[llvm] ea1f05e - [SelectionDAG] Fix lowering of IEEE 754 2019 minimum/maximum

David Majnemer via llvm-commits llvm-commits at lists.llvm.org
Mon Aug 19 22:10:27 PDT 2024


Author: David Majnemer
Date: 2024-08-20T05:09:49Z
New Revision: ea1f05e6189fbe2b567e795677e1d64acbadc0a2

URL: https://github.com/llvm/llvm-project/commit/ea1f05e6189fbe2b567e795677e1d64acbadc0a2
DIFF: https://github.com/llvm/llvm-project/commit/ea1f05e6189fbe2b567e795677e1d64acbadc0a2.diff

LOG: [SelectionDAG] Fix lowering of IEEE 754 2019 minimum/maximum

We used integer comparisons instead of floating point comparisons
resulting in very odd behavior.

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
    llvm/test/CodeGen/X86/fminimum-fmaximum.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index ae9f11d02a8ac4..4e796289cff0a1 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -8517,7 +8517,7 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
 
     // NaN (if exists) will be propagated later, so orderness doesn't matter.
     SDValue Compare =
-        DAG.getSetCC(DL, CCVT, LHS, RHS, IsMax ? ISD::SETGT : ISD::SETLT);
+        DAG.getSetCC(DL, CCVT, LHS, RHS, IsMax ? ISD::SETOGT : ISD::SETOLT);
     MinMax = DAG.getSelect(DL, VT, Compare, LHS, RHS, Flags);
   }
 
@@ -8534,7 +8534,7 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
   if (!MinMaxMustRespectOrderedZero && !N->getFlags().hasNoSignedZeros() &&
       !DAG.isKnownNeverZeroFloat(RHS) && !DAG.isKnownNeverZeroFloat(LHS)) {
     SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
-                                  DAG.getConstantFP(0.0, DL, VT), ISD::SETEQ);
+                                  DAG.getConstantFP(0.0, DL, VT), ISD::SETOEQ);
     SDValue TestZero =
         DAG.getTargetConstant(IsMax ? fcPosZero : fcNegZero, DL, MVT::i32);
     SDValue LCmp = DAG.getSelect(
@@ -8823,7 +8823,7 @@ SDValue TargetLowering::expandIS_FPCLASS(EVT ResultVT, SDValue Op,
   // Split the value into sign bit and absolute value.
   SDValue AbsV = DAG.getNode(ISD::AND, DL, IntVT, OpAsInt, ValueMaskV);
   SDValue SignV = DAG.getSetCC(DL, ResultVT, OpAsInt,
-                               DAG.getConstant(0.0, DL, IntVT), ISD::SETLT);
+                               DAG.getConstant(0, DL, IntVT), ISD::SETLT);
 
   // Tests that involve more than one class should be processed first.
   SDValue PartialRes;

diff  --git a/llvm/test/CodeGen/X86/fminimum-fmaximum.ll b/llvm/test/CodeGen/X86/fminimum-fmaximum.ll
index 55502f37da1a62..41d9a867c0a960 100644
--- a/llvm/test/CodeGen/X86/fminimum-fmaximum.ll
+++ b/llvm/test/CodeGen/X86/fminimum-fmaximum.ll
@@ -11,6 +11,8 @@ declare float @llvm.minimum.f32(float, float)
 declare double @llvm.minimum.f64(double, double)
 declare <2 x double> @llvm.minimum.v2f64(<2 x double>, <2 x double>)
 declare <4 x float> @llvm.maximum.v4f32(<4 x float>, <4 x float>)
+declare <4 x half> @llvm.maximum.v4f16(<4 x half>, <4 x half>)
+declare <4 x bfloat> @llvm.maximum.v4bf16(<4 x bfloat>, <4 x bfloat>)
 
 ;
 ; fmaximum
@@ -1381,3 +1383,1101 @@ define <4 x float> @test_fmaximum_v4f32_splat(<4 x float> %x, float %y) {
   %r = call <4 x float> @llvm.maximum.v4f32(<4 x float> %x, <4 x float> %vec) readnone
   ret <4 x float> %r
 }
+
+define <4 x half> @test_fmaximum_v4f16(<4 x half> %x, <4 x half> %y) {
+; SSE2-LABEL: test_fmaximum_v4f16:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    subq $104, %rsp
+; SSE2-NEXT:    .cfi_def_cfa_offset 112
+; SSE2-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    psrld $16, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    movdqa %xmm1, (%rsp) # 16-byte Spill
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrld $16, %xmm0
+; SSE2-NEXT:    callq __extendhfsf2 at PLT
+; SSE2-NEXT:    movd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Folded Spill
+; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE2-NEXT:    callq __extendhfsf2 at PLT
+; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 4-byte Folded Reload
+; SSE2-NEXT:    # xmm4 = mem[0],zero,zero,zero
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    testl %eax, %eax
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    js .LBB33_2
+; SSE2-NEXT:  # %bb.1:
+; SSE2-NEXT:    movdqa %xmm4, %xmm2
+; SSE2-NEXT:  .LBB33_2:
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    cmpunordss %xmm2, %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm3
+; SSE2-NEXT:    andps %xmm2, %xmm3
+; SSE2-NEXT:    js .LBB33_4
+; SSE2-NEXT:  # %bb.3:
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:  .LBB33_4:
+; SSE2-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1,1,1]
+; SSE2-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    movaps (%rsp), %xmm1 # 16-byte Reload
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1,1,1]
+; SSE2-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    maxss %xmm4, %xmm2
+; SSE2-NEXT:    andnps %xmm2, %xmm0
+; SSE2-NEXT:    orps %xmm3, %xmm0
+; SSE2-NEXT:    callq __truncsfhf2 at 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 at PLT
+; SSE2-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE2-NEXT:    callq __extendhfsf2 at PLT
+; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 4-byte Folded Reload
+; SSE2-NEXT:    # xmm4 = mem[0],zero,zero,zero
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    testl %eax, %eax
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    js .LBB33_6
+; SSE2-NEXT:  # %bb.5:
+; SSE2-NEXT:    movdqa %xmm4, %xmm2
+; SSE2-NEXT:  .LBB33_6:
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    cmpunordss %xmm2, %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm3
+; SSE2-NEXT:    andps %xmm2, %xmm3
+; SSE2-NEXT:    js .LBB33_8
+; SSE2-NEXT:  # %bb.7:
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:  .LBB33_8:
+; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE2-NEXT:    psrlq $48, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    movdqa (%rsp), %xmm1 # 16-byte Reload
+; SSE2-NEXT:    psrlq $48, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, (%rsp) # 16-byte Spill
+; SSE2-NEXT:    maxss %xmm4, %xmm2
+; SSE2-NEXT:    andnps %xmm2, %xmm0
+; SSE2-NEXT:    orps %xmm3, %xmm0
+; SSE2-NEXT:    callq __truncsfhf2 at 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 at PLT
+; SSE2-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE2-NEXT:    callq __extendhfsf2 at PLT
+; SSE2-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 4-byte Folded Reload
+; SSE2-NEXT:    # xmm4 = mem[0],zero,zero,zero
+; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    testl %eax, %eax
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    js .LBB33_10
+; SSE2-NEXT:  # %bb.9:
+; SSE2-NEXT:    movdqa %xmm4, %xmm2
+; SSE2-NEXT:  .LBB33_10:
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    cmpunordss %xmm2, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm3
+; SSE2-NEXT:    andps %xmm2, %xmm3
+; SSE2-NEXT:    js .LBB33_12
+; SSE2-NEXT:  # %bb.11:
+; SSE2-NEXT:    movdqa %xmm0, %xmm4
+; SSE2-NEXT:  .LBB33_12:
+; SSE2-NEXT:    maxss %xmm4, %xmm2
+; SSE2-NEXT:    andnps %xmm2, %xmm1
+; SSE2-NEXT:    orps %xmm3, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    callq __truncsfhf2 at 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 at PLT
+; SSE2-NEXT:    movss %xmm0, (%rsp) # 4-byte Spill
+; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE2-NEXT:    callq __extendhfsf2 at PLT
+; SSE2-NEXT:    movd (%rsp), %xmm4 # 4-byte Folded Reload
+; SSE2-NEXT:    # xmm4 = mem[0],zero,zero,zero
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    testl %eax, %eax
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    js .LBB33_14
+; SSE2-NEXT:  # %bb.13:
+; SSE2-NEXT:    movdqa %xmm4, %xmm2
+; SSE2-NEXT:  .LBB33_14:
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    cmpunordss %xmm2, %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm3
+; SSE2-NEXT:    andps %xmm2, %xmm3
+; SSE2-NEXT:    js .LBB33_16
+; SSE2-NEXT:  # %bb.15:
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:  .LBB33_16:
+; SSE2-NEXT:    maxss %xmm4, %xmm2
+; SSE2-NEXT:    andnps %xmm2, %xmm0
+; SSE2-NEXT:    orps %xmm3, %xmm0
+; SSE2-NEXT:    callq __truncsfhf2 at 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 $104, %rsp
+; SSE2-NEXT:    .cfi_def_cfa_offset 8
+; SSE2-NEXT:    retq
+;
+; AVX1-LABEL: test_fmaximum_v4f16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    subq $120, %rsp
+; AVX1-NEXT:    .cfi_def_cfa_offset 128
+; AVX1-NEXT:    vmovaps %xmm0, %xmm2
+; AVX1-NEXT:    vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vmovaps %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:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    vpsrld $16, %xmm2, %xmm0
+; AVX1-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    vmovaps %xmm1, (%rsp) # 16-byte Spill
+; AVX1-NEXT:    vpsrld $16, %xmm1, %xmm0
+; AVX1-NEXT:    callq __extendhfsf2 at PLT
+; AVX1-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX1-NEXT:    callq __extendhfsf2 at PLT
+; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    testl %eax, %eax
+; AVX1-NEXT:    js .LBB33_1
+; AVX1-NEXT:  # %bb.2:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; AVX1-NEXT:    jmp .LBB33_3
+; AVX1-NEXT:  .LBB33_1:
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX1-NEXT:  .LBB33_3:
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX1-NEXT:    vpsrlq $48, %xmm0, %xmm0
+; AVX1-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX1-NEXT:    vpsrlq $48, %xmm0, %xmm0
+; AVX1-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    vmaxss %xmm1, %xmm2, %xmm0
+; AVX1-NEXT:    vcmpunordss %xmm2, %xmm2, %xmm1
+; AVX1-NEXT:    vblendvps %xmm1, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    callq __truncsfhf2 at 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 at PLT
+; AVX1-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX1-NEXT:    callq __extendhfsf2 at PLT
+; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    testl %eax, %eax
+; AVX1-NEXT:    js .LBB33_4
+; AVX1-NEXT:  # %bb.5:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa (%rsp), %xmm2 # 16-byte Reload
+; AVX1-NEXT:    jmp .LBB33_6
+; AVX1-NEXT:  .LBB33_4:
+; AVX1-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX1-NEXT:  .LBB33_6:
+; AVX1-NEXT:    vmaxss %xmm1, %xmm2, %xmm0
+; AVX1-NEXT:    vcmpunordss %xmm2, %xmm2, %xmm1
+; AVX1-NEXT:    vblendvps %xmm1, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    callq __truncsfhf2 at 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 at PLT
+; AVX1-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX1-NEXT:    callq __extendhfsf2 at PLT
+; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    testl %eax, %eax
+; AVX1-NEXT:    js .LBB33_7
+; AVX1-NEXT:  # %bb.8:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa (%rsp), %xmm2 # 16-byte Reload
+; AVX1-NEXT:    jmp .LBB33_9
+; AVX1-NEXT:  .LBB33_7:
+; AVX1-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX1-NEXT:  .LBB33_9:
+; AVX1-NEXT:    vmaxss %xmm1, %xmm2, %xmm0
+; AVX1-NEXT:    vcmpunordss %xmm2, %xmm2, %xmm1
+; AVX1-NEXT:    vblendvps %xmm1, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    callq __truncsfhf2 at PLT
+; AVX1-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX1-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX1-NEXT:    callq __extendhfsf2 at PLT
+; AVX1-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX1-NEXT:    callq __extendhfsf2 at PLT
+; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    testl %eax, %eax
+; AVX1-NEXT:    js .LBB33_10
+; AVX1-NEXT:  # %bb.11:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; AVX1-NEXT:    jmp .LBB33_12
+; AVX1-NEXT:  .LBB33_10:
+; AVX1-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX1-NEXT:  .LBB33_12:
+; AVX1-NEXT:    vmaxss %xmm1, %xmm2, %xmm0
+; AVX1-NEXT:    vcmpunordss %xmm2, %xmm2, %xmm1
+; AVX1-NEXT:    vblendvps %xmm1, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    callq __truncsfhf2 at 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 $120, %rsp
+; AVX1-NEXT:    .cfi_def_cfa_offset 8
+; AVX1-NEXT:    retq
+;
+; AVX512-LABEL: test_fmaximum_v4f16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    pushq %rbp
+; AVX512-NEXT:    .cfi_def_cfa_offset 16
+; AVX512-NEXT:    pushq %r15
+; AVX512-NEXT:    .cfi_def_cfa_offset 24
+; AVX512-NEXT:    pushq %r14
+; AVX512-NEXT:    .cfi_def_cfa_offset 32
+; AVX512-NEXT:    pushq %r13
+; AVX512-NEXT:    .cfi_def_cfa_offset 40
+; AVX512-NEXT:    pushq %r12
+; AVX512-NEXT:    .cfi_def_cfa_offset 48
+; AVX512-NEXT:    pushq %rbx
+; AVX512-NEXT:    .cfi_def_cfa_offset 56
+; AVX512-NEXT:    .cfi_offset %rbx, -56
+; AVX512-NEXT:    .cfi_offset %r12, -48
+; AVX512-NEXT:    .cfi_offset %r13, -40
+; AVX512-NEXT:    .cfi_offset %r14, -32
+; AVX512-NEXT:    .cfi_offset %r15, -24
+; AVX512-NEXT:    .cfi_offset %rbp, -16
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[3,3,3,3]
+; AVX512-NEXT:    vcvtph2ps %xmm2, %xmm2
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[3,3,3,3]
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    xorl %eax, %eax
+; AVX512-NEXT:    vucomiss %xmm2, %xmm3
+; AVX512-NEXT:    movl $65535, %ecx # imm = 0xFFFF
+; AVX512-NEXT:    movl $0, %edx
+; AVX512-NEXT:    cmovpl %ecx, %edx
+; AVX512-NEXT:    movl $0, %edi
+; AVX512-NEXT:    cmoval %ecx, %edi
+; AVX512-NEXT:    vpsrldq {{.*#+}} xmm2 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512-NEXT:    vcvtph2ps %xmm2, %xmm2
+; AVX512-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    vucomiss %xmm2, %xmm3
+; AVX512-NEXT:    movl $0, %esi
+; AVX512-NEXT:    cmovpl %ecx, %esi
+; AVX512-NEXT:    movl $0, %r9d
+; AVX512-NEXT:    cmoval %ecx, %r9d
+; AVX512-NEXT:    vshufpd {{.*#+}} xmm2 = xmm1[1,0]
+; AVX512-NEXT:    vcvtph2ps %xmm2, %xmm2
+; AVX512-NEXT:    vshufpd {{.*#+}} xmm3 = xmm0[1,0]
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    vucomiss %xmm2, %xmm3
+; AVX512-NEXT:    movl $0, %r8d
+; AVX512-NEXT:    cmovpl %ecx, %r8d
+; AVX512-NEXT:    movl $0, %r11d
+; AVX512-NEXT:    cmoval %ecx, %r11d
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm1[3,3,3,3,4,5,6,7]
+; AVX512-NEXT:    vcvtph2ps %xmm2, %xmm2
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm0[3,3,3,3,4,5,6,7]
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    vucomiss %xmm2, %xmm3
+; AVX512-NEXT:    movl $0, %r10d
+; AVX512-NEXT:    cmovpl %ecx, %r10d
+; AVX512-NEXT:    movl $0, %ebp
+; AVX512-NEXT:    cmoval %ecx, %ebp
+; AVX512-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; AVX512-NEXT:    vcvtph2ps %xmm2, %xmm2
+; AVX512-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    vucomiss %xmm2, %xmm3
+; AVX512-NEXT:    movl $0, %ebx
+; AVX512-NEXT:    cmovpl %ecx, %ebx
+; AVX512-NEXT:    movl $0, %r14d
+; AVX512-NEXT:    cmoval %ecx, %r14d
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm1[1,1,1,1,4,5,6,7]
+; AVX512-NEXT:    vcvtph2ps %xmm2, %xmm2
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm0[1,1,1,1,4,5,6,7]
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    vucomiss %xmm2, %xmm3
+; AVX512-NEXT:    movl $0, %r15d
+; AVX512-NEXT:    cmovpl %ecx, %r15d
+; AVX512-NEXT:    movl $0, %r12d
+; AVX512-NEXT:    cmoval %ecx, %r12d
+; AVX512-NEXT:    vcvtph2ps %xmm1, %xmm2
+; AVX512-NEXT:    vcvtph2ps %xmm0, %xmm3
+; AVX512-NEXT:    vucomiss %xmm2, %xmm3
+; AVX512-NEXT:    movl $0, %r13d
+; AVX512-NEXT:    cmoval %ecx, %r13d
+; AVX512-NEXT:    vmovd %r13d, %xmm2
+; AVX512-NEXT:    vpinsrw $1, %r12d, %xmm2, %xmm2
+; AVX512-NEXT:    vpinsrw $2, %r14d, %xmm2, %xmm2
+; AVX512-NEXT:    vpinsrw $3, %ebp, %xmm2, %xmm2
+; AVX512-NEXT:    vpinsrw $4, %r11d, %xmm2, %xmm2
+; AVX512-NEXT:    vpinsrw $5, %r9d, %xmm2, %xmm2
+; AVX512-NEXT:    vpinsrw $6, %edi, %xmm2, %xmm2
+; AVX512-NEXT:    movl $0, %edi
+; AVX512-NEXT:    cmovpl %ecx, %edi
+; AVX512-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512-NEXT:    vcvtph2ps %xmm4, %xmm4
+; AVX512-NEXT:    vucomiss %xmm3, %xmm4
+; AVX512-NEXT:    movl $0, %r9d
+; AVX512-NEXT:    cmoval %ecx, %r9d
+; AVX512-NEXT:    vpinsrw $7, %r9d, %xmm2, %xmm2
+; AVX512-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm2
+; AVX512-NEXT:    vmovd %edi, %xmm3
+; AVX512-NEXT:    vpinsrw $1, %r15d, %xmm3, %xmm3
+; AVX512-NEXT:    vpinsrw $2, %ebx, %xmm3, %xmm3
+; AVX512-NEXT:    vpinsrw $3, %r10d, %xmm3, %xmm3
+; AVX512-NEXT:    vpinsrw $4, %r8d, %xmm3, %xmm3
+; AVX512-NEXT:    vpinsrw $5, %esi, %xmm3, %xmm3
+; AVX512-NEXT:    vpinsrw $6, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    movl $0, %edx
+; AVX512-NEXT:    cmovpl %ecx, %edx
+; AVX512-NEXT:    vpinsrw $7, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    vpbroadcastw {{.*#+}} xmm4 = [NaN,NaN,NaN,NaN,NaN,NaN,NaN,NaN]
+; AVX512-NEXT:    vpblendvb %xmm3, %xmm4, %xmm2, %xmm2
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm2[1,1,1,1,4,5,6,7]
+; AVX512-NEXT:    vcvtph2ps %xmm3, %xmm3
+; AVX512-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512-NEXT:    vucomiss %xmm4, %xmm3
+; AVX512-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; AVX512-NEXT:    cmovnel %eax, %edx
+; AVX512-NEXT:    cmovpl %eax, %edx
+; AVX512-NEXT:    vcvtph2ps %xmm2, %xmm3
+; AVX512-NEXT:    vucomiss %xmm4, %xmm3
+; AVX512-NEXT:    movl $65535, %esi # imm = 0xFFFF
+; AVX512-NEXT:    cmovnel %eax, %esi
+; AVX512-NEXT:    cmovpl %eax, %esi
+; AVX512-NEXT:    vmovd %esi, %xmm3
+; AVX512-NEXT:    vpinsrw $1, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
+; AVX512-NEXT:    vcvtph2ps %xmm5, %xmm5
+; AVX512-NEXT:    vucomiss %xmm4, %xmm5
+; AVX512-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; AVX512-NEXT:    cmovnel %eax, %edx
+; AVX512-NEXT:    cmovpl %eax, %edx
+; AVX512-NEXT:    vpinsrw $2, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm5 = xmm2[3,3,3,3,4,5,6,7]
+; AVX512-NEXT:    vcvtph2ps %xmm5, %xmm5
+; AVX512-NEXT:    vucomiss %xmm4, %xmm5
+; AVX512-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; AVX512-NEXT:    cmovnel %eax, %edx
+; AVX512-NEXT:    cmovpl %eax, %edx
+; AVX512-NEXT:    vpinsrw $3, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[2,3,2,3]
+; AVX512-NEXT:    vcvtph2ps %xmm5, %xmm5
+; AVX512-NEXT:    vucomiss %xmm4, %xmm5
+; AVX512-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; AVX512-NEXT:    cmovnel %eax, %edx
+; AVX512-NEXT:    cmovpl %eax, %edx
+; AVX512-NEXT:    vpinsrw $4, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm2[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512-NEXT:    vcvtph2ps %xmm5, %xmm5
+; AVX512-NEXT:    vucomiss %xmm4, %xmm5
+; AVX512-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; AVX512-NEXT:    cmovnel %eax, %edx
+; AVX512-NEXT:    cmovpl %eax, %edx
+; AVX512-NEXT:    vpinsrw $5, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[3,3,3,3]
+; AVX512-NEXT:    vcvtph2ps %xmm5, %xmm5
+; AVX512-NEXT:    vucomiss %xmm4, %xmm5
+; AVX512-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; AVX512-NEXT:    cmovnel %eax, %edx
+; AVX512-NEXT:    cmovpl %eax, %edx
+; AVX512-NEXT:    vpinsrw $6, %edx, %xmm3, %xmm3
+; AVX512-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512-NEXT:    vcvtph2ps %xmm5, %xmm5
+; AVX512-NEXT:    vucomiss %xmm4, %xmm5
+; AVX512-NEXT:    cmovnel %eax, %ecx
+; AVX512-NEXT:    cmovpl %eax, %ecx
+; AVX512-NEXT:    vpinsrw $7, %ecx, %xmm3, %xmm3
+; AVX512-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm5
+; AVX512-NEXT:    vpblendvb %xmm5, %xmm0, %xmm2, %xmm0
+; AVX512-NEXT:    vpcmpeqw %xmm4, %xmm1, %xmm4
+; AVX512-NEXT:    vpblendvb %xmm4, %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpblendvb %xmm3, %xmm0, %xmm2, %xmm0
+; AVX512-NEXT:    popq %rbx
+; AVX512-NEXT:    .cfi_def_cfa_offset 48
+; AVX512-NEXT:    popq %r12
+; AVX512-NEXT:    .cfi_def_cfa_offset 40
+; AVX512-NEXT:    popq %r13
+; AVX512-NEXT:    .cfi_def_cfa_offset 32
+; AVX512-NEXT:    popq %r14
+; AVX512-NEXT:    .cfi_def_cfa_offset 24
+; AVX512-NEXT:    popq %r15
+; AVX512-NEXT:    .cfi_def_cfa_offset 16
+; AVX512-NEXT:    popq %rbp
+; AVX512-NEXT:    .cfi_def_cfa_offset 8
+; AVX512-NEXT:    retq
+;
+; X86-LABEL: test_fmaximum_v4f16:
+; X86:       # %bb.0:
+; X86-NEXT:    subl $164, %esp
+; X86-NEXT:    .cfi_def_cfa_offset 168
+; 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:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero
+; X86-NEXT:    vmovd %xmm2, %eax
+; X86-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    testl %eax, %eax
+; X86-NEXT:    js .LBB33_1
+; X86-NEXT:  # %bb.2:
+; X86-NEXT:    vmovdqa %xmm2, %xmm1
+; X86-NEXT:    jmp .LBB33_3
+; X86-NEXT:  .LBB33_1:
+; X86-NEXT:    vmovdqa %xmm0, %xmm1
+; X86-NEXT:    vmovdqa %xmm2, %xmm0
+; X86-NEXT:  .LBB33_3:
+; X86-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %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:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X86-NEXT:    vmovd %xmm1, %eax
+; X86-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    testl %eax, %eax
+; X86-NEXT:    js .LBB33_4
+; X86-NEXT:  # %bb.5:
+; X86-NEXT:    vmovdqa %xmm1, %xmm2
+; X86-NEXT:    jmp .LBB33_6
+; X86-NEXT:  .LBB33_4:
+; X86-NEXT:    vmovdqa %xmm0, %xmm2
+; X86-NEXT:    vmovdqa %xmm1, %xmm0
+; X86-NEXT:  .LBB33_6:
+; X86-NEXT:    vmaxss %xmm2, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %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:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X86-NEXT:    vmovd %xmm1, %eax
+; X86-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    testl %eax, %eax
+; X86-NEXT:    js .LBB33_7
+; X86-NEXT:  # %bb.8:
+; X86-NEXT:    vmovdqa %xmm1, %xmm2
+; X86-NEXT:    jmp .LBB33_9
+; X86-NEXT:  .LBB33_7:
+; X86-NEXT:    vmovdqa %xmm0, %xmm2
+; X86-NEXT:    vmovdqa %xmm1, %xmm0
+; X86-NEXT:  .LBB33_9:
+; X86-NEXT:    vmaxss %xmm2, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %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:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X86-NEXT:    vmovd %xmm1, %eax
+; X86-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    testl %eax, %eax
+; X86-NEXT:    js .LBB33_10
+; X86-NEXT:  # %bb.11:
+; X86-NEXT:    vmovdqa %xmm1, %xmm2
+; X86-NEXT:    jmp .LBB33_12
+; X86-NEXT:  .LBB33_10:
+; X86-NEXT:    vmovdqa %xmm0, %xmm2
+; X86-NEXT:    vmovdqa %xmm1, %xmm0
+; X86-NEXT:  .LBB33_12:
+; X86-NEXT:    vmaxss %xmm2, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %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:    .cfi_def_cfa_offset 4
+; X86-NEXT:    retl
+  %r = call <4 x half> @llvm.maximum.v4f16(<4 x half> %x, <4 x half> %y)
+  ret <4 x half> %r
+}
+
+define <4 x bfloat> @test_fmaximum_v4bf16(<4 x bfloat> %x, <4 x bfloat> %y) {
+; SSE2-LABEL: test_fmaximum_v4bf16:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    pushq %rbp
+; SSE2-NEXT:    .cfi_def_cfa_offset 16
+; SSE2-NEXT:    pushq %r15
+; SSE2-NEXT:    .cfi_def_cfa_offset 24
+; SSE2-NEXT:    pushq %r14
+; SSE2-NEXT:    .cfi_def_cfa_offset 32
+; SSE2-NEXT:    pushq %rbx
+; SSE2-NEXT:    .cfi_def_cfa_offset 40
+; SSE2-NEXT:    subq $56, %rsp
+; SSE2-NEXT:    .cfi_def_cfa_offset 96
+; SSE2-NEXT:    .cfi_offset %rbx, -40
+; SSE2-NEXT:    .cfi_offset %r14, -32
+; SSE2-NEXT:    .cfi_offset %r15, -24
+; SSE2-NEXT:    .cfi_offset %rbp, -16
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:    movdqa %xmm0, %xmm5
+; SSE2-NEXT:    pextrw $0, %xmm1, %r14d
+; SSE2-NEXT:    pextrw $0, %xmm0, %r15d
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrld $16, %xmm0
+; SSE2-NEXT:    pextrw $0, %xmm0, %eax
+; SSE2-NEXT:    movdqa %xmm5, %xmm0
+; SSE2-NEXT:    psrld $16, %xmm0
+; SSE2-NEXT:    pextrw $0, %xmm0, %ecx
+; SSE2-NEXT:    shll $16, %ecx
+; SSE2-NEXT:    movd %ecx, %xmm3
+; SSE2-NEXT:    shll $16, %eax
+; SSE2-NEXT:    movd %eax, %xmm2
+; SSE2-NEXT:    testl %ecx, %ecx
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
+; SSE2-NEXT:    js .LBB34_2
+; SSE2-NEXT:  # %bb.1:
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:  .LBB34_2:
+; SSE2-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    shufps {{.*#+}} xmm4 = xmm4[1,1,1,1]
+; SSE2-NEXT:    movdqa %xmm5, (%rsp) # 16-byte Spill
+; SSE2-NEXT:    shufps {{.*#+}} xmm5 = xmm5[1,1,1,1]
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    cmpunordss %xmm1, %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm6
+; SSE2-NEXT:    andps %xmm1, %xmm6
+; SSE2-NEXT:    js .LBB34_4
+; SSE2-NEXT:  # %bb.3:
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:  .LBB34_4:
+; SSE2-NEXT:    pextrw $0, %xmm4, %ebp
+; SSE2-NEXT:    pextrw $0, %xmm5, %ebx
+; SSE2-NEXT:    maxss %xmm2, %xmm1
+; SSE2-NEXT:    andnps %xmm1, %xmm0
+; SSE2-NEXT:    orps %xmm6, %xmm0
+; SSE2-NEXT:    callq __truncsfbf2 at PLT
+; SSE2-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    shll $16, %r15d
+; SSE2-NEXT:    movd %r15d, %xmm3
+; SSE2-NEXT:    shll $16, %r14d
+; SSE2-NEXT:    movd %r14d, %xmm2
+; SSE2-NEXT:    testl %r15d, %r15d
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
+; SSE2-NEXT:    js .LBB34_6
+; SSE2-NEXT:  # %bb.5:
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:  .LBB34_6:
+; SSE2-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; SSE2-NEXT:    psrlq $48, %xmm5
+; SSE2-NEXT:    movdqa (%rsp), %xmm6 # 16-byte Reload
+; SSE2-NEXT:    psrlq $48, %xmm6
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    cmpunordss %xmm1, %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm4
+; SSE2-NEXT:    andps %xmm1, %xmm4
+; SSE2-NEXT:    js .LBB34_8
+; SSE2-NEXT:  # %bb.7:
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:  .LBB34_8:
+; SSE2-NEXT:    pextrw $0, %xmm5, %r15d
+; SSE2-NEXT:    pextrw $0, %xmm6, %r14d
+; SSE2-NEXT:    maxss %xmm2, %xmm1
+; SSE2-NEXT:    andnps %xmm1, %xmm0
+; SSE2-NEXT:    orps %xmm4, %xmm0
+; SSE2-NEXT:    callq __truncsfbf2 at PLT
+; SSE2-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE2-NEXT:    shll $16, %ebx
+; SSE2-NEXT:    movd %ebx, %xmm1
+; SSE2-NEXT:    shll $16, %ebp
+; SSE2-NEXT:    movd %ebp, %xmm3
+; SSE2-NEXT:    testl %ebx, %ebx
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    js .LBB34_10
+; SSE2-NEXT:  # %bb.9:
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:  .LBB34_10:
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    cmpunordss %xmm2, %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm4
+; SSE2-NEXT:    andps %xmm2, %xmm4
+; SSE2-NEXT:    js .LBB34_12
+; SSE2-NEXT:  # %bb.11:
+; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:  .LBB34_12:
+; SSE2-NEXT:    maxss %xmm3, %xmm2
+; SSE2-NEXT:    andnps %xmm2, %xmm0
+; SSE2-NEXT:    orps %xmm4, %xmm0
+; SSE2-NEXT:    callq __truncsfbf2 at PLT
+; SSE2-NEXT:    movaps %xmm0, (%rsp) # 16-byte Spill
+; SSE2-NEXT:    shll $16, %r14d
+; SSE2-NEXT:    movd %r14d, %xmm1
+; SSE2-NEXT:    shll $16, %r15d
+; SSE2-NEXT:    movd %r15d, %xmm3
+; SSE2-NEXT:    testl %r14d, %r14d
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    js .LBB34_14
+; SSE2-NEXT:  # %bb.13:
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:  .LBB34_14:
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    cmpunordss %xmm2, %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm4
+; SSE2-NEXT:    andps %xmm2, %xmm4
+; SSE2-NEXT:    js .LBB34_16
+; SSE2-NEXT:  # %bb.15:
+; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:  .LBB34_16:
+; SSE2-NEXT:    maxss %xmm3, %xmm2
+; SSE2-NEXT:    andnps %xmm2, %xmm0
+; SSE2-NEXT:    orps %xmm4, %xmm0
+; SSE2-NEXT:    callq __truncsfbf2 at 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:    .cfi_def_cfa_offset 40
+; SSE2-NEXT:    popq %rbx
+; SSE2-NEXT:    .cfi_def_cfa_offset 32
+; SSE2-NEXT:    popq %r14
+; SSE2-NEXT:    .cfi_def_cfa_offset 24
+; SSE2-NEXT:    popq %r15
+; SSE2-NEXT:    .cfi_def_cfa_offset 16
+; SSE2-NEXT:    popq %rbp
+; SSE2-NEXT:    .cfi_def_cfa_offset 8
+; SSE2-NEXT:    retq
+;
+; AVX1-LABEL: test_fmaximum_v4bf16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    pushq %rbp
+; AVX1-NEXT:    .cfi_def_cfa_offset 16
+; AVX1-NEXT:    pushq %r15
+; AVX1-NEXT:    .cfi_def_cfa_offset 24
+; AVX1-NEXT:    pushq %r14
+; AVX1-NEXT:    .cfi_def_cfa_offset 32
+; AVX1-NEXT:    pushq %r13
+; AVX1-NEXT:    .cfi_def_cfa_offset 40
+; AVX1-NEXT:    pushq %r12
+; AVX1-NEXT:    .cfi_def_cfa_offset 48
+; AVX1-NEXT:    pushq %rbx
+; AVX1-NEXT:    .cfi_def_cfa_offset 56
+; AVX1-NEXT:    subq $56, %rsp
+; AVX1-NEXT:    .cfi_def_cfa_offset 112
+; AVX1-NEXT:    .cfi_offset %rbx, -56
+; AVX1-NEXT:    .cfi_offset %r12, -48
+; AVX1-NEXT:    .cfi_offset %r13, -40
+; AVX1-NEXT:    .cfi_offset %r14, -32
+; AVX1-NEXT:    .cfi_offset %r15, -24
+; AVX1-NEXT:    .cfi_offset %rbp, -16
+; AVX1-NEXT:    vpsrlq $48, %xmm0, %xmm2
+; AVX1-NEXT:    vpsrlq $48, %xmm1, %xmm3
+; AVX1-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpextrw $0, %xmm4, %ebx
+; AVX1-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vpextrw $0, %xmm4, %ebp
+; 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, %xmm4
+; AVX1-NEXT:    js .LBB34_1
+; AVX1-NEXT:  # %bb.2:
+; AVX1-NEXT:    vmovdqa %xmm4, %xmm1
+; AVX1-NEXT:    jmp .LBB34_3
+; AVX1-NEXT:  .LBB34_1:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa %xmm4, %xmm0
+; AVX1-NEXT:  .LBB34_3:
+; AVX1-NEXT:    vpextrw $0, %xmm2, %r14d
+; AVX1-NEXT:    vpextrw $0, %xmm3, %r15d
+; AVX1-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vcmpunordss %xmm0, %xmm0, %xmm2
+; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    callq __truncsfbf2 at 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, %xmm2
+; AVX1-NEXT:    js .LBB34_4
+; AVX1-NEXT:  # %bb.5:
+; AVX1-NEXT:    vmovdqa %xmm2, %xmm1
+; AVX1-NEXT:    jmp .LBB34_6
+; AVX1-NEXT:  .LBB34_4:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa %xmm2, %xmm0
+; AVX1-NEXT:  .LBB34_6:
+; AVX1-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vcmpunordss %xmm0, %xmm0, %xmm2
+; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    callq __truncsfbf2 at PLT
+; AVX1-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    shll $16, %ebp
+; AVX1-NEXT:    vmovd %ebp, %xmm0
+; AVX1-NEXT:    shll $16, %ebx
+; AVX1-NEXT:    vmovd %ebx, %xmm2
+; AVX1-NEXT:    js .LBB34_7
+; AVX1-NEXT:  # %bb.8:
+; AVX1-NEXT:    vmovdqa %xmm2, %xmm1
+; AVX1-NEXT:    jmp .LBB34_9
+; AVX1-NEXT:  .LBB34_7:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa %xmm2, %xmm0
+; AVX1-NEXT:  .LBB34_9:
+; AVX1-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vcmpunordss %xmm0, %xmm0, %xmm2
+; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    callq __truncsfbf2 at PLT
+; AVX1-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX1-NEXT:    shll $16, %r15d
+; AVX1-NEXT:    vmovd %r15d, %xmm0
+; AVX1-NEXT:    shll $16, %r14d
+; AVX1-NEXT:    vmovd %r14d, %xmm2
+; AVX1-NEXT:    js .LBB34_10
+; AVX1-NEXT:  # %bb.11:
+; AVX1-NEXT:    vmovdqa %xmm2, %xmm1
+; AVX1-NEXT:    jmp .LBB34_12
+; AVX1-NEXT:  .LBB34_10:
+; AVX1-NEXT:    vmovdqa %xmm0, %xmm1
+; AVX1-NEXT:    vmovdqa %xmm2, %xmm0
+; AVX1-NEXT:  .LBB34_12:
+; AVX1-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vcmpunordss %xmm0, %xmm0, %xmm2
+; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    callq __truncsfbf2 at 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:    .cfi_def_cfa_offset 56
+; AVX1-NEXT:    popq %rbx
+; AVX1-NEXT:    .cfi_def_cfa_offset 48
+; AVX1-NEXT:    popq %r12
+; AVX1-NEXT:    .cfi_def_cfa_offset 40
+; AVX1-NEXT:    popq %r13
+; AVX1-NEXT:    .cfi_def_cfa_offset 32
+; AVX1-NEXT:    popq %r14
+; AVX1-NEXT:    .cfi_def_cfa_offset 24
+; AVX1-NEXT:    popq %r15
+; AVX1-NEXT:    .cfi_def_cfa_offset 16
+; AVX1-NEXT:    popq %rbp
+; AVX1-NEXT:    .cfi_def_cfa_offset 8
+; AVX1-NEXT:    retq
+;
+; AVX512-LABEL: test_fmaximum_v4bf16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    pushq %rbp
+; AVX512-NEXT:    .cfi_def_cfa_offset 16
+; AVX512-NEXT:    pushq %r15
+; AVX512-NEXT:    .cfi_def_cfa_offset 24
+; AVX512-NEXT:    pushq %r14
+; AVX512-NEXT:    .cfi_def_cfa_offset 32
+; AVX512-NEXT:    pushq %r13
+; AVX512-NEXT:    .cfi_def_cfa_offset 40
+; AVX512-NEXT:    pushq %r12
+; AVX512-NEXT:    .cfi_def_cfa_offset 48
+; AVX512-NEXT:    pushq %rbx
+; AVX512-NEXT:    .cfi_def_cfa_offset 56
+; AVX512-NEXT:    pushq %rax
+; AVX512-NEXT:    .cfi_def_cfa_offset 64
+; AVX512-NEXT:    .cfi_offset %rbx, -56
+; AVX512-NEXT:    .cfi_offset %r12, -48
+; AVX512-NEXT:    .cfi_offset %r13, -40
+; AVX512-NEXT:    .cfi_offset %r14, -32
+; AVX512-NEXT:    .cfi_offset %r15, -24
+; AVX512-NEXT:    .cfi_offset %rbp, -16
+; AVX512-NEXT:    vmovq %xmm1, %r13
+; AVX512-NEXT:    movq %r13, %rbx
+; AVX512-NEXT:    shrq $32, %rbx
+; AVX512-NEXT:    vmovq %xmm0, %rbp
+; AVX512-NEXT:    movq %rbp, %r14
+; AVX512-NEXT:    shrq $32, %r14
+; AVX512-NEXT:    movq %r13, %r15
+; AVX512-NEXT:    shrq $48, %r15
+; AVX512-NEXT:    movq %rbp, %r12
+; AVX512-NEXT:    shrq $48, %r12
+; AVX512-NEXT:    movl %ebp, %eax
+; AVX512-NEXT:    andl $-65536, %eax # imm = 0xFFFF0000
+; AVX512-NEXT:    sets %cl
+; AVX512-NEXT:    kmovw %ecx, %k1
+; AVX512-NEXT:    movl %r13d, %ecx
+; AVX512-NEXT:    andl $-65536, %ecx # imm = 0xFFFF0000
+; AVX512-NEXT:    vmovd %ecx, %xmm1
+; AVX512-NEXT:    vmovd %eax, %xmm0
+; AVX512-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX512-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; AVX512-NEXT:    vmovss %xmm0, %xmm1, %xmm1 {%k1}
+; AVX512-NEXT:    vmaxss %xmm2, %xmm1, %xmm0
+; AVX512-NEXT:    vcmpunordss %xmm1, %xmm1, %k1
+; AVX512-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1}
+; AVX512-NEXT:    callq __truncsfbf2 at PLT
+; AVX512-NEXT:    vpextrw $0, %xmm0, {{[0-9]+}}(%rsp)
+; AVX512-NEXT:    shll $16, %ebp
+; AVX512-NEXT:    sets %al
+; AVX512-NEXT:    kmovw %eax, %k1
+; AVX512-NEXT:    shll $16, %r13d
+; AVX512-NEXT:    vmovd %r13d, %xmm1
+; AVX512-NEXT:    vmovd %ebp, %xmm0
+; AVX512-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX512-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; AVX512-NEXT:    vmovss %xmm0, %xmm1, %xmm1 {%k1}
+; AVX512-NEXT:    vmaxss %xmm2, %xmm1, %xmm0
+; AVX512-NEXT:    vcmpunordss %xmm1, %xmm1, %k1
+; AVX512-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1}
+; AVX512-NEXT:    callq __truncsfbf2 at PLT
+; AVX512-NEXT:    vpextrw $0, %xmm0, (%rsp)
+; AVX512-NEXT:    shll $16, %r12d
+; AVX512-NEXT:    sets %al
+; AVX512-NEXT:    kmovw %eax, %k1
+; AVX512-NEXT:    shll $16, %r15d
+; AVX512-NEXT:    vmovd %r15d, %xmm1
+; AVX512-NEXT:    vmovd %r12d, %xmm0
+; AVX512-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX512-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; AVX512-NEXT:    vmovss %xmm0, %xmm1, %xmm1 {%k1}
+; AVX512-NEXT:    vmaxss %xmm2, %xmm1, %xmm0
+; AVX512-NEXT:    vcmpunordss %xmm1, %xmm1, %k1
+; AVX512-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1}
+; AVX512-NEXT:    callq __truncsfbf2 at PLT
+; AVX512-NEXT:    vpextrw $0, %xmm0, {{[0-9]+}}(%rsp)
+; AVX512-NEXT:    shll $16, %r14d
+; AVX512-NEXT:    sets %al
+; AVX512-NEXT:    kmovw %eax, %k1
+; AVX512-NEXT:    shll $16, %ebx
+; AVX512-NEXT:    vmovd %ebx, %xmm1
+; AVX512-NEXT:    vmovd %r14d, %xmm0
+; AVX512-NEXT:    vmovdqa %xmm0, %xmm2
+; AVX512-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; AVX512-NEXT:    vmovss %xmm0, %xmm1, %xmm1 {%k1}
+; AVX512-NEXT:    vmaxss %xmm2, %xmm1, %xmm0
+; AVX512-NEXT:    vcmpunordss %xmm1, %xmm1, %k1
+; AVX512-NEXT:    vmovss %xmm1, %xmm0, %xmm0 {%k1}
+; AVX512-NEXT:    callq __truncsfbf2 at PLT
+; AVX512-NEXT:    vpextrw $0, %xmm0, {{[0-9]+}}(%rsp)
+; AVX512-NEXT:    vmovaps (%rsp), %xmm0
+; AVX512-NEXT:    addq $8, %rsp
+; AVX512-NEXT:    .cfi_def_cfa_offset 56
+; AVX512-NEXT:    popq %rbx
+; AVX512-NEXT:    .cfi_def_cfa_offset 48
+; AVX512-NEXT:    popq %r12
+; AVX512-NEXT:    .cfi_def_cfa_offset 40
+; AVX512-NEXT:    popq %r13
+; AVX512-NEXT:    .cfi_def_cfa_offset 32
+; AVX512-NEXT:    popq %r14
+; AVX512-NEXT:    .cfi_def_cfa_offset 24
+; AVX512-NEXT:    popq %r15
+; AVX512-NEXT:    .cfi_def_cfa_offset 16
+; AVX512-NEXT:    popq %rbp
+; AVX512-NEXT:    .cfi_def_cfa_offset 8
+; AVX512-NEXT:    retq
+;
+; X86-LABEL: test_fmaximum_v4bf16:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    .cfi_def_cfa_offset 12
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    .cfi_def_cfa_offset 16
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    .cfi_def_cfa_offset 20
+; X86-NEXT:    subl $68, %esp
+; X86-NEXT:    .cfi_def_cfa_offset 88
+; X86-NEXT:    .cfi_offset %esi, -20
+; X86-NEXT:    .cfi_offset %edi, -16
+; X86-NEXT:    .cfi_offset %ebx, -12
+; X86-NEXT:    .cfi_offset %ebp, -8
+; X86-NEXT:    vpsrlq $48, %xmm0, %xmm2
+; X86-NEXT:    vpsrlq $48, %xmm1, %xmm3
+; X86-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm0[1,1,3,3]
+; X86-NEXT:    vpextrw $0, %xmm4, %esi
+; X86-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm1[1,1,3,3]
+; X86-NEXT:    vpextrw $0, %xmm4, %ebx
+; X86-NEXT:    vpextrw $0, %xmm0, %eax
+; X86-NEXT:    vpextrw $0, %xmm1, %ecx
+; X86-NEXT:    vpsrld $16, %xmm0, %xmm0
+; X86-NEXT:    vpextrw $0, %xmm0, %edx
+; X86-NEXT:    vpsrld $16, %xmm1, %xmm0
+; X86-NEXT:    vpextrw $0, %xmm0, %edi
+; X86-NEXT:    shll $16, %edi
+; X86-NEXT:    vmovd %edi, %xmm0
+; X86-NEXT:    shll $16, %edx
+; X86-NEXT:    vmovd %edx, %xmm4
+; X86-NEXT:    js .LBB34_1
+; X86-NEXT:  # %bb.2:
+; X86-NEXT:    vmovdqa %xmm4, %xmm1
+; X86-NEXT:    jmp .LBB34_3
+; X86-NEXT:  .LBB34_1:
+; X86-NEXT:    vmovdqa %xmm0, %xmm1
+; X86-NEXT:    vmovdqa %xmm4, %xmm0
+; X86-NEXT:  .LBB34_3:
+; X86-NEXT:    vpextrw $0, %xmm2, %edi
+; X86-NEXT:    vpextrw $0, %xmm3, %ebp
+; X86-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %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, %xmm2
+; X86-NEXT:    js .LBB34_4
+; X86-NEXT:  # %bb.5:
+; X86-NEXT:    vmovdqa %xmm2, %xmm1
+; X86-NEXT:    jmp .LBB34_6
+; X86-NEXT:  .LBB34_4:
+; X86-NEXT:    vmovdqa %xmm0, %xmm1
+; X86-NEXT:    vmovdqa %xmm2, %xmm0
+; X86-NEXT:  .LBB34_6:
+; X86-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %xmm2
+; X86-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %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, %ebx
+; X86-NEXT:    vmovd %ebx, %xmm0
+; X86-NEXT:    shll $16, %esi
+; X86-NEXT:    vmovd %esi, %xmm2
+; X86-NEXT:    js .LBB34_7
+; X86-NEXT:  # %bb.8:
+; X86-NEXT:    vmovdqa %xmm2, %xmm1
+; X86-NEXT:    jmp .LBB34_9
+; X86-NEXT:  .LBB34_7:
+; X86-NEXT:    vmovdqa %xmm0, %xmm1
+; X86-NEXT:    vmovdqa %xmm2, %xmm0
+; X86-NEXT:  .LBB34_9:
+; X86-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %xmm2
+; X86-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %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, %edi
+; X86-NEXT:    vmovd %edi, %xmm2
+; X86-NEXT:    js .LBB34_10
+; X86-NEXT:  # %bb.11:
+; X86-NEXT:    vmovdqa %xmm2, %xmm1
+; X86-NEXT:    jmp .LBB34_12
+; X86-NEXT:  .LBB34_10:
+; X86-NEXT:    vmovdqa %xmm0, %xmm1
+; X86-NEXT:    vmovdqa %xmm2, %xmm0
+; X86-NEXT:  .LBB34_12:
+; X86-NEXT:    vmaxss %xmm1, %xmm0, %xmm1
+; X86-NEXT:    vcmpunordss %xmm0, %xmm0, %xmm2
+; X86-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %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:    .cfi_def_cfa_offset 20
+; X86-NEXT:    popl %esi
+; X86-NEXT:    .cfi_def_cfa_offset 16
+; X86-NEXT:    popl %edi
+; X86-NEXT:    .cfi_def_cfa_offset 12
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    .cfi_def_cfa_offset 4
+; X86-NEXT:    retl
+  %r = call <4 x bfloat> @llvm.maximum.v4bf16(<4 x bfloat> %x, <4 x bfloat> %y)
+  ret <4 x bfloat> %r
+}


        


More information about the llvm-commits mailing list