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm5, %xmm3 ; AVX512BF16-NEXT: vshufps {{.*#+}} xmm4 = xmm1[3,3,3,3] ; AVX512BF16-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512BF16-NEXT: vshufps {{.*#+}} xmm5 = xmm0[3,3,3,3] ; AVX512BF16-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512BF16-NEXT: vmaxss %xmm4, %xmm5, %xmm6 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm6 = xmm6 & (xmm5 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm4, %xmm4, %k1 ; AVX512BF16-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm6, %xmm4 ; AVX512BF16-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3] ; AVX512BF16-NEXT: vpsrldq {{.*#+}} xmm4 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512BF16-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512BF16-NEXT: vpsrldq {{.*#+}} xmm5 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX512BF16-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512BF16-NEXT: vmaxss %xmm4, %xmm5, %xmm6 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm6 = xmm6 & (xmm5 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm4, %xmm4, %k1 ; AVX512BF16-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm6, %xmm4 ; AVX512BF16-NEXT: vshufpd {{.*#+}} xmm5 = xmm1[1,0] ; AVX512BF16-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512BF16-NEXT: vshufpd {{.*#+}} xmm6 = xmm0[1,0] ; AVX512BF16-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512BF16-NEXT: vmaxss %xmm5, %xmm6, %xmm7 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm7 = xmm7 & (xmm6 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm5, %xmm5, %k1 ; AVX512BF16-NEXT: vmovss %xmm6, %xmm7, %xmm7 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm7, %xmm5 ; AVX512BF16-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3] ; AVX512BF16-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1] ; AVX512BF16-NEXT: vpsrlq $48, %xmm1, %xmm4 ; AVX512BF16-NEXT: vcvtph2ps %xmm4, %xmm4 ; AVX512BF16-NEXT: vpsrlq $48, %xmm0, %xmm5 ; AVX512BF16-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512BF16-NEXT: vmaxss %xmm4, %xmm5, %xmm6 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm6 = xmm6 & (xmm5 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm4, %xmm4, %k1 ; AVX512BF16-NEXT: vmovss %xmm5, %xmm6, %xmm6 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm6, %xmm4 ; AVX512BF16-NEXT: vmovshdup {{.*#+}} xmm5 = xmm1[1,1,3,3] ; AVX512BF16-NEXT: vcvtph2ps %xmm5, %xmm5 ; AVX512BF16-NEXT: vmovshdup {{.*#+}} xmm6 = xmm0[1,1,3,3] ; AVX512BF16-NEXT: vcvtph2ps %xmm6, %xmm6 ; AVX512BF16-NEXT: vmaxss %xmm5, %xmm6, %xmm7 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm7 = xmm7 & (xmm6 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm5, %xmm5, %k1 ; AVX512BF16-NEXT: vmovss %xmm6, %xmm7, %xmm7 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm7, %xmm5 ; AVX512BF16-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3] ; AVX512BF16-NEXT: vcvtph2ps %xmm1, %xmm5 ; AVX512BF16-NEXT: vcvtph2ps %xmm0, %xmm6 ; AVX512BF16-NEXT: vmaxss %xmm5, %xmm6, %xmm7 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm7 = xmm7 & (xmm6 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm5, %xmm5, %k1 ; AVX512BF16-NEXT: vmovss %xmm6, %xmm7, %xmm7 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm7, %xmm5 ; AVX512BF16-NEXT: vpsrld $16, %xmm1, %xmm1 ; AVX512BF16-NEXT: vcvtph2ps %xmm1, %xmm1 ; AVX512BF16-NEXT: vpsrld $16, %xmm0, %xmm0 ; AVX512BF16-NEXT: vcvtph2ps %xmm0, %xmm0 ; AVX512BF16-NEXT: vmaxss %xmm1, %xmm0, %xmm6 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm6 = xmm6 & (xmm0 | xmm2) ; AVX512BF16-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512BF16-NEXT: vmovss %xmm0, %xmm6, %xmm6 {%k1} ; AVX512BF16-NEXT: vcvtps2ph $4, %xmm6, %xmm0 ; AVX512BF16-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3] ; AVX512BF16-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1] ; AVX512BF16-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0] ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_v4f16: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxph $21, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_v4f16: ; X86: # %bb.0: ; X86-NEXT: subl $164, %esp ; X86-NEXT: vmovdqa %xmm0, %xmm2 ; X86-NEXT: vmovdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vpsrlq $48, %xmm0, %xmm0 ; X86-NEXT: vmovdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovshdup {{.*#+}} xmm0 = xmm2[1,1,3,3] ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovshdup {{.*#+}} xmm0 = xmm1[1,1,3,3] ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vpsrlq $48, %xmm1, %xmm0 ; X86-NEXT: vmovdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vpsrld $16, %xmm2, %xmm0 ; X86-NEXT: vmovdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vpsrld $16, %xmm1, %xmm0 ; X86-NEXT: vmovdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vpextrw $0, %xmm1, (%esp) ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: fstpt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Spill ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vpextrw $0, %xmm0, (%esp) ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: fstpt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Spill ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vpextrw $0, %xmm0, (%esp) ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vpextrw $0, %xmm0, (%esp) ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: fldt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Reload ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm1 ; X86-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero ; X86-NEXT: vmaxss %xmm2, %xmm0, %xmm3 ; X86-NEXT: vandps %xmm3, %xmm1, %xmm1 ; X86-NEXT: vcmpunordss %xmm2, %xmm2, %xmm2 ; X86-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: vmovups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: fldt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Reload ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm1 ; X86-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero ; X86-NEXT: vmaxss %xmm2, %xmm0, %xmm3 ; X86-NEXT: vandps %xmm3, %xmm1, %xmm1 ; X86-NEXT: vcmpunordss %xmm2, %xmm2, %xmm2 ; X86-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: calll __truncsfhf2 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: calll __truncsfhf2 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vpextrw $0, %xmm0, (%esp) ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: fstpt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Spill ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vpextrw $0, %xmm0, (%esp) ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: fstpt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Spill ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vpextrw $0, %xmm0, (%esp) ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vpextrw $0, %xmm0, (%esp) ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: fldt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Reload ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm1 ; X86-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero ; X86-NEXT: vmaxss %xmm2, %xmm0, %xmm3 ; X86-NEXT: vandps %xmm3, %xmm1, %xmm1 ; X86-NEXT: vcmpunordss %xmm2, %xmm2, %xmm2 ; X86-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: calll __extendhfsf2 ; X86-NEXT: vmovups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: fldt {{[-0-9]+}}(%e{{[sb]}}p) # 10-byte Folded Reload ; X86-NEXT: fstps {{[0-9]+}}(%esp) ; X86-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm1 ; X86-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero ; X86-NEXT: vmaxss %xmm2, %xmm0, %xmm3 ; X86-NEXT: vandps %xmm3, %xmm1, %xmm1 ; X86-NEXT: vcmpunordss %xmm2, %xmm2, %xmm2 ; X86-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: calll __truncsfhf2 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vmovd %xmm0, (%esp) ; X86-NEXT: calll __truncsfhf2 ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload ; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload ; X86-NEXT: vpunpcklwd {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload ; X86-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3] ; X86-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],zero,zero ; X86-NEXT: addl $164, %esp ; X86-NEXT: retl %r = call <4 x half> @llvm.maximumnum.v4f16(<4 x half> %x, <4 x half> %y) ret <4 x half> %r } define bfloat @test_fmaximumnum_bf16(bfloat %x, bfloat %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_bf16: ; SSE2: # %bb.0: ; SSE2-NEXT: pushq %rax ; SSE2-NEXT: pextrw $0, %xmm1, %eax ; SSE2-NEXT: pextrw $0, %xmm0, %ecx ; SSE2-NEXT: shll $16, %ecx ; SSE2-NEXT: movd %ecx, %xmm1 ; SSE2-NEXT: shll $16, %eax ; SSE2-NEXT: movd %eax, %xmm0 ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: maxss %xmm0, %xmm2 ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: andps %xmm1, %xmm3 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: andps %xmm2, %xmm1 ; SSE2-NEXT: andnps %xmm1, %xmm0 ; SSE2-NEXT: orps %xmm3, %xmm0 ; SSE2-NEXT: callq __truncsfbf2@PLT ; SSE2-NEXT: popq %rax ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_bf16: ; AVX1: # %bb.0: ; AVX1-NEXT: pushq %rax ; AVX1-NEXT: vpextrw $0, %xmm0, %eax ; AVX1-NEXT: vpextrw $0, %xmm1, %ecx ; AVX1-NEXT: shll $16, %ecx ; AVX1-NEXT: vmovd %ecx, %xmm0 ; AVX1-NEXT: shll $16, %eax ; AVX1-NEXT: vmovd %eax, %xmm1 ; AVX1-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3 ; AVX1-NEXT: vandps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: callq __truncsfbf2@PLT ; AVX1-NEXT: popq %rax ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_bf16: ; AVX512F: # %bb.0: ; AVX512F-NEXT: pushq %rax ; AVX512F-NEXT: vpextrw $0, %xmm0, %eax ; AVX512F-NEXT: vpextrw $0, %xmm1, %ecx ; AVX512F-NEXT: shll $16, %ecx ; AVX512F-NEXT: vmovd %ecx, %xmm1 ; AVX512F-NEXT: shll $16, %eax ; AVX512F-NEXT: vmovd %eax, %xmm2 ; AVX512F-NEXT: vmaxss %xmm1, %xmm2, %xmm0 ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm3 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vorps %xmm3, %xmm2, %xmm3 ; AVX512F-NEXT: vandps %xmm0, %xmm3, %xmm0 ; AVX512F-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512F-NEXT: vmovss %xmm2, %xmm0, %xmm0 {%k1} ; AVX512F-NEXT: callq __truncsfbf2@PLT ; AVX512F-NEXT: popq %rax ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_bf16: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: pushq %rax ; AVX512DQ-NEXT: vpextrw $0, %xmm0, %eax ; AVX512DQ-NEXT: vpextrw $0, %xmm1, %ecx ; AVX512DQ-NEXT: shll $16, %ecx ; AVX512DQ-NEXT: vmovd %ecx, %xmm1 ; AVX512DQ-NEXT: shll $16, %eax ; AVX512DQ-NEXT: vmovd %eax, %xmm2 ; AVX512DQ-NEXT: vmaxss %xmm1, %xmm2, %xmm0 ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm3 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vorps %xmm3, %xmm2, %xmm3 ; AVX512DQ-NEXT: vandps %xmm0, %xmm3, %xmm0 ; AVX512DQ-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512DQ-NEXT: vmovss %xmm2, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: callq __truncsfbf2@PLT ; AVX512DQ-NEXT: popq %rax ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_bf16: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vpextrw $0, %xmm0, %eax ; AVX512BF16-NEXT: vpextrw $0, %xmm1, %ecx ; AVX512BF16-NEXT: shll $16, %ecx ; AVX512BF16-NEXT: vmovd %ecx, %xmm0 ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm1 ; AVX512BF16-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm2 = xmm2 & (xmm1 | m32bcst) ; AVX512BF16-NEXT: vcmpunordss %xmm0, %xmm0, %k1 ; AVX512BF16-NEXT: vmovss %xmm1, %xmm2, %xmm2 {%k1} ; AVX512BF16-NEXT: vcvtneps2bf16 %xmm2, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_bf16: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovw %xmm0, %eax ; AVX10_2-NEXT: vmovw %xmm1, %ecx ; AVX10_2-NEXT: shll $16, %ecx ; AVX10_2-NEXT: vmovd %ecx, %xmm0 ; AVX10_2-NEXT: shll $16, %eax ; AVX10_2-NEXT: vmovd %eax, %xmm1 ; AVX10_2-NEXT: vminmaxss $21, %xmm0, %xmm1 ; AVX10_2-NEXT: vcvtneps2bf16 %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_bf16: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: shll $16, %eax ; X86-NEXT: vmovd %eax, %xmm0 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: shll $16, %eax ; X86-NEXT: vmovd %eax, %xmm1 ; X86-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm3 ; X86-NEXT: vandps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: calll __truncsfbf2 ; X86-NEXT: popl %eax ; X86-NEXT: retl %z = call bfloat @llvm.maximumnum.bf16(bfloat %x, bfloat %y) ret bfloat %z } define <4 x bfloat> @test_fmaximumnum_v4bf16(<4 x bfloat> %x, <4 x bfloat> %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_v4bf16: ; SSE2: # %bb.0: ; SSE2-NEXT: pushq %rbp ; SSE2-NEXT: pushq %r15 ; SSE2-NEXT: pushq %r14 ; SSE2-NEXT: pushq %r13 ; SSE2-NEXT: pushq %r12 ; SSE2-NEXT: pushq %rbx ; SSE2-NEXT: subq $56, %rsp ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: psrlq $48, %xmm2 ; SSE2-NEXT: pextrw $0, %xmm2, %ebx ; SSE2-NEXT: movdqa %xmm0, %xmm2 ; SSE2-NEXT: psrlq $48, %xmm2 ; SSE2-NEXT: pextrw $0, %xmm2, %ebp ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[1,1] ; SSE2-NEXT: pextrw $0, %xmm2, %r14d ; SSE2-NEXT: movdqa %xmm0, %xmm2 ; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm0[1,1] ; SSE2-NEXT: pextrw $0, %xmm2, %r15d ; SSE2-NEXT: pextrw $0, %xmm1, %r12d ; SSE2-NEXT: pextrw $0, %xmm0, %r13d ; SSE2-NEXT: psrld $16, %xmm1 ; SSE2-NEXT: pextrw $0, %xmm1, %eax ; SSE2-NEXT: psrld $16, %xmm0 ; SSE2-NEXT: pextrw $0, %xmm0, %ecx ; SSE2-NEXT: shll $16, %ecx ; SSE2-NEXT: movd %ecx, %xmm1 ; SSE2-NEXT: shll $16, %eax ; SSE2-NEXT: movd %eax, %xmm0 ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: maxss %xmm0, %xmm2 ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: andps %xmm1, %xmm3 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: andps %xmm2, %xmm1 ; SSE2-NEXT: andnps %xmm1, %xmm0 ; SSE2-NEXT: orps %xmm3, %xmm0 ; SSE2-NEXT: callq __truncsfbf2@PLT ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: shll $16, %r13d ; SSE2-NEXT: movd %r13d, %xmm1 ; SSE2-NEXT: shll $16, %r12d ; SSE2-NEXT: movd %r12d, %xmm0 ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: maxss %xmm0, %xmm2 ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: andps %xmm1, %xmm3 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: andps %xmm2, %xmm1 ; SSE2-NEXT: andnps %xmm1, %xmm0 ; SSE2-NEXT: orps %xmm3, %xmm0 ; SSE2-NEXT: callq __truncsfbf2@PLT ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: shll $16, %r15d ; SSE2-NEXT: movd %r15d, %xmm1 ; SSE2-NEXT: shll $16, %r14d ; SSE2-NEXT: movd %r14d, %xmm0 ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: maxss %xmm0, %xmm2 ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: andps %xmm1, %xmm3 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: andps %xmm2, %xmm1 ; SSE2-NEXT: andnps %xmm1, %xmm0 ; SSE2-NEXT: orps %xmm3, %xmm0 ; SSE2-NEXT: callq __truncsfbf2@PLT ; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill ; SSE2-NEXT: shll $16, %ebp ; SSE2-NEXT: movd %ebp, %xmm1 ; SSE2-NEXT: shll $16, %ebx ; SSE2-NEXT: movd %ebx, %xmm0 ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: maxss %xmm0, %xmm2 ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: andps %xmm1, %xmm3 ; SSE2-NEXT: orps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: andps %xmm2, %xmm1 ; SSE2-NEXT: andnps %xmm1, %xmm0 ; SSE2-NEXT: orps %xmm3, %xmm0 ; SSE2-NEXT: callq __truncsfbf2@PLT ; SSE2-NEXT: movdqa (%rsp), %xmm1 # 16-byte Reload ; SSE2-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] ; SSE2-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload ; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3] ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] ; SSE2-NEXT: addq $56, %rsp ; SSE2-NEXT: popq %rbx ; SSE2-NEXT: popq %r12 ; SSE2-NEXT: popq %r13 ; SSE2-NEXT: popq %r14 ; SSE2-NEXT: popq %r15 ; SSE2-NEXT: popq %rbp ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fmaximumnum_v4bf16: ; AVX1: # %bb.0: ; AVX1-NEXT: pushq %rbp ; AVX1-NEXT: pushq %r15 ; AVX1-NEXT: pushq %r14 ; AVX1-NEXT: pushq %r13 ; AVX1-NEXT: pushq %r12 ; AVX1-NEXT: pushq %rbx ; AVX1-NEXT: subq $56, %rsp ; AVX1-NEXT: vpsrlq $48, %xmm0, %xmm2 ; AVX1-NEXT: vpextrw $0, %xmm2, %ebx ; AVX1-NEXT: vpsrlq $48, %xmm1, %xmm2 ; AVX1-NEXT: vpextrw $0, %xmm2, %ebp ; AVX1-NEXT: vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3] ; AVX1-NEXT: vpextrw $0, %xmm2, %r14d ; AVX1-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3] ; AVX1-NEXT: vpextrw $0, %xmm2, %r15d ; AVX1-NEXT: vpextrw $0, %xmm0, %r12d ; AVX1-NEXT: vpextrw $0, %xmm1, %r13d ; AVX1-NEXT: vpsrld $16, %xmm0, %xmm0 ; AVX1-NEXT: vpextrw $0, %xmm0, %eax ; AVX1-NEXT: vpsrld $16, %xmm1, %xmm0 ; AVX1-NEXT: vpextrw $0, %xmm0, %ecx ; AVX1-NEXT: shll $16, %ecx ; AVX1-NEXT: vmovd %ecx, %xmm0 ; AVX1-NEXT: shll $16, %eax ; AVX1-NEXT: vmovd %eax, %xmm1 ; AVX1-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3 ; AVX1-NEXT: vandps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: callq __truncsfbf2@PLT ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: shll $16, %r13d ; AVX1-NEXT: vmovd %r13d, %xmm0 ; AVX1-NEXT: shll $16, %r12d ; AVX1-NEXT: vmovd %r12d, %xmm1 ; AVX1-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3 ; AVX1-NEXT: vandps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: callq __truncsfbf2@PLT ; AVX1-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX1-NEXT: shll $16, %r15d ; AVX1-NEXT: vmovd %r15d, %xmm0 ; AVX1-NEXT: shll $16, %r14d ; AVX1-NEXT: vmovd %r14d, %xmm1 ; AVX1-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3 ; AVX1-NEXT: vandps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: callq __truncsfbf2@PLT ; AVX1-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill ; AVX1-NEXT: shll $16, %ebp ; AVX1-NEXT: vmovd %ebp, %xmm0 ; AVX1-NEXT: shll $16, %ebx ; AVX1-NEXT: vmovd %ebx, %xmm1 ; AVX1-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3 ; AVX1-NEXT: vandps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: callq __truncsfbf2@PLT ; AVX1-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload ; AVX1-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] ; AVX1-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX1-NEXT: vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload ; AVX1-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3] ; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],zero,zero ; AVX1-NEXT: addq $56, %rsp ; AVX1-NEXT: popq %rbx ; AVX1-NEXT: popq %r12 ; AVX1-NEXT: popq %r13 ; AVX1-NEXT: popq %r14 ; AVX1-NEXT: popq %r15 ; AVX1-NEXT: popq %rbp ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fmaximumnum_v4bf16: ; AVX512F: # %bb.0: ; AVX512F-NEXT: pushq %rbp ; AVX512F-NEXT: pushq %r15 ; AVX512F-NEXT: pushq %r14 ; AVX512F-NEXT: pushq %r13 ; AVX512F-NEXT: pushq %r12 ; AVX512F-NEXT: pushq %rbx ; AVX512F-NEXT: pushq %rax ; AVX512F-NEXT: vmovq %xmm1, %r13 ; AVX512F-NEXT: movq %r13, %rbx ; AVX512F-NEXT: shrq $32, %rbx ; AVX512F-NEXT: vmovq %xmm0, %rbp ; AVX512F-NEXT: movq %rbp, %r14 ; AVX512F-NEXT: shrq $32, %r14 ; AVX512F-NEXT: movq %r13, %r15 ; AVX512F-NEXT: shrq $48, %r15 ; AVX512F-NEXT: movq %rbp, %r12 ; AVX512F-NEXT: shrq $48, %r12 ; AVX512F-NEXT: movl %ebp, %eax ; AVX512F-NEXT: andl $-65536, %eax # imm = 0xFFFF0000 ; AVX512F-NEXT: vmovd %eax, %xmm1 ; AVX512F-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512F-NEXT: movl %r13d, %eax ; AVX512F-NEXT: andl $-65536, %eax # imm = 0xFFFF0000 ; AVX512F-NEXT: vmovd %eax, %xmm2 ; AVX512F-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512F-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512F-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512F-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512F-NEXT: callq __truncsfbf2@PLT ; AVX512F-NEXT: vpextrw $0, %xmm0, {{[0-9]+}}(%rsp) ; AVX512F-NEXT: shll $16, %ebp ; AVX512F-NEXT: vmovd %ebp, %xmm1 ; AVX512F-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512F-NEXT: shll $16, %r13d ; AVX512F-NEXT: vmovd %r13d, %xmm2 ; AVX512F-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512F-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512F-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512F-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512F-NEXT: callq __truncsfbf2@PLT ; AVX512F-NEXT: vpextrw $0, %xmm0, (%rsp) ; AVX512F-NEXT: shll $16, %r12d ; AVX512F-NEXT: vmovd %r12d, %xmm1 ; AVX512F-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512F-NEXT: shll $16, %r15d ; AVX512F-NEXT: vmovd %r15d, %xmm2 ; AVX512F-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512F-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512F-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512F-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512F-NEXT: callq __truncsfbf2@PLT ; AVX512F-NEXT: vpextrw $0, %xmm0, {{[0-9]+}}(%rsp) ; AVX512F-NEXT: shll $16, %r14d ; AVX512F-NEXT: vmovd %r14d, %xmm1 ; AVX512F-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512F-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512F-NEXT: shll $16, %ebx ; AVX512F-NEXT: vmovd %ebx, %xmm2 ; AVX512F-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512F-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512F-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512F-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512F-NEXT: callq __truncsfbf2@PLT ; AVX512F-NEXT: vpextrw $0, %xmm0, {{[0-9]+}}(%rsp) ; AVX512F-NEXT: vmovaps (%rsp), %xmm0 ; AVX512F-NEXT: addq $8, %rsp ; AVX512F-NEXT: popq %rbx ; AVX512F-NEXT: popq %r12 ; AVX512F-NEXT: popq %r13 ; AVX512F-NEXT: popq %r14 ; AVX512F-NEXT: popq %r15 ; AVX512F-NEXT: popq %rbp ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fmaximumnum_v4bf16: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: pushq %rbp ; AVX512DQ-NEXT: pushq %r15 ; AVX512DQ-NEXT: pushq %r14 ; AVX512DQ-NEXT: pushq %r13 ; AVX512DQ-NEXT: pushq %r12 ; AVX512DQ-NEXT: pushq %rbx ; AVX512DQ-NEXT: pushq %rax ; AVX512DQ-NEXT: vmovq %xmm1, %r13 ; AVX512DQ-NEXT: movq %r13, %rbx ; AVX512DQ-NEXT: shrq $32, %rbx ; AVX512DQ-NEXT: vmovq %xmm0, %rbp ; AVX512DQ-NEXT: movq %rbp, %r14 ; AVX512DQ-NEXT: shrq $32, %r14 ; AVX512DQ-NEXT: movq %r13, %r15 ; AVX512DQ-NEXT: shrq $48, %r15 ; AVX512DQ-NEXT: movq %rbp, %r12 ; AVX512DQ-NEXT: shrq $48, %r12 ; AVX512DQ-NEXT: movl %ebp, %eax ; AVX512DQ-NEXT: andl $-65536, %eax # imm = 0xFFFF0000 ; AVX512DQ-NEXT: vmovd %eax, %xmm1 ; AVX512DQ-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512DQ-NEXT: movl %r13d, %eax ; AVX512DQ-NEXT: andl $-65536, %eax # imm = 0xFFFF0000 ; AVX512DQ-NEXT: vmovd %eax, %xmm2 ; AVX512DQ-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512DQ-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512DQ-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512DQ-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: callq __truncsfbf2@PLT ; AVX512DQ-NEXT: vpextrw $0, %xmm0, {{[0-9]+}}(%rsp) ; AVX512DQ-NEXT: shll $16, %ebp ; AVX512DQ-NEXT: vmovd %ebp, %xmm1 ; AVX512DQ-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512DQ-NEXT: shll $16, %r13d ; AVX512DQ-NEXT: vmovd %r13d, %xmm2 ; AVX512DQ-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512DQ-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512DQ-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512DQ-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: callq __truncsfbf2@PLT ; AVX512DQ-NEXT: vpextrw $0, %xmm0, (%rsp) ; AVX512DQ-NEXT: shll $16, %r12d ; AVX512DQ-NEXT: vmovd %r12d, %xmm1 ; AVX512DQ-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512DQ-NEXT: shll $16, %r15d ; AVX512DQ-NEXT: vmovd %r15d, %xmm2 ; AVX512DQ-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512DQ-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512DQ-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512DQ-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: callq __truncsfbf2@PLT ; AVX512DQ-NEXT: vpextrw $0, %xmm0, {{[0-9]+}}(%rsp) ; AVX512DQ-NEXT: shll $16, %r14d ; AVX512DQ-NEXT: vmovd %r14d, %xmm1 ; AVX512DQ-NEXT: vpbroadcastd {{.*#+}} xmm0 = [NaN,NaN,NaN,NaN] ; AVX512DQ-NEXT: vpor %xmm0, %xmm1, %xmm0 ; AVX512DQ-NEXT: shll $16, %ebx ; AVX512DQ-NEXT: vmovd %ebx, %xmm2 ; AVX512DQ-NEXT: vmaxss %xmm2, %xmm1, %xmm3 ; AVX512DQ-NEXT: vpand %xmm3, %xmm0, %xmm0 ; AVX512DQ-NEXT: vcmpunordss %xmm2, %xmm2, %k1 ; AVX512DQ-NEXT: vmovss %xmm1, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: callq __truncsfbf2@PLT ; AVX512DQ-NEXT: vpextrw $0, %xmm0, {{[0-9]+}}(%rsp) ; AVX512DQ-NEXT: vmovaps (%rsp), %xmm0 ; AVX512DQ-NEXT: addq $8, %rsp ; AVX512DQ-NEXT: popq %rbx ; AVX512DQ-NEXT: popq %r12 ; AVX512DQ-NEXT: popq %r13 ; AVX512DQ-NEXT: popq %r14 ; AVX512DQ-NEXT: popq %r15 ; AVX512DQ-NEXT: popq %rbp ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fmaximumnum_v4bf16: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: pushq %rbp ; AVX512BF16-NEXT: pushq %r15 ; AVX512BF16-NEXT: pushq %r14 ; AVX512BF16-NEXT: pushq %r13 ; AVX512BF16-NEXT: pushq %r12 ; AVX512BF16-NEXT: pushq %rbx ; AVX512BF16-NEXT: vpextrw $7, %xmm1, %eax ; AVX512BF16-NEXT: vpextrw $7, %xmm0, %ebx ; AVX512BF16-NEXT: movl %ebx, %ecx ; AVX512BF16-NEXT: shll $16, %ecx ; AVX512BF16-NEXT: vmovd %ecx, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %eax, %ebx ; AVX512BF16-NEXT: movl %eax, %ecx ; AVX512BF16-NEXT: vpextrw $6, %xmm1, %eax ; AVX512BF16-NEXT: movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill ; AVX512BF16-NEXT: vpextrw $6, %xmm0, %r13d ; AVX512BF16-NEXT: movl %r13d, %edx ; AVX512BF16-NEXT: shll $16, %edx ; AVX512BF16-NEXT: vmovd %edx, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %eax, %r13d ; AVX512BF16-NEXT: vpextrw $5, %xmm1, %eax ; AVX512BF16-NEXT: movl %eax, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill ; AVX512BF16-NEXT: vpextrw $5, %xmm0, %r12d ; AVX512BF16-NEXT: movl %r12d, %esi ; AVX512BF16-NEXT: shll $16, %esi ; AVX512BF16-NEXT: vmovd %esi, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %eax, %r12d ; AVX512BF16-NEXT: vpextrw $4, %xmm1, %esi ; AVX512BF16-NEXT: vpextrw $4, %xmm0, %r15d ; AVX512BF16-NEXT: movl %r15d, %edi ; AVX512BF16-NEXT: shll $16, %edi ; AVX512BF16-NEXT: vmovd %edi, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %esi, %r15d ; AVX512BF16-NEXT: vpextrw $3, %xmm1, %edi ; AVX512BF16-NEXT: vpextrw $3, %xmm0, %r14d ; AVX512BF16-NEXT: movl %r14d, %r8d ; AVX512BF16-NEXT: shll $16, %r8d ; AVX512BF16-NEXT: vmovd %r8d, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %edi, %r14d ; AVX512BF16-NEXT: vpextrw $2, %xmm1, %r8d ; AVX512BF16-NEXT: vpextrw $2, %xmm0, %ebp ; AVX512BF16-NEXT: movl %ebp, %r9d ; AVX512BF16-NEXT: shll $16, %r9d ; AVX512BF16-NEXT: vmovd %r9d, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %r8d, %ebp ; AVX512BF16-NEXT: vpextrw $1, %xmm1, %r9d ; AVX512BF16-NEXT: vpextrw $1, %xmm0, %edx ; AVX512BF16-NEXT: movl %edx, %r10d ; AVX512BF16-NEXT: shll $16, %r10d ; AVX512BF16-NEXT: vmovd %r10d, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %r9d, %edx ; AVX512BF16-NEXT: vmovd %xmm0, %r10d ; AVX512BF16-NEXT: movl %r10d, %r11d ; AVX512BF16-NEXT: shll $16, %r11d ; AVX512BF16-NEXT: vmovd %r11d, %xmm0 ; AVX512BF16-NEXT: vucomiss %xmm0, %xmm0 ; AVX512BF16-NEXT: vmovd %xmm1, %r11d ; AVX512BF16-NEXT: cmovpl %r11d, %r10d ; AVX512BF16-NEXT: movl %ebx, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl %ecx, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm1 ; AVX512BF16-NEXT: vucomiss %xmm1, %xmm1 ; AVX512BF16-NEXT: cmovpl %ebx, %ecx ; AVX512BF16-NEXT: movl %ecx, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vmovd %r10d, %xmm1 ; AVX512BF16-NEXT: vpinsrw $1, %edx, %xmm1, %xmm1 ; AVX512BF16-NEXT: vpinsrw $2, %ebp, %xmm1, %xmm1 ; AVX512BF16-NEXT: vpinsrw $3, %r14d, %xmm1, %xmm1 ; AVX512BF16-NEXT: vpinsrw $4, %r15d, %xmm1, %xmm1 ; AVX512BF16-NEXT: vpinsrw $5, %r12d, %xmm1, %xmm1 ; AVX512BF16-NEXT: vpinsrw $6, %r13d, %xmm1, %xmm1 ; AVX512BF16-NEXT: vpinsrw $7, %ebx, %xmm1, %xmm1 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %ebx, %ecx ; AVX512BF16-NEXT: movl %r13d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl {{[-0-9]+}}(%r{{[sb]}}p), %ebx # 4-byte Reload ; AVX512BF16-NEXT: movl %ebx, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %r13d, %ebx ; AVX512BF16-NEXT: movl %ebx, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %r13d, %ebx ; AVX512BF16-NEXT: movl %r12d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl {{[-0-9]+}}(%r{{[sb]}}p), %r13d # 4-byte Reload ; AVX512BF16-NEXT: movl %r13d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %r12d, %r13d ; AVX512BF16-NEXT: movl %r13d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %r12d, %r13d ; AVX512BF16-NEXT: movl %r15d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl %esi, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %r15d, %esi ; AVX512BF16-NEXT: movl %esi, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %r15d, %esi ; AVX512BF16-NEXT: movl %r14d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl %edi, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %r14d, %edi ; AVX512BF16-NEXT: movl %edi, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %r14d, %edi ; AVX512BF16-NEXT: movl %ebp, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl %r8d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %ebp, %r8d ; AVX512BF16-NEXT: movl %r8d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %ebp, %r8d ; AVX512BF16-NEXT: movl %edx, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl %r9d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %edx, %r9d ; AVX512BF16-NEXT: movl %r9d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %edx, %r9d ; AVX512BF16-NEXT: movl %r10d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm0 ; AVX512BF16-NEXT: movl %r11d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm2 ; AVX512BF16-NEXT: cmovpl %r10d, %r11d ; AVX512BF16-NEXT: movl %r11d, %eax ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm2 ; AVX512BF16-NEXT: vucomiss %xmm2, %xmm0 ; AVX512BF16-NEXT: cmoval %r10d, %r11d ; AVX512BF16-NEXT: vmovd %r11d, %xmm0 ; AVX512BF16-NEXT: vpinsrw $1, %r9d, %xmm0, %xmm0 ; AVX512BF16-NEXT: vpinsrw $2, %r8d, %xmm0, %xmm0 ; AVX512BF16-NEXT: vpinsrw $3, %edi, %xmm0, %xmm0 ; AVX512BF16-NEXT: vpinsrw $4, %esi, %xmm0, %xmm0 ; AVX512BF16-NEXT: vpinsrw $5, %r13d, %xmm0, %xmm0 ; AVX512BF16-NEXT: vpinsrw $6, %ebx, %xmm0, %xmm0 ; AVX512BF16-NEXT: vpinsrw $7, %ecx, %xmm0, %xmm0 ; AVX512BF16-NEXT: vptestnmw %xmm1, %xmm1, %k1 ; AVX512BF16-NEXT: vpblendmw %xmm1, %xmm0, %xmm1 {%k1} ; AVX512BF16-NEXT: vpmovzxwd {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero ; AVX512BF16-NEXT: vpslld $16, %ymm2, %ymm2 ; AVX512BF16-NEXT: vxorps %xmm3, %xmm3, %xmm3 ; AVX512BF16-NEXT: vcmpeqps %ymm3, %ymm2, %k1 ; AVX512BF16-NEXT: vmovdqu16 %xmm1, %xmm0 {%k1} ; AVX512BF16-NEXT: popq %rbx ; AVX512BF16-NEXT: popq %r12 ; AVX512BF16-NEXT: popq %r13 ; AVX512BF16-NEXT: popq %r14 ; AVX512BF16-NEXT: popq %r15 ; AVX512BF16-NEXT: popq %rbp ; AVX512BF16-NEXT: vzeroupper ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_v4bf16: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vminmaxbf16 $21, %xmm1, %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_v4bf16: ; X86: # %bb.0: ; X86-NEXT: pushl %ebp ; X86-NEXT: pushl %ebx ; X86-NEXT: pushl %edi ; X86-NEXT: pushl %esi ; X86-NEXT: subl $68, %esp ; X86-NEXT: vpsrlq $48, %xmm0, %xmm2 ; X86-NEXT: vpextrw $0, %xmm2, %esi ; X86-NEXT: vpsrlq $48, %xmm1, %xmm2 ; X86-NEXT: vpextrw $0, %xmm2, %edi ; X86-NEXT: vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3] ; X86-NEXT: vpextrw $0, %xmm2, %ebx ; X86-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3] ; X86-NEXT: vpextrw $0, %xmm2, %ebp ; X86-NEXT: vpextrw $0, %xmm0, %eax ; X86-NEXT: vpsrld $16, %xmm0, %xmm0 ; X86-NEXT: vpsrld $16, %xmm1, %xmm2 ; X86-NEXT: vpextrw $0, %xmm2, %ecx ; X86-NEXT: shll $16, %ecx ; X86-NEXT: vmovd %ecx, %xmm2 ; X86-NEXT: vpextrw $0, %xmm0, %ecx ; X86-NEXT: shll $16, %ecx ; X86-NEXT: vmovd %ecx, %xmm0 ; X86-NEXT: vpextrw $0, %xmm1, %ecx ; X86-NEXT: vmaxss %xmm2, %xmm0, %xmm1 ; X86-NEXT: vbroadcastss {{.*#+}} xmm4 = [NaN,NaN,NaN,NaN] ; X86-NEXT: vorps %xmm4, %xmm0, %xmm3 ; X86-NEXT: vandps %xmm1, %xmm3, %xmm1 ; X86-NEXT: vcmpunordss %xmm2, %xmm2, %xmm2 ; X86-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: shll $16, %ecx ; X86-NEXT: vmovd %ecx, %xmm0 ; X86-NEXT: shll $16, %eax ; X86-NEXT: vmovd %eax, %xmm1 ; X86-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vorps %xmm4, %xmm1, %xmm3 ; X86-NEXT: vandps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: calll __truncsfbf2 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: shll $16, %ebp ; X86-NEXT: vmovd %ebp, %xmm0 ; X86-NEXT: shll $16, %ebx ; X86-NEXT: vmovd %ebx, %xmm1 ; X86-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm3 ; X86-NEXT: vandps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: calll __truncsfbf2 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: shll $16, %edi ; X86-NEXT: vmovd %edi, %xmm0 ; X86-NEXT: shll $16, %esi ; X86-NEXT: vmovd %esi, %xmm1 ; X86-NEXT: vmaxss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm3 ; X86-NEXT: vandps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: calll __truncsfbf2 ; X86-NEXT: vmovups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload ; X86-NEXT: vmovd %xmm0, (%esp) ; X86-NEXT: calll __truncsfbf2 ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload ; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] ; X86-NEXT: vmovdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload ; X86-NEXT: vpunpcklwd {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload ; X86-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3] ; X86-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],zero,zero ; X86-NEXT: addl $68, %esp ; X86-NEXT: popl %esi ; X86-NEXT: popl %edi ; X86-NEXT: popl %ebx ; X86-NEXT: popl %ebp ; X86-NEXT: retl %r = call <4 x bfloat> @llvm.maximumnum.v4bf16(<4 x bfloat> %x, <4 x bfloat> %y) ret <4 x bfloat> %r } define bfloat @test_fminimumnum_bf16(bfloat %x, bfloat %y) nounwind { ; SSE2-LABEL: test_fminimumnum_bf16: ; SSE2: # %bb.0: ; SSE2-NEXT: pushq %rax ; SSE2-NEXT: pextrw $0, %xmm1, %eax ; SSE2-NEXT: pextrw $0, %xmm0, %ecx ; SSE2-NEXT: shll $16, %ecx ; SSE2-NEXT: movd %ecx, %xmm1 ; SSE2-NEXT: shll $16, %eax ; SSE2-NEXT: movd %eax, %xmm0 ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: minss %xmm0, %xmm2 ; SSE2-NEXT: cmpunordss %xmm0, %xmm0 ; SSE2-NEXT: movaps %xmm0, %xmm3 ; SSE2-NEXT: andps %xmm1, %xmm3 ; SSE2-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 ; SSE2-NEXT: orps %xmm2, %xmm1 ; SSE2-NEXT: andnps %xmm1, %xmm0 ; SSE2-NEXT: orps %xmm3, %xmm0 ; SSE2-NEXT: callq __truncsfbf2@PLT ; SSE2-NEXT: popq %rax ; SSE2-NEXT: retq ; ; AVX1-LABEL: test_fminimumnum_bf16: ; AVX1: # %bb.0: ; AVX1-NEXT: pushq %rax ; AVX1-NEXT: vpextrw $0, %xmm0, %eax ; AVX1-NEXT: vpextrw $0, %xmm1, %ecx ; AVX1-NEXT: shll $16, %ecx ; AVX1-NEXT: vmovd %ecx, %xmm0 ; AVX1-NEXT: shll $16, %eax ; AVX1-NEXT: vmovd %eax, %xmm1 ; AVX1-NEXT: vminss %xmm0, %xmm1, %xmm2 ; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3 ; AVX1-NEXT: vorps %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; AVX1-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; AVX1-NEXT: callq __truncsfbf2@PLT ; AVX1-NEXT: popq %rax ; AVX1-NEXT: retq ; ; AVX512F-LABEL: test_fminimumnum_bf16: ; AVX512F: # %bb.0: ; AVX512F-NEXT: pushq %rax ; AVX512F-NEXT: vpextrw $0, %xmm0, %eax ; AVX512F-NEXT: vpextrw $0, %xmm1, %ecx ; AVX512F-NEXT: shll $16, %ecx ; AVX512F-NEXT: vmovd %ecx, %xmm1 ; AVX512F-NEXT: shll $16, %eax ; AVX512F-NEXT: vmovd %eax, %xmm2 ; AVX512F-NEXT: vminss %xmm1, %xmm2, %xmm0 ; AVX512F-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512F-NEXT: vandps %xmm3, %xmm2, %xmm3 ; AVX512F-NEXT: vorps %xmm0, %xmm3, %xmm0 ; AVX512F-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512F-NEXT: vmovss %xmm2, %xmm0, %xmm0 {%k1} ; AVX512F-NEXT: callq __truncsfbf2@PLT ; AVX512F-NEXT: popq %rax ; AVX512F-NEXT: retq ; ; AVX512DQ-LABEL: test_fminimumnum_bf16: ; AVX512DQ: # %bb.0: ; AVX512DQ-NEXT: pushq %rax ; AVX512DQ-NEXT: vpextrw $0, %xmm0, %eax ; AVX512DQ-NEXT: vpextrw $0, %xmm1, %ecx ; AVX512DQ-NEXT: shll $16, %ecx ; AVX512DQ-NEXT: vmovd %ecx, %xmm1 ; AVX512DQ-NEXT: shll $16, %eax ; AVX512DQ-NEXT: vmovd %eax, %xmm2 ; AVX512DQ-NEXT: vminss %xmm1, %xmm2, %xmm0 ; AVX512DQ-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0] ; AVX512DQ-NEXT: vandps %xmm3, %xmm2, %xmm3 ; AVX512DQ-NEXT: vorps %xmm0, %xmm3, %xmm0 ; AVX512DQ-NEXT: vcmpunordss %xmm1, %xmm1, %k1 ; AVX512DQ-NEXT: vmovss %xmm2, %xmm0, %xmm0 {%k1} ; AVX512DQ-NEXT: callq __truncsfbf2@PLT ; AVX512DQ-NEXT: popq %rax ; AVX512DQ-NEXT: retq ; ; AVX512BF16-LABEL: test_fminimumnum_bf16: ; AVX512BF16: # %bb.0: ; AVX512BF16-NEXT: vpextrw $0, %xmm0, %eax ; AVX512BF16-NEXT: vpextrw $0, %xmm1, %ecx ; AVX512BF16-NEXT: shll $16, %ecx ; AVX512BF16-NEXT: vmovd %ecx, %xmm0 ; AVX512BF16-NEXT: shll $16, %eax ; AVX512BF16-NEXT: vmovd %eax, %xmm1 ; AVX512BF16-NEXT: vminss %xmm0, %xmm1, %xmm2 ; AVX512BF16-NEXT: vpternlogd {{.*#+}} xmm2 = xmm2 | (xmm1 & m32bcst) ; AVX512BF16-NEXT: vcmpunordss %xmm0, %xmm0, %k1 ; AVX512BF16-NEXT: vmovss %xmm1, %xmm2, %xmm2 {%k1} ; AVX512BF16-NEXT: vcvtneps2bf16 %xmm2, %xmm0 ; AVX512BF16-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_bf16: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovw %xmm0, %eax ; AVX10_2-NEXT: vmovw %xmm1, %ecx ; AVX10_2-NEXT: shll $16, %ecx ; AVX10_2-NEXT: vmovd %ecx, %xmm0 ; AVX10_2-NEXT: shll $16, %eax ; AVX10_2-NEXT: vmovd %eax, %xmm1 ; AVX10_2-NEXT: vminmaxss $20, %xmm0, %xmm1 ; AVX10_2-NEXT: vcvtneps2bf16 %xmm0, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_bf16: ; X86: # %bb.0: ; X86-NEXT: pushl %eax ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: shll $16, %eax ; X86-NEXT: vmovd %eax, %xmm0 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: shll $16, %eax ; X86-NEXT: vmovd %eax, %xmm1 ; X86-NEXT: vminss %xmm0, %xmm1, %xmm2 ; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm3 ; X86-NEXT: vorps %xmm2, %xmm3, %xmm2 ; X86-NEXT: vcmpunordss %xmm0, %xmm0, %xmm0 ; X86-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0 ; X86-NEXT: vmovss %xmm0, (%esp) ; X86-NEXT: calll __truncsfbf2 ; X86-NEXT: popl %eax ; X86-NEXT: retl %z = call bfloat @llvm.minimumnum.bf16(bfloat %x, bfloat %y) ret bfloat %z } ; nnan minimumnum(Y, +inf) -> Y define float @test_fminimumnum_inf_nnan(float %x, float %y) nounwind { ; SSE2-LABEL: test_fminimumnum_inf_nnan: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_inf_nnan: ; AVX: # %bb.0: ; AVX-NEXT: vmovaps %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_inf_nnan: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovaps %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_inf_nnan: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = call nnan float @llvm.minimumnum.f32(float %y, float 0x7ff0000000000000) ret float %1 } ; nnan maximumnum(Y, -inf) -> Y define float @test_fmaximumnum_neg_inf_nnan(float %x, float %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_neg_inf_nnan: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_neg_inf_nnan: ; AVX: # %bb.0: ; AVX-NEXT: vmovaps %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_neg_inf_nnan: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovaps %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_neg_inf_nnan: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = call nnan float @llvm.maximumnum.f32(float %y, float 0xfff0000000000000) ret float %1 } ; Test we propagate the non-NaN arg, even if one arg is SNaN define float @test_fmaximumnum_snan(float %x, float %y) { ; SSE2-LABEL: test_fmaximumnum_snan: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_snan: ; AVX: # %bb.0: ; AVX-NEXT: vmovaps %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_snan: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovaps %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_snan: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = tail call float @llvm.maximumnum.f32(float 0x7ff4000000000000, float %y) ret float %1 } define float @test_fminimumnum_snan(float %x, float %y) { ; SSE2-LABEL: test_fminimumnum_snan: ; SSE2: # %bb.0: ; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_snan: ; AVX: # %bb.0: ; AVX-NEXT: vmovaps %xmm1, %xmm0 ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_snan: ; AVX10_2: # %bb.0: ; AVX10_2-NEXT: vmovaps %xmm1, %xmm0 ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_snan: ; X86: # %bb.0: ; X86-NEXT: flds {{[0-9]+}}(%esp) ; X86-NEXT: retl %1 = tail call float @llvm.minimumnum.f32(float 0x7ff4000000000000, float %y) ret float %1 } define fp128 @test_fmaximumnum_fp128(fp128 %x, fp128 %y) nounwind { ; SSE2-LABEL: test_fmaximumnum_fp128: ; SSE2: # %bb.0: # %start ; SSE2-NEXT: subq $40, %rsp ; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: callq __unordtf2@PLT ; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: jne .LBB41_2 ; SSE2-NEXT: # %bb.1: # %start ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; SSE2-NEXT: .LBB41_2: # %start ; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: callq __unordtf2@PLT ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; SSE2-NEXT: jne .LBB41_4 ; SSE2-NEXT: # %bb.3: # %start ; SSE2-NEXT: movaps (%rsp), %xmm1 # 16-byte Reload ; SSE2-NEXT: .LBB41_4: # %start ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill ; SSE2-NEXT: callq __gttf2@PLT ; SSE2-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movdqa %xmm0, %xmm1 ; SSE2-NEXT: jg .LBB41_6 ; SSE2-NEXT: # %bb.5: # %start ; SSE2-NEXT: movdqa (%rsp), %xmm1 # 16-byte Reload ; SSE2-NEXT: .LBB41_6: # %start ; SSE2-NEXT: movdqa %xmm1, (%rsp) # 16-byte Spill ; SSE2-NEXT: callq __trunctfsf2@PLT ; SSE2-NEXT: movaps (%rsp), %xmm2 # 16-byte Reload ; SSE2-NEXT: movd %xmm0, %eax ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: je .LBB41_8 ; SSE2-NEXT: # %bb.7: # %start ; SSE2-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: .LBB41_8: # %start ; SSE2-NEXT: pxor %xmm1, %xmm1 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: callq __eqtf2@PLT ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: je .LBB41_10 ; SSE2-NEXT: # %bb.9: # %start ; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload ; SSE2-NEXT: .LBB41_10: # %start ; SSE2-NEXT: addq $40, %rsp ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fmaximumnum_fp128: ; AVX: # %bb.0: # %start ; AVX-NEXT: subq $40, %rsp ; AVX-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX-NEXT: vmovaps %xmm0, %xmm1 ; AVX-NEXT: callq __unordtf2@PLT ; AVX-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovaps %xmm0, %xmm1 ; AVX-NEXT: jne .LBB41_2 ; AVX-NEXT: # %bb.1: # %start ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX-NEXT: .LBB41_2: # %start ; AVX-NEXT: vmovaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX-NEXT: vmovaps %xmm0, %xmm1 ; AVX-NEXT: callq __unordtf2@PLT ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX-NEXT: jne .LBB41_4 ; AVX-NEXT: # %bb.3: # %start ; AVX-NEXT: vmovaps (%rsp), %xmm1 # 16-byte Reload ; AVX-NEXT: .LBB41_4: # %start ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX-NEXT: callq __gttf2@PLT ; AVX-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovdqa %xmm0, %xmm1 ; AVX-NEXT: jg .LBB41_6 ; AVX-NEXT: # %bb.5: # %start ; AVX-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload ; AVX-NEXT: .LBB41_6: # %start ; AVX-NEXT: vmovdqa %xmm1, (%rsp) # 16-byte Spill ; AVX-NEXT: callq __trunctfsf2@PLT ; AVX-NEXT: vmovaps (%rsp), %xmm2 # 16-byte Reload ; AVX-NEXT: vmovd %xmm0, %eax ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: je .LBB41_8 ; AVX-NEXT: # %bb.7: # %start ; AVX-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX-NEXT: .LBB41_8: # %start ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX-NEXT: vmovaps %xmm2, %xmm0 ; AVX-NEXT: callq __eqtf2@PLT ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX-NEXT: je .LBB41_10 ; AVX-NEXT: # %bb.9: # %start ; AVX-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX-NEXT: .LBB41_10: # %start ; AVX-NEXT: addq $40, %rsp ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fmaximumnum_fp128: ; AVX10_2: # %bb.0: # %start ; AVX10_2-NEXT: subq $40, %rsp ; AVX10_2-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX10_2-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX10_2-NEXT: vmovaps %xmm0, %xmm1 ; AVX10_2-NEXT: callq __unordtf2@PLT ; AVX10_2-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovaps %xmm0, %xmm1 ; AVX10_2-NEXT: jne .LBB41_2 ; AVX10_2-NEXT: # %bb.1: # %start ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: .LBB41_2: # %start ; AVX10_2-NEXT: vmovaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX10_2-NEXT: vmovaps %xmm0, %xmm1 ; AVX10_2-NEXT: callq __unordtf2@PLT ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: jne .LBB41_4 ; AVX10_2-NEXT: # %bb.3: # %start ; AVX10_2-NEXT: vmovaps (%rsp), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: .LBB41_4: # %start ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX10_2-NEXT: callq __gttf2@PLT ; AVX10_2-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovdqa %xmm0, %xmm1 ; AVX10_2-NEXT: jg .LBB41_6 ; AVX10_2-NEXT: # %bb.5: # %start ; AVX10_2-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: .LBB41_6: # %start ; AVX10_2-NEXT: vmovdqa %xmm1, (%rsp) # 16-byte Spill ; AVX10_2-NEXT: callq __trunctfsf2@PLT ; AVX10_2-NEXT: vmovaps (%rsp), %xmm2 # 16-byte Reload ; AVX10_2-NEXT: vmovd %xmm0, %eax ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: je .LBB41_8 ; AVX10_2-NEXT: # %bb.7: # %start ; AVX10_2-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX10_2-NEXT: .LBB41_8: # %start ; AVX10_2-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vmovaps %xmm2, %xmm0 ; AVX10_2-NEXT: callq __eqtf2@PLT ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: je .LBB41_10 ; AVX10_2-NEXT: # %bb.9: # %start ; AVX10_2-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: .LBB41_10: # %start ; AVX10_2-NEXT: addq $40, %rsp ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fmaximumnum_fp128: ; X86: # %bb.0: # %start ; X86-NEXT: pushl %ebp ; X86-NEXT: movl %esp, %ebp ; X86-NEXT: pushl %esi ; X86-NEXT: andl $-16, %esp ; X86-NEXT: subl $80, %esp ; X86-NEXT: movl 8(%ebp), %esi ; X86-NEXT: vmovups 24(%ebp), %ymm0 ; X86-NEXT: vmovups %ymm0, {{[0-9]+}}(%esp) ; X86-NEXT: leal {{[0-9]+}}(%esp), %eax ; X86-NEXT: movl %eax, (%esp) ; X86-NEXT: vzeroupper ; X86-NEXT: calll fmaximum_numl ; X86-NEXT: subl $4, %esp ; X86-NEXT: vmovaps {{[0-9]+}}(%esp), %xmm0 ; X86-NEXT: vmovaps %xmm0, (%esi) ; X86-NEXT: movl %esi, %eax ; X86-NEXT: leal -4(%ebp), %esp ; X86-NEXT: popl %esi ; X86-NEXT: popl %ebp ; X86-NEXT: retl $4 start: %0 = tail call fp128 @llvm.maximumnum.f128(fp128 %x, fp128 %y) ret fp128 %0 } define fp128 @test_fminimumnum_fp128(fp128 %x, fp128 %y) nounwind { ; SSE2-LABEL: test_fminimumnum_fp128: ; SSE2: # %bb.0: # %start ; SSE2-NEXT: subq $40, %rsp ; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: callq __unordtf2@PLT ; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: jne .LBB42_2 ; SSE2-NEXT: # %bb.1: # %start ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; SSE2-NEXT: .LBB42_2: # %start ; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: movaps %xmm0, %xmm1 ; SSE2-NEXT: callq __unordtf2@PLT ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; SSE2-NEXT: jne .LBB42_4 ; SSE2-NEXT: # %bb.3: # %start ; SSE2-NEXT: movaps (%rsp), %xmm1 # 16-byte Reload ; SSE2-NEXT: .LBB42_4: # %start ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill ; SSE2-NEXT: callq __lttf2@PLT ; SSE2-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movdqa %xmm0, %xmm1 ; SSE2-NEXT: js .LBB42_6 ; SSE2-NEXT: # %bb.5: # %start ; SSE2-NEXT: movdqa (%rsp), %xmm1 # 16-byte Reload ; SSE2-NEXT: .LBB42_6: # %start ; SSE2-NEXT: movdqa %xmm1, (%rsp) # 16-byte Spill ; SSE2-NEXT: callq __trunctfsf2@PLT ; SSE2-NEXT: movaps (%rsp), %xmm2 # 16-byte Reload ; SSE2-NEXT: movd %xmm0, %eax ; SSE2-NEXT: negl %eax ; SSE2-NEXT: jo .LBB42_8 ; SSE2-NEXT: # %bb.7: # %start ; SSE2-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; SSE2-NEXT: .LBB42_8: # %start ; SSE2-NEXT: pxor %xmm1, %xmm1 ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: callq __eqtf2@PLT ; SSE2-NEXT: testl %eax, %eax ; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; SSE2-NEXT: je .LBB42_10 ; SSE2-NEXT: # %bb.9: # %start ; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload ; SSE2-NEXT: .LBB42_10: # %start ; SSE2-NEXT: addq $40, %rsp ; SSE2-NEXT: retq ; ; AVX-LABEL: test_fminimumnum_fp128: ; AVX: # %bb.0: # %start ; AVX-NEXT: subq $40, %rsp ; AVX-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX-NEXT: vmovaps %xmm0, %xmm1 ; AVX-NEXT: callq __unordtf2@PLT ; AVX-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovaps %xmm0, %xmm1 ; AVX-NEXT: jne .LBB42_2 ; AVX-NEXT: # %bb.1: # %start ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX-NEXT: .LBB42_2: # %start ; AVX-NEXT: vmovaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX-NEXT: vmovaps %xmm0, %xmm1 ; AVX-NEXT: callq __unordtf2@PLT ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX-NEXT: jne .LBB42_4 ; AVX-NEXT: # %bb.3: # %start ; AVX-NEXT: vmovaps (%rsp), %xmm1 # 16-byte Reload ; AVX-NEXT: .LBB42_4: # %start ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX-NEXT: callq __lttf2@PLT ; AVX-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovdqa %xmm0, %xmm1 ; AVX-NEXT: js .LBB42_6 ; AVX-NEXT: # %bb.5: # %start ; AVX-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload ; AVX-NEXT: .LBB42_6: # %start ; AVX-NEXT: vmovdqa %xmm1, (%rsp) # 16-byte Spill ; AVX-NEXT: callq __trunctfsf2@PLT ; AVX-NEXT: vmovaps (%rsp), %xmm2 # 16-byte Reload ; AVX-NEXT: vmovd %xmm0, %eax ; AVX-NEXT: negl %eax ; AVX-NEXT: jo .LBB42_8 ; AVX-NEXT: # %bb.7: # %start ; AVX-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX-NEXT: .LBB42_8: # %start ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX-NEXT: vmovaps %xmm2, %xmm0 ; AVX-NEXT: callq __eqtf2@PLT ; AVX-NEXT: testl %eax, %eax ; AVX-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX-NEXT: je .LBB42_10 ; AVX-NEXT: # %bb.9: # %start ; AVX-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX-NEXT: .LBB42_10: # %start ; AVX-NEXT: addq $40, %rsp ; AVX-NEXT: retq ; ; AVX10_2-LABEL: test_fminimumnum_fp128: ; AVX10_2: # %bb.0: # %start ; AVX10_2-NEXT: subq $40, %rsp ; AVX10_2-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX10_2-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX10_2-NEXT: vmovaps %xmm0, %xmm1 ; AVX10_2-NEXT: callq __unordtf2@PLT ; AVX10_2-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovaps %xmm0, %xmm1 ; AVX10_2-NEXT: jne .LBB42_2 ; AVX10_2-NEXT: # %bb.1: # %start ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: .LBB42_2: # %start ; AVX10_2-NEXT: vmovaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX10_2-NEXT: vmovaps %xmm0, %xmm1 ; AVX10_2-NEXT: callq __unordtf2@PLT ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: jne .LBB42_4 ; AVX10_2-NEXT: # %bb.3: # %start ; AVX10_2-NEXT: vmovaps (%rsp), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: .LBB42_4: # %start ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: vmovaps %xmm1, (%rsp) # 16-byte Spill ; AVX10_2-NEXT: callq __lttf2@PLT ; AVX10_2-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovdqa %xmm0, %xmm1 ; AVX10_2-NEXT: js .LBB42_6 ; AVX10_2-NEXT: # %bb.5: # %start ; AVX10_2-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload ; AVX10_2-NEXT: .LBB42_6: # %start ; AVX10_2-NEXT: vmovdqa %xmm1, (%rsp) # 16-byte Spill ; AVX10_2-NEXT: callq __trunctfsf2@PLT ; AVX10_2-NEXT: vmovaps (%rsp), %xmm2 # 16-byte Reload ; AVX10_2-NEXT: vmovd %xmm0, %eax ; AVX10_2-NEXT: negl %eax ; AVX10_2-NEXT: jo .LBB42_8 ; AVX10_2-NEXT: # %bb.7: # %start ; AVX10_2-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill ; AVX10_2-NEXT: .LBB42_8: # %start ; AVX10_2-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX10_2-NEXT: vmovaps %xmm2, %xmm0 ; AVX10_2-NEXT: callq __eqtf2@PLT ; AVX10_2-NEXT: testl %eax, %eax ; AVX10_2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: je .LBB42_10 ; AVX10_2-NEXT: # %bb.9: # %start ; AVX10_2-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload ; AVX10_2-NEXT: .LBB42_10: # %start ; AVX10_2-NEXT: addq $40, %rsp ; AVX10_2-NEXT: retq ; ; X86-LABEL: test_fminimumnum_fp128: ; X86: # %bb.0: # %start ; X86-NEXT: pushl %ebp ; X86-NEXT: movl %esp, %ebp ; X86-NEXT: pushl %esi ; X86-NEXT: andl $-16, %esp ; X86-NEXT: subl $80, %esp ; X86-NEXT: movl 8(%ebp), %esi ; X86-NEXT: vmovups 24(%ebp), %ymm0 ; X86-NEXT: vmovups %ymm0, {{[0-9]+}}(%esp) ; X86-NEXT: leal {{[0-9]+}}(%esp), %eax ; X86-NEXT: movl %eax, (%esp) ; X86-NEXT: vzeroupper ; X86-NEXT: calll fminimum_numl ; X86-NEXT: subl $4, %esp ; X86-NEXT: vmovaps {{[0-9]+}}(%esp), %xmm0 ; X86-NEXT: vmovaps %xmm0, (%esi) ; X86-NEXT: movl %esi, %eax ; X86-NEXT: leal -4(%ebp), %esp ; X86-NEXT: popl %esi ; X86-NEXT: popl %ebp ; X86-NEXT: retl $4 start: %0 = tail call fp128 @llvm.minimumnum.f128(fp128 %x, fp128 %y) ret fp128 %0 }